article stringlengths 1.98k 169k | summary stringlengths 1.01k 4.15k | section_headings listlengths 2 38 | keywords listlengths 0 12 | year stringclasses 11 values | title stringlengths 30 189 |
|---|---|---|---|---|---|
A puzzling property of synaptic transmission , originally established at the neuromuscular junction , is that the time course of transmitter release is independent of the extracellular Ca2+ concentration ( [Ca2+]o ) , whereas the rate of release is highly [Ca2+]o-dependent . Here , we examine the time course of release at inhibitory basket cell-Purkinje cell synapses and show that it is independent of [Ca2+]o . Modeling of Ca2+-dependent transmitter release suggests that the invariant time course of release critically depends on tight coupling between Ca2+ channels and release sensors . Experiments with exogenous Ca2+ chelators reveal that channel-sensor coupling at basket cell-Purkinje cell synapses is very tight , with a mean distance of 10–20 nm . Thus , tight channel-sensor coupling provides a mechanistic explanation for the apparent [Ca2+]o independence of the time course of release .
Calcium plays a key role in the control of transmitter release at chemical synapses ( Neher , 1998; Südhof , 2013 ) . As transmitter release is a Ca2+-dependent biochemical process , one would expect that both the extent and kinetics of release will depend on Ca2+ concentration . In contrast to this expectation , the time course of release ( TCR ) at the neuromuscular junction is independent of the extracellular Ca2+ concentration ( [Ca2+]o ) ( Datyner and Gage , 1980; Van der Kloot , 1988; Parnas et al . , 1989 ) . However , the apparent [Ca2+]o independence of the TCR is less well established at central synapses ( Sargent et al . , 2005 ) . Furthermore , the underlying mechanisms remain elusive . Several potential explanations were proposed , including adaptation of transmitter release ( Hsu et al . , 1996 ) and additional voltage sensors that control the timing of release ( Parnas et al . , 2000; See; Felmy et al . , 2003 ) . Recent results suggested that in several central synapses presynaptic Ca2+ channels and release sensors are tightly coupled ( Bucurenciu et al . , 2008; Eggermann et al . , 2012; Scimemi and Diamond , 2012; Schmidt et al . , 2013 ) . Tight coupling might be another factor contributing to the apparent [Ca2+]o independence of the TCR ( Yamada and Zucker , 1992 ) . However , this possibility has not been directly examined .
We studied the [Ca2+]o dependence of transmitter release at GABAergic synapses between cerebellar basket cells ( BCs ) and Purkinje cells ( PCs ) , using paired whole-cell recordings from synaptically connected neurons in mouse brain slices ( Caillard et al . , 2000; Sakaba , 2008; Eggermann and Jonas , 2012; Figure 1A–D ) . Action potentials were evoked in the presynaptic BC in whole-cell current-clamp or cell-attached voltage-clamp configurations , whereas IPSCs were recorded in the postsynaptic PC under whole-cell voltage-clamp conditions ( series resistance 3 . 8 ± 0 . 1 MΩ; mean ± SEM; 92 cells; ‘Materials and methods’ ) . Unitary IPSCs were initiated with short latency and high temporal precision . Synaptic latency was 1 . 20 ± 0 . 03 ms at ∼22°C ( 24 pairs ) and 0 . 47 ± 0 . 02 ms at ∼34°C ( 5 pairs; Figure 1C , D ) . Synaptic transmission was entirely blocked by bath application of the selective P/Q-type Ca2+ channel blocker ω-agatoxin IVa ( 1 µM ) , whereas the N-type Ca2+ channel blocker ω-conotoxin GVIa ( 1 µM ) had no detectable effect ( Figure 1E–H ) . Thus , transmitter release at this synapse is selectively mediated by P/Q-type Ca2+ channels . 10 . 7554/eLife . 04057 . 003Figure 1 . Fast and synchronous transmitter release at BC–PC synapses in the cerebellum is exclusively mediated by P/Q-type Ca2+ channels . ( A ) Light micrograph of a cerebellar basket cell filled with biocytin during recording and labeled using 3 , 3′-diaminobenzidine as chromogen . Similar morphological properties were obtained in 29 other biocytin-labeled cerebellar BCs . ( B ) Plot of IPSC peak amplitude ( top ) and corresponding series resistance ( bottom ) against time during recording . Action potentials were evoked in the presynaptic cell at time intervals of 4 s . Note that evoked IPSC shows only little rundown for more than 40 min with stable series resistance . ( C ) Presynaptic action potentials evoked in the whole-cell current clamp configuration ( top ) and evoked IPSCs ( bottom ) recorded at ∼22°C ( left ) and ∼34°C ( right ) . 10 consecutive individual traces ( gray ) and the corresponding average trace ( black ) are shown superimposed . ( D ) Summary bar graph . Left , latency ( time between steepest point in the rising phase of the presynaptic action potential and IPSC onset ) at ∼22°C and ∼34°C . Right , standard deviation of latency , a measure of synchrony of transmitter release . Bars indicate mean ± SEM; solid circles represent data from individual experiments ( ∼22°C: 24 pairs; ∼34°C: 5 pairs ) . ( E ) IPSCs in a BC–PC pair before ( left ) and after ( right ) application of 1 µM of the selective P/Q-type Ca2+ channel blocker ω-agatoxin IVa . Synaptic transmission was almost completely blocked . Top , presynaptic action currents evoked in the cell-attached voltage-clamp configuration; bottom , corresponding IPSCs . 10 consecutive individual traces ( gray ) and the corresponding average trace ( red ) are shown superimposed . ( F ) Similar recording as shown in ( E ) , but with 1 µM of the selective N-type Ca2+ channel blocker ω-conotoxin GVIa . ( G ) Plot of IPSC peak amplitude against time during application of 1 µM of ω-agatoxin IVa ( red ) or ω-conotoxin GVIa ( blue ) . The time of application of the different extracellular solutions is represented by horizontal bars . Black: mock application of control solution . Symbols indicate mean; error bars represent SEM . ( H ) Summary bar graph of the effects of Ca2+ channel blockers . Bars indicate mean ± SEM; solid circles represent data from individual experiments ( control: 4 pairs; ω-agatoxin IVa: 4 pairs; ω-conotoxin GVIa: 5 pairs ) . To achieve maximal stability , presynaptic BCs were noninvasively stimulated in the cell-attached configuration in all experiments . * and ** indicate p < 0 . 05 and 0 . 01 , respectively . All experiments except subsets in ( C ) , ( D ) were performed at ∼22°C . DOI: http://dx . doi . org/10 . 7554/eLife . 04057 . 003 Exploiting the technical advantages of this synapse ( including ideal voltage-clamp conditions conveyed by perisomatic synapse location , presynaptic accessibility due to short axon trajectories , and optimal signal-to-noise ratio because of large quantal size ) , we first examined the [Ca2+]o dependence of release efficacy . Analysis of the dependence of evoked IPSC peak amplitude on [Ca2+]o revealed that the amount of transmitter release was steeply [Ca2+]o-dependent , with a high power coefficient ( n = 3 . 02 in the low-concentration limit; Figure 2A , B ) . Thus , release at the BC–PC synapse was highly cooperative ( Dodge and Rahamimoff , 1967 ) . We next measured the [Ca2+]o dependence of release kinetics . To determine the TCR , we first recorded unitary IPSCs at a given [Ca2+]o , and subsequently in reduced [Ca2+]o to isolate quantal IPSCs ( Figure 2C ) . We then computed the TCR by deconvolution of unitary and quantal IPSC waveforms ( Diamond and Jahr , 1995; Neher and Sakaba , 2001; Sakaba , 2008; Figure 2D ) . When [Ca2+]o was changed in the range from 0 . 7–4 mM , peak release rate changed by a factor of 34 . 4 . In contrast , the half-duration of the TCR was only minimally affected ( 0 . 49 ± 0 . 05 ms at 0 . 7 mM; 0 . 43 ± 0 . 05 ms at 1 mM; 0 . 47 ± 0 . 01 ms at 2 mM; 0 . 47 ± 0 . 02 ms at 4 mM; 5–11 pairs; p = 0 . 56; Figure 2D , E ) . Furthermore , the decay time constant of the TCR was similar in the different conditions ( p = 0 . 38; Figure 2D , E ) . These results show that the TCR is largely [Ca2+]o-independent at a central synapse . Similar conclusions were reached at lower recording temperature , which would be expected to markedly slow down the kinetic rates of channels and sensors , but to only minimally change the rates of diffusional processes . At ∼12°C , changes in [Ca2+]o had no detectable effects on the half-duration of the TCR ( 2 . 84 ± 0 . 42 ms at 1 mM , 2 . 97 ± 0 . 24 ms at 2 mM , and 2 . 79 ± 0 . 25 ms at 4 mM [Ca2+]o; 5 pairs; p = 0 . 93 ) ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 04057 . 004Figure 2 . Changes in [Ca2+]o affect peak release rate , but leave the TCR unaffected at the cerebellar BC–PC synapse . ( A ) Presynaptic action potential ( top ) and evoked IPSCs ( bottom ) in different [Ca2+]o . 10 consecutive individual traces ( gray ) and the corresponding average trace ( red , 1 mM [Ca2+]o; black , 2 mM; blue , 4 mM ) are shown superimposed . Inset shows quantal IPSCs; 10 individual IPSCs recorded in 0 . 7 mM [Ca2+]o ( gray ) and the corresponding average trace ( green ) are shown superimposed . ( B ) Relationship between IPSC peak amplitude and [Ca2+]o . IPSC amplitudes were normalized to the value at 2 mM [Ca2+]o and averaged across pairs . Top , linear–logarithmic representation . Data were fit with a Hill equation , yielding a maximal value ( a ) of 3 . 73 , a half-maximal effective concentration ( EC50 ) of 3 . 09 mM , and an average Hill coefficient ( n ) of 2 . 39 . Bottom , double-logarithmic plot; data points for [Ca2+]o ≤ 2 mM were analyzed by linear regression , yielding a Hill coefficient of 3 . 02 in the low concentration limit . ( C ) Relationship between peak amplitude of IPSC successes and [Ca2+]o . Symbols indicate mean , error bars represent SEM . The peak amplitude of the successes reaches a constant level below 1 mM [Ca2+]o , suggesting that the level of quantal IPSCs was reached . ( D ) TCR in different [Ca2+]o obtained by deconvolution . Top , presynaptic action potential . Middle , average evoked IPSCs in different [Ca2+]o . Bottom , TCR in different [Ca2+]o . Inset , expanded waveforms of the TCR ( red , 1 mM [Ca2+]o; black , 2 mM; blue , 4 mM; gray , fit Gaussian curves ) . Data in ( A ) and ( D ) are from the same pair . ( E ) Peak release rate ( top ) , half-duration ( middle ) , and decay time course of release period ( bottom ) plotted vs [Ca2+]o . Open circles connected by lines indicate mean ± SEM; solid circles indicate data from individual experiments ( 0 . 7 mM: 5 pairs; 1 mM: 7 pairs; 2 mM: 11 pairs; 4 mM: 5 pairs ) . All experiments were performed at ∼22°C . ( Also see Figure 2—figure supplement 1 , 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04057 . 00410 . 7554/eLife . 04057 . 005Figure 2—figure supplement 1 . Lowering temperature slows TCR , but leaves its [Ca2+]o independence unaffected . ( A ) Presynaptic action potential ( top ) and evoked IPSCs ( bottom ) in different [Ca2+]o at ∼12°C . 10 consecutive individual traces ( gray ) and the corresponding average trace ( red , 1 mM [Ca2+]o; black , 2 mM; blue , 4 mM ) are shown superimposed . ( B ) Quantal IPSCs at ∼12°C . 10 individual IPSCs recorded in 1 mM [Ca2+]o ( gray ) and the corresponding average trace ( green ) are shown superimposed . Note that the action potentials , evoked IPSCs , and quantal IPSCs are markedly slower than those obtained at ∼22°C ( compare with Figure 2A , C ) . ( C ) TCR in different [Ca2+]o obtained by deconvolution at ∼12°C . Top , presynaptic action potential . Middle , average evoked IPSCs in different [Ca2+]o . Bottom , TCR in different [Ca2+]o ( red , 1 mM [Ca2+]o; black , 2 mM; blue , 4 mM ) . Data in ( A ) – ( C ) are from the same pair . ( D ) Summary bar graph of half-duration of TCR in different [Ca2+]o at ∼12°C . Bars indicate mean ± SEM; solid circles represent data from individual experiments ( 5 pairs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04057 . 00510 . 7554/eLife . 04057 . 006Figure 2—figure supplement 2 . TCR is unlikely to be distorted by postsynaptic receptor saturation or desensitization . ( A ) Presynaptic action potential ( top ) and evoked IPSCs in 4 mM [Ca2+]o ( bottom ) before and after application of 300 µM of the low-affinity GABAA receptor antagonist TPMPA . 2 µM CGP55845 were added during the entire experiment to block GABAB receptors . 10 consecutive individual traces ( gray ) and the corresponding average trace ( black , control; red , TPMPA ) are shown superimposed . ( B ) Summary bar graph of IPSC peak amplitude in control and 300 µM TPMPA . Bars indicate mean ± SEM; solid circles represent data from individual experiments; lines indicate measurements taken from the same experiment ( control: 12 pairs; TPMPA: 6 pairs ) . ( C ) TCR in 300 µM TPMPA obtained by deconvolution . Top , presynaptic action potential . Middle , average evoked IPSC . Bottom , TCR . Inset , average quantal IPSC obtained in 300 µM TPMPA . Data in ( A ) and ( B ) were obtained from different pairs . ( D ) Summary bar graph of half-duration of TCR in control and 300 µM TPMPA . Bars indicate mean ± SEM; solid circles represent data from individual experiments ( control: 8 pairs; TPMPA: 6 pairs; p = 0 . 38 ) . In all experiments , [Ca2+]o was 4 mM ( presumably maximizing the confounding effects of postsynaptic factors ) and the recording temperature was ∼22°C . DOI: http://dx . doi . org/10 . 7554/eLife . 04057 . 006 To explore the mechanisms underlying the paradoxical [Ca2+]o independence of the TCR , we used a realistic model of action potential-dependent transmitter release ( Eggermann and Jonas , 2012; Figure 3 ) . Ca2+ diffusion and buffering were computed by solving the full set of partial differential diffusion and reaction equations , and the TCR was simulated using a previously established release sensor model ( Lou et al . , 2005; Figure 3A , B ) . Finally , the half-duration of the TCR was plotted against the peak release rate ( PRR ) . Computational analysis revealed that the TCR was markedly dependent on the PRR in both low and high PRR limit . In the low PRR limit , the half-duration of the TCR decreased with PRR in a model with fixed coupling distance ( Figure 3C ) . This dependence was accentuated after slowing of release sensor rates ( Figure 3D , top ) , indicating that rate-limiting sensor kinetics are responsible . Furthermore , this dependence was inverted in a model with variable coupling distance ( Figure 3E ) . In such a configuration , small Ca2+ inflow may selectively release proximal vesicles , whereas large Ca2+ inflow may recruit all vesicles , resulting in a broadening of the TCR . In the high PRR limit , the half-duration of the TCR increased with PRR ( Figure 3C ) . This behavior was particularly prominent in small boutons ( Figure 3F , bottom ) , suggesting that saturation of the endogenous buffers leads to broadening of the TCR . In all configurations , the dependence of the TCR on the PRR was substantially more prominent in loose ( 200 nm , blue colors ) than in tight coupling configurations ( 20 nm , red colors ) . Similar conclusions were reached when the concentration and unbinding rate of fixed buffers were altered ( Figure 3—figure supplement 1; Gilmanov et al . , 2008 ) . Thus , whereas several factors influence the shape of the TCR–PRR relation , the coupling distance plays a key role in all conditions examined . 10 . 7554/eLife . 04057 . 007Figure 3 . Tight Ca2+ channel–release sensor coupling reduces the [Ca2+]o dependence of the TCR in a release model . ( A ) Schematic illustration of the release model . Top , location of Ca2+ source and release sensor within a schematic presynaptic terminal; bottom left , Ca2+ sensor model; bottom right , color code indicating coupling distance in subsequent plots ( C–F ) . The presynaptic terminal was modeled as a hemisphere . Ca2+ inflow was generated by a point source in the center ( red square and surrounding shaded area ) . The release sensor was placed at variable distance from the source ( red line ) . Ca2+ transients were calculated as the numerical solution to the full set of partial differential reaction-diffusion equations ( Smith , 2001 ) . Transmitter release was computed using a modified version of a previously described sensor model ( Lou et al . , 2005 ) . kon and koff are Ca2+-binding and unbinding rates , l+ is basal release rate , and f and b are the cooperativity factors for release and Ca2+ unbinding , respectively . ( B ) Left , [Ca2+]i plotted vs time . Right , corresponding release rate . Top , tight coupling ( 20 nm ) ; bottom , loose coupling ( 200 nm distance ) . Black: default Ca2+ inflow ( 3 . 5 and 104 . 4 Ca2+ channel equivalents , leading to a vesicular release rate of ∼2000 s−1 in both cases ) ; red: reduced Ca2+ inflow ( x 0 . 25 ) ; blue: increased Ca2+ inflow ( x 2 ) . Bottom , contour plots of peak release rate ( left ) and half-duration of the TCR ( right ) , plotted vs coupling distance ( horizontal axis ) and Ca2+ inflow ( vertical axis , normalized to that of single Ca2+ channel ) . Numbers right-adjacent to the contour lines indicate peak release rate and half-duration of the TCR , respectively . Bouton diameter 1 . 0 µm . ( C ) Plot of half-duration of the TCR vs peak release rate for different coupling distances ( individual curves for 10 to 200 nm in 5 nm steps; scale bar for color in ( A ) ) . Bottom graph shows expansion . Bouton diameter 1 . 0 µm . Red arrowheads indicate peak release rates at 0 . 7 , 1 , 2 , and 4 mM [Ca2+]o , estimated from the IPSC–[Ca2+]o curve in Figure 2B . ( D ) Half-duration of TCR–peak release rate relations for different sensor rates . Default sensor rates were slowed ( x 0 . 5 ) or accelerated ( x 20 , to make the kinetics of the release sensor very fast in comparison to all other processes ) . ( E ) Half-duration of TCR–peak release rate relations for variable coupling distance ( CV of 0 . 3 ) . Bottom graph shows expansion . ( F ) Half-duration of TCR–peak release rate relations for different bouton diameters . Dashed lines and points in ( C ) and ( E ) indicate the slope of the TCR–peak release rate relations for 20 nm and 200 nm coupling distance at a release rate of 200 quanta s−1 . Note that the absolute value of the slope is ∼12 times and ∼7 times higher , respectively , for loose than for tight coupling . Also see Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04057 . 00710 . 7554/eLife . 04057 . 008Figure 3—figure supplement 1 . Effects of concentration and affinity of the endogenous fixed buffer on the [Ca2+]o dependence of the TCR in the release model . ( A ) Plot of half-duration of the TCR vs peak release rate for different coupling distances in a model with increased concentration of the endogenous fixed buffer ( 0 . 5 mM ) . Red arrowheads indicate peak release rates at 0 . 7 , 1 , 2 , and 4 mM [Ca2+]o , estimated from the IPSC–[Ca2+]o curve in Figure 2B . ( B ) Plot of half-duration of the TCR vs peak release rate for different coupling distances in a model with reduced Ca2+-unbinding rate of the fixed buffer ( koff , 40 s−1; Eggermann and Jonas , 2012 ) . Individual curves were drawn for coupling distances ranging from 10 to 200 nm in 5 nm steps; scale bar for color shown in Figure 3A . Bouton diameter 1 . 0 µm in all simulations . Bottom graph shows expansion . Dashed lines and points indicate the slope of the TCR–peak release rate relations for 20 nm and 200 nm coupling distance at a release rate of 200 quanta s−1 . Note that the absolute value of the slope is higher for loose than for tight coupling in all cases . DOI: http://dx . doi . org/10 . 7554/eLife . 04057 . 008 Our modeling results suggest the intriguing possibility that tight coupling between Ca2+ channels and release sensors ( Christie et al . , 2011 ) might be a main reason for the apparent [Ca2+]o independence of the TCR at the cerebellar BC–PC synapse . To test this hypothesis quantitatively , we probed the distance between Ca2+ channels and release sensors , using exogenous Ca2+ chelators ( Adler et al . , 1991; Neher , 1998; Eggermann et al . , 2012; Figure 4 ) . Ca2+ chelators were loaded into presynaptic BC terminals through the somatic patch pipette , and the reversibility of their effects was tested by obtaining a second recording from the same presynaptic BC at later times . Whereas the fast Ca2+ chelator ethylenedioxybis- ( o-phenylenenitrilo ) -N , N , N' , N'-tetraacetic acid ( BAPTA ) markedly suppressed transmitter release ( Figure 4A , B ) , the slow Ca2+ chelator ethyleneglycol-bis ( 2-aminoethylether ) -N , N , N' , N'-tetraacetic acid ( EGTA ) was much less effective ( Figure 4C , D ) . Analysis of concentration–effect data revealed that the half-maximal inhibitory concentration ( IC50 ) for the steady-state effect was 0 . 6 mM for BAPTA ( 14 pairs total ) and 16 . 0 mM for EGTA ( 11 pairs total; Figure 4E ) . Thus , the IC50 value was ∼27-fold higher for EGTA than for BAPTA . 10 . 7554/eLife . 04057 . 009Figure 4 . Nanodomain coupling between Ca2+ channels and release sensors at the cerebellar BC–PC synapse . ( A ) Presynaptic action potentials and evoked IPSCs recorded immediately after presynaptic BC break-in ( left ) , 20 min after break-in to load 3 mM BAPTA into presynaptic terminals ( middle ) , and after the presynaptic recoding pipette was changed to one containing control solution ( right ) . 10 consecutive individual traces ( gray ) and the corresponding average trace ( red ) are shown superimposed . ( B ) Time course of the effect of 1 mM ( ○ , 5 pairs ) , 3 mM ( △ , 4 pairs ) and 10 mM ( ▿ , 5 pairs ) BAPTA . Data from each pair were normalized to the average amplitude obtained from the first 20 consecutive traces and averaged across pairs . ( C ) Similar recording as shown in ( A ) , except that the internal solution for the presynaptic BC contained 30 mM EGTA . ( D ) Time course of the effect of 0 . 1 mM ( ○ , control , 4 pairs ) , 10 mM ( △ , 5 pairs ) and 30 mM ( ▿ , 6 pairs ) EGTA . In all experiments included in ( B ) and ( D ) , the presynaptic BC was patched twice to demonstrate recovery . Symbols indicate mean , error bars represent SEM . ( E ) Plot of steady-state effects of chelators on IPSC peak amplitude against chelator concentration for BAPTA ( red circles ) and EGTA ( blue circles ) . Curves represent Hill equations fit to the data points . IC50 values were 0 . 6 mM and 16 . 0 mM , Hill coefficients were 0 . 96 and 1 . 51 , respectively . ( F ) Same data , fit with linearized models of Ca2+ diffusion and buffering ( continuous curve , constant coupling distance; small dashing , half-normally distributed coupling distance; large dashing , skewed-normally distributed coupling distance; corresponding distributions shown as insets on the right ) . For the single-channel model , the coupling distance was estimated as 11 . 4 nm . For the half-normal distribution , the standard deviation was 10 . 5 nm ( expectation value 13 . 9 nm; skewness 1 . 0 ) . For the skewed-normal distribution , location , scale , and shape parameter were 13 . 7 nm , 8 . 6 nm , and −1 . 11 , respectively ( expectation value 10 . 1 nm; skewness 0 . 40 ) . ( G ) Statistical errors . Histogram of coupling distance estimates in 1000 bootstrap replications . ( H ) Systematic errors . Plot of coupling distance against resting Ca2+ concentration and endogenous buffer product . Red point indicates default parameter values . In ( G ) and ( H ) , the model with constant coupling distance was used . All experiments were performed at ∼22°C . DOI: http://dx . doi . org/10 . 7554/eLife . 04057 . 009 To estimate the coupling distance , we fit the concentration–effect data for BAPTA and EGTA with a model of Ca2+ diffusion and buffering based on linear approximations ( Neher , 1998; Figure 4F ) . The main free parameter in the model was the distance between Ca2+ channels and release sensors , while several other parameters ( e . g . the physicochemical properties of the Ca2+ chelators and the cooperativity of transmitter release ) were well constrained . Analysis of the entire data set revealed a coupling distance of 10–20 nm ( 11 . 4 nm for constant coupling distance; 13 . 9 nm for half-normally distributed coupling distance; 10 . 1 nm for skewed-normally distributed coupling distance; Figure 4F ) . Statistical errors , as assessed by bootstrap analysis , were minimal ( 1 . 2 nm; Figure 4G ) . Furthermore , systematic errors were small , as indicated by the robustness of the estimate against variations in resting Ca2+ concentration and endogenous buffer product ( Figure 4H ) . These results reveal that the coupling between Ca2+ channels and release sensors at the cerebellar BC–PC synapse is tight .
Our results address a long-standing question in the field of synaptic transmission ( Yamada and Zucker , 1992; Parnas et al . , 2000 ) : If release probability is highly dependent on extracellular Ca2+ concentration ( Dodge and Rahamimoff , 1967 ) , how is it possible that the timing of release is [Ca2+]o-independent ( Datyner and Gage , 1980; Van der Kloot , 1988; Parnas et al . , 1989 ) ? To address this question , we measured the [Ca2+]o dependence of both the amount and the time course of transmitter release at the cerebellar BC–PC synapse , a central synapse in which the TCR can be precisely quantified . Our results demonstrate that while the amount of release is highly [Ca2+]o-dependent with a power coefficient of ∼3 , the TCR is largely [Ca2+]o-independent . Although the apparent [Ca2+]o independence of the TCR has been well established at the neuromuscular junction ( Datyner and Gage , 1980; Van der Kloot , 1988; Parnas et al . , 1989 ) , this phenomenon is less well documented at central synapses ( see Sargent et al . , 2005 for a notable exception ) . Thus , our results demonstrate that the [Ca2+]o independence of the TCR is a general phenomenon characteristic for synaptic transmission in both peripheral and central nervous system . We further measured the coupling distance between Ca2+ channels and release sensors at BC–PC synapses . Our results show that coupling is tight , with a coupling distance of 10–20 nm . Although several previous studies measured the coupling distance , the rules that define the coupling configuration remain elusive ( Hefft and Jonas , 2005; Bucurenciu et al . , 2008; Christie et al . , 2011; Nadkarni et al . , 2012; Scimemi and Diamond , 2012; Schmidt et al . , 2013; Vyleta and Jonas , 2014 ) . It has been suggested that tight coupling is primarily used at synapses designed for fast , reliable transmission , whereas loose coupling is utilized at synapses specialized on presynaptic plasticity ( Bucurenciu et al . , 2008; Eggermann et al . , 2012; Nadkarni et al . , 2012; Vyleta and Jonas , 2014 ) . The present results are fully consistent with this hypothesis , since GABAergic BC–PC synapses show fast transmitter release following single action potentials and reliable transmission during trains of spikes ( Caillard et al . , 2000; Sakaba , 2008; Eggermann and Jonas , 2012 ) . The present results identify novel links between the [Ca2+]o independence of the TCR and nanodomain coupling . Our model reveals that the [Ca2+]o dependence is substantially more prominent in loose than in tight coupling regimes in a variety of conditions . These include different Ca2+ sensor rates , different bouton diameters , various concentrations and affinity values of endogenous buffers , and uniform vs non-uniform coupling . Specifically , our model predicts that slowing of sensor rates should markedly enhance the [Ca2+]o dependence of the TCR in a microdomain , but not in a nanodomain coupling regime . We tested this prediction by lowering the temperature , which , among other potential effects , is expected to slow sensor rates . Whereas lowering the temperature had large effects on the absolute value of the half-duration of the TCR , its [Ca2+]o independence was unchanged . These findings provide experimental evidence that tight source–sensor coupling is a key factor that ensures the [Ca2+]o independence of the TCR . Previous studies highlighted several functional advantages of nanodomain coupling , including efficacy , speed , temporal precision , and energy efficiency of synaptic transmission ( Bucurenciu et al . , 2008; Eggermann et al . , 2012 ) . Our results suggest another functional benefit: conveying [Ca2+]o independence to the TCR . Is the invariance of the TCR relevant for microcircuit function under physiological conditions ? Maintenance of speed and temporal precision at the GABAergic BC–PC synapse is of critical importance for the operation of the cerebellum , since feedforward inhibition mediated by BCs sets the temporal window of signal integration in PCs ( Mittmann et al . , 2005; Bao et al . , 2010 ) . However , [Ca2+]o has been shown to fluctuate during repetitive activity , high-frequency network oscillations , and in pathophysiological conditions ( Heinemann et al . , 1977; Borst and Sakmann , 1999; Rusakov and Fine , 2003 ) . Furthermore , Ca2+ inflow will be changed by neuromodulators , which often act via inhibition of presynaptic Ca2+ channels ( Takahashi et al . , 1998 ) . Thus , nanodomain coupling at BC–PC synapses may ensure constant timing of fast feedforward inhibition under a variety of network conditions .
Slices were cut from the cerebellum of 14- to 16-day-old C57/Bl6 wild-type mice of either sex . Experiments were performed in strict accordance with institutional , national , and European guidelines for animal experimentation . Mice were maintained under light ( 7 am–7 pm ) and dark cycle ( 7 pm–7 am ) conditions and were kept in a litter of 8 animals together with the mother in a single cage . Animals were lightly anesthetized using isoflurane ( Forane , AbbVie , Austria ) and sacrificed by rapid decapitation . The brain was rapidly dissected out and immersed in ice-cold slicing solution containing 87 mM NaCl , 25 mM NaHCO3 , 2 . 5 mM KCl , 1 . 25 mM NaH2PO4 , 10 mM D-glucose , 75 mM sucrose , 0 . 5 mM CaCl2 , and 7 mM MgCl2 , ( pH 7 . 4 in 95% O2 / 5% CO2 , ∼326 mOsm ) . Parasagittal 250-µm-thick cerebellar slices from the vermis region were cut using a custom-built vibratome . After ∼20 min incubation at ∼35°C , the slices were stored at room temperature . Experiments were performed at 21–23°C , unless specified differently , or at either 11–13°C or 32–35°C in subsets of experiments as indicated . During experiments , slices were superfused with a bath solution containing 125 mM NaCl , 2 . 5 mM KCl , 25 mM NaHCO3 , 1 . 25 mM NaH2PO4 , 25 mM D-glucose , 2 mM CaCl2 , and 1 mM MgCl2 ( pH 7 . 4 in 95% O2 / 5% CO2 , ∼316 mOsm ) . To investigate the relationship between IPSC peak amplitude and [Ca2+]o , different combinations of [Ca2+]o / [Mg2+]o were used ( 0 . 5 / 2 . 5 , 0 . 6 / 2 . 4 , 0 . 7 / 2 . 3 , 0 . 8 / 2 . 2 , 1 / 2 , 3 / 1 , 4 / 1 , and 10 / 1 mM ) . Paired recordings from synaptically connected BCs and PCs were performed as previously described ( Caillard et al . , 2000; Sakaba , 2008; Eggermann and Jonas , 2012 ) . Intracellular solution used for presynaptic BCs contained 125 mM K-gluconate , 20 mM KCl , 10 mM HEPES , 10 mM phosphocreatine , 2 mM MgCl2 , 0 . 1 mM EGTA , 2 mM ATP , 0 . 4 mM GTP ( pH adjusted to 7 . 3 with KOH , ∼310 mOsm ) . In a subset of experiments , 0 . 2% biocytin was added . For experiments using Ca2+ chelators , 0 . 1 mM EGTA was replaced with different concentrations of BAPTA ( 1 , 3 , or 10 mM ) or EGTA ( 10 or 30 mM ) ; the concentration of K-gluconate was reduced accordingly to maintain osmolarity . Presynaptic pipettes were fabricated from borosilicate glass tubing . Presynaptic pipette resistance was 12–15 MΩ . Intracellular solution for postsynaptic PCs contained 140 mM KCl , 10 mM HEPES , 2 mM MgCl2 , 10 mM EGTA , 2 mM ATP , and 2 mM QX-314 ( pH adjusted to 7 . 3 with KOH , ∼310 mOsm ) . To achieve small postsynaptic series resistance , leaded glass ( PG10165-4 , WPI , Sarasota , FL ) was used to fabricate large tip-sized recording pipettes . To minimize capacitance , pipettes were coated with dental wax . Postsynaptic pipette resistance was 0 . 8–1 . 5 MΩ , resulting in a mean series resistance of 3 . 8 ± 0 . 1 MΩ ( range: 2 . 5–7 . 5 MΩ , 92 cells ) . Postsynaptic series resistance was not compensated , but continuously monitored using 5-mV test pulses . Experiments were only analyzed if changes in series resistance in the entire recording period were less than 2 MΩ . Similarly , experiments were discarded if there was a detectable rundown of IPSC peak amplitude during the control period . For intracellular stimulation of BCs in the whole-cell configuration under current-clamp conditions , single pulses ( 400 pA , 4 ms at ∼22°C and ∼34°C; 500 pA , 5 ms at ∼12°C ) were injected into the BC every 4 s ( ∼22°C and ∼34°C ) or 15 s ( ∼12°C ) . A holding current of ∼ –50 pA was applied to maintain the resting membrane potential at ∼–65 mV and to avoid spontaneous action potential generation . For cell-attached stimulation under voltage-clamp conditions ( Perkins , 2006; Vyleta and Jonas , 2014 ) , the presynaptic pipette contained a K+-based intracellular solution . Action potentials were evoked by brief voltage pulses ( amplitude <1 V , duration 0 . 1–0 . 2 ms ) . Pipette holding potential was set to −60 or −80 mV to minimize the holding current and to avoid spontaneous action potential generation . Above a threshold value , action currents in BCs and IPSCs in PCs were evoked , demonstrating reliable all-or-none activation of the synapse . Experiments in which presynaptic seal resistance went below 1 GΩ during recording were discarded . In all experiments , PCs were recorded under voltage-clamp conditions at a holding potential of −70 mV . Membrane potentials given were not corrected for liquid junction potentials . Temperature control of bath solution was achieved using a temperature controller ( Sigmann , Germany ) in combination with a high precision electronic thermometer ( GHM , Germany ) placed near the specimen . For cooling , the inflow tubing was placed in an ice reservoir . Peptide toxins were applied using a recirculation system with a peristaltic pump ( Ismatec , Germany ) . The total volume of the system was ∼4 ml , and the solution was equilibrated with 95% O2 / 5% CO2 . Bovine serum albumin ( Sigma-Aldrich , St . Louis , MI ) was added at a concentration of 1 mg ml−1 to prevent adsorption of peptides to the surfaces of the perfusion system . ω-agatoxin IVa and ω-conotoxin GVIa were from Bachem ( Switzerland ) , ( 1 , 2 , 5 , 6-tetrahydropyridine-4-yl ) -methylphosphinic acid ( TPMPA ) and CGP55845 hydrochloride were from Tocris ( UK ) , other chemicals were obtained from Sigma–Aldrich or Merck ( Germany ) , unless specified differently . For analysis of neuron morphology , slices were fixed >24 hr in 2 . 5% paraformaldehyde , 1 . 25% glutaraldehyde , and 15% saturated picric acid in 100 mM phosphate buffer ( PB; pH 7 . 35 ) . After fixation , slices were washed , incubated in 2% hydrogen peroxide , and shock-frozen in liquid nitrogen . Subsequently , the tissue was treated with PB containing 1% avidin–biotinylated horseradish peroxidase complex ( ABC; Vector Laboratories , Burlingame , CA ) overnight at 4°C . Excess ABC was removed by several rinses with PB , before development with 0 . 05% 3 , 3′-diaminobenzidine tetrahydrochloride and 0 . 01% hydrogen peroxide . Subsequently , slices were rinsed in PB several times and embedded in Mowiol ( Roth , Germany ) . Data were acquired with a Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , CA ) . Signals were filtered at 6 kHz ( 4-pole low-pass Bessel filter ) and digitized at 20 or 50 kHz using a CED 1401plus interface ( Cambridge Electronic Design , UK ) . Pulse generation and data acquisition were performed using FPulse 3 . 33 ( U . Fröbe , Physiological Institute Freiburg , Germany ) running under Igor Pro 6 . 3 . 2 ( Wavemetrics , Portland , OR ) on a PC . Data were analyzed with Igor Pro 6 . 3 . 2 and Mathematica 8 . 0 . 1 or 9 . 01 ( Wolfram Research , Champaign , IL ) . Synaptic latency ( the steepest point of the rise phase of the presynaptic action potential to the onset of the IPSC ) , and proportion of failures were determined from 50 to 800 single traces . To quantify the block of transmitter release by BAPTA and EGTA , the peak amplitude of the evoked IPSCs against experimental time was fit with an exponential or sigmoidal function , and the amount of suppression was quantified as the ratio of steady-state to initial values of the fit curve . Concentration–effect curves ( IPSC vs [Ca2+]o or [chelator] ) were fit with a Hill equation of the form f ( c ) = a [1 + ( c50 / c ) n]−1 , where c is concentration , a is maximal amplitude , c50 is half-maximal effective ( EC50 ) or inhibitory concentration ( IC50 ) , and n is Hill coefficient . The time course of release ( TCR ) was determined by deconvolution ( Diamond and Jahr , 1995; Neher and Sakaba , 2001; Sakaba , 2008 ) . Average unitary IPSCs ( IPSCunitary ) were deconvolved from average quantal IPSCs ( IPSCquantal ) as F−1[F ( IPSCunitary ) / F ( IPSCquantal ) ] , where F is the discrete Fourier transform and F−1 is the inverse . Unitary IPSCs were aligned to the steepest point in the rising phase of the corresponding presynaptic action potentials ( to account for minimal jitter in the timing of action potential initiation ) . Quantal IPSCs recorded in conditions of reduced [Ca2+]o were aligned by the rising phase of the individual events and then averaged . For subsequent analysis and display , the TCR was filtered at 5 kHz ( for recordings at ∼22°C or ∼34°C ) or 0 . 5 kHz ( for measurements at ∼12°C ) using a digital filter . Finally , the filtered TCR was fit with a Gaussian function . The effects of filtering were corrected by subtracting the variance of an impulse response of a Gaussian filter ( Colquhoun and Sigworth , 1995 ) . At 2 mM [Ca2+]o , the quantal content , estimated as the integral under the TCR , was 10 . 4 ± 1 . 8 . Analysis was performed using Mathematica 8 . 0 . 1 running under Windows 7 on a PC . To test the possibility that postsynaptic factors , such as receptor desensitization , receptor saturation , or GABA spillover affected our measurements , we measured the TCR in the presence of the low-affinity competitive antagonist 300 µM TPMPA ( Jones et al . , 2001 ) and 2 µM CGP55845 to avoid effects of TPMPA on GABAB receptors . In 4 mM [Ca2+]o ( presumably maximizing the confounding effects of postsynaptic factors ) , the half-duration of the TCR was 0 . 51 ± 0 . 04 in the presence ( Figure 2—figure supplement 2; 6 pairs ) vs 0 . 47 ± 0 . 02 in the absence of TPMPA ( Figure 2E; p = 0 . 38 ) . To quantify the coupling distance between Ca2+ channels and release sensors , concentration–effect data for both BAPTA and EGTA were first fit with a Hill equation . Next , data were analyzed with a model of Ca2+ diffusion and buffering based on linear approximations ( Neher , 1998; Bucurenciu et al . , 2008 ) . The ratio of Ca2+ transients in the presence and absence of chelators was converted into the ratio of release probabilities , using a power function with the Hill coefficient set according to the slope of the double-logarithmic IPSC–[Ca2+]o relation in the low-concentration limit ( Figure 2B , bottom ) . Three different model variants were used: ( 1 ) a model with a constant coupling distance , ( 2 ) a model with a half-normally distributed coupling distance , and ( 3 ) a model with skewed-normally distributed coupling distance . In the two latter cases , the average coupling distance was specified as the expectation value of the distribution . For BAPTA , the Ca2+ binding and unbinding rates were assumed as kon = 4 108 M-1s−1 and koff = 88 s−1 ( affinity 220 nM ) . For EGTA , the rates were taken as kon = 1 107 M-1s−1 and koff = 0 . 7 s−1 ( affinity 70 nM ) . The diffusion coefficients for Ca2+ , EGTA , and BAPTA were assumed to be 220 µm2 s−1 ( Neher , 1998 ) . The endogenous buffer product was set to 5500 s−1 and the resting Ca2+ concentration was assumed as 40 nM ( Collin et al . , 2005 ) . Confidence intervals of coupling distance were obtained by bootstrap procedures . 1000 artificial data sets were generated from the means and SEMs of the original data set and analyzed as the original ( Efron and Tibshirani , 1998 ) . Error estimates were given as half of the 15 . 9–84 . 1 percentile range . All simulations were performed using Mathematica 8 . 0 . 1 running under Windows 7 on a PC . All values are given as mean ± SEM . Error bars in the figures also indicate SEM ( shown only if larger than symbol size ) . Statistical significance was tested using a two-sided Wilcoxon signed rank test for paired data , a two-sided Wilcoxon rank sum test for unpaired data , and a Kruskal–Wallis test for multiple comparisons ( Igor Pro 6 . 3 . 2 ) . Differences with p < 0 . 05 were considered significant . Ca2+ diffusion and binding to mobile and fixed buffers were modeled using the full set of partial differential equations ( PDEs ) of the reaction–diffusion problem , including all necessary boundary and initial conditions ( Smith , 2001; Bucurenciu et al . , 2008; Eggermann and Jonas , 2012; Vyleta and Jonas , 2014 ) . PDEs were solved numerically with NDSolve of Mathematica 8 . 01 running under Windows 7 on a PC . A release unit was implemented as a hemisphere ( Figure 3A ) . The diameter was assumed to be 1 µm , approximately corresponding to the radius of inhibitory boutons , unless specified differently . PDEs were integrated over the radial coordinate ( 2000 grid points ) and solved at a concentration accuracy of 0 . 01 nM . Brief single action potentials were applied as stimuli . A cluster of Ca2+ channels was represented as a point source . A previously published Hodgkin-Huxley-type gating model of P/Q-type Ca2+ channels was used to calculate the Ca2+ inflow ( Borst and Sakmann , 1998 ) . The Ca2+ inflow , relative to that of a single Ca2+ channel , was varied between 1 and 100 . The single-channel conductance was assumed as 2 . 2 pS ( Li et al . , 2007 ) . The coupling distance was varied between 10 and 200 nm . The standard parameters of the model were as follows: For the fixed endogenous Ca2+ buffer , the rates were chosen as kon = 5 108 M−1 s−1 and koff = 1000 s−1 ( affinity 2 µM ) . For standard simulations , both a fixed buffer ( 100 µM ) and a mobile buffer with BAPTA-like properties ( 10 µM ) were incorporated . Ca2+ buffer concentrations were considered to be spatially uniform . The resting Ca2+ concentration was set to 40 nM ( Collin et al . , 2005 ) . Vesicular release rate was computed using a model of transmitter release originally established at the calyx of Held ( Lou et al . , 2005 ) . This model was preferred over alternative models because it is based on the most extensive set of experimental data . The occupancies for the different states of the model were obtained by solving the corresponding first-order ordinary differential equations with a Q-matrix approach . Release rate was computed as the sum of the product of occupancy and release rate for each state; pool depletion was not considered . To account for the rapid timing of transmitter release at BC–PC synapses , the previously used presynaptic action potential ( Meinrenken et al . , 2002; Bucurenciu et al . , 2008 ) was time-compressed by a factor of two , Ca2+ channel gating rates were multiplied by a factor of two , and the binding and unbinding rates of the sensor were increased by a factor of two . The maximal release rate in the model was 6008 s−1 ( Lou et al . , 2005 ) . In a subset of simulations , a distributed arrangement of release sensors was assumed , with a coefficient of variation for the coupling distance of 0 . 3 ( Figure 3E ) . In these cases , release rates were obtained at various distances , and the average release rate was computed as the weighted mean , with weight factors set according to a normal distribution . | The nervous system sends information around the body in the form of electrical signals that travel through cells called neurons . However , these electrical signals cannot cross the synapses between neurons . Instead , the information is carried across the synapse by molecules called neurotransmitters . Calcium ions control the release of neurotransmitters . There is a high concentration of calcium ions outside the neuron but they are not able to pass through the cell membrane under normal conditions . However , when an electrical impulse reaches the synapse , ion channels in the membrane open and allow calcium ions to enter the cell . Once inside , the ions activate the release of neurotransmitters by binding to proteins called release sensors . Several experiments on the release of neurotransmitters have studied the synapses between neurons and muscle fibers . These studies found that the higher the concentration of ions outside the neuron , the higher the rate at which the neurotransmitters were released . However , the timing of release—the length of time over which the neurotransmitters were released—did not depend on the concentration of calcium ions . Arai and Jonas have now studied neurotransmitter release at a synapse in a region of the brain called the cerebellum . These experiments also found that the timing of the release did not depend on the ion concentration , suggesting that this may be a general property of neurotransmitter release . To find out more , Arai and Jonas created a mathematical model of neurotransmitter release . This model suggests that for the timing of release to remain the same , the ion channel and the release sensor must be located close together in the presynaptic terminal . If they are not close together , the timing of release becomes blurred and more dependent on the external calcium concentration . Further experiments confirm the prediction of the model by showing that the calcium channels and the release sensors in these synapses are very close together . The next challenge will be to find out whether the conclusions are also valid for other synapses where the calcium channels and release sensors are further apart . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"neuroscience"
] | 2014 | Nanodomain coupling explains Ca2+ independence of transmitter release time course at a fast central synapse |
Chronic endoplasmic reticulum ( ER ) stress results in toxicity that contributes to multiple human disorders . We report a stress resolution pathway initiated by the nuclear receptor LRH-1 that is independent of known unfolded protein response ( UPR ) pathways . Like mice lacking primary UPR components , hepatic Lrh-1-null mice cannot resolve ER stress , despite a functional UPR . In response to ER stress , LRH-1 induces expression of the kinase Plk3 , which phosphorylates and activates the transcription factor ATF2 . Plk3-null mice also cannot resolve ER stress , and restoring Plk3 expression in Lrh-1-null cells rescues ER stress resolution . Reduced or heightened ATF2 activity also sensitizes or desensitizes cells to ER stress , respectively . LRH-1 agonist treatment increases ER stress resistance and decreases cell death . We conclude that LRH-1 initiates a novel pathway of ER stress resolution that is independent of the UPR , yet equivalently required . Targeting LRH-1 may be beneficial in human disorders associated with chronic ER stress .
It has been estimated that close to one-third of newly synthesized proteins cannot properly fold and are therefore subjected to rapid intracellular degradation ( Schubert et al . , 2000 ) . Both the large scale of this process and the toxicity of misfolded proteins dictate that protein folding must be tightly controlled . When cellular protein folding demands in the cell exceed capacity , misfolded proteins accumulate within the endoplasmic reticulum ( ER ) , causing ER stress . In many human diseases , especially metabolic and neurodegenerative disorders , cellular folding capacities are overwhelmed by increased protein synthesis rates and/or accumulation of abnormal proteins , resulting in ER stress-mediated cell toxicity or death ( Lin et al . , 2008 ) . A set of highly conserved signaling pathways collectively termed the unfolded protein response ( UPR ) functions to resolve ER stress . The UPR consists of three distinct pathways , each of which is initiated by a specific ER-bound protein that acts as a sensor for misfolded proteins . Activation of these sensors elicits downstream signaling pathways that ultimately produce three transcription factors , ATF4 , ATF6 , and XBP-1 , which are active at genes involved in protein folding , expansion of the ER , protein degradation , and , in the case of severe stress , apoptosis . In recent years , many additional components of ER stress resolution have been identified , often with roles in maintenance of metabolic homeostasis following exposure to stress . Despite the increasing number of factors involved in the ER stress response , only a few are directly responsible for ER stress resolution . Treatment of knockout mice for any of the upstream UPR pathway components , Atf6 , Ire1α ( product of Ern1 ) , or Perk ( product of Eifak3 ) , with tunicamycin , a chemical stressor that primarily affects the liver , elicits a striking phenotype . All knockouts exhibit an inability to resolve ER stress upon exposure , as well as a striking metabolic phenotype of strong and persistent accumulation of hepatic triglycerides ( Rutkowski et al . , 2008; Teske et al . , 2011; Zhang et al . , 2011 ) . While tunicamycin is a supraphysiological stressor , these results are important because no other mouse models have been reported to share these phenotypes following ER stress , suggesting that the canonical UPR pathways are functionally overlapping , yet non-redundant , and also that the UPR contains the only required pathways for ER stress resolution . The orphan nuclear hormone receptor liver receptor homolog-1 ( LRH-1; product of Nr5a2 ) is expressed in secretory tissues or tissues with high rates of protein production , including liver , pancreas , intestine , and reproductive tissue ( Higashiyama et al . , 2007 ) . None of the divergent roles of LRH-1—notably bile acid production , development and maintenance of pluripotency , and local steroidogenesis—directly imply a role in ER stress resolution . However , the fact that metabolic diseases such as type II diabetes and fatty liver disease are associated with chronic ER stress ( Ozcan et al . , 2006 ) , but can be alleviated by activation of LRH-1 ( Lee et al . , 2011 ) , suggested a potential connection . In the present study , this connection was substantiated by our discovery that LRH-1 is indispensible for ER stress resolution . In this unexpected role , LRH-1 initiates the kinase-driven activation of a CREB-like transcription factor following exposure to ER stress , specifically through polo-like kinase 3 ( Plk3 ) induction of activating transcription factor 2 ( ATF2 ) phosphorylation . This response is absent in Lrh-1 liver-specific knockout mice , but restoration of Plk3 induction to Lrh-1 null cells reestablishes ER stress resolution . Similarly , expression of a constitutively active Atf2 restores ER stress resolution capacity to Lrh-1 null cells , and reduction of ATF2 activity sensitizes control cells to ER stress . Treatment of hepatocytes with LRH-1 agonists increases the capacity for ER stress resolution and diminishes toxicity resulting from severe ER stress . Thus , we report the existence of an unexpected kinase-driven and drug-targetable pathway responding to ER stress that lies outside of the classical UPR pathways , but is equally required for ER stress resolution .
To test whether LRH-1 is involved in ER stress resolution , we treated Lrh-1 liver-specific knockout ( Lrh-1LKO ) mice and control littermates ( Lrh-1f/f ) with the ER stress inducer tunicamycin ( TM ) , which primarily affects the liver . Lrh-1LKO mice exhibited profound hepatic lipid accumulation by 48 hr following stress , as evidenced by macroscopic evaluation ( Figure 1A ) and measurement of increased hepatic triglycerides and free fatty acids ( Figure 1B ) . We also observed increased TUNEL staining by 72 hr following stress in Lrh-1LKO mice , confirming that the prolonged ER stress had resulted in increased apoptosis ( Figure 1C , D ) . As expected , Lrh-1LKO mice exhibit increased caspase and PARP cleavage following TM ( Figure 1E ) . To confirm that this response was associated with misfolded proteins , we stained primary hepatocytes from TM-treated control and Lrh-1LKO mice with Thioflavin T , which fluoresces when bound to protein aggregates and therefore can be used to quantitate ER stress ( Beriault and Werstuck , 2013 ) . We observed little increase in staining for control cells treated with TM , suggesting that resolvable ER stress induced by TM does not result in significant protein aggregation , but observed strong staining in Lrh-1LKO cells increasing over time treated with TM ( Figure 1F ) . 10 . 7554/eLife . 01694 . 003Figure 1 . Lrh-1 is required for ER stress resolution and for protection against stress-induced lipid accumulation and cell death . ( A ) Macroscopic visualization of steatosis following ER stress in Lrh-1 LKO mice . Mice were i . p . injected with 1 mg/kg tunicamycin ( TM ) or vehicle and livers photographed following sacrifice . Representative of 3–6 mice per group . ( B ) Quantification of hepatic triglycerides and non-esterified free fatty acids of control ( Lrh-1f/f ) and Lrh-1LKO mice ( n = 3–6 ) injected with TM or vehicle . ( C ) Representative hepatic TUNEL staining for apoptosis of control and Lrh-1LKO mice injected with 1 mg/kg TM and sacrificed at 72 hr . Green fluorescence represents TUNEL-positive and blue represents DAPI-positive ( merged image on right ) . Magnification at 40x ( objective ) . Representative of three mice per group . ( D ) Quantification of TUNEL-positive cells for control and Lrh-1LKO mice treated with TM and sacrificed at 72 hr . Ratio of TUNEL-positive to DAPI-positive cells was calculated from three fields of each slide ( n = 3 ) . Significance at p<0 . 05 . ( E ) Immunoblot of cleaved PARP , cleaved caspase 3 , and cleaved caspase 6 in cytoplasmic fractions generated from control and Lrh-1LKO mice ( n = 3–6; pooled ) injected with 1 mg/kg tunicamycin ( TM ) or vehicle . β-actin was used as a loading control . ( F ) Thioflavin T fluorescence of protein aggregates in primary hepatocytes prepared from control and Lrh-1LKO mice , treated with vehicle or 0 . 01 µg/ml TM , and fixed in 4% PFA before staining with 500 µM Thioflavin T . Magnification at 10x ( objective ) . Representative of three mice per group . ( G ) Immunoblot of nuclear spliced XBP-1 , cleaved ATF6 , and ATF4 for control and Lrh-1LKO mice ( n = 3–6; pooled ) injected with 1 mg/kg tunicamycin ( TM ) or vehicle . TBP was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 00310 . 7554/eLife . 01694 . 004Figure 1—figure supplement 1 . Loss of Lrh-1 does not result in loss of UPR target genes in response to stress . Relative expression by quantitative PCR for genes dependent on each of the three UPR pathways . RNA was collected at designated timepoints for control and Lrh-1LKO mice treated with vehicle or 1 mg/kg tunicamycin ( TM ) ( n=3-6 ) . Data was normalized to Tbp expression . DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 004 The hepatic lipid accumulation phenotype observed in Lrh-1LKO mice is identical to that observed in mice lacking any of the three UPR branches following TM treatment , suggesting that there could be a significant deficit in one or more of these pathways in our Lrh-1LKO mice . Thus , we assessed the nuclear accumulation of UPR transcription factors that represent the most downstream effector of each branch . In contrast to our expectation of a UPR deficiency , we observed comparable nuclear accumulation of downstream transcription factors for all three UPR branches in response to ER stress in both control and Lrh-1LKO mice , suggesting that all three branches were functional ( Figure 1G ) . Target genes dependent on each of the three UPR pathways were similarly induced following TM stress in both control and Lrh-1LKO mice , confirming that XBP-1 , ATF6 , and ATF4 were transcriptionally active in addition to being nuclear localized in Lrh-1LKO mice ( Figure 1—figure supplement 1 ) . Importantly , however , we observed sustained signaling of these in Lrh-1LKO mice , indicating that ER stress could not be resolved in Lrh-1LKO mice despite a functional UPR ( Figure 1E ) . In addition to tunicamycin , we utilized two additional chemical ER stress inducers , dithiothreitol ( DTT ) , and Brefeldin A ( BFA ) , which induce ER stress via different mechanisms and kinetics in comparison to TM . Primary hepatocytes from control mice treated with DTT or BFA exhibit a mild or nonexistent UPR response at earlier times , and no UPR signaling at later times ( Figure 2A ) . In contrast , hepatocytes from Lrh-1LKO mice exhibit a more robust initial UPR response as well as sustained signaling at later times , indicating failure to resolve these stresses . Cell death in response to DTT and BFA was significantly higher in Lrh-1LKO cells as compared to controls ( Figure 2B ) . In addition , there is a trend towards increased fat accumulation in Lrh-1LKO cells treated with BFA ( Figure 2C ) , consistent with that observed in Lrh-1LKO mice treated with TM ( Figure 1A , B ) . 10 . 7554/eLife . 01694 . 005Figure 2 . Loss of Lrh-1 sensitizes mice to ER stress resulting from chemical and physiological inducers . ( A ) Immunoblot of nuclear spliced XBP-1 , cleaved ATF6 , and ATF4 for primary hepatocytes from control and Lrh-1LKO mice treated with 2 mM DTT or 0 . 05 µg/ml Brefeldin A ( BFA ) . TBP was used as a loading control . Results representative of three independent experiments . ( B ) Quantification of TUNEL-positive cells for primary hepatocytes from control and Lrh-1LKO mice treated with 2 mM DTT or 0 . 05 µg/ml BFA . Ratio of TUNEL-positive to DAPI-positive cells was calculated from four fields of each slide ( n = 3 ) . Significance at p<0 . 01 . ( C ) Quantification of triglycerides from primary hepatocytes from control and Lrh-1LKO mice ( n = 3 ) treated with vehicle or 0 . 05 µg/ml BFA for 48 hr . TG was normalized to total protein by Bradford assay . ( D ) Partial hepatectomy ( PH ) was used as a non-chemical ER stress inducer . Control and Lrh-1LKO mice underwent surgical removal of 70% of liver weight or sham surgery and were sacrificed 48 hr post surgery . Oil Red O staining ( 200x ) for neutral lipid accumulation on PH or sham surgery liver samples . Results representative of four independent samples . ( E ) Nuclear protein was extracted and immunoblotted for XBP-1s , ATF6 , and ATF4 with TBP as a loading control for control and Lrh-1LKO mice 48 hr after sham or PH surgery . Results representative of four independent samples . DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 005 To confirm that the effect of loss of Lrh-1 in failure to resolve ER stress is not due to indirect effects on drug metabolism of the ER stressors utilized , we also evaluated the effect of a non-chemical inducer of ER stress . Liver regeneration in response to partial hepatectomy induces transient ER stress and the response of UPR pathway-deficient mice to partial hepatectomy is similar to their response to TM treatment , in that liver fat accumulates due to unresolved ER stress ( Zhang et al . , 2011 ) . We performed partial hepatectomy or sham surgery on control and Lrh-1LKO mice and observed dramatic fat accumulation following partial hepatectomy in Lrh-1LKO mice ( Figure 2D ) . In addition , we assessed nuclear UPR transcription factor accumulation as a marker for ER stress . Unlike control mice , Lrh-1LKO mice exhibited accumulation of these factors following partial hepatectomy ( Figure 2E ) , indicating that they were not able to resolve ER stress caused by liver regeneration . Following ER stress , Lrh-1 mRNA expression is moderately induced ( Figure 3A ) , although this does not appear to be reflected in LRH-1 protein levels ( Figure 3B ) . Interestingly , we still observe a TM-dependent induction of LRH-1 target genes Cyp7a1 and Cyp8b1 ( Figure 3C ) , which are involved in bile acid biosynthesis , although no change in total hepatic bile acid in response to stress or genotype ( Figure 3D ) . To determine whether components of the UPR could affect the transcriptional activity of LRH-1 , we used siRNA to knockdown Ire1a , Perk , and Atf6 in primary hepatocytes from control mice . We observed robust silencing of these genes and subsequently treated cells with tunicamycin . The LRH-1 target genes Shp ( product of Nr0b2 ) and Plk3 ( introduced below ) are induced in response to ER stress , but this induction is significantly blunted following knockdown of Ire1a , similar to the loss of Lrh-1 ( Figure 3E ) . Knockdown of either Ire1a or Atf6 also decreased the modest stress-dependent induction of Lrh-1 mRNA expression . However , the loss of Atf6 did not decrease induction of Shp by TM , indicating a specific effect of IRE1a on Lrh-1 transactivation . It is quite unlikely that ER-bound IRE1a directly regulates LRH-1 , as GFP-tagged LRH-1 is exclusively nuclear in the presence or absence of TM ( Figure 3F ) . 10 . 7554/eLife . 01694 . 006Figure 3 . An increase in LRH-1 transcriptional activity and expression is observed following ER stress , with heightened expression dependent on UPR components . ( A ) Relative expression by quantitative PCR for Lrh-1 . RNA was collected at designated timepoints for control and Lrh-1LKO mice treated with vehicle or 1 mg/kg tunicamycin ( TM ) ( n = 3–6 ) . Data were normalized to Tbp expression . ( B ) Western blot analysis of nuclear LRH-1 for control and Lrh-1LKO mice injected with 1 mg/kg TM or vehicle and sacrificed at designated timepoints . TBP was used as a loading control . Results representative of 3–6 individual samples . ( C ) Relative expression by quantitative PCR for Cyp7A1 and Cyp8B1 , which are LRH-1 target genes . RNA was collected at designated timepoints for control and Lrh-1LKO mice treated with vehicle or 1 mg/kg TM ( n = 3–6 ) . Data were normalized to Tbp expression . Significance at p<0 . 05 between genotypes . ( D ) Hepatic bile acid was extracted from tissue from control and Lrh-1LKO mice treated with vehicle or 1 mg/kg TM ( n = 3–6 ) and sacrificed at designated timepoints . ( E ) Primary hepatocytes from control and Lrh-1LKO mice were transfected with siRNA against Ire1a , Perk , or Atf6 , or a nonsilencing siRNA . Percent knockdown was quantified by qPCR for Ire1a , Perk , and Atf6 in control cell samples ( n = 3–4 ) 52 hr after transfection . Relative expression for Shp , Plk3 , and Lrh-1 was by qPCR for primary hepatocytes from control and Lrh-1LKO mice treated with vehicle or 5 ng/ml TM ( n = 3–4 ) . TBP was used as a loading control . Significance at p<0 . 05 as compared with control siRNA samples from control mice . ( F ) N-terminal-tagged GFP-LRH-1 fluorescence ( green ) in TLR-3 cells transfected with GFP-LRH-1 and treated with vehicle or 1 µg/ml TM for timepoints indicated . DNA was stained with DAPI ( blue ) . Magnification at 100x ( objective ) and images cropped . Results representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 006 The striking phenotype of the Lrh-1LKO mice in the absence of any apparent defect in the three canonical UPR pathways predicted that LRH-1 could initiate a novel pathway of ER stress resolution . We performed microarray analysis on control and Lrh-1LKO mice treated with vehicle or TM to identify LRH-1-dependent TM responsive genes , but initial analysis of the results did not identify obvious candidates that could account for the Lrh-1LKO phenotype . To identify potentially relevant signaling pathways , we used the Molecular Signatures Database ( MSigDB ) program to identify transcription factor binding motifs that were significantly overrepresented in promoters of the top 100 genes differentially responsive to TM between genotypes . The predominant output of this analysis was the cyclic AMP response element ( CRE ) ( Figure 4A ) . Among the multiple transcription factors that can bind this motif , activating transcription factor 2 ( ATF2 ) was a promising candidate for further investigation based on its known roles in responding to genotoxic and oxidative stresses ( van Dam et al . , 1995; Kurata , 2000 ) . Atf2 levels were slightly lower in Lrh-1LKO mice compared to controls during basal conditions , but there were no significant differences in levels of Atf2 between genotypes following treatment with TM ( Figure 4B ) . Because ATF2 requires phosphorylation to be transcriptionally active ( Livingstone et al . , 1995 ) , we assessed its phosphorylation status . In TM-treated primary hepatocytes from control and Lrh-1LKO mice , we observed a rapid induction of ATF2 phosphorylation at the residues ( T51/53 , corresponding to human T69/71 ) required for ATF2 transcriptional activity in control mice ( Figure 4C ) , and this response was absent in Lrh-1LKO mice . 10 . 7554/eLife . 01694 . 007Figure 4 . Microarray analysis suggests loss of ATF2 transcriptional ability in Lrh-1 deficient mice . ( A ) Illumina Mouse Ref-8 arrays were performed for control and Lrh-1LKO mice treated with vehicle or tunicamycin ( TM ) for 24 hr ( n = 3 ) . Data were normalized and top 100 genes differentially induced by TM as assessed by fold change was analyzed by the Molecular Signatures Database software to identify overrepresented binding motifs . Transcription factors with binding sites significantly enriched in input gene promoters shown in chart . Motifs shown with significance at p<0 . 05 . ( B ) Relative expression for Atf2 . RNA was collected at designated timepoints for control and Lrh-1LKO mice treated with vehicle or 1 mg/kg TM ( n = 3–6 ) . Data were normalized to Tbp expression . Significance at p<0 . 01 . ( C ) Immunoblot for total ATF2 and phospho-ATF2 ( mT51/53; hT69/71 ) , with pATF2 representing sites required to be phosphorylated for ATF2 activity . Primary hepatocytes were isolated from control and Lrh-1LKO mice and treated with 5 µg/ml TM or vehicle and protein collected at designated timepoints . Results representative of three independent experiments . ( D ) Relative expression for B3gat3 , Creld1 , and Mcfd2 . RNA was collected at designated timepoints for control and Lrh-1LKO mice treated with vehicle or 1 mg/kg TM ( n = 3–6 ) . Data were normalized to Tbp expression . ( E ) Chromatin immunoprecipitation was performed from livers of control and Lrh-1LKO mice treated with vehicle or TM ( 1 mg/kg ) for 6 hr ( n = 4 ) . Immunoprecipation was done with an anti-ATF2 , pATF2 ( 69/71 ) , or acetyl histone H4 antibody or rabbit IgG as a control . qPCR was used to determine binding by use of primers flanking binding site previously identified by ENCODE Projects . DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 007 To further link ATF2 and the Lrh-1-dependent ER stress response , we screened for genes that were induced at least 1 . 5-fold by TM in control mice with significantly different induction in Lrh-1LKO mice , and secondarily screened for those with significantly different expression levels following TM treatment between genotypes . From this filtering we obtained a list of 65 genes ( Table 1 ) and further investigated their capacity to be bound and regulated by ATF2 . Genome-wide ATF2 binding data from the ENCODE Project ( ENCODE Project Consortium , 2011 ) was used to identify genes with ATF2 binding sites −500 bp to +275 bp from the transcriptional start site ( TSS ) . The same analysis was performed with genome-wide LRH-1 binding data ( Chong et al . , 2012 ) . Our gene list was compared with known ATF2 and LRH-1 binding sites , and genes bound by ATF2 or LRH-1 −500 bp to +275 bp from the TSS were determined ( Table 1 ) . Hypergeometric distribution tests determined our list was highly enriched for ATF2-bound genes ( p=3 . 83e−08 ) and enriched , although less strikingly so , for LRH-1-bound genes ( p=1 . 40e−04 ) . 10 . 7554/eLife . 01694 . 008Table 1 . Genes differentially induced by TM between control and Lrh-1-null mice and their regulation by ATF2 and LRH-1DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 008GeneKnown target ofATF2LRH-1PLK3✓✓GDF15FGF21*PMM1CRELD2✓NFIL3IGSF11GSTM2GOT1DPP9✓✓ST5CCBL1✓✓TMEM66✓KRTCAP2✓✓LRRC59✓✓CDK2AP2✓✓ST3GAL1TES✓CCDC134✓✓UGT1LACOT2✓PLIN5✓LITAF✓RPS13✓LGALS3BPCRYMSUPT5H✓B3GAT3✓GRN✓ARRDC4SLC30A7✓CRELD1✓NT5MARSGSEP15MKNK1✓DDX52EXOSC5✓RABAC1✓RPS15✓✓D830014E11RIKAVPI1✓RPS9✓CRAT✓BHMT2✓B230217C12RIKEIF3G✓✓SLC25A28✓HRCCL9ANXA4✓✓SMCO4SMOXARL14EP✓SLC39A7✓ICA1ENTPD5✓PIWIL2ANG✓MCFD2✓✓SOAT2SLC41A3✓✓MFGE8CYP4A14D12ERTD647EMicroarray analysis was performed for control and Lrh-1LKO mice treated with vehicle or 1 mg/kg tunicamycin for 24 hr ( n = 3 ) . Genes were screened for those induced at least 1 . 5 fold by TM in control mice with significantly different induction in Lrh-1LKO mice by t-test ( p<0 . 05 ) . This list was filtered for those with differential expression between genotypes with TM treatment by t-test ( p<0 . 05 ) . Previously published genome-wide ATF2 and LRH-1 binding datasets were analyzed to identify genes with ATF2 or LRH-1 binding sites −500 to +275 bp from the TSS . Genes in our set were compared with these sets and genes that contain ATF2 or LRH-1 binding sites meeting the above criteria are marked . *No binding site for ENCODE data set but a known ATF2 direct target ( Hondares et al . , 2011 ) . We chose three direct ATF2 target genes—B3gat3 , likely involved in proteoglycan synthesis ( Baasanjav et al . , 2011 ) , Creld1 , an ER stress-inducible protein of unknown function ( Hansen et al . , 2012 ) , and Mcfd2 , a receptor for ER–Golgi transport ( Zhang et al . , 2003 ) —for further studies . These genes were induced by TM in control mice but showed blunted induction in Lrh-1LKO mice ( Figure 4D ) . To determine whether ATF2 is bound , phosphorylated , and enzymatically active on these genes , we performed chromatin immunoprecipitation ( ChIP ) for ATF2 , pATF2 ( m51/53; h69/71 ) , and acetylated histone H4 ( AcH4 ) , a marker for activated ATF2 histone acetyltransferase activity ( Bruhat et al . , 2007 ) . ChIP was performed on chromatin from control and Lrh-1LKO mice treated with vehicle or TM . While there was little difference in ATF2 binding between treatments or genotypes for B3gat3 , Creld1 , and Mcfd2 , binding of pATF2 and AcH4 was increased following TM treatment in control mice alone for these three genes ( Figure 4E ) . Overall , we conclude that a significant number of genes differentially regulated by ER stress in control and Lrh-1LKO mice are ATF2 targets dependent on activation of ATF2 by phosphorylation . We did not observe significantly reduced activation of kinases known to phosphorylate ATF2 , particularly the MAP kinases JNK and p38 ( Ouwens et al . , 2002 ) , in Lrh-1LKO mice in response to ER stress . However , polo-like kinase 3 ( Plk3 ) , which we found to be differentially induced by TM between genotypes ( Table 1 ) , has also been reported to be capable of phosphorylating ATF2 ( Wang et al . , 2011b ) , although not in the context of ER stress . In line with our microarray findings , Plk3 mRNA expression was robustly induced in control mice following ER stress , and this induction was significantly diminished in Lrh-1LKO mice ( Figure 5A ) . This ER stress-induced increase in control mice alone was also observed at the level of PLK3 protein ( Figure 5B ) . A time course of Plk3 induction performed in primary hepatocytes from control and Lrh-1LKO mice ( Figure 5C ) indicates that Plk3 induction occurs in a similar time frame to ATF2 phosphorylation , suggesting that this induction could be sufficient for the stress-inducible ATF2 phosphorylation observed ( Figure 4C ) . We predicted that Plk3 was a direct transcriptional target of LRH-1 and identified a strong LRH-1 binding site 285 base pairs upstream of the Plk3 TSS . Chromatin immunoprecipitation of LRH-1 identified an enrichment of LRH-1 occupancy at this site shortly after TM treatment in control mice ( Figure 5D ) . 10 . 7554/eLife . 01694 . 009Figure 5 . Plk3 is a direct LRH-1 target that is required for induction of ATF2 target genes and ER stress resolution . ( A ) Quantitative PCR for Plk3 . RNA was collected at designated timepoints for control and Lrh-1LKO mice treated with vehicle or 1 mg/kg tunicamycin ( TM ) ( n = 3–6 ) . Data were normalized to Tbp expression . Significance at p<0 . 01 . ( B ) Immunoblot for PLK3 protein in control and Lrh-1LKO mice treated with vehicle or 1 mg/kg TM for designated timepoints . Liver tissue was boiled in Laemmli buffer to extract insoluble protein and samples were pooled ( n = 2 ) prior to gel loading . β-actin was used as a loading control . ( C ) Quantitative PCR for Plk3 . RNA was collected at designated timepoints for primary hepatocytes isolated from control and Lrh-1LKO mice treated with vehicle or 0 . 5 µg/ml TM ( n = 3 ) . Data were normalized to Tbp expression . Significance at p<0 . 01 . ( D ) A LRH-1 binding site in the Plk3 promoter was identified 285 bases upstream of the TSS . Chromatin immunoprecipitation was performed from livers of control mice treated with vehicle or TM ( 1 mg/kg ) for 6 hr ( n = 3 ) . Immunoprecipation was done with an anti-LRH-1 antibody or mouse IgG as a control . qPCR was used to determine binding by use of primers flanking binding site . Significance at p<0 . 05 between antibodies with error bars representing SEM . ( E ) TLR-3 cells were transfected with nonsilencing siRNA or siRNA against Lrh-1 and nuclear protein was prepared at designated timepoints . Samples were immunoblotted for total ATF2 and phospho-ATF2 ( mT51/53; hT69/T71 ) following treatment with vehicle or 1 ug/ml TM . Results representative of three independent experiments . ( F ) Relative luciferase activity for TLR-3 cells transfected with a cAMP response element ( CRE ) -luciferase reporter , Atf2 , and Lrh-1 . 48 hr after transfection , cells were treated with vehicle or 5 µg/ml TM and the following inhibitors: 10 µM D-JNKi for JNKs , 1 µM SB202190 for p38 , 10 µM GW84362X for PLK1/PKL3 , or 1 µM GSK650394A for SGK . 24 hr after treatment , cells were lysed , and luciferase activity was measured and normalized to β-galactosidase activity . Significance at p<0 . 01 as compared with TM treated cells ( n = 3 ) . Results representative of three independent experiments . ( G ) Atf3 expression by qPCR in TLR-3 cells transfected with control or siRNA targeting Lrh-1 ( knockdown efficiency same as 5E ) , along with overexpression of constitutively active Atf2 ( C2/Atf2 ) , Plk3 , or an empty vector . 48 hr post transfection , cells were treated with 1 µg/ml TM for 24 hr . Data were normalized to Tbp expression . Significance at p<0 . 01 for TM treated vs vehicle treated samples ( n = 3 ) . Results representative of three independent experiments . ( H ) Atf3 expression by qPCR from primary hepatocytes from control and Lrh-1LKO mice treated with vehicle or 5 µg/ml TM and 10 µM PLK3 inhibitor GW843682X for 24 hr . Data were normalized to Tbp expression . Significance at p<0 . 01 between genotypes ( n = cells from 2–3 mice/group ) . ( I ) Wild-type mice were i . p . injected with PLK3 inhibitor GW843682X ( 1 mg/kg BW ) or vehicle ( DMSO ) . Mice were also i . p . injected with TM ( 1 mg/kg BW ) or vehicle ( DMSO ) . 48-hr post injection , liver tissue was collected and nuclear protein was isolated to assess accumulation of UPR transcription factors spliced XBP-1 , cleaved ATF6 , and ATF4 . TBP was used as a loading control . Results representative of results from three mice . ( J ) Wild-type and Plk3−/− mice were treated with TM ( 0 . 5 mg/kg ) or vehicle . Nuclear UPR proteins were assessed by immunoblot at designated timepoints . TBP was used as a loading control . Results representative of four independent samples . DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 009 To determine the requirement for PLK3 in induction of CRE-containing ATF2 target genes , we utilized the cell line TLR-3 , which was reported to have stress-inducible ATF2 phosphorylation in response to environmental toxins ( Muguruma et al . , 2008 ) . We observed that ATF2 was phosphorylated in response to TM dependent on Lrh-1 expression ( Figure 5E ) . Luciferase reporter assays utilizing ATF2 and a cAMP response element reporter demonstrated an increase in reporter activity under ER stress conditions , suggesting increased ATF2 phosphorylation ( Figure 5F ) . Inhibition of JNK , p38 , and SGK did not significantly reduce reporter activity whereas PLK3 inhibition did , arguing that PLK3 is the predominant driver of ATF2 phosphorylation following ER stress . For endogenous gene expression , we quantified ATF2 target gene Atf3 ( Mayer et al . , 2008 ) induction following TM treatment , and observed that it was significantly blunted with Lrh-1 knockdown . However , overexpression of a constitutively active Atf2 ( C2/Atf2; Steinmuller and Thiel , 2003 ) or Plk3 restored stress-inducible Atf3 induction ( Figure 5G ) . We then assessed Atf3 induction in primary hepatocytes from control and Lrh-1LKO mice treated with TM or GW843682X , a PLK3 inhibitor . We observed that hepatocytes from Lrh-1LKO mice had blunted Atf3 induction in response to stress , and this was not affected by treatment with GW843682X ( Figure 5H ) . However , in control hepatocytes , the robust stress-dependent induction of Atf3 was blunted to levels comparable to Lrh-1LKO mice when PLK3 was inhibited . These results predict that PLK3 is critical for ER stress resolution . To address this , we first treated control mice with tunicamycin and GW843682X . PLK3 inhibition resulted in decreased ability to resolve ER stress , as evidenced by sustained UPR signaling at 48 hr ( Figure 5I ) , reminiscent of unresolved ER stress observed in Lrh-1LKO mice ( Figure 1G ) . We then obtained control and Plk3−/− mice ( Myer et al . , 2011 ) and treated them with TM . UPR transcription factors were comparably induced observed between genotypes at 24 hr , indicating that the UPR branches are functional in Plk3−/− mice . As expected , wild-type mice had resolved ER stress by 72 hr , as evidenced by reduction of UPR transcription factor accumulation , but the sustained UPR signaling in Plk3−/− mice indicated that they were unable to resolve ER stress ( Figure 5J ) , as also observed with chemical PLK3 inhibition . Thus , we conclude that loss of Plk3 confers ER stress sensitivity similar to that observed in Lrh-1LKO mice . As loss of Plk3 phenocopied the loss of Lrh-1 , we predicted that restoration of stress-inducible Plk3 induction in Lrh-1LKO mice should rescue ATF2 phosphorylation and resolve ER stress . We generated a tetracycline-inducible adenovirus overexpressing mouse Plk3 ( Ad-Plk3 ) , with LacZ used as a control ( Ad-control ) . Primary hepatocytes from control and Lrh-1LKO mice were transduced and later treated with vehicle or TM , as well as doxycycline to induce expression of Plk3 or LacZ , and nuclear UPR accumulation of spliced XBP-1 , ATF4 , and cleaved ATF6 was assessed . Following TM treatment , control cells transduced with Ad-Plk3 or Ad-control showed comparable UPR responses at 24 hr , with resolution of ER stress by 48 hr . Treatment of Lrh-1LKO cells with Ad-control resulted in no improvement in ER stress resolution capacity , as demonstrated by failure to resolve ER stress by 48 hr , similar to what was observed in Lrh-1LKO mice treated with TM . In contrast , transduction of Lrh-1LKO cells with Ad-Plk3 restored their ability to resolve ER stress , as evidenced by the absence of nuclear UPR transcription factor accumulation at 48 hr ( Figure 6A ) . Thus , restoration of Plk3 induction in Lrh-1LKO cells is sufficient for the restoration of ER stress resolution . 10 . 7554/eLife . 01694 . 010Figure 6 . Restoration of Plk3 induction rescues ATF2 phosphorylation and ER stress resolution in Lrh-1LKO mice , and loss or gain of ATF2 transcriptional activity also alters ER stress resolution capacity . ( A ) Primary hepatocytes were prepared from Lrh-1f/f and Lrh-1LKO mice and transduced with Ad-Plk3 or Ad-control at a MOI of 100 . Cells were treated 36 hr later with vehicle or tunicamycin ( TM ) ( 0 . 01 µg/ml ) and doxycycline ( 1 µg/ml ) to induce Plk3 or LacZ control . Nuclear protein was obtained at timepoints indicated and immunoblotted for UPR transcription factors , with TBP as a loading control . Results are representative of three independent experiments . ( B ) Primary hepatocytes were prepared from Lrh-1LKO mice and transduced with Ad-Plk3 or Ad-control at a MOI of 100 . Cells were treated 36 hr later with vehicle or TM ( 1 µg/ml ) and doxycycline ( 1 µg/ml ) to induce Plk3 or LacZ control . Nuclear protein was collected at indicated timepoints and immunoblotted for pATF2 ( 69/71 ) and total ATF2 . Samples are from same gel and brightness/contrast was adjusted equally prior to cropping . Results are representative of three independent experiments . ( C ) Primary hepatocytes were prepared from Lrh-1f/f and Lrh-1LKO mice and transduced with Ad-Plk3 or Ad-control at a MOI of 100 . Cells were treated 36 hr later with vehicle or TM ( 0 . 01 µg/ml ) and doxycycline ( 1 µg/ml ) to induce Plk3 or LacZ control . Cells were fixed at various timepoints and stained for lipids ( red ) using Lipidtox dye and counterstained with DAPI . Magnification is 40X ( objective ) . Results are representative of three independent experiments . ( D ) Relative expression of Plk3 by qPCR . Primary hepatocytes from control and Lrh-1LKO mice were transduced with Ad-Plk3 or Ad-control at a MOI of 100 and treated with 1 µg/ml doxycycline and vehicle or 0 . 01 µg/ml TM . Data were normalized to Tbp expression . Significance at p<0 . 01 between genotypes ( n = cells from 3–4 mice ) . ( E ) Primary hepatocytes from control mice were transduced with Ad-DN Atf2 ( Atf2 T51A/T53A to serve as a dominant negative ) or Ad-control at a MOI of 100 and treated with 5 ng/ml TM . Primary hepatocytes from control and Lrh-1LKO mice were transduced with Ad-C2/Atf2 ( expressing a constitutively active Atf2 ) or Ad-control at a MOI of 100 and treated with 5 ng/ml TM . Nuclear protein was obtained at timepoints indicated and immunoblotted for UPR transcription factors , with TBP as a loading control . Results are representative of samples from three mice . ( F ) Atf2 expression by qPCR to quantify viral Atf2 overexpression . Primers were chosen to amplify regions identical between wildtype Atf2 , C2/Atf2 , and DN Atf2 . Primary hepatocytes from control mice ( n = 3–4 ) were transduced at a MOI of 100 and treated with vehicle for 24 hr prior to RNA collection . ( G ) Spliced Xbp-1 expression by qPCR for primary hepatocytes from control and Lrh-1LKO mice ( n = 3–4 ) transduced with Ad-control , Ad-DN Atf2 , or Ad-C2/Atf2 and treated with vehicle or 5 ng/µl TM . Data were normalized to Tbp expression . DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 010 To confirm that the restoration of Plk3 induction in primary liver cells from Lrh-1LKO mice could restore stress-inducible ATF2 phosphorylation , primary hepatocytes from Lrh-1LKO mice were transduced and treated with TM as described above . In accordance with what we observed in non-transduced hepatocytes from Lrh-1LKO mice , Lrh-1LKO cells transduced with Ad-control were unable to phosphorylate ATF2 in response to stress ( Figure 6B ) . However , when transduced with Ad-Plk3 , cells from Lrh-1LKO mice regained the ability to phosphorylate ATF2 in a stress-inducible manner , consistent with what we had previously observed in control mice ( Figure 4C ) . Lastly , we asked whether the striking phenotype of liver lipid accumulation observed in Lrh-1LKO mice following ER stress could be resolved by restoration of Plk3 induction . Primary hepatocytes from control and Lrh-1LKO mice were isolated , transduced with adenovirus , and treated with TM as described above . Cells were stained with Lipidtox to visualize neutral lipids ( red ) , and counterstained with DAPI . No differences in fat accumulation between genotypes or transduction with Ad-Plk3 or Ad-control were observed when cells were treated with vehicle or TM for 24 hr ( Figure 6C ) . However , by 48 hr post stress , lipid staining was reduced in control cells , irrespective of transduction with either Ad-Plk3 or Ad-control . In Lrh-1LKO cells transduced with Ad-control , lipid accumulation was sustained at later timepoints , reminiscent of what we had observed in Lrh-1LKO mice treated with TM ( Figure 1A ) . In contrast , Lrh-1LKO cells transduced with Ad-Plk3 exhibited lipid levels comparable to those in control cells . Since the overexpression of Plk3 is quite minimal in our system ( Figure 6D ) , we expect that conclusions drawn from these experiments are physiologically relevant . Overall , these studies demonstrate that Plk3 expression is sufficient to rescue the defects in the Lrh-1LKO mice of ATF2 target gene induction and ER stress resolution in response to TM , as well as protect against metabolic derangement following stress . Our bioinformatics analyses indicated that ATF2 is the predominant downstream effector of PLK3 . However , PLK3 is known to phosphorylate other transcription factors , including c-Jun and p53 ( Table 2 ) . Further analysis of our gene set ( Table 1 ) for overlap with known binding sites for these transcription factors in promoters of our genes indicated that , while c-Jun and p53 sites are enriched , ATF2 remains the most significant factor . To investigate the functional significance of ATF2 in ER stress resolution , we generated adenoviruses expressing mouse Atf2 with mutations in essential phosphorylation sites ( T51/53 to A51/53 ) , similar to constructs previously reported ( Hayakawa et al . , 2003 ) , to serve as a dominant negative ( Ad-DN Atf2 ) . We also generated adenovirus expressing a constitutively active Atf2 ( Steinmuller and Thiel , 2003 ) ( Ad-C2/Atf2 ) or LacZ ( Ad-control ) . Primary hepatocytes from control mice were transduced with Ad-control or Ad-DN Atf2 and treated with tunicamycin . We observed resolution of TM-induced ER stress by 48 hr when transduced with Ad-control , but sustained UPR activation when transduced with Ad-DN Atf2 ( Figure 6E ) , even at moderate levels of viral overexpression ( Figure 6F ) . We also prepared primary hepatocytes from control and Lrh-1LKO mice and transduced them with Ad-control or Ad-C2/Atf2 . When transduced with Ad-control , control hepatocytes are able to resolve ER stress , whereas hepatocytes deficient in Lrh-1 cannot ( Figure 6E ) . However , transduction with Ad-C2/Atf2 increases the stress resolution capacity of Lrh-1LKO cells . Taken together , along with spliced Xbp-1 expression quantifying these experiments ( Figure 6G ) , we conclude that the loss of ATF2 activity sensitizes cells to ER stress , while overexpression of active ATF2 facilitates ER stress resolution . 10 . 7554/eLife . 01694 . 011Table 2 . Enrichment of known transcription factor binding sites in our gene set ( Table 1 ) of differentially regulated genes by ER stress between control and Lrh-1LKO miceDOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 011Transcription factor name:Known phosphorylation by PLK3:Overlap of target genes with our gene set ( Table 1 ) :Dataset used for analysis:ATF2T71 ( Wang et al . , 2011b ) 3 . 83E-08ATF2 binding in G12878 cells ( ENCODE EH002306 ) Lrh-1none1 . 40E-04Lrh-1 binding in mouse liver ( Chong et al . , 2012 ) p53S20 ( Xie et al . , 2001 ) 1 . 37E-04p53 binding in U2OS cells treated with Nutlin-3 ( Menendez et al . , 2013 ) , which results in S20 phosphorylation ( Valentine et al . , 2011 ) c-JunS63 and S73 ( Wang et al . , 2011a ) 2 . 95E-02c-Jun binding in CH12 cells ( ENCODE EM001943 ) , in which S63 may be constitutively phosphorylated like other B-lymphoma lines ( Gururajan et al . , 2005 ) NRF2NoneNSNRF2 binding in lymphoid cell lines treated with sulforaphane ( Chorley et al . , 2012 ) Transcription factors known to be phosphorylated by PLK3 ( ATF2 , p53 , and c-Jun ) are summarized here , along with NRF2 to represent the oxidative stress response and LRH-1 . Overlap of known transcription factor binding sites −500 to +b250 bp of the TSS for genes in Table 1 was calculated , and significance was determined using hypergeometric distribution tests . Based on the increased expression of Plk3 and ATF2 target genes by tunicamycin with LRH-1 agonist treatment compared to TM alone ( Table 1 ) , we hypothesized that LRH-1 activation would promote increased ER stress resolution and cell survival . We treated primary hepatocytes from mice transgenic for human Lrh-1 ( hLrh-1 TG ) and null for mouse Lrh-1 ( mLrh-1 LKO ) with the non-lipid LRH-1 agonist RJW100 ( Whitby et al . , 2011 ) . As predicted ( Table 1 ) , co-treatment with RJW100 and TM increased expression of Plk3 above that by TM alone ( Figure 7A ) , and this was dependent on hLrh-1 expression . The ATF2 target genes Mcfd2 and Atf3 could also be induced by RJW100 alone , which was also dependent on hLrh-1 ( Figure 7B ) . We observed no effect of RJW100 on induction of target genes dependent on any UPR transcription factor ( Figure 7C ) , leading us to conclude that the effect of RJW100 is specific to the ER stress responsive pathway initiated by LRH-1 . We then treated hLrh-1 TG; mLrh-1 LKO mouse primary hepatocytes with a low or high dose of TM , along with RJW100 . For both low and high doses of TM , co-treatment with RJW100 increased ability of cells to resolve ER stress , as evidenced by less sustained UPR signaling by 48 hr post treatment ( Figure 7D ) . To evaluate cell survival , we treated primary hepatocytes from hLrh-1 TG; mLrh-1 LKO and mLrh-1 LKO mice with a range of TM doses , along with RJW100 . RJW100 decreased cell death in hLrh-1 TG; mLrh-1 LKO cells back to untreated levels for all but the highest TM dose , with significant reduction even at this dose ( Figure 7E ) . Mice lacking mLrh-1 were more sensitive to TM at all doses and RJW100 had no beneficial effects on cell survival ( Figure 7E ) . Taken together , our results indicate that hLrh-1 is able to compensate for loss of mLrh-1 in ER stress resolution , and that LRH-1 agonism potently increases the ability of cells to resolve high levels of ER stress . 10 . 7554/eLife . 01694 . 012Figure 7 . LRH-1 agonism promotes Plk3 and ATF2 target gene expression and increases resistance to ER stress independent of the UPR . ( A ) Quantitative PCR for Plk3 . RNA was collected at designated timepoints for primary hepatocytes isolated from hLrh-1 TG; mLrh-1 LKO and mLrh-1LKO mice treated with 10 uM RJW100 and/or 0 . 01 ug/ml TM ( n = cells from 3 mice ) . Data were normalized to Tbp expression . Significance at p<0 . 05 as compared with vehicle treatment . ( B ) Quantitative PCR for Mcfd2 and Atf3 . RNA was collected at designated timepoints for primary hepatocytes isolated from hLrh-1 TG; mLrh-1 LKO and mLrh-1LKO mice treated with 10 µM RJW100 ( n = cells from 3 mice ) . Data were normalized to Tbp expression . Significance at p<0 . 05 as compared with vehicle treatment . ( C ) Primary hepatocytes were isolated from hLrh-1 TG; mLrh-1 LKO mice ( n = cells from 3 mice ) and treated with 10 µM RJW100 and/or 0 . 01 µg/ml TM . RNA was collected at designated timepoints and qPCR performed for UPR target genes . Data were normalized to Tbp expression . No significance between groups . ( D ) Primary hepatocytes were isolated from hLrh-1 TG; mLrh-1 LKO mice and treated with 10 µM RJW100 and/or 0 . 01 µg/ml or 0 . 1 µg/ml TM . Nuclear protein was obtained at timepoints indicated and immunoblotted for UPR transcription factors , with TBP as a loading control . Results are representative of three independent experiments . ( E ) Primary hepatocytes were isolated from hLrh-1 TG; mLrh-1 LKO and mLrh-1LKO mice ( n = cells from three mice ) and treated with 10 µM RJW100 and/or 0 . 01–1 µg/ml TM . Cells were fixed 48 hr post treatment and TUNEL staining was performed and quantified . Ratio of TUNEL-positive to DAPI-positive cells was calculated from four fields of each slide ( n = 3 ) . Significance at p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 012
Our studies have identified an unexpected yet essential pathway for hepatic ER stress resolution initiated by the nuclear receptor LRH-1 ( Figure 8 ) . Following ER stress , LRH-1 is recruited to the Plk3 promoter and dramatically induces transcription of this atypical kinase . This is essential for ATF2 phosphorylation , which is profoundly deficient in Lrh-1LKO hepatocytes in response to stress . PLK3 is required for ATF2 activation following ER stress , and that Plk3−/− mice , like Lrh-1LKO mice , are defective in ER stress resolution . The similar impact of loss of Lrh-1 and loss of Plk3 predicted that restoration of Plk3 induction to Lrh-1LKO mice could be sufficient to rescue their ability to resolve ER stress . We found that restoring Plk3 induction to primary Lrh-1LKO hepatocytes did restore their ability to phosphorylate ATF2 and also resulted in decreased fat accumulation at later times after ER stress . More importantly , the Lrh-1LKO hepatocytes resolved ER stress similar to wild-type cells when Plk3 induction was reinstated . 10 . 7554/eLife . 01694 . 013Figure 8 . Mechanism of LRH-1’s requirement in ER stress resolution . DOI: http://dx . doi . org/10 . 7554/eLife . 01694 . 013 This led us to consider whether activation of ATF2 is also required for ER stress resolution . Since functional redundancy with the transcription factors Atf7 and Cre-bpa ( Creb5 ) complicates study of ATF2 function in mice ( Breitwieser et al . , 2007 ) , we overexpressed a dominant negative ATF2 in hepatocytes . Transduction of wild-type hepatocytes with this construct diminishes ability to resolve ER stress , consistent with that observed following the loss of Lrh-1 or Plk3 . We also found that overexpression of a constitutively active ATF2 resulted in improved ability to resolve ER stress . These results indicate that ATF2 activation is sufficient for ER stress resolution , and that activation of ATF2 or its heterodimeric partners , including ATF7 and/or CRE-BPa , is necessary . Determining genome-wide binding of ATF2 and other PLK3-responsive transcription factors in the context of ER stress should enhance our understanding of the downstream targets of this essential pathway . The induction of LRH-1 target genes following ER stress and its increased recruitment to the Plk3 promoter suggests that LRH-1 is transcriptionally activated in response to TM treatment . In general , this could result from either signaling pathways that phosphorylate , or otherwise activate LRH-1 or its coregulators , or to increased production of an endogenous LRH-1 agonist . It could also result from increased expression of Lrh-1 following ER stress . Unfortunately , both the very limited information on specific signaling pathways that promote LRH-1 transactivation and the absence of information on endogenous LRH-1 ligands severely limits approaches to determine whether LRH-1 is post-translationally regulated following ER stress . However , we observed that the stress-dependent induction of Lrh-1 expression is dependent on the UPR components IRE1a and ATF6 , and IRE1a also appears to impact LRH-1 transactivation . It is possible that transcriptional regulation of LRH-1 following ER stress is at least partially under control of the UPR in this regard . However , it is quite unlikely that IRE1a directly regulates LRH-1 protein , based on the constitutive nuclear localization of LRH-1 following ER stress . Future studies will be needed to characterize post-translational control of LRH-1 in a variety of contexts , including ER stress . More broadly , the ligand responsiveness of LRH-1 suggests that it could become a therapeutic target in human disorders associated with chronic ER stress . Our results suggest that activation of LRH-1 by RJW100 heightens induction of Plk3 and ATF2 target genes . Additionally , treatment of cells transgenic for human Lrh-1 with RJW100 results in the ability to resolve high levels of ER stress and protection from cell death following stress . Therefore , targeting LRH-1 in vivo with agonists or newly developed antagonists ( Rey et al . , 2012 ) may allow selective augmentation or inhibition of ER stress resolution , which could prove beneficial in numerous human disorders . This is of particular interest since other strategies for targeting the UPR have proven to be difficult , and no chemical compounds have been successfully used in mice that directly activate downstream UPR components . Although it has not been previously linked to the UPR , ATF2 has been associated with other cellular stress responses , including those that mitigate DNA damage ( van Dam et al . , 1995 ) and oxidative stress ( Kurata , 2000 ) , with these roles dependent on ATF2 phosphorylation . This raises the interesting question of whether this LRH-1 initiated pathway contributes to a broader , kinase-driven cell defense system that can be activated by diverse stresses . Highly conserved , kinase-driven pathways are increasingly being recognized for not only being activated by ER stress , but also for facilitating ER stress resolution . In mammalian cells , specific MAP kinases have been identified that are activated by ER stress and diminish cell death following stress ( Hu et al . , 2004 ) . A screen for genes whose loss confers sensitivity to ER stress in S . cerevisae also uncovered twelve MAP kinase components , remarkable in scale considering that the yeast UPR only contains a single branch consisting of IRE1 and substrate HAC1 ( Chen et al . , 2005 ) . We provide an additional example of such a kinase-driven ER stress resolution pathway , in which LRH-1 protects the cell through Plk3 induction and subsequence ATF2 phosphorylation . While PLK3 is not a MAP kinase , it is important to note that ATF2 is otherwise phosphorylated by MAP kinases p38 and JNK . This suggests that the presence of LRH-1 adds an additional layer of control onto these MAP kinase pathways through activation of the atypical kinase . We are currently investigating the conservation of this response utilizing simpler model organisms , as well as investigating whether LRH-1 and downstream factors PLK3 and ATF2 are required for resolution of broader cell stresses . Overall , we conclude that this novel LRH-1-Plk3 pathway represents a core pathway that is required for hepatic ER stress resolution . Although it is as essential as any of the three canonical arms of the UPR , and appears to function independently of them , it cannot be considered a fourth arm because there is no evidence that it directly contributes to restoration of ER function , for example by promoting protein folding , and also because LRH-1 is not universally expressed . Nonetheless , our results open new directions for therapeutic targeting of ER stress in human disease , and suggest that this pathway , which links highly conserved kinase signaling and resolution of ER stress , is a prime example of the emerging integration of broad cell stress responses and regulation of protein homeostasis .
Lrh-1 liver specific knockout ( Lrh-1LKO ) mice were obtained by crossing mice with an Lrh-1 allele flanked by LoxP sites ( Lrh-1f/f ) with albumin-Cre transgenic mice . Lrh-1f/f mice were provided by the Kliewer/Mangelsdorf lab and have been previously described ( Lee et al . , 2008 ) ; albumin-Cre transgenic mice were provided by Bert O’Malley’s laboratory at Baylor College of Medicine . Wild-type and Plk3−/− mice were obtained from Peter Stambrook’s lab and have been previously described ( Myer et al . , 2011 ) . Male mice 8–12 weeks of age were injected intraperitoneally with tunicamycin ( 1 mg/kg body weight ) or vehicle ( 2% DMSO ) in 150 mM dextrose . A dose of 0 . 5 mg/kg TM was used for Plk3 WT and Plk3−/− mice due to sensitivity of this strain to the drug . Conditionally expressed humanized Lrh-1 transgenic ( hLrh-1 TG ) mice were a kind gift from Franco DeMayo and were generated as previously described for hCoup-TF1 ( Wu et al . , 2010 ) unless otherwise indicated . In short , cDNA for human Lrh-1 was cloned into a shuttle vector , yielding an N terminus-FLAG-myc tandem-tagged protein . The shuttle vector was homologously recombined with a base vector containing two ROSA26 genomic sequences and a loxP-STOP-loxP cassette and transformed into 294-Flp cells for generation of the targeting construct . The targeting construct was linearized and electroporated into AB ES cells , instead of R1 cells as previously described . ES clones were screened and C57 chimeras were produced as previously described . Mice were maintained in a mixed background and crossed with Lrh-1f/f mice with one allele of albumin-Cre . We confirmed that hLrh-1 TG; Lrh-1LKO mice ( hLrh-1 TG/+; Lrh-1f/f; albumin-Cre/+ ) had robust expression of hLrh-1 by PCR . Methods were approved by Baylor College of Medicine’s Institutional Animal Care and Use Committee . Nuclear and cytoplasmic fractions were obtained from fresh liver tissue or primary hepatocytes . Fractionation was adapted from a protocol written for cultured cells obtained from the lab website of David Ron ( http://ron . medschl . cam . ac . uk/protocols/NucCyto . html ) . Briefly , tissue was dounce homogenized in a solution containing 10 mM HEPES , 50 mM NaCl , 0 . 5M sucrose , 0 . 1 mM EDTA , and 0 . 5% Triton X100 . This suspension was spun at 1000 rpm at 4°C to pellet nuclei . Primary hepatocytes were simply harvested in the above solution and spun as above . The cytoplasmic supernatant was re-spun and resulting supernatant was collected . Nuclei were washed in a solution containing 10 mM HEPES , 10 mM KCl , 0 . 1 mM EDTA , and 0 . 1 mM EGTA and pellets were then suspended in a solution of 10 mM HEPES , 500 mM NaCl , 0 . 1 mM EDTA , 0 . 1 mM EGTA , and 0 . 1% NP-40 . This suspension was vortexed and spun to yield a supernatant containing nuclear proteins . To obtain insoluble proteins ( containing PLK3 ) , we homogenized liver tissue in Laemmli buffer with B-mercaptoethanol and boiled for 10 min . Samples were separated on 4–12% Bis-Tris gels ( Invitrogen , Carlsbad , CA ) . For cytoplasmic fractions or whole cell lysates , 100 μg of total protein was loaded . For nuclear fractions , 10 μg of total protein was loaded . The following antibodies and conditions were used: cleaved PARP ( 9544; 1:500; Cell Signaling , Danvers , MA ) , caspase-6 ( 9762; 1:500; Cell Signaling ) , cleaved caspase-3 ( 9664; 1:500; Cell Signaling ) , XBP-1 ( sc-7160; 1:200; Santa Cruz , Dallas , TX ) , ATF4 ( sc-200; 1:200; Santa Cruz ) , ATF6 ( IMG-273; 1:200; Imgenex , San Diego , CA ) ; LRH-1 ( PP-H2325-10; 1:200; Perseus Proteomics , Tokyo , Japan ) , pATF2 ( 9225S; 1:500; Cell Signaling [note: we have experienced difficulties with recent batches] ) , ATF2 ( ab47476; 1:1000; Abcam , Cambridge , England ) ; and PLK3 ( 4896S; 1:400; Cell Signaling ) . Immobilon Western substrate ( Millipore , Billerica , MA ) was used for detection . RNA was isolated from snap-frozen liver tissue by homogenization in TriZol reagent ( Invitrogen ) and precipitation with ethanol . cDNA was synthesized from 1 μg RNA using Qiagen’s QuantiTect reverse transcription kit . Quantitative PCR was run on a Roche LightCycler 480 with Perfecta SYBR Green FastMix obtained from Quanta BioSciences ( Gaithersburg , MD ) . Samples were run in triplicate from 3 to 6 samples/group and expression was normalized to Tbp . Standard curves were ran for each primer set and relative fold changes were calculated with the ΔΔCt method . Primer sequences are available in Supplementary file 1 . Mice were anesthetized with a lethal dose of tribromoethanol and perfused with Earle’s balanced salt solution ( EBSS ) containing 5 mM EGTA . Perfusion was accomplished by inserting a 25 G needle into the inferior vena cava , cutting the portal vein , and using a peristaltic pump to deliver perfusion solutions . Following perfusion with EBSS , Hank’s balanced salt solution ( HBSS ) with 100 U/ml collagenase and containing trypsin inhibitor was perfused . The liver was then removed and massaged to obtain dissociated cells in hepatocyte wash media ( 17704-024; Invitrogen ) . The cells were passed through a mesh to obtain a single cell suspension . This suspension was layered onto Percoll ( P4937; Sigma , St . Louis , MO ) and spun at 600 rpm for 10 min to isolate dead cells . Live cells pelleted were washed and viability was assessed . The cells were plated in collagen-coated dishes in Williams E media ( 12551; Invitrogen ) containing insulin-transferrin-selenium supplementation . In studies in which cells were not transduced with adenovirus , cells were treated 12 hr later . In studies in which cells that were transduced with adenovirus , adenovirus was added 6 hr post plating at a multiplicity of infection of 100 . The cells were kept for 36 hr in the same media prior to treatment . We used Clontech’s Adeno-X Adenoviral System 3 to generate tetracycline-inducible adenovirus according to manufacturer’s instructions . We cloned the coding sequence for mouse Plk3 from cDNA , along with supplied LacZ control fragment , into the pAdenoX-Tet3G vector . Plasmids were linearized with PacI and transfected into 293T/17 cells with Lipofectamine 2000 ( Invitrogen ) . Titre was determined by use of Adeno-X Rapid Titer Kit ( Clontech , Mountain View , CA ) . Virus was delivered experimentally as crude lysate . We used the modified AdEasy system ( Luo et al . , 2007 ) to generate non-inducible adenoviruses expressing LacZ , DN Atf2 , and C2/Atf2 . To construct Ad-DN Atf2 , we cloned the coding sequence for mouse Atf2 ( NM_001025092 ) and used overlap extension PCR to introduce threonine to alanine mutations ( T51A; T53A ) corresponding to amino acids T69/71 in human ATF2 . To construct Ad-C2/Atf2 , we cloned C2/Atf2 from the pCMV-Flag-C2/Atf2 vector ( Steinmuller and Thiel , 2003 ) generously provided by Dr Gerald Thiel . As previously described , this vector contains the DNA-binding domain of ATF2 fused with the activation domain of ATF4 to generate a constitutively active ATF2 mutant . These sequences , along with the coding sequence for LacZ , were digested with XhoI and KpnI and ligated into pShuttle-CMV ( Addgene , Cambridge , MA ) . These vectors were electroporated into BJ5183-AD-1 cells ( Stratagene , La Jolla , CA ) , which contain AdEasy-1 encoding the adenoviral backbone , for recombination . Potential recombinants were screened and successful recombinants were PacI linearized and transfected in 293 cells by calcium phosphate transfection . Titre was determined by use of Adeno-X Rapid Titer Kit ( Clontech ) . Virus was delivered experimentally as crude lysate . Primary hepatocytes were plated on collagen-coated glass coverslips and treated with TM . The cells were washed and fixed in 4% paraformaldehyde for 20 min at room temperature . The cells were again washed and stained with 500 μM Thioflavin T in PBS for 3 min at room temperature . Fluroscence was visualized and imaged using a Zeiss Axioplan 2 microscope ( GFP filter ) . TLR-3 cells were obtained from the Health Science Research Resources Bank ( HSRRB ) , part of the Japan Health Sciences Foundation . These cells were derived from C57Bl/6 mouse hepatocytes immortalized with a temperature sensitive large T antigen ( tsSV40 large T antigen ) . We cultured cells at the permissive temperature of 33°C on collagen-coated plates . Cells were cultured in DMEM supplemented with 5% FBS , 1x insulin-transferrin-selenium supplement ( Invitrogen ) , and 10 ng/ml EGF . TLR-3 cells were grown on collagen-coated plates and transfected with N-terminal tagged EGFP-LRH-1 ( Atanasov et al . , 2008 ) , a kind gift of Dr Thomas Brunner . The cells were treated with TM and fixed with 4% paraformaldehyde for 20 min at room temperature . The cells were counterstained with DAPI and visualized with a Zeiss Axioplan 2 microscope ( GFP and DAPI filters ) . For primary hepatocyte experiments , we plated cells at half the typical density ( 31 . 19 thousand per 24-well plate well ) . siRNAs were purchased from Invitrogen ( Stealth siRNA ) and validated with the exception of siRNA to Ire1a , which was purchased as a pool from Dharmacon ( Lafayette , CO ) ( ON-TARGETplus SMARTpool ) . 50 pmol siRNA was transfected into each well using Lipofectamine RNAiMAX transfection reagent ( Invitrogen ) . Cells were treated with TM 48 hr post transfection and knockdown validated at 56 hr ( representing the latest timepoint collected ) . For experiments in TLR-3 cells , cells were plated at 70% confluency in collagen-coated plates . Lrh-1 siRNA was purchased from Invitrogen ( Stealth siRNA ) and 200 pmol was transfected in each well of a 12 well plate . Cells were treated with TM 48 hr post transfection . TLR-3 cells were plated in collagen-coated 24-well plates and transfected with 200 ng cAMP reponse element-luciferase ( CRE-luciferase ) reporter ( Qiagen , Valencia , CA ) , 150 ng β-galactosidase , 200 ng mouse LRH-1 , and 200 ng ATF2 ( Addgene ) . 48 hr later , the cells were treated with TM or the following kinase inhibitors: 10 μM D-JNKi for JNKs ( Sigma ) , 1 μM SB202190 for p38 ( Tocris ) , 10 μM GW84362X for PLK1/PKL3 ( Tocris , Bristol , UK ) , or 1 μM GSK650394A for SGK ( Tocris ) . 24 hr after treatment , cells were lysed in Tropix lysis buffer ( 100 mM potassium phosphate , 0 . 2% TritonX-100 , pH 7 . 8 ) plus DTT . Lysates were plated in triplicates in 96-well plates and 85 μl of reaction buffer was automatically injected by the luminometer . Reaction buffer for each well was prepared as follows: 0 . 7 μl galacton ( Applied Biosystems , Foster City , CA ) , 88 μM luciferin , 2 . 4 mM ATP , and 11 . 9 mM MgCl2 in 0 . 11M Tris-phosphate buffer , pH 7 . 8 . Firefly luciferase activity was measured and samples were incubated 1 hr at room temperature . 100 μl Tropix Accelerator ( Applied Biosystems ) was then automatically injected and measured to quantify β-galactosidase activity , which was used to normalize luciferase activity values . Lipid was extracted from liver by homogenizing in 9 volumes PBS and was added to a chloroform:methanol ( 2:1 ) mixture . PBS was added and the samples were spun at 3000 rpm for 10 min at 4°C . The bottom layer was removed and solute evaporated overnight . Lipid was resuspended in 1% Triton-X in ethanol for 4 hr with rotation . This was then used to calculate liver triglycerides and non-esterified fatty acids using Thermo Scientific’s ( Waltham , MA ) Infinity Triglyceride kit and Wako’s ( Richmond , VA ) NEFA kit , respectively . Triglyceride was extracted from primary hepatocytes by scraping cells into PBS and shaking at 2500 rpm for 15 s ( 2x ) with a MagNA Lyser bead homogenizer ( Roche , Basel , Switzerland ) . The samples were sonicated for 10 s prior to spinning at 4000 rpm 5 min to clear sample . Triglycerides were quantified with Thermo Scientific’s Infinity Triglyceride kit and protein quantified by Bradford assay . Liver tissue was frozen in optimum cutting temperature compound ( Sakura Finetek , Torrance , CA ) on dry ice . Tissue was finely cut using a cryostat , fixed , and stained with Oil Red O solution . Nuclei were also counterstained with hematoxylin . Tissue cutting and staining was performed by the Comparative Pathology Laboratory at Baylor College of Medicine . Liver tissue was frozen in optimum cutting temperature compound ( Sakura Finetek ) on dry ice . Tissue was finely cut using a cryostat by the Comparative Pathology Laboratory at Baylor College of Medicine . We then performed TUNEL staining in our lab with use of Takara’s In Situ Apoptosis Detection Kit ( Otsu , Japan ) . Briefly , tissue was fixed in acetone for 30 min and washed . Tissue was permeablized with buffer contained in kit for 5 min and incubated with labeling mix ( containing TdT enzyme and fluorescein-dUTP ) for 90 min at 37°C . Tissue was then washed and mounted with a medium containing DAPI . Fluorescent images were captured using a Zeiss Axioplan 2 microscope at 400X using OpenLab 3 . 1 . 5 software . For primary hepatocytes , the cells were fixed in 4% PFA for 1 hr and permeabilized ( 0 . 1% Triton-X100 in 0 . 1% sodium citrate ) for 5 min . The cells were washed and stained with TUNEL reaction mixture ( 2 . 5 mM cobalt chloride , 0 . 4 U/μl Tdt enzyme , and 2 μM fluorescein-dUTP ) for 60 min at 37°C . The cells were washed and stained with DAPI and counted with a Zeiss Axioplan 2 microscope . Surgical removal of 2/3 liver mass was achieved by anesthetizing mice with tribromoethanol and resecting the three most anterior lobes of the liver . This was accomplished by tying off the lobes with silk suture to stop blood flow and cutting immediately below the suture . The right and left medial lobe were sutured together and the left lateral lobe was sutured separately . Mice undergoing sham surgery underwent anesthesia and laparotomy but no liver lobes were removed . Mice were treated with TM ( 1 mg/kg body weight ) or vehicle for 24 hr . Total RNA from liver tissue was extracted using Trizol reagent ( Invitrogen ) and purity of the RNA was assessed by Agilent 2100 Bioanalyzer . 500 ng of RNA was reverse transcribed into cRNA and biotin-UTP labeled using the Illumina TotalPrep RNA Amplification Kit ( Ambion , Austin , TX ) . cRNA was quantified using an Agilent Bioanalyzer 2100 and hybridized to the Illumina mouseRefseq-8v2 Expression BeadChip using standard protocols ( Illumina , San Diego , CA ) . Image data was converted into unnormalized Sample Probe Profiles using the Illumina BeadStudio software . Arrays were normalized with Chipster software ( Kallio et al . , 2011 ) using default settings for Illumina arrays . To identify over-represented transcription factor binding sites , we used the Molecular Signatures Database , part of the Gene Set Enrichment Analysis software ( Subramanian et al . , 2005 ) . We analyzed the top 100 genes induced at least 1 . 5-fold by tunicamycin in control mice with significantly different induction in Lrh-1LKO mice . Promoters from the input gene list were analyzed for enriched presence of 615 different known transcription factor binding sites ( as defined in the TRANSFAC version 7 . 4 database ) . p<0 . 05 was considered significant . To generate a table of genes with significantly different induction by TM between genotypes , we filtered for genes induced at least 1 . 5-fold by TM in control mice with significantly different induction in Lrh-1LKO mice by t-test ( p<0 . 05 ) . We secondarily filtered for those with significantly different expression when treated with TM by t-test ( p<0 . 05 ) . For genome-wide binding of ATF2 , we downloaded ENCODE data ( wgEncodeEH002306 ) as a BED file and used HOMER software ( Heinz et al . , 2010 ) to annotate distance to TSS for each set of peak coordinates . For genome-wide binding of LRH-1 , we obtained a BED file from the lab of Tim Osborne ( Chong et al . , 2012 ) and annotated distance to TSS with HOMER . For DLPC and TM-treated samples , control mice were treated with 100 mg/kg DLPC by oral gavage as previously described ( Lee et al . , 2011 ) every 12 hr ( twice before TM and twice afterwards ) and 1 mg/kg TM by i . p . injection . Microarray for these samples was performed alongside previously described samples . Fold change was calculated for TM-treated samples and compared to TM- and DLPC-treated samples ( n = 3 ) . Primary hepatocytes were washed and subsequently fixed in fresh 4% paraformaldehyde for 30 min . Following this , the cells were washed and then stained with Deep Neutral Red Lipidtox ( Invitrogen ) for 1 hr . The cells were then counterstained with DAPI and imaged with a Zeiss Axioplan 2 microscope at 400X using OpenLab 3 . 1 . 5 software . Merged images were enhanced for contrast using identical settings with GNU Image Manipulation Program . 2-way ANOVA with Bonferroni post-hoc tests was performed for experiments using SPSS software . Significant differences between genotypes were marked by asterisks on graphs with significance at the level of p<0 . 01 unless otherwise indicated . Error bars on graphs represent standard error of the mean ( SEM ) . | A protein can only work properly if it has been folded into the correct shape . However , it is estimated that about one third of new proteins have the wrong shape . This is a major challenge for cells because misfolded proteins are often toxic , and cause many neurodegenerative and metabolic disorders . In eukaryotic cells , most protein folding takes place inside a part of the cell called the endoplasmic reticulum ( ER ) . If an incorrectly folded protein is detected , it is prevented from leaving the ER until it is refolded correctly , or destroyed . If too many proteins are misfolded , a process called the unfolded protein response helps the cell to cope with this ‘ER stress’ by expanding the ER and producing more of the molecules that assist protein folding . If this does not relieve the ER stress , the cell self-destructs . Neighboring cells then have to increase protein production to compensate for what would have been produced by the dead cell , thereby increasing the chance that they will also experience ER stress . Activation of a protein called LRH-1 ( short for liver receptor homolog-1 ) that is produced in the liver , pancreas and intestine can relieve the symptoms of the various metabolic diseases that are associated with chronic ER stress , including type II diabetes and fatty liver disease . However , researchers have been puzzled by the fact that although LRH-1 performs many different roles , its molecular structure provides few clues as to how it can do this . Mamrosh et al . now confirm the speculated link between LRH-1 and ER stress relief in mice . LRH-1 triggers a previously unknown pathway that can relieve ER stress and is completely independent of the unfolded protein response . Targeting LRH-1 with certain chemical compounds alters its activity , suggesting that drug treatments could be developed to relieve ER stress . As similar targets for drugs have not been found in the unfolded protein response , the discovery of the LRH-1 pathway could lead to new approaches to the treatment of the diseases that result from ER stress . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2014 | Nuclear receptor LRH-1/NR5A2 is required and targetable for liver endoplasmic reticulum stress resolution |
The NLRP3 inflammasome is a multi-molecular protein complex that converts inactive cytokine precursors into active forms of IL-1β and IL-18 . The NLRP3 inflammasome is frequently associated with the damaging inflammation of non-communicable disease states and is considered an attractive therapeutic target . However , there is much regarding the mechanism of NLRP3 activation that remains unknown . Chloride efflux is suggested as an important step in NLRP3 activation , but which chloride channels are involved is still unknown . We used chemical , biochemical , and genetic approaches to establish the importance of chloride channels in the regulation of NLRP3 in murine macrophages . Specifically , we identify LRRC8A , an essential component of volume-regulated anion channels ( VRAC ) , as a vital regulator of hypotonicity-induced , but not DAMP-induced , NLRP3 inflammasome activation . Although LRRC8A was dispensable for canonical DAMP-dependent NLRP3 activation , this was still sensitive to chloride channel inhibitors , suggesting there are additional and specific chloride sensing and regulating mechanisms controlling NLRP3 .
Inflammation is an important protective host-response to infection and injury , and yet is also detrimental during non-communicable diseases ( Dinarello et al . , 2012 ) . Inflammasomes are at the heart of inflammatory responses . Inflammasomes are formed by a soluble pattern recognition receptor ( PRR ) , in many cases the adaptor protein ASC ( apoptosis-associated speck-like protein containing a CARD ) , and the protease caspase-1 ( 2 ) . Inflammasomes form in macrophages in response to a specific stimulus to drive the activation of caspase-1 , facilitating the processing of the cytokines pro-interleukin ( IL ) 1β and pro-IL-18 to mature secreted forms , and the cleavage of gasdermin D to cause pyroptotic cell death ( Evavold and Kagan , 2019 ) . A number of different inflammasomes have been described , but potentially the inflammasome of greatest interest to non-communicable disease is formed by NLRP3 ( NACHT , LRR and PYD domains-containing protein 3 ) ( Mangan et al . , 2018 ) . The mechanisms of NLRP3 activation remain poorly understood . The NLRP3 inflammasome is activated through several routes which have been termed the canonical , non-canonical , and the alternative pathways ( Mangan et al . , 2018 ) . Activation of the canonical NLRP3 pathway , which has received greatest attention thus far , typically requires two stimuli; an initial priming step involving a pathogen associated molecular pattern ( PAMP ) , typically bacterial endotoxin ( lipopolysaccharide , LPS ) , to induce expression of pro-IL-1β and NLRP3 , and a second activation step usually involving a damage associated molecular pattern ( DAMP ) , such as adenosine triphosphate ( ATP ) ( Mariathasan et al . , 2006 ) . In 1996 , Perregaux and colleagues discovered that hypotonic shock was effective at inducing the release of mature IL-1β when applied to LPS-treated human monocytes and suggested the importance of a volume-regulated response ( Perregaux et al . , 1996 ) . It was later discovered that hypotonicity induced release of IL-1β via activation of the NLRP3 inflammasome ( Compan et al . , 2012 ) , and that this was linked to the regulatory volume decrease ( RVD ) , which is a regulated reduction in cell volume in response to hypo-osmotic-induced cell swelling , and was inhibited by the chloride ( Cl- ) channel blocker NPPB ( 5-nitro- ( 3-phenylpropylamino ) benzoic acid ) ( Compan et al . , 2012 ) . The RVD is regulated by the Cl- channel VRAC ( volume regulated anion channel ) . The molecular composition of the VRAC channel was established to consist of an essential LRRC8A sub-unit in combination with other ( B-E ) LRRC8 sub-units ( Qiu et al . , 2014; Voss et al . , 2014 ) . We recently reported that fenamate NSAIDs could inhibit the canonical NLRP3 inflammasome by blocking a Cl- channel , which we suggested could be VRAC ( Daniels et al . , 2016 ) . We also further characterised the importance of Cl- flux in the regulation of NLRP3 , showing that Cl- efflux facilitated NLRP3-dependent ASC oligomerisation ( Green et al . , 2018 ) . Given the poor specificity of many Cl- channel inhibitors , we set out to systematically determine the importance of VRAC and the RVD to NLRP3 inflammasome activation . Given that hypertonic buffers inhibit ATP and other NLRP3 activating stimuli-induced IL-1β release from macrophages ( Perregaux et al . , 1996; Compan et al . , 2012 ) , we hypothesised that cell swelling , and the RVD , would be central to all NLRP3 activating stimuli . Using pharmacological and genetic approaches , we discovered that VRAC exclusively regulated RVD-dependent NLRP3 activation in response to hypotonicity , and not NLRP3 activation in response to other canonical stimuli . Thus , we provide genetic evidence for the importance of Cl- in regulating NLRP3 via the VRAC dependence of the hypotonicity response , and suggest the presence of additional Cl- sensing mechanisms regulating NLRP3 in response to DAMPs .
Following publication of the cryo-electron microscopy ( cryo-EM ) structure of VRAC ( Deneka et al . , 2018 ) with inhibitor 4- ( 2-butyl-6 , 7-dichloro-2-cyclopentyl-indan-1-on-5-yl ) oxobutyric acid ( DCPIB ) ( Figure 1A; Kern et al . , 2019 ) , we were able to investigate the interaction of established VRAC inhibitors with the channel using molecular modelling . DCPIB was first computationally redocked using Molecular Operating Environment ( MOE 2015 . 08 , Chemical Computing Group , Canada ) into the homohexameric VRAC structure ( PDB code 6NZW , resolution 3 . 2 Å ) ( Kern et al . , 2019 ) . The resulting pose produced a reasonable overlay with the cryo-EM conformation of DCPIB , giving a root-mean-square deviation in atomic position of 2 . 6 Å ( Figure 1B , C ) . DCPIB exhibits an ionic interaction of its carboxylate group with the cationic side-chain of one of the Arg103 residues comprising the electropositive selectivity filter of VRAC ( Kern et al . , 2019 ) . Known VRAC inhibitors ( Daniels et al . , 2016; Hélix et al . , 2003; Droogmans et al . , 1999 ) possessing carboxylic acid groups ( flufenamic acid ( FFA ) ) , mefenamic acid ( MFA ) and N- ( ( 4-methoxy ) −2-naphthyl ) −5-nitroanthranilic acid ( MONNA ) , the tetrazole moiety ( NS3728 ) and sulfonic acid groups ( 4-sulfonic calix[6]arene ) were then docked into the DCPIB site of VRAC . The most favourably bound poses of these ligands were similarly found to block the pore in a cork-in-bottle manner ( Kern et al . , 2019 ) at the selectivity filter; the ligands’ ionised acidic groups formed strong electrostatic interactions with Arg103 ( Figure 1D–H ) . Tamoxifen , a basic inhibitor of VRAC was also docked into the cryo-EM structure ( Figure 1I ) . Accordingly , tamoxifen docked with its cationic tertiary amino group remote to the Arg103 side-chains ( Figure 1l and Figure 1—figure supplement 1 ) ; these side-chains instead formed cation-π interactions with the phenyl group of tamoxifen ( Figure 1l ) . The interaction of the tamoxifen pose was computed as having a calculated ligand-binding affinity of −6 . 5 kcal mol−1 via the molecular mechanics/generalised Born volume integration ( MM/GBVI ) method ( Labute , 2008; Galli et al . , 2014 ) . The binding energies of the anionic ligands were also predicted as favourable , ranging from −5 . 1 ( DCPIB ) to −5 . 8 kcal mol−1 ( MONNA ) . This range excludes the larger 4-sulfonic calix[6]arene ( calixarene ) , which gave a binding energy of −9 . 9 kcal mol−1; we note that the GBVI implicit solvent model may be underestimating the high desolvation cost of this polyanionic ligand and therefore overestimating the magnitude of the corresponding binding energy of this compound . We then tested the ability of six of the compounds described in Figure 1 ( DCPIB , calixarene ( Calix ) , tamoxifen , FFA , MONNA , NS3728 ) to inhibit hypotonicity-induced VRAC-dependent Cl- flux using the iodide ( I- ) quenching of halide-sensitive YFP H148Q/I152L ( Galietta et al . , 2001 ) in live HeLa cells ( Figure 2A ) . In this model , I- enters the cell through open Cl- channels to induce quenching of a mutant EYFP . In response to hypotonic shock to induce VRAC opening , YFP fluorescence was immediately quenched , which was significantly inhibited by tamoxifen ( 10 µM ) , MONNA ( 50 µM ) , DCPIB ( 10 µM ) , FFA ( 100 µM ) , and NS3728 ( 10 µM ) , but not calixarene ( 100 µM ) ( Figure 2A , B ) . VRAC also regulates RVD in response to cell swelling ( Qiu et al . , 2014; Voss et al . , 2014 ) . We measured the RVD by measuring the change in cellular fluorescence in calcein-loaded primary mouse bone-marrow-derived macrophages ( BMDMs ) in response to hypotonicity . Hypotonicity caused a rapid increase in cell volume which declined over time , characteristic of an RVD response ( Figure 2C ) . Similar to the quenching assay , RVD was also significantly inhibited in the presence of tamoxifen , MONNA , DCPIB , FFA and NS3728 , but not calixarene ( Figure 2C , D ) . These data suggest that all the molecules in our panel , except calixarene , are bona-fide VRAC inhibitors at the concentrations tested . We tested whether the panel of VRAC inhibitors characterised above could block NLRP3 inflammasome activation and release of IL-1β in response to DAMP stimulation . Primary BMDMs were primed with LPS ( 1 µg mL−1 , 4 hr ) , before activation of NLRP3 by ATP ( 5 mM , 2 hr ) . Inhibitors were given at the same dose they inhibited VRAC above in HeLa cells 15 min before the addition of ATP and were then present for the duration of the experiment . Of the panel of verified VRAC inhibitors , only MONNA , FFA and NS3728 consistently inhibited ATP-induced IL-1β release ( Figure 3A ) . At the dose used in this assay , DCPIB did not consistently inhibit ATP-induced IL-1β release ( Figure 3A ) , but at higher concentrations did inhibit NLRP3 activation ( Figure 3—figure supplement 1 ) . Likewise , pyroptosis , as measured by LDH release , was significantly reduced by MONNA and FFA , and was unaffected by tamoxifen or calixarene ( Figure 3B ) . The inhibitors that blocked ATP-induced IL-1β release also inhibited ASC oligomerisation , caspase-1 activation , and gasdermin D cleavage ( Figure 3C ) . These data show that some very effective VRAC inhibitors failed to inhibit activation of the NLRP3 inflammasome and release of IL-1β , suggesting that VRAC may not be the molecular target of these molecules inhibiting the inflammasome . ATP-induced NLRP3 activation is dependent upon K+ efflux , whereas NLRP3 activation by treatment with the imidazoquinoline compound imiquimod is K+ efflux-independent ( Groß et al . , 2016 ) . We therefore sought to test if VRAC inhibitors that were effective at blocking ATP-induced inflammasome activation were specific to K+ efflux-sensitive mechanisms . FFA ( 100 µM ) and NS3728 ( 10 µM ) were effective at blocking IL-1β release after treatment with the K+ ionophore nigericin ( 10 µM , 2 hr ) ( Figure 3D ) , but were unable to block imiquimod ( 75 µM , 2 hr ) -induced IL-1β release ( Figure 3E ) . Similarly , ASC oligomerisation , caspase-1 activation and gasdermin D cleavage induced by nigericin were sensitive to FFA and NS3728 pre-treatment ( Figure 3F ) . However , ASC oligomerisation , caspase-1 activation and gasdermin D cleavage induced by imiquimod were not affected by FFA and NS3728 pre-treatment ( Figure 3F ) . Increased extracellular KCl ( 25 mM ) was sufficient to block nigericin-induced activation , but not imiquimod , demonstrating the K+ dependency of nigericin ( Figure 3F ) . These data suggest that these Cl- channel inhibiting compounds exclusively target the K+-dependent canonical pathway of NLRP3 activation . Many Cl- channel inhibiting drugs are known to inhibit multiple Cl- channels , and we established that very effective VRAC inhibitors ( tamoxifen and DCPIB ) had negligible effect on NLRP3 activation at VRAC inhibiting concentrations . Thus , to conclusively determine the role of VRAC in NLRP3 inflammasome activation we generated a macrophage specific LRRC8A knockout ( KO ) mouse using CRISPR/Cas9 ( Figure 4A ) . The generation of a macrophage-specific LRRC8A KO was required as whole animal LRRC8A KO mice do not survive beyond 4 weeks and have retarded growth ( Kumar et al . , 2014 ) . Lrrc8afl/fl mice were bred with mice constitutively expressing Cre under the Cx3cr1 promoter , as previously shown to be expressed in monocyte and macrophage populations ( Yona et al . , 2013 ) . This generated mice with the genotype Lrrc8afl/fl:Cx3cr1cre ( KO ) with littermates Lrrc8afl/fl:Cx3cr1WT ( WT ) . Cell lysates were prepared from BMDMs and peritoneal macrophages isolated from WT and KO mice and were western blotted for LRRC8A confirming that Lrrc8a KO cells were knocked out for LRRC8A ( Figure 4B ) . Functional loss of LRRC8A was confirmed using the calcein RVD assay described above . BMDMs were subjected to a hypotonic shock and changes in calcein fluorescence measured over time . In WT cells , there was a characteristic RVD ( Figure 4C ) . However , in Lrrc8a KO cells there was complete loss of the RVD response ( Figure 4C , D ) . The absence of RVD was also strikingly evident by observation of the cells by phase contrast microscopy ( Figure 4E , Videos 1 and 2 ) . Treatment of Lrrc8a KO BMDMs with DCPIB and subsequent hypotonic shock resulted in a further dysregulation of cell volume compared to WT cells or KO cells treated with hypotonic shock alone , suggesting that additional Cl- channels act to constrain cell volume in the absence of a functional RVD ( Figure 4—figure supplement 1 ) . These data confirm functional KO of the VRAC channel in macrophages . We next used the Lrrc8a KO macrophages to test the hypothesis that VRAC and the RVD were important for NLRP3 inflammasome activation and IL-1β release in response to DAMP stimulation . WT BMDMs and Lrrc8a KO BMDMs were primed with LPS ( 1 µg mL−1 , 4 hr ) and then treated with the NLRP3 inflammasome activators ATP ( 5 mM , 2 hr ) , nigericin ( 10 µM , 2 hr ) , silica ( 300 µg mL−1 , 2 hr ) , or imiquimod ( 75 µM , 2 hr ) . Knocking out LRRC8A had no effect on the release of IL-1β ( Figure 5A ) or cell death ( Figure 5B ) . We then used western blotting to determine ASC oligomerisation and caspase-1 activation . In response to the NLRP3 inflammasome activators nigericin , ATP , and imiquimod , there was no effect of LRRC8A KO on ASC oligomerisation or caspase-1 activation ( Figure 5C ) . Furthermore , IL-1β release in response to ATP or nigericin was still inhibited by the VRAC inhibitors flufenamic acid ( FFA , 100 µM ) , and NS3728 ( 10 µM ) in the Lrrc8a KO BMDMs , confirming that these inhibitors are inhibiting NLRP3 inflammasome activation by a VRAC-independent mechanism ( Figure 5D ) . Flufenamic acid and NS3728 also inhibited ASC oligomerisation and caspase-1 activation as determined by western blot in the Lrrc8a KO BMDMs to the same extent as in the WT ( Figure 5E ) . We then used a murine peritonitis model described previously ( Daniels et al . , 2016 ) to investigate the role of LRRC8A in vivo . First , we tested if the VRAC inhibitor NS3728 was effective at blocking NLRP3 in vivo . Wild-type C57BL6/J mice were injected intraperitoneally with NS3728 ( 50 mg kg−1 ) , the NLRP3 inhibitor MCC950 ( 50 mg kg−1 ) , or vehicle control , at the same time as LPS ( 1 µg , 4 hr ) . NLRP3 was then further activated by intraperitoneal injection of ATP ( 100 mM , 500 µL , 15 min ) and IL-1β release was measured by ELISA of the peritoneal lavage ( Figure 5F ) and plasma ( Figure 5G ) . Addition of ATP induced a significant increase in the release of IL-1β into peritoneal lavage and this was inhibited by MCC950 and NS3728 , indicating an NLRP3-dependent response . IL-6 levels were unaltered in both peritoneal lavage and plasma by addition of NS3728 ( Figure 5—figure supplement 1A , B ) . These data show that NS3728 was able to inhibit NLRP3 in vivo . To determine the role of VRAC in this model , we repeated this experiment in our macrophage Lrrc8a KO mice and their littermate controls . Macrophage Lrrc8a KO mice exhibited normal proportions of myeloid cells in the peritoneum as assessed by flow cytometry ( Figure 5—figure supplement 1C–G ) . Loss of macrophage LRRC8A had no effect on the IL-1β levels in the peritoneal lavage in response to LPS and ATP ( Figure 5H ) , or in the plasma ( Figure 5I ) . Moreover , similar to our in vitro findings , NS3728 was still effective at inhibiting this response in the absence of LRRC8A ( Figure 5H , I ) . These data suggested that VRAC was dispensable for NLRP3 activation by DAMP stimulation , and that the VRAC inhibitors are effective at inhibiting NLRP3 in the absence of VRAC , suggesting the presence of another target . RVD in response to hypo-osmotic-induced cell swelling is documented as an inducer of NLRP3 inflammasome activation and IL-1β release ( Perregaux et al . , 1996; Compan et al . , 2012 ) . Since Lrrc8a KO BMDMs could no longer control their volume in response to hypotonic shock , we tested whether NLRP3 inflammasome activation by hypotonicity was altered . LPS-primed ( 1 µg mL−1 , 4 hr ) BMDMs were incubated in a hypotonic solution ( 4 hr ) which caused IL-1β and LDH release from WT cells , and which was significantly inhibited in Lrrc8a KO BMDMs ( Figure 6A , B ) . There was no difference in IL-1β release or cell death between ATP-stimulated WT and KO BMDMs ( Figure 6A , B ) . Caspase-1 cleavage and IL-1β processing induced by hypotonicity were also completely inhibited in the absence of LRRC8A ( Figure 6C ) , indicating the response was completely dependent on both NLRP3 and VRAC . Moreover , hypotonicity-induced ASC oligomerisation was also dependent on VRAC ( Figure 6D ) . These data show that in response to hypo-osmotic stress , VRAC was essential for NLRP3 inflammasome activation .
Pharmacological and biochemical evidence supporting an important role of Cl- ions in the activation of the NLRP3 inflammasome has been provided by various studies over the years ( Perregaux et al . , 1996; Compan et al . , 2012; Daniels et al . , 2016; Green et al . , 2018; Verhoef et al . , 2005 ) , although conclusive genetic evidence has been lacking . The promiscuous nature of many Cl- channel inhibiting drugs , and an unresolved molecular identity of major Cl- channels , have prevented the emergence of conclusive genetic proof . However , the discovery that the Cl- channel regulating the RVD ( VRAC ) was composed of LRRC8 sub-units , and that LRRC8A was essential for channel activity , offered us the opportunity to investigate the direct importance of VRAC in the regulation of NLRP3 . Hypotonicity induces cell swelling which is corrected by the VRAC-dependent RVD ( Qiu et al . , 2014; Voss et al . , 2014 ) . The RVD was previously linked to NLRP3 activation ( Compan et al . , 2012 ) . Thus by knocking out LRRC8A , and thus VRAC , we would discover that VRAC was essential for RVD-induced NLRP3 inflammasome activation , providing strong evidence for the direct requirement of a Cl- channel in NLRP3 inflammasome activation . Given that there are a number of inflammatory conditions and models that can be targeted by administration of a hyper-tonic solution ( e . g . Theobaldo et al . , 2012; Schreibman et al . , 2018; Shields et al . , 2003; Petroni et al . , 2015 ) , it is possible that the VRAC-dependent regulation of NLRP3 in response to hypotonicity could represent a therapeutic target . However , VRAC was only essential for RVD-induced NLRP3 activation and was not involved in the NLRP3 response to DAMP stimulation . The fact that our VRAC channel-inhibiting drugs block DAMP-induced NLRP3 activation suggests that additional Cl- channels ( or alternative targets ) are involved in coordinating NLRP3 responses to other stimuli . Chloride intracellular channel proteins ( CLICs 1–6 ) form anion channels and regulate a variety of cellular processes ( Littler et al . , 2010; Argenzio and Moolenaar , 2016 ) . Localisation of CLIC1 and 4 to membrane fractions in macrophages is increased by LPS stimulation , and RNAi knockdown of both CLIC1 and 4 impaired LPS and ATP-induced IL-1β release from macrophages ( Domingo-Fernández et al . , 2017 ) . In addition to CLICs 1 and 4 , CLIC5 is also implicated in NLRP3-dependent IL-1β release ( Tang et al . , 2017 ) . Knockdown of CLICs 1 , 4 , and 5 inhibits NLRP3 inflammasome activation in response to the soluble agonists ATP and nigericin , and also the particulate DAMP monosodium urate crystals ( Tang et al . , 2017 ) . Thus , it appears that multiple Cl- channels encode diverse signals arising from DAMP stimulation , or from altered cellular homeostasis , to trigger NLRP3 inflammasome activation . Importantly our data suggest that Cl- channels are only important to NLRP3 activation dependent upon K+ efflux , highlighting the further potential for selective pathway modulation and therapeutic development . The relationship between K+ and Cl- efflux requires further investigation , and whether , in RVD , Cl- efflux is a pre-requisite for K+ efflux . Although activation of VRAC is best understood under hypotonic conditions , VRAC activation has also been reported to occur in response to a variety of stimuli ( Osei-Owusu et al . , 2018 ) . It is also possible that VRAC may regulate NLRP3 inflammasome activation in response to stimuli other than hypotonicity that were not tested here . For example , sphingosine-1-phosphate ( S1P ) activates VRAC in mouse macrophages ( Burow et al . , 2015 ) , and we previously reported that sphingosine , and S1P , could activate NLRP3 ( Luheshi et al . , 2012 ) . VRAC is also thought to be important for cell death induced by apoptosis inducing drugs ( Sørensen et al . , 2016; Planells-Cases et al . , 2015 ) , and we also previously reported that apoptosis inducing drugs could activate the NLRP3 inflammasome in activated macrophages ( England et al . , 2014 ) suggesting another potential area of relevance . While VRAC is directly permeable to Cl- it is also possible that additional Cl- channels are involved in the hypotonicity-induced swelling response . For example , VRAC activation caused by swelling can activate other Cl- channels , notably anoctamin 1 ( e . g . Liu et al . , 2019; Benedetto et al . , 2016 ) . Furthermore , whilst our research suggests an importance of Cl- , VRAC is also permeable to small molecules including cGAMP ( Zhou et al . , 2020 ) and ATP ( Dunn et al . , 2020 ) , highlighting additional ways through which VRAC could contribute to inflammation . Inhibiting the NLRP3 inflammasome has become an area of intense research interest due to the multiple indications of its role in disease ( Mangan et al . , 2018 ) . The inhibitor MCC950 is now thought to bind directly to NLRP3 to cause inhibition ( Coll et al . , 2019; Tapia-Abellán et al . , 2019 ) , although it has also been reported to inhibit Cl- flux from macrophages treated with nigericin ( Jiang et al . , 2017 ) , and was found to bind directly to CLIC1 ( Laliberte et al . , 2003 ) , so it is possible that some of its inhibitory activity may be attributable to an effect on Cl- . We found that Cl- channel inhibition blocked IL-1β release in a NLRP3-dependent model of peritonitis , and previously reported protective effects of the fenamate NSAIDs in rodent models of Alzheimer’s disease that we attributed to an effect on Cl- channel inhibition ( Daniels et al . , 2016 ) . Thus , it is possible that targeting Cl- channels offers an additional route to inhibit NLRP3-dependent inflammation in disease . In summary , we have reported that hypotonicity-induced NLRP3 inflammasome activation depends exclusively on the Cl- channel VRAC , and that different Cl- sensing and regulating systems coordinate the activation of NLRP3 in response to DAMPs . This opens the possibility of discrete Cl- regulating mechanisms conferring selectivity and information about the nature of the NLRP3 activating stimulus . Thus , this investigation has opened the door to further studies on Cl- regulation of NLRP3 and identified the possibility of selective therapeutic intervention strategies informed by the nature of the disease or DAMP stimulus , potentially minimising complications of immunosuppression caused by a blanket NLRP3 inhibition .
Docking of ligands to VRAC employed the recently solved cryo-EM structure of VRAC ( PDB: 6NZW , resolution 3 . 2 Å ) ( Kern et al . , 2019 ) . Tautomeric and ionisation states of VRAC amino acid residues at pH 7 . 4 were assigned using MOE ( MOE 2015 . 08 , Chemical Computing Group , Canada ) . Similarly , ligands were modelled in their ionised forms according to physiological conditions . Docking was performed with the Triangle Matcher placement method of MOE using the London dG scoring function ( MOE 2015 . 08 , Chemical Computing Group , Canada ) . The pocket into which the VRAC inhibitors were docked was that occupied by the ( S ) -isomer of DCPIB in the cryo-EM structure of VRAC . Rescoring of poses used the molecular mechanics ( MM ) /generalised Born/volume integral ( GBVI ) potential ( Labute , 2008 ) . Primary bone-marrow-derived macrophages ( BMDMs ) and peritoneal macrophages were isolated from male and female wild-type C57BL6/J mice . Bone marrow was harvested from both femurs , red blood cells were lysed and resulting marrow cells were cultured in 70% DMEM ( 10% v/v FBS , 100 U/mL penicillin , 100 μg/mL streptomycin ) supplemented with 30% L929 mouse fibroblast-conditioned media for 6–7 days . BMDMs were seeded out the day before at a density of 1 × 106 mL−1 in DMEM ( 10% v/v FBS , 100 U mL−1 penicillin , 100 μg mL−1 streptomycin ) . Peritoneal macrophages were isolated by peritoneal lavage and seeded out overnight at a density of 1 × 106 mL−1 in DMEM ( 10% v/v FBS , 100 U mL−1 penicillin , 100 μg mL−1 streptomycin ) . HeLa cells were seeded out at 0 . 1 × 106 mL−1 in DMEM ( 10% v/v FBS , 100 U mL−1 penicillin , 100 μg mL−1 streptomycin ) . HeLa cells were obtained from ATCC ( HeLa ( ATCC CCL-2 ) ) and are periodically tested for mycoplasma . Primary BMDMs were primed with LPS ( 1 µg mL−1 , 4 hr ) in DMEM ( 10% v/v FBS , 100 U mL−1 penicillin , 100 μg mL−1 streptomycin ) . After priming , the media was replaced with serum-free DMEM , or when specified an isotonic buffer ( 132 mM NaCl , 2 . 6 mM KCl , 1 . 4 mM KH2PO4 , 0 . 5 mM MgCl2 , 0 . 9 mM CaCl2 , 20 mM HEPES , 5 mM NaHCO3 , 5 mM Glucose , pH 7 . 3 , 340 mOsm/kg ) or hypotonic buffer ( 27 mM NaCl , 0 . 54 mM KCl , 0 . 3 mM KH2PO4 , 0 . 5 mM MgCl2 , 0 . 9 mM CaCl2 , 20 mM HEPES , 5 mM NaHCO3 , 5 mM Glucose , pH 7 . 3 , 117 mOsm kg−1 ) . When used , VRAC inhibitors were added 15 min before stimulation of the NLRP3 inflammasome . For analysis of IL-1β release and pyroptosis , cell supernatants were collected . IL-1β release was determined by ELISA ( DuoSet , R and D Systems ) according to the manufacturer’s instructions . Cell death was assessed by lactate dehydrogenase ( LDH ) release using CytoTox 96 nonradioactive cytotoxicity assay ( Promega ) according to manufacturer’s instructions . For western blotting , total cell lysates were made by directly adding protease inhibitor cocktail and Triton x-100 ( 1% v/v ) to each well containing cells and supernatant . 1 × 106 primary BMDMs were seeded out overnight into 12-well plates . After LPS priming ( 1 µg mL−1 , 4 hr ) , cells were incubated in either serum-free DMEM , an isotonic or a hypotonic buffer ( as described above ) and stimulated as described . BMDMs were lysed directly in-well by addition of protease inhibitor cocktail and Triton x-100 ( 1% v/v ) and lysed on ice . Total cell lysates were then spun at 6800xg for 20 min at 4°C to separate the lysate into Triton x-100 soluble and insoluble fractions . The Triton x-100 insoluble fraction ( pellet ) was then chemically crosslinked by addition of disuccinimidyl suberate ( DSS , 2 mM , 30 min , RT ) in PBS . Following crosslinking , the insoluble fraction was spun at 6800xg for 20 min and the resulting pellet was resuspended and boiled in 40 µL 1X Laemlli buffer . The Triton x-100 soluble fraction was concentrated by trichloroacetic acid ( TCA ) precipitation . Triton x-100 soluble lysate was mixed 1:1 with TCA ( 20% w/v ) and spun at 14 , 000xg for 10 min at 4°C . The pellet was then washed in acetone , spun at 14 , 000xg for 10 min at 4°C , and resuspended in 2X Laemlli buffer . Cell lysates were separated by Tris-glycine SDS PAGE and transferred onto nitrocellulose or PVDF membranes using a semidry Trans-Blot Turbo system ( Bio-Rad ) . Membranes were blocked ( 1 hr , RT ) in milk ( 5% w/v ) in PBS containing Tween 20 ( 0 . 1% v/v , PBS-T ) before incubation ( overnight , 4°C ) with indicated primary antibodies in bovine serum albumin ( 5% w/v ) in PBS-T . Membranes were washed three times for 5 min in PBS-T before incubation ( 1 hr , RT ) with appropriate HRP-conjugated secondary antibodies ( Dako ) . After a further three washes in PBS-T , membranes were incubated with Amersham ECL prime detection reagent ( GE healthcare ) and chemiluminescence was visualised using a G:Box Chemi XX6 ( Syngene ) . 5 × 104 BMDMs were seeded out into black walled 96-well plates overnight . Cells were loaded with calcein ( 10 µM , 1 hr , 37°C ) in an isotonic buffer ( 132 mM NaCl , 2 . 6 mM KCl , 1 . 4 mM KH2PO4 , 0 . 5 mM MgCl2 , 0 . 9 mM CaCl2 , 20 mM HEPES , 5 mM NaHCO3 , 5 mM Glucose , pH 7 . 3 , 340 mOsm kg−1 ) . Following loading , BMDMs were washed three times with isotonic buffer before incubation with VRAC inhibitors or vehicle control at indicated concentrations for 5 min . GFP fluorescence was then imaged for a further 5 min before hypotonic shock was induced by a fivefold dilution with a hypotonic buffer ( 0 . 9 mM CaCl2 , 20 mM HEPES , 5 mM NaHCO3 , 5 mM Glucose , pH 7 . 3 ) , resulting in a final osmolarity of 117 mOsm kg−1 . GFP fluorescence was measured on an Eclipse Ti inverted microscope ( Nikon ) and analysed using Image J software . Point visiting was used to allow multiple positions to be imaged within the same time-course and cells were maintained at 37°C and 5% CO2 . For experiments with VRAC inhibitors , combined treatment with hypotonicity and VRAC inhibitors resulted in some cells undergoing lytic cell death over the course of the experiment and loss of calcein fluorescence . Therefore , GFP fluorescence was used to identify the area of living cells . HeLa cells were seeded at a density of 0 . 1 × 106 ml−1 in black-walled , clear bottom 96-well plates ( Greiner ) . Transient transfection with the halide-sensitive YFP mutant pcDNA3 . 1 EYFP H148Q/I152L , a gift from Peter Haggie ( Addgene plasmid # 25872 ) , was performed using Lipofectamine 3000 ( Thermo Fisher ) . 18–24 hr post-transfection , HeLa cells were washed twice with isotonic buffer ( 140 mM NaCl , 5 mM KCl , 20 mM HEPES , pH 7 . 4 , 310 mOsm kg−1 ) before 5 min incubation in 50 µL isotonic buffer containing either drug at indicated concentrations , or vehicle . 50 µL isotonic or hypotonic ( 5 mM KCl , 20 mM HEPES , 90 mM mannitol , pH 7 . 4 , 120 mOsm kg−1 ) buffer containing either drug or vehicle was then added and cells were incubated for a further 5 min . NaI ( 200 mM , 25 µL ) was then added directly to the well , and fluorescence readings were take every 2 s using the FlexStation3 plate reader . We used CRISPR-Cas9 to generate the floxed LRRC8A allele on C57BL/6J background . LRRC8A is a four exon gene spanning 26 kb on mouse chromosome 2 . Only two of these exons contain coding sequence , with exon three harbouring >85% of the coding sequence and possessing large introns , and thus an ideal candidate for floxing . We initially attempted the 2-sgRNA , 2-oligo approach described previously ( Yang et al . , 2013 ) , but failed to obtain mice with both loxP integrated on the same allele ( Gurumurthy et al . , 2019 ) . Instead , a colony from a single founder with the 5’ LoxP integrated was established , bred to homozygosity , and used as a background to integrate the second 3’ loxP . For both steps , we used the Sanger WTSI website ( http://www . sanger . ac . uk/htgt/wge/ , Hodgkins et al . , 2015 ) to design sgRNA that adhered to our criteria for off target predictions ( guides with mismatch ( MM ) of 0 , 1 or two for elsewhere in the genome were discounted , and MM3 were tolerated if predicted off targets were not exonic ) . sgRNA sequences were purchased as crRNA oligos , which were annealed with tracrRNA ( both oligos supplied IDT; Coralville , USA ) in sterile , RNase-free injection buffer ( TrisHCl 1 mM , pH 7 . 5 , EDTA 0 . 1 mM ) by combining 2 . 5 µg crRNA with 5 µg tracrRNA and heating to 95°C , which was allowed to slowly cool to room temperature . For 5’ targeting the sgRNA GTCTAGTTAGGGACTCCTGG-GGG was used , with the ssODN repair template 5’-tccttgacttgctgtttaccgctctcttccccacaccacagttatccacaggaagttacccataacctccctcgtgcacccctacccccaATAACTTCGTATAGCATACATTATACGAAGTTATGGTACCggagtccctaactagacctgctgtctctccatagccctgtctacacct-3’ , where capitals indicate the LoxP sequence with a KpnI site , and lower case the homology arms . For embryo microinjection , the annealed sgRNA was complexed with Cas9 protein ( New England Biolabs ) at room temperature for 10 min , before addition of ssODN ( IDT ) donor ( final concentrations; sgRNA 20 ng µL−1 , Cas9 protein 20 ng µL−1 , ssODN 50 ng µL−1 ) . CRISPR reagents were directly microinjected into C57BL6/J ( Envigo ) zygote pronuclei using standard protocols ( Demayo et al . , 2012 ) . Zygotes were cultured overnight and the resulting two-cell embryos surgically implanted into the oviduct of day 0 . 5 post-coitum pseudopregnant mice . Potential founder mice were identified by extraction of genomic DNA from ear clips , followed by PCR using primers that flank the homology arms and sgRNA sites ( Geno F1 tcagatggcgaaccagaagtc and Geno R1 tacaatgtagtcaggtgtgacg ) . WT sequences produced a 833 bp band , and loxP knock in 873 bp , which is also susceptible to KpnI digest . Pups with a larger band were reserved , the band isolated and amplified using high fidelity Phusion polymerase ( NEB ) , gel extracted and subcloned into pCRblunt ( Invitrogen ) . Colonies were mini-prepped and Sanger sequenced with M13 Forward and Reverse primers , and aligned to predicted knock-in sequence . Positive pups were bred with a WT C57BL6/J to confirm germline transmission and a colony established . To integrate the 3’ LoxP we used sgRNA ACTACCCCATTACCTCTTGG-TGG with the ssODN repair template 5’- gagggccaaaactgtggaaagcaacacccttgaagtgtaggtggcccctgtgcaccagctctgtgtgtgactgcaaagcccccaccaagaATAACTTCGTATAGCATACATTATACGAAGTTATGGTACCggtaatggggtagttagacgggctgagggcagagcacttgtgtggctt-3’ . Again , capitals indicate the LoxP sequence with a KpnI site , and lower case the homology arms . For this second round of targeting , we generated embryos from the LRRC8A-5’fl colony by IVF using homozygous LRRC8A-5’fl mice , and used electroporation ( Nepa21 electroporator , Sonidel ) to deliver the sgRNA:Cas9 RNP complex and ssODN to the embryos , AltR crRNA:tracrRNA:Cas9 complex ( 200 ng μL−1; 200 ng μL−1; 200 ng μL−1 respectively ) and ssDNA HDR template ( 500 ng μL−1 ) ( Kaneko , 2017 ) . Zygotes were cultured overnight and the resulting two-cell embryos surgically implanted into the oviduct of day 0 . 5 post-coitum pseudopregnant mice . Potential founder mice were identified by extraction of genomic DNA from ear clips , followed by PCR using primers that flank the homology arms and sgRNA sites ( Geno F2 atccccactgcttttctgga and Geno R2 ccactcaagagccagcaatg ) . WT sequences produced a 371 bp band , and loxP knock in 411 bp , which is also susceptible to KpnI digest . As before , Pups with a larger band were reserved , the band isolated and amplified using high fidelity Phusion polymerase ( NEB ) , gel extracted and subcloned into pCRblunt ( Invitrogen ) . Colonies were mini-prepped and Sanger sequenced with M13 Forward and Reverse primers , and aligned to predicted knock-in sequence . Positive pups were bred with a WT C57BL6/J to confirm germline transmission and a colony established ( Lrrc8afl/fl ) . Lrrc8afl/fl mice were bred with Cx3cr1cre mice ( as previously described Yona et al . , 2013 ) to generate Lrrc8afl/fl Cx3cr1cre/+ mice , which specifically induce removal of LRRC8A in cells expressing CX3CR1 ( Lrrc8a KO ) . CX3CR1-cre mice were obtained from a breeding colony at the University of Manchester managed by John Grainger . Experiments using Lrrc8a KO cells were compared to wild-type littermate controls ( Lrrc8afl/fl Cx3cr1+/+ ) . All procedures were performed with appropriate personal and project licenses in place , in accordance with the Home Office ( Animals ) Scientific Procedures Act ( 1986 ) , and approved by the Home Office and the local Animal Ethical Review Group , University of Manchester . Eight to ten week-old male wild type ( WT ) C57BL/6J mice ( Charles River ) were used to test efficacy of NS3728 and MCC950 . Mice were treated intraperitoneally with LPS ( 2 µg mL−1 , 500 µL ) and either a vehicle control ( 5% DMSO ( v/v ) , 5% Cremophor ( v/v ) , 5% ethanol ( v/v ) in PBS ) , NS3728 ( 50 mg kg−1 ) or MCC950 ( 50 mg kg−1 ) for 4 hr . Mice were then anesthetised with isofluorane ( induced at 3% in 33% O2 , 67% NO2 , maintained at 1–2% ) before injection with a vehicle control , NS3728 or MCC950 as before and ATP ( 0 . 5 mL , 100 mM in PBS , pH 7 . 4 ) or PBS for 15 min . The peritoneum was then lavaged with RPMI 1640 ( 3 mL ) and blood was collected via cardiac puncture . For experiments using Lrrc8a knockout ( KO ) mice , 8- to 10-week-old male and female Lrrc8a KO and WT littermate controls were used and treated as described above . Plasma and lavage fluid was used for cytokine analysis . BCA analysis was performed on the peritoneal lavage fluid to normalise cytokine release to total protein level . Murine studies were performed with the researcher blinded to genotype and treatment for the duration of the experiment . Eight to ten-week-old male and female Lrrc8a KO and WT littermate controls were anesthetised with isofluorane ( induced at 3% in 33% O2 , 67% NO2 , maintained at 1–2% ) and the peritoneal cavity was lavaged with 6 mL RPMI ( 3% v/v FBS , 100 U mL−1 penicillin , 100 μg mL−1 streptomycin , 1 mM EDTA ) before sacrifice . Red blood cells were lysed using Pharmlyse ( BD Biosciences ) in H2O . Cells were surface stained with fluorescence-conjugated anti-CD45 , anti-CD11b , anti-Ly6G , anti-MHCII , anti-F4/80 , anti-Ly6C and anti-CX3CR1 antibody cocktail containing Fc block ( anti-CD16/CD32 ) and Tris-EDTA ( 1 mM ) . Cells were then fixed ( 10 min , room temperature ) with paraformaldehyde ( 2% w/v ) . Live/Dead Fixable Blue stain was used to exclude dead cells . Samples were analysed on an LSRII flow cytometer ( Becton-Dickinson ) and cell populations characterised as follows:- neutrophils ( CD45hi/CD11bhi/Ly6Ghi ) , monocyte-derived-macrophages ( CD45hi/CD11bhi/Ly6G-/MHCIIhi/F4/80- ) , resident macrophages ( CD45hi/CD11bhi/Ly6G-/F4/80hi ) , Ly6Chi monocytes ( CD45hi/CD11bhi/Ly6G-/MHCII-/F4/80-/CX3CR1hi/Ly6Chi ) , using FlowJo software . Data are presented as mean values plus the SEM . Accepted levels of significance were *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 . Statistical analyses were carried out using GraphPad Prism ( version 8 ) . Data where comparisons were made against a vehicle control , a one way ANOVA was performed with a Dunnett’s post hoc comparison was used . Experiments with two independent variables were analysed using a two-way ANOVA followed by a Tukey’s post hoc corrected analysis . Equal variance and normality were assessed with Levene’s test and the Shapiro–Wilk test , respectively , and appropriate transformations were applied when necessary . n represents experiments performed on individual animals or different passages for experiments involving HeLa cells . | Inflammation is a critical part of a healthy immune system , which protects us against harmful pathogens ( such as bacteria or viruses ) and works to restore damaged tissues . In the immune cells of our body , the inflammatory process can be activated through a group of inflammatory proteins that together are known as the NLRP3 inflammasome complex . While inflammation is a powerful mechanism that protects the human body , persistent or uncontrolled inflammation can cause serious , long-term damage . The inappropriate activation of the NLRP3 inflammasome has been implicated in several diseases , including Alzheimer’s disease , heart disease , and diabetes . The NLRP3 inflammasome can be activated by different stimuli , including changes in cell volume and exposure to either molecules produced by damaged cells or toxins from bacteria . However , the precise mechanism through which the NLRP3 becomes activated in response to these stimuli was not clear . The exit of chloride ions from immune cells is known to activate the NLRP3 inflammasome . Chloride ions exit the cell through proteins called anion channels , including volume-regulated anion channels ( VRACs ) , which respond to changes in cell volume . Green et al . have found that , in immune cells from mice grown in the lab called macrophages , VRACs are the only chloride channels involved in activating the NLRP3 inflammasome when the cell’s volume changes . However , when the macrophages are exposed to molecules produced by damaged cells or toxins from bacteria , Green et al . discovered that other previously unidentified chloride channels are involved in activating the NLRP3 inflammasome . These results suggest that it might be possible to develop drugs to prevent the activation of the NLRP3 inflammasome that selectively target specific sets of chloride channels depending on which stimuli are causing the inflammation . Such a selective approach would minimise the side effects associated with drugs that generically suppress all NLRP3 activity by directly binding to NLRP3 itself . Ultimately , this may help guide the development of new , targeted anti-inflammatory drugs that can help treat the symptoms of a variety of diseases in humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"immunology",
"and",
"inflammation"
] | 2020 | LRRC8A is essential for hypotonicity-, but not for DAMP-induced NLRP3 inflammasome activation |
As the closest unicellular relatives of animals , choanoflagellates serve as useful model organisms for understanding the evolution of animal multicellularity . An important factor in animal evolution was the increasing ocean oxygen levels in the Precambrian , which are thought to have influenced the emergence of complex multicellular life . As a first step in addressing these conditions , we study here the response of the colony-forming choanoflagellate Salpingoeca rosetta to oxygen gradients . Using a microfluidic device that allows spatio-temporal variations in oxygen concentrations , we report the discovery that S . rosetta displays positive aerotaxis . Analysis of the spatial population distributions provides evidence for logarithmic sensing of oxygen , which enhances sensing in low oxygen neighborhoods . Analysis of search strategy models on the experimental colony trajectories finds that choanoflagellate aerotaxis is consistent with stochastic navigation , the statistics of which are captured using an effective continuous version based on classical run-and-tumble chemotaxis .
Taxis , the physical migration towards preferred or away from undesired conditions , is a feature shared by virtually all motile organisms . Taxis comes in many forms , and in common is an underlying field of attractant ( or repellent ) and an ability to react and navigate along gradients of this field . Bacteria do chemotaxis towards nutrients ( Adler , 1969; Berg , 1993 ) and away from toxins ( Tso and Adler , 1974 ) . Algae do phototaxis towards light ( Yoshimura and Kamiya , 2001; Drescher et al . , 2010 ) and gyrotaxis along gravitational potentials ( Kessler , 1985 ) . Chemotaxis provides a mechanism for the recognition and attraction of gametes ( Vogel et al . , 1982 ) and for complex behavioural patterns such as in the slime mould Dictyostelium discoideum , where cAMP-driven chemotaxis is a critical part of the formation of the multicellular stage of the life cycle ( Bonner , 1947 ) . Aerotaxis , defined as oxygen-dependent migration , is well-characterized in bacteria ( Taylor et al . , 1999 ) , but is poorly studied in more complex organisms . This is despite the essentiality of oxygen for all aerobic life , and the important role that Precambrian oxygen levels played in the emergence and evolution of multicellular animal life ( Nursall , 1959 ) . One group of aquatic heterotrophic protists , the choanoflagellates , are of particular interest for the study of how multicellularity evolved . Choanoflagellates are a class of unicellular microorganisms that are the closest relatives of the animals ( Lang et al . , 2002 ) . This relationship was first proposed by James-Clark in 1866 ( James-Clark , 1866 ) , on the basis of the resemblance between choanoflagellates and the choanocytes of sponges . The sister relationship between choanoflagellates and animals was further confirmed in the genomic era by molecular evidence ( King et al . , 2008 ) . All choanoflagellates share the same basic unicell structure: a prolate cell body with a single beating flagellum that is surrounded by a collar of microvilli . The beating of the flagellum creates a current in the surrounding fluid that guides suspended prey such as bacteria through the collar ( Pettitt and Orme , 2002 ) where they can be caught and ultimately phagocytosed . The flagellar current also has the effect of causing the choanoflagellate cell to swim . The choanoflagellate Salpingoeca rosetta can form colonies through incomplete cytokinesis ( Fairclough et al . , 2010 ) . In the presence of certain bacteria ( Dayel et al . , 2011; Levin et al . , 2014 ) , these colonies have an eponymous rosette-like shape as shown in Figure 1 . The colony morphology is variable , and the constituent flagella beat independently of one another ( Kirkegaard et al . , 2016 ) . The random and independent flagellar motion argues against there being any coordination between cells in a colony , and as yet no evidence of any form of taxis for choanoflagellate colonies has been reported . 10 . 7554/eLife . 18109 . 003Figure 1 . Micrograph of S . rosetta colonies ( left ) with schematic illustration ( right , collars in blue ) . Scale bar: 50 µm . Cell body diameters are ~5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 003 The geometry , flagella independence and lack of taxis observed in S . rosetta colonies contrast with other lineages , such as the Volvocales , a group of green algae ( Goldstein , 2015 ) . Phototaxis is clearly observable in both unicellular ( Chlamydomonas ) ( Yoshimura and Kamiya , 2001 ) and colonial ( Volvox ) ( Drescher et al . , 2010 ) species , in order to maintain optimum light levels for photosynthesis . Volvocalean phototaxis is deterministic , requiring precise tuning between the internal biochemical timescales and the rotation period of the organism as a whole . Although S . rosetta colonies also rotate around an internal axis , due to the variable colony morphology and the independent beating of the individual flagella , this rotation rate will itself be random ( Kirkegaard et al . , 2016 ) , rendering a strategy similar to that of the Volvocales unlikely in S . rosetta . An alternative strategy is stochastic taxis , sometimes referred to as kinesis . The classic example of stochastic taxis is the run-and-tumble chemotaxis of certain peritrichous bacteria ( Berg , 1993 ) . By spinning their left-handed helical flagella in different directions , such bacteria can alternate between swimming in straight lines ( running ) and randomly reorienting themselves ( tumbling ) . Through biasing tumbles to be less frequent when going up the gradient , the bacteria exhibit biased motion towards a chemoattractant without directly steering towards it ( Berg , 1993 ) . Here , we study S . rosetta and show that it exhibits aerotaxis , i . e . navigation along gradients of oxygen . We further examine and statistically analyse aerotaxis of S . rosetta colonies under spatio-temporal variations of oxygen at the level of total colony populations and at the level of the trajectories of individual colonies . From these experiments we establish two key features of the aerotactic response of choanoflagellates: they employ a stochastic reorientation search strategy and the sensing of oxygen concentration gradients is logarithmic . Finally , we render these results quantitative through the use of mathematical analysis of a modified Keller-Segel model ( Keller and Segel , 1971 ) .
The study of aerotaxis in bacteria has led to numerous methods for creating spatial oxygen gradients ( Shioi et al . , 1987; Wong et al . , 1995; Zhulin et al . , 1996; Taylor et al . , 1999 ) , one of which is the exploitation of soft lithography techniques ( Adler et al . , 2012; Rusconi et al . , 2014 ) . Since PDMS , the most commonly used material for microfluidic chambers , is permeable to gases , gas channels can be introduced in the devices to allow gaseous species to diffuse into the fluid . For example , an oxygen gradient can be created using a source channel flowing with normal air and a sink channel flowing with pure nitrogen . Our device , shown schematically in Figure 2 , is a modified version of that used by Adler , et al . ( Adler et al . , 2012 ) . Viewed from above , the sample channel ( yellow ) consists of a wide observation chamber with thin inlet and outlet channels . The outlet leads to a serpentine channel that hinders bulk fluid flows . On each side of the sample channel are gas channels , the inlets of which are connected to a valve system allowing for the flow of air ( 20% oxygen ) and nitrogen . The flow of air and nitrogen can be conducted in any combination and configuration , e . g . oxygen in one channel and nitrogen in the other , and can be easily swapped over . The PDMS chamber is plasma etched to a glass slide , and an extra glass slide is etched on top of the device , preventing air from diffusing in from the surrounding environment . 10 . 7554/eLife . 18109 . 004Figure 2 . Microfluidic device . ( a ) Top view of the device . The sample channel ( yellow ) is loaded with culture and observed in the middle chamber . The side channels ( red , blue ) are gas channels in which oxygen and nitrogen may be flown . Scale bar: 10 mm . ( b ) Side view of the device . PDMS is plasma etched to a glass slide , and a cover slip is plasma etched on top , centered on the imaging chamber , also shown in ( a ) . Thickness of the channels are ≈115 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 004 Experiments were carried out immediately after plasma etching as the permeability of gases slowly decreases thereafter . Cultures of S . rosetta were introduced at the inlet of the device , and both the inlet and outlet were then closed to prevent evaporative flows . A gradient of oxygen was set up by having air flowing in one of the side channels and nitrogen in the other . Our main experimental result , shown in Figure 3a , b , is the observation that S . rosetta colonies accumulate at the oxygen-rich side and away from the oxygen-poor side , i . e . that they are aerotactic . We also found aerotaxis in the unicellular fast swimmer form ( Dayel et al . , 2011 ) of S . rosetta , showing that this is not an exclusive phenomenon to colonies . 10 . 7554/eLife . 18109 . 005Figure 3 . Aerotaxis of S . rosetta colonies . ( a–b ) Micrographs near an oxygen-rich wall at twice the resolution of that used in the density experiments . Scale bar: 50 μµm ( a ) Colonies approach a wall where the oxygen-concentration is high . ( b ) Colonies staying near this wall . ( c ) Density evolution of S . rosetta during experiment . At each time step the distribution is normalized to a probability distribution [colorbar units in µm−1] . Colors on the side indicate what gas is flowing in that side channel , red for oxygen and blue for nitrogen . Ncolonies~150 , concentration ~5 · 106 mL−1 . ( d ) Keller-Segel model with log-concentration input given by Equation 4 , D = 865 µm2/s , α = 1850 µm , vdrift = 5 . 2 µm/s . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 00510 . 7554/eLife . 18109 . 006Figure 3—figure supplement 1 . Seperate aerotaxis experiment . ( a ) Density evolution of S . rosetta during experiment [colorbar units in µm−1] . Ncolonies∼60 , concentration ∼2⋅106mL−1 . ( b ) Keller-Segel model with linear concentration input 𝒱[c]=vdrifttanh ( αcy ) , with α=8 . 2 , and otherwise same parameters as in main text Figure 3 . ( c ) Keller-Segel model with log concentration input , same parameters as in main text Figure 3 . Experiment and simulation was started in steady-state configuration of oxygen down and nitrogen up . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 006 With the present microfluidic device we can explore more details of choanoflagellate aerotaxis by dynamically changing the oxygen boundary conditions , for instance by flipping the gradient direction or by removing all oxygen influx after a uniform distribution has been reached . Figure 3c shows the result of such a dynamic experiment over the course of ∼3 . 5 hr . The density is normalized for each frame and the noise present is partly due to colonies missing in the tracking in some frames . Many repetitions of the experiment show that the behaviour in Figure 3c is highly repeatable and robust to changes in the details of the cycling protocol ( See Figure 3—figure supplement 1 ) . For consistency the figures in the main text are based on this specific protocol . Whenever one gas channel contains oxygen and the other nitrogen , the colonies swim towards the oxygen-rich side as further shown in Video 1 . In the time after a gas channel swap , the slope of the maximum density reveals the ensemble drift velocity vdrift . When there is oxygen in both gas channels , we observe that the density reaches an approximately uniform distribution within the time frame of the experiment . For periods in which nitrogen flows in both channels , this is not the case . Under these experimental conditions , the colonies accumulate in the middle of the chamber , where there is still some residual oxygen , as further shown in Video 2 . The fact that in this nitrogen-only configuration the colonies accumulate mid-chamber shows that accumulation does not depend on the presence of a nearby surface . With only nitrogen flowing , eventually there will be no oxygen gradient . Nonetheless , we observed the colonies to stay in the middle of chamber even after 90 min ( see Figure 3—figure supplement 1 ) . At that time , the highest oxygen levels are estimated by the diffusion equation to be less than ~0 . 2% . 10 . 7554/eLife . 18109 . 007Video 1 . Experimental videos of aerotaxis ( top ) and oxygen gas simulation ( bottom ) ( as in Figure 6 ) . Experimental videos are colored by the output of the gas simulation . Colonies migrating from one side to other after a swap of nitrogen and air ( 138–148 min . in experiment of Figure 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 00710 . 7554/eLife . 18109 . 008Video 2 . Colonies migrating from the middle to the sides after a change from nitrogen only in the two gas channels to air only ( 45–55 min in experiment of Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 008 This contrasting behaviour between the oxygen-only and nitrogen-only configurations suggests an asymmetry or non-linearity in the aerotactic response . If the response to oxygen concentration had been linear , the observation of the density band in the nitrogen-only section would imply similar density bands at the chamber edges in the oxygen-only section , which is not observed . Instead one might hypthesize that the colonies navigate along relative ( ∇c/c ) instead of absolute ( ∇c ) gradients , i . e . reacting to gradients that are comparable in magnitude to the background concentration . This is also known as logarithmic sensing , and we will confirm in the modelling section that this hypothesis can quantitatively explain the experiments . Strategies of taxis can be categorized into two main classes: deterministic and stochastic . In both strategies the swimming organism measures the attractant gradient ( for small organisms by some temporal filter [Block et al . , 1982; Celani and Vergassola , 2010] ) . A deterministic strategy , then , is one in which the organism directly steers towards the attractant , such as seen in sperm cells that modulate their flagellar beat to adjust directly the curvature and torsion of its swimming path in the gradient direction ( Friedrich and Jülicher , 2007; Jikeli et al . , 2015 ) . Contrasting is a stochastic strategy such as bacterial run-and-tumble locomotion ( Berg , 1993 ) , where modulation of the frequency of random reorientations biases the motion in the gradient direction without directly steering towards it . One simple method of taxis results from an organism swimming faster when it is moving up the gradient , creating an overall bias towards the attractant . With the detailed colony-tracking in the present study it is possible to test whether this mechanism is in operation with S . rosetta . Figure 4—figure supplement 1 shows examples of tracks during periods of uniform swimming ( t = 70 min ) and after a gas channel swap ( t = 142 min ) . Figure 4 shows the evolution of the mean colony swimming speed v ( green ) as well as the component velocities vx ( yellow ) and vy ( purple ) , averaged over ~150 colonies in each frame . For most times , the component velocities average to zero , but after a gas channel swap the y-component peaks . The ensemble average swimming speed in these sections , however , does not show an increase , suggesting that a velocity modulation is not the method of taxis . To quantify this further , the inset of Figure 4 shows the swimming speed in these sections plotted against the alignment to the gradient c^vy/v where c^=±1 signifies the direction of the gradient . The plot shows a very small ( ~3% ) change in swimming speed going up the gradient . Velocity-biased taxis can be described by v ( t ) =v ( p^ ) p^ , where e . g . v ( p^ ) =v ( 1+γp^⋅∇c/|∇c| ) , γ being the velocity-modulation taxis parameter . p^ , the direction of swimming , is unbiased by the attractant field c and evolves by rotational diffusion . To obtain a drift velocity ~1/3 of the swimming velocity , as we find for S . rosetta in the following section , the velocity modulation would have to be γ= 2/3 for a two-dimensional swimmer and γ = 1 in three dimensions , much larger than the ~3% observed . We conclude that the primary mechanism of aerotaxis in S . rosetta is therefore not a modulation of swimming speed . 10 . 7554/eLife . 18109 . 009Figure 4 . Running mean velocity statistics , showing that the primary mechanism of aerotaxis is not by modulation of swimming speed . Evolution of mean speed ( green , right axis ) and velocity in the x-direction ( yellow , left axis ) and y-direction ( purple , left axis ) , y being along the gradient of oxygen . Left and right axes have equal ranges . Side bars indicate gas flowing , oxygen ( red ) or nitrogen ( blue ) . The peaks of vy do not quite reach the true drift velocity due to smoothing of the curves . Inset shows the speed as function of alignment with the gas gradient c^vy/v at times after a swap . c^=1 if the gradient is up and = −1 if down . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 00910 . 7554/eLife . 18109 . 010Figure 4—figure supplement 1 . Example tracks . Trajectories are from the experiment of the main text . ( a ) t = 70 min . Trajectories at uniform oxygen concentration . ( b ) t = 142 min . Migration downwards after a swap to oxygen in the lower channel . Tracks were obtained by the algorithm described in methods . Scale bars: 150 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 010 S . rosetta colonies swim along noisy helical paths , and each colony displays distinct helix parameters ( Kirkegaard et al . , 2016 ) . To perform any kind of statistical angle analysis , we consider ensemble average quantities: the speed v and rotational diffusion dr , and average helix rotations out . Figure 5a shows the angular distribution data during the swaps , where , for the purposes of displaying all the data in a single graph , we have let θ→−θ for times when the oxygen gradient were pointing down . This distribution favors the up-direction θ=π/2 . More interesting is the distribution of reorientations . For this we define the angle turned by a colony in a time Δt as Δϕ=|θ ( t+Δt ) -π/2|-|θ ( t ) -π/2| such that it is positive if the turn is in the direction of the gradient and negative otherwise , and choose Δt low enough that −π<Δϕ<π . Figure 5b shows this distribution . The distribution is centred on zero , revealing that the colonies do as many turns in the wrong direction as in the correct direction . This indicates that the colonies navigate by a stochastic strategy . 10 . 7554/eLife . 18109 . 011Figure 5 . Angle statistics . Experimental data in grey bars . Deterministic model in purple and stochastic in green . ( a ) Distribution of θ . θ=π/2 is along the gradient . ( b ) Change in angle Δϕ for Δt = 0 . 65 s . Positive change corresponds to a turn towards the gradient . Deterministic parameters: ϵd=0 . 28s-1 , dr=0 . 52s-1 . Stochastic parameters: ϵs=0 . 55 , dr=0 . 33s-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 011 We study the spatio-temporal evolution of the choanoflagellate colony population within the observation chamber with the Keller-Segel model ( Keller and Segel , 1971 ) , which has broad applicability for taxis ( Tindall et al . , 2008 ) . In describing the phenomenon of aerotaxis the two quantities of interest are the colony population density ρ ( x , t ) , and the oxygen concentration c ( x , t ) . The former obeys the Keller-Segel equation ( 1 ) ∂ρ∂t=D∇2ρ−∇⋅ ( 𝒱[c]ρ ) and the latter follows the diffusion equation , ( 2 ) ∂c∂t=∇⋅ ( Dc∇c ) . The functional 𝒱 specifies the population drift velocity’s dependency on the local oxygen concentration , i . e . the drift of cells due to taxis . For a fixed response , the functional would equal a constant 𝒱=vdrift . D and Dc ( x ) are the colony and oxygen diffusion constants , the latter of which varies with position inside the microfluidic device , with values Dc , PDMS=3 . 55×10−3mm2/s in PDMS ( Cox and Dunn , 1986 ) and Dc , water=2 . 10×10−3mm2/s in water ( Cussler , 1997 ) . Using in-house software , we solve Equation 2 on a cross-section of the microfluidic device with time-dependent boundary conditions corresponding to the experimental protocols . Gas channels with oxygen flowing have the condition c=20% ( for theory the unit of oxygen is not important and we find percentage to be the most intuitive measure ) . For channels with nitrogen flowing c=0 . Glass interfaces have no-flux conditions n^⋅∇c=0 , where n^ is the surface normal . A snapshot from the numerical studies is shown in Figure 6a at a time following a swap of nitrogen and oxygen channels , thus showing residual oxygen above the channels now filled with nitrogen and vice versa . The simulation can now be evaluated at the position of the observation chamber . Note that the no-flux conditions at the glass interface render the concentration gradients in the z-direction very small , so the precise height of evaluation is not significant . The simulation with boundary conditions corresponding to those in the experiment of Figure 3c is shown in Figure 6b . To a very good approximation the concentration field is constant along the x-direction . Figure 6c shows the oxygen field evaluated at y = −250 µm ( red curve ) . Neglecting the consumption of oxygen by the colonies , these results provide the input concentration field c for the Keller-Segel model . 10 . 7554/eLife . 18109 . 012Figure 6 . Simulation of oxygen concentration in microfluidic device . ( a ) Simulation of 2D cross-section of the device . Oxygen concentration boundary conditions are imposed at the gas channel positions . Snapshot shows t = 110 min , ~1 . 5 min after the swap . White line indicates evaluation location at the observation chamber . ( b ) Evolution of oxygen concentration at z = 100 µm . ( c ) Simulation at y = −250 µm . Oxygen percentage in red ( left axis ) , and spatial gradient in purple normalized to fit in [−1 , 1] ( right axis ) , response function tanh ( α∇c ( x , t ) /c ( x , t ) ) in green ( right axis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 012 The simplest and most widely-used response functional in the Keller-Segel model is linear in spatial gradients , 𝒱=β∇c ( x , t ) , where β is termed the taxis coefficient . Such a response can , however , reach unrealistic drift velocities , i . e . higher than the swimming velocity , if the oxygen gradients are large . This defect can be eliminated by various functional forms ( Tindall et al . , 2008 ) , such as the choice ( 3 ) 𝒱[c]=vdrifttanh ( α|∇c| ) ∇c|∇c| , where vdrift must be smaller than the swimming velocity . This choice behaves linearly for small gradients , but tends asymptotically to a maximum value for large gradients . Most of the behaviour of Figure 3c can be explained by this model ( See Figure 3—figure supplement 1 ) , but not the section where nitrogen is flowing in both side channels . The reason is the aforementioned asymmetry between the sensing-signalling in low versus high oxygen concentrations , which we argued may be explained by relative gradient sensing , also known as logarithmic sensing since ∇logc=∇c/c . To examine quantitatively this hypothesis we consider ( 4 ) 𝒱[c]=vdrifttanh ( α|∇cc| ) ∇c|∇c| . For unidirectional gradients ( say , in the y-direction ) this expression reduces to 𝒱[c]=vdrifttanh ( αcy/c ) y^ , where cy=∂c/∂y . Since the logarithmic sensing cannot be maintained at infinitely small concentrations , there must naturally be some lower cutoff to this expression in absolute concentration levels . Nonetheless , from a modelling perspective we can ignore this for the present experiments with nitrogen-only sections lasting less than 1 . 5 hr as discussed . Figure 6c shows that the spatial oxygen gradient cy ( purple curve ) compares to y^⋅𝒱[c] of Equation 4 ( green curve ) at all times except in the nitrogen-only section , where the log-response function does not decay towards zero . A positive value of the response function means a positive ( y ) drift velocity . The Keller-Segel equation with log-sensing is able to explain all sections of the experiment , as demonstrated in Figure 3d . The parameters obtained from a numerical fit include the drift velocity vdrift = 5 . 2 µm/s ( which should be compared to the ensemble average swimming speed v = 16 . 5 µm/s , the speed averaged over all colonies ) and the diffusion constant D = 865 µm2/s . Using the ensemble-averaged speed , the diffusion constant can be related to an effective rotational diffusion constant Dr=v2/2D=0 . 16s−1 . The parameter vdrift represents the coupling between the oxygen-gradient response and the resulting population drift , but it is a purely phenomenological quantity in which the underlying microscopic mechanism of aerotaxis is hidden . To explain the origin of vdrift we must consider the navigation strategy . To distinguish deterministic and stochastic strategies we introduce two effective models and in the following consider them in the context of a constant oxygen gradient along the y-axis , but the generalization is immediate . We furthermore ignore translational diffusion due to thermal noise , since this contribution is orders of magnitude smaller than that of active diffusion . Thus in a quasi-2D system , an organism’s path is described by ( 5 ) dx= ( cosθ ( t ) sinθ ( t ) ) vdt , where θ is the instantaneous swimming direction , and we choose motion along the positive y-axis ( θ=π/2 ) to be toward the attractant . In the deterministic model , the organisms actively steer towards the gradient . We model this with the Langevin equation ( 6 ) dθ=ϵdcosθdt+2drdW ( t ) , where W ( t ) is a Wiener process for which ⟨dW ( t ) dW ( t′ ) ⟩=δ ( t−t′ ) . This process is illustrated in Figure 7a , yellow arrows showing ϵdcosθΔt and purple arrows 2drΔW . For the stochastic model we take a ‘continuous’ version of run-and-tumble , in which the rotational diffusion is modulated ( 7 ) dθ=2dr ( 1−ϵssinθ ) dW ( t ) , 10 . 7554/eLife . 18109 . 013Figure 7 . Illustration of deterministic and stochastic strategies based on discretised simulations with exaggerated steps . Time evolves from left to right . Orientations , shown by green arrows , are trying to align to up-motion , θ=π/2 , indicated by red ( oxygen ) at the top and blue ( nitrogen ) at the bottom . ( a ) : Deterministic strategy , described by Equation 6 . Deterministic part in yellow and stochastic part in purple . The deterministic part is always in the correct direction . ( b ) : Stochastic strategies , described by Equation 7 . All steps are stochastic , but largest when furthest away from θ=π/2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18109 . 013 where −1≤ϵs≤1 and the multiplicative noise is interpreted in the Itō sense . This process is illustrated in Figure 7b . In both models , ϵ is an effective ensemble average response . To compare to the experiments , ϵ should be replaced by a function coupled to the gas concentration field through 𝒱[c] , thus coupling to Equation 5 . The steady state gradient in the experiment is approximately linear , and thus as a first approximation Equations 6 , 7 should describe the angle statistics in the time between the swap of gas channels and reaching the opposite side . In our experiments , after a flip in the oxygen gradient direction , the colonies reach the opposite wall in a time comparable to that for oxygen to diffuse across the chamber . Thus , a true steady state is not reached , but during the intermediate times a steady state approximation is accurate . Solving the Fokker-Planck equations corresponding to the systems Equations 6 and 7 in steady-state , we obtain theoretical angle distributions ( see Materials and methods ) . Both models are able to describe the data in Figure 5a well , with the deterministic model ( purple ) fitting the down-gradient swimming best , and the stochastic model ( green ) fitting the up-gradient swimming best . Both fits involve a single parameter , kd=ϵd/2dr in the deterministic model and ϵs in the stochastic one . We now move beyond steady-state distributions to examine the detailed statistics of the trajectories themselves . We recall our definition of the angle turned by a colony in a time Δt as Δϕ=|θ ( t+Δt ) −π/2|−|θ ( t ) −π/2| such that it is positive if the turn is in the direction of the gradient and negative otherwise . The Equations ( 6 ) and ( 7 ) imply distributions of Δϕ ( see Materials and methods ) . Figure 5b shows the best fit of both models to the data . For the stochastic model ( green ) ϵs is known and the fit is in dr and matches well . The deterministic ( purple ) is constrained by kd=ϵd/2dr , and the fit can be done in dr as well . We see clearly that the deterministic model does not provide a satisfactory fit to the data . In detail , the value of ϵd needed to fit the data in Figure 5a shifts the mean of pd ( Δϕ ) in the positive direction . This result persists with any amount of smoothing applied to the data , averaging out active rotations . We thus conclude that the colonies navigate by a stochastic strategy , and that the ensemble angle statistics can be captured by this simple model . The navigation strategy model must be consistent with the Keller-Segel population dynamics model . Having shown that the data favor a stochastic model , we may now couple Equation 7 to Equation 5 and let the Fokker-Planck equation ( Materials and methods ) replace the Keller-Segel model . Such an approach leads to similar results as Figure 3d . Furthermore , we now recognize that the Keller-Segel model is a quasi-stationary approximation and we can calculate the stationary-approximation drift velocity ( 8 ) vdrift=v⟨sinθ⟩=v[1/ϵs−1/ϵs2−1] . In other words ϵs= ( 2vvdrift ) / ( v2+vdrift2 ) is the ratio of the squared geometric mean to the quadratic mean of the average and drift velocities . For the fitted ϵs=0 . 55 , vdrift≈0 . 3v≈5 µm/s , consistent with the fitted value in the Keller-Segel model .
We have shown that colonies of S . rosetta can navigate along gradients of oxygen , thus exhibiting positive aerotaxis . The cells navigate along relative oxygen gradients and the navigation strategy is stochastic in nature , achieved by modulating not speed but direction of swimming . The experimental observation that choanoflagellates are aerotactic raises a number of questions . One concerns the actual sensing mechanism , and how this compares to those of animals , given the close evolutionary relationship between choanoflagellates and metazoans . In animals , oxygen concentrations can be sensed by the highly-conserved hypoxia-inducible factor ( HIF ) transcription factor pathway ( Loenarz et al . , 2011; Kaelin and Ratcliffe , 2008; Rytkönen et al . , 2011 ) . At normal oxygen levels , the activity of specific prolyl-hydroxylase ( PHD ) enzymes labels the HIF protein complexes for degradation . At low oxygen levels , however , PHD activity is inhibited , leading to elevated HIF levels . The transcription factor activity of HIF up-regulates expression of genes involved in hypoxia response ( e . g . glycolysis enzymes ) for survival in low-oxygen conditions ( Greer et al . , 2012 ) . The genes involved in HIF signalling are widespread in metazoans , with evidence suggesting that some components of this pathway are descended from prokaryotic ancestors ( Scotti et al . , 2014 ) . Preliminary experiments involving exposing S . rosetta cultures to DMOG ( an inhibitor of the prolyl-hydroxylase step in the HIF pathway [Fong and Takeda , 2008] ) had no observable effect on aerotaxis within our experimental system ( data not shown ) . This is perhaps unsurprising , given that choanoflagellates lack some important components of the HIF pathway ( Rytkönen et al . , 2011 ) . Additionally , the aerotactic response observed here is acute , within a timeframe on the order of seconds to minutes , rather than the being a longer-term response to lowered oxygen levels by modulation of gene expression . In certain aerotactic bacteria , it is known that oxygen concentrations are measured indirectly via energy-sensing , which can then influence the rotation of the bacterial flagella between running and tumbling ( Taylor et al . , 1999 ) . An analogous mechanism may be at play in choanoflagellates , possibly via reactive oxygen species ( Cash et al . , 2007 ) and oxygen-sensitive ion channels ( Lahiri et al . , 2006; Ward , 2008 ) modulating the beating rate of the flagella of each S . rosetta cell . Thus the question of the molecular and cellular mechanisms underpinning aerotaxis in choanoflagellates is a promising avenue for further research . Oxygen is implicated as having an important influence on animal evolution ( Nursall , 1959; Lenton et al . , 2014 ) , with some hypotheses that the emergence of complex animal ecosystems ( Sperling et al . , 2013 ) were only triggered when oxygen levels rose above certain thresholds . Sponges , believed to be the most basal animal group , were recently shown to have very low oxygen requirements , disputing the importance of oxygen in early animal evolution ( Mills et al . , 2014 ) . Our results raise the possibility that the common ancestor from which choanoflagellates and animals evolved was aerotactic , and that oxygen sensing and responding have thus been under strong selection throughout the holozoans . If the ancestors of animals and choanoflagellates were strongly aerotactic , this would be indicative of the importance of oxygen as a resource during the Precambrian ( Lyons et al . , 2014 ) . Equally , it could be the case that choanoflagellates themselves have evolved aerotaxis after the split from the animal stem lineage . To answer this question , the oxygen requirements and aerotactic capabilities of other opisthokont groups , e . g . ichthyosporeans , filasterians ( Ruiz-Trillo et al . , 2008; Sebé-Pedrós et al . , 2013 ) , as well as basal animals such as sponges , ctenophores and placozoans ( Dunn et al . , 2008; Ryan et al . , 2013; Pisani et al . , 2015; Whelan et al . , 2015 ) , requires investigation . Such an analysis can be further complimented by determining the oxygen sensing and signalling mechanisms across the opisthokont phylogeny . If aerotaxis was key to the ancestral unicellular holozoan , it is crucial that the evolution to multicellularity did not hinder this ability . Our experimental results show that both the unicellular and colonial morphotypes of S . rosetta can perform efficient aerotaxis and navigation in general , despite lacking coordination between the constituent cells of the colony ( Kirkegaard et al . , 2016 ) . Therefore , an evolutionary transition to a multicellularity resembling the unicellular-to-colonial transition in S . rosetta ( Dayel et al . , 2011 ) would not require additional cell-cell communication mechanisms to coordinate navigation . Not only would this allow aerotaxis towards oxygen , but also taxis in response to other stimulants , such as bacterial signals ( Woznica et al . , 2016 ) . This is a particular feature of the stochastic navigation strategy , which works equally well for both single cells and colonies formed from multiple units of the same cells . We have described the stochastic strategy of S . rosetta in terms of an effective model . The effective bias parameter ϵs is the result of flagella modulation . Flagella can be imaged on colonies stuck to the microscope slide ( Kirkegaard et al . , 2016 ) , but measuring directly the flagella modulation is challenging , since the oxygen changes that can be induced in a microfluidic device are on a time scale of minutes , whereas the flagella beating is on a time scale of tens of milliseconds ( Kirkegaard et al . , 2016 ) . The swimming trajectories suggest that this modulation is a mixture of many types , ranging from slow to vigorous , as also seen in other organisms ( Jikeli et al . , 2015 ) , but direct imaging is needed to make more quantitative statements . We have shown that in spatio-temporal varying environments , considerations of population dynamics can distinguish between linear- and logarithmic-sensing mechanisms , and concluded that choanoflagellates do logarithmic sensing . The fact that conclusions of logarithmic sensing can be made from analysis of the population dynamics alone shows that individual tracks are not needed and thus allows for such analysis in dense experiments . Logarithmic sensing is a key attribute for survival: it allows sensing of and reacting to gradients in very low concentration environments , while still being able to effectively navigate along large gradients . Logarithmic sensing has also been experimentally observed for other species , e . g . bacteria ( Mesibov et al . , 1973; Kalinin et al . , 2009 ) . With logarithmic sensing , cells only navigate along oxygen gradients that are significant compared to the absolute concentration . This is a well-known phenomenon also from human behaviour , where , for instance , dim lights are seen only when it is dark and weak sounds heard only when it is quiet . It is known as Weber’s law ( Ross , 1996 ) , which states that the magnitude of just-noticeable differences of a stimuli is proportional to the stimuli magnitude itself . This is closely related to the Weber-Fechner law , stating that stimuli magnitude grows logarithmically with the actual signal , which we have found to be in agreement with the experimental observation of S . rosetta aerotaxis . For microorganisms , a Weber lower limit can be understood , at least partly , as a physical limitation . Because of thermal fluctuations , the error on any concentration gradient measurement increases with the local absolute concentration ( Berg and Purcell , 1977; Endres and Wingreen , 2008 ) , and thus it immediately follows that the limit of just-noticeable gradients must decrease with absolute concentration . Quantitatively , the noise scales as ∼c1/2 in the absolute concentration ( Endres and Wingreen , 2008 ) . Such a scaling is not observed for bacteria’s just-noticeable limits , where instead Weber’s law hold ( Mesibov et al . , 1973 ) . Sensor adaptation enables the bacteria to have the linear scaling ∼c ( Sourjik and Wingreen , 2012 ) leading to logarithmic sensing and high dynamic range , but it is nonetheless the case that for purely physical reasons a lower limit scaling with concentration must be present . The fitting of the Keller-Segel model to the population data were optimal precisely for relative gradient sensing . Sensing functionals such as ∇c/c or ∇c/c2 only decreased the fit quality . This implies that Weber’s law and logarithmic sensing , at least to a good approximation , is occurring in chaonoflagellates . This scaling could be further studied by measuring the aerotactic response to an order-of-magnitude range of concentration gradients under a range of absolute concentrations . Precision control of gas mixtures would allow the extraction of any biological deviations from Weber’s law and potentially reveal a transition to the physical limit of c scaling for very low concentrations , i . e . at the limit of sensing .
S . rosetta were cultured polyxenicly in artificial seawater ( 36 . 5 g/L Marin Salts [Tropic Marin , Germany] ) with organic enrichment ( 4 g/L Proteose Peptone [Sigma-Aldrich , USA] , 0 . 8 g/L Yeast Extract [Fluka Biochemika] ) at 15 µl/mLμ , and grown at 23°C , split weekly . Cultures were centrifuged to reach the high concentrations , ∼5×106ml−1 , used in experiments . Microfluidic devices were manufactured using standard soft-lithography techniques . The master was produced by spinning SU8-2075 ( MicroChem , USA ) at 1200 rpm to a thickness of ~115 µm . Chambers were cast in PDMS , Sylgard 184 ( Dow Corning , USA ) , and plasma etched to the glass slides . Cultures were concentrated by centrifugation before loaded into the device . Gas cylinders containing pure nitrogen and air ( 20% oxygen ) were connected via a system of valves to the gas channels of the device . Experiments were filmed ( Imaging Source , Germany ) in bright field at 10 fps on an inverted Zeiss LSM 700 Microscope . To track colonies , we first generated a running-median video , where each pixel in each frame of the experimental video is the median of that pixel taken over the neighbouring ~2 . 5 s of video . This method extracts an estimated background , i . e . a video without the colonies present . Subtracting this from the original video , the resulting video contains only the colonies and noise . Band-pass filters were used to remove the noise , and finally the colony positions were found by locating local maxima in the Gaussian filtered video . Density distributions were estimated by first calculating ( 9 ) η ( y ) =∑iexp ( − ( yi−y ) 2/2σ2 ) ∫y0y1exp ( − ( y′−y ) 2/2σ2 ) dy′ , where yi are the tracked positions , σ a standard deviation of separation , and the denominator adjusts for boundary effects . Hereafter ρ ( y ) =η ( y ) /∫y0y1η ( y′ ) dy′ is the normalized density . For velocity and angle statistics , the tracked positions were linked solely by proximity . If the image analysis algorithm failed to identify a given colony over fewer than three successive frames the integrity of the track was preserved by keeping a running memory . After the trajectories were obtained , spurious trajectories less than three frames in length were removed . Examples of trajectories are shown in Figure 4—figure supplement 1 . The final tracks contain ~150 trajectories in each frame , varying slightly over the course of the experiment due to loss of colonies in the tracking and swimming in and out of observation chamber . Oxygen diffusion and Keller-Segel equations were numerically solved using in-house software with finite difference spatial discretisation and implicit time-stepping . Given the stochastic dynamics ( 6 ) for individual particles following a deterministic strategy , the probability distribution function pd ( θ , t ) for the population obeys the Fokker-Planck equation ( 10 ) ∂pd∂t=dr∂2pd∂θ2−ϵd∂∂θ ( cos ( θ ) pd ) . The steady state distribution is found to be a von-Mises distribution ( 11 ) pd ( θ ) =12πI0 ( ϵd/2dr ) exp ( ϵdsinθ2dr ) , where I0 is the modified Bessel function of order zero . On the other hand , for the Fokker-Planck equation for the stochastic model , ( 12 ) ∂ps∂t=∂2∂θ2 ( dr ( 1−ϵssinθ ) ps ) , we find the steady-state distribution ( 13 ) ps ( θ ) =12π1-ϵs21-ϵssinθ . In the time right before a channel swap , we have p ( θ ) ≈1/2π , since the colonies stay near the wall . In the deterministic model , for small Δt , Δϕ=|θ ( t+Δt ) −π/2|−|θ ( t ) −π/2| is composed of deterministic δ=ϵdcos ( θ ) Δt and stochastic ξ=2drΔW , where ΔW∼Δt . Since we are assuming p ( θ ) =1/2π , we have p ( δ ) =1/π ( ϵdΔt ) 2−δ2 for δ∈ ( −ϵdΔt;ϵdΔt ) . The distribution of Δϕ=|δ|+ξ is then found as the convolution ( 14 ) pd ( Δϕ ) =∫0ϵdΔtexp ( − ( Δϕ−δ ) 2/4drΔt ) π3drΔt[ ( ϵdΔt ) 2−δ2]dδ , which can be evaluated numerically by Gaussian quadrature . In the stochastic model there is only ξ=2dr ( 1−ϵssinθ ) ΔW , but this is conditional on θ . We marginalize for the final distribution ( 15 ) ps ( Δϕ ) =∫−ππexp ( −Δϕ2/4dr ( 1−ϵssinθ ) Δt ) 16π3dr ( 1−ϵssinθ ) Δtdθ . | Most animals are made up of millions of cells , yet all animals evolved from ancestors that spent their whole lives as single cells . Today the closest single-celled relatives of animals are a group of aquatic organisms called choanoflagellates . Certain species of choanoflagellates can also form swimming colonies . This kind of multicellularity might resemble that seen in the earliest of animals . As such , studies into modern-day choanoflagellates can give insights into how the first animals to evolve might have behaved . Many organisms can find their way towards favorable areas using different strategies . For instance , bacteria can bias their tumbling to gradually swim towards food , and algae can turn and move directly towards light . While choanoflagellates require oxygen , it was not known if they could also actively navigate towards it , or any other resource . Now , Kirkegaard et al . find that the choanoflagellate Salpingoeca rosetta can indeed navigate towards oxygen – an ability called aerotaxis . This was true for both individual cells and for colonies made up of many cells . This discovery suggests that the transition from living as a single cell to living as a simple multicellular organism could still have allowed the earliest animals to seek out and move towards resource-rich areas . Aerotaxis requires cells to both sense oxygen and react appropriately to changes in its concentration . Kirkegaard et al . watched choanoflagellate colonies swimming under controlled conditions and varied the oxygen concentration in the water over time . These experiments revealed that the colonies navigate based on the logarithm of the oxygen concentration , so that at low oxygen levels the cells were even more sensitive to small changes in oxygen concentration . This type of ‘logarithmic sensing’ is similar to how our ears sense sounds and our eyes sense light . Kirkegaard et al . went on to conclude that the colonies were not actively steering in the correct direction directly . Instead , the colonies appeared to choose directions at random and later decide whether such a turn was correct . It remains unclear whether the common ancestor of animals and choanoflagellates could also perform aerotaxis , and if so what mechanisms this involved . Further studies to compare aerotaxis and aerotaxis-related genes in simple animals and other single-celled relatives of animals would be needed to illuminate this . Future studies could also explore the maximum and minimum oxygen concentrations that choanoflagellates can detect , and how well they navigate at these upper and lower limits . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"physics",
"of",
"living",
"systems"
] | 2016 | Aerotaxis in the closest relatives of animals |
Copy number alterations ( CNAs ) in cancer patients show a large variability in their number , length and position , but the sources of this variability are not known . CNA number and length are linked to patient survival , suggesting clinical relevance . We have identified genes that tend to be mutated in samples that have few or many CNAs , which we term CONIM genes ( COpy Number Instability Modulators ) . CONIM proteins cluster into a densely connected subnetwork of physical interactions and many of them are epigenetic modifiers . Therefore , we investigated how the epigenome of the tissue-of-origin influences the position of CNA breakpoints and the properties of the resulting CNAs . We found that the presence of heterochromatin in the tissue-of-origin contributes to the recurrence and length of CNAs in the respective cancer type .
Genomic alterations in cancer show considerable heterogeneity across different tumor types and even across patients with the same type of cancer . For point mutations , we are beginning to understand the determinants of this variation: the epigenomic profile of the tissue-of-origin highly influences local mutation rates along the chromosome ( Schuster-Böckler and Lehner , 2012; Polak et al . , 2015; Supek and Lehner , 2015 ) , different mutagens induce characteristic mutational signatures ( Alexandrov et al . , 2013 ) , and tissue-specific exposure to environmental factors affects the selection of mutations during tumourigenesis ( Schaefer and Serrano , 2016 ) . The driving forces behind copy number alterations ( CNAs ) , that is , amplifications or deletions of genomic regions , are much less understood than the causes of point mutations . Furthermore , we do not know why some cancer types are associated with many CNAs and other types with only a few . This is partly due to the fact that CNAs tend to affect several genes at the same time [in the dataset from The Cancer Genome Atlas ( TCGA; http://cancergenome . nih . gov/ ) used in this analysis , 59 genes on average are affected by a single CNA] . Therefore , it is often difficult to tell whether , and on which of the genes in the amplified or deleted region , selection is acting . In addition , cancer samples usually carry a much lower number of CNAs ( on average 46 CNAs in the patient samples considered in this study ) than single nucleotide variants ( SNVs; usually 10 . 000s per cancer genome ) . The sparse number of CNAs hinders the detection of statistical associations between CNAs and genetic and epigenetic features , work that has previously been carried out for SNVs ( Schuster-Böckler and Lehner , 2012; Polak et al . , 2015; Supek and Lehner , 2015 ) . Like other alterations , CNAs show a large variation in position , length and number across cancer types ( Zack et al . , 2013 ) . Authors have reported that CNA breakpoints are preferentially located in close proximity to DNA-methylation-depleted G-quadruplex sequences ( De and Michor , 2011 ) . This suggests that DNA secondary structure contributes to the CNA distribution . In addition , CNAs that are close to telomeres are longer than those found in internal regions . This suggests that there are several different mechanisms of CNA generation ( Zack et al . , 2013 ) . It has also been observed that DNA contact points in genome-wide chromosome conformation capture ( HiC ) proximity maps are more likely to become CNA breakpoints . Thus , the length distribution of CNAs reflects chromosomal interactions ( Fudenberg et al . , 2011 ) . The observation that certain genes tend to be mutated in CNA-rich ( TP53 and SPOP [Ciriello et al . , 2013; Boysen et al . , 2015] ) or CNA-poor ( CTCF and ARID1A [Ciriello et al . , 2013] ) cancers implies that , besides epigenetic factors , the genetic background of the cell influences CNA variation . Here , we make use of the wealth of cancer genomics data provided by TCGA , to understand how the genetic background influences the CNA count per sample . We identify mutations in genes that are statistically linked to the number of CNAs in cancer patients . We refer to the identified gene set as CONIM genes ( COpy Number Instability Modulators; Figure 1A ) . The encoded proteins form a densely interacting network of epigenetic modifiers and DNA repair genes . To test whether this network is associated with the cancer-type-specific preference for CNAs in certain regions , we investigate how the chromatin organisation in the healthy tissue-of-origin relates to the occurrence of CNAs in cancer . 10 . 7554/eLife . 16519 . 003Figure 1 . Mechanisms of CNA number modulation and clinical importance . ( A ) Schematic showing how CONIM gene mutations can result in a higher or lower CNA number . ( B ) We performed Kaplan-Meier statistics on data from lower grade glioma ( LGG ) patients with deviating CNA numbers and lengths . LGG patients with fewer CNAs have a significantly better survival prognosis as compared to patients with many CNAs . ( C ) LGG patients with shorter CNAs have a significantly better survival prognosis when compared to patients with longer CNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 003
To estimate the relevance of CNA number and length for clinical outcome , we performed Kaplan-Meier survival analyses . To this end , we grouped the patients of each cancer type into quartiles with respect to the distributions of CNA number and average length . We then compared the survival frequencies of patients in the top quartile with those of patients in the bottom quartile . It has been shown previously that cancer cells that have undergone whole genome duplications are associated with higher CNA rates ( Zack et al . , 2013 ) and poor prognosis ( Dewhurst et al . , 2014 ) , thus we removed aneuploid samples . As CNA numbers have been linked to mutation rate ( Ciriello et al . , 2013 ) , we additionally excluded highly mutated samples . We observed that for five of the 19 cancer types ( brca , lgg , hnsc , paad and ucec ) for which we had CNA and survival data , fewer CNAs were significantly associated with a longer survival period ( p < 0 . 05; chi-square test; see Figure 1B as an example ) . In addition , in two out of the 19 cancer types ( lgg and lihc ) , samples in the bottom quartile of the average CNA length were associated with a longer survival compared to samples from the top quartile ( p < 0 . 05; chi-square test; see Figure 1C as an example ) , again controlling for mutation number and ploidy . In none of the cancer types were fewer or shorter CNAs significantly associated with shorter survival . We investigated the relation between the mutational background and the CNA number of a patient . To this end , we set up a computational pipeline in order to detect genes that are associated with significantly different CNA numbers , comparing samples in which the gene was non-silently mutated with those that were mutation-free . We corrected for potential confounding factors such as tumor-type- and gene-specific alteration rates ( see Materials and methods ) . We applied our pipeline to a pan-cancer set consisting of 5 , 734 samples from 20 different cancer types ( see Materials and methods ) . This resulted in a list of 63 genes that are associated with significantly different CNA numbers . To acknowledge the potential impact of mutations of these genes on the overall number of CNAs , we termed this gene set COpy Number Instability Modulator ( CONIM ) genes . Mutations in 62 of these genes are associated with significantly fewer CNAs , whereas one gene ( TP53 ) is associated with a significantly higher number of CNAs ( see Supplementary file 1 for the full gene list and Figure 2A for two examples ) . 10 . 7554/eLife . 16519 . 004Figure 2 . Detection and functional properties of CONIM genes . ( A ) CNA numbers in samples in which CTCF ( left box ) or TP53 ( right box ) are mutated versus samples in which the respective gene is not mutated . The CNA number distributions are shown for all cancers types ( left whiskers within each box ) and for a single cancer type ( right whiskers within each box ) . ( B ) Mutations in CONIM genes tend to have a higher functional impact than mutations found in genes with an equal mutation frequency . Even CONIM genes not previously reported ( Lawrence et al . , 2014 ) to be frequently mutated in cancer tend to host mutations with a higher functional impact score ( mean 17 . 23 ) as compared to random gene sets having matched mutation numbers ( p = 0 . 029; randomisation test ) . For comparison , the most frequently mutated cancer-driver genes have a mean score of 18 . 31 . ( C ) The functional categories most significantly overrepresented among the CONIM genes are shown . Among the most highly enriched categories are several terms related to DNA damage ( green ) , chromatin organisation ( blue ) and complex formation ( red ) . Significance levels are indicated as follows: **q < 0 . 01 , ***q < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 00410 . 7554/eLife . 16519 . 005Figure 2—figure supplement 1 . Variant allele fractions ( VAFs ) of different gene groups . VAFs of mutations in CONIM genes that have not previously been implicated in cancer ( red ) are compared to those of a random set of equally often mutated genes ( green ) . Additionally , known cancer driver genes are shown in yellow ( CONIM ) and blue ( non-CONIM ) . Out of the five cancer types tested , two ( hnsc , luad ) have non-cancer CONIM genes that are associated with significantly lower VAFs as compared to random genes and cancer drivers . Non-cancer CONIM genes were not associated with significantly higher VAFs in any of the tested cancer types . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 005 Of the 63 CONIM genes , 15 are known to be frequently mutated in cancer ( Lawrence et al . , 2014 ) , and as such are likely to be drivers of malignant transformation . Their fraction among CONIM genes is higher than expected by chance ( p < e-16; chi-square test ) . We contemplated whether mutations in the remaining 48 genes contribute to the progression of the cancer or are just a by-product of the increased mutation rates found in cancer cells . Accordingly , we used functional impact scores to estimate the pathogenicity ( Kircher et al . , 2014 ) of the mutations found in CONIM genes that had not been previously implicated in cancer progression . The scores were compared to those of mutations found in genes that have an equal missense mutation frequency ( Figure 2B ) . We found that mutations in CONIM genes have , on average , a stronger functional impact than those in genes not associated with a change in CNA number . To estimate the temporal order of somatic events , we compared the variant allele fractions ( VAFs ) of mutations in non-cancer CONIM genes to the VAFs of mutations from equally often mutated genes . We found that in two out of five cancer types tested , mutations in CONIM genes were associated with a lower VAF ( Figure 2—figure supplement 1 ) . This suggests that mutations in CONIM genes tend to arise later in time but are more likely to be pathogenic than those in genes having similar mutation frequencies . To investigate the potential mechanisms through which mutations in genes encoding CONIM proteins affect the amount of CNAs in a tumor , we explored the functions of the CONIM gene set . We tested for functions , pathways , and complexes enriched among CONIM genes ( Kamburov et al . , 2013 ) . Interestingly , we found several interrelated functions to be most strongly enriched ( Figure 2C ) . Among the most frequent GO terms were chromosome organisation ( q < e-4; all functional enrichments FDR corrected ) and chromatin modification ( q < 0 . 001 ) , suggesting that CONIM genes might alter CNA numbers through structural changes in the genome . More specifically , eight CONIM genes were involved in histone modification ( q < 0 . 001 ) . Of these , three genes were related to histone deacetylation ( q < 0 . 01 ) and another three to histone methylation ( q < 0 . 05 ) . Together , 17 of the 63 genes had functions related to the structural organisation of the chromosomes or to epigenetic modifications ( Supplementary file 1 ) . Several pathways related to DNA damage were strongly enriched [e . g . , 'DNA Damage/Telomere Stress Induced Senescence' ( q < 0 . 01 ) and 'DNA Damage Response ( only ATM dependent ) ' ( q < e-4 ) ] . Notably , the ATM ( Ataxia Telangiectasia Mutated ) DNA damage response plays an important role in the repair of double-stranded DNA breaks from which CNAs originate . TP53 is a member of both pathways . The greater number of CNAs in TP53-mutated samples might reflect the incapability of the affected cells to repair DNA breaks or to initiate apoptosis upon damage . Another group of the most strongly enriched terms centered on complex formation [e . g . 'macromolecular complex binding' ( q < e-4 ) and 'macromolecular complex subunit organisation' ( q < 0 . 001 ) ] . We tested whether we can recover the same CONIM genes when we vary the underlying CNA data or algorithmic details of the detection pipeline ( see the section 'Robustness of CONIM gene discovery and properties' for details of three alternative CONIM gene detection pipelines ) . Even though we found that some of the CONIM genes are specifically detected by a single pipeline or in only a subset of cancer types , we found three genes that come up in all conditions and 21 genes that are recovered by at least two pipelines . Also , the enrichment of epigenetic modifiers among CONIM genes from the different pipelines is very robust . To investigate whether CNA properties other than their number depend on the genetic background , we tested whether the average length of CNAs differs between samples with and without a mutation in each gene . We found 540 genes that are significantly associated with shorter or longer CNAs ( FDR corrected q < 0 . 01; Mann-Whitney-Wilcoxon test ) . Out of these 540 genes , 122 were also associated with a significantly different CNA number ( FDR corrected q < 0 . 01; Mann-Whitney-Wilcoxon test on pan-cancer set without applying any additional filters or corrections ) . The overlap between the mutated genes found in samples with a differential CNA length and those found in samples with a differential CNA number was larger than expected by chance ( p < e-16; chi-square test ) . The majority ( 98 . 5% ) of these genes were associated with fewer and shorter CNAs , suggesting that the same cellular mechanisms might influence both CNA number and length . As the functional enrichment analysis revealed a tendency of CONIM proteins to participate in the formation of protein complexes ( Figure 2C ) , we investigated the network organisation of this protein group . When linking CONIM proteins with protein-protein interaction ( PPI ) information [from HIPPIE version 1 . 8 ( Schaefer et al . , 2012 ) ] , we observed that 32 of the CONIM proteins ( 50 . 8% ) are part of a large connected network ( Figure 3A ) . To test whether the degree of connectivity among CONIM proteins is greater than one would expect by chance , we performed a network randomisation test . We found that both the observed numbers of PPIs ( p = 0 . 001; randomisation test; Figure 3B ) and the size of the largest connected component ( p = 0 . 003; randomisation test; Figure 3C ) were significantly larger in the original network than in the randomised networks . 10 . 7554/eLife . 16519 . 006Figure 3 . CONIM proteins form a dense network . ( A ) All interactions between CONIM proteins are shown . A total of 32 CONIM proteins are connected to each other via 42 physical interactions . Several complexes are highlighted . ( B ) The observed number of PPIs between CONIM proteins is greater than that for randomly sampled networks of proteins forming as many interactions as the CONIM proteins ( p = 0 . 001; randomisation test ) . ( C ) Using the same network randomisation approach , we establish that the size of the largest connected component exceeds random expectation ( p = 0 . 003 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 006 We found that CONIM proteins of the largest connected component are significantly enriched in several complexes – the four complexes with the strongest enrichment are highlighted in Figure 3A . In agreement with the functional enrichment , we found an enrichment of CONIM proteins in the SWI/SNF complex ( EP400 , ARID1A , PBRM1 and ATRX ) , which is involved in chromatin remodeling by restructuring nucleosomes . Mutations in components of the SWI/SNF complex have been observed in different tumor types , but their contribution to carcinogenesis is only poorly understood ( Masliah-Planchon et al . , 2015 ) . Given previous reports on the link between chromatin structure and the genomic position of CNAs ( see Introduction ) , we hypothesise that epigenetic modifiers are enriched among CONIM genes because they influence structural instability through chromatin modifications . In this way , CONIM genes could alter the susceptibility of chromosomal regions to DNA double-strand breaks that , when not repaired properly , would result in CNAs . CNAs are around four orders of magnitude less abundant in patients than are SNVs . This prevented us from correlating CNA numbers from different cancer types with epigenetic marks in the respective tissue-of-origin using windows with a sufficient genomic resolution , as has been done for SNVs ( Polak et al . , 2015 ) . Instead , we explicitly tested whether epigenetic marks around breakpoints are enriched in those tissues where the breakpoint frequently occurs during cancer development versus those tissues where the breakpoint does not occur . To this end , we assembled a list of recurrent CNAs ( Mermel et al . , 2011 ) that are significantly more frequent in a certain cancer type than would be expected by chance ( q < 0 . 1; FDR corrected ) , resulting in 1 , 036 unique CNA breakpoints . As a first analysis , we compared the frequency of 18 chromatin states ( Kundaje et al . , 2015 ) around the breakpoint in the tissue from which the cancer originated ( 'associated tissues' ) with the frequency in other tissues ( 'non-associated tissues' ) : Figure 4A shows the frequency ratios for the most abundant states . The strongest enrichment was observed for ‘Heterochromatin’ ( p = 0 . 009; chi-square test ) . The only other significantly enriched state is 'ZNF genes and repeats' ( p = 0 . 03; chi-square test; Figure 4—figure supplement 1 ) . However , the frequency of this state at CNA breakpoints is more than five times lower than that of ‘Heterochromatin’ . 10 . 7554/eLife . 16519 . 007Figure 4 . Epigenetic properties of CNA breakpoint regions . ( A ) Ratio of the number of breakpoints falling into different chromatin regions in tissues where the CNA event is significantly recurrent to the number in other tissues . States coinciding at least 100 times with breakpoints in non-associated tissues are shown . The number of CNA breakpoints in 'Heterochromatin' is significantly enriched ( p = 0 . 009; chi-square test ) . ( B ) The average fraction of genomic windows centering on CNA breakpoints that is associated with different histone marks is compared between tissues where the CNA region drives cancer ( observed ) and other tissues ( expected ) . Black dots represent bin sizes with significant enrichment ( Bonferroni-corrected p < 0 . 05; Mann-Whitney-Wilcoxon test ) . ( C ) CNAs originating from 343 H3K9me3-enriched breakpoints are significantly longer than those originating from 738 H3K9me3-depleted breakpoints ( **p < 0 . 01; Mann-Whitney-Wilcoxon test; 10 kb window ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 00710 . 7554/eLife . 16519 . 008Figure 4—figure supplement 1 . Enrichment of chromatin states at breakpoints for different cell-of-origin associations . For the original and alternative selections of reference epigenomes , the states 'ZNF genes and repeats' ( 12_ZNF/Rpts ) and ‘Heterochromatin’ ( 13_Het ) show the most significant enrichments [chi-square test; chromatin state labels according to 18-state model ( Kundaje et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 00810 . 7554/eLife . 16519 . 009Figure 4—figure supplement 2 . Enrichment of H3K9me3 for different cell-of-origin associations . Independently of the reference epigenome selection , H3K9me3 at CNA breakpoints is enriched in tissues where the CNA event is recurrent as compared to other tissues ( p < 0 . 003; Mann-Whitney-Wilcoxon test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 009 As both of these states are characterised by the presence of H3K9me3 ( Kundaje et al . , 2015 ) , we specifically investigated the enrichment of histone marks in the proximity of CNA breakpoints . For this purpose , we defined windows of different sizes centering on CNA breakpoints , and computed the total length of regions corresponding to a specific histone mark in tissues where the CNA region is recurrent and in other tissues . As expected , we observed the strongest enrichment for H3K9me3 ( Bonferroni-corrected p < 0 . 001 for windows between 10 kb and 10 Mb; Mann-Whitney-Wilcoxon test; Figure 4B ) . This enrichment decreases with increasing distance from the CNA breakpoint , suggesting a colocalisation of H3K9me3 marks with recurrent breakpoints in the tissue-of-origin . For all other histone marks considered ( H3K4me1 , H3K4me3 , H3K27me3 , H3K36me3 , H3K9ac , and H3K27ac ) , we find much weaker effects . We next studied whether CNAs originating from H3K9me3-enriched breakpoints have any properties that distinguish them from CNAs at H3K9me3-depleted sites . As we observed the strongest H3K9me3 enrichment in 10 kb windows around CNA breakpoints ( Figure 4B ) , we considered tri-methylated H3K9 in this range . We found that CNAs with a H3K9me3 enrichment in close proximity to the breakpoint were longer than CNAs originating from H3K9me3-depleted breakpoints ( p = 0 . 001; Mann-Whitney-Wilcoxon test; Figure 4C ) . As telomere-bounded CNAs have previously been reported to be longer than others ( Zack et al . , 2013 ) , we tested whether our observation could be an artifact of higher heterochromatin content towards the chromosome ends ( even though CNAs originating from telomeres were excluded from our analyses ) . When comparing the positions relative to the chromosome end , we did not detect any differences between breakpoints in H3K9me3-enriched and H3K9me3-depleted genomic windows ( p = 0 . 8; Mann-Whitney-Wilcoxon test ) . Additionally , we tested the effect of excluding breakpoints located within 1 Mb or 10 Mb of the chromosome ends . Both tests re-confirmed significant differences in the length distributions of the two CNA groups ( p < 0 . 005; Mann-Whitney-Wilcoxon test ) . The difference in length distributions might suggest distinct mechanisms of generation that depend on the epigenetic features present at the position where the DNA breakpoint appears . To establish a link between tissue-specific chromatin at the CNA breakpoints and CONIM gene mutations , we sought to demonstrate that tissues with highly H3K9me3-enriched breakpoints also have more mutations in chromatin-modifying CONIM genes . None of the CONIM proteins specifically methylates H3K9 , but the CONIM proteins ATRX , EP400 and NIPBL bind to H3K9me3 directly or form H3K9me3-binding complexes ( Eustermann et al . , 2011; Lai et al . , 2013; Oka et al . , 2011; Vermeulen et al . , 2010; Kunowska et al . , 2015; Nikolov et al . , 2011 ) . We found that non-silent mutations in these genes affect a greater proportion of samples in cancer types ( luad , lusc , lihc and skcm ) that show a strong H3K9me3 enrichment ( > 2-fold change in 10 kb windows around breakpoints; p < 0 . 05; Mann-Whitney-Wilcoxon test ) in their tissue-of-origin ( p < e-6; chi-square test ) . An overrepresentation of mutated samples in these cancer types was again observed when considering each gene individually ( ATRX: p = 0 . 02; EP400: p < 0 . 001; NIPBL: p = 0 . 09; chi-square test ) . To better understand how gain- or loss-of-function mutations in CONIM genes could affect CNAs , we investigated the relationship between CONIM gene activity and heterochromatin amount in healthy tissues . For this purpose , we compared tissue-specific RNA abundance levels ( as a proxy for gene activities ) with the percentage of DNA in a heterochromatic state in the same tissue . We computed the Pearson correlation between the expression of all human protein-coding genes with the percentage of heterochromatin in 48 cell lines and tissues ( Kundaje et al . , 2015 ) . We found that the absolute correlation between total heterochromatin amount and expression of either CONIM histone modifiers or all CONIM genes is significantly larger than that of non-CONIM genes ( p < 0 . 05 and p < e-5; Mann-Whitney-Wilcoxon test; Figure 5A ) . One possible explanation for this observation is that ( under healthy conditions ) CONIM genes are implicated in controlling the overall amount of heterochromatin . We decided to focus on NIPBL , the CONIM histone modifier that showed the strongest absolute correlation ( −0 . 53 ) between tissue-specific expression and amount of heterochromatin in the same tissue . This gene has been implicated in the developmental disorder Cornelia de Lange syndrome ( CdLS ) ( Krantz et al . , 2004 ) . Mutations in NIPBL have been associated with chromatin decompaction and , indeed , mutations that are predicted to have a more severe effect on NIPBL exhibit a stronger effect on chromatin ( Nolen et al . , 2013 ) . We therefore tested whether mutations in the HEAT domain , which is necessary to target NIPBL to sites of DNA damage ( Oka et al . , 2011 ) , have a stronger effect on CNA number in cancers than do other missense mutations . We also checked whether cancers with truncating mutations in the N-terminus of NIPBL are associated with a significantly lower CNA number as compared to those with truncating mutations in the C-terminus ( Figure 5B ) . In both cases , we observed a significant difference , with mutations that have an anticipated stronger functional or structural impact on NIPBL being associated with fewer CNAs . 10 . 7554/eLife . 16519 . 010Figure 5 . CONIM genes modify the CNA amount via the epigenome . ( A ) The absolute correlation between heterochromatin amount and expression of either CONIM histone modifiers or all CONIM genes is significantly larger than that of non-CONIM genes . ( B ) In the NIPBL gene , nonsense or frameshift mutations in the N-terminal third of the protein , and missense mutations in the HEAT repeat , have a stronger effect on the CNA number in the respective samples than do those mutations that have a smaller effect on protein structure and function . The average ( C ) CNA number and ( D ) CNA length per cancer type is correlated with the percentage of heterochromatin in the associated healthy tissue . Significance levels are indicated as follows: *: q < 0 . 05 , **: q < 0 . 01 , ***: q < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 01010 . 7554/eLife . 16519 . 011Figure 5—figure supplement 1 . Average CNA number and heterochromatin percentage for alternative reference epigenomes . The Spearman correlation between the average number of CNAs per cancer type and the heterochromatin percentage in the tissue-of-origin is significantly larger when all possible combinations of reference epigenomes are considered than for 1 , 000 randomly sampled associations between cancer types and healthy tissues ( p < e-10; Mann-Whitney-Wilcoxon test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 01110 . 7554/eLife . 16519 . 012Figure 5—figure supplement 2 . Average CNA length and heterochromatin percentage for alternative reference epigenomes . The Spearman correlation between average length of CNAs per cancer type and heterochromatin percentage in the tissue-of-origin is significantly larger when all possible combinations of reference epigenomes are considered than for 1 , 000 randomly sampled associations between cancer types and healthy tissues ( p < e-16; Mann-Whitney-Wilcoxon test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16519 . 012 These analyses suggest that the condensation state of chromatin influences the occurrence of DNA breaks . We therefore investigated whether the overall amount of heterochromatin in each tissue is linked to the amount of CNAs in the cancer type originating from the respective tissues . For most cancer types , we observed that the average number of CNAs is highly correlated to the percentage of heterochromatin in the associated tissue ( Figure 5C ) . Ovarian cancer does not follow the general trend , but for other cancer types ( for which we had CNA numbers and heterochromatin information available ) , we observed a Spearman correlation of 0 . 72 ( p = 0 . 02 ) . This suggests that the distribution of CNA numbers over cancer types is linked to the chromatin organisation of the tissue-of-origin . As we found CNAs originating from breakpoints in heterochromatin to be longer , we also compared the mean length of CNAs for each cancer type with the percentage of heterochromatin in the tissue from which the cancer originated . Again , we observed a good correlation for most cancer types except for ovarian cancer ( Spearman's rho = 0 . 85; p = 0 . 002 ) , which decreases but remains significant when ovarian cancer is included ( Spearman's rho = 0 . 62; p = 0 . 04; Figure 5D ) . These observations provide a possible explanation for how mutations in CONIM genes could affect CNA numbers and lengths: the altered activity of CONIM genes affects the amount of heterochromatin , with more heterochromatin leading to more and on average longer CNAs , and with less heterochromatin having the opposite effect . The tissue-specific differences in CNA number seem to reflect the tissue-specific differences in heterochromatin .
Here , we describe a new class of cancer-related genes: the CONIM genes . They are characterised by being associated with the amount of chromosomal gain or loss in a cancer cell , but only about 24% of these genes have previously been associated with cancer . Our study highlights their possible role as copy number instability modulators and suggests a mechanism for how they contribute to cancer development . Mutations in all but one of the CONIM genes are associated with a smaller number of CNAs . One explanation for this observation could be that mutations in CONIM genes tend to occur late during cancer development . This is supported by the low VAFs of CONIM genes that we observe in two cancer types . When many alterations have already been accumulated , high proliferation rates increase the risk of further damage which – at this point – would be detrimental to the cancer . The exception is TP53 , which is associated with a higher number of CNAs when mutated . Inactivation of TP53 decreases sensitivity to apoptosis , and therefore more DNA damage ( including CNAs ) is tolerated . Previously , an inverse relation between the number of CNAs and the number of point mutations has been described ( Ciriello et al . , 2013 ) , subdividing tumors into two groups: one CNA-rich and one mutation-rich . The CNA-rich group has been associated with recurrent mutations in TP53 and the mutation-rich ( and CNA-depleted ) group with mutations in ARID1A and CTCF . These three genes are also in our CONIM gene list . Several other studies investigated relations between point mutations and CNA numbers in single cancer types: a higher number of CNAs has been reported in SPOP-mutated prostate cancer ( Boysen et al . , 2015 ) . Lower CNA numbers have been detected in CASP8-mutated oral squamous cell carcinoma ( Pickering et al . , 2013 ) and in CTNNB1-mutated endometrial cancer ( Kandoth et al . , 2013 ) . Unlike CASP8 and CTNNB1 , which are part of our CONIM list , SPOP did not pass our pan-cancer CNA enrichment filter criteria because the effect of SPOP on CNAs is highly cancer-type-specific . However , SPOP was recovered by our cancer-type-specific alternative detection pipeline ( see Materials and methods ) . Our study goes beyond these previous studies by also considering the influence on CNA occurrence of the epigenome in the tissue from which the cancer originated . As the inverse relation between CNA and point mutations might affect the detection of CONIM genes , we apply different strategies to correct for this potential confounder ( regressing out mutation rates , removing highly mutated samples and applying a mutation-number-matched permutation test ) . We found that the greater amount of CONIM genes associated with lower CNA number , the enrichment of epigenetic modifiers and the high connectivity can be reproduced with different CONIM gene detection pipelines . We also tested whether a gene that is associated with an elevated point mutation rate would automatically end up in our CONIM gene list due to the inverse relation between CNA and mutation counts . POLE has been described in the literature to cause a hypermutation phenotype when somatically mutated ( Roberts and Gordenin , 2014; Briggs and Tomlinson , 2013 ) . We can confirm that samples with POLE mutations have higher point mutation counts as compared to randomly selected samples ( carrying mutations in genes with similar mutation frequencies as POLE ) . However , we do not find a reduced number of CNAs in POLE mutated samples . The most strongly enriched pathway among CONIM genes is ATM-dependent DNA repair . ATM is required for the repair of DNA double-strand breaks in heterochromatic regions , a process which is characterised by slow repair kinetics ( Goodarzi et al . , 2010 ) . ATM-mediated phosphorylation of KAP1 ( KRAB-associated protein 1 ) triggers local decondensation of heterochromatin and thereby facilitates efficient repair . This suggests that it is not only the amount of cellular heterochromatin but also the cell’s ability to decondense it that is important . Other studies have begun to investigate the causes of variation in the frequency of CNAs throughout the genome by comparing distributions of CNAs to those of genomic and epigenomic features ( De and Michor , 2011; Fudenberg et al . , 2011; Zack et al . , 2013 ) . These analyses have suggested , among other features , the involvement of chromatin formation in determining the distribution of CNAs . However , none of the previous studies have systematically compared tissue-of-origin chromatin conformation to cancer-type-specific recurrence of CNAs ( in a similar manner as it has been done for epigenetic marks and point mutations [Polak et al . , 2015] ) . Our study complements these previous efforts by showing that not just the distribution of SNVs but also the CNA breakpoint distribution seem to be influenced by local chromatin structure . Here , we establish a link between heterochromatin enrichment and the variation in CNA number , length , and position across cancer types . In accordance with other studies , density of chromatin ( for example , differences in mechanical forces or exposure to mutagens resulting from the localisation of dense chromatin at the nuclear periphery [Misteli , 2007] ) determines where CNAs occur or persist ( the high degree of condensation might hinder the detection and repair of DNA damage [Peterson and Côté , 2004] ) . These factors are governed by the properties of the tissue-of-origin ( which contribute to the variability in the number , length and distribution of CNAs over cancer types ) and could be influenced by abnormal activity of epigenetic modifiers through mutation or differential expression ( contributing to the variation on the patient-level ) . With respect to possible mechanisms of heterochromatin formation interruption , it is worth mentioning that CONIM genes encode rather more H3K9me3 ‘readers’ than ‘writers’ . Interestingly , we found that the local epigenome not only impacts where a DNA breakpoint occurs but also the length of the resulting CNAs . CNAs originating from H3K9me3-enriched regions tend to be longer than those without neighboring H3K9me3 marks . This increased average length is probably due to the fact that a greater degree of packaging of the interphase genome into heterochromatin facilitates long-range contacts between distant parts of the DNA , which then serve as end points for CNAs . This interpretation is in agreement with the observation that the chromatin shapes the length distribution of CNAs ( Fudenberg et al . , 2011 ) . It has previously been reported that CNAs originating from telomeres are longer than chromosome-internal CNAs ( Zack et al . , 2013 ) . We found that breakpoints in H3K9me3-enriched regions are associated with longer CNAs than other breakpoints , independent of their positions with respect to the chromosome ends . As telomeres in fact form heterochromatin ( Blasco , 2007 ) , our findings might explain the previously observed position-dependent length differences . Regarding the link between cancer-type-specific CNA numbers or lengths and heterochromatin proportions in the corresponding tissue-of-origin , ovarian cancer may not follow the same trend as other cancer types due to very high mutation fractions in TP53 ( 94% [Lawrence et al . , 2014] ) . In accordance with previous studies ( Ciriello et al . , 2013 ) , we show that TP53 deficiency is strongly associated with high CNA numbers . More research needs to be done on the mechanistic details of CNA breakpoint generation by chromatin disorganisation . To this end , our study highlights several interesting candidate genes that could be valuable drug targets as our analyses suggest that CNA number and size are clinically relevant . In summary , our observations suggest that the epigenome impacts CNA occurrence in a tissue- and patient-specific manner . CNA breakpoints are overrepresented in heterochromatic regions , so the epigenome of the tissue from which a cancer originates has a large impact on where CNAs arise during carcinogenesis . In addition , we identified genes in which mutations are associated with differential CNA number and length . Interestingly , this gene set is enriched in chromatin-modifying genes , which could suggest that these genes influence CNA properties through chromatin modifications .
The Kaplan-Meier analysis was performed with the survival R package ( https://cran . r-project . org/web/packages/survival/index . html ) . To prevent results from being confounded by high mutation rates , we removed samples with a mutation number of more than two standard deviations higher than the cancer-type-specific median . We also controlled for the effect of whole-genome duplications . We retrieved ploidy information for most cancer types from the COSMIC database ( Forbes et al . , 2015; RRID:SCR_002260 ) and the literature ( Ceccarelli et al . , 2016 ) . For the remaining cancer types ( kidney renal clear cell carcinoma and pheochromocytoma and paraganglioma ) , we estimated ploidy using the ABSOLUTE tool ( Carter et al . , 2012; RRID:SCR_005198 ) . We removed samples with an estimated ploidy of more than 2 . 9 . We retrieved somatic mutations in coding regions for 20 cancer types [Bladder Urothelial Carcinoma ( blca ) , Breast invasive carcinoma ( brca ) , Cervical squamous cell carcinoma and endocervical adenocarcinoma ( cesc ) , Colon adenocarcinoma ( coad ) , Glioblastoma multiforme ( gbm ) , Head and Neck squamous cell carcinoma ( hnsc ) , Kidney renal clear cell carcinoma ( kirc ) , Kidney renal papillary cell carcinoma ( kirp ) , Acute Myeloid Leukemia ( laml ) , Brain Lower Grade Glioma ( gbm ) , Liver hepatocellular carcinoma ( lihc ) , Lung adenocarcinoma ( luad ) , Lung squamous cell carcinoma ( lusc ) , Pancreatic adenocarcinoma ( paad ) , Pheochromocytoma and Paraganglioma ( pcpg ) , Prostate adenocarcinoma ( prad ) , Skin Cutaneous Melanoma ( skcm ) , Stomach adenocarcinoma ( stad ) , Thyroid carcinoma ( thac ) , and Uterine Corpus Endometrial Carcinoma ( ucec ) ] from TCGA comprising a set of 5 , 960 samples . CNA coordinates for each sample were retrieved from SNP6 array data through firehose ( gdac . broadinstitute . org; run 2014/10/17 ) . Only CNAs with a segment copy number larger than 0 . 1 or smaller than −0 . 1 ( in units of log2 ( copy number ) – 1 ) and with a minimum length of 100 bp were considered . We used 5 , 734 cancer samples from 20 different cancer types for which both mutation and CNA information were available , and tested whether samples with non-silent mutations in certain genes carry significantly more or less CNAs than samples without mutations in the respective genes . We excluded genes with less than 60 non-silent mutations in the set of 5 , 734 samples . We only considered the cancer types that have at least 200 available samples; with at least five of the samples carrying a non-silent mutation in the respective gene . To ensure a strong effect , we only considered cases where the absolute log ratio difference was above 0 . 5 and applied a q-value cutoff of 0 . 01 ( Mann-Whitney-Wilcoxon test; FDR corrected ) . Among the genes associated with a higher number of CNAs , we observed a strong enrichment of those encoding large membrane-bound proteins . Among them were the largest human protein , TTN , and several olfactory receptors . As mutations in these genes are thought to be spurious passenger mutations and to confound statistics through locally elevated mutation rates ( Lawrence et al . , 2013 ) , we additionally removed genes associated with a differential number of CNAs when carrying silent mutations . This was done by dividing samples into those carrying a silent mutation versus those not carrying a silent mutation and performing the same test as before . This implicitly corrects for local differences in mutation rates and gene length . CNA numbers differ across cancer types and are anti-correlated with the number of mutations ( Ciriello et al . , 2013 ) . We therefore aimed to control for these confounding factors by ( a ) testing in each cancer type separately for genes that when mutated are associated with higher or lower CNA number and ( b ) including mutation rates into a multiple regression model . For each genei and each cancer type , we fitted a linear model with sample-specific CNA number as the predictor variable and with both mutation status of genei and mutation number per sample as predictor variables . We then tested whether the mutation status alone significantly contributes to the CNA number ( t-test ) . We kept only genes in the result list for which there was at least one cancer type with a FDR corrected q < 0 . 1 . We used the web tool ConsensusPathDB ( Kamburov et al . , 2013 ) ; RRID:SCR_002231 ) to assess the significance of GO term and pathway enrichment . We restricted the analysis to GO terms , pathways and complexes from the pathway databases Reactome ( Croft et al . , 2014; RRID:SCR_003485 ) and WikiPathways ( Kutmon et al . , 2016; RRID:SCR_002134 ) as well as the protein complex database CORUM ( Ruepp et al . , 2008 ) ; RRID:SCR_002254 ) . The enrichment of all discussed functions related to epigenetic modifications and DNA repair remained significant ( q < 0 . 05; FDR corrected ) when we computed the enrichment with respect to highly mutated genes in cancer ( genes with at least 100 non-silent mutations in the pooled cancer set ) instead of to the entire genomic background . We observed several genes that are involved in signaling among the CONIM set ( e . g . , KRAS and BRAF ) . However , the enrichment of signaling-related GO terms was much weaker than , for example , terms related to chromatin organisation: among the 50 most significantly enriched GO terms , none had ‘signaling’ but eight had ‘chromosome’ or ‘chromatin’ in the name . The functional impact of the mutations was estimated using the Phred-transform of the CADD score ( Kircher et al . , 2014 ) . To estimate the significance of the higher mean damage score associated with CONIM genes , a randomisation test was applied: CONIM genes not previously involved in cancer were replaced by other genes with the same number of missense mutations . We excluded genes if less than 25 other genes had exactly the same mutation count . We computed VAFs as the read count supporting mutation divided by the total read count for each mutation in ucec , hnsc , luad , brca and skcm , as these cancer types had at least 100 mutations in non-cancer CONIM genes ( considering genes with at least 15 non-silent mutations ) , read count information and cancer gene classification ( Lawrence et al . , 2014 ) available . We retrieved PPIs from the integrated human PPI resource HIPPIE v1 . 8 ( Schaefer et al . , 2012; RRID:SCR_014651 ) . To test whether the observed number of PPIs between CONIM proteins and the size of the largest connected component ( the subnetwork in which every pair of proteins is connected by paths through the network ) were larger than expected by chance , we performed a randomisation test . We randomly sampled 1 , 000 protein sets of size equal to that of the CONIM protein set by replacing each CONIM protein by a protein of the same degree ( forming as many interactions as the replaced protein ) . This approach avoids an overestimate of connectivity statistics due to highly interacting proteins in the original protein set . In the few cases where there were less than 15 proteins with the same interaction degree we successively increased the margin around the interaction degree of the replaced proteins until there were at least 15 proteins with the same or similar interaction degree . For each random network , we counted the size of the largest connected component and the number of PPIs formed within the random network ( not considering self-interactions ) . In order to exclude the possibility that the higher number of PPIs and size of the largest connected component in the original set compared to the random sets is caused by the presence of a few highly-interacting hub proteins , we removed the protein with the highest interaction degree ( TP53 , the only protein forming more than 500 PPIs ) . We then repeated the randomisation test . We again observed a larger connected component size and a higher amount of interactions than would be expected by chance ( both p < 0 . 001; randomisation test ) . To estimate how robust the definition of the CONIM genes is with respect to algorithmic details and technical variation in the experimental determination of CNAs , we set up three additional pipelines . First , we designed an approach that primes cancer-type-specific CONIM genes high . To diminish the impact of variation in mutation frequency , we removed highly mutated samples ( more than two standard deviations away from the median – affecting 167 samples ) . We then computed the significant deviation in CNA number for each cancer type separately . As before , we removed genes that show an association with CNA number when considering only silent mutations . A resulting 25 genes had a FDR corrected Mann-Whitney-Wilcoxon p-value below 0 . 01– and 10 of those were associated with a higher CNA number ( TP53 again with the strongest effect ) . From the 15 genes associated with lower CNA number , 11 were found in the previous CONIM gene list . The greater proportion of CONIM genes associated with greater CNA number in this run indicates that this effect might be more cancer-type-specific . Second , to test the degree to which the CONIM gene definition is affected by the experimental method used to detect CNAs , we retrieved another CNA set from TCGA , employing Illumina HiSeq whole-genome sequencing . This dataset is generated on a much smaller set of tumor and normal samples ( 856 sample pairs for which we also had mutation data ) , covering only 10 of the initial 20 cancer types . Applying the initial CONIM detection pipeline to this dataset revealed a smaller number of only three genes ( TP53 , ARID1A and PTEN ) fully contained in the initial CONIM set and the cancer-type-specific CONIM gene set . However , relaxing the parameters increases the overlap with the results of the two other pipelines ( for example , applying only a q-value cutoff of 0 . 1 on the pan-cancer CNA set results in 24 genes , nine of which are in the original CONIM set ) . Third , we applied a different way of controlling for the sample-specific mutation frequencies: we permuted the observed mutations over samples and genes while keeping the number of mutations in a given gene over samples and the number of mutations in a given sample constant [following the approach described in ( Ding et al . , 2008 ) ] . In each permutation , we computed the absolute difference in the mean CNA number between samples with and without non-silent mutations in the respective gene as a test statistic . We performed 1 , 000 permutations and computed an empirical p-value as the fraction of times in which the absolute CNA difference was larger than the observed difference in the original data . As before , we included only genes with at least 60 mutations in the 20 cancer types considered . For each gene , we considered only cancer types with at least five non-silent mutations in the respective gene . This resulted in a list of 48 genes that when mutated were associated with a higher or lower CNA number in the same sample ( q < 0 . 01; permutation test ) . Seventeen of these genes overlapped with our initial CONIM gene definition; two of the genes were associated with higher CNA number ( TP53 and OR6N1 ) , 46 with lower CNA number . We tested whether the genes from the alternative pipelines have the same properties as the original CONIM set: the 25 genes from the cancer-type-specific pipeline were most strongly enriched in 'DNA Damage Response ( only ATM dependent ) ' ( q < e-6 ) . Several categories related to chromatin modification were found to be significantly enriched ( q < 0 . 01 ) : e . g . 'chromatin binding' and 'chromatin assembly or disassembly' . There is an enrichment of PPIs among these genes and a largest connected component exceeding random expectation ( both p < 0 . 001; randomisation test ) . Likewise , 'DNA Damage Response ( only ATM dependent ) ' was significant ( q < e-4 ) among the genes from the permutation-based pipeline . Also , several chromatin-modification-related categories were enriched: e . g . 'chromatin silencing' , 'chromatin modification' and 'histone methylation' ( all q < 0 . 01 ) . The number of PPIs formed among these genes exceeded random expectation ( p < 0 . 05 ) . We did not test functional or PPI enrichment among the sequencing-based pipeline as it contains only three genes , which are fully contained in the result sets of the other three pipelines . Supplementary file 1 gives information on the number of pipelines in which each CONIM gene can be reproduced . Significantly recurrent CNAs per cancer type were retrieved from FireBrowse [firebrowse . org; SNP6 Copy number analysis ( GISTIC2 ) ] applying a q-value cutoff of 0 . 1 . The GISTIC2 algorithm ( Mermel et al . , 2011; RRID:SCR_000151 ) separates arm-level and focal copy-number events , models background rates for CNA formation and defines boundaries with a predetermined confidence level . We assigned 13 cancer types [Acute Myeloid Leukemia ( laml ) , Breast invasive carcinoma ( brca ) , Colon adenocarcinoma ( coad ) , Esophageal carcinoma ( esca ) , Glioblastoma multiforme ( gbm ) , Liver hepatocellular carcinoma ( lihc ) , Lung adenocarcinoma ( luad ) , Lung squamous cell carcinoma ( lusc ) , Ovarian serous cystadenocarcinoma ( ov ) , Rectum adenocarcinoma ( read ) , Skin cutaneous melanoma ( skcm ) , Stomach adenocarcinoma ( stad ) , Thymoma ( thym ) ] to their tissues of origin in the Roadmap Epigenomics project ( Kundaje et al . , 2015; RRID:SCR_008924 ) . Identifiers of selected reference epigenomes used here as well as alternative epigenomes that likewise represent potential cell types of origin are listed in Supplementary file 2 . We defined CNA regions as being associated with a specific healthy tissue if they were significantly recurrent in the corresponding cancer type . CNA breakpoints falling into centromere or telomere regions , as retrieved from UCSC [human genome assembly hg19 ( February 2009 ) ; ( Rosenbloom et al . , 2015; RRID:SCR_005780 ) ] , and breakpoints that were associated with more than three healthy tissues were excluded from the analyses . It should be noted that the number of breakpoints for which both exclusion criteria apply is larger than expected by chance ( p < e-16; Fisher's test ) , suggesting that most CNA breakpoints that fall into centromere or telomere regions are not tissue-specific . For each healthy tissue , we used data from the Roadmap Epigenomics project ( Kundaje et al . , 2015 ) to quantify epigenetic marks for associated CNAs that are recurrent in the corresponding cancer type as compared to non-associated CNAs that promote cancer in other tissues . We assigned a CNA breakpoint to a chromatin state if it colocalised with the genomic region corresponding to that state as defined in the 18-state model by the Roadmap Epigenomics Consortium ( Kundaje et al . , 2015 ) . To test whether the chromatin state enrichments we observe depend on the specific reference epigenome selection , we repeated our analysis by replacing any number of reference epigenomes with equivalent cell types of origin ( Supplementary file 2 ) . This confirmed that the states 'ZNF genes and repeats' and ‘Heterochromatin’ show the most significant effects ( chi-square test; Figure 4—figure supplement 1 ) . To analyze the density of histone modifications in the vicinity of CNA breakpoints , we counted the total number of base pairs that overlap with ChIP-seq peaks ( ENCODE NarrowPeak format ) in genomic windows centering on the breakpoint . The enrichment that we found for tri-methylated H3K9 adjacent to CNA breakpoints can be reproduced when simply counting the number of ChIP-seq peaks in a genomic window . Moreover , an enrichment of H3K9me3 can be observed for all possible cell-of-origin associations ( Supplementary file 2; Figure 4—figure supplement 2; Bonferroni-corrected p < 0 . 005; Mann-Whitney-Wilcoxon test ) , suggesting that the results are independent of the reference epigenome selection . To investigate a potential link between H3K9me3 enrichment and CNA length , we compared the length of CNAs originating from breakpoints with at least one H3K9me3 ChIP-seq peak in a 10 kb window around the breakpoint to those without neighboring H3K9me3 marks . To test whether the results of this analysis depend on the reference epigenomes that we selected , we performed this comparison for different tissue-of-origin associations ( Supplementary file 2 ) and observed a significant or marginally significant difference in length distributions in all cases ( p ≤ 0 . 05 ) . All results are described using GISTIC2 'region limits' . In most cases , the results hold true independent of whether 'wide peak boundaries' or 'region limits' are used to define breakpoints and independent of excluding only one or both breakpoints of CNA regions that are bounded by a genomic coordinate that falls into centromeric or telomeric regions . Exceptions are the enrichment of the chromatin state 'ZNF genes and repeats' and the link between CNA length and H3K9me3 enrichment , where we found significant differences only when defining CNA breakpoints as GISTIC2 'region limits' . We computed the proportion of heterochromatin from the 18-state chromatin model as defined by the Roadmap Epigenomics project ( Kundaje et al . , 2015 ) . Likewise , we retrieved RNA expression data for protein-coding genes from Roadmap Epigenomics and we computed the Pearson correlation between the heterochromatin fraction and RNA expression for each healthy cell type for which we had RNA expression and chromatin state data . For the analyses in which we correlated the heterochromatin fractions of tissues with CNA number and length in the corresponding cancer type , p-values testing for significance of Spearman's rho were computed with the R function cor . test , which implements the Algorithm AS 89 with Edgeworth series approximation ( Best and Roberts , 1975 ) . We repeated the analysis for all possible combinations of healthy tissues as well as for 1 , 000 random associations between heterochromatin proportions and cancer-type-specific CNA numbers and lengths . The Spearman correlation between heterochromatin percentage and CNA number ( p < e-10; Mann-Whitney-Wilcoxon test; Figure 5—figure supplement 1 ) or CNA length ( p < e-16; Mann-Whitney-Wilcoxon test; Figure 5—figure supplement 2 ) within the permutated reference epigenome set are significantly higher than in the random set . This holds true irrespective of whether ovarian cancer is excluded from the test statistic or not . | Cancer is a genetic disease that develops when a cell’s DNA becomes altered . There are several different types of DNA alterations and one that is frequently seen in cancer cells is known as a “copy number alteration” ( or CNA for short ) . These CNAs arise when breaks in the DNA are repaired incorrectly , leading to some pieces of DNA being multiplied while others are lost . Ultimately , CNAs contribute to cancer growth either by providing extra copies of genes that drive tumour development or by deleting genes that normally protect against cancer . However , it is not known why patients with some types of cancer tend to have more CNAs than others and why some DNA regions are particularly susceptible to this type of alteration . Cramer et al . asked whether cancer patients have any other genetic mutations that might be linked with having many or few CNAs . Analysing datasets from almost 6000 patients with 20 different types of cancer showed that mutations in several genes are linked to a higher or lower number of CNAs in patients . Cramer et al . called the proteins encoded by these genes “copy number instability modulators” ( or CONIMs for short ) . Further investigation revealed that several of these CONIM proteins can change the way DNA is packaged inside cells . Furthermore , many of the regions of DNA that are vulnerable to CNAs in cancer cells are tightly packaged within healthy cells . These data suggest that the three-dimensional arrangement of DNA in cells influences where CNAs occur . The next step following on from this work is to find out exactly how the CONIM proteins influence the formation of CNAs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology"
] | 2016 | A network of epigenetic modifiers and DNA repair genes controls tissue-specific copy number alteration preference |
Ribosome assembly is a complex process involving the folding and processing of ribosomal RNAs ( rRNAs ) , concomitant binding of ribosomal proteins ( r-proteins ) , and participation of numerous accessory cofactors . Here , we use a quantitative mass spectrometry/electron microscopy hybrid approach to determine the r-protein composition and conformation of 30S ribosome assembly intermediates in Escherichia coli . The relative timing of assembly of the 3′ domain and the formation of the central pseudoknot ( PK ) structure depends on the presence of the assembly factor RimP . The central PK is unstable in the absence of RimP , resulting in the accumulation of intermediates in which the 3′-domain is unanchored and the 5′-domain is depleted for r-proteins S5 and S12 that contact the central PK . Our results reveal the importance of the cofactor RimP in central PK formation , and introduce a broadly applicable method for characterizing macromolecular assembly in cells .
The ribosome catalyzes protein biosynthesis and is essential for cell growth . In Escherichia coli ( E . coli ) , the 70S ribosome is a large ( 2 . 4 MDa ) ribonucleoprotein consisting of a small ( 30S ) and large ( 50S ) subunit . Because ribosome biogenesis is complex and taxing on the metabolic resources of the cell , the process is tightly regulated . The efficiency of ribosome assembly is so directly tied to cell growth that even slight defects in assembly confer a significant selective disadvantage , and strong defects can threaten cell survival . In addition , ribosomes must be assembled accurately to ensure the fidelity of protein synthesis . A large number of accessory factors have been implicated in the regulation and efficiency of ribosomal production , although the precise roles for many of these factors remain unknown ( Reviewed in Wilson and Nierhaus , 2007; Shajani et al . , 2011 ) . Remarkably , the 30S subunit can be reconstituted in vitro from 16S ribosomal RNA ( rRNA ) and 20 ribosomal proteins ( r-proteins ) in a high temperature , high Mg2+ environment ( Traub and Nomura , 1969 ) . Early work by Nomura and colleagues established the order and dependencies of r-protein binding in the assembling 30S subunit under equilibrium conditions ( Figure 1A ) ( Mizushima and Nomura , 1970; Held et al . , 1973 , 1974 ) . In the early stages of assembly , primary r-proteins bind directly to the 5′- , central and 3′-domains of 16S rRNA . These initial r-protein binding events lead to changes in the rRNA structure , and facilitate subsequent binding of secondary and tertiary r-proteins ( Held et al . , 1973; Stern et al . , 1989 ) . More recent studies using time-resolved hydroxyl radical RNA structure probing ( Adilakshmi et al . , 2008 ) , fluorescence correlation spectroscopy ( Ridgeway et al . , 2012 ) , single-molecule fluorescence resonance energy transfer ( Kim et al . , 2014 ) , pulse-chase monitored by quantitative mass spectrometry ( Talkington et al . , 2005; Bunner et al . , 2010a ) , and time-resolved negative stain electron microscopy ( Mulder et al . , 2010 ) added valuable insight into the dynamics and kinetics of RNA folding , r-protein binding , and immature subunit conformations throughout the assembly process . These studies have revealed that even in the presence of r-protein binding dependencies , assembly can proceed through multiple parallel pathways . In addition , a large body of evidence indicates that misfolded rRNA structure leads to stable kinetic traps during in vitro 30S reconstitution , inhibiting the binding of several secondary and tertiary r-proteins and limiting the efficiency of the reconstitution ( Reviewed in Sykes and Williamson , 2009 ) . 10 . 7554/eLife . 04491 . 003Figure 1 . High-throughput qMS/EM analysis of assembly intermediates from WT E . coli . ( A ) Nomura assembly map; adapted . Three major regions of 16S rRNA are labeled at top . Arrows represent binding dependencies , with primary , secondary and tertiary proteins labeled on left . ( B ) Sucrose gradient chromatogram ( absorbance at 254 nm ) for WT E . coli lysate . 30S peak fractions analyzed by qMS and EM are labeled . ( C ) R-proteins clustered by relative abundance in 30S particles across sucrose gradient fractions 1–5 with a blue to red gradient representing high to low relative abundance . Relative abundance of each r-protein was normalized to that of S4 . Gray boxes indicate r-proteins for which no peptides were detected . Clusters of r-proteins from more abundant to less abundant across the gradient are highlighted on the right as blue , green , yellow , and red . ( D ) Negative stain EM class averages for sucrose gradient fractions 2–5 ( labeled at bottom ) . Classes were obtained by reference-free maximum likelihood alignment and classification and are sorted by Group . Histogram at top shows the fractional contribution of particles from each dataset to each Group . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 00310 . 7554/eLife . 04491 . 004Figure 1—figure supplement 1 . Raw micrographs and initial class averages for WT F2 negative stain EM data set . ( A ) Exemplar image collected on FEI T12 transmission electron microscope operating at 120 keV and equipped with a Tietz TemCam-F416 4k × 4k CMOS camera at nominal magnification of 52000× . Scale bar represents 200 nm . ( B ) Particle picks selected by reference-free DoG picking . ( C ) Initial class averages obtained by reference-free maximum likelihood alignment and classification . Several E . coli complexes are readily observed , including pili ( 26 ) , GroEL ( 27 and 32 ) , glutamine synthetase complex ( 33 ) and pyruvate dehydrogenase and 2-oxoglutarate dehydrogenase complexes ( 49 ) . Components of these complexes were detected by MS-MS analysis of the sucrose gradient fractions ( Supplementary file 1 ) . 30S intermediates were identified by comparison with previous EM studies ( Mandiyan et al . , 1991; Mulder et al . , 2010 ) , and by further experiments including affinity purification described in text . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 00410 . 7554/eLife . 04491 . 005Figure 1—figure supplement 2 . Hierarchical clustering of class averages . ( A ) Class averages obtained from data sets for four WT sucrose gradient fractions ( ML2D , 15 classes ) . The final classes were aligned to one another using EMAN align2d ( Ludtke et al . , 1999 ) . ( B ) Dendrogram showing clustering of class averages from ( A ) . Groups are labeled on bottom , and classes are ordered based on fraction and Group in Figure 1D . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 005 In the cell , 30S assembly is fast and efficient , proceeding with the help of numerous assembly factors , including enzymes that directly modify the 16S rRNA and r-proteins , as well as a number of RNA-binding chaperones and GTPases that assist in RNA folding ( Reviewed in Wilson and Nierhaus , 2007; Shajani et al . , 2011 ) . Previous studies suggest that some assembly factors , such as RimM and RbfA , may promote efficient assembly by binding co-transcriptionally to the nascent rRNA to facilitate folding and to prevent the formation of kinetic traps ( Williamson , 2003; Clatterbuck Soper et al . , 2013 ) . In addition , factors may guide rearrangement of rRNA structure at specific points during the assembly process , as is likely the case during the re-structuring and cleavage of the 16S 5′-leader sequence late in assembly ( Dammel and Noller , 1995 ) . Cryo-EM reconstructions indicate that several assembly factors , including RbfA ( Datta et al . , 2007; Jomaa et al . , 2011b ) , the GTPases Era ( Sharma et al . , 2005 ) and RsgA ( Guo et al . , 2011 ) , and the 16S rRNA methyltransferase KsgA ( Boehringer et al . , 2012 ) , bind immature subunits and block them from prematurely entering the translation cycle . Many assembly factors appear to be functionally related , forming a complex network of interconnected activities ( Bylund et al . , 1998; Lu and Inouye , 1998; Inoue et al . , 2003; Campbell and Brown , 2008; Goto et al . , 2011; Connolly and Culver , 2013 ) . Other factors , such as RimP , have been implicated in 30S assembly but have no known connection to the overall assembly factor network ( Nord et al . , 2009; Bunner et al . , 2010b ) . One of the major obstacles hindering studies of in vivo ribosomal biogenesis stems from the complexity of the assembly process and heterogeneity of the incompletely assembled intermediates . Multiple parallel pathways are operative for assembly , giving rise to a variety of intermediates containing distinct sets of r-proteins ( Talkington et al . , 2005; Mulder et al . , 2010 ) . Quantitative mass spectrometry ( qMS ) provides a high-throughput method for precisely measuring the relative levels of proteins in a complex mixture ( Charollais et al . , 2003 ) . Recently , qMS has been used to determine the composition of r-proteins within ribosomal assembly intermediates isolated from cells , revealing the binding dependencies of late-associating r-proteins on earlier binding r-proteins and assembly factors ( Chen and Williamson , 2013; Clatterbuck Soper et al . , 2013; Guo et al . , 2013; Leong et al . , 2013 ) . Single-particle EM is also an ideal technique for analyzing heterogeneous samples due to the development of powerful alignment and classification schemes that enable the identification of sub-populations of particle conformations within a single sample ( Frank , 2006 ) . Automated data collection and downstream image processing has greatly improved the throughput of single-particle EM analysis , facilitating the rapid analysis of even highly heterogeneous samples in a short period of time ( Suloway et al . , 2005; Lander et al . , 2009 ) . In addition , automated random conical tilt ( RCT ) data collection enables three-dimensional reconstructions of each subpopulation identified in a sample ( Radermacher et al . , 1987; Yoshioka et al . , 2007; Voss et al . , 2010 ) . These advances in automation were previously applied to study the in vitro assembly of the 30S subunit , resulting in a quantitative visualization of assembly intermediate conformations present at various stages of the in vitro reconstitution process ( Mulder et al . , 2010 ) . In order to visualize the distributions of 30S assembly intermediates present in the cell we have developed a hybrid approach combining qMS and single-particle EM to provide unprecedented insight into the composition and structure of ribosome assembly intermediates from cellular lysates . Our approach exploits the exquisite facility of both techniques to analyze the heterogeneous samples generated by fractionating crude E . coli lysates using sucrose gradient ultracentrifugation . This one-step sample preparation eliminates the use of affinity tags or purification steps that may inadvertently lead to the exclusion of early intermediates . For each gradient fraction , ribosomal protein levels were measured using qMS , revealing r-proteins present or depleted within assembly intermediates . In parallel , gradient fractions were analyzed using single-particle negative stain EM to elucidate the structures of assembly intermediates and mature ribosomes present in each sample . These techniques were used to analyze E . coli 30S subunit assembly in wild type ( WT ) cells and in several assembly factor deletion strains . These studies revealed a novel intermediate , where the central pseudoknot ( PK ) that connects the 5′-body domain of the 16S rRNA with the 3′-head domain , is unformed , although the 3′-head domain is partially assembled . Deletion of the assembly factor RimP causes a striking defect in central PK stability and results in the depletion of central PK-adjacent proteins S5 and S12 , and late-binding proteins S2 , S3 and S21 . Together , our data suggest that central PK formation can occur either before or after head domain formation . Furthermore , our data implicate RimP in efficient central PK formation and in the subsequent incorporation of tertiary r-proteins S2 , S3 , S5 , S12 and S21 . In addition to providing novel insights into ribosome assembly , our approach represents a generalizable toolkit for studying the assembly of supramolecular structures in heterogeneous cellular samples .
We previously demonstrated that ribosome assembly intermediates could be separated from mature subunits by sucrose gradient centrifugation ( Chen and Williamson , 2013 ) . Quantitative MS ( qMS ) analysis of sucrose gradient fractions revealed that individual r-protein levels in early 30S and early 50S fractions are variable and consistent with the expected binding dependencies based on the Nomura ( Held et al . , 1974 ) ( Figure 1A ) and Nierhaus ( Herold and Nierhaus , 1987 ) maps , respectively . For example , in early 30S fractions , r-protein levels cluster into four groups with 5′-domain primary and secondary binders such as S4 and S8 in the most abundant group and late binders such as S2 , S3 and S21 in the most depleted group . The presence of four distinct groups of r-protein levels suggests that several intermediate species with varied r-protein compositions are present in early 30S fractions . In order to investigate the subpopulations of intermediates that accumulate in vivo , we combined our qMS analysis with single particle EM , applied to fractions collected from the 30S peak . E . coli BW25113 ( Keio collection background strain/WT ) ( Baba et al . , 2006 ) grown in M9 minimal media was harvested during exponential growth to ensure active production of ribosomes and the steady-state presence of assembly intermediates in the culture . The cell lysate was resolved by sucrose gradient centrifugation and five fractions encompassing the entire 30S peak were collected for qMS and EM analysis , allowing for a direct comparison of r-protein compositions with the observed particle conformations ( Figure 1B ) . For qMS analysis , an equimolar amount of 15N-labeled ribosomes was added to each fraction prior to trypsin digestion . The resulting combination of 14N- and 15N-labeled peptides was quantified by LC-MS with the 15N-labeled peptides used as a reference . Peptides were detected for all r-proteins with the exception of later binding proteins with very low abundance in fraction 1 ( S2 , S3 , S13 , S19 , S21 ) and S17 for fractions 1 , 4 and 5 ( Figure 1C ) . The isotope distribution fits for S17 peptides are often poor for both the experimental and reference sample , preventing unambiguous assignment of these peptides and necessitating their exclusion from the qMS analysis for some fractions . For convenience , the abundance of each r-protein in the experimental sample was normalized to that of the early binder S4 , which is expected to be present in all assembling 30S particles . R-proteins were then grouped by the profile of their relative abundances using hierarchal clustering . The protein abundance data is consistent with the Nomura map and previous qMS analysis , with the earliest fractions containing 30S particles with the primary and secondary binders of the 5′- and central domain bound ( Figure 1A , C ) ( Chen and Williamson , 2013 ) . In contrast , most tertiary binders of the 5′- and central domain ( S5 , S11 , S12 ) and most 3′-domain binders ( S7 , S9 , S10 , S13 , S14 , S19 ) are only abundant in later fractions . Moreover , tertiary central and 3′-domain binders ( S2 , S3 , S21 ) are the least abundant r-proteins in 30S particles across all fractions . In parallel , fractions from the sucrose gradient were prepared for EM analysis . Images of negatively stained sample were collected for each fraction using automated methods ( Suloway et al . , 2005; Yoshioka et al . , 2007 ) and analyzed using single-particle methods ( Lander et al . , 2009; Mulder et al . , 2010 ) . A mixture of 30S intermediate particles and other abundant large cellular complexes , such as GroEL , is readily observable in raw images collected for the sucrose gradient fractions ( Figure 1—figure supplement 1A ) . Fraction 1 contained a particularly low abundance of 30S particles relative to other complexes , and as a result this fraction was omitted from further EM analysis . Given the heterogeneity of particles observed in the raw images , a reference free Difference of Gaussian particle picking method was used to select particles with diameters ranging from 100–300 Å ( Voss et al . , 2009 ) ( Figure 1—figure supplement 1B ) . Particles were aligned and classified using iterative rounds of reference-free alignment , removing non-ribosomal particles between each iteration ( Figure 1—figure supplement 1C , also See ‘Materials and methods’ section ) . The final set of 60 class averages were compared using hierarchical clustering , revealing five major groups of 30S particles at various stages of assembly ( Figure 1B , Figure 1—figure supplement 2 ) . Four of these Groups ( I , III , IV and V ) resemble the four groups characterized in previous time-resolved EM studies of in vitro 30S reconstitution ( Mulder et al . , 2010 ) , while the Group II class averages were observed in the previous study , but were uncharacterized . In addition to this fraction-by-fraction two-dimensional analysis , random conical tilt ( RCT ) analysis was performed for fractions from the center of the 30S peak ( fractions 3–4 ) , enabling the reconstruction of 3D maps ( Yoshioka et al . , 2007; Voss et al . , 2010 ) from representative 2D class averages from each Group ( Figure 2 ) . Three-dimensional volumes provided additional insight into the stage of assembly of each conformation observed in the 2D class averages . 10 . 7554/eLife . 04491 . 006Figure 2 . RCT reconstructions of assembly intermediates from WT E . coli . Representative RCT reconstructions aligned with crystal structure of the mature 30S subunit ( PDB 2AVY ( Schuwirth et al . , 2005 ) shown in gray ) . ( A ) Group I intermediate . All density for 3′-domain is missing . ( B ) Group II intermediate . Head domain density is detached from the body/platform domain . ( C ) Group III intermediate . The head density is angled away from the body domain , and density for S2 and S3 are missing . A 30 Å filter was applied to the PDB chains for S2 and S3 , and the resulting volumes ( gray surfaces ) lie outside of the RCT volume . ( D ) Group IV intermediate . S2 and S21 density is missing , and the PDB chains for these proteins are shown as gray 30 Å filtered maps as in ( C ) . ( E ) Group V intermediate missing S2 and S21 , with the PDB chains for these proteins shown as gray 30 Å filtered maps as in ( C ) . ( F ) Group V intermediate missing S3 density . The PDB chain for S3 is shown as a gray 30 Å filtered map as in ( C ) . ( G ) Fully mature Group V . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 006 Consistent with the qMS data , the majority of early assembly intermediates ( Groups I–III ) observed across the 30S peak are present in the early fractions , with a gradual build up of late intermediates and mature subunits ( Groups IV–V ) toward the end of the peak ( Figure 1D ) . The predominant conformation ( Group I ) present in early 30S peak fraction 2 is the earliest identifiable intermediate , encompassing the 5′-body and platform regions , but wholly lacking density for the 3′-head domain ( Figure 1D , Figure 2A ) . The abundance of Group I particles is consistent with the depletion of 3′-domain proteins ( S2 , S3 , S7 , S9 , S10 , S13 , S14 , S19 ) observed by qMS ( Figure 1C ) . Both Groups II and III contain head domain density , although the location of the domain relative to the body/platform regions is strikingly different in the two conformations ( Figure 1D , Figure 2B , C ) . In Group II , the head appears to be completely unanchored from the platform domain , swinging well away from the helix 23/S11 interaction region ( Figure 2B ) . In contrast , in Group III the head appears to be docked along the platform domain as it is in mature subunits , but angled away from its final resting spot along the body domain ( Figure 2C ) . The vastly different locations of the head domain in Group II and Group III can be readily observed in RCT reconstructions of classes from these Groups ( Figure 2B , C ) . RCT volumes for classes from both Groups suggest that late-binding proteins S2 and S3 may be missing from these particles , consistent with the relatively low levels of the proteins observed by qMS ( Figure 1C , Figure 2B , C ) . In the three RCT reconstructions obtained for classes from WT Group II , the location of the head varies in relation to the 5′-domain . Similarly , the location of head density varies substantially in class averages assigned to Group III , as revealed by focused 2D classification of the head region using custom masks created with Maskiton ( Video 1 ) ( Yoshioka et al . , 2013 ) . This type of ‘hinged’ head movement is similar to conformations observed in in vitro assembly intermediates , as well as intermediates observed in strains lacking the late-acting assembly factor RimM ( Mulder et al . , 2010; Guo et al . , 2013; Leong et al . , 2013 ) . 10 . 7554/eLife . 04491 . 007Video 1 . Analysis of Group III head density movement using Maskiton . A total of 3490 Group III particles were aligned to a reference image using SPIDER ( Frank et al . , 1996 ) . The aligned stack was uploaded to the Maskiton server ( www . maskiton . scripps . edu , [Yoshioka et al . , 2013] ) , and the Maskiton web interface was used to apply a mask to the head region of the averaged stack . Classifications of the masked region were run for 1000 iterations . The resulting 16 class average images were compiled into a movie using QuickTime Pro 7 ( Apple ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 007 Classes belonging to Groups IV and V represent very late assembly intermediates and mature subunits , as observed by 2D class averages and 3D RCT reconstructions ( Figure 1D , Figure 2D–H ) . Groups IV and V are highly represented in later fractions from the 30S peak , consistent with qMS data showing nearly stoichiometric levels of most r-proteins in these fractions . RCT reconstructions of Group IV classes appear to lack some density in the platform region , suggesting that these particles may primarily be late intermediates depleted of very late-binders such as S21 and S2 ( Figure 2D ) . Similarly , for Group V , one subgroup of classes that appears to be missing density in the same region of the platform was detected ( Figure 2E ) . A second subgroup of classes clearly missing density for S3 but containing density for S2 and S21 was observed ( Figure 2F ) , in agreement with previous in vitro observations showing that S2 can bind prior to S3 ( Mulder et al . , 2010 ) . Other Group V particles appear to be fully mature ( Figure 2G ) , suggesting that 70S ribosomes may have disassociated during sample preparation . Together , these late-intermediate classes account for the depletion of S2 , S3 and S21 observed in the qMS analysis of the late fractions of the 30S peak . Among the intermediates observed in WT E . coli , the class averages present in Group II were the most intriguing . The vastly different location of the head in these particles compared to Group III particles suggests two alternate pathways for formation of the 3′-domain , either before or after docking of the head onto the platform region of the central domain . In 16S rRNA , the central domain ( 16S nt:567-912 ) and 3′-major domain comprising the head ( 16S nt:920-1396 ) are connected by a short linker ( 16S nt:913-919 ) , which forms a long range pseudoknot interaction with helix 1 ( h1 ) at the 5′-end ( Figure 3A , B ) . This central pseudoknot ( PK ) is formed by helix 2 ( h2 ) , and is the primary structural feature that leads to the positioning of the head domain upon the platform domain ( Figure 3A , B ) . The detachment of the head from the platform in Group II classes suggests that these particles may have unstable or unformed central PK regions . To probe the stability of the RNA in the central PK region , an RNase H cleavage assay was developed using a DNA oligonucleotide anti-sense to the 3′-end of helix 27 ( h27 ) and the 3′-strand of h2 ( 16S nt:906-920 , Figure 3A ) . This ‘anti-PK’ oligo was incubated with samples encompassing the 30S sucrose gradient peak , allowing the oligo to specifically anneal to assembly intermediates in which the central PK is unformed . The hybridized RNA was subsequently digested with RNase H , resulting in two 16S products encompassing the 5′- ( ∼900 nt ) and 3′-domains ( ∼600 nt ) ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 04491 . 008Figure 3 . Survey of central pseudoknot stability upon deletion of assembly factors . ( A ) 16S rRNA secondary structure ( red: 5′-body; green: central; blue: 3′-head; yellow: 3′-minor ) , with central PK region boxed . The sequences of helix 2 ( h2 ) , which forms the central PK , and adjacent secondary structures helix 1 ( h1 ) , helix 27 ( h27 ) and helix 28 ( h28 ) are shown at right . The sequence targeted by an anti-sense DNA oligo ( 16S rRNA nt 906-920 ) is highlighted in yellow . ( B ) Crystal structure ( PDB: 2AVY ( Schuwirth et al . , 2005 ) ) of the 30S subunit from E . coli . 16S rRNA is shown as backbone ribbon and colored as in ( A ) . Only r-proteins S5 ( orange ) and S12 ( blue ) are shown . A close-up of the central pseudoknot and adjacent rRNA helices and r-proteins is shown at right . ( C ) Anti-PK hybridization ( 500 pmol oligo ) /RNase H cleavage of 16S rRNA from 30S peak sucrose gradient fractions for seven different E . coli strains . The average fraction of rRNA cleaved ( product/total RNA ) from three replicates is plotted . Error bars represent the standard deviation of fraction cleaved between the three replicates . ( D ) Abundance of Group II particles in seven E . coli strains as measured by negative stain EM . 10000 30S assembly intermediate particles from each strain were combined into a single stack of 70 , 000 particles . The stack was subjected to reference-free maximum likelihood alignment ( Figure 3—figure supplement 2 ) . The number of particles from each strain contributing to Group II classes is plotted in the histogram . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 00810 . 7554/eLife . 04491 . 009Figure 3—figure supplement 1 . Oligo hybridization/RNase H assay . ( A ) Sucrose gradient chromatograms ( absorbance at 254 nm ) for seven strains . Lines indicate the portion of the 30S peak that was collected and used for RNase H assay and EM analysis . ( B ) RNase assay cleavage detected on 2% agarose gel stained with ethidium bromide . Products are labeled on right . The anti-h21 oligo was used as a control , targeting a region of the rRNA that should be stable and inaccessible in all strains . ( C ) Quantitation of RNase H cleavage ( average of three replicates , error bars represent standard deviation ) . For each lane , the intensity of intact 16S rRNA and cleavage products was measured using image quant . Cleavage products detected in the ‘no oligo’ lane were subtracted from other lanes for the same strain , to account for background cleavage that may have occurred before or during the cleavage assay . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 00910 . 7554/eLife . 04491 . 010Figure 3—figure supplement 2 . Class averages from seven E . coli strains . ( A ) Direct comparison of the distribution of assembly intermediate Groups in data sets collected for samples from seven different strains . A combined stack of 70 , 000 particles ( 10 , 000 particles from each strain ) was subjected to reference-free maximum likelihood alignment . The resultant classes were aligned to one another , then clustered using the Mathematica script described in the experimental methods . The number of particles from each strain contributing to each Group is plotted in the histogram . ( B ) Reference-free maximum likelihood class averages obtained from individual data sets collected for each sample from Figure 3—figure supplement 1A . The classes are sorted by Group within each strain . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 010 Previously , several 30S assembly factors have been implicated in central PK formation providing a motivation for comparing the PK accessibility for both WT E . coli and deletion strains using the RNase H assay . The level of central PK formation was determined in E . coli BW-25113 and deletion strains of five assembly factors: RimM , RbfA , RimP , RsgA ( YjeQ ) and KsgA . Alongside these strains , the effects of the plasmid pU23 ( Dammel and Noller , 1993 ) , that contains a copy of the rRNA operon rrnB bearing a C23U mutation within h1 of 16S RNA , was determined . This mutation destabilizes h1 and favors formation of an alternate stem-loop structure within the leader sequence of the pre-cleavage 17S rRNA , thus disrupting formation of h2 . When transformed into E . coli strain BW-25113 , pU23 displays a similar dominant negative cold-sensitive phenotype observed previously in E . coli strain DH1 ( Dammel and Noller , 1993 ) . Each strain was grown at 37°C and harvested during exponential growth , and the 30S peak from the sucrose gradient of each lysate was collected for analysis ( Figure 3—figure supplement 1A ) . RNase H cleavage in the presence of the anti-PK oligo led to accumulation of two products of the expected size ( Figure 3—figure supplement 1B ) . Very little product was observed for WT BW-25113 , indicating that the majority of intermediates contain fully formed and stable central PKs in this strain ( Figure 3C , Figure 3—figure supplement 1B , C ) . In contrast , BW-25113+pU23 and all deletion strains displayed substantial but variable amounts of cleavage , suggesting varying degrees of central PK exposure ( Figure 3C , Figure 3—figure supplement 1B , C ) . Among the deletion strains , only E . coli lacking RimP displayed greater cleavage than BW-25113+pU23 , suggesting that the ΔrimP strain may have the strongest defect in central PK formation . Consistent with a potential role in PK formation , RimP has previously been shown to accelerate binding of the central PK-associated r-proteins S5 and S12 in in vitro 30S assembly assays ( Bunner et al . , 2010b ) . Negative stain EM datasets for each of the strains were collected in parallel to the RNase H assay , in order to observe the extent of Group II particle accumulation across the 30S peak ( Figure 3—figure supplement 2 ) . Group II particles were most abundant in ΔrimP and BW-25113+pU23 ( Figure 3C , Figure 3—figure supplement 2 ) , and overall the levels of particles belonging to Group II in each strain was consistent with the amount of cleavage observed for anti-PK-oligo-dependent RNase H cleavage ( Figure 3C , D ) . Based on these results , the ΔrimP strain was chosen for further investigation into the composition and conformation of Group II assembly intermediates . In order to determine the effect of RimP deletion on the abundance of specific r-proteins in assembling 30S particles , cell lysate from WT and ΔrimP strains were prepared for qMS analysis . WT and ΔrimP cells were grown in 14N- and 50% 15N-labeled M9 media respectively . Equivalent amounts of lysates from each strain were purified using sucrose gradient centrifugation . In agreement with previous studies , the ΔrimP strain shows an increase in 30S and 50S particles with a concomitant decrease in 70S particles , compared to the WT strain ( Figure 4A ) ( Nord et al . , 2009 ) . To directly compare the abundance of specific r-proteins in assembling 30S particles between the WT and ΔrimP strains , equivalent amounts of each cell lysate were combined and purified using sucrose gradient centrifugation ( Figure 4B ) . Fractions were collected across the 30S peak and analyzed by qMS as previously described , using 15N-labeled 70S particles as a reference ( Chen and Williamson , 2013 ) . Hierarchal clustering of r-protein abundances normalized to that of S4 , reveals significant depletion of S3 and S21 relative to other r-proteins in both strains , across all fractions ( Figure 4C ) . Furthermore , S2 and S12 are more depleted in the ΔrimP strain relative to the WT strain . 10 . 7554/eLife . 04491 . 011Figure 4 . Comparison of 30S assembly in WT and ΔrimP by qMS . ( A ) Overlay of sucrose gradient chromatograms ( absorbance at 254 nm ) for WT ( black ) and ΔrimP ( red ) . ( B ) Sucrose gradient chromatogram of combined WT and ΔrimP ( blue ) lysates , with fractions analyzed by qMS labeled 1–4 . ( C ) R-proteins in 30S particles in WT ( black ) and ΔrimP ( red ) , across fractions 1–4 ( labeled at top ) clustered by relative abundance , with a red to blue gradient representing high to low relative abundance . Relative abundance of each r-protein was normalized to that of S4 . Gray boxes indicate r-proteins for which no peptides were detected . The cluster comprising r-proteins that are the least abundant in both strains ( S3 and S21 ) and preferentially depleted in ΔrimP ( S2 and S12 ) is highlighted by a red box . ( D ) Fraction labeling of 30S r-proteins in 70S particles versus their relative abundance in 30S particles in WT ( black ) compared to ΔrimP ( red ) . Data collected from cells labeled for 45 min . ( E ) Representative labeling kinetics for an early binder , S4 ( black ) compared to late binder , S12 ( red ) in ΔrimP . The maximum expected labeling rate is represented by the dashed grey line . Time course of experiment was fit ( bold black line for S4 ) to a previously reported pulse-labeling model to determine the precursor pool size ( P ) of each r-protein ( Chen et al . , 2012 ) . ( F ) Precursor pool size compared to relative abundance of r-proteins in 30S assembly intermediates in ΔrimP . R-proteins with large pool sizes and high abundance in 30S assembly intermediates are boxed in black while those with small pool sizes and low abundance in 30S assembly intermediates are boxed in red . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 01110 . 7554/eLife . 04491 . 012Figure 4—figure supplement 1 . Pulse labeling experiment to monitor 30S assembly . ( A ) Model of flow of 15N-label in 30S subunits during 70S assembly highlighting domain formation ( body-red , platform-green , head-blue ) . ( B ) Isotope distribution of representative peptide with material synthesized pre-pulse ( 100% 14N ) in red , material synthesized post-pulse ( 50% 15N ) in green and material from the reference ( 100% 15N ) , used for peptide identification in orange . ( C ) Fraction labeled values of r-proteins in 30S particles compared to 70S particles for WT ( black ) and ΔrimP ( red ) with a blue to red gradient representing high to low fraction labeled value . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 012 The observed incomplete 30S particles could either be degradation products , dead-end assembly intermediates or on-pathway intermediates that eventually mature into 30S and 70S ribosomes . To distinguish among these distinct particle types , we performed a previously described pulse-labeling approach that monitors the flow of 15N-label during 70S assembly ( Figure 4—figure supplement 1A ) ( Chen et al . , 2012 ) . First , WT and ΔrimP cells were pulse-labeled over a time-course at mid-log phase and the 30S particles and 70S ribosomes were purified and analyzed by qMS . The fraction of label incorporated into 70S ribosomes ( fL ) was determined for each r-protein by comparing the isotope amplitude of the isotope distributions pre-pulse ( 14N ) and post-pulse ( 50% 15N ) ( Figure 4—figure supplement 1B ) . Fully assembled 70S particles greatly outnumber assembling 30S intermediates during exponential growth , resulting in faster labeling of the intermediates relative to 70S ribosomes and their degradation products . Therefore , assembly intermediates should have a higher fL value than 70S ribosomes and their degradation products . The qMS analysis of the fractions containing 30S particles and 70S ribosomes from the WT and ΔrimP strains reveals that the 30S particles in the early fractions of both strains have higher fL values than the 70S ribosomes ( Figure 4—figure supplement 1C ) . This indicates that the early fractions of both strains contain significant amounts of 30S assembly intermediates , and that these intermediates are competent to mature into 70S ribosomes . As assembly progresses , early binding proteins are incorporated into intermediates before late binding proteins , thereby spending more time bound to intermediates than the late binders . Therefore , post-pulse , r-proteins sequentially incorporated into on-pathway intermediates would have different fL values . Early binding r-proteins would have lower fL values than late binding r-proteins . Furthermore , any delay in binding of specific r-proteins to on-pathway intermediates would be reflected in their fL values in 70S ribosomes . By comparing the fL values of r-proteins in 70S ribosomes in the WT and ΔrimP strains to their relative abundance in the 30S intermediates , it can be seen that on average , WT r-proteins are more labeled than those in ΔrimP ( Figure 4D ) . This indicates that there is a delay in 30S assembly in the ΔrimP strain compared to the WT strain , corresponding to an accumulation of assembly intermediates . Moreover , the data show that in the ΔrimP strain , the most depleted r-proteins in the 30S intermediates are those with the highest fraction labeled values in the 70S ribosomes , and these correspond to the latest binding r-proteins . In order to confirm that the incomplete 30S particles are on-pathway assembly intermediates , the ΔrimP strain was pulse-labeled and the fL values of r-proteins in 70S ribosomes were measured for various time periods post-pulse . The fL values of the r-proteins as a function of the duration of pulse labeling were fit to equations describing the time-course of pulse labeling ( Figure 4E ) ( Chen et al . , 2012 ) . From these fits , the magnitude of the precursor pool size ( P ) of each r-protein was calculated , reflecting the quantity of unbound r-protein as well as r-protein bound to assembly intermediates . Since dead-end particles do not assemble into 70S ribosomes , they have no effect on the P measured for each r-protein . However , r-proteins in on-pathway intermediates would have P related to their abundance in the precursor pool , including both free protein and assembly intermediates . According to this model , r-proteins that bind early in assembly are the most abundant in intermediates , and are expected to have larger values of P than those that bind later and are less abundant . The data show that primary binding r-proteins such as S4 that are highly abundant in assembly intermediates in the ΔrimP strain have large values of P ( p = 0 . 12 or 12% ) , confirming that the assembly intermediates are on-pathway ( Figure 4E , F ) . Previous studies have shown that WT cells have precursor pool sizes less than 2% ( Chen et al . , 2012 ) . The large precursor pool sizes of early binders in the ΔrimP strain are further confirmation of a delay in assembly . In contrast , depleted r-proteins such as S12 have small precursor pools ( P ∼ 0 ) ( Figure 4E , F ) . The incorporation of these r-proteins is delayed during 30S assembly in the ΔrimP strain , and apparently , their synthesis and/or degradation is regulated such that they do not accumulate in their unbound form , resulting in a negligible pool size for these proteins . The pool sizes for r-proteins in the ΔrimP strain cluster into two distinct groups when compared to their relative abundance in the 30S assembly intermediates ( Figure 4F ) . One cluster contains r-proteins known to bind early in WT that are highly abundant and have large pools . In contrast , the second group is composed of the latest binders in WT , with one exception , S12 . This indicates a marked delay in S12 binding in the ΔrimP strain relative to WT . These late binding r-proteins are depleted in intermediates in the ΔrimP strain and have small pools . Outliers include S10 , which is known to have extra-ribosomal functions ( Friedman et al . , 1981; Mason and Greenblatt , 1991 ) , S14 , which is exchangeable ( Pulk et al . , 2010 ) and S20 , which is non-stoichiometric ( Hardy , 1975; Tal et al . , 1990 ) . Together , these pulse-labeling data show that the intermediate assembly species are on-pathway and that incorporation of S12 and late-binding proteins S2 , S21 and S3 is delayed during 30S assembly in the absence of RimP . Having determined the composition of on-pathway intermediates that accumulate upon deletion of RimP , fractions from the 30S peak of a ΔrimP strain sucrose gradient were additionally analyzed by negative stain EM as described above for WT to determine the distribution of conformations present across the gradient ( Figure 5A ) . Clustering analysis of class averages from the ΔrimP strain revealed five Groups of similar conformations to those observed in the WT dataset ( Figure 5B ) . However , the relative number of particles within each Group differs significantly between the two strains , with a dramatic increase in the abundance of Group II intermediates in ΔrimP relative to WT ( Figure 1D , Figure 5B ) . In contrast , the number of particles classified as Group I and III intermediates is similar in the two strains , while the number of late intermediates and mature subunits in Groups IV and V is decreased in ΔrimP relative to WT . In order to directly compare the particle conformations and distributions between WT and ΔrimP datasets , a combined stack of 10 , 000 randomly selected particles from each fraction from the two strains was classified using a reference-free maximum-likelihood protocol , followed by clustering analysis of the resulting class averages . For each Group from the cluster analysis , the number of particles contributed from each fraction of either WT or ΔrimP was calculated ( Figure 5C ) . Similar to the individual fractional analysis , this direct comparison of the combined datasets reveals a substantial relative accumulation of Group II intermediates across the entire 30S peak in ΔrimP . 10 . 7554/eLife . 04491 . 013Figure 5 . Direct comparison of WT and ΔrimP assembly intermediates by EM . ( A ) Overlay of sucrose gradient chromatograms ( absorbance at 254 nm ) for WT ( blue ) and ΔrimP ( red ) lysates , with 30S peak fractions analyzed by EM indicated . ( B ) Negative stain EM class averages for fractions 2–5 of ΔrimP sucrose gradient ( labeled at bottom ) . Classes were obtained by reference-free maximum likelihood alignment and classification and are sorted by Group . Histogram at top shows the fractional contribution of particles from each dataset to each Group . ( C ) Direct comparison of assembly intermediate abundance in WT ( shades of blue ) and ΔrimP ( shades of red ) strains . 10000 particles from each fraction for each strain were combined into a single stack with 80 , 000 particles . The stack was subjected to reference-free maximum likelihood alignment . For each strain , the number of particles from each fraction contributing to each Group are plotted as a stacked bar in the histogram , showing the contribution from each fraction and the overall number of particles in each group throughout the 30S peak . ( D ) Two-dimensional class averages and resulting 3D RCT volumes of Group II intermediates from the ΔrimP strain . The 3′-head domain location is highly variable between the different volumes . ( E ) Average density of the ten RCT volumes shown in ( D ) . ( F ) Variance analysis of the 10 RCT volumes shown in ( D ) . The average density from ( E ) is shown in gray , and the variance map is shown in red . ( G ) PDB model of the unanchored head conformation based on location of head in average density of RCT volumes . The 16S rRNA is shown and colored as in Figure 3A–B . A 50 Å filter was applied to the PDB , and the density is shown at right . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 013 The relative abundance of Group II particles in ΔrimP fractions enabled a more in depth analysis of the unusual conformation adopted by this intermediate . In both 2D class averages and 3D RCT volumes , the location of the head volume is highly variable ( residues 912–920 ) ( Figure 5D ) . To facilitate the 3D structural analysis , the RCT volumes in Figure 5D were aligned based on the body/platform density , and the average density and variance maps between the 10 volumes were calculated ( Figure 5E , F ) . The regions of high variance are mainly localized to the head domain , which can sample a substantial range of motion , from locations close to the S11-binding region in the platform domain to the S4-binding region of the body domain . Analysis of Group II particles in 2D by Maskiton ( Yoshioka et al . , 2013 ) recapitulates this result and additionally indicates that head movement is constrained by a short but highly flexible linker ( Video 2 ) , likely comprising the 3′-end of helix 27 and the 3′-strand of the unformed h2 ( 16S residues 910–919 ) . Indeed , the distance between the head and body among Group II RCT volumes is generally 20–40 Å , well within the range of lengths that could be accommodated by a 10-nt ssRNA . PDB models of the 16S rRNA for the body/platform region ( nt 1-909 ) and the head domain ( nt 920-1396 ) were docked into the average density from the 10 RCT volumes ( Figure 5E , G ) , and the distance between the two domains could be accounted by the length of the 910–919 linker . A 50-Å filter was additionally applied to the PDB model containing the 16S rRNA and r-proteins ( excluding S2 ) , revealing striking similarities to several RCT volumes in the amount of density observed for both the body/platform and head domain ( Figure 5G ) . The similarity in size suggests that Group II particles may contain a nearly complete complement of r-proteins in the head domain , and that head domain assembly can occur prior to central PK formation . 10 . 7554/eLife . 04491 . 014Video 2 . Analysis of Group II head density movement using Maskiton . Movie was generated as described for Video 1 , using a total of 3660 Group II particles . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 014 The qMS and negative stain EM analysis of ΔrimP 30S fractions revealed the r-protein levels and distribution of particle conformations in a complex mixture of various 30S particles and other large complexes . However , the sample complexity prohibited a detailed characterization of assembly intermediates in which the central PK is unformed . To reduce this sample complexity , an affinity purification protocol was developed using a biotinylated oligonucleotide anti-sense to the 3′-strand of h2 , similar to the anti-PK oligo used in the RNase H assays described above . This ‘capture oligo’ was incubated with samples containing 30S subunits and/or intermediates , then bound by NeutrAvidin agarose resin . After thorough washing of the resin , 30S particles annealed to the capture oligo were displaced by adding an excess of a DNA oligo bearing complete complementarity to the capture oligo . Using this purification strategy , 30S intermediates were enriched both from 30S peak sucrose gradient fractions and directly from ΔrimP crude lysate ( Figure 6—figure supplement 1A ) . In contrast , no 16S rRNA could be detected when purified mature 30S subunits were incubated with the capture oligo ( Figure 6—figure supplement 1A ) . The eluent from the ΔrimP intermediate affinity purification was first analyzed by negative stain EM . Whereas samples taken directly from sucrose gradient fractions yielded images containing a large number of non-ribosomal E . coli complexes , raw images obtained from the affinity purified sample contained no readily observable non-ribosomal particles , confirming the specific purification of 30S particles . Particle classification further indicated the specific enrichment of early assembly intermediates , with the majority of particles classifying into Group I and II class averages and very few Group III-V classes observed . In addition to the previously identified Groups , an additional class was observed that might be partial degradation products of Group II intermediates corresponding to the 3′-domain of the 30S subunit . Forward projections of the 30S 3′-domain filtered to 30 Å strongly resemble the observed class averages ( Figure 6—figure supplement 1B , C ) . In addition , the putative 3′-domain particles varied in abundance based on the amount of 16S rRNA degradation observed in the pulldown sample ( Figure 6—figure supplement 1D ) . Together , these observations suggest that particles in these classes contain the final ∼600 nt of the 16S rRNA , including the head domain and the 3′-minor domain containing helices 44 ( h44 ) and 45 . Indeed , density for h44 could be observed in some negative stain class averages , and was readily observed by cryo-EM ( see below , Figure 6—figure supplement 1C ) . The 3′-domain particles appear to be preferentially enriched , suggesting that the 16S:906-920 region is more exposed in these particles than in Group II particles . The 3′-domain particles likely result from non-specific cleavage of the exposed central PK region in Group II particles by contaminating RNases in the sample used for affinity purification . Efforts were made to limit sample degradation using RNase inhibitors , with limited success , further indicating the extent of rRNA exposure in the ΔrimP intermediates . Next , the protein composition of the affinity purified intermediates was analyzed by qMS as described above , using 15N-labeled 70S particles as a reference . The relative abundance of each r-protein , normalized with respect to S4 , shows that S2 and S12 are very depleted in 30S particles with PK instability , with partial depletion of S3 and S5 ( Figure 6A ) . The depletion of S2 , S3 and S12 in particles with PK instability is consistent with the earlier analysis of all the particles found in the 30S peak . The observed low abundance of S5 in the affinity purified particles could have been masked by the presence of a significant amount of particles containing S5 in the untreated sample . Furthermore , the affinity purified particles show a high abundance of most of the 3′-domain r-proteins relative to early 5′-domain binder , S4 . This is consistent with the observation that 3′-domain particles are preferentially enriched by the purification procedure . With the exception of S2 and S3 , the uniform abundance of all the 3′-domain r-proteins in the purified intermediates suggests that head domain formation is not perturbed until the very late stages of assembly . Some r-proteins , S17 and S21 could not be accurately quantitated in the qMS analysis due to poor fits of their isotope distributions , while the significantly high abundance of S18 is possibly due to its exchange in the 70S particles used as a reference . 10 . 7554/eLife . 04491 . 015Figure 6 . Cryo-EM and qMS analysis of affinity purified pre-central PK intermediates . ( A ) Relative abundance of 30S r-proteins grouped by domain bound ( body-red , platform-green , head-blue ) . Relative abundance of each r-protein was normalized to that of S4 . No peptides were detected for S17 and S21 . ( B ) Representative cryo-EM structure of Group I intermediate . All 3′-domain density is missing , beginning with h27 ( green ) and continuing through the head and the 3′-minor domain ( h44 and h45 , yellow ) . Close-ups of missing body domain r-proteins S5 ( orange ) and S12 ( magenta ) are shown at center . The PDB chains for S5 and S12 were filtered to 20 Å , and the resulting maps are located outside of the cryo-EM density . ( C and D ) Codimensional PCA variance analysis for Group I cryo-EM particles . ( C ) The average density for all 12 , 425 Group I cryo-EM particles . ( D ) Variance map for Group I cryo-EM particles ( red ) overlaid on average map ( gray ) . Regions of high variance are mainly localized in the platform domain . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 01510 . 7554/eLife . 04491 . 016Figure 6—figure supplement 1 . Affinity purification of pre-central PK intermediates using an anti-PK capture oligo . ( A ) Agarose gels ( stained with ethidium bromide ) showing results for affinity purification for ΔrimP sucrose gradient fractions 2–3 , ΔrimP lysate , and purified 30S subunits . 16S rRNA is not visible in later washes , but is visible in elution fractions for ΔrimP samples . ( B ) Class average of 3′-domain degradation product versus a forward projection of the 3′-domain filtered to 30 Å resolution . ( C ) Class averages from negative stain and cryoEM data sets with helix 44 density clearly visible , compared with a similar forward projection of the 3′-domain model . ( D ) Comparison of particle distribution between two affinity purification samples . In sample 1 , the input 16S rRNA was already heavily degraded , and the 3′-domain was preferentially enriched based on agarose gel analysis . In sample 2 , degradation was limited by the addition of RNasin ( Promega ) and reducing the amount of time for sample preparation . 5000 particles from negative stain data sets for each sample were combined into a single stack ( 10 , 000 particles ) , and subjected to reference-free maximum likelihood classification . The fraction of particles from each data set contributing to various conformations is plotted in the histogram . Putative 3′-domain classes are enriched in the degraded sample 1 , while Group II classes are enriched in the intact sample 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 01610 . 7554/eLife . 04491 . 017Figure 6—figure supplement 2 . 3D classification of Group I particles from cryo-EM data set of affinity-purified sample . ( A ) Reference-free class averages for all cryo-EM particles . Despite improvement of heterogeneity , several species are present including 3′-domain degradation products ( for example , classes 0–5 ) . ( B ) Fourier shell correlation curves for the four Frealign 9 classes . Resolutions are reported in ( C ) . The values are based on the 0 . 143 cutoff criterion . ( C ) Comparison of four structures obtained from classification using Frealign 9 , as in Figure 6B of the main text . The resolution of each structure is reported below the class number . In the bottom row , the structures are rotated by 180° . Differences are observed in the platform region , suggesting that protein content and rRNA structure may vary in this region . All structures lack density for S5 ( orange ) and S12 ( magenta ) and all rRNA residues starting with h27 ( green ) and including h44-45 ( yellow ) . ( D ) Comparison of reference-free class averages ( odd columns ) with re-projections ( even columns ) of Class 4 model from ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 017 The initial negative stain and qMS analysis indicated that purification of 30S intermediates from mature subunits and other E . coli complexes substantially reduced sample heterogeneity , allowing for cryo-EM analysis of the affinity-purified sample . Similar to the negative stain analysis , class averages from the cryo-EM data set revealed that the majority of particles are early intermediates ( Figure 6—figure supplement 2A ) . Despite efforts to limit sample degradation , a substantial number of 3′-domain degradation products were observed in the class averages , leading to an increase in the compositional heterogeneity of the data set . In addition , flexibility between the head and body domains in Group II intermediates led to significant conformational heterogeneity for these particles , and limited the ability to reconstruct 3D volumes for these particles . Efforts were therefore concentrated on 3D classification of the Group I cryo-EM particles . Following extensive sorting and classification of Group I particles , four reconstructions of this early intermediate were generated ( Figure 6B , Figure 6—figure supplement 2B–D ) . As expected , the Group I structures lack density for the entire 3′-domain , including all RNA and r-proteins comprising the head domain and helices 44 and 45 of the 3′-minor domain ( Figure 6B , Figure 6—figure supplement 2C ) . In addition , 16S rRNA helix 27 lies completely outside the density in all Group I reconstructions , indicating that this final secondary structural element of the central domain is not present in these particles . Notably , the Group I reconstructions also clearly lack density for both S5 and S12 , consistent with their observed depletion by qMS . The relatively high levels of S5 in comparison to S12 suggest that S5 may be present in Group II particles , while S12 is likely missing from both intermediates . Although the four Group I reconstructions are overall very similar in structure , some variability was apparent in the platform region . To identify regions of heterogeneity due to compositional or conformational variability , codimensional principal component analysis ( PCA ) was performed for the 12 , 425 Group I particles ( Penczek et al . , 2011 ) ( Figures 6C , D ) . In agreement with variations observed by 3D classification of the particles , this codimensional PCA revealed that variability between these structures mainly arises from differences in the platform region ( Figure 6D , Figure 6—figure supplement 2C ) . Given the limited resolution of these reconstructions , it is difficult to discern whether these differences are due to compositional or conformational variability in Group I particles . However , the levels of S11 measured by qMS are slightly depleted in comparison to other central domain proteins , in agreement with the variability in S11 density observed in the Group I reconstructions . Together , these analyses of Group I cryo-EM particles suggest that structural stability within the platform region may be dependent on assembly and docking of the 3′-domain .
Over the past 50 years , bacterial ribosome assembly has been studied extensively in vitro using a variety of biochemical and biophysical techniques . These previous studies provided insight into the order , hierarchies and kinetics of r-protein binding and rRNA folding , the fundamental underpinnings of ribosome biogenesis . In contrast , the understanding of in vivo ribosome assembly is relatively modest , owing in part to a lack of tools for the efficient study of this process at a molecular level . Recent developments in biophysical techniques have facilitated more detailed studies into the molecular mechanisms of cellular ribosome biogenesis , and especially the roles of various biogenesis factors . We have developed a high-throughput hybrid qMS/EM approach to study the composition and structure of cellular ribosome assembly intermediates . Our approach allows for the direct comparison of data sets from multiple samples , enabling quantitation of assembly intermediate distribution upon perturbation of the biogenesis pathway . Both qMS and single-particle EM are ideal methods for the analysis of heterogeneous samples , and both methods were applied to understanding the assembly intermediates present in samples taken directly from a sucrose gradient of crude E . coli lysate . Theoretically , all soluble proteins and complexes present in the cell could be observed across the gradient , including all stable ribosomal assembly intermediates . However , other dense cellular complexes that co-elute with 30S assembly intermediates and mature subunits increase sample heterogeneity . Indeed , several abundant proteins and complexes were readily observed in gradient fractions by proteomic analysis and by negative stain EM ( Supplementary file 1 , Figure 1—figure supplement 1 ) . At the center of the 30S peak , the majority of the particles ( 65–70% ) were ribosomal; however , fractions on the leading and lagging edges of the 30S peak were far more heterogeneous , containing only 30–35% ribosomal particles . The amount of data required to overcome this sample heterogeneity made high-throughput data collection indispensable for the detection of a wide range of 30S assembly intermediates . The combination of EM and qMS allowed for the classification of these intermediates along the 30S assembly pathway . Moreover , the application of stable isotope pulse-labeling and qMS facilitated the determination of assembly intermediates as on-pathway . Our analysis of sucrose gradient fractions from the E . coli 30S peak revealed five distinct groups of assembly intermediates distinguished by their conformation and r-protein content ( Figure 1C , D ) . Group I particles comprised the earliest observed assembly intermediates with the body and platform domains intact but no head domain density ( Figure 1D , Figure 2A ) . Group II and III particles all contained head domain density , with the head unanchored from the platform domain in Group II particles and slightly askew in Group III particles , when compared to mature 30S subunits ( Figure 1D , Figure 2B , C , Figure 4D ) . Particles in Group IV and V were the most mature 30S intermediates , containing almost all r-proteins except for those last to be incorporated , namely S2 , S3 and S21 ( Figure 1D , Figure 2D–H ) . Previous studies have shown that the r-protein binding and rRNA folding can proceed through multiple parallel pathways ( Talkington et al . , 2005; Mulder et al . , 2010 ) . The difference in the location of head density between Group II and Group III particles suggests that these conformations may result from two such parallel assembly pathways . In particular , the central PK , the long-range tertiary interaction within 16S rRNA that anchors the head domain to the body and platform , appears to be formed in Group III but unformed in Group II particles . Central PK formation occurs late in the assembly pathway ( Powers et al . , 1993; Besancon and Wagner , 1999; Holmes and Culver , 2004 ) , and could potentially occur after the modular assembly of the 5′-body , central , and 3′-head domains ( Sykes and Williamson , 2009 ) . This appears to be the case in Group II particles , in which nearly complete density is observed for all three domains in RCT reconstructions ( compare Figures 5D and 5G ) . In contrast , Group III particles have partially formed heads that are anchored to the body and platform domains , suggesting that assembly of the head domain proceeds following central PK formation in these intermediates . The presence of both types of intermediates in WT E . coli suggests that both pathways are possible in vivo . A number of assembly factors , including RimM and RbfA , have previously been implicated in central PK formation ( Clatterbuck Soper et al . , 2013 ) . In order to further examine the roles of various factors in this step , we used an anti-PK oligonucleotide hybridation/RNase H assay to test the degree of rRNA exposure in the central PK region . Surprisingly , we found that deletion of RimP has a much stronger effect on central PK exposure than either RimM or RbfA , suggesting that RimP may have a more direct role in central PK formation ( Figure 3B , Figure 3—figure supplement 1B , C ) . Indeed , EM analysis of intermediates purified from assembly factor deletion strains revealed that Group II particles are most abundant in ΔrimP ( Figure 3C , Figure 3—figure supplement 2 ) . In contrast , ΔrimM predominantly contains Group III intermediates in which the head domain is anchored and only partially formed ( Figure 3—figure supplement 2 ) , consistent with head-formation defects observed in cryo-EM structures and by in vivo 16S hydroxyl radical footprinting ( Clatterbuck Soper et al . , 2013; Guo et al . , 2013; Leong et al . , 2013 ) . RbfA has previously been implicated in the re-structuring of the 5′-leader sequence of 16S rRNA , which must be refolded in order for the central PK to form ( Dammel and Noller , 1995 ) . Deletion of RbfA leads to defects in folding of the 3′-head domain and the 5′-body domain , including the final placement of h44 ( Clatterbuck Soper et al . , 2013 ) . Our EM analysis of ΔrbfA shows an accumulation of Group II , III and late Group V intermediates ( Figure 3—figure supplement 2 ) , consistent with the role of RbfA in several stages of assembly . Similarly , ΔrsgA and ΔksgA primarily lead to accumulation of late Group V intermediates ( Figure 3—figure supplement 2 ) , consistent with previous observations that these factors act in h44 placement during the very late stages of 30S assembly ( Jomaa et al . , 2011a; Boehringer et al . , 2012 ) . An in-depth qMS analysis was performed on intermediate 30S particles extracted from the ΔrimP strain and directly compared to those from the WT strain to determine the composition of ΔrimP intermediates relative to those from an unperturbed assembly pathway . The ΔrimP strain was significantly depleted in S2 , S3 , S12 and S21 when compared to the WT strain . EM analysis of the ΔrimP sucrose gradient fractions revealed that these abundant intermediates are mainly Group I and II particles ( Figure 5B , C ) . In particular , Group II is enriched by >3-fold in ΔrimP samples when directly compared to WT ( Figure 5C ) . This result suggests that Group II intermediates may either be more long-lived in ΔrimP cells than in WT , or that RimP normally prevents Group II intermediates from forming in WT cells . The ΔrimP intermediates are on-pathway and are eventually incorporated into 70S ribosomes , based on pulse-labeling analysis ( Figure 4E , F ) . However , incorporation of depleted proteins occurs at a relatively slow rate , suggesting that the completion of these intermediates is kinetically unfavorable . RimP may act as a chaperone to prevent the assembling 30S subunit from falling into this kinetic trap . RimP has previously been shown to bind to free 30S subunits resolved on sucrose gradients , but not complete 70S ribosomes , indicating that it directly interacts with 30S assembly intermediates ( Nord et al . , 2009 ) . To directly measure the protein content of pre-central PK intermediates from the ΔrimP strain , we devised a strategy for affinity purification using a biotinylated anti-PK oligonucleotide . This early intermediate sample contained relatively low levels of S5 and S3 and significantly depleted levels of S2 and S12 ( Figure 6A ) . In addition , the late binding tertiary protein S21 could not be accurately quantitated , likely due to its extremely low levels in the purified intermediates . Cryo-EM structures of the Group I intermediate lack density for all five of these proteins , in addition to all 16S rRNA beginning with h27 ( Figure 6B , Figure 6—figure supplement 2C ) . It is likely that the severely depleted S2 , S12 and S21 are missing from Group II intermediates as well , given that Group II particles are the predominant species in the sample . The specific depletion of S5 and S12 in early ΔrimP intermediates is notable , as both proteins contact the central PK region in mature 30S subunits ( Figure 3B ) . Interestingly , RimP has previously been shown to accelerate binding of S5 and S12 by twofold and sixfold , respectively , in in vitro reconstitution assays ( Bunner et al . , 2010b ) . Together with our findings , these previous results suggest that addition of RimP to 30S reconstitution experiments may help to promote central PK formation and avoid the kinetically unfavorable Group I and Group II intermediates . In previous EM studies of assembly intermediates present during in vitro 30S reconstitution , Group I-like class averages were highly abundant at early time-points during assembly ( Mulder et al . , 2010 ) . Intriguingly , Group II-like classes are also present during the early stages of assembly , although they were uncharacterized in that study ( See Figure 1B in Mulder et al . , 2010 ) . The presence of Group I and II-like classes subsides with the incorporation of S5 and S12 , as measured by pulse-chase followed by qMS . These previous in vitro results agree with the present in vivo observations that kinetically unfavorable pre-central PK conformations remain viable on-pathway intermediates . The central PK is essential for translation ( Brink et al . , 1993; Poot et al . , 1998 ) and is conserved in all kingdoms of life . The accurate formation and stability of the central PK is a critical step during small subunit assembly in both prokaryotes and eukaryotes . Recently , the essential ribosomal biogenesis factor Mrd1 was implicated in central PK formation in Saccharomyces cerevisiae ( Segerstolpe et al . , 2013 ) . Mrd1 contains multiple RNA-binding domains ( RBDs ) and binds directly to 18S rRNA helices h27 and h28 , two secondary structural elements that reside in close proximity to the central PK in the mature small subunit structure ( Segerstolpe et al . , 2013 ) . Similarly , RimP is composed of two RBDs , and may play an analogous role in binding to 16S rRNA regions adjacent to the central PK . We propose that RimP acts during the early and late stages of 30S subunit biogenesis to assist in the stabilization of the central pseudoknot , allowing for the subsequent incorporation of central-PK binding r-proteins S5 and S12 and late binding r-proteins S2 , S3 and S21 ( Figure 7 ) . During the early stages of 30S biogenesis , premature central PK formation is blocked by a structure within the 16S 5′-leader that is mutually exclusive with h1 . RbfA is thought to bind to the leader sequence and promote formation of h1 during the late stages of assembly , and may act synergistically with RimP to stabilize the central PK . In contrast , RimM may act independently of RimP to facilitate assembly of the head domain regardless of the status of central PK formation . Overall , our findings suggest that RimP might be one of the earliest factors to act upon the assembling 30S subunit . The combined EM/qMS approach employed here should have immediate and broad applicability to study of the role of other ribosome assembly factors as well as macromolecular assembly processes involving other bacterial and eukaryotic cellular machines . 10 . 7554/eLife . 04491 . 018Figure 7 . A model for 3′-domain formation during in vivo 30S biogenesis . Co-transcriptional folding and binding of 5′-body ( red ) and central domain ( green ) r-proteins results in the formation and accumulation of Group I intermediates . The 3′-head domain ( blue ) can fold and r-proteins , including both primary and secondary binders , can bind prior to or following formation of the central PK , resulting in the accumulation of Group II and Group III intermediates , respectively . In the absence of RimP , the central PK is destabilized and the flux of 30S intermediates flows mainly through the Group II pathway , in which the 3′ domain is nearly fully formed prior to formation of the central PK . These intermediates are on pathway and eventually all remaining r-proteins , including S5 and S12 , are incorporated into the mature 30S subunit . DOI: http://dx . doi . org/10 . 7554/eLife . 04491 . 018
All E . coli strains used in this study were in the BW25113 background . BW25113 was used as WT , while assembly factor knockout strains were part of the Keio collection ( Baba et al . , 2006 ) . E . coli strains were obtained from either the Yale E . coli Genetic Stock Center or Thermo Scientific ( Waltham , MA ) , and genotypes were confirmed by PCR using primers flanking the gene of interest by ∼100 bp on either side . The plasmid pU23 encodes a copy of the rrnB rRNA operon containing C23U and C1192U ( confers spectinomycin resistance , used for selection purposes in the original study ) mutations within the 16S rRNA gene ( Dammel and Noller , 1993 ) . The plasmid was a generous gift from the Noller lab . E . coli cultures were grown aerobically in M9 media ( glucose , trace vitamins and minerals ) supplemented with either 14N- , 50% 15N- or 15N-labeled ammonium sulfate as the only source of nitrogen . Cultures were grown to mid-log phase ( 0 . 4-0 . 5 OD600 ) at 37°C , then quickly cooled by direct addition to ice and harvested by centrifugation ( 20 min at 6000×g , 4°C ) . For pulse labeling experiments , cultures were grown to mid-log phase in 14N M9 media , then pulsed with an equivalent volume of 15N M9 media for 15 , 20 , 30 or 45 min . Cells were lysed in Buffer A ( 20 mM Tris , pH 7 . 5 , 100 mM NH4Cl , 10 mM MgCl2 , 0 . 5 mM EDTA , 6 mM β-Me ) by bead beating ( 0 . 1 mm Zirconia/Silica beads , BioSpec mini-bead beater ) , then cleared by two successive centrifugation steps at 22 , 000×g . The lysate was resolved on a 35 ml 12 . 9–51 . 5% sucrose gradient by centrifugation at 4°C in a Beckman SW32 rotor for 16 hr at 26 , 000 rpm . For samples prepared for qMS analysis comparing the ΔrimP and WT strains , equimolar amounts of the respective lysates were combined prior to sucrose gradient centrifugation . The gradients were eluted using a Brandel Gradient Fractionator and prepared for qMS and EM analysis . Fractions containing 70S particles in the WT strain grown in 15N M9 media were pooled and stored at −80°C for use as a reference sample . Samples used for fraction-by-fraction EM analysis were concentrated to 100 nM if necessary ( early and late fractions from 30S peak ) , and frozen at −80°C . For the assembly factor deletion EM studies , RNase H assays and affinity purification , fractions from the 30S peak were pooled ( Figure 3—figure supplement 1A ) , then dialyzed against Buffer A for three hours at 4°C . The samples were concentrated to 250 nM , flash frozen and stored at −80°C . For WT fraction-by-fraction analysis , particles in each fraction under the 30S peak were prepared for LC-MS investigation as previously described ( Chen and Williamson , 2013 ) . In brief , 20 pmol of each fraction was combined with 20 pmol of 70S spike and trichloroacetic acid ( TCA ) precipitated ( 13% vol/vol final concentration ) at 4°C overnight . Precipitates were isolated by centrifugation ( 30 min at 14 , 000 rpm at 4°C ) , washed with 10% TCA , then cold acetone and left to air dry . TCA precipitates were resuspended in 20 μl 100 mM NH4HCO3 , 5% acetonitrile ( ACN ) , 2 μl 50 mM dithiothreitol was added and the mixture was incubated in a 65°C water bath for 10 min . The samples were then treated with 2 μl 100 mM iodoacetamide and incubated for 30 min at 30°C , then digested overnight with 2 μl 0 . 1 μg trypsin at 37°C . The trypsinized peptides were purified over a Pierce C18 column , eluting across a 5–50% ACN gradient over 105 min . Peptides were then detected on a coupled Agilent G1969A ESI-TOF mass spectrometer over a set detection range of 250–1300 m/z . For samples from WT cells only , peptides were detected by an Agilent Q-TOF G6520B and initially processed with Agilent Qualitative Analysis software . Peak lists from the raw LC-MS data were generated using the Aglient Mass Hunter and Mass Profiler programs , and the 14N/15N peak pairs were identified and quantified as described previously ( Sperling et al . , 2008; Sykes et al . , 2010 ) . The relative abundance of each r-protein was then calculated by comparing the amplitude of its 14N peptides to that of the sum of its 14N and 15N peptides [14N/ ( 14N + 15N ) ] . The isotope distribution of each peptide was examined and fits with low signal-to-noise ratios were excluded from further analysis . The relative abundance of each r-protein was normalized to that of the primary binder S4 to flatten any differences in total r-protein amount of each sample . For fraction-by-fraction analysis of the ΔrimP strain as compared to the WT strain , qMS analysis was carried out as described above with one exception . In this case , the relative abundance of each r-protein in the ΔrimP strain was calculated by comparing the amplitude of the 50% 15N peptides to that of the sum of the 50% 15N and 15N peptides [50% 15N/ ( 50% 15N + 15N ) ] . The normalized relative abundance of each r-protein across the 30S peak were hierarchically clustered using Euclidean distance scoring and average linkage in Gene Cluster 3 . 0 . The resulting cluster trees were visualized using Java TreeView . For proteomic data ( Supplementary file 1 ) , equimolar amounts of 14N peptides were prepared as described above for each sucrose gradient fraction from WT and ΔrimP strains . Samples were submitted to an Agilent G6520B QTOF mass spectrometer for LC-MS/MS analysis as previously described ( Chen and Williamson , 2013 ) . Briefly , peptides were separated by a 90 min 5–60% concave acetonitrile gradient and detected over a precursor detection range of 400 to 2000 m/z and a product ion detection range of 80–2000 m/z . Data were analyzed using Mascot ( precursor mass error tolerance = 0 . 05 Da , product mass error tolerance = 0 . 10 Da ) , and identified peptides were subject to a significance threshold of 0 . 05 and ion score cutoff of 0 . 05 . The data provided in Supplementary file 1 represent the highest-scoring match for each peptide . For pulse-labeling experiments , WT or ΔrimP samples were grown in 14N-labeled media to mid-log phase , then pulsed with an equivalent volume of 15N-labeled media for 15 , 20 , 30 or 45 min . At each time-point , 100 ml of culture was rapidly removed and quenched and the cell pellet was stored at −80°C . Time-point samples were then purified by sucrose gradient centrifugation to isolate 30S and 70S particles . Each sample containing either 30S or 70S particles was combined with an equimolar amount of 15N-labeled 70S particles ( reference for accurate peptide identification ) , and prepared for qMS analysis as described above . For each peptide , the observed raw LC-MS data comprised three isotope distribution envelopes ( Figure 4—figure supplement 1B ) . The leftmost ( low m/z ) envelope corresponds to r-proteins synthesized prior to the pulse ( 100% 14N ) while the middle envelope corresponds to r-proteins synthesized post-pulse ( 50% 15N ) . The rightmost ( high m/z ) envelope corresponds to r-proteins from the reference 70S particles ( 100% 15N ) . The fraction labeled ( fL ) value of each r-protein was calculated by comparing the abundance of r-proteins synthesized post-pulse to that of the sum of r-proteins pre- and post-pulse [50% 15N/ ( 100% 14N + 50% 15N ) ] . For each r-protein , the time course of 15N-labeling was fit to Equation ( 1 ) below using Igor Pro ( WaveMetrics Inc . ) as previously reported ( Chen et al . , 2012 ) . ( 1 ) fL ( t ) =1+P·exp[−k· ( 1+1/P ) ·t]− ( 1+P ) ·exp[−k·t]where fL is the fraction labeled value , P is the precursor pool size , t is the length of 15N pulse and k is the growth rate , with P set as the only free parameter . The curve representing the maximum expected labeling was calculated using Equation ( 2 ) , ( 2 ) fmax ( t ) =1−exp[−k·t] For untilted negative stain EM , samples were applied to plasma-cleaned ( 20 s , Gatan Solarus ) carbon-coated copper mesh grids ( Ted Pella , Inc . ) . For RCT negative stain EM and cryo-EM , samples were applied to plasma-cleaned ( 5 s ) C-flat grids ( Protochips ) coated with a thin ( 2–5 nm ) layer of continuous carbon . Sucrose gradient fraction samples were diluted with Buffer A to a concentration yielding optimal particle distribution and homogeneity on the grid surface , generally to a concentration of ∼10 nM ( based on absorbance reading at 260 nm of ∼0 . 13 and 30S extinction coefficient of 12 . 8 × 106 M-1cm-1 ) . The affinity purified sample was diluted with Buffer A 1:5 for negative stain analysis and 1:3 for cryo-EM analysis . Negative stain grids were prepared by applying the sample ( 3 µl ) to the grid for 1 min , then blotting from the side to remove excess sample . The grid was washed immediately with 3 µl Buffer A , then blotted from the side . Concurrent with blotting , 3 µl of fresh 2% uranyl formate was applied to the grid , then blotted from the side . This step was repeated twice , then the grid was allowed to dry for at least 10 min . For cryo-EM grid preparation , 3 µl of sample was applied for 1 min , blotted for 3 s , then plunge-frozen in liquid ethane using a Gatan CP3 . All EM images were collected using Leginon ( Suloway et al . , 2005 ) . Data for WT and ΔrimP fractional analysis were acquired using an FEI T12 transmission electron microscope operating at 120 keV and equipped with a Tietz TemCam-F416 4k × 4k CMOS camera . Images were collected at a nominal magnification of 52000× and pixel size of 2 . 05 Å with a dose of ∼30 e-/Å2 and a nominal focus range from 0 . 8–1 . 8 µm under focus . Image tilt pairs ( −50°/0° ) for RCT data were collected at a dose of ∼20 e-/Å2 ( Yoshioka et al . , 2007 ) . Data for 30S peak samples from the WT and knockout strains were acquired using a Tecnai F20 Twin transmission electron microscope operating at 200 keV equipped with a Tietz TemCam-F416 4k × 4k CMOS camera . Images were collected at a nominal magnification of 62 , 000× and a pixel size of 1 . 36 Å with a dose of ∼30 e-/Å2 and a nominal focus range from 0 . 8–1 . 8 µm under focus . Cryo-EM data were acquired using a Tecnai F20 Twin transmission electron microscope operating at 200 keV equipped with a Gatan K2 Summit direct detection device . Cryo-EM images were collected at a nominal magnification of 29000× and pixel size of 1 . 21 Å with a nominal focus range from 2 . 5–5 . 0 µm under focus . Images frame sets ( 1253 ) were collected for 6 s with a dose of 33 . 67 e-/Å2 for 30 frames ( 200 ms each ) , followed by whole frame alignment as previously described ( Li et al . , 2013 ) . For negative stain EM data , all image processing was carried out in Appion ( Lander et al . , 2009 ) . The CTF for all images was estimated using CTFFind3 ( Mindell and Grigorieff , 2003 ) . For all datasets , particle picking was carried out using DoG picker ( Voss et al . , 2009 ) . Parameters were adjusted to ensure that all particles were selected from each image , in order to eliminate particle selection bias based on size in the initial stack . Following particle extraction ( with box sizes ranging from 350–380 Å ) , the initial stack was subjected to a first round of 2D reference-free alignment and classification using Xmipp ML2D to obtain classes with <2000 particles/class ( Scheres et al . , 2005a , 2005b ) . This initial alignment allowed for identification and removal of classes lacking any identifiable features ( generally false positive particle picks ) or clearly identifiable as a non-ribosomal E . coli complex . Identification of non-ribosomal complexes was validated by comparison with known structures of the complexes , and by their presence in proteomic analysis of sucrose gradient fractions ( Figure 1—figure supplement 1C , Supplementary file 1 ) . The cleaned stack was then subjected to reference-free alignment and clustering using Xmipp CL2D to obtain classes with <200 particles/class ( Sorzano et al . , 2010 ) . This finer classification revealed additional false positive and non-ribosomal classes , which were removed . A final round of cleaning was implemented following a second ML2D classification ( <500 particles/class ) . The alternating use of CL2D and ML2D strategies revealed additional class averages containing false positive particles or non-ribosomal classes , although these classes were generally low in abundance and population . The final stack was subjected to ML2D classification with the resultant classes shown in Figure 1C , Figure 4B and Figure 3—figure supplement 2B . The classes were aligned to a reference , and the aligned classes were imported into Mathematica ( Wolfram Research ) for hierarchical clustering analysis . Dendrograms were constructed using agglomerative hierarchical clustering of the class images , using a correlation distance metric and average linkage clustering . For direct comparison of particle distribution between strains , substacks of 10 , 000 random particles were created for each data set and combined into a single stack . These combined stacks were then subjected to ML2D alignment and classification . The resultant classes were clustered into Groups in Mathematica as described above . For RCT data sets , particle tilt pairs were identified using TiltPicker ( Voss et al . , 2009 ) . Untilted and tilted particles were extracted ( box size 224 , pixel size 2 . 05 Å ) into two separate stacks . For the untilted stack , bad particles were identified and removed using an initial ML2D alignment into 100 classes followed by a CL2D alignment into 256 classes . The cleaned untilted stacks were subjected to ML2D alignment into 15 classes , which were subsequently clustered using Mathematica . Substacks were created for every Group based on the clustering , and each substack was aligned using ML2D . RCT volumes were reconstructed from the resultant classes using the Create RCT Volume function in Appion ( Voss et al . , 2010 ) . For cryo-EM image analysis , CTF estimation and particle picking were carried out as described above . The micrographs were contrast-inverted , then particles were extracted with a box size of 288 pixels at 1 . 21 Å/pixel . This box size was optimized for Group I particles , but we also performed a parallel analysis with a larger box size of 320 pixels to examine the larger Group II particles . The initial stack was binned by four and subjected to ML2D alignment and classification to obtain 100 classes . False positive peak picks were discarded , and the cleaned stack was aligned and clustered into 128 classes using CL2D for initial evaluation of the conformations and views present in the sample . Given the heterogeneity of the sample , it was difficult to distinguish between various conformations and views based solely on visual inspection . We therefore employed a sorting algorithm that compared a set of models to our experimental class averages using Xmipp projection-matching refinement ( Sorzano et al . , 2004 ) . Each class average was assigned to one of the models based on the highest correlation value following projection-matching refinement , as previously described in ( Lyumkis et al . , 2013b ) . The following five models used for projection matching were created from PDB 2AVY ( Schuwirth et al . , 2005 ) and low pass filtered to 30 Å: the 3′-domain comprising S3 , S7 , S9 , S10 , S13 , S14 , S19 , and 16S nt 921-1534; a Group I model comprising S4 , S6 , S8 , S11 , S15 , S16 , S17 , S18 , S20 , S21 , and 16S nt 26-909; a Group II model in which the 3′-domain model ( with 16S nt 1398-1534 removed ) and the Group I model were fit into a representative RCT volume; a late intermediate missing only S2 , S3 and S21; and the fully mature 30S subunit . Notably , no class averages generated by CL2D of the cleaned stack were matched with the mature model . Group I classes were identified from the initial ML2D classes using this projection-matching sorting algorithm aided by visual inspection . These particles were subjected to two further rounds of CL2D to remove bad particles , resulting in a final cleaned stack of 12 , 425 Group I particles . An initial model was generated from the final CL2D classes using the OptiMod common lines/refinement package in Appion ( Lyumkis et al . , 2013c ) . An initial set of angles was assigned to the Group I particle stack using Xmipp projection-matching refinement . These angles were further refined and particles were classified through 200 rounds into four models using Frealign 9 ( Lyumkis et al . , 2013a ) . The final distribution of particles and Fourier shell correlations for the 4 models were as follows: Model 1 – 3407 particles , 27 . 6 Å; Model 2 – 2975 particles , 26 . 1 Å; Model 3 – 2722 particles , 27 . 0 Å; Model 4 – 3241 particles , 25 . 4 Å . Variance analysis for the Group I cryo-EM particles was performed using the codimensional PCA application in SPARX ( Penczek et al . , 2011 ) . All structure figures were created in UCSF Chimera ( Pettersen et al . , 2004 ) . DNA oligonucleotides used for these experiments were designed to anneal to 16S rRNA positions 906–920 ( anti-PK 5′-ATTCATTTGAGTTTT-3′ ) to test for central PK accessibility or 589–603 ( anti-h21 5′- ATCTGACTTAACAAA-3′ ) as a negative control targeting a highly stable region of the 16S rRNA . RNase H assays were performed in Buffer A , with 10 mM DTT substituted for the 6 mM β-Me . Cleavage reactions were initiated on ice by adding 0 . 5 pmol 30S subunits ( final concentration 33 nM ) to 50 or 500 pmol ( final concentration 3 . 3 or 33 µM ) anti-PK or anti-h21 oligo ( or buffer A for mock reactions ) and 5U RNase H ( New England Biolabs ) ( or buffer A for mock reactions ) . Samples were incubated on ice at 4°C for 16 hr , then resolved on a 2% agarose/TAE gel and visualized by ethidium bromide staining . Intact 16S rRNA and cleavage products were quantified using ImageQuant software , with the two cleavage bands treated as a single product . The intensity of ‘cleavage products’ detected in the mock reaction lane was subtracted from the cleavage band for each reaction containing oligo , to account for background cleavage that may have occured before or during the RNase H reaction . Fraction cleaved was calculated by dividing the volume of the cleavage products by the total RNA in the lane ( cleavage products plus uncleaved rRNA ) . The average of three replicates was plotted with error bars representing the standard deviation between the three replicates . Affinity purification protocol was adapted from ( Schnapp et al . , 1998; Clatterbuck Soper et al . , 2013 ) . Oligonucleotides were designed based on the anti-PK oligo used for the RNase H assay . The capture oligo comprised a 5′-biotin followed by 10 random DNA nucleotides and finally the 2′-O-methylated anti-PK sequence ( 5′-biotin-CTACAGGTGCAAmAmUmUmCmAmUmUmUmGmAmGmUmUmU-3′ ) , in order to promoted annealing of the anti-PK sequence to the 16S rRNA . The displacement DNA oligonucleotide was completely complementary to the capture oligo ( 5′-AAACTCAAATGAATTTGCACCTGTAG-3′ ) . Samples used for affinity purification contained 200 pmol of F2-3 from the ΔrimP sucrose gradient ( 30S peak ) , 20 OD260 of ΔrimP lysate , or 200 pmol purified 30S subunits in 350 µl Buffer A . Each sample was incubated with 17 . 5 µl 1 mg/ml yeast tRNA , 2 µl 100 µM capture anti-PK oligo ( 200 pmol ) , and 5 µl RNasin at 30°C for 15 min . NuetrAvadin agarose beads ( 100 µl per sample ) ( Thermo Scientific ) were blocked with 0 . 5 mg/ml BSA in Buffer A twice , then washed with Buffer A , and finally incubated at 30°C for 10 min . Samples were added to beads , then incubated at 30°C for 10 min . Samples were then transferred to 4°C , and incubated with rocking for 2 hr . Beads were centrifuged for 5 min at 500×g , and the supernatant was removed . Beads were washed four times at 4°C with Buffer A + 0 . 01% Nikkol , then twice more at room temperature with 5 min of incubation with rocking for each wash . Samples were eluted by adding 5 pmol of displacement oligo to 150 µl of Buffer A + 0 . 01% Nikkol . The buffer and beads were incubated with gentle rocking at room temperature for 30 min , then centrifuged for 5 min at 500×g , and the eluent was removed . This elution was repeated up to three times , but very little 16S rRNA was observed in later elution fractions . Samples were visualized on a 2% agarose/TAE gel stained with ethidium bromide . The first elution fraction was aliquoted and flash frozen in liquid nitrogen , then stored at −80°C prior to EM analysis . Samples for qMS were generated in the same manner , except 275 pmol 30S ribosomes , 500 pmol capture oligo and 250 µl NeutrAvidin Agarose beads were used for affinity purification , and 2 . 5 µM of displacement oligo was used for elution . Electron microscopy maps for the 30S ribosomal intermediates have been deposited to the 3D-Electron Micrscopy Data Bank ( EMDB http://www . ebi . ac . uk/pdbe/emdb/ ) EMDB ID code 6125-6145 . | The proteins in cells are made by complex organelles called ribosomes . These organelles are made of two subunits: the small ribosomal subunit , which reads the messenger RNA that contains the genetic code for the protein , and the large ribosomal subunit , which links amino acids together to form a protein . But how are the ribosomes themselves—which contain several ribosomal RNA molecules and dozens of ribosomal proteins—put together ? Various aspects of the assembly of ribosomes have been studied in the test tube , but the complexity of the assembly process means there is little data from experiments performed on living cells . Now Sashital et al . have used a combination of two techniques—mass spectrometry and electron microscopy—to study the assembly of ribosomes in living Escherichia coli cells . Mass spectrometry measures the relative amounts of the different ribosomal proteins in each sample , while electron microscopy provides information on the shape of the ribosome , including the shape of some of the intermediate structures formed during the assembly process . Sashital et al . analyzed the composition and structure of the small ribosomal subunits in wild type E . coli , and also in mutant E . coli cells in which the genes for various proteins thought to be involved in the assembly process had been deleted . These experiments revealed that a protein called RimP had a key role in stabilizing an important central structure called a pseudoknot . The approach developed by Sashital et al . should be able to reveal other details about the assembly of ribosomes , and also about other macromolecular complexes that are found inside the cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | A combined quantitative mass spectrometry and electron microscopy analysis of ribosomal 30S subunit assembly in E. coli |
Many ‘non-enveloped’ viruses , including hepatitis A virus ( HAV ) , are released non-lytically from infected cells as infectious , quasi-enveloped virions cloaked in host membranes . Quasi-enveloped HAV ( eHAV ) mediates stealthy cell-to-cell spread within the liver , whereas stable naked virions shed in feces are optimized for environmental transmission . eHAV lacks virus-encoded surface proteins , and how it enters cells is unknown . We show both virion types enter by clathrin- and dynamin-dependent endocytosis , facilitated by integrin β1 , and traffic through early and late endosomes . Uncoating of naked virions occurs in late endosomes , whereas eHAV undergoes ALIX-dependent trafficking to lysosomes where the quasi-envelope is enzymatically degraded and uncoating ensues coincident with breaching of endolysosomal membranes . Neither virion requires PLA2G16 , a phospholipase essential for entry of other picornaviruses . Thus naked and quasi-enveloped virions enter via similar endocytic pathways , but uncoat in different compartments and release their genomes to the cytosol in a manner mechanistically distinct from other Picornaviridae .
The presence or absence of an external lipid envelope has featured strongly in the systematic classification of animal viruses for decades . However , many viruses that have previously been considered to be ‘non-enveloped’ are now known to be released non-lytically from infected cells in a ‘quasi-enveloped’ form , enclosed in small extracellular vesicles ( EVs ) devoid of virus-encoded surface proteins . This phenomenon was recognized first among members of the Picornaviridae , including hepatitis A virus ( HAV , genus Hepatovirus ) ( Feng et al . , 2013 ) , poliovirus and coxsackievirus B ( genus Enterovirus ) ( Bird et al . , 2014; Chen et al . , 2015; Jackson et al . , 2005; Robinson et al . , 2014 ) , but it has been demonstrated also for hepatitis E virus ( Hepeviridae ) , rotaviruses ( Reoviridae ) , and noroviruses ( Caliciviridae ) ( Nagashima et al . , 2017; Santiana et al . , 2018 ) . The size of the virus-containing EVs varies widely among different viruses , as does the number of virus capsids enclosed in each vesicle , most likely reflecting different mechanisms of biogenesis . However , these membrane-wrapped , quasi-enveloped virions share the capacity to infect cells , and contribute to pathogenesis either by cloaking capsids in membranes such that they are sequestered from the host immune system , or possibly by increasing the number of viral genomes delivered to newly infected cells , thereby facilitating genetic complementation ( Chen et al . , 2015; Feng et al . , 2013 ) . HAV provides a prime example of viral quasi-envelopment . An ancient pathogen that remains a common cause of enterically-transmitted hepatitis globally ( Lemon et al . , 2017 ) , it is hepatotropic in vivo and released without cell lysis in small EVs containing 1–3 capsids ( Feng et al . , 2013 ) . In contrast to the naked , nonenveloped virus that is shed in feces , these quasi-enveloped virions ( eHAV ) are the only form of virus found in sera from infected humans and account for most viruses in supernatant fluids of permissive cell cultures ( Feng et al . , 2013 ) . They are fully infectious , ~50–110 nm in diameter , and possess a buoyant density of ~1 . 100 g/cm3 in iodixanol . The protein composition of the quasi-envelope resembles that of exosomes , suggesting a multivesicular body ( MVB ) origin , and their biogenesis is dependent on ALIX ( ALG-2-interacting protein 1 , also known as PDCD6IP ) and other components of the endosomal sorting complexes required for transport ( ESCRT ) ( Feng et al . , 2013; González-López et al . , 2018; McKnight et al . , 2017 ) . The capsids enclosed within the eHAV vesicle are , like other picornaviral capsids , comprised of 60 copies of each of 4 proteins ( Wang et al . , 2015 ) . However , they differ from the naked , nonenveloped capsids shed in feces in that they contain an unprocessed form of the VP1 capsid protein ( VP1pX ) retaining a 71 amino acid carboxy-terminal domain absent in naked virions ( Feng et al . , 2013 ) . Interactions between picornaviral capsids and their receptors are critical for promoting endocytosis , virion uncoating , and safe delivery of the viral RNA genome across endosomal membranes into the cytoplasm to establish a productive infection ( Baggen et al . , 2018; Groppelli et al . , 2017; Strauss et al . , 2015 ) . The phosphotidylserine ( PtdSer ) receptor TIM1 ( T cell immunoglobulin and mucin-containing domain one protein , also known as HAVCR1 ) was identified as a receptor for HAV twenty years ago ( Feigelstock et al . , 1998; Kaplan et al . , 1996 ) , prior to the discovery of quasi-enveloped virions . TIM1 has since been shown to facilitate the binding of eHAV but not naked HAV virions to the cell surface ( Das et al . , 2017 ) , presumably through binding PtdSer displayed on the surface of the eHAV membrane ( Feng et al . , 2015 ) . However , TIM1 is not essential for attachment or entry of either HAV or eHAV , nor is it required for infection of permissive strains of mice ( Das et al . , 2017 ) . Thus far , an essential receptor has yet to be identified for either HAV or eHAV . Little is known about how these two virion types enter cells , although prior studies point to the existence of distinct entry pathways for naked versus quasi-enveloped virions as might be expected from the presence of the limiting lipid membrane in eHAV . eHAV is selectively sensitive to the lysosomal poison chloroquine ( Feng et al . , 2013 ) , and slower to enter cells and begin replication than naked HAV capsids ( Das et al . , 2017; Feng et al . , 2013 ) . eHAV is resistant to anti-capsid neutralizing antibodies in quantal , plaque reduction-like assays , but neutralizing antibodies restrict its replication when added to cells 4–6 hr after adsorption of the virus , suggesting a delay in uncoating within an endocytic compartment ( Feng et al . , 2013 ) . Here , we report detailed roadmaps for the entry of these different types of infectious hepatitis A virions in hepatocytes , identify a key role for integrin β1 in endocytosis of both , and demonstrate distinct trafficking of these virion types through the endocytic system . We demonstrate critical temporal and spatial differences in the uncoating of HAV and eHAV capsids , and show that eHAV entry is uniquely dependent on the ESCRT accessory protein ALIX as well as lysosomal proteins involved in lipid metabolism .
To identify the endocytic pathways responsible for internalization of HAV and eHAV virions in Huh-7 . 5 human hepatoma cells , we used pharmacological and genetic approaches to disrupt the function of regulators of several canonical endocytic routes . Inhibition of clathrin- and dynamin-mediated endocytosis by the drugs chlorpromazine and dynasore , respectively , strongly inhibited the uptake of both gradient-purified HAV and eHAV virions , as measured by RT-PCR quantitation of the viral genome in cell lysates 6 hr post-infection ( hpi ) ( Figure 1A , Figure 1—figure supplement 1 ) . In contrast , inhibiting cavaeolae-dependent endocytosis with filipin resulted in mild impairment of eHAV entry only , whereas inhibiting cytosolic dynein , actin- and/or Rac1-dependent micropinocytosis , or heparan sulfate proteoglycan binding with various compounds had no effect on the entry of either virion type ( Figure 1A ) . Roles for clathrin and dynamin in the uptake of both HAV and eHAV were confirmed by siRNA-mediated depletion of clathrin heavy chain ( CLTC ) , the µ1 subunit of the clathrin-associated adaptor complex 2 ( AP2M1 ) , or dynamin-2 ( DNM2 ) ( Figure 1B , Figure 1—figure supplement 2A ) . Caveolin-1 ( CAV1 ) depletion minimally inhibited entry of eHAV only , consistent with the effect of filipin treatment , while depleting the clathrin- and caveolae-independent endocytosis regulators ADP ribosylation factor 6 ( ARF6 ) and flotillin-1 ( FLOT1 ) had no significant effect . Consistent with these results , confocal fluorescence microscopy revealed a high-degree of co-localization of both HAV and eHAV capsid antigen with clathrin-coated vesicles between ~0 . 5–1 hpi , and minimal co-localization of the eHAV capsid only with caveolin-1 ( Figure 1C ) . Thus , both HAV and eHAV entry occur primarily through clathrin- and dynamin-dependent endocytosis , although caveolin-dependent endocytosis may play a minor role in eHAV uptake . The host proteins associated with the eHAV quasi-envelope are similar to those identified in exosomes ( McKnight et al . , 2017 ) . This suggests that eHAV entry might be mediated by integrins or adhesion molecules previously reported to be involved in the uptake of EVs ( van Dongen et al . , 2016 ) . Consistent with this hypothesis , siRNA-mediated depletion of integrin β1 ( ITGB1 ) resulted in a striking and highly significant reduction in the uptake of quasi-enveloped as well as naked virions ( Figure 1D , Figure 1—figure supplement 2B ) . Depleting integrin β1 did not reduce the quantity of eHAV or HAV bound to the cell surface at 4°C , but significantly reduced the amount of eHAV and HAV RNA present in Huh-7 . 5 cells 6 hpi at 37˚C ( Figure 1E ) . CRISPR/Cas9 knockout of ITGB1 also reduced both eHAV and HAV uptake and spread in H1-HeLa cells ( Figure 1F , G ) . Consistent with these results , pre-treating Huh-7 . 5 cells with an RGD peptide containing an integrin β1-binding motif reduced uptake of both virion types by about 50% ( Figure 1H ) . On the other hand , pre-treating cells with antibodies that activate integrin β1 by binding to and stabilizing specific β1 conformations ( Su et al . , 2016 ) increased viral uptake compared to an inert integrin β1 antibody ( K-20 ) , and revealed differences in the interaction of integrin β1 with eHAV versus HAV ( Figure 1H ) . The activating antibody TS2/16 , which binds an open conformation of β1 ( Su et al . , 2016 ) , enhanced eHAV but not HAV entry , whereas 8E3 and HUTS-4 , which bind extended and open headpiece β1 conformations , respectively , had the opposite effect , enhancing naked HAV but not quasi-enveloped eHAV entry . These data hint at differences in the ligands , yet to be identified , that are bound by integrin β1 during eHAV and HAV entry . In contrast to the impact of integrin β1 depletion , depletion experiments failed to confirm a requirement for any specific α integrin in the uptake of either virion ( Figure 1D , Figure 1—figure supplement 2B ) . While RNAi-mediated depletion of integrin α1 caused a modest but statistically significant decrease in HAV uptake in Huh-7 . 5 cells , this was not confirmed in H1-HeLa cells with CRISPR/Cas9 knockout of ITGA1 ( Figure 1D , Figure 1—figure supplement 3 ) . Confocal microscopic imaging also suggested eHAV was associated with integrin β1 , both at the surface of Huh-7 . 5 cells at 4°C and during virion internalization at 37˚C ( Figure 1I ) , but not with either α5 or αV integrins ( Figure 1E , Figure 1—figure supplement 4 ) . Collectively , these results demonstrate that HAV and eHAV are dependent on distinct integrin β1 interactions for uptake by clathrin- and dynamin-mediated endocytosis , but leave unanswered the role of α integrins . Several GTPases are well-known for their role in the sorting of cargo through functionally distinct endosomes , with Rab5A and Rab7a involved in trafficking through early and late endosomes , respectively ( Mellman , 1996; Mercer et al . , 2010 ) . Confocal microscopy of infected Huh-7 . 5 cells revealed transient co-localization of the capsid antigen in both naked and quasi-enveloped virions with Rab5A+ and Rab7a+ compartments around~1–2 hpi ( Figure 2A ) . In contrast , neither type of virion was associated with Rab11A+ recycling endosomes . RNAi-mediated depletion of Rab5A or Rab7a , but not Rab11A , resulted in a significant reduction in the accumulation of intracellular HAV RNA ( Figure 2B , Figure 2—figure supplement 1 ) . Thus , both types of HAV virions traffic through early and late endosomes shortly after uptake into the cell through clathrin-mediated endocytosis . Our earlier studies suggested that infection with eHAV , but not naked HAV , requires endosomal acidification since it was specifically inhibited by the lysosomal poison , chloroquine ( Feng et al . , 2013 ) . Consistent with this , confocal microscopy demonstrated that the capsid antigen associated with quasi-enveloped eHAV , but not naked HAV , was selectively trafficked to LAMP1+ and VAMP8+ lysosomes as early as 4 hpi , remaining there up to 12 hpi ( Figure 2C , Figure 2—figure supplement 2 ) . Notably , naked virion capsid antigen was never found to be associated with lysosomes , suggesting that Rab7a+ late endosomes represent the final trafficking destination of naked HAV . Importantly , sorting of both eHAV and naked HAV virions was associated with co-internalization of integrin β1 to these endolysosomal compartments ( Figure 2D , Figure 2—figure supplement 3 ) . Similar results were obtained when integrin β1 expressed on the cell surface was labeled prior to virus adsorption with the activating monoclonal antibody TS2/16 ( Figure 2D , Figure 2—figure supplement 4 ) , which triggers the endocytosis and trafficking of integrin β1 to lysosomes ( Margadant et al . , 2012 ) . Since components of the ESCRT machinery , particularly ALIX , are involved in endosomal sorting to the lysosome ( Dores et al . , 2016; Murrow et al . , 2015 ) , we asked whether eHAV trafficking to the lysosome is dependent on ESCRT . Strikingly , quasi-enveloped eHAV virions failed to reach the lysosome in cells depleted of ALIX , but not the ESCRT-III proteins CHMP1B or CHMP2A ( Figure 2E , Figure 2—figure supplement 5 ) . The apparent lack of a requirement for these ESCRT-III proteins could reflect less robust depletion of the targeted mRNA than that achieved with ALIX ( Figure 2—figure supplement 5 ) , or possibly the existence of functionally redundant homologs such as CHMP2B . Consistent with the imaging studies , depletion of ALIX had a strong negative effect on the early replication of eHAV , but not naked HAV ( Figure 2F ) . Altogether , these results demonstrate that while both types of virions reach the late endosome , only eHAV is trafficked to the lysosome through an ALIX-dependent mechanism . Although an essential receptor molecule has yet to be identified for HAV ( Das et al . , 2017 ) , studies with other picornaviruses ( Strauss et al . , 2015 ) suggest that the entry of both naked and quasi-enveloped virions is likely to involve binding of the capsid to a specific receptor that triggers uncoating . With quasi-enveloped eHAV , however , this can only occur after the membrane cloaking the capsid is degraded or fuses with a cellular membrane . Fusion seems unlikely given the absence of any virus-encoded proteins in the quasi-envelope ( McKnight et al . , 2017 ) , whereas the selective targeting of eHAV to lysosomes suggests that the quasi-envelope , despite being stable at pH 5 . 0 , might be degraded by cholesterol transporter proteins and hydrolytic enzymes expressed within late endosomes and lysosomes ( Feng et al . , 2013; Kolter and Sandhoff , 2010 ) . A similar process has been suggested recently to facilitate the entry of phylogenetically-distinct , quasi-enveloped hepeviruses ( Yin et al . , 2016 ) . Consistent with this hypothesis , partial siRNA-mediated depletion of the cholesterol transporter Niemann-Pick disease type C1 ( NPC1 ) protein and lysosomal acid lipase ( LAL ) , but not LAMP1 , significantly impaired eHAV but not naked HAV infection , likely through altering the kinetics of quasi-envelope degradation ( Figure 3A , Figure 3—figure supplement 1 ) . Pharmacological inhibition of NPC1 and LAL with U18666A and Lalistat-2 ( Lu et al . , 2015; Rosenbaum et al . , 2010 ) , respectively , recapitulated these effects individually , and demonstrated an additive effect when combined ( Figure 3B ) . To confirm that the quasi-envelope is degraded within the lysosome , we harvested eHAV from supernatant fluids of infected Huh-7 . 5 cells and labelled the virions with the membrane-intercalating , red fluorescent dye PKH26 ( Figure 3C ) . PKH26 irreversibly stains membrane lipids , allowing the labelled virions to be purified subsequently by isopycnic ultracentrifugation , and the fate of the quasi-envelope tracked by confocal microscopy following uptake into cells . We combined this approach with immunostaining cells with a monoclonal antibody to the capsid ( K24F2 ) under minimal permeabilization conditions . This allowed us to visualize capsid antigen associated with PKH26-labeled membranes , and to differentiate eHAV from other EVs with similar density that co-purify in iodixanol gradients ( Feng et al . , 2013; McKnight et al . , 2017 ) . Confocal microscopy of Huh-7 . 5 cells inoculated with the gradient-purified , PKH26-labeled eHAV showed that capsid antigen was surrounded by PKH26-labeled membranes within lysosomes at ~6 hpi ( Figure 3D ) . Thus , eHAV reaches the lysosome cloaked in membranes . At later time points , however , the PKH26 fluorescence was absent although the capsid antigen was still detected within lysosomes , consistent with the quasi-envelope being degraded within the lysosome to produce a naked capsid . As HAV replicates slowly in cell culture ( Whetter et al . , 1994 ) , newly synthesized capsid antigen associated with the generation of progeny virions was not detected until after ~18–24 hpi , at which time there was minimal localization of capsid antigen within lysosomes . As the data presented above suggest that LAL is important for eHAV entry ( Figure 3A , B ) , we monitored loss of the eHAV membrane over time by following the decay of PKH26 fluorescence in cells infected in the presence or absence of the LAL inhibitor , Lalistat-2 ( Figure 3E ) . As expected , inhibition of LAL delayed the kinetics of eHAV membrane loss without altering its trafficking to the lysosome , further supporting a model in which degradation of the eHAV membrane is facilitated by lysosomal enzymes . Interestingly , similar experiments using PKH26-labeled EVs collected from supernatant fluids of uninfected Huh-7 . 5 cells showed targeting of the PKH26 dye to CD63+ endosomes , not lysosomes , without any decay in the fluorescence signal even as late as 24 hr post-inoculation ( Figure 3D , Figure 3—figure supplement 2 ) . This suggests that there may be specific targeting signals present within the eHAV membrane that are absent in unrelated EVs . Although the quasi-envelope protects the virus against neutralization in quantal infectious focus-reduction neutralization assays , our previous studies show that eHAV ( but not naked HAV virions ) can be neutralized within an endocytic compartment when neutralizing IgG or IgA antibodies are added to cells as late as 4–6 hr after adsorption of the virus ( Feng et al . , 2013 ) . To determine whether such neutralization is dependent upon LAL-mediated degradation of the quasi-envelope within lysosomes , we pre-treated Huh-7 . 5 cells with Lalistat-2 prior to infection with naked HAV or eHAV , and added anti-HAV-positive human plasma ( ‘JC’ plasma ) at intervals following removal of the inoculum . Consistent with our previous results ( Feng et al . , 2013 ) , neutralizing antibodies had no effect on replication of the nonenveloped , naked HAV under these conditions ( Figure 3F , top ) , whereas replication of quasi-enveloped eHAV was substantially reduced when antibody was added as late as ~4 hr after adsorption ( Figure 3F , bottom ) . Importantly , however , the period of time during which eHAV was vulnerable to neutralization was extended significantly in cells treated with Lalistat-2 ( Figure 3F , bottom ) . These results are consistent with Lalistat-2 slowing the transition of eHAV to a neutralization-susceptible state . Collectively , these data show that the quasi-enveloped eHAV capsid remains wrapped in membranes until the virion reaches the lysosome , where the quasi-envelope is degraded by lysosomal enzymes and cholesterol transporter proteins , rendering the capsid susceptible to antibody-mediated neutralization and , presumably , interactions with a yet-to-be-identified receptor . To determine whether there are differences in the kinetics of uncoating of naked and quasi-enveloped capsids , we dually immunostained infected cells with a murine monoclonal antibody ( K34C8 ) that recognizes an epitope expressed only on fully assembled capsids , and polyclonal human antibody ( JC plasma ) that recognizes both assembled capsids and assembly intermediates ( 14S pentamers ) ( González-López et al . , 2018; Stapleton et al . , 1993 ) . Infections were done in the presence of cycloheximide to prevent synthesis of new viral proteins , such that uncoating would lead to loss of K34C8 , but not JC , antigenicity . Confocal imaging of cells inoculated with naked HAV particles showed that the K34C8 signal was lost ~1–2 hpi without the capsid ever reaching the lysosome ( Figure 4A ) . In contrast , K34C8-labeled capsid antigen was readily detected at ~4 hpi within lysosomes in cells infected with eHAV . This was followed by a progressive loss of the K34C8 signal , while JC antibody continued to detect capsid antigen within lysosomes up to 12 hpi . Thus , naked virions uncoat relatively rapidly upon entry , likely in a late endosomal compartment , whereas the capsids enclosed within eHAV vesicles do not uncoat until the virus reached the lysosome 4 hr or more following adsorbtion . To assess how differences in the kinetics of uncoating of naked versus quasi-enveloped virions influence the onset and rate of polyprotein translation and viral RNA replication , we inoculated H1-HeLa cells with gradient-purified virions produced by a recombinant reporter virus that expresses nanoluciferase ( HAV-NLuc ) from within its polyprotein . Cells were infected in the presence or absence of the picornaviral RNA synthesis inhibitor , guanidine hydrochloride ( GnHCl ) , and production of nanoluciferase measured over time ( Figure 4B ) . Nanoluciferase expression resulting from translation of the incoming naked virus genome could be detected as early as ~4 hpi , while translation of the eHAV genome was not detectable until ~8 hpi . In both cases , nanoluciferase expression increased similarly in the presence or absence of GnHCl for ~10 hr once translation had commenced , following which accelerating increases in the absence of GnHCl indicated the production of new viral transcripts ( Figure 4B ) . Thus , translation of the genomic RNA as well as the first round of RNA replication occurs sooner with naked HAV than with quasi-enveloped eHAV , consistent with the relatively rapid uncoating of the naked virion . The data presented above indicate that late endosomes and lysosomes are the terminal trafficking destinations of naked and quasi-enveloped HAV virions , respectively , and that the capsids associated with these different virion types uncoat and release their RNA genomes within these distinct endolysosomal compartments . However , it is not clear in either case how the RNA genome released from the capsid is then translocated from the endolysosomal lumen to the cytosol where it is translated on ribosomes . The VP4 capsid peptide possesses membrane pore forming activity in vitro ( Shukla et al . , 2014 ) , but neither HAV nor eHAV has been shown previously to disrupt the integrity of endolysosomal membranes during infection . To determine whether hepatoviruses induce pores in endolysosomal membranes during entry as observed with other picornaviruses , we inoculated cells with eHAV or HAV in the presence of α-sarcin or restrictocin A , membrane-impermeable ribotoxins that are released into the cytoplasm only if endolysosomal membranes are compromised ( Figure 4C ) ( Cuadras et al . , 1997; Fernández-Puentes and Carrasco , 1980; Staring et al . , 2017 ) . Global protein synthesis , quantified by puromycin incorporation , was significantly reduced in cells ~ 6 hr after adsorbtion of eHAV but not naked HAV , in the presence of either α-sarcin or restrictocin A ( Figure 4D , Figure 4—figure supplement 1A , B ) . Reductions in protein synthesis were similar but not as strong as those observed in cells infected with human rhinovirus 14 ( HRV14 ) , included as a positive control in these experiments , and were not observed in cells inoculated with eHAV in the presence of neutralizing anti-HAV antibody which abrogated the ability of eHAV to induce endosomal escape of the ribotoxins ( Figure 4D , Figure 4—figure supplement 1B ) . This effect was specific to eHAV , and naked HAV was never found to induce ribotoxin escape from endosomes at any time post-infection , even under conditions in which it was able to initiate translation of its genome ( Figure 4D , Figure 4—figure supplement 2A , B ) . To confirm that eHAV induces endolysosomal membrane injury and to determine more specifically that it occurs within lysosomes , as expected from the trafficking studies described above , we pre-loaded the lysosomes of Huh-7 . 5 cells with fluorophore-conjugated dextran prior to virus infection . Dextran is a complex , branched glucan that enters cells through fluid-phase pinocytosis and accumulates following its internalization in endolysosomal vesicles positive for LAMP1 and Rab7/LAMP1 ( Humphries et al . , 2011 ) . The release of this pre-loaded dextran from lysosomes into the cytoplasm can be induced by lysosomotropic agents like L-leucyl-L-leucine methyl ester ( LLOMe ) ( Figure 4E , Figure 4—figure supplement 3 ) , providing a useful measure of lysosomal membrane permeability . As expected , cells infected with eHAV demonstrated significantly reduced numbers of dextran-positive compartments as early as six hpi ( Figure 4E ) , a time at which eHAV was seen to be accumulating within lysosomes and sometimes co-localizing with dextran . In contrast , naked virions were never observed in lysosomes and did not alter the number of dextran-containing compartments , even at 12 hpi . Collectively , these data indicate that eHAV uncoats within the lysosomal lumen and induces membrane damage congruent with release of its genome to the cytoplasm , whereas uncoating of naked HAV virions takes place within a late endosomal compartment in the absence of detectable endosomal membrane damage . Whether the absence of detectable endosomal rupture during HAV entry reflects a process that is mechanistically different from that by which eHAV releases its RNA genome across endolysosomal membranes is uncertain . It could be simply that late endosomal membranes breached by HAV are more capable of repair than the lysosomal membrane breached by eHAV . PLA2G16 was identified recently as an essential entry factor for several members of the Picornaviridae ( Staring et al . , 2017 ) . A phospholipase , it facilitates the safe translocation of the RNA genome from the endosome to the ribosome , providing for its escape from autophagosome-dependent degradation initiated by galectin-8 recruited to sites of endosomal membrane damage . To determine whether either HAV or eHAV entry is similarly dependent on PLA2G16 , wild-type and CRISPR/Cas9-edited H1-Hela cells lacking expression of PLA2G16 ( ΔPLA2G16 cells ) or galectin-8 ( ΔLGALS8 cells ) ( Staring et al . , 2017 ) were infected with the nanoluc reporter virus . Surprisingly , neither PLA2G16 or LGALS8 knockout resulted in a difference in nanoluciferase expression 12 hpi with either HAV or eHAV ( Figure 5A ) . Thus , unlike enteroviruses and cardioviruses ( Staring et al . , 2017 ) , PLA2G16 is not required for safe transport of the hepatovirus genome from the endosomal lumen to ribosomes to initiate viral protein synthesis . Although PLA2G16 knockout reduces the permissiveness of H1-HeLa cells for enterovirus infection ( Staring et al . , 2017 ) , longer term studies demonstrated that the replication of both HAV and eHAV was enhanced in ΔPLA2G16 cells , with increased hepatovirus RNA abundance , more dsRNA , and greater viral protein synthesis ( Figure 5B-Figure 5—figure supplement 1A , B ) . The replication of naked HAV was similarly boosted in Huh-7 . 5 cells with siRNA- or CRISPR/Cas9-mediated depletion of PLA2G16 ( Figure 5B , Figure 4—figure supplement 2A , B ) . Thus , PLA2G16 restricts , rather than promotes hepatovirus infection . Further experiments demonstrated that this restriction occurs at a post-entry step in replication of the HAV genome , as replication of a subgenomic reporter replicon RNA ( HAV-FLuc ) lacking capsid-coding sequence was similarly enhanced in PLA2G16 knockout Huh-7 . 5 cells ( Figure 5C ) . Although PLA2G16 is not required for entry of either virion type , confocal imaging showed co-localization of PLA2G16 with eHAV capsid antigen six hpi , presumably at sites of damaged lysosomal membranes ( Figure 5D ) . Co-localization was not observed in cells infected with naked HAV . Thus , PLA2G16 appears to be recruited to sites of endolysosomal membrane damage induced by eHAV , behaving as it does in response to entry of other picornaviruses ( Staring et al . , 2017 ) . However , PLA2G16 is not required to protect the RNA genome from autophagy during its delivery from the endolysosome to ribosomes , suggesting a fundamental difference in how hepatoviruses and other picornaviruses manage the final step in viral entry and deliver their RNA genomes across endolysosomal membranes .
Naked HAV and quasi-enveloped eHAV virions play distinct but equally important roles in the pathogenesis of hepatitis A , with naked HAV virions responsible for fecal-oral transmission of the virus between individuals , and quasi-enveloped eHAV mediating subsequent spread within the newly infected host ( Feng et al . , 2013; Hirai-Yuki et al . , 2016 ) . Here , we describe the entry pathways followed by these two virion types . The early endocytic trafficking routes for these different types of infectious virions are quite similar , but they are differentially sorted within the late endosome and uncoat their encapsidated RNA genomes in different endocytic compartments . The entry of both types of virions requires clathrin- and dynamin-dependent endocytosis and results in trafficking through Rab5+ and Rab7+ endosomal compartments , but only eHAV continues its trafficking to reach the lysosome , where degradation of the quasi-envelope and uncoating of the genome ensues ( Figure 5E ) . In contrast , naked HAV uncoats in late endosomes shortly after internalization , resulting in relatively rapid translation of its genomic RNA . These results provide new insight into how quasi-enveloped hepatoviruses infect the cell , and are likely relevant to pathogenic quasi-enveloped viruses from other families , notably hepeviruses and noroviruses , that are released from infected cells in EVs of comparable size ( Nagashima et al . , 2017; Santiana et al . , 2018 ) . The quasi-envelope represents an elegant strategy for evading antibody-mediated immune responses ( Feng et al . , 2013; Takahashi et al . , 2010 ) , but it imposes a need for additional steps in cellular entry prior to uncoating of the genome . PtdSer is displayed on the eHAV surface and the initial attachment of eHAV to cells occurs in part through the PtdSer receptor , TIM1 ( HAVCR1 ) ( Das et al . , 2017; Feng et al . , 2015 ) . This interaction likely promotes virus spread within the liver , but TIM1 is not essential for infection ( Das et al . , 2017 ) . We show here that integrin β1 is also involved in quasi-enveloped virus entry . It is not required for attachment to the cell at 4˚C , but is essential for efficient internalization of eHAV at 37˚C ( Figure 1D , E ) . It acts similarly in uptake of the naked virion , but must do so through interactions with a different ligand since integrin β1 co-localized with both virion types on the cell surface , and within endosomes , hours before degradation of the quasi-envelope ( Figure 1I ) . Consistent with this , activating antibodies that stabilize specific conformations of integrin β1 ( Su et al . , 2016 ) differentially enhanced the uptake of the two virion types ( Figure 1H ) . Integrin β1 facilitates uptake of several other picornaviruses through interactions with their capsids ( Merilahti et al . , 2012 ) , and the VP3 capsid protein of HAV contains conserved RGD and KGE integrin recognition motifs . However , neither motif is exposed on the surface of the naked HAV capsid ( Wang et al . , 2015 ) , and thus both are unlikely ligands for integrin β1 . We conclude that integrin β1 binds elsewhere on the HAV capsid , likely in association with an α integrin , and also interacts with a host protein on the surface of eHAV . We were unable to identify a specific α integrin involved in entry of either virion type , but sub-optimal knockdown conditions and/or promiscuity of integrin β1 for an α integrin partner , may have masked a specific role for an α integrin ( s ) . Collectively , our data show trafficking of quasi-enveloped virus to the lysosome is essential for entry and uncoating of its genome . In addition to Rab5 and Rab7 GTPases , this requires the ESCRT accessory protein , ALIX ( Figure 2 ) , likely due to the critical role it plays in regulating trafficking from late endosomes to the lysosome ( Murrow et al . , 2015 ) . ALIX mediates the ubiquitin-independent sorting and trafficking of certain G-protein coupled receptors ( GPCRs ) to lysosomes through interactions with YPX3L motifs ( Dores et al . , 2016 ) . The VP2 capsid protein of HAV possesses two such YPX3L ALIX interaction motifs , and we have shown that these have an essential role in the biogenesis of eHAV ( Feng et al . , 2013; González-López et al . , 2018 ) . Thus , ALIX is required for both efficient entry and release of quasi-enveloped hepatovirus . However , ALIX mediates sorting of eHAV to the lysosome prior to degradation of its membrane , and thus promotes the entry process at a point during which the VP2 YPX3L motifs are occluded by the quasi-envelope , and moreover are within the lumen of the endosome and not available to interact with cytoplasmic ALIX . Rather than a direct interaction with the virus , the requirement for ALIX in eHAV entry is more likely to reflect a role for the ESCRT-associated protein in maturation and trafficking of the late endosome , akin to its suggested role in entry of human papillomavirus and arenaviruses ( Gräßel et al . , 2016; Pasqual et al . , 2011 ) . Is there a specific signal that directs the endocytosed eHAV virion to lysosomes ? We found that nonspecific EVs released from uninfected Huh-7 . 5 cells ( most likely exosomes ) did not traffic to lysosomes , and that PKH26-labelled membranes associated with these vesicles decayed within the cell much more slowly than the eHAV membrane ( Figure 3D , Figure 3—figure supplement 2 ) . These results are consistent with previous studies of exosome entry that show PKH26-labeled exosome membranes can be tracked within cells for 24 hr or longer after entry with little or no accumulation in lysosomes ( Dutta et al . , 2014; Ringuette Goulet et al . , 2018; Svensson et al . , 2013 ) . This suggests the existence ( or possibly the absence ) of a specific targeting signal within the eHAV membrane that results in these virions being routed to the lysosomal lumen for degradation of the quasi-envelope . Quantitative proteomics studies provide some support for this hypothesis , as such studies show differential enrichment of host proteins associated with eHAV versus exosomes released from the same cells ( McKnight et al . , 2017 ) . Within the lysosome , both LAL and NPC1 contribute to the degradation of the quasi-envelope required for uncoating of the genome ( Figure 3 ) , recapitulating the function of these lysosomal proteins in the quasi-enveloped hepevirus life cycle ( Yin et al . , 2016 ) . No other picornavirus is known to be trafficked to the lysosome for uncoating . What triggers uncoating of the eHAV capsid after degradation of the quasi-envelope within the lysosome , and is this trigger the same as that for naked capsids within late endosomes ? For picornaviruses of the Aphthovirus genus , low pH alone is sufficient to promote the dissociation of the capsid into pentameric subunits ( Tuthill et al . , 2009 ) . However , the HAV capsid is highly resistant to acid pH ( Siegl et al . , 1981 ) . An alternative model is provided by poliovirus , which interacts with a specific receptor ( CD155 ) that triggers a massive conformational rearrangement of the capsid providing for safe transfer of genomic RNA across the endosomal membrane ( Groppelli et al . , 2017; Strauss et al . , 2015 ) . This is accompanied by evidence of endosomal pore formation in ribotoxin assays , such as that we show here for quasi-enveloped hepatoviruses ( Figure 4D ) ( Schober et al . , 1998; Staring et al . , 2017 ) . However , we did not detect pore formation in cells infected with naked HAV , even under conditions in which we documented translation of the genomic RNA , and thus successful translocation of the genome across the endosomal membrane ( Figure 4—figure supplements 1 and 2 ) . Crystallographic studies have identified substantial differences in the structures of the hepatovirus and poliovirus capsids , including a domain swap in VP2 and the absence of a lipid ‘pocket factor’ in HAV ( Wang et al . , 2015 ) . Interactions of the poliovirus capsid with its cellular receptor lead to the release of this pocket factor and an irreversible expansion of the capsid ( Strauss et al . , 2015 ) . Whether a similar expansion of the HAV capsid occurs during the process of its uncoating is unknown . The HAV capsid is exceptionally stable , and how it uncoats is enigmatic ( Stuart et al . , 2018 ) . Recent studies show that monoclonal antibodies that bind with high affinity are capable of destabilizing the HAV capsid , possibly mimicking a specific receptor interaction ( Wang et al . , 2017 ) . The nature of that putative receptor remains unknown , but data we present here suggest that it is likely to be present in late endolysosomal membranes . We found that the phospholipase , PLA2G16 , is not required for safe translocation of the RNA genome in either virion type from the endolysosomal lumen to the cytoplasm . These observations stand in sharp contrast to the essential role of PLA2G16 in the entry of multiple other picornaviruses ( Staring et al . , 2017 ) . Our data point collectively to fundamental differences in the mechanism ( s ) by which hepatoviruses and other picornaviruses accomplish the endgame in entry , delivering their RNA genome to the ribosome where synthesis of the viral polyprotein can commence . The data leave open the possibility that this process differs for eHAV and HAV not only in where it occurs spatially within the endolysosomal system , but also in its molecular details . This question is likely to be resolved only after it is determined whether a specific receptor protein exists that is capable of triggering uncoating of the hepatovirus capsid .
Huh-7 . 5 cells ( obtained from Charles Rice , Rockefeller University ) were maintained in Dulbecco’s modified Eagle medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) . Wild type , ΔPLA2G16 , and ΔLGALS8 H1-HeLa cells have been previously characterized ( Staring et al . , 2017 ) and were maintained in DMEM supplemented with 10% FBS and 2 mM GlutaMAX . Cell lines were validated by phenotypic screening and confirmed to be mycoplasma-free using a PCR detection kit ( Sigma-Millipore , # MP0035 ) . To produce knockout Huh-7 . 5 cells , CRISPR/Cas9-expressing lentiviruses were generated by co-transfection of 293FT cells with sgRNA lentivectors ( Applied Biological Materials , Supplementary file 1 ) and a third-generation lentivirus packaging mix kit ( Applied Biological Materials , #LV053-G074 ) . Supernatant fluids were collected at 48–72 hr post-transfection , spun to remove cell debris , and filtered through a 0 . 45 µm filter syringe . Lentivirus transduction of Huh-7 . 5 or H1-HeLa cells was performed by supplementation of 8 µg . ml−1 polybrene followed by antibiotic selection with 6 or 10 µg . ml−1 puromycin , respectively , was performed as previously described ( Das et al . , 2017 ) . All cells were maintained at 37˚C in a 5% CO2 atmosphere . Infectious clones of HM175/p16 . 2 virus ( low-passage , non-cytopathic , cell culture-adapted , GenBank KP879217 . 1 ( McKnight et al . , 2017 ) ) and HM175/18 . 2 ( high cell culture passage , rapid replication , cytopathic , GenBank KP879216 . 1 ( González-López et al . , 2018; Zhang et al . , 1995 ) are variants of the HM175 strain ( Jansen et al . , 1988; Taylor et al . , 1993 ) and have been previously described . The HM175/18 f . 2-NanoLuc ( HAV-NLuc ) plasmid was created by PCR amplifying the NLuc ORF using pNL1 . 1 ( Promega ) as template and oligos containing the tri-glycine sequence flanked by XbaI and BamH1 restriction sites . The PCR amplicon was enzymatically digested and ligated into digested pSK-2A-Zeo-2B plasmid . The resulting plasmid was further digested with SacI/PflMI to release the entire 2A-NLuc-2B fragment which was then ligated into a similarly digested HM175/18 f parental plasmid . Infectious HAV mRNA transcripts were generated in vitro using the T7 RiboMAXTM Express Large-Scale RNA Production System ( Promega ) as per manufacturer’s protocol and transfected into Huh-7 . 5 cells by electroporation in a Gene Pulser Xcell Total System ( Bio-Rad ) as previously described ( Feng et al . , 2013 ) . Cell culture supernatant fluids were then collected ( 9 to 15 days post-transfection ) and centrifuged at 1 , 000 ×g for 10 min at 4˚C to remove debris and further clarified at 10 , 000 ×g for 30 min at 4˚C . The virus was concentrated by ultracentrifugation at 100 , 000 ×g for 60 min at 4˚C , and the resulting pellet was resuspended in 250 µl phosphate buffer saline ( PBS ) and loaded on top of a five-step gradient of 8% to 40% iodixanol ( OptiPrep , Sigma ) and centrifuged at 165 , 915 ×g ( 37 , 000 rpm ) for 24 hr at 4˚C in a Beckman SW55i rotor using a Beckman Optima LE-80K ultracentrifuge . Approximately 20 fractions were collected from the top of the isopycnic gradient , and HAV RNA content and density were quantified by reverse transcription-quantitative PCR ( RT-qPCR ) and refractometry , respectively , as previously described ( Feng et al . , 2013; McKnight et al . , 2017 ) . Fractions containing eHAV and HAV at the appropriate buoyant densities ( for eHAV , approximately 1 . 08 g/cm3 , fractions 9 to 11; for naked HAV , approximately 1 . 22 g/cm3 , fractions 18–19 ) were stored in aliquots at –80˚C until use . Huh-7 . 5 cells were seeded on 96-well clusters and transfected with in vitro transcribed subgenomic HAV-FLuc replicon or a replication-incompetent mutant ( González-López et al . , 2018; Yi and Lemon , 2002 ) using TransIT-mRNA transfection kit ( Mirus Bio , #MIR2250 ) according to the manufacturer’s instructions . Cells were harvested in 1 × passive lysis buffer ( PLB , Promega ) at the indicated times post-transfection and luciferase activity was measured using a firefly luciferase assay system ( Promega , #E1501 ) . For HAV-NanoLuc assays , cells were lysed for 5–10 min in 1 × PLB and mixed with 1 × substrate for Oplophorus luciferase ( NanoLight Technology , #325 ) according to the manufacturer’s instructions . All luciferase readings were obtained on a BioTek Synergy two microplate reader . For viral entry assays , 5 × 104 Huh-7 . 5 cells seeded on 12-well chamber slides were pre-treated with the indicated inhibitors at the following concentrations: 80 µM dynasore ( Millipore , #324410 ) , 10 µg . ml−1 chlorpromazine ( Sigma-Aldrich , #C8138 ) , 1 µg . ml−1 filipin ( Sigma-Aldrich , #F9765 ) , 20 µM cytochalasin D ( Sigma-Aldrich , #C2618 ) , 1 µM latrunculin A ( Sigma-Aldrich , #428026 ) , 1 µM EIPA ( 5- ( N-Ethyl-N-isopropyl ) amiloride ) ( Sigma-Aldrich , #A3085 ) , 1 µM wortmannin ( Sigma-Aldrich , #W1628 ) , 300 µM NSC23766 ( Sigma-Aldrich , #SML0952 ) , 25 µM dynarrestin ( kindly provided by Dr . Jared Sterneckert , Max Planck Institute for Molecular Biomedicine , ( Höing et al . , 2018 ) ) , 10 µg . ml−1 heparin ( Sigma-Aldrich , #H3149 ) , 200 µM Lalistat-2 ( kindly provided by Paul Helquist and Bruce Malencon , University of Notre Dame ) , 2 µg . ml−1 U18666A ( Sigma-Aldrich , #U3633 ) , or dimethyl sulfoxide ( DMSO ) solvent control in supplemented DMEM for 60–120 min at 37˚C prior to virus adsorption . For functional integrin assays , cells were incubated with 100 µM RGD peptide ( Santa Cruz Biotechnology , #sc-201176 ) for 2 hr; or 10 µg . ml−1 of either mouse IgG ( Abcam , #ab37355 ) , K-20 ( Santa Cruz Biotechnology , #sc-18887 ) , TS2/16 ( Santa Cruz Biotechnology , #sc-53711 ) , 8E3 ( Millipore-Sigma , #MABT199 ) , or HUTS-4 ( Millipore-Sigma , #MAB2079Z ) for 20 min on ice prior to virus adsorption at 37˚C . Other inhibitors include 5 mM guanidine hydrochloride ( Sigma , #G3272 ) , 25 µg . ml−1 cycloheximide ( Sigma , #C7698 ) , 5–20 µg . ml−1 puromycin ( InvivoGen ) , 100 µg . ml−1 α-sarcin ( Santa Cruz Biotechnology , #sc-204427 ) , 50 µg . ml−1 Restrictocin A ( Sigma-Aldrich , #R0389 ) , and 2 mM L-Leucyl-L-Leucine-methyl ester ( Cayman Chemical , #16008 ) . RNA was extracted from cell lysates with the RNeasy Kit ( Qiagen ) and cDNA was synthesized with oligo ( dT ) 20 followed by RNaseH digestion . HAV RNA GEs were quantified in a SYBR Green Real-Time qPCR ( Bio-Rad ) assay against a synthetic RNA standard curve using primers targeting the HAV 5’ untranslated region as previously described ( Feng et al . , 2013 ) and HAV RNA levels were normalized to total µg RNA . For siRNA-mediated knockdown efficiency , host mRNA target abundance was determined using gene-specific primers ( Supplementary file 1 ) and normalized to glyceraldehyde-3-phosphate dehydrogenase levels; efficiency was calculated as the percent mRNA expression relative to non-targeting control siRNA samples . Huh-7 . 5 cells were transfected with 50–75 nM gene-specific SMARTPool ON-TARGETplus siRNAs ( Dharmacon , Supplementary file 1 ) using the Lipofectamine RNAiMAX transfection reagent ( Thermo Fisher Scientific ) according to the manufacturer’s instructions . Three-to-four days post-transfection , cells were adsorbed with equal quantities of HAV genome equivalents ( GEs ) of purified naked or quasi-enveloped HAV ( ~10 GEs per cell ) for 1 hr at 37˚C . The inoculum was removed , the cells were rinsed with PBS and incubated at 37˚C in fresh medium . Cells were lysed in radioimmunoprecipitation assay ( RIPA ) buffer ( 50 mM Tris-HCl [pH 7 . 4] , 1% NP-40 , 0 . 25% sodium deoxycholate , 150 mM NaCl , 1 mM EDTA , 1% sodium dodecyl sulfate [SDS] ) supplemented with a cocktail of protease and phosphatase inhibitors for 20 min on ice and then clarified at 14 , 000 ×g for 10 min at 4˚C . Total protein concentration was determined using a bicinchoninic acid assay ( Thermo Fisher Scientific ) . A total of 5–20 µg of protein was boiled for 5 min in Laemmli sample buffer , resolved by SDS-polyacrylamide gel electrophoresis ( SDS-PAGE ) , and transferred to a polyvinylidene fluoride ( PVDF ) membrane by standard methods . Membranes were blocked with Odyssey blocking buffer ( LI-COR Bioscience ) , probed with the indicated primary antibodies , and incubated with infrared-conjugated ( IRDye ) secondary antibodies ( LI-COR Biosciences ) . Proteins were visualized using an Odyssey Infrared Imaging System ( LI-COR Biosciences ) . Huh-7 . 5 or H1-HeLa cells seeded on 8-well chamber slides were adsorbed with equal quantities of HAV genome equivalents ( GEs ) of purified naked or quasi-enveloped HAV ( ~1000 GEs per cell ) for 1 hr at 37˚C . The inoculum was removed , the cells were rinsed with PBS and reincubated at 37˚C in fresh medium . Cells were then fixed in 4% paraformaldehyde ( PFA ) for 20 min at the indicated hours post-inoculation ( hpi ) and , unless stated otherwise , permeabilized with 0 . 25% Triton X-100 in PBS for 10 min . Slides were blocked with 5–10% normal goat serum ( Sigma-Aldrich ) in PBS for 60 min , and incubated with the indicated primary antibodies diluted in 0 . 1% IgG- and protease-free bovine serum albumin ( BSA ) ( Jackson ImmunoResearch , #001–000 ) for 1–2 hr at room temperature . Slides were extensively rinsed in PBS and incubated with the appropriate species-specific Alexa Fluor-conjugated secondary antibodies ( Thermo Fisher Scientific ) diluted in 0 . 1% IgG- and protease-free BSA for 1 hr at room temperature . Nuclei were counterstained with 300 nM DAPI ( 4’ , 6-diamidino-2-phenylindole ) and coverslips were mounted on slides using ProLong Gold ( Thermo Fisher Scientific , #P36930 ) . Antibodies used for immunoblotting and their corresponding dilutions were anti-clathrin heavy chain ( Abcam , ab21679 , 1:2000 ) , anti-AP2M1 ( GeneTex , GTX113332 , 1:1000 ) , anti-DNM2 ( GeneTex , GTX113171 , 1:1000 ) , anti-CAV1 ( Abcam , ab2910 , 1:2000 ) , anti-FLOT1 ( GeneTex , GTX104769 , 1:1000 ) , anti-ARF6 ( GeneTex , GTX112872 , 1:1000 ) , anti-Rab5A ( Abcam , ab18211 , 1:1000 ) , anti-Rab7a ( Cell Signaling , 9367 , 1:1000 ) , anti-Rab11 ( Cell Signaling , 5589 , 1:1000 ) , anti-NPC1 ( Abcam , ab134113 , 1:2000 ) , anti-LAL ( GeneTex , GTX101169 , 1:1000 ) , anti-ALIX ( Santa Cruz Biotech , sc-53540 , 1:500 ) , anti-puromycin ( Millipore , MABE343 , 1:10000 ) , anti-PLA2G16 ( Cayman Chemical , 10337 , 1:200 ) , anti-Galectin-8 ( R and D Systems , AF1305 , 1:500 ) , anti-tubulin ( Sigma , T6199 , 1:20000 ) , and anti-actin ( Sigma , A2066 , 1:5000 ) , anti-VCAM-1 ( R and D Systems , AF809 , 1:250 ) , anti-ICAM-1 ( R and D Systems , AF720 , 1:250 ) , anti-Tspan8 ( R and D Systems , MAB4734 , 1:150 ) , anti-integrin β1 ( Cell Signaling , 9699 , 1:1000 ) , anti-integrin β3 ( Cell Signaling , 13166 , 1:500 ) , anti-integrin α1 ( R and D Systems , MAB5676 , 1:250 ) , anti-integrin α2 ( Abcam , ab133557 , 1:500 ) , anti-integrin α3 ( Millipore , AB1920 , 1:250 ) , anti-integrin α4 ( Cell Signaling , 8440 , 1:250 ) , anti-integrin α5 ( Cell Signaling , 4705 , 1:700 ) , anti-integrin αV ( Cell Signaling , 4711 , 1:700 ) . anti-integrin α6 ( GeneTex , GTX100565 , 1:500 ) , anti-integrin α7 ( Thermo Fisher Sci , PA5-37435 , 1:250 ) , anti-integrin α8 ( Novus Biologicals , NBP1-59940 , 1:250 ) , and anti-integrin α9 ( R and D Systems , MAB4574 , 1:250 ) . Antibodies user for indirect immunofluorescence and their corresponding dilutions were anti-clathrin heavy chain ( Abcam , ab21679 , 1:1000 ) , anti-CAV1 ( Abcam , ab2910 , 1:500 ) , anti-integrin β1 ( Abcam , ab30394 , 1:100 ) , anti-integrin α5 ( Abcam , ab150361 , 1:250 ) , anti-integrin αV ( Abcam , ab179475 , 1:500 ) , anti-Rab5A ( Cell Signaling , 3547 , 1:200 ) , anti-Rab7a ( Cell Signaling , 9367 , 1:100 ) , anti-Rab11 ( Cell Signaling , 5589 , 1:100 ) , anti-LAMP1 ( Cell Signaling , 9091 , 1:200 ) , anti-VAMP8 ( Abcam , ab76021 , 1:250 ) , anti-CD63 ( BD Biosciences , #556019 , 1:50 ) , anti-PLA2G16 ( Sigma , H8290 , 1:50 ) , J2 anti-dsRNA ( Scicons , J2 clone , 1:1 , 000 ) , postconvalescent polyclonal anti-HAV human plasma ‘JC’ ( Feng et al . , 2013 , 1:600 ) , anti-HAV capsid K24F2 and K34C8 ( MacGregor et al . , 1983 , 1:100 and 1:300 , respectively ) . Huh-7 . 5 cells seeded on 8-well chamber slides were treated with DMEM supplemented with the indicated inhibitors for 1 hr . Cells were then rinsed , placed on ice for 10 min , and incubated with 10–25 µg . ml−1 Alexa 594-conjugated Transferrin ( Thermo Fisher , #T13343 ) or cholera toxin subunit B ( Thermo Fisher , #34777 ) diluted in supplemented DMEM for 15–20 min at 37˚C . Cells were then fixed with 4% paraformaldehyde ( PFA ) and nuclei were counterstained with 300 nM DAPI . Supernatant fluids from uninfected or HAV-infected Huh-7 . 5 cells were clarified and concentrated by ultracentrifugation at 100 , 000 ×g for 1 hr as described above . The pellet was then resuspended 250 µl of Diluent C and mixed with 2 µM PKH26 red fluorescent cell linker ( Sigma-Aldrich , MIDI26 ) diluted in Diluent C for 5 min according to the manufacturer’s instructions . The staining was blocked with 500 µl FBS for 3 min and the labeled vesicles/eHAV were loaded onto an iodixanol gradient and ultracentrifuged at 165 , 915 ×g ( 37 , 000 rpm ) for 24 hr at 4˚C as described above . Fractions containing eHAV and EVs at the appropriate buoyant densities ( approximately 1 . 08 g/cm3 , fractions 9 to 11 ) were stored in aliquots at 4˚C until use . Recipient Huh-7 . 5 cells were inoculated with a 1:10 dilution of the fractions diluted in complete DMEM and fixed in 4% PFA . Slides were blocked in 10% normal goat serum , and incubated simultaneously with anti-HAV capsid ( K24F2 ) and anti-LAMP1 for 1 hr diluted in 0 . 01% saponin ( Sigma-Aldrich , S2149 ) , carefully rinsed , and incubated with a mix of DAPI and Alexa fluor-conjugated secondary antibodies diluted in 0 . 01% saponin for 45 min at room temperature . Slides were mounted on ProLong Gold . Huh-7 . 5 cells seeded on 12-well clusters ( 1 × 105 cells per well ) were pre-treated with 200 µM Lalistat-2 or DMSO solvent control for 1 hs at 37˚C and then adsorbed with equal quantities of HAV genome equivalents ( GEs ) of gradient-purified naked or quasi-enveloped HAV ( ~1 GE per cell ) for 1 hr at 37˚C . The inoculum was removed , cells were rinsed three times with PBS , and replaced with fresh DMEM supplemented with 10% FBS and 200 µM Lalistat-2 or DMSO . At the indicated times post-infection , media was replaced with postconvalescent human plasma ( ‘JC plasma’ ) collected several months following symptomatic acute hepatitis A ( Feng et al . , 2013 ) or normal human serum control diluted 1:50 in DMEM . Intracellular viral RNA was harvested at 48 hpi , cDNA was synthesized , and HAV RNA levels were quantified by RT-qPCR as described above . For endosomal membrane integrity , H1-HeLa cells were inoculated with equal amounts ( ~1000 HAV GEs per cell ) of gradient-purified HAV or eHAV , or 10 plaque-forming units ( PFU ) per cell of human rhinovirus-14 ( HRV-14 ) ( McKnight and Lemon , 1996 ) in supplemented DMEM presence of α-sarcin or Restrictocin A previously reconstituted in sterile ultrapure water and incubated for 6 hr at 37˚C ( for HAV ) or 33˚C ( for HRV-14 ) . Cells were then incubated with supplemented DMEM containing 20 µg . ml−1 puromycin for 20 min , rinsed twice with PBS , and total protein lysates were harvested as described above . Specific puromycin incorporation was validated by pre-treating cells with cycloheximide for 30 min prior to the puromycin pulse ( Figure 4—figure supplement 1 ) . For lysosomal membrane integrity analysis , Huh-7 . 5 cells seeded in 8-well chamber slides were loaded with 100 µg . ml−1 anionic-lysine fixable Alexa Fluor 594-conjugated dextran ( 10 kDa ) ( Thermo Fisher Scientific , #D22913 ) diluted in supplemented DMEM for 16 hr at 37˚C . Cells were rinsed with PBS , inoculated with purified HAV or eHAV ( 1000 HAV GEs per cell ) , and fixed as indicated . Slides were examined with an Olympus FV10000 laser-scanning confocal microscope equipped with a super corrected 60×/1 . 4 NA oil-immersion objective and a dichroic mirror DM405/488/543/635 was used for all experiments . The pinhole was maintained at 1 Airy unit and images were acquired in two separate channels to prevent bleed-through . The excitation/emission wavelengths were 405 nm/425–520 nm for DAPI , 488 nm/500–520 nm for Alexa Fluor 488 , 543 nm/555–647 nm for Alexa Fluor 594 or PKH26 , and 635 nm/647–700 nm for Alexa Fluor 647 . Intensity plot profiles were generated using the ImageJ software and co-localization indexes ( Mander’s coefficients ) were obtained with the Just Another Colocalisation Plugin ( JACoP ) module for ImageJ . All micrographs are representative of at least 10 images for each sample per experiment , and each experiment was performed at least twice . Images were processed for presentation using Photoshop CS4 . Unless stated otherwise , significance was assessed by unpaired t tests or ANOVA calculated with GraphPad Prism seven for Windows software . Significance values are shown as ****p<0 . 0001 , ***p<0 . 001 , **p<0 . 01 , *p<0 . 05 . | The Hepatitis A virus is a common cause of liver disease in humans . It is unable to multiply on its own so it needs to enter the cells of its host and hijack them to make new virus particles . Infected human cells produce two different types of Hepatitis A particles . The first , known as ‘naked’ virus particles , consist of molecules of ribonucleic acid ( or RNA for short ) that are surrounded by a protein shell . Naked virus particles are shed in the feces of infected individuals and are very stable , allowing the virus to spread in the environment to find new hosts . At the same time , a second type of particle , known as the ‘quasi-enveloped’ virus , circulates in the blood of the infected individual . In a quasi-enveloped particle , the RNA and protein shell are completely enclosed within a membrane that is released from the host cell . This membrane protects the protein shell from human immune responses , enabling quasi-enveloped virus particles to spread in a stealthy fashion within the liver . It was not clear how these two different types of virus particle are both able to enter cells despite their surface being so different . To address this question , Rivera-Serrano et al . used a microscopy approach to observe Hepatitis A particles infecting human liver cells . The experiments showed that both types of virus particle actually use similar routes . First , the external membrane of the cell folded around the particles , creating a vesicle that trapped the viruses and brought them within the cell . Inside these vesicles , the naked virus particles soon fell apart , and their RNA was released directly into the interior of the cell . However , the vesicles that carried quasi-enveloped virus travelled further into the cell and eventually delivered their contents to a specialized compartment , the lysosome , where the virus membrane was degraded . This caused the quasi-enveloped viruses to fall apart and release their RNA into the cell more slowly than the naked particles . Several viruses , such as the one that causes polio , also have quasi-enveloped forms . Studying how these particles are able to infect human cells while hiding behind membranes borrowed from the host may help us target these viruses better . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | Cellular entry and uncoating of naked and quasi-enveloped human hepatoviruses |
Correlations in brain activity between two areas ( functional connectivity ) have been shown to relate to their underlying structural connections . We examine the possibility that functional connectivity also reflects short-term changes in synaptic efficacy . We demonstrate that paired transcranial magnetic stimulation ( TMS ) near ventral premotor cortex ( PMv ) and primary motor cortex ( M1 ) with a short 8-ms inter-pulse interval evoking synchronous pre- and post-synaptic activity and which strengthens interregional connectivity between the two areas in a pattern consistent with Hebbian plasticity , leads to increased functional connectivity between PMv and M1 as measured with functional magnetic resonance imaging ( fMRI ) . Moreover , we show that strengthening connectivity between these nodes has effects on a wider network of areas , such as decreasing coupling in a parallel motor programming stream . A control experiment revealed that identical TMS pulses at identical frequencies caused no change in fMRI-measured functional connectivity when the inter-pulse-interval was too long for Hebbian-like plasticity .
Temporal correlations in activity between brain areas can be measured with functional magnetic resonance imaging ( fMRI ) and are often referred to as indices of ‘functional connectivity’ ( Friston , 1994 ) . Statistical dependencies between remote cortical regions exist both in the absence of external stimuli or task demands ( i . e . , during the resting state ) and during execution of a task ( Hampson et al . , 2002 ) . In the following , the term functional connectivity therefore simply describes a statistical relationship of neural elements with each other . In addition to providing insights into the basic anatomical and physiological organization of healthy neural networks ( Fox and Raichle , 2007; Bullmore and Sporns , 2009; Smith et al . , 2009; O'Reilly et al . , 2013 ) , functional connectivity has been used to identify pathological changes occurring in neural circuits in conditions such as stroke ( Wang et al . , 2010 ) , traumatic brain injury ( Bonnelle et al . , 2011; Ham and Sharp , 2012 ) , neurodegeneration ( Seeley et al . , 2009 ) , or psychiatric disorders ( Bassett et al . , 2008 ) . Although changes in fMRI-based functional connectivity can be highly specific , their underlying biological mechanisms are less clear . It is thought that functional connectivity patterns are shaped largely by the relatively stable underlying skeleton of structural connections ( O'Reilly et al . , 2013 ) . However , modifications in functional connectivity might also be influenced by changes in synaptic efficacy , for example through changes in the quantity of neurotransmitter release , changes in astrocytes or dendritic spine stabilization . Here , we aimed to elucidate the contribution of changes in short-term synaptic efficacy to fMRI-based functional connectivity . To this aim , we modulated synaptic efficacy in a specific corticocortical pathway using repetitive paired pulses of transcranial magnetic stimulation ( TMS ) with a brief inter-pulse interval ( IPI; 8 ms ) that evoked synchronous pre- and post-synaptic activity and monitored whether those changes were reflected in altered functional connectivity ( Experiment 1 ) ( Figure 1A ) . Several TMS protocols have been shown to induce changes in excitability in primary motor cortex ( M1 ) using repetitive stimulation of M1 itself ( Chen et al . , 1997 ) or stimulation of premotor regions projecting to M1 ( Munchau et al . , 2002 ) . These changes are often thought to reflect frequency-dependent potentiation of synaptic transmission . Furthermore , it has been shown that repetitive paired stimulation of an input into M1—such as the median nerve—and then of M1 itself can change M1 cortico-spinal excitability ( Stefan et al . , 2000; Wolters et al . , 2003 ) . These paired associative stimulation ( PAS ) protocols are based upon Hebbian principles of synaptic plasticity and appear to modify connectivity in a controlled manner . Investigations that applied paired-pulse TMS over interconnected sites—for example , homotopical M1 sites ( Rizzo et al . , 2009 ) , M1 and the supplementary motor area ( SMA ) ( Arai et al . , 2011 ) , and M1 and posterior parietal cortex ( Koch et al . , 2013 ) —demonstrated altered motor cortical excitability . Notably , the current protocol of repetitive paired-pulse TMS has been shown to induce a causal and directional change of influence of the first brain region ( ventral premotor cortex: PMv ) over the anatomically connected second region ( M1 ) ( Buch et al . , 2011 ) . This is important since it is such pathway-specific changes that occur in animal models of synaptic plasticity ( Markram et al . , 1997; Jackson et al . , 2006 ) and these are argued to underlie the self-organization proposed to occur in mono-synaptically connected networks in response to regularly occurring input ( Sussillo and Abbott , 2009 ) . 10 . 7554/eLife . 04585 . 003Figure 1 . Experimental procedure . ( A ) 90 paired pulses were applied over ventral frontal cortex near PMv and M1 ( mean MNI coordinates [−56 19 19] and [−40 −18 59] respectively ) at 0 . 1 Hz for 15 min . ( B ) Individual stimulation locations for 8 ms IPI ( red ) and 500 ms IPI ( blue ) . ( C ) Participants performed visually guided grasping movements towards one of two objects ( small or large; see inset ) while lying supine in the MR scanner . The head coil was tilted forward by 30° to allow for direct line of sight of the objects to be grasped . A response button box was positioned on the upper leg . ( D ) Experimental design and setup for all experiments ( for both 8 ms IPI and 500 ms IPI experiment ) . The order of resting-state and grasping task fMRI as well as of pre-TMS ( baseline ) and post-TMS sessions was counterbalanced . DOI: http://dx . doi . org/10 . 7554/eLife . 04585 . 003 PMv and M1 are a part of the so-called ‘dorsolateral circuit’ of areas composed of the anterior intraparietal ( AIP ) area , areas PF and PFG in the inferior parietal lobule , and PMv and M1 in the frontal lobes . During complex motor behaviour such as reaching and grasping this dorsolateral sensorimotor stream is complemented by a ‘dorsomedial circuit’ composed of dorsal premotor ( PMd ) , medial intraparietal area ( MIP ) , and posterior superior parietal cortex ( pSPL ) ( Jeannerod et al . , 1995; Wise et al . , 1997; Tanne-Gariepy et al . , 2002; Galletti et al . , 2003; Brochier and Umiltà , 2007; Grafton , 2010; Turella and Lingnau , 2014 ) . As is the case for other inter-regional connections , the connections between premotor cortex and M1 are glutamatergic , excitatory ones , but there are synapses on both pyramidal neurons and inhibitory interneurons within M1 ( Tokuno and Nambu , 2000 ) . This means that although paired stimulation of PMv and M1 leads to strengthening of the excitatory connections between PMv and M1 , such strengthening can lead to both enhanced facilitatory and enhanced inhibitory influences of PMv on M1 . Enhanced facilitatory influences are more apparent when subjects are subsequently tested while performing a simple reaching and grasping task , and enhanced inhibitory influences are more apparent when subjects are subsequently tested while at rest ( Buch et al . , 2011 ) . These different effects appear as a function of the subject's behavioural and cognitive state at the time of testing ( Bäumer et al . , 2009; Buch et al . , 2010 ) , but they have not been shown to depend on the subject's cognitive state at the time of plasticity induction ( Buch et al . , 2011 ) . PMv microwire stimulation in macaques has also been shown to exert both facilitatory and inhibitory effects on corticospinal outputs as a function of the animal's state ( Prabhu et al . , 2009 ) . In the following , this causal and directional influence as quantified by motor-evoked potentials ( MEPs ) is referred to as ‘effective connectivity’ ( Friston , 1994 ) . Classic Hebbian synaptic learning rules such as pathway specificity , spike timing dependency , rapid evolution , persistence for several hours , and reversibility have been demonstrated for this directed pathway manipulation . Using TMS in this way entails a direct and specific inter-areal manipulation distinct from the compensatory plasticity that occurs following single-site manipulations of neural activity by means of TMS ( Lee et al . , 2003; O'Shea et al . , 2007; Grefkes et al . , 2010; Hartwigsen et al . , 2012 ) . Recent studies have shown that lesions and disruption of brain areas as well as lesions to connections between brain areas can affect distant areas and connections ( O'Shea et al . , 2007; Hartwigsen et al . , 2012; O'Reilly et al . , 2013 ) . These changes are thought to be partly compensatory . For example , in the study by O'Shea et al . ( 2007 ) , it is suggested that ‘activity’ in contralateral ‘non-dominant’ PMd is increased after interruption of ipsilateral PMd . This enhancement of contralateral PMd is accompanied by preserved performance in a stimulus-response matching task . Similarly , Hartwigsen et al . ( 2012 ) show that action reprogramming can be preserved after PMd interference if the supramarginal gyrus is uncompromised . This study suggests a rapid redistribution of functional weights in order to compensate for interference . Moreover , it has been shown that the interruption of specific pathways has effects far beyond the regions that are directly connected by the pathway ( O'Reilly et al . , 2013 ) . Here by contrast , we study the functional enhancement of a pathway , rather than the disruption of a region or pathway , and its effect on coupling within and outside the targeted network . To ensure that the changes in functional connectivity we observed could be attributed to plasticity induction , we performed a control experiment of paired-pulse TMS over the same cortical regions ( Experiment 2 ) stimulating with the same number of pulses at the same frequency but with an IPI which precluded spike timing-dependent plasticity ( STDP ) ( IPI: 500 ms ) . We decided on a 500 ms IPI for the control condition following the exclusion of several other alternative IPIs; we decided against reversing the order of conditioning and test stimulus because we have demonstrated in a previous study that this stimulation order leads to long-term depression-like effects ( as assessed by examining the impact of further PMv TMS pulses on M1 ( Buch et al . , 2011 ) ; against stimulating both areas at the same time because I-wave interactions may occur at such IPIs ( Prabhu et al . , 2009 ) ; against any time interval below 50 ms because there is evidence of plasticity induction at such intervals within the motor system of freely behaving monkeys ( Jackson et al . , 2006 ) . Moreover , we noted that long-interval intracortical inhibition ( LICI ) within M1 has been demonstrated with TMS using IPIs of up to 200 ms ( Valls-Solé et al . , 1992 ) . Admittedly , other intervals in the hundred milliseconds range might equally have been chosen . Targeting the same cortical areas controlled for the impact of stimulation on brain activity in each component node that the pathway interconnects; if changes in connectivity are simply attributable to stimulation of each area , rather than increased pathway efficacy , then changes in functional connectivity ought to be comparable in Experiment 1 and 2 . Here , we show that pathway functional connectivity was not modulated in Experiment 2 . By increasing synaptic efficacy in a corticocortical connection—PMv-M1—involved in complex motor behaviour , we were able to study the relationship of induced plasticity and functional connectivity during the performance of a motor task as well as during the resting state . Further , investigation of a wider motor network provided information about functional reorganisation in response to pathway-specific plasticity induction . In Experiment 3 , we directly tested whether estimates of pathway connectivity based on fMRI data share construct validity with measures of effective connectivity indexed by MEP amplitude ratios across subjects . We demonstrated a direct correlation between the strengths of the two measures for both cognitive states ( i . e . , resting and grasping ) , however , the sign of net effective influence of one neural node over another was only determined by TMS-evoked measures and not by fMRI functional connectivity .
Since increased functional connectivity between two areas is difficult to distinguish from increased mean firing in the two areas , it is possible that the measured connectivity changes observed following repeated paired PMv-M1 TMS at 8 ms IPI in Experiment 1 could have been the result of an increase in activity in each of the stimulated areas instead of a induced change in functional connectivity ( Chawla et al . , 2000 ) . In Experiment 2 , we therefore applied identical numbers of pulses at identical frequencies over the identical brain regions , but we did so at 500 ms IPI . This interval is many times longer than the longest one at which PMv-M1 interactions have been observed ( Davare et al . , 2008; Neubert et al . , 2010 ) . While such a protocol ought to induce similar changes in each stimulated region , it should not result in their co-activation or in STDP . Using a higher-level analysis ( mixed-model ANOVA ) with between-subjects factor ‘PROTOCOL’ , we directly contrasted the effects from Experiment 1 and Experiment 2 for each of the analyses conducted . We present the results in Table 2 and will go through the findings in the following order: ( 1 ) functional connectivity analysis between PMv-M1; ( 2 ) PPI analysis between PMv-M1; ( 3 ) partial correlation analysis between pairwise network nodes; ( 4 ) multiple regression PPI analysis between pairwise network nodes; and ( 5 ) dual-regression analysis . 10 . 7554/eLife . 04585 . 011Table 2 . Summary of results from hypothesis-driven analyses conducted on 8 ms-IPI Experiment 1 and control Experiment 2 ( IPI of 500 ms ) DOI: http://dx . doi . org/10 . 7554/eLife . 04585 . 011Expt 1 ( IPI 8 ms ) Expt 2 ( IPI 500 ms ) Expt 1 vs Expt 2PMv-M1AIP-PMvpSPL-PMdPMd-M1PMv-M1AIP-PMvpSPL-PMdPMd-M1PMv-M1AIP-PMvpSPL-PMdPMd-M1Functional connectivity ( fc ) graspt ( 13 ) = −2 . 59; p = 0 . 023*t ( 13 ) = 0 . 94; p = 0 . 36F ( 1 , 26 ) = 4 . 64; p = 0 . 041*restt ( 14 ) = −0 . 07; p = 0 . 94t ( 14 ) = 0 . 07; p = 0 . 95n . a . partial correlation fcgraspt ( 13 ) = −3 . 72; p = 0 . 003*n . s . n . s . n . s . t ( 13 ) = 1 . 00; p = 0 . 34n . s . n . s . n . s . F ( 1 , 26 ) = 7 . 76; p = 0 . 011*n . s . n . s . n . s . restt ( 14 ) = −0 . 07; p = 0 . 95t ( 14 ) = −2 . 50; p = 0 . 025*t ( 14 ) = 2 . 22; p = 0 . 04*t ( 14 ) = 2 . 84; p = 0 . 013*t ( 14 ) = −0 . 39; p = 0 . 70t ( 14 ) = 1 . 08; p = 0 . 30t ( 14 ) = −1 . 24; p = 0 . 24t ( 14 ) = 0 . 47; p = 0 . 65n . a . F ( 1 , 28 ) = 7 . 15; p = 0 . 012*F ( 1 , 28 ) = 5 . 29; p = 0 . 029*F ( 1 , 28 ) = 5 . 92; p = 0 . 08Psycho-physiological interaction ( PPI ) graspt ( 13 ) = −4 . 78; p = 0 . 0004*t ( 13 ) = 0 . 98; p = 0 . 35F ( 1 , 26 ) = 6 . 92; p = 0 . 014*restt ( 14 ) = 0 . 08; p = 0 . 93t ( 14 ) = 0 . 20; p = 0 . 85n . a . multiple regression PPIgraspt ( 13 ) = −2 . 53; p = 0 . 0064*n . s . n . s . n . s . t ( 13 ) = 1 . 18; p = 0 . 26n . s . n . s . n . s . F ( 1 , 26 ) = 7 . 47; p = 0 . 011*n . s . n . s . n . s . restn . a . t ( 14 ) = −2 . 55; p = 0 . 023*t ( 14 ) = 1 . 78; p = 0 . 097t ( 14 ) = 2 . 84; p = 0 . 013*n . a . t ( 14 ) = 0 . 41; p = 0 . 96t ( 14 ) = −1 . 18; p = 0 . 26t ( 14 ) = 0 . 01; p = 0 . 99n . a . F ( 1 , 28 ) = 5 . 74; p = 0 . 024*F ( 1 , 28 ) = 3 . 66; p = 0 . 066F ( 1 , 28 ) = 4 . 44; p = 0 . 044*Analyses were conducted on rest and task data . Moreover in order to show that specific effects relate to plasticity induction ( 8 ms IPI ) several higher-level analyses contrasting Experiment 1 and 2 are presented . T-tests were conducted as two-tailed paired t-tests ( within subjects ) . Mixed-model ANOVAs were conducted between experiments ( across subjects ) . Detailed information on all analyses is provided in the ‘Materials and methods’ section . Asterisks indicate significant results , p < 0 . 05 . Abbreviations: n . s . = non-significant . A higher-level analysis of functional connectivity between PMv-M1 confirmed that PMv-M1 coupling was not changed during task performance following paired TMS with a 500 ms IPI; this is in contrast to significantly greater connectivity following paired TMS with an 8 ms IPI ( mixed-model ANOVA: TIME by PROTOCOL interaction: F ( 1 , 26 ) = 4 . 64 , p = 0 . 041; Experiment 2 during task: paired t-test: t ( 13 ) = 0 . 94 , p = 0 . 36 ) . At rest , functional connectivity was not changed in the PMv-M1 connection following either protocol ( Experiment 2 at rest: paired t-test: t ( 14 ) = 0 . 07 , p = 0 . 95 ) . A higher-level PPI analysis of PMv-M1 connectivity supports the finding from the functional connectivity analysis ( mixed-model ANOVA: TIME by PROTOCOL interaction during task: F ( 1 , 26 ) = 6 . 92 , p = 0 . 014; Experiment 2 during task: paired t-test: t ( 13 ) = 0 . 98 , p = 0 . 35 ) . At rest , no changes in PMv-M1 connectivity were found either ( Experiment 2 at rest: paired t-test: t ( 14 ) = 0 . 20 , p = 0 . 85 ) . In parallel to the analyses for Experiment 1 , we then examined the wider dorsolateral and dorsomedial sensorimotor circuits . A partial correlation analysis contrasting Experiment 1 with Experiment 2 confirmed that during task , PMv-M1 coupling was only changed in the grasping condition following plasticity induction with an 8 ms IPI ( mixed-model ANOVA: TIME by PROTOCOL interaction: F ( 1 , 26 ) = 7 . 47 , p = 0 . 011; Experiment 2 during task: paired t-test: t ( 13 ) = 1 . 18 , p = 0 . 26 ) . At rest , no changes in coupling in the wider dorsolateral and dorsomedial sensorimotor circuits were found after 500 ms IPI TMS ( Experiment 2 at rest: paired t-tests: AIP-PMv: t ( 14 ) = 1 . 08 , p = 0 . 30; pSPL-PMd: t ( 14 ) = -1 . 24 , p = 0 . 24; PMd-M1: t ( 14 ) = 0 . 47 , p = 0 . 65 ) . Further statistical testing showed that the strengthening of functional connectivity in the AIP-PMv pathway was significantly stronger after 8 ms IPI TMS ( Experiment 1 ) than was after 500 ms IPI TMS ( Experiment 2 ) ( mixed-model ANOVA: TIME by PROTOCOL interaction: F ( 1 , 28 ) = 7 . 15 , p = 0 . 012 ) as were the decreases in the PMd-pSPL pathway strength ( mixed-model ANOVA: TIME by PROTOCOL interaction: F ( 1 , 28 ) = 5 . 29 , p = 0 . 029 ) . The PMd-M1 connection showed a similar trend ( mixed-model ANOVA: TIME by PROTOCOL interaction: F ( 1 , 28 ) = 5 . 92 , p = 0 . 08 ) . The lack of reorganisation within the dorsolateral and dorsomedial circuits in Experiment 2 was confirmed by employing the same multiple regression PPI analysis used for Experiment 1 . Table 2 . Finally , employing a dual-regression analysis , we confirmed that it was only after 8 ms IPI TMS in Experiment 1 that the left frontoparietal network was found to be more coherently coupled with itself and co-active with PMv in the resting state , but not after 500 ms IPI TMS in Experiment 2 ( mixed-model ANOVA: TIME by PROTOCOL interaction: p = 0 . 022 ) . The fronto-parietal network did not significantly alter its coupling pattern in Experiment 2 ( p = 0 . 446 ) . In Experiment 3 , we investigated how both TMS-based effective connectivity and fMRI-based functional connectivity indices relate to each other within the same subjects , focussing on the PMv-M1 pathway in ten of the subjects tested in Experiment 1 . We measured the size of MEPs evoked by TMS of M1 alone and evoked by M1 TMS applied 8 ms after a PMv pulse . Such PMv TMS pulses are known to either augment or diminish the size of the MEP induced by M1 TMS depending on whether or not subjects are making grasping movements or are at rest , respectively ( Davare et al . , 2008; Buch et al . , 2010 , 2011 ) . To quantify the influence of PMv over M1 , we compared the MEPs induced by M1 stimulation alone with MEPs induced by M1 stimulation that was preceded by PMv-stimulation . A TMS-based index of effective connectivity between PMv and M1 was calculated as the ratio of the difference in MEP amplitudes evoked by paired-pulse TMS and single-pulse TMS divided by single-pulse-evoked MEP amplitudes . The PMv-M1 TMS ratio is positive when PMv TMS augments the size of M1 TMS-induced MEPs , but negative when PMv TMS diminishes M1 TMS MEP size . Moreover , the ratio was measured both while subjects were at rest and during the reaching task . In this way , the influence of PMv over M1 could be quantified for both cognitive states , and the direction ( facilitatory or inhibitory ) and magnitude of effective connectivity could then be compared to functional connectivity as measured with fMRI in Experiment 1 . During task performance , the two connectivity measures ( a TMS-based effective connectivity index and the fMRI-based functional connectivity measure ) for the PMv-M1 pathway were positively correlated across subjects at baseline ( Pearson's correlation coefficient: R = 0 . 74 , p = 0 . 01 ) ( Figure 5A ) . A positive correlation indicates that the greater the facilitatory influence of PMv on M1 as measured with TMS , the greater the fMRI-derived functional connectivity . Furthermore , during task performance , paired-pulse TMS-derived effective connectivity was still significantly correlated with fMRI-derived connectivity after repeated paired 8 ms IPI TMS ( R = 0 . 87 , p = 0 . 0008 ) ( Figure 5B ) . 10 . 7554/eLife . 04585 . 012Figure 5 . Experiment 3: correlation of PMv-M1 connectivity measures before and after 8 ms IPI paired TMS ( N = 10 ) . When subjects were making grasping movements , there was a significant correlation between functional connectivity ( derived from partial-correlation analysis of fMRI ) in the baseline ( A ) and post-TMS session ( B ) and the baseline effective connectivity measure derived from the paired pulse TMS MEP ratio at baseline . There was a significant negative correlation between functional connectivity in the post-TMS session and the baseline effective connectivity measure derived from the paired-pulse TMS MEP ratio at baseline ( D ) . The correlation did not reach significance when the functional connectivity measure as well as the effective connectivity measure was taken from the baseline session ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04585 . 012 At rest , the correlation between the two measures was not significant at baseline ( R = −0 . 21 , p = 0 . 57 ) ( Figure 5C ) . However , a significant correlation was observed between the baseline TMS-derived effective connectivity measure and the fMRI-derived functional connectivity measure following repeated paired 8 ms IPI TMS ( R = −0 . 68 , p = 0 . 03 ) ( Figure 5D ) . Intriguingly , the correlation was negative which implies that when PMv had a stronger net inhibitory influence on M1 ( as indexed by neurophysiological measurements ) , fMRI indicated stronger net positive functional connectivity between the two areas across participants . Finally , we note that baseline effective connectivity strength at rest ( i . e . , inhibitory ) and during grasping ( i . e . , excitatory ) were also negatively correlated across individuals ( R = −0 . 63 , p = 0 . 049 ) .
In this study , we describe the functional connectivity signature in fMRI data of short-term synaptic potentiation within a specific anatomical pathway . Using two different paired-pulse TMS manipulations , we demonstrated that application of a TMS protocol known to change synaptic efficacy within a motor pathway ( PMv-M1 ) results in increases in functional connectivity along the same pathway that can be measured with fMRI . Furthermore , we established a significant correlation between the size of the causal influence of PMv on M1 ( effective connectivity as assessed by paired-pulse TMS ) and the fMRI-based index of functional connectivity across individuals . PMv provides one of the principal inputs into M1 , and it exerts a powerful influence over M1 output ( Shimazu et al . , 2004; Dum and Strick , 2005 ) but the degree of interaction between PMv and M1 can be modulated by repeated paired pulse TMS with an 8 ms IPI over PMv and M1 ( Buch et al . , 2011 ) . By means of two correlation analyses and two PPI analyses on fMRI data acquired during the performance of a grasping task , we confirmed our a priori hypothesis that augmented pathway efficacy is mirrored in increases in inter-regional functional connectivity . We specifically show that the activity in PMv was more tightly related to activity in M1 following intervention ( Figure 2C ) and that the influence of PMv on M1 ( or the responsiveness of M1 to input from PMv ) was increased in response to short-term pathway manipulation ( Figure 2C and Figure 2—figure supplement 1A ) . In the current study , we did not track the duration of these changes in functional coupling after the intervention . However , we note that in a previous study changes in effective connectivity were shown to last more than 1 hr ( Buch et al . , 2011 ) . An alternative interpretation of the increased correlation in PMv and M1 activity might be attempted not by referring to synaptic change involving the pathway between them but by simply referring to the changes that pulses over each area induce even when applied in isolation . Such an explanation , however , is unlikely to be correct . First , there is no empirical reason to think that TMS stimulation of any one area at a frequency of 0 . 1 Hz would lead to a protracted change in that area's activity which is detectable by fMRI many minutes later . Moreover , Experiment 2 employed a control procedure , a repeated paired-pulse TMS protocol which precludes the temporal contiguity required for pathway plasticity induction by using a longer IPI ( 500 ms vs 8 ms in Experiment 1 ) even though it involved stimulation of the same areas at the same frequency and intensity . Functional connectivity between PMv and M1 was not altered in response to the control procedure in Experiment 2 suggesting that the influence of PMv on M1 in Experiment 1 was indeed attributable to short-term changes in synaptic efficacy . These observations extend previous studies that described acute compensatory plasticity of the motor system following single-site TMS manipulations albeit in the context of task performance ( Lee et al . , 2003; O'Shea et al . , 2007; Grefkes et al . , 2010; Hartwigsen et al . , 2012 ) . The results also extend the understanding of the effect of the repeated paired pulse TMS plasticity induction procedure that we previously examined in the absence of fMRI data ( Buch et al . , 2011 ) . For example , greater M1 output was previously observed by measuring MEP sizes with M1 TMS during grasping after the paired pulse plasticity induction procedure but the origin of the effect was unclear . The new results make it clear that it is driven by M1 being more responsive to activity in PMv . Moreover , the current results reveal that the increase in PMv-M1 connectivity was very specific . Although several analyses demonstrated that functional connectivity also increased between PMv and its principal parietal input in the dorsolateral sensorimotor pathway , AIP and adjacent parts of the parietal cortex ( Godschalk et al . , 1984; Davare et al . , 2010 ) ( Figures 3A , D , E , 4B , Figure 4—figure supplement 1B ) , increased functional connectivity was not seen between M1 and other premotor areas . In fact the reverse was true; functional connectivity between PMd and M1 significantly decreased . In addition , functional connectivity in other parts of the dorsomedial sensorimotor circuit , between PMd and pSPL , also declined ( Figure 3D , E ) . The exact functional role of these accessory decreases in functional coupling in distant connections requires further investigation . It is unclear whether they should be thought of as ‘compensatory’ as they were more prominent at rest than during the grasping task . Inhibitory plasticity might accompany excitatory plasticity in order to stabilise neural networks involved in learning ( Vogels et al . , 2011 ) . They suggest that inhibitory spike timing-dependent plasticity could balance excitatory inputs . Learning or the formation of associative excitatory connections in such networks would require the co-adaptation of excitatory and inhibitory synapses . Although Vogels' et al . ideas largely make predictions about structural and functional properties of local neural circuits , the results of this experiment could be taken to suggest that similar principles apply to the network and systems level . Additionally enhancement of one pathway might be accompanied by diminution of a parallel pathway if both of them compete to influence a particular target structure such as M1 . It has been argued that two pathways for movement preparation—the dorsomedial visuomotor stream ( pSPL–PMd ) and the dorsolateral stream ( AIP–PMv ) —complement each other by driving movement selection proportional to the amount of information available in each stream ( Verhagen et al . , 2008 ) . It remains to be determined how exactly movement selection is biased towards dorsolateral or dorsomedial streams and whether there are categorical or graded differences . The study of multi-sensory integration has generated proposals concerning how integration of information from two different streams might be achieved ( Ernst and Banks , 2002 ) . Future research needs to understand the relation of these different pathways and how they interact and potentially compete to guide movement selection . With more detailed knowledge about the structural skeleton and the functional relationship of these streams , we might be able to predict the complex effects of learning and plasticity not only on the particular network primarily involved in learning and plasticity but also on other parallel streams and networks . More generally this might eventually contribute to a better understanding of network effects relating to learning , development and degeneration ( Fair et al . , 2008; Seeley et al . , 2009; Dayan and Cohen , 2011 ) . For this line of research whole-brain approaches such as fMRI or magnetoencephalography ( MEG ) might have some advantages in some contexts in comparison to examining more local effects of plasticity , such as changes in MEPs ( Buch et al . , 2011 ) . The absence of reduced dorsomedial pathway coupling in Experiment 2 suggests that it cannot simply be due to the repeated asynchrony of activity in PMd and M1 that is induced by the paired TMS protocol ( repeated TMS-induced activation of M1 without corresponding activation of PMd ) . If this were the case , then one would expect the protocol in Experiment 2 , in which PMd and M1 activity was also stimulated asynchronously , to induce similar decrements in dorsomedial pathway coupling , but this was not observed in the current study . Some of the changes in functional connectivity seen in Experiment 1 were more apparent when subjects were engaged in the motor task while others were more apparent when subjects were at rest . Broadly speaking , changes in the interactions between the stimulated areas themselves , PMv and M1 , areas known to be intimately involved in the making of grasping movements , were most apparent when subjects were actively engaged in just such motor activity ( Figure 2C , Figure 4A , Figure 2—figure supplement 1A , Figure 4—figure supplement 1A ) . By contrast , changes in interactions between PMv and adjacent ventral frontal areas such as IFJ with other prefrontal and parietal areas concerned with high-level cognitive control and attention were more apparent when subjects were at rest ( Figure 4B ) . Changes in interactions between PMv and the parietal areas surrounding AIP were apparent both when subjects were engaged in grasping and when they were at rest ( Figure 4A , B ) . The differential sensitivity of the two conditions to different aspects of functional connectivity change may be related to the varying roles of these pathways in the control , planning , and coordination of movement and the actual implementation and execution of movements . Finally , Experiment 3 demonstrated that similar patterns of fMRI-measured functional connectivity are associated with either net facilitatory or net inhibitory influences being exerted by PMv over M1 when subjects are engaged in a grasping task or at rest ( Figure 5 ) . Despite these differences in the sign of the relationship between separate indices of connectivity , the sizes of functional connectivity indices were correlated across subjects , that is , subjects with stronger functional connectivity between PMv and M1 also showed higher degrees of TMS-measured effective connectivity between the two areas . The paired pulse TMS approach provides a particularly direct and simple assay of the effective connectivity that exists between two brain areas but it complements other techniques , such as DCM , that attempt to recover effective connectivity estimates from fMRI data in particular behavioural contexts ( Friston , 1994; Friston et al . , 2003 ) . From our experiments , we infer that functional connectivity is not only shaped by structural connections but also by short-term plastic changes in synaptic efficacy . It still , however , remains a challenge to link the changes seen with neuroimaging measures to specific cellular and molecular level changes at the synapse . Paired stimulation of two brain regions led to increased functional connectivity between the two regions but also to a limited set of other functional connectivity changes , both positive and negative , in other parts of the cortical sensorimotor circuits . In addition , we showed that positive functional connectivity between two areas may reflect either facilitatory or inhibitory effective connectivity . Such changes in functional connectivity are not only interesting in their own right but also because different patterns of premotor–M1 interaction are seen in patients who do and who do not recover motor skills after stroke ( Gerloff et al . , 2006; Lotze et al . , 2006 ) . An interesting possible future avenue for research is to employ pathway-specific non-invasive stimulation protocols in patients to induce directed changes in connectivity and thereby potentially drive neural network reorganisation so as to assist in recovery of motor function .
15 subjects ( eight males ) participated in Experiment 1; fifteen subjects ( nine males ) participated in Experiment 2 . For Experiment 3 , paired-pulse TMS data were obtained for 10 participants from Experiment 1 ( five males ) . The overall mean age of all participants was 24 ± 4 years ( mean ± SD ) . The study was approved by the local ethics committee and informed consent was obtained from all subjects . TMS was applied using two Magstim 200 stimulators each of which was connected to a 50 mm figure-8 coil . On a day prior to the day of the combined TMS-fMRI experiment , resting motor threshold ( RMT ) was determined for each participant for the left M1 ‘hot spot’ , which is the scalp location where TMS evoked the largest MEP amplitude in right first dorsal interosseous ( FDI ) ( Rossini et al . , 1994 ) ( mean ± SD: 40 ± 7% stimulator output ) . Electromyographic ( EMG ) activity in right FDI was recorded with bipolar surface Ag-AgCl electrode montages . Responses were bandpass filtered between 10 and 1000 Hz , with additional 50 Hz notch filtering , sampled at 5000 Hz , and recorded using a CED 1902 amplifier , a CEDmicro1401 Mk . II A/D converter , and PC running Spike2 ( Cambridge Electronic Design ) . To stimulate left M1 , one coil was placed over the scalp location of the left FDI ‘hot spot’ at average MNI coordinates [−40 −18 59] . The location was projected onto the high-resolution , T1-weighted MRI brain scan of each participant using frameless stereotactic neuronavigation ( Brainsight; Rogue Research ) . The second coil , over left PMv , was positioned so as to be ventral to the convergence of the inferior frontal sulcus and inferior precentral sulcus on each individual's MRI scan . The mean MNI location [−56 19 19] was within the region defined previously as human PMv ( Mayka et al . , 2006 ) but which more precisely corresponds to the border between IFJ and 6v ( Neubert et al . , 2014 ) ( Figure 1B ) . As in previous studies PMv was stimulated with 110% of RMT and M1 with a stimulation intensity sufficient to elicit a 1-mV MEP following a single TMS pulse ( Neubert et al . , 2010; Buch et al . , 2010 and 2011 ) . For the duration of the experiment , TMS coils were fixed in place tangentially to the skull by means of adjustable metal arms and monitored throughout the experiment . In all three experiments , an attempt was made to induce plasticity between PMv and M1 by repeated paired stimulation of the two areas . Paired TMS lasted for 15 min and was applied at a frequency of 0 . 1 Hz ( i . e . , 90 pairs of pulses ) , with an IPI of 8 ms ( Experiments 1 and 3 ) and an IPI of 500 ms ( Experiment 2 ) . In Experiment 3 , 8 ms IPI paired pulses were also used in a second way in order to provide a neurophysiological index of effective connectivity between PMv and M1 . Ten participants drawn from Experiment 1 took part on a day separate from the two MRI image acquisition days . MEPs were recorded from the right FDI muscle in response to either M1 TMS ( 20 trials ) or paired-pulse TMS delivered over both PMv and , 8 ms later , over M1 ( 20 trials ) . Trials were administered in pseudorandom order . The ratio of MEP sizes in the paired pulse trials compared to the single M1 pulse trials provided an index of the modulatory influence of PMv over M1 . Experiment 3 was conducted under two conditions . In the grasping condition , volunteers sat in a darkened room and made right-hand reaching and grasping movements cued by illumination of one of two concentrically arranged cylinders ( 15 and 65 mm diameter ) located 30 cm in front of the starting hand position ( Buch et al . , 2010 and 2011 ) . Each trial was initiated by pressing a touch bar with the right hand . Intertrial intervals were therefore variable ( mean ± SD , 6 . 90 ± 0 . 79 s ) but did not differ significantly across phases of each experiment . Following a variable delay of 5–7 s ( uniformly distributed ) , one cylinder was illuminated . Volunteers responded by grasping it with their thumb and index finger before lifting it from its pedestal . Reaction and movement times were recorded . All trials were accompanied by either M1 TMS or paired PMv-M1 TMS , with the pulse applied to M1 always occurring 100 ms after cylinder illumination , which was before movement onset . For TMS when at rest , volunteers still attended to cylinder illumination , as they had done during the motor task , but now they simply maintained a static hand posture . To control for the overall temporal distribution of the TMS pulses , ITIs for rest blocks were defined as the sum of the ITIs used in task blocks plus a reaction and movement time sample drawn from probability density functions for these variables ( Buch et al . , 2010 and 2011 ) . ITIs were therefore variable ( mean ± SD , 6 . 23 ± 0 . 07 s ) but did not differ significantly across phases of each experiment . MRI data were acquired on a Siemens 3T Trio MRI scanner at the Oxford Centre for Clinical Magnetic Resonance Imaging ( OCMR ) . For purposes of neuronavigation-guided TMS , all volunteers underwent high-resolution , T1-weighted structural MRI scans that included nose and ears . For each condition—resting state and grasping task—5 min of whole-brain T2*-weighted gradient echo planar images ( EPIs ) sensitive to BOLD were acquired ( repetition time = 3 . 000 ms , echo time = 30 ms , flip angle = 87° , isotropic voxels of 3 . 0 mm , no slice gap , 45 slices in axial direction ) . Participants were instructed to keep their eyes closed during resting-state fMRI . During the grasping task , which was based on a previous study ( Grol et al . , 2007; Majdandzić et al . , 2007 ) , participants performed 66 reaching-and-grasping trials towards either a small or a large cube positioned in front of them . A new trial sequence was generated for every participant and for each session , with an inter-trial interval of 4295 . 5 ms–4795 . 5 ms ( mean ± SD: 4545 . 5 ms ± 145 . 5 ms ) which allowed every participant to complete the movement . Participants lay supine in the MR scanner with the eight-channel head coil tilted forward by 30° enabling them to perform a naturalistic visually guided reaching-and-grasping task in front of their bodies ( Figure 1C ) . Participants were allowed to move their eyes in order to guide their movements . An optical response button box was placed on their right upper leg and served as a start-and-finish position . Reaction times and total movement times were recorded . With the aim of avoiding movement artefacts , the participant's upper arm lay on a wedge-shaped polyfoam cushion and was firmly , but comfortably strapped to the side of the participant's chest . This setup constrained rotation movements in the plane between the button box and the target objects . The head was supported with foam wedges . The participants had received extensive training in the reaching-and-grasping task at least one day prior to the first MRI acquisition outside the MRI scanner . The target object , which consisted of a large red cube and a small green cube ( Figure 1C , inset ) , was held in place through an arc-shaped device positioned over each participant's hips . Participants had been instructed to grasp one of the two cubes , to slide it out of its supporting rail on a rectangular box , and to return it into the same supporting rail . On a given trial , either the large red or the small green cube was to be grasped . A red or green light-emitting diode ( LED ) in the middle of the rectangular box instructed the participant which cube to grasp . MRI-compatible switches on the device recorded the time at which the object was removed from the supporting rail and the time at which the object was returned into the supporting rail . Control of LEDs and recording of movement-related responses was performed with a computer running Presentation 15 . 0 ( Neurobehavioral Systems , San Francisco , CA ) . TMS was applied outside the MRI scanner room . Participants walked to the MRI scanner and scanning commenced within 3 to 4 min . Note , previous neurophysiological experiments ( Buch et al . , 2011 ) suggest plasticity induction should last at least 1 hr with this protocol and that there were no differences in efficacy immediately after intervention in comparison to +30 min or +60 min post-intervention . FMRI data were pre-processed using tools from the FMRIB Software Library ( FSL; www . fmrib . ox . ac . uk/fsl; Smith et al . , 2004 ) . Imaging volumes were registered to the individuals' structural scan using boundary-based registration ( BBR ) ( Greve and Fischl , 2009 ) and to standard space using FMRIBs Linear Image Registration Tool ( FLIRT ) with 12° of freedom . Pre-processing involved: motion correction ( McFLIRT ) , brain extraction ( BET ) , spatial smoothing with a Gaussian 5 mm full-width at half-maximum ( FWHM ) kernel , and high-pass temporal filtering at 100 s . Individual subject independent-component analysis ( ICA ) fMRI analysis was carried out on baseline data of twelve Experiment 1 and eleven Experiment 2 data sets using Multivariate Exploratory Linear Optimized Decomposition into Independent Components ( MELODIC ) ( Beckmann and Smith , 2005 ) . Individual pre-statistical processing consisted of motion correction ( McFLIRT ) , brain extraction ( BET ) , spatial smoothing using a Gaussian kernel of full-width at half maximum ( FWHM ) of 5 mm , and high-pass temporal filtering . Imaging volumes were registered to the individuals' structural scan using boundary-based registration ( BBR ) ( Greve and Fischl , 2009 ) and to standard space using FMRIBs Linear Image Registration Tool ( FLIRT ) with 12° of freedom . Pre-processed functional data were temporally concatenated across subjects . SBCA maps the functional connectivity of one ‘seed’ ROI across the entire brain in a voxel-wise manner on the basis of the correlation between the seed ROI's BOLD time series and the BOLD time series at each voxel in the rest of the brain ( O'Reilly et al . , 2010 ) . We employed SBCA to assess if paTMS-based modulation of the PMv-M1 pathway dynamically altered the functional interactions of either of these two nodes with each other and/or with other nodes within the reaching and grasping network . We assessed the functional connectivity of a 6 mm diameter seed mask in left PMv with the whole brain ( target mask ) before and immediately after paTMS and contrasted PMv-M1 connectivity at baseline vs connectivity during post-TMS ( for details about statistical analyses see below ) . The analyses of resting state and grasping task fMRI data were conducted independently . All analyses conducted for Experiment 1 and Experiment 2 were identical , which allowed us to directly contrast the effects in a higher-level analysis . For the first step of SBCA , statistical connectivity maps for every individual and for each of the four conditions ( resting-state baseline/resting-state plasticity expression and task baseline/task plasticity expression ) were created using the SBCA tool implemented in FSL ( fsl sbca ) . The time series for the left PMv seed mask was calculated . The SBCA model also accounted for the time series resulting from structured noise in the average BOLD signal in white matter ( WM ) , grey matter ( GM ) , and cerebrospinal fluid ( CSF ) and head movement ( six regressors resulting from McFLIRT motion correction ) . WM , GM , and CSF masks were derived from individual T1-weighted structural images using the FSL segmentation tool FAST and registered to EPI space using FLIRT ( Jenkinson and Smith , 2001; Zhang et al . , 2001 ) . The resulting connectivity map described the correlation between the average BOLD time series of the PMv mask and the time series for each voxel within the whole brain . The individual correlation maps were transformed into MNI space , using FLIRT affine registration of EPI to structural space and subsequently FNIRT non-linear registration to MNI space . Standard space group correlation maps ( z-score maps ) were generated by entering SBCA-derived individual correlation maps into a group general linear model ( GLM ) and thresholding at Z > 2 . 3 with a significance threshold of p < 0 . 05 . These thresholded group z-score maps were projected onto the Midthickness . 32k CaretBrain as provided by the Human Connectome Project Workbench using the ‘surf proj’ algorithm as implemented in FSL and then visualized using the Human Connectome Project Workbench ( http://www . humanconnectome . org/connectome/get-connectome-workbench . html ) . As the next step , we computed the average time series resulting from these statistical connectivity maps for the M1 ROI . We then compared time series correlations of PMv with M1 at baseline and during post-TMS . For statistical comparisons , we conducted a paired t-test , contrasting PMv-M1 interactions at baseline vs during post-TMS . Prior to statistical analysis , correlation coefficients were Fisher z-transformed . We also conducted a higher-level analysis , contrasting Experiment 1 with Experiment 2 , using a mixed-model ANOVA with within-subject factors TIME ( baseline / post-TMS ) and between-subjects factor PROTOCOL ( Experiment 1/Experiment 2 ) . We used a significance level of p < 0 . 05 . 15 participants contributed to the resting-state group z-score map ( for both Experiment 1 and Experiment 2 ) ; 14 participants contributed to the grasping task group z-score map , since the data of one participant had to be removed due to excessive head movement during data acquisition ( for both Experiment 1 and Experiment 2 ) . To investigate changes in functional connectivity between pairs of grasping network nodes , we conducted a partial correlation analysis between the BOLD time series of directly connected nodes of the left hemisphere using Matlab R2013b ( MathWorks ) . Partial correlation analysis generated correlations represent only correlations specific to the pair of cortical regions in question by regressing out the time series of all other network nodes under investigation . We focussed on pairs of regions thought to be monosynaptically connected ( Matelli et al . , 1986; Johnson et al . , 1997; Wise et al . , 1997; Matelli et al . , 1998; Luppino et al . , 1999; Geyer et al . , 2000; Tanne-Gariepy et al . , 2002; Dum and Strick , 2005; Rushworth et al . , 2006; Grol et al . , 2007; Tomassini et al . , 2007; Mars et al . , 2011; Sallet et al . , 2013; Neubert et al . , 2014 ) : M1-PMv , PMv-AIP , AIP-V3A , M1-PMd , PMd-pSPL , and pSPL-V3A . Individual BOLD time series for each network node mask ( 6 mm diameter ) were generated using a GLM-based design that incorporated regressors denoting potentially confounding factors such as variation in WM , GM , and CSF , and whole brain BOLD signal as implemented in FSL ( fsl glm ) . Individual partial correlations were normalised using Fisher's z-transform . Analogous to statistical tests used in SBCA , we conducted paired t-tests contrasting pairwise interactions at baseline vs during post-TMS on Fisher z-transformed partial correlation coefficients ( independently for Experiment 1 and Experiment 2 ) . At a later stage , we also subjected Experiment 1 and Experiment 2 to a direct comparison by means of a mixed-model ANOVA with factors PROTOCOL ( Experiment 1/Experiment 2; between-subjects factor ) and TIME ( baseline/post-TMS; within-subjects factor ) . We used a significance level of p < 0 . 05 . Resting state and grasping task MRI data sets were analysed in an identical way , but were not compared directly due to a categorical difference in movement artefacts ( movement artefacts were larger in the grasp task than in the resting-state MRI ) . The severity of movement artefacts required the removal of one grasping task data set for both Experiment 1 and Experiment 2 . Psychophysiological interaction ( PPI ) analysis refers to the interaction between physiological activity and experimental context and thereby identifies brain areas ( specifically , voxels ) in which activity is more related to activity in a seed region of interest in a given experimental context . To test whether there is a change in the influence PMv ( seed region ) has on M1 , the analysis tested for differences in the regression slope of activity in M1 on the activity in the seed region ( PMv ) under the experimental contexts of ‘baseline’ and ‘post-TMS’ . The change in influence of PMv on M1 can also be understood as a change in responsiveness of M1 to input from PMv . PPI analysis requires an a priori hypothesis about directionality; from physiological models it is well established that PMv provides a major input into M1 ( Dum and Strick , 2005 ) . Directionality of the predominant information flow from PMv to M1 was also supported by a feed-forward model validated on fMRI data acquired during performance of a grasping task ( Grol et al . , 2007 ) and paired-pulse TMS studies ( Davare et al . , 2008; Buch et al . , 2010 ) . To test the hypothesis that repeated paired-pulse TMS stimulation of PMv and M1 altered the responsiveness of M1 to activity in PMv , we conducted a regression analysis between BOLD time series of the network nodes using Matlab R2013b ( MathWorks ) . Individual BOLD time series for each of the two network node masks ( 6 mm diameter ) were generated using a GLM-based design that incorporated regressors denoting potentially confounding factors such as variation in white matter ( WM ) , grey matter ( GM ) , and cerebrospinal fluid ( CSF ) , and whole brain BOLD signal as implemented in FSL ( fsl glm ) . Time series were demeaned and variance-normalised . Analogous to statistical tests used in SBCA , we conducted paired t-tests contrasting pairwise interactions at baseline vs during post-TMS on regression coefficients ( independently for Experiment 1 and Experiment 2 ) . At a later stage , we also subjected Experiment 1 and Experiment 2 to a direct comparison by means of a mixed-model ANOVA with factors PROTOCOL ( Experiment 1/Experiment 2; between-subjects factor ) and TIME ( baseline/post-TMS; within-subjects factor ) . We used a significance level of p < 0 . 05 . Resting-state and grasping task MRI data sets were analysed in an identical way but were not compared directly due to a categorical difference in movement artefacts ( movement artefacts were larger in the grasp task than in the resting-state MRI ) . The severity of movement artefacts required the removal of one grasping task data set for both the Experiment 1 and Experiment 2 condition . In analogy to a partial correlation analysis , we conducted a multiple linear regression analysis on the reaching-and-grasping network nodes ( V3A , pSPL , PMd , AIP , PMv , and M1; for MNI coordinates see above in ‘Regions of interest [ROI]’ ) to understand the influence of one network node upon a specific other network node in terms of the interaction of activity in the remaining network nodes and the experimental context . Time series from the six seed masks ( 6 mm diameter ) were generated as described above in ‘Psychophysiological interaction ( PPI ) analysis’ . To analyse the influence of a given brain area upon another , the time series of all other brain areas of interest are entered as a regressor into the multiple linear regression analysis . Statistical tests on regression coefficients were conducted as described in ‘Psychophysiological interaction ( PPI ) analysis’ . To understand if co-activation patterns in large-scale networks of functional connectivity change dynamically in response to plasticity induction , we investigated networks defined by their shared spontaneous low-frequency fluctuations ( <0 . 1 Hz ) . Coherence within resting-state networks ( RSNs ) ( Friston , 1994 ) and networks during task performance ( Hampson et al . , 2002 ) were analysed before and after paired pulse TMS intervention using a whole-brain corrected approach . Whereas SBCA and partial correlation analyses focused on nodes of the fronto-parietal grasping-network , this approach has the potential to identify any networks ( defined as areas sharing BOLD signal temporal correlations ) in which connectivity is changing as a result of the TMS intervention . This procedure was carried out completely separately for resting-state fMRI and fMRI during task performance . The approach proceeds in three stages . To begin , concatenated multiple fMRI data sets are decomposed using ICA to identify large-scale spatial patterns of functional connectivity . We used the baseline fMRI data sets of all 23 participants who participated in this study and obtained group-averaged ICA-network maps . For seven participants who contributed to both experimental conditions , only one baseline data set was randomly selected to generate the ‘group-averaged baseline’ network masks; specifically , 12 ‘baseline’ data sets were drawn from Experiment 1 and 11 ‘baseline’ data sets were drawn from Experiment 2 . By identifying ICA components based on data from both experiments , we avoided biasing our analysis as a result of any possible differences in the two groups of subjects . At the second stage , two regressions are carried out in which the ICA-derived components are regressed back against the BOLD time series from the baseline and post-plasticity induction periods in the two experiments ( 8 ms IPI and 500 ms IPI ) : firstly , to identify subject-specific temporal dynamics for each group-averaged ICA spatial component via a linear model fit ( spatial regression ) and , secondly , to compute subject-specific associated spatial maps , by using the generated time series as a regressor against the associated fMRI data ( temporal regression ) . At the final stage , the resulting individual spatial component maps are collected across subjects into single 4D files ( 1 per original ICA network map , with the fourth dimension being subject to identification ) . The resulting maps—baseline and post-TMS—with one for each of the two experimental conditions , that is , Experiment 1 and Experiment 2—were then tested for voxel-wise statistical significance against the ICA maps generated from all 23 participants ( ‘group-averaged baseline’ fMRI data sets ) using nonparametric permutation testing ( 5000 permutations ) ( Nichols and Holmes , 2002 ) and cluster-based thresholding and normalisation of the design matrix columns to unit standard deviation . Voxel-wise testing excluded the cerebellum . The only difference in how resting-state fMRI and grasping task fMRI were treated lay in the dimensionality estimation of ICA . The number of components was estimated automatically for both resting-state fMRI and task-positive fMRI using the Laplace approximation to the Bayesian evidence for a probabilistic principal component model ( Beckmann and Smith , 2005 ) , which resulted in 22 and 15 independent components for resting-state and task-positive fMRI , respectively . | When a person has their brain scanned , the resulting images show that regions with similar roles tend to be active at the same time . These coordinated patterns of activity are often altered in the brains of patients with neurological or psychiatric disorders . However , relatively little is known about how the patterns are generated . The degree to which brain regions are active at the same time is thought to depend partly on how well they are connected by brain cells . However , it is also possible that the coordinated activity reflects the extent to which one brain region is able to influence the activity of another . More than 50 years ago , it was demonstrated that this is the case between individual brain cells . If one brain cell repeatedly helps to activate another , the connection between the two cells will be strengthened . This process—known as synaptic plasticity—is thought to support learning and memory . Now , Johnen , Neubert et al . have shown that the same process can also act between different brain regions . A technique called transcranial magnetic stimulation—in which magnetic fields are applied to specific areas of the scalp to excite brain tissue—was used on human volunteers to activate two regions involved in producing grasping movements with their hands . If the first region of the brain was repeatedly activated a few milliseconds before the second region as the volunteers reached towards objects , the ability of the first region to activate the second increased . Notably , the effect was not seen when the interval between the activation of the regions was increased to 500 milliseconds: a delay long enough to ensure that brain cells in the first region were no longer active when the second region was stimulated . This suggests that coordinated changes in the activity of brain regions might reflect the same plasticity processes as changes in activity seen between individual brain cells . This finding raises the possibility that , by deliberately altering the degree of coordinated activity between specific brain regions , it might be possible to recover abilities that have been lost as a result of disorders such as stroke . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Causal manipulation of functional connectivity in a specific neural pathway during behaviour and at rest |
Although alternative pre-mRNA splicing ( AS ) significantly diversifies the neuronal proteome , the extent of AS is still unknown due in part to the large number of diverse cell types in the brain . To address this complexity issue , we used an annotation-free computational method to analyze and compare the AS profiles between small specific groups of Drosophila circadian neurons . The method , the Junction Usage Model ( JUM ) , allows the comprehensive profiling of both known and novel AS events from specific RNA-seq libraries . The results show that many diverse and novel pre-mRNA isoforms are preferentially expressed in one class of clock neuron and also absent from the more standard Drosophila head RNA preparation . These AS events are enriched in potassium channels important for neuronal firing , and there are also cycling isoforms with no detectable underlying transcriptional oscillations . The results suggest massive AS regulation in the brain that is also likely important for circadian regulation .
Organisms ranging from cyanobacteria to mammals contain circadian clocks that synchronize physiology and behavior to time-of-day environmental cues , such as light and temperature . The fruit fly , Drosophila , is no exception and exhibits robust circadian clocks that control a diverse range of behaviors such as activity , sleep , mating and feeding . The core molecular machinery of the circadian clock is a set of transcription factors that act together to drive cycling or time-of-day dependent transcription of at least 20% of genes in the genome ( Abruzzi et al . , 2011 ) . The heterodimeric transcription factor CLOCK/CYCLE ( CLK/CYC ) binds to the promoters of many direct target genes and activates their transcription late in the day . Two of these direct target genes , Timeless ( Tim ) and Period ( Per ) , encode transcriptional repressors that re-enter the nucleus in the late night and act as part of a negative feedback loop to turn off CLK/CYC transcription ( Edery et al . , 1994; Naidoo et al . , 1999; Sehgal et al . , 1994 ) . In the early morning , the photoreceptor , Cryptochrome ( CRY ) and the kinase , Doubletime ( Dbt ) contribute to the degradation of Tim/Per ( Kim et al . , 2007; Kloss et al . , 1998; Emery et al . , 1998; Stanewsky et al . , 1998; Ceriani et al . , 1999; Hunter-Ensor et al . , 1996; Syed et al . , 2011 ) . Without these repressors , CLK/CYC-mediated transcription begins again restarting the daily cycle of transcription . In the Drosophila brain , this molecular clock resides in a neural network of ~150 neurons . These neurons are divided into seven subgroups including three groups of dorsal neurons ( DNs; DN1 , DN2 , and DN3 ) , three groups of lateral neurons ( ventral and dorsal lateral neurons; large and small LNvs and LNds , respectively ) and the lateral posterior neurons ( LPNs ) with discrete behavior functions; reviewed in ( Peschel and Helfrich-Förster , 2011 ) . These neurons make up a small portion of the Drosophila brain ( ~0 . 1% ) and head ( ~0 . 05% ) . As a result , it is difficult to profile these specialized neurons as part of studies that focus on Drosophila brain and head tissues ( Hughes et al . , 2012; Rodriguez et al . , 2013 ) . To learn more about the function of these neurons within the circadian neuronal circuit , three subgroups ( DN1s , LNds , and LNvs ) of these neurons as well as one non-circadian outgroup ( Dopaminergic or tyrosine-hydroxylase ( TH ) expressing neurons ) were labeled with GFP using neuron-specific drivers and manually isolated ( Abruzzi et al . , 2015; Abruzzi et al . , 2017 ) . Illumina RNA-seq datasets from these neurons were analyzed to examine the global mRNA landscape at the transcription level . This analysis revealed many cycling transcripts whose abundance changes with time of day that were not identified in previous studies of brain or head tissues ( McDonald and Rosbash , 2001; Claridge-Chang et al . , 2001; Wijnen et al . , 2006; Rodriguez et al . , 2013 ) . In addition , many neuron-specific transcripts including novel circadian neuropeptides were revealed . Besides transcriptional level gene regulation , post-transcriptional level regulation , such as alternative pre-mRNA splicing , is also known to be essential for the normal functioning of the nervous system . Alternative pre-mRNA splicing ( AS ) is a major gene regulatory mechanism that enables a single gene locus to produce populations of often functionally distinct mRNAs in a tissue- or cell-type-specific manner , greatly diversifying the metazoan transcriptome . The nervous system is well recognized to exhibit extraordinarily complex and diverse AS patterns in a variety of metazoan organisms ( Li et al . , 2007; Wang et al . , 2008; Irimia et al . , 2014; Li et al . , 2015 ) . Importantly , these diversified AS patterns have been shown to play important roles in modulating numerous neuronal states and activities ( Vuong et al . , 2016; Norris and Calarco , 2012 ) . For example , many neuronal receptors such as the NMDA receptors undergo AS that results in functionally distinct proteins that are regulated in different ways to directly affect neuronal membrane depolarization and action-potential firing ( Chandrasekar , 2013 ) . Recent findings reported several cases where AS is involved in the modulation of the circadian clock in plants , mice and Drosophila ( Sanchez et al . , 2010; Petrillo et al . , 2011; McGlincy et al . , 2012; Hughes et al . , 2012; Preußner et al . , 2014 ) . However , a comprehensive understanding of AS profiles and their role in circadian clock regulation remains largely unexplored . As most studies profiling the Drosophila neuronal transcriptome originate from head samples , it has been difficult to detect cycling of alternatively spliced transcripts because heterogeneous tissues can often mask cell-type specific transcript cycling ( Kula-Eversole et al . , 2010; Nagoshi et al . , 2010; Abruzzi et al . , 2017 ) . In addition , currently most AS analysis software for short-read Illumina RNA-seq data depend on pre-annotated libraries of known spliced transcripts , making it difficult to detect and quantitate novel neuronal AS events that prevail in neuronal tissues . Here , we have comprehensively analyzed the global AS profiles in the transcriptomes of manually dissected DN1 , LNd , and LNv circadian neurons , as well as the non-circadian , TH control neurons described above , using a new computational tool called JUM ( the Junction Usage Model ) ( Wang and Rio , 2017a ) . JUM is able to identify , quantitate and categorize tissue-specific AS patterns from RNA-seq datasets without the need for or use of prior genome or transcriptome annotations . Using JUM , we identify hundreds of previously unannotated pre-mRNA isoforms . They differ extensively among the four neuronal subgroups and were totally missed when RNA-seq data from the heterogeneous fly head sample was analyzed using identical methods . Gene ontology analysis of differentially spliced mRNAs in circadian neurons versus non-circadian neurons revealed that they are enriched for transcripts encoding potassium channels , a hallmark of neuronal excitability . In addition , we discovered many new alternatively spliced variants and cycling AS patterns in transcripts encoding proteins of the molecular clock including the photoreceptor Cryptochrome and the kinase Shaggy . Importantly , we also identified a large set of transcripts that exhibit cycling patterns of alternative splicing throughout the day-night cycle , potentially contributing to the normal functioning of the circadian clock . This study highlights and reinforces the idea that post-transcriptional processes like AS impact cell type specification , diversity and function in the brain .
To investigate how AS regulation differs with circadian neuron identity and function , we compared the pre-mRNA splicing profiles from three small groups of circadian neurons , the DN1s , LNds and LNvs , as well as from a non-circadian outgroup , the dopaminergic or tyrosine hydroxylase-expressing ( TH ) neurons . Total RNA and then mRNA from ~100 manually isolated neurons were purified from entrained flies every four hours across two days ( 12 samples for each group ) . A mixture of oligo-dT and random-primed cDNA was used to create RNA-seq libraries from each sample , as previously described ( Abruzzi et al . , 2015 ) . To profile the AS patterns from each neuron group , we first collapsed the time-series RNA-seq data for each neuron group into two pooled datasets , one for each day , and compared the profiles of alternatively spliced junctions using JUM ( Wang and Rio , 2017a ) . We chose the JUM software because of its unique ability to identify , categorize and quantitate global splicing patterns without any a priori knowledge of , or need for , a genome or transcriptome annotation . Since neurons often exhibit exceptionally diverse AS patterns that are not documented in current transcriptome annotations ( Li et al . , 2007; Wang et al . , 2008; Irimia et al . , 2014; Li et al . , 2015 ) , JUM allows for the discovery of novel and unannotated splicing events and patterns . JUM exclusively utilizes RNA-seq reads that map to splice junctions for AS analysis and defines alternatively spliced junctions ( AS junctions ) as splice junctions that have alternative 5’ - or 3’- splice sites ( Figure 1A; left panel ) . To further evaluate the tissue-specificity and diversity of AS from these neuron groups , we compared their profiles of AS junctions to those from a separately prepared Drosophila head sample , the most common source of adult nervous system RNA . We identified tens of thousands of AS junctions in each distinct neuronal sample with extensive length variation ( Figure 1—figure supplement 1 ) . Importantly , almost all AS junctions contain canonical splice sites ( Supplementary file 1 ) . Below , we will focus our description on the events that are typical of canonical AS junctions , which are located within single genes . We identified a total of ~6 , 000 AS junctions in each of the distinct neuronal subtypes that reside in ~2000 genes , and ~14 , 000 AS junctions in heads that reside in ~3000 genes ( Figure 1B; Figure 1—source data 1; see Materials and methods ) . Approximately 20% of these head AS junctions are novel , that is , they were unannotated in the current Drosophila genome annotation , and these novel AS junctions are found in 1320 genes ( Figure 1B; light blue ) . Importantly , many AS junctions were identified exclusively in each of the neuronal subpopulations compared to fly heads ( ~2300 in total , in 839 genes; Figure 1B; red ) , and nearly all of them ( ~95% ) were novel , that is , they had not been previously annotated in the current Drosophila genome annotation ( genome version: FB2017_05; Figure 1C ) . This analysis provides a glimpse of the transcriptome diversity in these neuron groups , which is undetectable in total fly head RNA . Comparing the AS junctions in the DN1 , LNd , LNv , TH neurons and heads identifies not only many AS junctions that are specific to a particular neuronal group but also many with extensive overlap among the neuron subgroups ( Figure 1D ) . Indeed , approximately 60% of the novel AS junctions ( 1398 within 383 genes ) are present in all 4 groups of neurons but not present in heads ( Figure 1D; bottom left ) . These may be common splice junctions of transcripts that are not ubiquitous and therefore less abundant and not detectable in head transcriptomes . The remaining 40% of the novel AS junctions ( 889 ) are specific to one or more of the neuron groups ( Figure 1D; bottom leftmost ) and are from 610 genes . These junctions may therefore specify cell-type-specific protein isoforms resulting in functional differences between the neurons . Another 1 , 436 AS junctions were identified in fly heads but not in any of the four neuronal groups ( Figure 1C , D ) and are from 812 genes; these junctions may be non-neuronal , for example , from glia , or they may derive from neurons other than the four groups characterized here . Finally , 4 , 374 AS junctions are identified in fly heads as well as in a subset of the neuronal groups and are from 2084 genes ( Figure 1D; bottom rightmost ) . Importantly , these identified novel AS junctions are predicted to encode protein isoforms that may impact the function of their specific neuron groups . Gene ontology analysis reveals that the novel AS junctions impact crucial functions in each of these neuron subgroups including neurotransmitter secretion in LNvs , acetylcholine-activated cation-selective channel activity ( covering five different subunits of the nicotinic acetylcholine receptors ) and synaptic target recognition in LNds , potassium ion transporters and locomotor rhythms in DN1s , and chemical synaptic transmission in THs ( Figure 1—figure supplement 2; Supplementary file 2 ) . One of the circadian genes that shows a novel AS junction in DN1s is the circadian photoreceptor cryptochrome ( cry ) . cry has a novel splice junction in DN1 and LNd neurons , which is absent in the LNv , TH and head samples ( Figure 2A; the sashimi plot indicates the number of sequencing reads spanning the splice junctions [Katz et al . , 2015] ) . This novel AS junction spans from the 5’- splice site right after the first exon to an alternative 3’-splice site in the first intron of cry and results in a truncated , short transcript and presumptive short protein isoform that lacks the functional domains of cry . Approximately 20% of the DN1 ( Figure 2A ) and approximately 28% of the LNd ( Figure 2A ) cry transcripts are spliced into this short transcript isoform . Furthermore , CG10483 transcripts encoding a putative G-protein-coupled receptor ( GPCR ) experience a novel exon skipping event ( Figure 2B; skipped exon marked by ‘*’ ) only in LNds and DN1 ( Figure 2B ) . This event cleanly removes ~65 aa from the receptor transmembrane regions in the middle of the protein . To explore the potential regulatory role of alternative pre-mRNA splicing in circadian rhythms , we compared the global AS patterns in the three subtypes of circadian neurons ( DN1 , LNd and LNv ) with the non-circadian TH neurons using JUM . To do this , JUM grouped all identified AS junctions into the basic AS quantitation unit called AS structures , which are a set of AS junctions that share the same 5’ splice site or 3’ splice site ( Figure 1A; middle panel ) . Each AS junction within an AS structure is defined as a sub-AS-junction of the corresponding AS structure . JUM then quantifies the ‘usage’ of each sub-AS-junction in every AS structure , that is the relative abundance of each sub-AS-junction in the AS structure and identifies AS structures whose usage of sub-AS-junctions are significantly different between the circadian and non-circadian neuron groups . Finally , JUM faithfully assembles the AS structures into the conventionally recognized AS patterns ( cassette exon - SE , alternative 5’ or 3’ splice site – A5SS/A3SS , mutually exclusive exons - MXE , and intron retention - IR ) based on the unique topological features of each pattern ( Figure 1A; right panel ) ( Wang and Rio , 2017a ) . While our first analysis identified ‘all or none’ neuronal group-specific splice junction usage , this approach allows for the identification of differential splice junction usage in which AS junctions are present in both neuronal subgroups but used at significantly different frequencies . Using this method , we found 249 , 194 and 70 AS events that are significantly differentially spliced in the DN1 , LNv and LNd neurons compared to the TH neurons , respectively , which cover all five conventionally classified AS patterns ( Figure 3A; Materials and methods; Figure 3—source data 1–3 ) . To test whether the differentially spliced AS events identified are specific to a particular group of neurons , we examined the overlap of the differentially spliced AS events in each neuronal subgroup ( Figure 3B , top right; Figure 3—figure supplement 1 ) . Remarkably , the majority of the differentially spliced AS events in DN1 , LNv or LNd neurons , compared to TH neurons , are unique to each of the circadian neuronal subtypes ( Figure 3B ) . Only approximately 1 . 6–6 . 6% ( 9-36 ) of these differentially spliced AS events were found to overlap between 2 of the three circadian neuronal subgroups , which is significantly lower than expected ( Hypergeometric Test; p value = 2 . 38e-07 for DN1/TH and LNd/TH; p value = 3 . 68e-39 for DN1/TH and LNv/TH; p value = 1 . 7e-04 for LNd/TH and LNv/TH ) . Only 1 . 8% ( 8 ) of these differentially spliced AS events are found in all three circadian neuron groups . Three of these eight genes are involved in regulating neuronal plasticity , remodeling and synaptic transmission ( Pten , Sap47 , Rim ) . This result suggests that each circadian neuronal subgroup possesses a unique pattern of AS isoforms that contribute to the identity of each distinct neuronal group ( see Discussion ) . To further support this conclusion , we also directly compared differential AS patterns within the circadian neuron subgroups; as predicted by the above results , we found many neuronal group-specific AS events ( Figure 3—figure supplement 2; Figure 3—source data 4–6 ) . To explore mechanisms that might contribute to these circadian neuron group-specific AS profiles , we identified RNA binding proteins ( RBPs ) that are differentially expressed in each of the circadian neurons compared to the non-circadian TH neurons ( Figure 3B; Figure 3—source data 7–9 ) . Interestingly , each circadian neuron subpopulation expresses its own unique set of differentially expressed RBPs compared to TH neurons , with very limited overlap ( Figure 3B , lower left panel ) . Further profiling of the identified targets of a subset of these RBPs using publicly available CLIP and RIP-seq experimental data ( Stoiber et al . , 2015; Hansen et al . , 2015 ) suggest that these specific RBPs could account for many AS events . For example , Syb and mub , two RBPs that are differentially expressed in LNv neurons compared to TH neurons , target 23 and 32 LNv-specific differentially spliced AS target RNAs in LNv versus TH neurons , respectively , with five overlaps , which covers ~36% of total LNv neuron specific differentially spliced AS events . Although speculative , this analysis highlights the possibility that circadian neuron-specific RBP expression could drive much of the observed circadian neuron-specific AS patterns compared to TH neurons . One example of a differentially spliced transcript in LNv neurons compared to TH neurons is the neuronal synaptobrevin ( N-syb ) transcript . It shows a much higher inclusion frequency of an exon ( Figure 4A; marked by ‘*’ ) present in N-syb variant J . 85% of n-Syb transcripts in LNv cells utilize this exon compared to 25–36% in LNd , DN1 and TH neurons ( Figure 4A ) . Inclusion of this exon should give rise to a unique protein of 206 amino acids containing an alternative 85 bp C-terminus . The potassium channel Shab also undergoes alternative 3’-splice site use . In TH and LNd neurons the Shab transcripts preferentially utilize an alternative terminal exon ( Figure 4B; marked by ‘*’ ) . It results in a significantly shorter isoform and encodes a protein lacking the sixth transmembrane region of the potassium channel protein ( Figure 4B ) , which function as a dominant negative protein . 61–74% of Shab transcripts encode this short isoform in TH and LNd neurons , whereas with the ratio of the short isoform decreases significantly , to 10 and 40% , in LNvs and DN1s ( Figure 4B ) . To further investigate which molecular functions in circadian neurons might be impacted by AS , we performed gene ontology ( GO ) analysis ( Figure 5A ) on those genes with significantly differentially spliced AS events in each of the circadian neuronal subtypes . Interestingly , potassium ion transport was the top GO term for both the DN1 and LNv neurons ( Figure 5A and B ) . The differentially spliced AS events include six different potassium channels , as well as two sodium-potassium-exchanging ATPases ( see Shab ( above ) and Figure 5C ) . These ion channels modulate neuronal excitability and many of them have been implicated in controlling both circadian locomotor activity and sleep ( Jaramillo et al . , 2004; Cirelli et al . , 2005; Pimentel et al . , 2016; Fogle et al . , 2015 , see Discussion ) . Other notable GO terms in the LNv neuronal population are neuronal projection ( Fas2 , unc-13 , Slo , Dscam1 , Stai ) , terminal bouton ( cpx , Rab3-GEF , Syn , unc-13 and n-syb ) and synaptic vesicle exocytosis ( Tomosyn , cpx , syn , unc-13 , Rim ) . These GO terms capture some of the unique features of the LNv neuronal population: they exhibit highly dynamic neuronal projections that undergo time-of-day dependent morphological changes and synaptic vesicle localization ( Fernández et al . , 2008; Gorostiza et al . , 2014; Petsakou et al . , 2015 ) . Other notable GO terms for the DN1s include the calcium-calmodulin-dependent family of protein kinases ( CASK , CAMKI , CAMKII , and CG17528 ) , as well as sodium ion transporters ( Nhe3 , Nckx30c , NaCP60E , Atpalpha , nrv3 ) . In contrast , the lower number of differentially alternatively spliced transcripts in the LNds resulted in no statistically significant GO terms . To further explore the functional importance of the differentially spliced AS events in circadian neurons versus TH neurons , we examined the cross-species conservation of identified differentially spliced cassette exons and the preservation of reading frame from the AS of these cassette exons . We first plotted the PhastCons conservation score ( Siepel et al . , 2005 ) across 27 species of insects for the sets of alternatively spliced and non-alternatively spliced cassette exons , respectively , in each of the circadian neuron subgroup versus TH comparisons ( Figure 3—figure supplement 3A ) . Interestingly , the alternatively spliced cassette exons are somewhat more conserved than the non-alternatively spliced cassette exons , although the result is not statistically significant ( Mann-Whitney U Test; p value 0 . 17 ) . We also found that the inclusion/exclusion of the alternatively spliced cassette exons are more prone to preserve reading frame than non-alternatively spliced exons ( Figure 3—figure supplement 3B ) , further indicating that these AS events are important to the identity and function of the neuron subpopulation . To further investigate the role of AS regulation in circadian rhythms , we used JUM to identify AS events that exhibit cycling alternative splicing in each of the neuronal subtypes . To do this , we queried AS structures identified in each of the neuronal subgroups across the six time-point RNA-seq data ( two independent replicas for each time point ) to profile the changes in sub-AS- junction usage throughout the day ( F24 and JTK-cycle; see Materials and methods ) . We identified cycling AS structures in all 3 groups of circadian neurons: 173 AS structures in DN1 neurons ( 5 . 7% of all AS structures ) , 92 in LNv neurons ( 5 . 0% of all AS structures ) , and 48 in LNd neurons ( 3 . 7% of all AS structures ) ( Figure 6—source data 1 ) . These events affect 147 , 81 , 43 , and nine genes in DN1 , LNv , LNd and TH neurons , respectively ( Figure 6A ) . Importantly , 85% of the transcripts that exhibit time-dependent changes in alternative splicing do not cycle at the total mRNA level ( Figure 6A; RPKM based on ESAT quantification [Derr et al . , 2016; Abruzzi et al . , 2017] ) . The non-circadian TH neurons serve as a negative control for this analysis ( only 11 cycling AS structures ) because they have very few cycling mRNAs and probably do not express the clock genes ( Abruzzi et al . , 2017 ) . The majority of these cycling AS structures are specific to a particular group of circadian neurons ( Figure 6B ) . Some of this specificity is because certain AS structures are only found in one subgroup of circadian neurons ( ~40% in DN1 , 14% in LNd and 25% in LNv ) . Moreover , only two transcripts show cycling alternative splicing in all three groups of circadian neurons: still life ( sif ) that regulates synaptic differentiation and the type II Camp-dependent protein kinase , pKa-R1 ( Figure 6B ) . This striking specificity suggests that time-of-day changes in alternative pre-mRNA splicing acts as an additional and previously unappreciated level of circadian regulation in Drosophila clock neurons , as previously suggested in mammals based on microarrays ( McGlincy et al . , 2012 ) . One striking example of a cycling AS structure is a transcript isoform encoding circadian kinase , Shaggy ( from the fly gene sgg; known in mammals as GSK3 ) . Alternative splicing of the sgg transcript is due to the use of two different 3’- splice sites generating distinct transcript and protein isoforms differing at the C terminus: a shorter isoform utilizing terminal Exon b and a longer isoform utilizing terminal Exon a that includes a specific additional 63 amino acids ( Figure 6C; marked by ‘*’ and ‘**’ , respectively ) . In LNv and DN1 neurons , the longer sgg isoform is almost exclusively used ( Figure 6—figure supplement 1 ) . However , in LNd neurons , the longer isoform is utilized almost exclusively in the morning ( ZT0-ZT11 ) and then by late night ( ZT18-24 ) there is approximately equal usage of the long and short sgg transcripts ( Figure 6C; two replicates quantified in Figure 6D , red line and circle ) . The potassium channel SLOWPOKE is encoded by the Slo gene and also undergoes neuron-specific AS cycling . There are two mutually exclusively used exons in the second alternatively spliced exonic regions in the Slo transcript: exon 2a and exon 2b that alter the cytosolic face of the channel ( Figure 6—figure supplement 2A , marked by ‘*’ and ‘**’ , respectively ) . In DN1 neurons only , exon 2b is included in the mornings and then there is a shift toward exon 2a inclusion in the late nighttime ( Figure 6—figure supplement 2A , B ) . SLOWPOKE has been shown to affect circadian behavior in Drosophila ( Ceriani et al . , 2002 ) , and its mammalian homolog , Kcnma1 , shows similar phenotypes in mammals ( Meredith et al . , 2006 ) . It is possible that the different protein isoforms created by time-of-day changes in alternative splicing mRNA isoforms are important for these functions ( see Discussion ) . To explore how cycling AS could impact neuronal functions , we performed gene ontology analysis on transcripts undergo cycling AS in the circadian neurons . Interestingly , modulation of locomotor behavior is enriched for cycling AS transcripts in all three circadian neuron groups , although a distinct group of transcripts is implicated in each circadian neuronal group ( Figure 6—figure supplement 3; Supplementary file 3 ) . In LNvs and LNds neurons , the most significantly enriched GO term for cycling AS transcripts is male courtship behavior and olfactory learning , respectively ( Figure 6—figure supplement 3A , B; Supplementary file 3 ) . In DN1 neurons , the top enriched GO term for cycling AS transcripts is mRNA binding proteins and 8 out of the 12 genes in this category encode alternative splicing regulators . The cycling AS of these splicing regulators could further facilitate downstream time of day-dependent changes in global splicing profiles ( Figure 6—figure supplement 3C; Supplementary file 3 ) . Previous analysis of mRNA expression cycling in the circadian neurons revealed that mRNAs cycle with distinct phases in each subgroup of neurons ( Abruzzi et al . , 2017 ) . Interestingly , the phases of the cycling AS structures also show neuron-group specific distributions ( Figure 7A; Figure 7—figure supplement 1; blue ) . Most interesting were the DN1 neurons in which the majority of the time-of-day-dependent cycling AS structures peak in the late night ( Figure 7A; blue ) . This peak is almost anti-phase to that of DN1 mRNA cycling ( Figure 7A , red ) . In contrast , mRNA cycling and AS cycling show similar phases in the LNv neurons ( Figure 7—figure supplement 1A ) . The LNd neurons have a third pattern: cycling mRNAs peak in the early evening , but cycling AS structures have a random distribution throughout the day ( Figure 7—figure supplement 1B ) . To probe potential mechanisms contributing to this neuron-specific AS cycling , we queried the list of cycling mRNAs within these neurons ( Abruzzi et al . , 2017 ) and identified cycling transcripts encoding RBPs in all three neuronal subgroups ( Figure 7B ) . None of the identified RBP mRNAs cycle in more than one circadian neuron subgroup . Although there was no clear global correlation between the phases of these RBP mRNAs and cycling AS in any neuronal group , we did identify a cycling splicing factor in LNds , Qkr54B , that has been shown to target Sgg transcripts ( Stoiber et al . , 2015 ) . Interestingly , Qkr54B mRNA peaks 3–4 hr prior to the time of day dependent inclusion of the alternative spliced Sgg terminal Exon b ( Figure 6D; green line and triangle ) , suggesting that cycling RBP expression might contribute to the cycling AS patterns .
Tissues of the nervous and germline systems , such as brain , testes and ovaries , have more complex transcriptomes than other cell types due to extensive alternative pre-mRNA splicing or AS ( Wang et al . , 2008; Pan et al . , 2008 ) . The nervous system especially exhibits vast numbers of AS isoforms , many of which are novel and are only beginning to be comprehensively identified ( Li et al . , 2007; Irimia et al . , 2014 ) . This increase in transcript isoform complexity likely contributes to the specification and functional diversity of cell types within the nervous system . Here , we have applied a novel computational algorithm called JUM to characterize the transcript isoform diversity generated by alternative splicing in three circadian neuronal subtypes ( LNv , LNd and DN1 ) , as well as a non-circadian dopaminergic neuron population ( TH neurons ) of the Drosophila central nervous system . JUM can comprehensively analyze , quantitate and compare tissue- or cell-type-specific AS patterns without requiring a priori annotations of known transcripts or transcriptomes ( Wang and Rio , 2017a ) . Our analysis revealed a previously unappreciated diversity and complexity of alternatively spliced transcript isoform patterns in these four neuronal subtypes , suggesting that they contribute to neuronal identity , connectivity , activity and circadian functions . This is because many of these novel , previously undetected and unannotated isoforms were unique to a given neuronal population and occurred in transcripts from genes implicated in neuronal activity or circadian rhythms . For example , the kinase Shaggy and the blue light photoreceptor cryptochrome play central roles in circadian clock regulation and have novel AS patterns in discrete subsets of the circadian neurons . In addition , nine different transcripts involved in potassium transport undergo differential AS in circadian neurons compared to non-circadian neurons . These transcripts encode six different potassium channels ( Figure 5 ) . Many of these genes have a complex organization known to encode populations of functionally distinct proteins isoforms , which change the activation kinetics as well as calcium sensitivity of the channels ( Johnson et al . , 2011 ) . Neuronal firing is known to play a key role in the circadian circuit with recent studies illustrating that different subgroups of circadian neurons have characteristic time-of-day neuronal firing patterns ( Flourakis et al . , 2015; Liang et al . , 2016; Guo et al . , 2017 ) . Although it is not yet fully understood which potassium channels play a critical role in each circadian neuron subgroup , several channel pre-mRNAs that undergo differential splicing in circadian neurons impact circadian behavior and sleep , such as slowpoke ( Slo [Jaramillo et al . , 2004] ) , Shaker ( Sh; [Cirelli et al . , 2005; Pimentel et al . , 2016] ) and Hyperkinetic ( Hk; [Fogle et al . , 2015] ) . It is therefore likely that AS adds diversity and distinct physiological properties to these protein isoforms , which then impacts neuron-specific firing patterns . From a more general perspective , AS augments transcriptional regulation in giving different circadian neurons individual identities and distinct functions . Approximately 5% of the AS events identified in circadian neurons also undergo time-of-day dependent changes in alternative splicing ( cycling splicing ) . It is important to note that all our experiments were conducted under 12 hr of light and 12 hr of dark conditions , making it impossible to distinguish between light and clock control . Nonetheless , these data indicate that splicing adds a dramatic layer of gene regulation to diurnal changes in gene expression . Moreover , many of the cycling AS transcripts show constant overall mRNA levels , which suggests the existence of neuron-specific splicing factors that are expressed or activated only at specific times of the day . Indeed , we have identified several candidate cycling neuron-enriched transcripts that encode RBPs that may help to drive cycling AS patterns . A recent trend in biological research is to generate transcriptome profiles from single cells . For example , this strategy is part of the ‘human cell atlas’ project aimed at personalized genomic medicine or the ‘brain initiative’ project to generate profiles of all neurons in the mouse brain ( Jorgenson et al . , 2015; Regev et al . , 2017 ) . One recent study was able to obtain about 20M sequence reads per isolated human iPS cell but only managed to analyze splicing patterns for the most highly expressed genes ( Song et al . , 2017 ) . Our study in contrast used ~100 isolated Drosophila neurons for each of the four neuron subtypes along with judicious use of both oligo-dT and random hexamer priming of the cDNA libraries . This strategy obtained about 10–30M sequence reads for each sample , including substantial information from the 5’ ends of transcripts , and JUM was able to detect and classify a large number of previously unannotated pre-mRNA isoforms . Many of them are missing from our fly head RNA-seq data assayed and analyzed in parallel , indicating that these new isoforms are cell-type specific . Not surprisingly , the novel isoforms from the three circadian neuron groups fall into many gene ontology ( GO ) categories associated with specific circadian clock activity and function . Taken together , the work presented here indicates that the number of alternative splicing events that take place in neuronal tissues is grossly underestimated , even though publically-funded genome projects , such as the NIH modENCODE projects deeply sequenced transcriptomes from a variety of Drosophila tissues and developmental stages . This is despite the appreciation of how much AS occurs in the nervous system , for example recent comprehensive analysis of splicing patterns through deep sequencing of ~50 mouse and human tissues revealed about 2500 neuronally-regulated alternative splicing events ( Irimia et al . , 2014; Li et al . , 2015 ) . We therefore suggest that these events will need to be comprehensively evaluated by much deeper sequencing than is currently afforded by most contemporary single cell RNA-seq studies and by AS analysis software like JUM that is not constrained by a priori knowledge of known splicing events .
Neuron dissection , RNA extraction and library preparation were performed as described in ( Abruzzi et al . , 2015 ) . The sequencing data used in this study is publically-available at Gene Expression Omnibus ( Accession numbers GSE77451 ( neurons ) and GSE79916 ( heads ) ) . RNA-seq reads were mapped to the Drosophila ( dm3 ) genome using STAR ( Dobin et al . , 2013 ) in the 2-pass mode; the exact commands used are listed in Supplementary file 5 . Only uniquely mapped reads are kept for downstream analysis . The mapping statistics for each sample are listed in Supplementary file 4 . For pooled data analysis , fastq files from each time points were pooled before subjecting the pooled data to mapping as described above . The mapping parameters for splice junction profiling are set as default in STAR: the minimum overhang length for splice junctions on both sides are set to be the 30 bp for non-canonical motif junctions , and 12 bp for the canonical GT/AG ( CT/AC ) , GC/AG ( CT/GC ) , AT/AC ( GT/AT ) motif junctions , respectively . The minimum uniquely mapped read count per junction is set to be three for non-canonical motif junctions and one for the canonical motif junctions . The minimum allowed distance to other splice junctions’ donor/acceptor is set to be 10 for non-canonical motif junctions and 0 , 5 , 10 for the canonical GT/AG ( CT/AC ) , GC/AG ( CT/GC ) , AT/AC ( GT/AT ) motif junctions , respectively . The maximum gap allowed for junctions are set to be as follows: junctions supported by one read can have gaps ≤ 50000 b; if supported by two reads then gaps ≤ 100000 b; if supported by three reads then gaps ≤ 200000; if supported by more than four reads then gaps ≤ alignIntronMax ( see the STAR manual for details ) . Splice junctions with alternative 5’ or 3’ splice sites in the sample were profiled from the pool of total STAR-identified splice junctions for each neuronal sample , and defined as AS junctions . The novelty of each AS junction was compared against the library of annotated junctions in the UCSC genome browser RefSeq transcriptome annotation ( genome version: FB2017_05 ) . Gene Ontology analyses were performed using David Bioinformatics Resources 6 . 8 ( https://david . ncifcrf . gov/home . jsp ) ( Huang et al . , 2009a ) . For neuron-subgroup-specific analyses , a list of transcripts expressed at greater than five reads was used as a background data set . A p-value of less than 0 . 05 was required for a gene to be considered enriched in the dataset . All RNA-seq data are visualized using IGV ( Thorvaldsdóttir et al . , 2013; Robinson et al . , 2011 ) and the Sashimi plots tool ( Katz et al . , 2015 ) . Visualized tracks are further organized using ImageJ ( Schneider et al . , 2012 ) . All boxplots in this paper were plotted using BoxPlotR ( Spitzer et al . , 2014 ) . For each comparison of a circadian neuron subtype ( DN1 , LNv and LNd ) and the non-circadian , TH neuron subtype , total splice junctions that receive more than 10 unique mapped reads in both collapsed datasets in the circadian or non-circadian neuron subtype samples were pooled together , and AS structures were constructed based on the pooled splice junction set . The usage of each sub-AS-junction in every AS structure was calculated and AS structures with significantly differentially ‘used’ sub-AS-junctions between each of the circadian neuron subtypes and the non-circadian neuron subtype were profiled as specified ( Wang and Rio , 2017a ) . All AS structures were then assembled into the five conventionally recognized AS patterns – cassette exon ( SE ) , alternative 5’/3’ splice sites ( A5SS/A3SS ) , mutually exclusive exons ( MXE ) , and intron retention ( IR ) , as well as a Composite category , which represents inseparable combinations of the conventional AS patterns . Here we are only focusing on AS events in the conventional AS pattern categories . Only AS events that showed more than 10% of change in alternative splicing and had a differential splicing test statistical pvalue ≤ 0 . 05 were considered significantly differentially spliced AS events . For more details see ( Wang and Rio , 2017a ) and the GitHub page of JUM ( Wang and Rio , 2017b ) : https://github . com/qqwang-berkeley/JUM ( copy archived at https://github . com/elifesciences-publications/JUM ) . The same procedure was performed to analyze differential AS patterns among circadian neuron subgroups as well . Differential gene expression analysis between each circadian neuron subgroup and the non-circadian TH neurons were performed using DESeq2 ( Love et al . , 2014 ) . Adjusted-p value 0 . 05 was chosen as the statistical cutoff . Conservation scores ( PhastCon ) ( Siepel et al . , 2005 ) for each single base in the cassette exons for alignments of 26 insects with D . melanogaster was downloaded from the UCSC Genome browser and an average PhastCon score for each cassette exon was calculated and compared . For each neuron subtype , total splice junctions that receive more than 10 unique mapped reads in at least one time point from both replicas were pooled . AS structures were constructed based on the pooled splice junction set . The relative usage of each sub-AS-junction in every AS structure is calculated , and subject to analysis using fourier analysis ( F24; [Wijnen et al . , 2005] ) and JTK-cycle ( Hughes et al . , 2010 ) . To be considered cycling using fourier transformation the following cutoffs were used: F24 score greater than 0 . 5 , >1 . 5 fold change in splice junction usage , and the average transcript reads greater than 10 for at least one timepoint in each independent dataset . JTK cycle identified transcripts as cycling that had >1 . 5 fold change in splice junction usage , the average transcript reads greater than 10 for at least one timepoint in each independent dataset , and a p-value cutoff of less than 0 . 05 . Phase determination was done using fourier transformation . | The life of nearly all creatures on Earth follows the rhythm of day and night . For example , in fruit flies , darkness and light dictate when the insects feed , rest , move or mate . This is possible thanks to the circadian clock , an internal program which is synchronized with the environment to tell cells in the body when to perform certain roles . In fruit flies , the structure that keeps the body clock ticking is formed of about 150 ‘circadian neurons’ , which are divided into several subgroups . In these cells , a complex genetic programis at work , with networks of genes being ‘switched on’ in a cyclical way . To understand how this program works , scientists need to know which genes are turned on and when , as well as which proteins are created based on the information contained in these genes . This can be difficult because one gene does not necessarily code for only one protein . Indeed , when a gene is turned on , it gets copied into a pre-messenger RNA ( pre-mRNA ) , which is formed of several modules . The pre-mRNA can then go through a process called alternative splicing that shuffles or removes the different modules . This means that one gene can give rise to different pre-mRNA molecules that will each serve as a template to build a distinct protein . Until now , there have been few studies that examine the different types of pre-mRNAs found in circadian neurons , and how these change with the time of day . Here , Wang , Abruzzi et al . extract three subgroups of circadian neurons , and one subgroup of non-circadian neurons , from the brain of fruit flies . The pre-mRNAs are isolated , and then a new computational method , called JUM , identifies , counts and categorizes the pre-mRNA molecules in the different groups of neurons . This analysis reveals hundreds of previously unknown pre-mRNA molecules , many of which differed extensively between the types of brain cells . When comparing circadian and non-circadian neurons , Wang , Abruzzi et al . show that the circadian cells had more pre-mRNAs that code for proteins that help the cell communicate with other neurons . Finally , many genes in the circadian neurons use alternative splicing to turn on different types of pre-mRNA molecules at different times of the day in a cyclical way; this suggests that these pre-mRNAs might be participating in the genetic circadian program . Many human disorders , such as certain forms of insomnia , emerge when the circadian clock is thrown off balance . The results reported by Wang , Abruzzi et al . show that alternative splicing may be an overlooked mechanism that shapes this complex program . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"neuroscience"
] | 2018 | Striking circadian neuron diversity and cycling of Drosophila alternative splicing |
The AAA+ family ATPase TRIP13 is a key regulator of meiotic recombination and the spindle assembly checkpoint , acting on signaling proteins of the conserved HORMA domain family . Here we present the structure of the Caenorhabditis elegans TRIP13 ortholog PCH-2 , revealing a new family of AAA+ ATPase protein remodelers . PCH-2 possesses a substrate-recognition domain related to those of the protein remodelers NSF and p97 , while its overall hexameric architecture and likely structural mechanism bear close similarities to the bacterial protein unfoldase ClpX . We find that TRIP13 , aided by the adapter protein p31 ( comet ) , converts the HORMA-family spindle checkpoint protein MAD2 from a signaling-active ‘closed’ conformer to an inactive ‘open’ conformer . We propose that TRIP13 and p31 ( comet ) collaborate to inactivate the spindle assembly checkpoint through MAD2 conformational conversion and disassembly of mitotic checkpoint complexes . A parallel HORMA protein disassembly activity likely underlies TRIP13's critical regulatory functions in meiotic chromosome structure and recombination .
The assembly and disassembly of specific protein complexes underlies many important signaling pathways in the cell . The HORMA domain ( Aravind and Koonin , 1998 ) is a conserved , structurally unique signaling module that forms complexes through a characteristic ‘safety-belt’ interaction in which the C-terminus of the domain wraps entirely around a short region of a binding partner ( Luo et al . , 2002; Sironi et al . , 2002; Hara et al . , 2010; Kim et al . , 2014 ) . HORMA domain protein complexes participate in multiple cellular signaling pathways , including meiotic recombination control , DNA repair , and the spindle assembly checkpoint ( SAC ) . While the regulated assembly of HORMA domain protein complexes has been extensively characterized , the mechanisms underlying their disassembly , which requires significant conformational changes to disrupt the extremely stable safety-belt interaction , are largely unknown . In meiosis , homologous chromosomes must recognize one another and recombine , forming physical links called crossovers ( COs ) that enable their bi-orientation and segregation in meiosis I ( Zickler and Kleckner , 1999 ) . CO formation is promoted and regulated by a conserved family of HORMA domain proteins termed HORMADs . Early in meiotic prophase , HORMADs localize to chromosomes along their entire lengths , where they promote the introduction of DNA double-strand breaks and bias the repair of those breaks toward the homologous chromosome ( Subramanian and Hochwagen , 2014 ) . In both yeast and mammals , excess recombination is limited by a feedback mechanism that removes or redistributes HORMADs along the chromosome after sufficient COs have formed . The removal of HORMADs depends on a conserved AAA+ family ATPase , Pch2/TRIP13 , without which the frequency and spatial distribution of COs is disrupted ( San-Segundo and Roeder , 1999; Börner et al . , 2008; Joshi et al . , 2009; Wojtasz et al . , 2009; Roig et al . , 2010; Chen et al . , 2014 ) . We have previously shown that the HORMADs assemble into higher-order oligomers through head-to-tail safety-belt interactions , and that these interactions are critical for their meiotic functions ( Kim et al . , 2014 ) . As a predominant family of AAA+ ATPases function to disaggregate or disassemble protein complexes ( Erzberger and Berger , 2006; Sauer and Baker , 2011 ) , it has been proposed that Pch2/TRIP13 mediates HORMAD removal from chromosomes through specific recognition and disassembly of chromosome-associated HORMAD complexes ( Chen et al . , 2014 ) . Recently , mammalian TRIP13 has been shown to regulate the SAC , which monitors kinetochore-microtubule attachment in both mitosis and meiosis ( Musacchio and Salmon , 2007 ) . In this pathway , unattached kinetochores generate an inhibitor of the anaphase promoting complex/cyclosome ( APC/C ) called the mitotic checkpoint complex ( MCC ) , which is composed of the MAD2 , CDC20 , BUBR1 , and BUB3 proteins ( Hardwick et al . , 2000; Fraschini et al . , 2001; Sudakin et al . , 2001 ) . MAD2 is a relative of the meiotic HORMADs , and exists in one of two conformers: an inactive ‘open’ state ( O-MAD2 ) , and an active ‘closed’ state ( C-MAD2 ) ( Figure 8—figure supplement 1A ) that binds CDC20 through a safety-belt interaction to form the core of the MCC ( Sironi et al . , 2002; Luo et al . , 2004; Shah et al . , 2004; Chao et al . , 2012 ) . After all kinetochores become properly attached to microtubules , new MCC assembly is halted and the SAC is inactivated . Timely SAC inactivation requires two factors , TRIP13 ( Wang et al . , 2014 ) and the HORMA domain protein p31 ( comet ) ( Habu et al . , 2002; Xia et al . , 2004; Hagan et al . , 2011; Varetti et al . , 2011; Westhorpe et al . , 2011; Ma et al . , 2012 ) , which recent evidence suggests may act together to directly disassemble the MCC . p31 ( comet ) specifically recognizes and binds C-MAD2 , and the p31 ( comet ) -MAD2 interface overlaps MAD2's interface with BUBR1 in the intact MCC ( Xia et al . , 2004; Yang et al . , 2007; Tipton et al . , 2011b; Chao et al . , 2012 ) , suggesting that p31 ( comet ) may compete with BUBR1 for MAD2 binding . Further , the combined activities of p31 ( comet ) and TRIP13 can cause the dissociation of MAD2 from immunoprecipitated CDC20 or BUBR1 complexes in vitro ( Teichner et al . , 2011; Eytan et al . , 2014 ) . Intriguingly , human TRIP13 has also been identified as an oncogene: TRIP13 is overexpressed in a number of human cancers ( Larkin et al . , 2012; van Kester et al . , 2012; Banerjee et al . , 2014; Wang et al . , 2014 ) , and can promote proliferation and invasion when overexpressed in human cell lines ( Banerjee et al . , 2014 ) . The source of TRIP13's oncogenic activity is unknown , but may stem from effects on chromosome structure and DNA repair pathways ( as its meiotic functions would suggest ) , or may instead arise from aberrant regulation of the SAC . Pch2/TRIP13 is thus directly implicated in the regulation of HORMA domain-mediated signaling in two separate pathways , meiotic recombination and the SAC . The mechanistic basis for this regulation , however , remains unknown . Here , we show that Pch2/TRIP13 comprises a new family of AAA+ ATPase protein remodelers , with a substrate-recognition domain similar to the NSF/p97/PEX1 remodeler family and a physical mechanism closely related to the bacterial ClpX unfoldase . We show that TRIP13 converts closed , active MAD2 to its inactive open conformer , and that p31 ( comet ) functions as an adapter to recognize closed MAD2 and deliver it to TRIP13 . Thus , TRIP13 regulates the SAC through MAD2 conformational conversion and safety belt disengagement , and a similar mechanism for HORMAD complex disassembly likely underlies the enzyme's regulatory functions in meiosis .
Pch2/TRIP13 proteins are members of the functionally diverse AAA+ ATPase family ( Erzberger and Berger , 2006; Wendler et al . , 2012 ) . These proteins share a common architecture , with a family-specific N-terminal domain ( NTD ) responsible for localization or substrate recognition , and one or two AAA+ ATPase modules that typically assemble into a hexameric ring . AAA+ ATPases are extremely diverse and include DNA and RNA helicases , DNA replication initiators , and a large family termed the ‘classic remodelers , ’ which disaggregate or unfold proteins; these include the SNARE complex disassembly factor NSF , the ubiquitin-directed disaggregase p97/Cdc48 , and the ATPase component of the eukaryotic proteasome ( Erzberger and Berger , 2006 ) . Sequence comparisons of the Pch2/TRIP13 AAA+ ATPase module fail to clearly classify it within any well-characterized AAA+ family ( Figure 1A ) . Moreover , sequence comparisons of the Pch2/TRIP13 NTD fail to identify homology to any known proteins . Therefore , we took a structural approach to determine the relationship of Pch2/TRIP13 to other AAA+ ATPases . We overexpressed and purified Mus musculus TRIP13 and its C . elegans ortholog PCH-2 , and found that while TRIP13 adopts a range of oligomeric states from monomer to hexamer , PCH-2 forms a stable hexamer both with and without added nucleotides ( Figure 1C , D ) . We next performed negative-stain electron microscopy ( EM ) on PCH-2; low-resolution class averages reveal a distinctly asymmetric hexamer in the absence of nucleotides , which becomes more symmetric and compact when ATP is added ( Figure 1E , Figure 1—figure supplement 1 ) . We attempted crystallization both in the presence and absence of nucleotides , and successfully determined the crystal structure of PCH-2 without added nucleotide to a resolution of 2 . 3 Å . The structure reveals an elongated hexamer with an approximate ‘dimer of trimers’ symmetry and an overall shape similar to our EM class averages of this state ( Figure 2A , Table 1 ) . 10 . 7554/eLife . 07367 . 003Figure 1 . PCH-2/TRIP13 is a distinct class of hexmeric AAA+ ATPase . ( A ) Phylogenetic tree of selected AAA+ ATPases , colored by clade ( Erzberger and Berger , 2006 ) . ( B ) Conserved AAA+ sequence motifs in Pch2/TRIP13 , the ‘classic remodelers’ , and E . coli ClpX . Pch2/TRIP13 and ClpX lack the first of two conserved arginine residues in the Arg finger region ( yellow ) , and possess a Sensor 2 arginine ( R385 , red ) , which the classic remodelers lack . ( C ) Size-exclusion chromatography coupled to multi-angle light scattering ( SEC-MALS ) analysis of C . elegans PCH-2 in the absence of nucleotides . Hexamer molecular weight = 288 . 1 kDa; measured molecular weight = 252 kDa ( red line ) . ( D ) SEC-MALS analysis of M . musculus TRIP13 in the absence of nucleotides . The wild-type protein ( black ) adopts a mixture of oligomeric states from monomer to hexamer , consistent with findings from S . cerevisiae Pch2 ( Chen et al . , 2014 ) . The proportion of higher-molecular weight oligomers increases upon the addition of ATP or non-hydrolyzable analogs ( not shown ) . The ATP hydrolysis-defective TRIP13E253Q mutant ( gray , molecular weight measurements red ) is predominantly hexameric both in the presence ( shown ) and absence of ATP . Molecular weight measurements by SEC-MALS ( red ) are shown for TRIP13E253Q; WT measurements ( not shown ) are consistent . ( E ) Selected negative-stain EM class averages of C . elegans PCH-2 without added nucleotides ( Apo ) or with added ATP . For example raw images , see Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 00310 . 7554/eLife . 07367 . 004Figure 1—figure supplement 1 . Negative-stain electron microscopy ( EM ) of C . elegans PCH-2 . ( A ) Example negative-stain EM image of C . elegans PCH-2 without added nucleotide ( Apo ) , and selected class averages from XMIPP clustering 2D alignment ( from 32 classes , 5916 total particles ) . ( B ) Example negative-stain EM image of PCH-2 in the presence of 1 mM ATP , and selected class averages ( from 16 classes , 4297 total particles ) . PCH-2 hexamers adopt a more compact and sixfold symmetric conformation after ATP addition . Because of the strong bias toward top-down views of the hexamer , 3D reconstructions were not attempted for either state . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 00410 . 7554/eLife . 07367 . 005Figure 2 . Structure of C . elegans PCH-2 . ( A ) Overall structure of PCH-2 . The hexamer shows a ‘dimer of trimers’ symmetry with chains A/B/C equivalent to chains D/E/F . Chains A/D and C/F are bound to SO4− in the ATPase active site , and chains B/E are bound to ADP ( space-fill representation ) . For data collection and refinement statistics , see Table 1 . ( B ) Structural comparison of the PCH-2 NTD ( residues 1–99 ) with the NSF N-C subdomain ( residues 92–189; [May et al . , 1999] ) ; Cα r . m . s . d . 1 . 94 Å over 55 residues . ( C ) Schematic of Pch2/TRIP13 domain structure vs NSF/p97 . Pch2/TRIP13 shares these proteins' N-C subdomain ( blue ) and one of their tandem AAA+ ATPase regions ( green/red ) . ( D ) Close-up view of PCH-2 chain B , with domains colored as in ( C ) , showing its packing against subunits A and C . ( E ) Close-up of ADP bound to PCH-2 chain B , with AAA+ motifs shown as sticks . For close-up views of all six active sites , see Figure 2—figure supplement 1 . ( F ) Stereo view of refined 2Fo − Fc electron density at 2 . 3 Å resolution , contoured at 1 . 0 σ , for the bound ADP and surrounding residues in PCH-2 chain B . View is equivalent to ( E ) ; the small AAA domain has been removed for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 00510 . 7554/eLife . 07367 . 006Figure 2—figure supplement 1 . X-ray crystallographic analysis of C . elegans PCH-2 . ( A ) Mean diffraction intensity divided by standard deviation ( I/σ ) ( left axis , dotted lines ) and the half–set correlation coefficient ( CC1/2 ) ( right axis , solid lines ) for PCH-2 native diffraction data ( see Table 1 ) along the three principal axes a* ( red ) , b* ( yellow ) , and c* ( blue ) . Resolution cutoffs corresponding to an I/σ of 1 . 0 and a CC1/2 of 0 . 5 for c* ( 3 . 2 Å ) and a*/b* ( 2 . 3 Å ) are shown as dashed lines . ( B ) Simulated-annealing Fo − Fc electron density map , contoured at 3 . 0 σ , calculated from a model missing all bound nucleotides and SO4− ions . Views roughly equivalent to Figure 2E are shown for each chain in the PCH-2 hexamer . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 00610 . 7554/eLife . 07367 . 007Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 007PCH-2 SeMetPCH-2 nativeData collection Synchrotron/BeamlineAPS 24ID-ESSRL 12-2 Resolution ( Å ) 3 . 232 . 3 Wavelength ( Å ) 0 . 979210 . 9795 Space groupC2221C2221 Unit cell dimensions ( a , b , c ) Å126 . 1 , 239 . 5 , 198 . 2126 . 7 241 . 0 197 . 9 Unit cell angles ( α , β , γ ) °90 , 90 , 9090 , 90 , 90 I/σ ( last shell ) 9 . 3 ( 1 . 0 ) 17 . 9 ( 0 . 8 ) * Rsym ( last shell ) 0 . 198 ( 2 . 166 ) 0 . 098 ( 3 . 143 ) † Rmeas ( last shell ) 0 . 213 ( 2 . 326 ) 0 . 102 ( 3 . 297 ) ‡ Isotropic CC1/2 , last shell0 . 5920 . 275§ Directional CC1/2 , last shell ( Å ) a*–0 . 498 ( 2 . 3 Å ) b*–0 . 532 ( 2 . 3 Å ) c*–0 . 608 ( 3 . 2 Å ) Completeness ( last shell ) %99 . 9 ( 99 . 9 ) 99 . 5 ( 90 . 8 ) Number of reflections33 , 4621 , 808 , 343 unique4410134 , 133 Multiplicity ( last shell ) 7 . 5 ( 7 . 6 ) 13 . 5 ( 10 . 6 ) Number of sites68–§ Anisotropic scaling B-factors ( Å2 ) a*–−8 . 09 b*–−8 . 02 c*–16 . 11 isotropic B-factor correction–−19 . 65Refinement Resolution range ( Å ) –40 - 2 . 3 No . of reflections–96 , 084 working–91 , 200 free–4884 # Rwork ( % ) –22 . 97 # Rfree ( % ) –26 . 42Structure/Stereochemistry Number of atoms–18 , 017 ligands ( ADP , SO4 ) –89 solvent–55 r . m . s . d . bond lengths ( Å ) –0 . 004 r . m . s . d . bond angles ( ° ) –0 . 730 ¶ PDB ID–4XGU*Rsym = ∑∑j|Ij − 〈I〉|/∑Ij , where Ij is the intensity measurement for reflection j and 〈I〉 is the mean intensity for multiply recorded reflections . †Rmeas = ∑h [√ ( n/ ( n − 1 ) ) ∑j [Ihj − 〈Ih〉]/∑hj 〈Ih〉 where Ihj is a single intensity measurement for reflection h , 〈Ih〉 is the average intensity measurement for multiply recorded reflections , and n is the number of observations of reflection h . ‡CC1/2 is the Pearson correlation coefficient between the average measured intensities of two randomly-assigned half-sets of the measurements of each unique reflection ( Karplus and Diederichs , 2012 ) . §High-resolution native data were anisotropically scaled and elliptical data cutoffs were applied according to directional intensity and CC1/2 data ( see ‘Materials and methods’ and Figure 3—figure supplement 1A for details on data anisotropy and resolution cutoffs ) . #Rwork , free = ∑||Fobs| − |Fcalc||/|Fobs| , where the working and free R-factors are calculated using the working and free reflection sets , respectively . ¶Coordinates and structure factors have been deposited in the RCSB Protein Data Bank ( www . pdb . org ) . While the PCH-2 NTD ( residues 1–99 of 424 ) lacks detectable sequence homology to other proteins , the structure of this domain shows a clear relationship to the N-terminal substrate recognition domains of a AAA+ ‘classic remodeler’ sub-family that includes NSF , p97 , and PEX1 . These proteins possess two-part NTDs with tightly associated N-N and N-C subdomains ( May et al . , 1999; Yu et al . , 1999 ) . A hydrophobic cleft between the two subdomains binds either directly to substrates , or alternatively to ‘adapter’ proteins that aid localization and specific substrate recognition ( Kloppsteck et al . , 2012 ) . In PCH-2 , the NTD contains a single folded domain similar to the NSF/p97/PEX1 N-C subdomain ( Figure 2B ) , and as such does not share these proteins' substrate-binding hydrophobic cleft . Nonetheless , the similarity in NTD structure indicates a hitherto unappreciated evolutionary link between Pch2/TRIP13 and the NSF/p97/PEX1 remodeler family ( Figure 2C ) , and strongly suggests that the PCH-2 NTD is involved in substrate recognition , either directly or indirectly through one or more adapter proteins . PCH-2's single AAA+ ATPase module is composed of two domains , termed the large and small AAA domains . Structural comparisons using the DALI server ( Holm and Rosenström , 2010 ) indicate that the ATP-binding large AAA domain of PCH-2 ( residues 100–323 ) is most structurally similar to the ‘classic remodelers , ’ including NSF/p97/PEX1 , Vps4 , Katanin p60 , and the proteasome ATPase subunits ( 2 . 7–3 . 4 Å r . m . s . d . comparing 150–170 Cα atoms ) . The domain also shows strong similarity to other AAA+ ATPase families including the ‘HCLR’ clade that includes ClpX , the unfoldase component of the bacterial ClpXP protease ( 3 . 2–3 . 3 Å r . m . s . d . comparing ∼150 Cα atoms ) ( Glynn et al . , 2009; Sauer and Baker , 2011 ) . The small AAA domain ( residues 324–424 ) is most similar to ‘classic remodeler’ family members ( 1 . 5–2 . 0 Å r . m . s . d . comparing 60–70 Cα atoms ) . As in other AAA+ ATPases , the PCH-2 hexamer assembles through interactions between each subunit's large AAA domain and the small AAA domain of a neighboring subunit , with the ATP-binding sites situated near the subunit interfaces ( Figure 2A , D ) . Although no nucleotides were added during purification or crystallization of PCH-2 , we observed that two subunits in the hexamer ( chains B and E ) are bound to ADP ( Figure 2A , E , Figure 2—figure supplement 1B ) , enabling a close analysis of PCH-2 active site structure . PCH-2 possesses the characteristic Walker A , Walker B , Sensor-1 , and ‘arginine finger’ motifs in the large AAA domain that cooperate to bind nucleotide ( Figures 1B , 2E ) ( Wendler et al . , 2012 ) . In many AAA+ ATPases , nucleotide binding is sensed by an additional ‘Sensor-2’ motif , typically an arginine residue , reaching from the small AAA domain into the active site . This motif is involved in nucleotide binding , hydrolysis , and nucleotide-dependent inter-domain conformational changes in various AAA+ ATPases ( Ogura et al . , 2004 ) . Curiously , the ‘classic remodelers’ family , including NSF/p97/PEX1 , uniformly lacks the Sensor-2 motif and also possesses a second arginine adjacent to the arginine finger ( Figure 1B ) . These differences indicate that this family's mechanism for ATP-driven conformational changes may have diverged somewhat from other AAA+ ATPases ( Ogura et al . , 2004; Erzberger and Berger , 2006 ) . Pch2/TRIP13 proteins , in contrast , possesses only a single arginine finger ( R312 in C . elegans PCH-2 ) , and our PCH-2 structure shows that a conserved arginine ( R385 ) is properly positioned to act as a Sensor-2 motif ( Figures 1B , 2E ) . This finding suggests that despite sharing a common NTD with a family of ‘classic remodelers’ , the detailed mechanism for nucleotide-dependent conformational changes in Pch2/TRIP13 may more closely resemble other AAA+ families . The distinctly asymmetric hexamer architecture of PCH-2 provides clues to conformational changes that likely occur during ATP binding , hydrolysis , and release . Within the hexamer , four PCH-2 subunits adopt a ‘closed’ conformation equivalent to that observed in most AAA+ ‘classic remodeler’ structures , with the large and small AAA domains tightly associated around the ATP-binding site . Two of these subunits ( chains B and E ) are bound to ADP in our structure , while the other two ( chains A and D ) contain a SO4− ion from the crystallization buffer . Asymmetry in the PCH-2 hexamer arises from large conformational differences in the remaining two subunits ( chains C and F , also bound to SO4− ) , situated on opposite ends of the extended hexamer . Compared to the four closed subunits , these chains adopt an ‘open’ conformation , in which the small AAA domain is rotated ∼70° away from the large AAA domain ( Figure 3A–B ) . 10 . 7554/eLife . 07367 . 008Figure 3 . Conformational changes within the PCH-2 hexamer . ( A ) Structural basis for nucleotide binding-dependent conformational changes . All six subunits ( A/D dark blue , B/E light blue , C/F green ) are overlaid based on their large AAA domains , and their small AAA domains are represented by a single α-helix , residues 323–342 . ( B ) Relative orientation of large and small AAA domains in different subunit types . Bound ADP and SO4− ions are shown as sticks . While the small AAA domain position varies widely between subunits , all six subunit–subunit interfaces are equivalent , forming six rigid-body units within the hexamer ( see Figure 3—figure supplement 1 ) . ( C ) ‘Closed’ ( blue ) and ‘open’ ( green ) ClpX monomers in the nucleotide-free ClpX hexamer ( PDB ID 3HTE; [Glynn et al . , 2009] ) . Later work showed that the ‘closed’ conformation is compatible with nucleotide binding ( Stinson et al . , 2013 ) . ( D ) Top view of the asymmetric PCH-2 hexamer , with subunits colored as in ( A ) and ( B ) , and pore loops ( residues 217–226 ) colored magenta . ( E ) Pore-side view of PCH-2 D/E/F chains ( A/B/C chains removed ) , showing the axial staggering of these subunits' pore loops . ( F ) Top view of the nucleotide-free ClpX hexamer ( Glynn et al . , 2009 ) , with ‘closed’ and ‘open’ subunits colored as in PCH-2 and pore loops ( residues 145–153 ) colored magenta . ( G ) Pore-side view of ClpX D/E/F chains ( A/B/C removed ) , colored as in ( F ) . ( H ) Sequence alignment of pore loop region in PCH-2 orthologs , and equivalent region of human p97 and NSF , and E . coli ClpX . Magenta box: PCH-2 pore loop; Yellow boxes: NSF ‘YVG’ and ClpX ‘GYVG’ motifs . ( I ) Schematic model for ATP-driven conformational changes in PCH-2 , with pore-side view equivalent to panel F . As the left-most subunit binds ATP ( blue; represented by the closed ‘ATP’-like state in chain D ) , hydrolyzes ATP to ADP ( light blue; represented by PCH-2 chain E ) , then releases hydrolyzed ADP ( green; represented by PCH-2 chain F ) , its pore loop ( magenta ) undergoes axial motions that drive substrate remodeling . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 00810 . 7554/eLife . 07367 . 009Figure 3—figure supplement 1 . The asymmetric PCH-2 hexamer is composed of equivalent rigid-body groups . ( A ) Overlaid large AAA domains of the six protein chains in the PCH-2 hexamer , as in Figure 3A , with associated small AAA+ domains from each neighboring subunit . While the small AAA domains of each subunit show significant conformational differences , the interface between each large AAA domain and the neighboring small AAA domain remains fixed . ( B ) Views as in panel ( A ) of each subunit type . Outlines in yellow , pink , and purple illustrate rigid-body rotation units within the hexamer , with NTDs removed for clarity ( see below ) . Each rigid body is proposed to rotate as a unit in response to the nucleotide-binding state of its ATPase active site . ( C ) Left: top view of the PCH-2 hexamer ( NTDs omitted for clarity ) , colored by protein chain as in Figure 3 . Center: PCH-2 hexamer colored as in ( A ) , with outlines in yellow , pink , and purple as in ( B ) highlighting rigid-body rotation units in the hexamer . Right: PCH-2 with rigid-body units colored as in ( B ) . Each rigid-body unit comprises the large AAA domain from one chain ( e . g . , chain B ) and the small AAA domain from its neighbor ( e . g . , chain A ) . Cyan ovals indicate the locations of ATPase active sites , positioned between each rigid-body unit . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 009 The dramatic conformational differences between subunits , and the resulting overall hexamer architecture of PCH-2 , are distinct from most existing structures of AAA+ ‘classic remodelers’ , which are typically either symmetric or display a subtle helical pitch resulting in a ‘lock-washer’ conformation , and usually lack the significant rotations between large and small AAA domains seen in PCH-2 ( Davies et al . , 2008; Lander et al . , 2012; Zhao et al . , 2015 ) . Instead , the PCH-2 structure closely resembles several prior structures of the ClpX unfoldase , which contain four ‘closed’ and two ‘open’ subunits arranged in the same pattern as in PCH-2 ( Figure 3C , F , G ) ( Glynn et al . , 2009; Stinson et al . , 2013 ) . This conformation was observed both in the absence of nucleotides ( Glynn et al . , 2009 ) and in the presence of a non-hydrolyzable ATP analog ( ATP-γ-S ) , which was found to bind the four ‘closed’ subunits but not the two ‘open’ subunits ( Stinson et al . , 2013 ) . Detailed biochemical analysis of the ClpX mechanism has demonstrated that sequential ATP binding , hydrolysis , and release drive cyclical open → closed → open conformational changes within each subunit ( Glynn et al . , 2009; Stinson et al . , 2013 ) . These motions in turn drive axial movement of loops lining the hexamer pore ( pore loops ) , which contain aromatic residues that directly engage substrate proteins during unfolding ( Siddiqui et al . , 2004; Iosefson et al . , 2015 ) . Despite a potentially diverged mechanism for ATP-powered conformational changes , the classic remodelers possess functionally equivalent pore loops ( Zhao et al . , 2010 ) . Within each half-hexamer of PCH-2 ( chains A/B/C and D/E/F ) , the pore loops ( residues 217–226; Figure 3H ) are axially staggered to create a ‘spiral staircase’ of likely substrate-engaging groups . We interpret these pore loop positions as representing structural intermediates adopted during ATP binding , hydrolysis , and release within each PCH-2 subunit that drive substrate remodeling ( Figure 3I ) . To test the physical mechanism of Pch2/TRIP13 and the roles of active-site and pore-loop residues , we measured nucleotide binding and hydrolysis by PCH-2 and its M . musculus ortholog TRIP13 . As in other AAA+ ATPases , a mutation in the Walker B motif of both PCH-2 and TRIP13 ( E253Q in both enzymes ) retains high-affinity nucleotide binding , while PCH-2 Walker A and Sensor-1 mutants do not bind nucleotide ( Figure 4A , B ) . We found that TRIP13E253Q also forms stable hexamers , in contrast to the predominantly monomeric wild-type enzyme ( Figure 1D ) . Mutation of PCH-2 R385 also results in the loss of nucleotide binding , illustrating that this residue is likely to be functionally analogous to the Sensor-2 motifs in other AAA+ ATPases . 10 . 7554/eLife . 07367 . 010Figure 4 . Nucleotide binding and hydrolysis by PCH-2 and TRIP13 . ( A ) Binding of PCH-2 active-site mutants to BODIPY-FL ATP . ( B ) Binding of M . musculus TRIP13E253Q to BODIPY-FL ATP . ( C ) Basal ATP hydrolysis rates of wild-type and mutant C . elegans PCH-2 at pH 8 . 5 ( optimal for ATPase activity; ATPase stimulation assays ( Figure 6 ) were performed at pH 7 . 5 , where basal activity is lower but stimulation is more robust ) . E253Q: Walker B ATPase mutant; W221A/F222A: pore loop mutants; WT ΔNTD: residues 100–424 . ( D ) Basal ATP hydrolysis rates of wild-type and mutant M . musculus TRIP13 at pH 8 . 5 . Residue numbering for mutants is identical to C . elegans PCH-2 . ( E ) Km/kcat values ( reported as ATP min−1 per hexameric enzyme ) for wild-type and mutant PCH-2 and TRIP13 . For PCH-2T186A , PCH-2N300A , and PCH-2R385A , rates were measured at a single ATP concentration of 2 mM , so Km was not determined ( N/D ) . For PCH-2E253Q and TRIP13E253Q , very low ATPase activity precluded a reliable Km determination ( N/A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 010 We used an enzyme-coupled assay to measure ATP hydrolysis by PCH-2 and TRIP13 . Both enzymes showed modest but reproducible ATPase activity , and mutation of conserved active site residues resulted in the complete loss of activity ( Figure 4C–E ) . We next tested the importance of the enzymes' pore loops , which in other AAA+ ATPases are necessary to properly couple ATP hydrolysis to substrate engagement ( Siddiqui et al . , 2004 ) . We generated alanine mutants of PCH-2 W221 and F222 , which align with the ClpX/NSF ‘YVG’/‘GYVG’ motifs ( Figure 3H ) , and also created a ‘pore loop AG’ mutant ( PCH-2AG ) in which residues 218-227 were replaced by an Ala-Gly linker of equal length . These mutants all formed soluble hexamers as in wild-type PCH-2 , but showed variable ATPase activity: while PCH-2F222A showed a fivefold reduction in basal ATPase activity , PCH-2W221A and PCH-2AG showed a striking increase in activity , with the kcat of PCH-2W221A around threefold higher than that of the wild-type enzyme ( Figure 4C , E ) . Similarly , M . musculus TRIP13W221A had a more than twofold higher basal kcat than wild-type TRIP13 ( Figure 4D , E ) . The variable effect of pore loop mutations in PCH-2/TRIP13 suggests that this element may play a role in coupling ATP hydrolysis to substrate engagement . Timely SAC inactivation relies on both TRIP13 and p31 ( comet ) , and prior work has shown that the two proteins can together dissociate MAD2:CDC20 complexes in vitro ( Eytan et al . , 2014 ) . As p31 ( comet ) is known to bind both MAD2 ( Habu et al . , 2002; Yang et al . , 2007 ) and TRIP13 ( Tipton et al . , 2012 ) , we reasoned that the protein may act as an adapter . We mapped sequence conservation onto the structure of human p31 ( comet ) :MAD2 ( Yang et al . , 2007 ) , and identified a highly conserved surface on p31 ( comet ) opposite its MAD2-binding interface ( Figure 5A ) . To test whether this surface , which includes residues on the p31 ( comet ) ‘safety-belt’ and a short loop bordering this motif , is responsible for TRIP13 interaction , we generated a series of mutations in M . musculus p31 ( comet ) . Several mutations to this conserved surface disrupted TRIP13 binding in a yeast two-hybrid assay , while retaining MAD2 binding ( Figure 5B ) . Conversely , a previously-characterized mutant at the MAD2 interface ( Yang et al . , 2007 ) disrupted MAD2 binding but did not affect the interaction with TRIP13 ( Figure 5B , D ) . To test whether p31 ( comet ) is able to simultaneously interact with MAD2 and TRIP13 , we performed a yeast three-hybrid assay . This assay showed an interaction between TRIP13 and MAD2 that depends on the presence of untagged p31 ( comet ) ( Figure 5C ) , showing that p31 ( comet ) can indeed function as an adapter between MAD2 and TRIP13 ( Figure 5G ) . 10 . 7554/eLife . 07367 . 011Figure 5 . p31 ( comet ) functions as an adapter between TRIP13 and MAD2 . ( A ) Two views of the crystal structure of human p31 ( comet ) ( colored by conservation ) bound to MAD2 ( yellow ) ( Yang et al . , 2007 ) . Residue numbers shown are of M . musculus p31 ( comet ) ( 76% identity with Homo sapiens p31 ( comet ) ; all noted residues are conserved ) . ( B ) Yeast two-hybrid assay for M . musculus p31 ( comet ) binding to TRIP13 and MAD2 . The p31 ( comet ) -MAD2 interaction can also be detected using purified proteins ( panel D ) . BD: Gal4 DNA-binding domain fusion; AD: Gal4 activation domain fusion . N/S: no selection; -HIS: selection for interaction between BD- and AD-fused proteins . ( C ) Yeast three-hybrid assay showing interaction of BD-TRIP13 and AD-MAD2 in the presence of untagged p31 ( comet ) . ( D ) Ni2+-pulldown assay using purified His6-tagged M . musculus p31 ( comet ) pulling down untagged M . musculus MAD2 . ( E ) Size exclusion chromatography traces and gels from M . musculus TRIP13E253Q ( yellow ) , p31 ( comet ) :C-MAD2R133A ( blue ) , and an equimolar mixture of TRIP13E253Q + p31 ( comet ) :C-MAD2R133A ( green ) . Quantitation of Coomassie-stained bands in bottom gel ( lanes 2 and 3 ) show an ∼6:1 molar ratio of TRIP13E253Q to p31 ( comet ) :MAD2 ( 6 copies of TRIP13 , 1 . 06 copies of p31 ( comet ) and 0 . 74 copies of C-MAD2R133A ) . See Figure 5—figure supplement 1 for full gels and analysis of different protein combinations including p31 ( comet ) mutants , and Figure 5—figure supplement 2 for purification of p31 ( comet ) and MAD2 . ( F ) SEC-MALS analysis of TRIP13E253Q ( yellow ) and TRIP13E253Q:p31 ( comet ) :C-MAD2R133A ( green ) . TRIP13E253Q migrates as a single peak with measured molecular weight of 251 kDa , close to the calculated hexamer molecular weight of 290 . 2 kDa . Upon the addition of p31 ( comet ) and C-MAD2R133A , the measured molecular weight shifts to 325 kDa . The shift of 74 kDa is close to the weight of a p31 ( comet ) :C-MAD2R133A complex ( 54 . 7 kDa ) . Excess p31 ( comet ) :C-MAD2R133A elutes after the complex with TRIP13 . ( G ) Schematic illustrating p31 ( comet ) functioning as an adapter between C-MAD2 ( via blue surface ) and TRIP13 ( via red surface ) . See Figure 8—figure supplement 1C for the p31 ( comet ) crystal structure colored equivalently . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 01110 . 7554/eLife . 07367 . 012Figure 5—figure supplement 1 . Interactions between M . musculus TRIP13 , p31 ( comet ) , and MAD2 . Size exclusion chromatography traces and gels from mixtures of separately purified M . musculus TRIP13E253Q , p31 ( comet ) , and C-MAD2R133A . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 01210 . 7554/eLife . 07367 . 013Figure 5—figure supplement 2 . Purification and characterization of M . musculus MAD2 and p31 ( comet ) . ( A ) Anion-exchange purification of M . musculus MAD2R133A . O-MAD2R133A and C-MAD2R133A peaks are noted . The MAD2 ‘loopless’ mutant ( residues 109–117 replaced by GSG ( adopts monomeric open form ) [Mapelli et al . , 2007] ) eluted at the same salt concentration as the O-MAD2R133A peak ( not shown ) . ( B ) SDS-PAGE analysis of fractions from ( A ) . ( C ) SEC-MALS analysis of M . musculus MAD2 . Wild-type MAD2 ( top , black ) is predominantly dimeric , while O-MAD2R133A ( green ) , C-MAD2R133A ( blue ) , and ‘loopless’ MAD2 are predominantly monomeric . ( D ) SDS-PAGE analysis of final purified MAD2 WT , O-MAD2R133A , C-MAD2R133A , and ‘loopless’ MAD2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 013 We next sought to reconstitute the TRIP13:p31 ( comet ) :MAD2 complex in vitro . We separately purified M . musculus TRIP13E253Q , which is catalytically inactive and forms more stable hexamers than wild-type TRIP13 ( Figure 1D ) , wild-type p31 ( comet ) , and a monomeric variant of MAD2 ( MAD2R133A ) in which open ( O-MAD2 ) and closed ( C-MAD2 ) monomers can be separately purified ( Figure 5—figure supplement 2 ) ( Sironi et al . , 2001; Luo et al . , 2004; Mapelli et al . , 2006 ) . We mixed TRIP13E253Q , p31 ( comet ) , and C-MAD2R133A in the presence of ATP , and measured complex formation using size-exclusion chromatography . In agreement with prior studies ( Yang et al . , 2007 ) , p31 ( comet ) and C-MAD2R133A formed a stable heterodimeric complex . When this complex was pre-incubated with TRIP13E253Q , a small amount of both proteins was shifted into the TRIP13 hexamer peak ( Figure 5E , Figure 5—figure supplement 1 ) . Semi-quantitative analysis of Coomassie-stained gels , and light-scattering based molecular-weight measurements on this peak , revealed that for each TRIP13E235Q hexamer , about 1 copy of p31 ( comet ) :C-MAD2R133A was shifted into the TRIP13 peak ( Figure 5E , F , Figure 5—figure supplement 1 ) . MAD2 did not shift in the absence of p31 ( comet ) , and only a very small amount of p31 ( comet ) shifted in the absence of MAD2 ( Figure 5—figure supplement 1 ) . p31 ( comet ) mutants that disrupt either MAD2 or TRIP13 binding also largely eliminated p31 ( comet ) :MAD2 co-migration with TRIP13E253Q ( Figure 5—figure supplement 1 ) . Taken together with our yeast two-hybrid results , these data suggest that a single p31 ( comet ) :MAD2 complex associates with a TRIP13 hexamer , first through transient TRIP13-p31 ( comet ) binding ( likely mediated by the TRIP13 NTD ) , then through direct interactions between TRIP13 and MAD2 ( at the TRIP13 hexamer pore ) . We were unable to directly test the role of the TRIP13 pore loops in binding , however , as the TRIP13W221A/E253Q double mutant did not form stable hexamers ( data not shown ) . We next examined whether PCH-2 or TRIP13 ATPase activity is stimulated by MAD2 , p31 ( comet ) , or the p31 ( comet ) :MAD2 complex . While PCH-2 ATPase activity was mostly unaffected by the addition of either MAD-2 or the recently-identified p31 ( comet ) ortholog CMT-1 ( Vleugel et al . , 2012 ) , it was modestly stimulated in the presence of both proteins ( Figure 6A ) . A PCH-2 construct missing the NTD was not stimulated by the addition of MAD-2 + CMT-1 , supporting the proposed role for this domain in substrate recognition ( Figure 6A ) . The PCH-2W221A pore-loop mutant was also not stimulated by MAD-2 + CMT-1 , supporting the idea that this mutant uncouples ATP hydrolysis from productive substrate engagement ( Figure 6A ) . 10 . 7554/eLife . 07367 . 014Figure 6 . ATP hydrolysis in PCH-2/TRIP13 is stimulated by p31 ( comet ) + MAD2 . ( A ) Stimulation of C . elegans PCH-2 ATPase activity by CMT-1 ( p31 ( comet ) ) and MAD-2 . See Figure 6—figure supplement 1 for purification of C . elegans CMT-1 and MAD-2 . Cdc20: N-terminal MBP fusion of C . elegans FZY-1 residues 98-140 . Substrates were in sixfold molar excess of PCH-2 hexamer . ( B ) Stimulation of M . musculus TRIP13 ATPase activity by p31 ( comet ) and MAD2 . WT: wild-type MAD2 dimer; ‘open’: O-MAD2R133A monomer; ‘closed’: C-MAD2R133A monomer; ‘loopless’: residues 109–117 replaced by GSG ( adopts monomeric open form ) ( Mapelli et al . , 2007 ) . CDC20: N-terminal MBP fusion of CDC20 residues 111–150 , sufficient for MAD2 binding ( Luo et al . , 2000 ) . Substrates were in sixfold molar excess of TRIP13 hexamer . ( C ) Stimulation of TRIP13 ATPase activity in the presence of p31 ( comet ) mutants . ( D ) Schematic illustrating requirements for TRIP13 stimulation . PCH-2/TRIP13 is stimulated by the combination of C-MAD2 and p31 ( comet ) ( scheme 1 ) ; mutation of either binding surface of p31 ( comet ) ( schemes 2 and 3 ) eliminates stimulation , as does replacement of C-MAD2 with O-MAD2 ( scheme 4 ) . p-values in ( A ) and ( C ) were calculated using an unpaired Student's T-test . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 01410 . 7554/eLife . 07367 . 015Figure 6—figure supplement 1 . Purification and characterization of C . elegans MAD-2 and CMT-1 . ( A ) SEC-MALS analysis of C . elegans MAD-2 shows a mixture of monomer and dimer states . These were separately pooled for PCH-2 ATPase stimulation assays; while both pools stimulated PCH-2 in the presence of CMT-1 , the dimer peak more strongly stimulated PCH-2 ( not shown ) . ( B ) SDS-PAGE analysis of purified C . elegans MAD-2 . ( C ) Size exclusion chromatography and SDS-PAGE analysis of C . elegans CMT-1 , showing that the protein is monomeric ( elution volume of size standards shown for comparison ) . ( D ) SDS-PAGE analysis of fractions from ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 015 M . musculus TRIP13 was strongly stimulated by the addition of MAD2 + p31 ( comet ) ( Figure 6B ) . Stimulation did not depend on the presence of CDC20 , but was highly sensitive to MAD2 conformation: while C-MAD2R133A stimulated TRIP13 equivalently to wild-type MAD2 , O-MAD2R133A showed minimal stimulation , and the locked-open ‘loopless’ MAD2 mutant ( Mapelli et al . , 2007 ) showed no stimulation ( Figure 6B ) . As with PCH-2 , neither MAD2 nor p31 ( comet ) alone had a stimulatory effect on TRIP13 . Also consistent with PCH-2 , the TRIP13W221A pore-loop mutant showed a very high basal level of ATP hydrolysis , and was not further stimulated by MAD2 + p31 ( comet ) ( Figure 6B ) . Finally , we found that p31 ( comet ) mutations that disrupt either MAD2 or TRIP13 binding also significantly reduce stimulation of TRIP13 by p31 ( comet ) + MAD2 ( Figure 6C , D ) . The above results suggest a model in which p31 ( comet ) recognizes C-MAD2 and delivers it to TRIP13 for remodeling . If TRIP13 functions by unfolding MAD2 , or alternatively converting C-MAD2 to O-MAD2 , a prolonged incubation with TRIP13 should eliminate the stimulatory effect of p31 ( comet ) + MAD2 . This is indeed the case: we found that the ability of p31 ( comet ) + MAD2 to stimulate TRIP13 ATPase activity was almost eliminated after a 2-hr pre-incubation , in an ATP-dependent manner ( Figure 7A , compare samples #2 , 3 , and 4 ) . Importantly , p31 ( comet ) and MAD2 were in sixfold molar excess to TRIP13 hexamers in this assay , illustrating that each TRIP13 hexamer acted on multiple p31 ( comet ) :MAD2 complexes during the pre-incubation period . Addition of fresh MAD2 to the pre-incubated samples rescued TRIP13 stimulation ( Figure 7A , sample #5 ) . We interpret these findings to indicate that TRIP13 is stimulated by the p31 ( comet ) :MAD2 complex but acts on MAD2 specifically , likely converting C-MAD2 into the non-stimulatory open state ( Figure 7B ) . 10 . 7554/eLife . 07367 . 016Figure 7 . TRIP13 converts C-MAD2 to O-MAD2 . ( A ) Stimulation of TRIP13 ATPase activity by p31 ( comet ) + MAD2 , before ( samples 1–2 ) or after ( samples 3–5 ) a 2-hr pre-incubation period . All proteins were at 4 μM ( sixfold molar excess of substrate:TRIP13 hexamer ) . For sample 5 , an additional 4 μM MAD2 was added after the pre-incubation period . ( B ) Schematic illustrating results from ( A ) in terms of complex formation and TRIP13 stimulation . ( C ) Anion-exchange elution profiles for O-MAD2 ( green ) , C-MAD2 ( gray ) , and the p31 ( comet ) :C-MAD2 complex ( blue ) . ( D ) p31 ( comet ) and MAD2 in anion-exchange fractions from the indicated pre-incubated reaction mixtures . p31 ( comet ) and MAD2 were at 30 μM , and TRIP13 at 5 μM ( hexamer concentration ) except for starred sample ( third from top ) , where TRIP13 was at 1 . 25 μM . The observed C-MAD2 to O-MAD2 conversion requires active TRIP13 , ATP , and p31 ( comet ) . Neither TRIP13E253Q nor TRIP13W221A supported MAD2 conversion . At equilibrium , MAD2R133A is predominantly in the C-MAD2 state ( Figure 7—figure supplement 1 ) , further supporting that the observed C-MAD2 to O-MAD2 conversion is an active process . ( E ) Rate of TRIP13-mediated C-MAD2 to O-MAD2 conversion in limiting TRIP13 . Reactions with 30 μM p31 ( comet ) + MAD2R133A ( initially ∼10 μM O-MAD2 and ∼20 μM C-MAD2 ) were incubated at 37°C for 30 min with the indicated amounts of TRIP13 , and separated by ion-exchange as in ( D ) . SDS-PAGE band intensities were quantified , converted to [O-MAD2] , and plotted . Linear regression fitting indicates a rate of 57 . 4 ± 6 . 7 MAD2 conversions in 30 min per TRIP13 hexamer , or ∼1 . 9 ± 0 . 2 min−1 ( F ) TRIP13-mediated MAD2 conversion in the presence of p31 ( comet ) mutants . The high concentration of p31 ( comet ) + MAD2 in this assay ( 30 μM ) allowed single mutants to support limited MAD2 conversion , but mutant combinations effectively eliminated MAD2 conversion . PK: P230A/K231A; QF: Q86A/F193A . ( G ) ATPase activity of M . musculus TRIP13 at 37°C ( all other ATPase assays were performed at 27°C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 01610 . 7554/eLife . 07367 . 017Figure 7—figure supplement 1 . MAD2R133A is predominantly in the C-MAD2 state at equilibrium . ( A ) Purified samples of O-MAD2R133A ( 30 μM total concentration ) were separated by anion-exchange chromatography as in Figure 7C–D , either as purified ( 0 hr ) or after a 24 hr incubation at 37°C . Quantitation of SDS-PAGE bands in fractions 2–3 ( O-MAD2 ) vs fractions 5–6 ( C-MAD2 ) indicate that the sample is 13% C-MAD2 prior to incubation , and 87% C-MAD2 after incubation . ( B ) Purified samples of C-MAD2R133A were analyzed as in ( A ) . The sample is 64% C-MAD2 prior to incubation , and 85% C-MAD2 after incubation . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 017 To directly assay whether TRIP13 converts C-MAD2 to O-MAD2 , we took advantage of the fact that in the dimerization-defective MAD2R133A mutant , the open and closed conformers are separable by anion-exchange chromatography ( Figure 7C , Figure 5—figure supplement 2A ) ( Luo et al . , 2004 ) . When we incubated p31 ( comet ) + C-MAD2R133A with TRIP13 , the p31 ( comet ) :C-MAD2 complex that initially forms was dissociated , and C-MAD2R133A was converted to the open conformer ( Figure 7D ) . This conversion depended on active TRIP13 , ATP , and p31 ( comet ) . Despite a high rate of ATP hydrolysis , the TRIP13W221A pore-loop mutant was unable to catalyze MAD2 conversion , illustrating that pore loop integrity is critical for MAD2 conformational conversion . To directly measure the catalytic activity of TRIP13 , we titrated the enzyme and monitored C-MAD2R133A to O-MAD2R133A conversion . At 37°C , a single TRIP13 hexamer catalyzes the conversion of 1 . 9 ± 0 . 2 MAD2 molecules per minute ( Figure 7E ) . Combining this measurement with TRIP13's fully-stimulated ATPase activity in these conditions ( 16 . 7 ± 0 . 3 ATP min−1 per hexamer; Figure 7G ) , we estimate that TRIP13 hydrolyzes 8–10 ATP's per MAD2 conformational conversion . This number is similar to several prior measurements of NSF-mediated SNARE complex disassembly ( 6–50 ATP per event , depending on experimental conditions ) ( Cipriano et al . , 2013; Ryu et al . , 2015; Shah et al . , 2015 ) or ClpX-mediated unfolding of a small model substrate ( ∼150 ATP ) ( Burton et al . , 2001 ) , likely reflecting that MAD2 conformational conversion may require only a local perturbation of the safety-belt motif , rather than complete unfolding ( see ‘Discussion’ ) . This idea would fit with recent work indicating that NSF unfolds its substrates in a single step , using the energy from multiple ATP hydrolysis events to build up tension within the hexamer , then promoting a critical conformational change that results in SNARE complex disassembly ( Ryu et al . , 2015 ) . A similar ‘spring-loaded’ mechanism in TRIP13 could enable the enzyme to catalyze conversion/disassembly of unliganded C-MAD2 or its complexes with partner proteins such as CDC20 with similar efficiency; alternatively , if MAD2 conformational conversion requires several rounds of ATP hydrolysis , the energy requirements for disassembly of ligand-stabilized MAD2 may be significantly higher than the 8–10 ATP we measured for C-MAD2R133A conformational conversion . Finally , we tested the ability of p31 ( comet ) mutants that disrupt binding to either MAD2 or TRIP13 to support MAD2 conformational conversion . The MAD2-binding mutant ( Q86A/F193A ) and two TRIP13-binding mutants ( Δ159–164 or P230A/K231A ) each modestly reduced MAD2 conversion compared to wild-type p31 ( comet ) ( Figure 7F ) . Combinations of these mutants , however , almost completely eliminated MAD2 conversion ( Figure 7F ) , illustrating that the adapter function of p31 ( comet ) is critical for TRIP13 to recognize and convert C-MAD2 to O-MAD2 .
Our data show that Pch2/TRIP13 is a AAA+ ATPase with structural and mechanistic properties similar to both the ‘classic remodelers’ and the bacterial protein unfoldase ClpX . First , the PCH-2 hexamer architecture shows close mechanistic parallels with ClpX , suggesting a shared mechanism for nucleotide-dependent conformational changes driving pore loop motions and substrate remodeling . Our structural and biochemical data suggest that Pch2/TRIP13 engages its substrates within the hexamer pore , and undergoes coordinated ATP hydrolysis-coupled conformational changes to mediate substrate unfolding . In contrast to ClpX , however , Pch2/TRIP13 does not completely unfold its HORMA domain protein substrates . Rather , we have shown that TRIP13 catalyzes a much more subtle structural change , converting closed MAD2 to its open state . We propose that TRIP13 specifically unfolds the C-terminal safety belt region of MAD2 , then allows it to refold into the open state . Given the mechanistic similarities between Pch2/TRIP13 and the processive unfoldase ClpX , however , an obvious question is how unfolding by Pch2/TRIP13 is controlled to achieve HORMA domain conformational conversion instead of complete unfolding . The answer to this question may lie in the second aspect of Pch2/TRIP13's hybrid nature: its mode of substrate recognition , which is mediated by an NTD related to a family of ‘classic remodelers’ including NSF , p97 , and PEX1 . Importantly , as in NSF and p97 , substrate recognition by Pch2/TRIP13 is indirect with p31 ( comet ) acting as an adapter to deliver MAD2 to TRIP13 . p31 ( comet ) binds specifically to C-MAD2 , meaning that once TRIP13 engages and unfolds the MAD2 safety belt , p31 ( comet ) would cease to bind MAD2 . This could destabilize the ternary complex , releasing partially-unfolded MAD2 and allowing its re-folding into the open conformation . As mentioned above , MAD2 safety belt unfolding could occur in a processive manner accompanied by multiple rounds of ATP hydrolysis by TRIP13 , or could occur similarly to NSF , where recent work has indicated a single-step ‘spring-loaded’ mechanism for SNARE complex disassembly ( Ryu et al . , 2015 ) . In the SAC , we propose that p31 ( comet ) and TRIP13 catalyze a two-step MCC disassembly mechanism to inactivate the SAC ( Figure 8A , B ) . First , p31 ( comet ) displaces BUBR1 from MAD2 , potentially causing its dissociation from MAD2:CDC20 . The resulting p31 ( comet ) :MAD2:CDC20 complex is then recognized by TRIP13 , which converts C-MAD2 to O-MAD2 , thus disrupting binding to both p31 ( comet ) and CDC20 , and also preventing MCC re-assembly . 10 . 7554/eLife . 07367 . 018Figure 8 . Model for SAC inactivation by p31 ( comet ) and TRIP13 . ( A ) Unattached kinetochores catalyze the assembly of the mitotic checkpoint complex ( MCC ) through the conversion of O-MAD2 to C-MAD2 and assembly with CDC20 ( blue ) , BUBR1 ( pink ) , and BUB3 ( not shown ) . ( B ) After kinetochore-microtubule attachment , MCC assembly is halted . p31 ( comet ) binds existing MCC and displaces BUBR1 , then delivers C-MAD2:CDC20 to TRIP13 for conformational conversion and disassembly . The CDC20:BUBR1 interaction may be disrupted directly by p31 ( comet ) or at a later point . ( C ) Scheme for TRIP13-mediated disassembly of HORMAD oligomers ( blue ) in meiosis . It is unknown whether HORMADs possess an open state analogous to O-MAD2 . See Figure 8—figure supplement 1 for structures of O-MAD2 , C-MAD2 , C-HORMAD , and p31 ( comet ) showing the safety-belt conformation in each state . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 01810 . 7554/eLife . 07367 . 019Figure 8—figure supplement 1 . HORMA domain structures in open and closed conformation . ( A ) Structures of human MAD2 in the open ( left panel; PDB ID 1DUJ; [Luo et al . , 2000] ) , closed ( middle panel; PDB ID 1S2H [Luo et al . , 2004] ) , and closed/peptide bound ( right panel; PDB ID 1KLQ [Luo et al . , 2002] ) conformations , with MAD2 in yellow , MAD2 safety belt ( residues 160:205 ) colored dark green , and the MAD2-binding peptide ( MBP-1 ) in blue . MAD2 residues 109–117 , replaced with ‘GSG’ in loopless MAD2 ( Mapelli et al . , 2007 ) , are shown in blue in the O-MAD2 structure ( left panel ) . ( B ) Structure of a meiotic HORMAD in the closed conformation ( C . elegans HIM-3 bound to a peptide from HTP-3; PDB ID 4TZJ [Kim et al . , 2014] ) . The safety belt is colored dark green as in panel ( A ) . ( C ) Structure of human p31 ( comet ) ( PDB ID 2QYF [Yang et al . , 2007] ) , with safety-belt motif ( closed around its own C-terminus ) shown in red , and the MAD2 binding surface ( with residues 86 and 193 ( numbering according to M . musculus p31 ( comet ) ) ) shown in blue . Side-chains for residues involved in TRIP13 binding ( 159–164 and 230–231 ) are shown in stick view . Cartoon representations of all structures are shown for reference . DOI: http://dx . doi . org/10 . 7554/eLife . 07367 . 019 Essentially all free MAD2 in HeLa cells is in the open state ( Luo et al . , 2004 ) , but prior work ( Luo et al . , 2004 ) and our own analysis ( Figure 7—figure supplement 1 ) indicates that while the two conformations are relatively stable at 4°C , at physiological temperatures essentially all O-MAD2 spontaneously converts to C-MAD2 within several hours . These data strongly suggest that cellular factors actively maintain MAD2 in the open state . We propose that a major role for TRIP13 and p31 ( comet ) may be to counteract spontaneous O-MAD2 to C-MAD2 conversion , thus guarding against improperly-timed MCC assembly ( which can occur outside mitosis given a supply of soluble C-MAD2 [Tipton et al . , 2011a] ) and also ensuring a sufficient supply of O-MAD2 for SAC activation in prometaphase . An important remaining question is how the competing pathways for MCC assembly and disassembly are balanced and regulated throughout the cell cycle: do p31 ( comet ) and TRIP13 constantly disassemble MCC at a low level during prometaphase and metaphase , with this activity becoming dominant only after new MCC assembly is ceased , or is the activity of p31 ( comet ) and TRIP13 suppressed during metaphase by additional mechanisms ? Recently it was shown that human p31 ( comet ) is phosphorylated specifically in mitosis , and that phosphorylation lowers the affinity of p31 ( comet ) for MAD2 ( Date et al . , 2014 ) . While the phosphorylated residue ( Ser102 ) is not universally conserved in p31 ( comet ) orthologs , this result nonetheless represents one potential mechanism for suppressing TRIP13-mediated MCC disassembly specifically during mitosis . The ability of TRIP13 to disengage the safety-belt motif of MAD2 strongly suggests a parallel mechanism for its remodeling/removal of HORMAD proteins along chromosomes in meiotic prophase ( Figure 8C ) . We have previously shown that the meiotic HORMADs assemble into hierarchical head-to-tail complexes through safety-belt interactions , and that these interactions are crucial for meiotic DNA break formation , inter-homolog recombination , and chromosome segregation ( Kim et al . , 2014 ) . We have been unable , however , to detect direct interactions between PCH-2/TRIP13 and their putative HORMAD substrates , nor do these proteins stimulate PCH-2/TRIP13 ATPase activity ( not shown ) . Thus , how the enzyme recognizes HORMAD complexes , whether a p31 ( comet ) -like adapter is needed for this recognition , and what signals coordinate crossover formation with HORMAD complex remodeling and removal , remain important open questions . Finally , given the additional association of human TRIP13 with a number of cancer types ( Larkin et al . , 2012; van Kester et al . , 2012; Banerjee et al . , 2014 ) , addressing the fundamental mechanistic questions regarding how this enzyme recognizes and remodels its substrates will be important for understanding TRIP13's multiple roles in human health and disease .
For sequence analysis of AAA+ ATPases , isolated AAA+ regions ( large plus small domains , isolated D1 domain for p97/Cdc48 and NSF ) were aligned with MAFFT ( Katoh and Standley , 2013 ) , a phylogenetic tree was constructed in JalView ( Waterhouse et al . , 2009 ) , and the tree was visualized with Dendroscope ( Huson and Scornavacca , 2012 ) . Full-length C . elegans PCH-2 and M . musculus TRIP13 were cloned from cDNA into a bacterial expression vector with an N-terminal TEV protease-cleavable His6 tag . Mutant constructs were generated by PCR-based mutagenesis: C . elegans PCH-2 ΔNTD consisted of residues 100–424 , and the ‘Pore loop AG’ mutant replaced residues 218–227 with the protein sequence ‘AGAAGAAAGA’ . All PCH-2/TRIP13 mutants used for activity assays were expressed at levels similar to the wild-type proteins and migrated equivalently on a size-exclusion column , indicating that they are soluble and folded . The Walker A motif mutant K185Q of both PCH-2 or TRIP13 , and the ΔNTD mutant of TRIP13 , were not solubly expressed , precluding their analysis . For C . elegans MAD-2 ( MDF-2 ) and CMT-1 ( C41D11 . 5 ) , and M . musculus MAD2 and p31 ( comet ) , full-length proteins were cloned from cDNA into a bacterial expression vector with an N-terminal TEV protease-cleavable His6 tag . Mutant constructs were generated by PCR-based mutagenesis: M . musculus MAD2 ‘loopless’ replaced residues 109–117 with the protein sequence ‘GSG’ as in ( Mapelli et al . , 2007 ) ( see Figure 8—figure supplement 1A ) . For identification of highly conserved surface residues in p31 ( comet ) , 226 animal/plant p31 ( comet ) sequences were aligned with MAFFT ( Katoh and Standley , 2013 ) and conservation was mapped on the structure of p31 ( comet ) bound to MAD2 ( Yang et al . , 2007 ) using the CONSURF server ( Ashkenazy et al . , 2010 ) . All mutant constructs were generated by PCR-based mutagenesis . All mutant constructs used here ( p31 ( comet ) , MAD2 , and TRIP13 ) expressed at levels similar to wild-type , and migrated equivalently on a size-exclusion column ( not shown ) , indicating that they were soluble and folded . Proteins were expressed in Escherichia coli strain Rosetta 2 ( DE3 ) pLysS ( EMD Millipore , Billerica MA ) at 20°C for 16 hr , then cells were harvested by centrifugation and resuspended in buffer A ( 25 mM Tris pH 7 . 5 , 10% glycerol , 5 mM MgCl2 ) plus 300 mM NaCl , 5 mM imidazole , and 5 mM β-mercaptoethanol . Protein was purified by Ni2+-affinity ( Ni-NTA agarose , Qiagen ) then ion-exchange ( Hitrap Q HP , GE Life Sciences , Piscataway NJ ) chromatography . Tags were cleaved with TEV protease ( Tropea et al . , 2009 ) , and cleaved protein was passed over a size exclusion column ( Superdex 200 , GE Life Sciences ) in buffer A plus 300 mM NaCl and 1 mM dithiothreitol ( DTT ) . Purified protein was concentrated by ultrafiltration ( Amicon Ultra , EMD Millipore ) to ∼10 mg/ml and stored at 4°C . For selenomethionine derivatization of PCH-2 , protein expression was carried out in M9 minimal media supplemented with amino acids plus selenomethionine prior to IPTG induction ( Van Duyne et al . , 1993 ) , and proteins were exchanged into buffer containing 1 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) after purification to maintain the selenomethionine residues in the reduced state . For size-exclusion chromatography coupled multi-angle light scattering ( SEC-MALS ) , proteins were separated on a Superdex 200 10/300 GL size exclusion column ( GE Life Sciences ) , their light scattering and refractive index profiles collected by miniDAWN TREOS and Optilab T-rEX detectors ( Wyatt Technology , Santa Barbara CA ) , respectively , and their molecular weights calculated using ASTRA v . 6 software ( Wyatt Technology ) . For negative-stain EM , PCH-2 was passed over a size-exclusion column in EM buffer ( buffer A without glycerol , and with added 1 mM DTT ) , then peak fractions were diluted to ∼0 . 01 mg/ml in EM buffer with or without 1 mM ATP . Samples were applied onto freshly glow discharged carbon coated copper grids , reduced to a thin film by blotting , and a 2% solution of uranyl formate negative stain was then immediately applied to the grid and blotted off from the same side . The negative stain process was repeated 3 times . Data were acquired using a Tecnai F20 Twin transmission electron microscope ( FEI , Hillbsboro OR ) operating at 200 kV . Images were automatically collected using the LEGINON system ( Suloway et al . , 2005 ) . Images were recorded using a Tietz F416 4k × 4k pixel CMOS camera ( TVIPS , Gauting , Germany ) . Experimental data were processed by the APPION software package ( Lander et al . , 2009 ) . The defoci were estimated using ctffind3 ( Mindell and Grigorieff , 2003 ) and ACE2 ( Lander et al . , 2009 ) and CTF correction was done by phase flipping the whole micrograph . Particles were selected automatically in APPION using DogPicker ( Roseman , 2004; Voss et al . , 2009 ) . After stack creation in APPION all datasets were prealigned and classified using 2-D maximum likelihood procedures and multivariate statistical analysis as implemented in XMIPP and IMAGIC ( Radermacher et al . , 1986; van Heel et al . , 1996; Sorzano et al . , 2004; Scheres et al . , 2008 ) . Resulting class averages were manually inspected and classes that represented noise or distorted particles were rejected . Final clustering was performed using XMIPP clustering 2D alignment ( cl2d ) . For crystallization , PCH-2 was exchanged into crystallization buffer ( 25 mM Tris pH 7 . 5 , 5 mM MgCl2 , 200 mM NaCl , 1 mM tris ( 2-carboxyethyl ) phosphine [TCEP] ) , either without added nucleotide ( Apo ) or with 1 mM ADP or non-hydrolyzable ATP analogs ( ATP-γS or AMP-PCP ) . Regardless of added nucleotides , PCH-2 formed large prism-shaped crystals after mixing 1:1 with 100 mM sodium citrate pH 5 . 6 , 200 mM ammonium sulfate , and 15% PEG 3350 . Crystals were cryoprotected by the addition of 20% glycerol , and flash-frozen in liquid nitrogen . Diffraction data were collected at synchrotron sources ( see Table 1 ) , and processed with HKL2000 ( Otwinowski and Minor , 1997 ) or XDS ( Kabsch , 2010 ) . All crystals were in space group C2221 , with one PCH-2 hexamer per asymmetric unit . The structure was determined using phases obtained from a single-wavelength anomalous diffraction ( SAD ) dataset from a crystal grown from selenomethionine-derivatized protein . Automated XDS → SHELX → PHENIX for the SAD dataset was performed by the RAPD data-processing pipeline at the Advanced Photon Source NE-CAT beamline 24ID-E ( https://rapd . nec . aps . anl . gov/rapd ) . 46 selenomethionine sites were identified using SHELX as implemented in hkl2map ( Sheldrick , 2010 ) , then supplied to the AutoBuild module of PHENIX ( Terwilliger et al . , 2009; Adams et al . , 2010 ) , which located an addition 22 sites ( for 68 total—66 sites would be expected for a hexamer of PCH-2 , but in several cases two sites represented alternate rotamers for a single methionine residue ) , and calculated and refined phases using PHASER ( McCoy et al . , 2007 ) and RESOLVE ( Terwilliger et al . , 2009 ) ( http://www . solve . lanl . gov ) . Initial sequence-threaded models of the large and small AAA domains were generated by the PHYRE2 server ( Kelley and Sternberg , 2009 ) ( http://www . sbg . bio . ic . ac . uk/phyre2 ) and manually placed to generate an initial model . Initial placement and refinement of this model allowed identification of twofold non-crystallographic symmetry , which was then used during early-stage map generation , model building , and refinement ( final refinement was performed without non-crystallophic symmetry ) . Numerous rounds of manual rebuilding in Coot ( Emsley et al . , 2010 ) and refinement in phenix . refine ( Adams et al . , 2010 ) against a high resolution native dataset resulted in improved maps , allowing us to manually build the NTDs . Data were highly anisotropic , showing significantly lower intensity ( I/σ ) and half–set correlation ( CC1/2 ) ( Karplus and Diederichs , 2012 ) along the c* axis than along a* and b* ( Table 1 and Figure 2—figure supplement 1A ) . For refinement , the high-resolution native dataset processed to 2 . 3 Å with XDS was submitted to the UCLA Diffraction Anisotropy Server ( Strong et al . , 2006 ) ( http://services . mbi . ucla . edu/anisoscale/ ) for application of anisotropic cutoffs ( 2 . 3 Å along a* and b* , 3 . 2 Å along c* ) . The final model consists of six PCH-2 monomers , with a total of 2214 residues modeled out of 2544 ( 6 × 424 residues ) ; the model displays good geometry with 98 . 05% of residues in favored , and 99 . 77% of residues in allowed Ramachandran space ( Table 1 ) . All crystallographic software was installed and maintained through the SBGrid program ( Morin et al . , 2013 ) . For yeast two-hybrid analysis , full-length sequences for M . musculus TRIP13 , MAD2 , and p31 ( comet ) ( wild-type and mutants ) were cloned into pGADT7 ( Gal4 activation domain fusion: ‘AD’ ) and pBridge ( Gal4 DNA binding domain fusion: ‘BD’ ) vectors ( Clontech Laboratories , Mountain View CA ) . Plasmids were transformed into AH109 and Y187 yeast strains , and transformants selected on CSM -Leu ( pGADT7 ) or CSM -Trp ( pBridge ) media . Haploid strains were mated overnight at room temperature , and diploids were selected on CSM -Leu-Trp media . Diploids were then patched onto CSM -Leu-Trp-His ( low stringency; shown in Figure 5B–C ) or CSM -Leu-Trp-His-Ade ( high stringency; not shown , results consistent with low-stringency results ) media , grown 3 days at 30°C , and imaged . For yeast three-hybrid analysis , p31 ( comet ) was cloned into multiple cloning site #2 of pBridge to express the untagged protein alongside the BD- and AD-fusion proteins . For Ni2+ pulldown assays , 300 picomoles ( 9 . 3 μg ) His6-tagged p31 ( comet ) was mixed with 450 picomoles untagged MAD2 ( 10 . 7 μg ) in 50 μl binding buffer ( 20 mM Tris-HCl pH 7 . 5 or 8 . 5 , 200 mM NaCl , 20 mM imidazole , 1 mM β-mercaptoethanol , 5% glycerol , 0 . 1% NP-40 ) , incubated 60 min 20°C , then ‘load’ samples ( 5 μl , 10% ) were removed and samples were mixed with Ni-NTA magnetic beads ( 10 μl 5% suspension , Qiagen , Hilden , Germany ) for 20 min 20°C . Samples were washed 3× with 1 ml binding buffer , then 25 μl SDS-PAGE loading buffer was added , samples were boiled , run on 12 . 5% SDS-PAGE gels and imaged by Coomassie staining . For size-exclusion chromatography analysis of TRIP13 plus p31 ( comet ) :MAD2 , equimolar amounts ( 10 nanomoles ) of TRIP13E253Q , p31 ( comet ) , and C-MAD2R133A were mixed in 300 μl total volume of gel filtration buffer ( 20 mM Tris-HCl pH 7 . 5 , 300 mM NaCl , 10% Glycerol , 1 mM DTT ) plus 2 mM ATP , and incubated on ice for 30 min before application on a size exclusion column ( Superdex 200 Increase 10/300 GL , GE Life Sciences ) in gel filtration buffer plus 0 . 1 mM ATP . For SEC-MALS analysis of selected complexes , the same protocol was followed except for addition of nucleotide to the column running buffer . For ATPase assays with C . elegans PCH-2 , optimal basal ATPase rates were obtained from protein treated during the Ni2+-affinity purification step with 0 . 8 M urea , which removes the two ADP molecules bound to each hexamer ( as determined by UV absorbance; not shown ) , followed by addition of 50 mM ammonium sulfate to all subsequent purification steps ( necessary for protein stability after ADP removal ) . ATPase activity was determined at 27°C ( except where indicated ) using an enzyme-coupled assay ( Nørby , 1988 ) adapted for a microplate reader ( Kiianitsa et al . , 2003 ) . 100 μl reactions contained assay buffer ( 25 mM Tris-HCl at pH 7 . 5 or 8 . 5 ( see below ) , 200 mM NaCl , 10 mM MgCl2 , 1 mM DTT , 5% glycerol ) plus 2 mM ATP , 3 mM phosphoenolpyruvate , 20 U/ml lactate dehydrogenase ( Sigma Aldrich , St . Louis MO ) , 20 U/ml pyruvate kinase ( Sigma Aldrich ) , and 0 . 3 mM NADH . All PCH-2/TRIP13 constructs showed a strong dependence on pH with almost undetectable activity at pH 7 . 0 and below , and full activity at pH 8 . 5–9 . 5 . For assays measuring stimulation of ATPase activity by p31 ( comet ) and MAD2 , assays were performed at pH 7 . 5 where basal activity was lower but substrate stimulation was robust . The assayed concentration of each PCH-2/TRIP13 construct was adjusted between 0 . 5 and 20 μM monomer , for the most accurate measurement of ATPase activity , depending on the assay . For TRIP13 , which adopts multiple oligomeric states in solution , we verified that the ATP hydrolysis rate varies linearly between 0 . 625 and 10 μM TRIP13 ( monomer concentration ) at both pH 7 . 5 and 8 . 5 , indicating that the protein is predominantly hexameric ( and thus fully active ) in our ATPase assay conditions ( data not shown ) . Unless otherwise indicated , TRIP13 was equimolar with added p31 ( comet ) and MAD2 ( judging by TRIP13 monomer concentration ) . The decline of NADH absorbance at 340 nm was measured using a TECAN ( Mannedorf , Switzerland ) Infinite M1000 spectrophotometer in 384-well microplates . NADH oxidation rate was calculated from a linear fit to each time course and converted to ATP hydrolysis rates . For calculation of Km and kcat , sampled were performed in triplicate with ATP concentration varying from 20 μM ( lower bound for measuring decline in NADH absorbance ) to 1 mM , and data were fit to the Michaelis-Menten equation ( Y = ( Vmax × X ) / ( Km + X ) ) using PRISM v . 6 ( GraphPad Software , La Jolla CA ) . For samples measuring TRIP13 ATPase activity after pre-incubation , proteins ( 4 μM concentration for all proteins ) were pre-incubated for 2 hr at 20°C in assay buffer with or without ATP as above . Samples were then passed through a desalting spin column ( Zeba-Spin , Thermo Scientific , Waltham MA ) to remove remaining ATP and hydrolyzed ADP . Fresh ATP and coupled-assay master mix were then added and ATP hydrolysis measured as above . To examine MAD2 conformational conversion by TRIP13 , separately purified C-MAD2 monomer ( R133A mutant ) , p31 ( comet ) , and TRIP13 were incubated at 20°C for 2 hr at 30 μM concentration ( sixfold molar excess of p31 ( comet ) and MAD2 to TRIP13 hexamers ) , in ATPase assay buffer ( pH 7 . 5 ) with or without 2 mM ATP ( 166 μl reaction volume ) . Samples were diluted to 50 mM NaCl by the addition of buffer without NaCl , then loaded onto a 1 ml HiTrap Q HP column ( GE Life Sciences ) and eluted with a gradient to 400 mM NaCl . Fractions were collected , analyzed by SDS-PAGE and visualized by Coomassie staining . For measurement of MAD2 conversion rate by TRIP13 ( Figure 7E , F ) , pre-incubations were performed for 30 min at 37°C with 30 μM p31 ( comet ) , 30 μM MAD2 enriched for C-MAD2 ( approximately 10 μM O-MAD2 and 20 μM C-MAD2 ) , and the indicated amounts of TRIP13 . Samples were separated by ion-exchange chromatography , and quantitation of Coomassie blue-stained SDS-PAGE bands was performed . Background-subtracted relative intensities of O-MAD2 vs C-MAD2 ( lanes 6–7 ) was performed using ImageJ ( Schneider et al . , 2012 ) , and ratios were converted to quantities based on the total ( MAD2 ) of 30 μM . ( O-MAD2 ) was plotted vs ( TRIP13 hexamer ) in the range of ( TRIP13 ) where the reaction was not saturated ( up to 0 . 2 μM TRIP13 hexamer ) . Linear regression fitting was performed with PRISM v . 6 ( GraphPad Software ) . | The genetic material inside human and other animal cells is made of DNA and is packaged in structures called chromosomes . Before a cell divides , the entire set of chromosomes is copied so that each chromosome is now made of two identical sister ‘chromatids’ . Next , the chromosomes line up on a structure called the spindle , which is made of filaments called microtubules . Cells have a surveillance system known as the spindle assembly checkpoint that halts cell division until every chromosome is correctly aligned on the spindle . Once the chromosomes are in place , the checkpoint is turned off and the spindle pulls the chromatids apart so that each daughter cell receives a complete set of chromosomes . A protein called MAD2 plays an important role in the spindle assembly checkpoint . It can adopt two distinct shapes: in the ‘closed’ shape it is active and halts cell division , but in the ‘open’ shape it is inactive and allows cell division to proceed . Another protein called TRIP13 can help turn off the checkpoint , but it is not clear how this works or whether TRIP13 acts on MAD2 directly . Here , Ye et al . studied these proteins using a technique called X-ray crystallography and several biochemical techniques . The experiments show that TRIP13 belongs to a family of proteins known as ‘AAA-ATPases’ , which can unfold proteins to alter their activity . Ye et al . found that TRIP13 binds to an adaptor protein that allows it to bind to the closed form of MAD2 . TRIP13 then unfolds a part of the MAD2 protein , converting MAD2 into the open shape . Ye et al . propose that , once all chromosomes are lined up on the spindle , TRIP13 turns off the spindle assembly checkpoint by converting closed MAD2 to open MAD2 . Also , when cells are not undergoing cell division , TRIP13 may maintain MAD2 in the open shape to prevent cells from turning on the spindle assembly checkpoint at the wrong time . Further work will be needed to show how TRIP13 recognizes the closed form of MAD2 , and whether it can act in a similar way on other proteins in the cell . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | TRIP13 is a protein-remodeling AAA+ ATPase that catalyzes MAD2 conformation switching |
Haem and iron homeostasis in most eukaryotic cells is based on a balanced flux between haem biosynthesis and haem oxygenase-mediated degradation . Unlike most eukaryotes , ticks possess an incomplete haem biosynthetic pathway and , together with other ( non-haematophagous ) mites , lack a gene encoding haem oxygenase . We demonstrated , by membrane feeding , that ticks do not acquire bioavailable iron from haemoglobin-derived haem . However , ticks require dietary haemoglobin as an exogenous source of haem since , feeding with haemoglobin-depleted serum led to aborted embryogenesis . Supplementation of serum with haemoglobin fully restored egg fertility . Surprisingly , haemoglobin could be completely substituted by serum proteins for the provision of amino-acids in vitellogenesis . Acquired haem is distributed by haemolymph carrier protein ( s ) and sequestered by vitellins in the developing oocytes . This work extends , substantially , current knowledge of haem auxotrophy in ticks and underscores the importance of haem and iron metabolism as rational targets for anti-tick interventions .
Haem , the heterocyclic tetrapyrrole that conjugates divalent iron , is an essential molecule for most aerobic organisms , as a prosthetic group of numerous enzymes involved in a variety of biological processes such as cellular respiration , detoxification of xenobiotics or redox homeostasis ( Furuyama et al . , 2007; Kořený et al . , 2013 ) . Most organisms synthesise their own haem by an evolutionarily conserved multi-enzymatic pathway occurring in the mitochondria and cytosol . Only a few haem auxotrophs lacking functional haem biosynthesis have been described to date . Among these rare organisms that are reliant on the acquisition of exogenous haem are , for instance , a protozoan parasitic apicomplexan Babesia bovis ( Brayton et al . , 2007 ) , and kinetoplastid flagellates of the genus Trypanosoma and Leishmania ( Kořený et al . , 2010 ) . Some haem auxotrophs , such as the filarial nematode parasite Brugia malayi ( Ghedin et al . , 2007; Wu et al . , 2009 ) , acquire haem from their endosymbionts , while others , such as the free-living nematode Caenorhabditis elegans ( Rao et al . , 2005 ) obtain haem from ingested bacteria . The inability to synthesise haem de novo was also biochemically demonstrated for the cattle tick Rhipicephalus ( Boophilus ) microplus ( Braz et al . , 1999 ) . In contrast to its benefits , haem is also cytotoxic , where free haem catalyses the generation of reactive oxygen species ( ROS ) , causing cellular damage , mainly through lipid peroxidation ( Jeney et al . , 2002; Klouche et al . , 2004; Graca-Souza et al . , 2006 ) . Therefore , in all living organisms , free intracellular haem has to be maintained at a low level via strictly regulated homeostasis ( Ryter and Tyrrell , 2000; Khan and Quigley , 2011 ) . This task is a critical challenge for haematophagous parasites , such as the malarial Plasmodium , blood flukes or triatominae insects that acquire large amounts of haem from digested haemoglobin ( Oliveira et al . , 2000; Pagola et al . , 2000; Paiva-Silva et al . , 2006; Toh et al . , 2010 ) . Maintenance of haem balance is even more demanding for ticks , as their blood meal exceeds their own weight more than one hundred times ( Sonenshine and Roe , 2014 ) . Despite its importance , the knowledge of haem acquisition , inter-tissue transport and further utilisation in ticks is fairly limited . Haemoglobin , the abundant source of haem for these animals , is processed intracellularly in tick gut digest cells by a network of cysteine and aspartic peptidases ( Sojka et al . , 2013 ) . Excessive haem is detoxified by aggregation in specialised organelles termed haemosomes ( Lara et al . , 2003; 2005 ) and its movement from digestive vesicles is mediated by a recently described ATP-binding cassette transporter ( Lara et al . , 2015 ) . Only a small proportion of acquired haem is destined for systemic distribution to meet the metabolic demands of tick tissues ( Maya-Monteiro et al . , 2000 ) . In the present work , we have screened available tick and mite genomic databases and found that ticks have lost most genes encoding the haem biosynthetic pathway . All mites also commonly lack genes coding for haem oxygenase ( HO ) that catalyzes haem catabolism , raising the question of iron source for these organisms . Using in vitro membrane feeding of the hard tick Ixodes ricinus ( Kröber and Guerin , 2007 ) , the European vector of Lyme disease and tick-borne encephalitis , we performed differential feeding of females on haemoglobin-rich and haemoglobin-depleted diets . These experiments conclusively proved that ticks completely rely on the supply of exogenous haem to accomplish successful reproduction and that iron required for metabolic processes in tick tissues does not originate from haem . We propose that the unique maintenance of systemic and intracellular haem homeostasis in ticks represents a specific adaptation to their parasitic life style , and as such offers promising targets for anti-tick intervention .
The availability of the genome-wide database for the deer tick Ixodes scapularis ( Gulia-Nuss et al . , 2016 ) made it possible to analyse the overall genetic make-up for enzymes possibly participating in haem biosynthesis and compare this data with other mites and insects ( Hexapoda ) . Complete haem biosynthetic and degradative pathways are present in insects , represented by the genomes of the fruit fly Drosophila melanogaster ( Adams et al . , 2000 ) and the blood-feeding malaria mosquito , Anopheles gambiae ( Holt et al . , 2002 ) ( Figure 1A , B ) . The canonical haem biosynthetic pathway is also fully conserved in the genomes of the herbivorous mite Tetranychus urticae , and the predatory mite Metaseiulus occidentalis , but is substantially reduced in the genome of the obligatory blood-feeding tick , I . scapularis ( Figure 1B ) . The tick genome contains only genes encoding the last three mitochondrial enzymes of haem biosynthesis , namely , coproporphyrinogen-III oxidase ( CPOX , [Vectorbase: ISCW010977] , Figure 1—figure supplement 1 ) , protoporphyrinogen oxidase ( PPOX , [Vectorbase: ISCW023396 , Figure 1—figure supplement 2 ) , and ferrochelatase ( FECH , [Vectorbase: ISCW016187] , Figure 1—figure supplement 3 ) . Corresponding orthologues could be also found in the I . ricinus transcriptome ( Kotsyfakis et al . , 2015 ) ( GenBank Ac . Nos JAB79008 , JAB84046 and JAB74800 , respectively ) . Phylogenetic analyses confirmed that these genes cluster together with other Acari homologues ( Figure 1—figure supplements 1–3 , respectively ) . Another two gene sequences related to 5-aminolevulinate synthase ( ALAS , Vectorbase: ISCW020754 ) and uroporphyrinogen decarboxylase ( UROD , Vectorbase: ISCW020804 ) are clearly bacterial and most likely originate from bacterial contamination of the genomic DNA ( Figure 1—figure supplement 4 and Figure 1—figure supplement 5 , respectively ) . This conclusion was further corroborated by the fact that these genes do not contain introns and are flanked by other bacterial genes in the corresponding genomic regions . 10 . 7554/eLife . 12318 . 003Figure 1 . Evolution of haem biosynthetic and degradative pathways . ( A ) General scheme of haem biosynthetic and degradative pathways in the eukaryotic cell . Haem biosynthesis ( upper ) is a series of eight reactions beginning in the mitochondria by condensation of succinyl coenzyme A with glycine , continuing in the cell cytoplasm , and finishing in the mitochondria with the final synthesis of the haem molecule . Haem degradation ( lower ) is mediated by haem oxygenase in the cell cytoplasm , releasing a ferrous iron , biliverdin , and carbon monoxide . ( B ) Evolution of haem biosynthetic and degradative pathways in arthropods , according to the available genomic projects . Similarly to vertebrates , hexapods ( insects ) including blood feeding mosquitoes ( red-coloured body ) , possess all enzymes for haem biosynthesis and degradation . Chelicerates lack haem oxygenase , indicating iron acquisition from sources other than haem . Plant-feeding mites ( green-coloured body ) of the superorder Acariformes , as well as mite-predating mites ( black-coloured body ) of the superorder Parasitiformes , possess a complete set of genes for haem biosynthesis . Ticks , which feed solely on blood ( red-coloured body ) retained only the last three enzymes ( mitochondrial ) of the pathway . CO - carbon monoxide , Fe2+ - ferrous iron , Gly - glycine , Suc-CoA - succinyl coenzyme A , ALAS - 5-aminolevulinate synthase , PBGS - porphobilinogen synthase , HMBS - hydroxymethylbilane synthase , UROS - uroporphyrinogen synthase , UROD - uroporphyrinogen decarboxylase , CPOX - coproporphyrinogen oxidase , PPOX - protoporphyrinogen oxidase , FECH - ferrochelatase; HO - haem oxygenase . Enzyme nomenclature and abbreviations according to ( Hamza and Dailey , 2012 ) DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 00310 . 7554/eLife . 12318 . 004Figure 1—figure supplement 1 . Phylogenetic tree of selected coproporphyrinogen-III oxidases . Unrooted tree of coproporphyrinogen-III oxidase ( CPOX ) amino acid sequences reconstructed using the Neighbor Joining method ( NJ ) based on alignment using ClustalX . The Ixodes scapularis and Ixodes ricinus CPOXs are distant from bacterial , but also from vertebrate , and invertebrate homologues , whose phylogenies cannot be clearly resolved ( low bootstrap ) . Red dots indicate CPOX of ticks and green dots indicate CPOX of other chelicerates . Numbers at branches represent bootstrap supports using NJ criteria with 1000 replicates . The horizontal bar represents a distance of 0 . 05 substitutions per site . R . ricketsii ( Rickettsia rickettsii , bacteria , WP_012151472 ) , E . coli ( Escherichia coli , bacteria , WP_001625620 ) , T . castaneum ( Tribolium castaneum , red flour beetle , XP_008201513 ) , C . gigas ( Crassostrea gigas , pacific oyster , EKC32626 ) , H . sapiens ( Homo sapiens , ENSG00000080819 ) , D . melanogaster ( Drosophila melanogaster , fruitfly , FBgn0021944 ) , A . gambiae ( Anopheles gambiae , malaria mosquito , AGAP004749 ) , S . mimosarum ( Stegodyphus mimosarum , social spider , KFM71890 ) , M . occidentalis ( Metaseiulus occidentalis , western predatory mite , XP_003744828 ) , T . urticae ( Tetranychus urticae , two-spotted spider mite , tetur04g09527 ) , I . ricinus ( Ixodes ricinus , castor been tick , JAB79008 ) , I . scapularis ( Ixodes scapularis , deer tick , ISCW010977 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 00410 . 7554/eLife . 12318 . 005Figure 1—figure supplement 2 . Phylogenetic tree of selected protoporphyrinogen oxidases . Unrooted tree of protoporphyrinogen oxidase ( PPOX ) amino acid sequences reconstructed using the Neighbor Joining method ( NJ ) based on alignment using ClustalX . The Ixodes scapularis , Ixodes ricinus , vertebrate , and invertebrate PPOXs , whose phylogenies cannot be clearly resolved ( low bootstraps ) are distant from bacterial homologues . Red dots indicate CPOX of ticks and green dots indicate CPOX of other chelicerates . Numbers at branches represent bootstrap supports using NJ criteria with 1000 replicates . The horizontal bar represents a distance of 0 . 1 substitutions per site . H . sapiens ( Homo sapiens , ENSG00000143224 ) , D . melanogaster ( Drosophila melanogaseter , fruitfly , FBgn0020018 ) , A . gambiae ( Anopheles gambiae , malaria mosquito , AGAP003704 ) , I . ricinus ( Ixodes ricinus , castor been tick , JAB84046 ) , I . scapularis ( Ixodes scapularis , deer tick , ISCW023396 ) , M . occidentalis ( Metaseiulus occidentalis , western predatory mite , XP_003740594 ) , T . urticae ( Tetranychus urticae , two-spotted spider mite , tetur10g04900 ) , S . mimosarum ( Stegodyphus mimosarum , social spider , KFM82234 ) , S . pneumoniae ( Streptococcus pneumoniae , bacteria , CGG00621 ) , B . subtilis ( Bacillus subtilis , bacteria , WP_032725328 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 00510 . 7554/eLife . 12318 . 006Figure 1—figure supplement 3 . Phylogenetic tree of selected ferrochelatases . Unrooted tree of ferrochelatase ( FECH ) amino acid sequences reconstructed using the Neighbor Joining method ( NJ ) based on alignment using ClustalX . The Ixodes scapularis and Ixodes ricinus FECHs clusters together with other chelicerate homologues . Red dots indicate CPOX of ticks and green dots indicate CPOX of other chelicerates . Numbers at branches represent bootstrap supports using NJ criteria with 1000 replicates . The horizontal bar represents a distance of 0 . 1 substitutions per site . H . sapiens ( Homo sapiens , ENSG00000066926 ) , C . gigas ( Crassostrea gigas , pacific oyster , EKC30122 ) , D . melanogaster ( Drosophila melanogaster , fruitfly , FBgn0266268 ) , T . castaneum ( Tribolium castaneum , red flour beetle , XP_008193416 ) , A . gambiae ( Anopheles gambiae , malaria mosquito , AGAP003719 ) , I . ricinus ( Ixodes ricinus , castor been tick , JAB74800 ) , I . scapularis ( Ixodes scapularis , deer tick , ISCW016187 ) , T . urticae ( Tetranychus urticae , two-spotted spider mite , tetur04g02210 ) , M . occidentalis ( Metaseiulus occidentalis , western predatory mite , XP_003748486 ) , R . rickettsii ( Rickettsia rickettsii , bacteria , WP_012262655 ) , E . coli ( Escherichia coli , bacteria , ACI87485 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 00610 . 7554/eLife . 12318 . 007Figure 1—figure supplement 4 . Phylogenetic tree of selected 5-aminolevulinate synthases . Unrooted tree of 5-aminolevulinate synthase ( ALAS ) amino acid sequences reconstructed using the Neighbor Joining method ( NJ ) based on alignment using ClustalX . The ISCW020754 annotated in the Ixodes scapularis genome as a putative serine palmitoyltransferase is clearly a bacterial gene , homologous to Rickettsia , symbionts of ticks . Red dot indicates ISCW020754 sequence from the tick genome , green dots indicate chelicerate ALASs . Numbers at branches represent bootstrap supports using NJ criteria with 1000 replicates . The horizontal bar represents a distance of 0 . 05 substitutions per site . H . sapiens ( Homo sapiens , CAA42916 ) , D . melanogaster ( Drosophila melanogaster , fruitfly , CAA74915 ) , A . gambiae ( Anopheles gambiae , malaria mosquito , AGAP003184 ) , T . castaneum ( Tribolium castaneum , red flour beetle , TC013340 ) , L . polyphemus ( Limulus polyphemus , atlantic horseshoe crab , AAD20805 ) , S . mimosarum ( Stegodyphus mimosarum , social spider , KFM81891 ) , M . occidentalis ( Metaseiulus occidentalis , western predatory mite , XP_003744200 ) , T . urticae ( Tetranychus urticae , two-spotted spider mite , tetur32g00320 ) , ISCW020754 ( annotated Ixodes scapularis gene , deer tick , ISCW020754 ) , R . ricketsii ( Rickettsia rickettsii , bacteria , WP_014363330 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 00710 . 7554/eLife . 12318 . 008Figure 1—figure supplement 5 . Phylogenetic tree of selected uroporphyrinogen decarboxylases . Unrooted tree of uroporphyrinogen decarboxylase ( UROD ) amino acid sequences reconstructed using the Neighbor Joining method ( NJ ) based on alignment using ClustalX . The ISCW020804 annotated in the Ixodes scapularis genome is clearly a bacterial gene homologous to Rickettsia , symbionts of ticks . Red dot indicates ISCW020804 sequence from the tick genome , green dots indicate chelicerate URODs . Numbers at branches represent bootstrap supports using NJ criteria with 1000 replicates . The horizontal bar represents a distance of 0 . 05 substitutions per site . H . sapiens ( Homo sapiens , NP_000365 ) , T . urticae ( Tetranychus urticae , two-spotted spider mite , tetur19g03090 ) , M . occidentalis ( Metaseiulus occidentalis , western predatory mite , XP_003740745 ) , T . castaneum ( Tribolium castaneum , red flour beetle , XP_972457 ) , A . gambiae ( Anopheles gambiae , malaria mosquito , XP_320631 ) , D . melanogaster ( Drosophila melanogaster , fruitfly , ACH92415 ) , I . scapularis ( Ixodes scapularis , deer tick , ISCW020804 ) , R . ricketsii ( Rickettsia rickettsii , bacteria , WP_014362650 ) , E . coli ( Escherichia coli , bacteria , WP_000137647 ) , S . aureus ( Staphylococcus aureus , bacteria , KLN00580 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 008 Despite an incomplete haem biosynthetic pathway , the I . scapularis genome contains at least 225 genes encoding a variety of enzymes utilizing haem as a cofactor , such as respiratory chain cytochromes , catalase , and a large family of cytochrome P450 genes ( Supplementary file 1 ) . Hence , ticks must possess efficient mechanisms for the acquisition of exogenous haem , together with its intra- and extra-cellular transport to produce endogenous haemoproteins . In order to determine the origin of haem required for tick basal metabolism and development , we exploited an in vitro membrane feeding system developed by Kröber and Guerin ( Kröber and Guerin , 2007 ) . We fed I . ricinus females with whole blood ( BF ticks ) , and , in parallel , with haemoglobin-free serum ( SF ticks ) ( Figure 2 and Figure 2—figure supplement 1 ) . Serum-fed ticks were capable of fully engorging and laying eggs similar to BF ticks ( Figure 2 ) . However , striking differences were observed in embryonic development and larval hatching . Embryos in eggs laid by BF females developed normally as described for naturally-fed ticks ( Santos et al . , 2013 ) and gave rise to living larvae ( Figure 2 ) . In contrast , no embryonic development was observed in colourless eggs laid by SF ticks , and accordingly , no larvae hatched from these eggs ( Figure 2 ) . To prove that haemoglobin alone , and no other component of red blood cells , is required for successful tick development , a rescue experiment was performed . From the fifth day of membrane feeding ( prior to the females commencing the rapid engorgement phase ) , the serum diet was supplemented with 10% , 1% , or 0 . 1% pure bovine haemoglobin and ticks were allowed to complete feeding ( S+Hb-F ticks ) . The presence of haemoglobin in the diet rescued the competence of embryos to develop normally and the number of larvae hatching from eggs laid by S+Hb-F ticks was comparable with BF ticks ( Figure 2 , bottom panels ) . The same rescue effect was observed for ticks fed on 1% and 0 . 1% haemoglobin ( Figure 2—figure supplement 2 ) demonstrating that as little as one hundredth of the physiological concentration of haemoglobin in the diet is sufficient to maintain tick reproduction . 10 . 7554/eLife . 12318 . 009Figure 2 . Impact of dietary haemoglobin on tick feeding , oviposition , embryogenesis , and larval hatching . ( Membrane feeding ) - membrane feeding in vitro of Ixodes ricinus females on whole blood ( Blood-fed ) , serum ( Serum-fed ) and on serum supplemented with 10% bovine haemoglobin ( Serum + 10% Hb ) . For dietary composition , see Figure 2—figure supplement 1 . ( Oviposition ) – representative females laying eggs . ( Embryogenesis ) – microscopic examinations of embryonal development in eggs laid by differentially fed females; 1w , 2w – 1 week , 2 weeks after oviposition , respectively . Note , no embryos developed in eggs from serum-fed ticks , while embryogenesis was rescued in serum + 10% Hb-fed ticks . ( Larval hatching ) – Laid eggs were incubated to allow larval hatching . Note , no larvae hatched out of eggs laid by serum-fed females and the hatching was fully rescued in serum + 10% Hb-fed ticks . Similar rescue effects were also observed for ticks fed on serum supplemented with 1% and 0 . 1% Hb ( see Figure 2—figure supplement 2 ) DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 00910 . 7554/eLife . 12318 . 010Figure 2—figure supplement 1 . Diets used for tick membrane feeding and faecal examination . ( A ) Females of I . ricinus were membrane fed until full engorgement ( FE ) using whole blood , serum , and serum supplemented with 10% ( physiological concentration ) of pure bovine haemoglobin ( Serum+Hb ) . ( B ) Composition of diets . Equal levels of haemoglobin in whole blood and reconstituted Serum+Hb were verified by spectrophotometry ( absorbance at ~ 400 nm - Soret peak ) and by SDS-PAGE of diets ( arrow points to haemoglobin band ) . ( C ) Faecal examination . To ensure complete passage of Hb through the digestive tract , faeces were inspected 12 hr after serum supplementation with Hb . Examination of faecal extracts by spectrophotometry ( absorbance at ~ 400 nm - Soret peak ) and by SDS PAGE confirmed the availability of supplemented Hb before ticks commence a rapid engorgement phase ( ‘big sip’ ) . Note that the protein profile of faeces was almost identical to that of the applied meal . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 01010 . 7554/eLife . 12318 . 011Figure 2—figure supplement 2 . Rescue experiments with sub-physiological levels of haemoglobin . Embryonal development and larval hatching was fully rescued in I . ricinus females fed on serum supplemented with 1% or 0 . 1% bovine haemoglobin . ( Embryogenesis ) – microscopic examination of embryonal development in eggs laid by differentially fed females; 1w , 2w – 1 week , 2 weeks after oviposition , respectively . ( Larval hatching ) – Laid eggs were incubated to allow larval hatching . Note that tick reproduction was not affected if only one hundredth of the natural haemoglobin concentration was present in the tick diet . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 011 After a blood meal , the physiology of an adult female tick is dominated by its reproductive effort as up to half of the weight of a fully engorged female is used in the production of thousands of eggs ( Sonenshine and Roe , 2014 ) . To disclose the importance of haemoglobin in tick reproduction , we first determined haem levels in eggs obtained from both BF and SF ticks . The concentration of haem b ( the form of haem present in haemoglobin ) was determined by reverse-phase HPLC ( Figure 3A and Figure 3—figure supplement 1 ) . Eggs laid by BF ticks contained 669 ± 45 pmol haem b/mg eggs , whereas eggs laid by SF ticks contained virtually no haem ( only 3 ± 1 . 6 pmol haem b/mg eggs ) . Eggs from the rescue experiment ( S+Hb-F ticks ) contained only slightly decreased haem levels ( 508 ± 79 pmol haem b/mg eggs ) compared to BF ticks . Eggs from ticks fed with sub-physiological levels ( 1% and 0 . 1% ) of haemoglobin contained gradually decreasing haem levels ( 471 ± 17 and 229 ± 97 pmol haem b /mg eggs , respectively ) ( Figure 3A ) , but were still capable completing development and producing viable larvae ( Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 12318 . 012Figure 3 . Determination of haemoglobin-derived nutrients in ticks ( haem , amino acids , iron ) . ( A ) Levels of haem b were determined by HPLC in egg homogenates from ticks fed on whole blood ( BF ) serum ( SF ) , and serum supplemented with 10% , 1% or 0 . 1% bovine haemoglobin ( S+10%Hb , S+1%Hb and S+0 . 1% Hb , respectively; rescue experiments ) . Data ( mean values ± SEM ) were acquired from homogenates of three independent clutches of eggs . Representative chromatograms detecting haem b in egg homogenates are shown for BF ticks , SF ticks , and S+10% Hb - fed ticks , see Figure 3—figure supplement 1 . ( B ) Quantitative Western blot analyses detecting levels of vitellin 1 and vitellin 2 in egg homogenates using antibodies raised against vitellin precursors - vitellogenins ( IrVg1 , IrVg2 ) . Bar charts depict the mean levels ± SEM of the particular vitellin in the egg homogenates from three different clutches of BF ticks or SF ticks ( see Figure 3—figure supplement 2 ) . Representative Western blot detection is shown below the bar chart . ( C ) Quantitative Western blot analyses detecting ferritin1 ( IrFer1 ) in the gut , ovary , and salivary gland homogenates from BF and SF ticks . Bar charts depict the mean ± SEM levels of IrFer1 in the tissue homogenates prepared from three independent tissue pools ( see also Figure 3—figure supplement 2 ) . Representative Western blot detections for guts , ovaries and salivary glands are shown below the bar charts . ( D ) GF-AAS elemental analysis of iron in ovaries and salivary glands pools . Each data point represents a pool of five tissues dissected from BF and SF partially engorged ticks ( fed for 6 days ) . Iron content is expressed in ppm ( ng Fe per mg of dry tissue ) . Main and error bars indicate group means and SEM , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 01210 . 7554/eLife . 12318 . 013Figure 3—figure supplement 1 . HPLC analysis of haem b in tick egg homogenates . Homogenates prepared from 10 mg of eggs were collected from three independent egg clutches laid by I . ricinus females fed on different diets . Representative chromatograms are shown detecting haem b in egg homogenates of ticks fed on the whole blood ( BF ) and haemoglobin-free serum ( SF ) and serum supplemented with 10% of haemoglobin ( S+10% Hb ) . The inset shows the zoom-in of haem b detection in SF ticks; note the different y-axis scales . For details , see Material and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 01310 . 7554/eLife . 12318 . 014Figure 3—figure supplement 2 . Full appearance of SDS-PAGE and Western blot analyses shown in the Figure 3 . SDS-PAGE analyses were carried out on homogenates prepared from three independent tissue preparations from ticks fed on whole blood ( BF ) or serum ( SF ) . Protein profiles were visualized using the TGX Stain-Free technology ( TGX ) and the BioRad ChemiDoc MP imager . ( A ) SDS-PAGE of homogenates of freshly laid eggs , and Western blot analyses detecting levels of vitellin 1 ( Vn1 ) and vitellin 2 ( Vn2 ) using specific antibodies against vitellogenin 1 ( αIrVg1 ) and vitellogenin 2 ( αIrVg2 ) . ( B ) Control SDS-PAGE ( upper panel ) and corresponding Western blot ( lower panel ) for identification of ferritin 1 ( Fer1 ) by RNAi . Western blot analysis of Fer1 levels was performed using specific antibodies against recombinant I . ricinus ferritin1 ( αIrFer1 ) . No cross-reacting band was present in the whole blood or serum diet . The Fer1-specific band was clearly present in gut homogenates from naturally fed ticks , pre-injected with gfp dsRNA ( gfp ) but completely absent in ticks pre-injected with ferritin 1 dsRNA ( fer1-KD ) . RNAi-mediated silencing of iron-regulatory protein ( irp-KD ) caused a marked Fer1 up-regulation . ( C–E ) SDS-PAGE ( upper panels ) and corresponding Western blots ( lower panels ) used for quantification of Fer1 levels in tissue homogenates dissected from partially engorged BF and SF ticks ( fed for 6 days ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 01410 . 7554/eLife . 12318 . 015Figure 3—figure supplement 3 . Detection of biliverdin IX derivatives in Ixodes ricinus and Aedes aegypti . The HPLC using a diode array detector was set to enable a simultaneous determination of haem b and biliverdin IX compounds at wavelengths of 375 nm and 660 nm , respectively . For details , see Materials and methods . ( A ) I . ricinus gut extracts from fully engorged females 5 days after detachment from the host . ( B ) I . ricinus gut extracts from fully engorged females 5 days after detachment from the host and spiked with 50 pmol of biliverdin IX standard . ( C ) Whole body extracts from naturally fed Aedes aegypti mosquitoes allowed to digest blood for three days were used as a positive control . The presence of biglutaminyl-biliverdin IX as a haem b degradation product ( Pereira et al . , 2007 ) was detected . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 015 Vitellins , the major tick egg yolk proteins , account for more than 90% of the protein content of a mature egg ( James and Oliver , 1997; Logullo et al . , 2002 ) . In contrast to haem concentrations , no apparent differences were observed in vitellin levels in eggs from BF and SF ticks , as determined by quantitative Western blot analysis ( Figure 3B ) with specific antibodies raised against recombinant vitellin precursors , I . ricinus vitellogenin 1 ( IrVg1 ) and vitellogenin 2 ( IrVg2 ) ( Supplementary file 2; Figure 3—figure supplement 2 ) . This result implies that haemoglobin is replaceable by serum proteins as a nutritional source of amino acids needed for vitellogenesis . Genome-wide analyses of I . scapularis and other mites revealed a common unique feature; the gene encoding haem oxygenase ( HO ) is missing , pointing to a lack of enzymatic degradation of haem in these Acari representatives ( Figure 1B ) . HO-mediated haem degradation results in the equimolar release of iron and the linear tetrapyrrole product , biliverdin ( Khan and Quigley , 2011 ) . Gut homogenates from fully engorged I . ricinus females were analysed by HPLC for the presence of biliverdin IX ( Figure 3—figure supplement 3 ) . With the detection limit as low as 5 pmol , no trace of biliverdin IX or modified biliverdin showing a bilin-like light absorbance near 660nm was detected in I . ricinus gut homogenates . In contrast to ticks , the presence of biglutamyl biliverdin IX in whole body extracts of the blood-fed mosquito , Aedes aegypti ( Pereira et al . , 2007 ) , was confirmed by our method exploiting diode-array detection ( Figure 3—figure supplement 3 ) . The lack of HO thus poses a question of the iron source for ticks . Iron availability in tick tissues was examined using two independent methods: ( i ) The presence of iron was indirectly tested by monitoring the levels of intracellular Ferritin 1 ( IrFer1 ) . Under iron deficiency , the translation of ir-fer1mRNA is suppressed by binding of the iron regulatory protein ( IRP1 ) to its 5’-located iron-responsive element , whereas at high iron levels , the proteosynthesis of IrFer1 is up-regulated ( Kopáček et al . , 2003; Hajdusek et al . , 2009 ) . Homogenates of guts , ovaries , and salivary glands were analysed by quantitative Western blotting using IrFer1-specific antibody ( Figure 3C and Figure 3—figure supplement 2 ) . IrFer1 levels were lower in guts and about equal in ovaries and salivary glands of BF compared to SF ticks ( Figure 3C and Figure 3—figure supplement 2 ) ; ( ii ) The elemental iron concentration in tick tissues was determined directly by graphite furnace atomic absorption spectrometry ( GF-AAS ) . As this method is not able to distinguish between iron of haem and non-haem origins , only salivary glands and ovaries dissected from partially engorged BF and SF ticks were used for the analysis to avoid distortions caused by the presence of haemoglobin in the samples . Despite large variations within individual biological replicates , the average iron concentration in either tissue was independent of haemoglobin in the tick diet ( Figure 3D ) . These results conclusively proved that the bioavailable iron in tick tissues originates from host serum components rather than from haemoglobin-derived haem . Guts dissected from partially-engorged I . ricinus females , and ovaries dissected 6 days after detachment ( AD ) from both BF and SF ticks displayed similar overall morphologies , except for colour ( Figure 4 ) . Accordingly , haem-containing haemosomes were not observed in the digest cells from SF ticks ( Figure 4 ) . Haemolymph collected from BF ticks displayed a typical haem light absorbance maximum ( Soret peak ) around 400 nm , which is not present in haemolymph from SF ticks ( Figure 5A ) . This observation demonstrates that haem present in the haemolymph of fully engorged females originated only from the blood meal of adults , and not from previous feeding at the nymphal stage . We estimate that out of approximately 10 µmol of total haem acquired from a tick blood meal , only about 100 nmol ( ~1% ) needs to be transported to the ovaries within a period of several days . 10 . 7554/eLife . 12318 . 016Figure 4 . Appearance of the tick gut , digest cells , and ovaries from blood- and serum-fed ticks . Whole guts from blood-fed ( BF ) and serum-fed ( SF ) partially engorged females ( fed for 6 days ) were dissected and semi-thin sections of digest cells were prepared and stained with toluidine blue . L - lumen; N - nucleus; arrows point to developing haemosomes that were present only in digest cells of BF ticks . Ovaries were dissected from BF and SF fully engorged females 6 days after detachment from the membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 01610 . 7554/eLife . 12318 . 017Figure 5 . IrCP3 is the major haem-binding protein in I . ricinus haemolymph . ( A-C ) Ir-CP3 and haem levels in haemolymph collected from blood-fed ( BF ) and serum-fed ( SF ) partially engorged females . ( A ) Absorbance spectra of haemolymph samples from BF and SF females . ( B ) SDS-PAGE of haemolymph samples from BF and SF ticks . Protein profiles were visualized using the TGX Stain-Free technology ( TGX ) . ( C ) Native pore-limit PAGE of heamolymph proteins stained with Coomassie ( CBB ) and specific co-detection of haem using peroxidase reaction with 3 , 3´-diaminobenzidine ( DAB ) . ( D-F ) Effect of RNAi-mediated silencing of ir-cp3 on the Ir-CP3 and haem levels in tick haemolymph . Unfed I . ricinus females were injected with gfp dsRNA ( gfp , control group ) or with ir-cp3 dsRNA ( ir-cp3 KD group ) and ticks were allowed to feed naturally on guinea pigs until partial engorgement ( fed for 6 days ) . ( D ) Absorbance spectra of haemolymph samples from from gfp control and ir-cp3 KD silenced ticks . ( E ) SDS-PAGE of haemolymph proteins ( 10 µl , 1:20 dilution ) collected from gfp control and ir-cp3 KD ticks . Protein profiles were stained with Coomassie ( CBB ) . ( F ) Native pore-limit PAGE of heamolymph proteins from gfp control and ir-cp3 KD ticks . Protein profiles were stained with Coomassie ( CBB ) and haem was co-detected using DAB . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 01710 . 7554/eLife . 12318 . 018Figure 5—figure supplement 1 . Stage and tissue expression of I . ricinus haemolymph carrier protein ( IrCP3 ) . ( A ) qPCR analyses of ir-cp3 expression in developmental stages of I . ricinus . ( B ) qPCR analyses of ir-cp3 expression in tissues dissected from fully engorged females . Data were obtained from three independent cDNA sets , and normalised to elongation factor 1 ( ef1 ) or actin . UF - unfed; FE - fully engorged; SG - salivary glands , OVA - ovaries; TRA - trachea-fat body complex; MT - Malpighian tubules; REST - remaining tissues . ( C ) SDS-PAGE separation of tissues dissected from I . ricinus females 6 days after detachment , visualized using the TGX Stain-Free technology ( TGX ) , and corresponding Western blot analyses of IrCP3 detected with specific antibodies ( αIrCP3 ) . SG - salivary glands , OVA - ovaries; TRA - trachea-fat body complex; HEM - haemolymph . Gut homogenate ( 50 µg of protein ) or other tissue homogenates ( 10 µg of protein ) were loaded per lane . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 018 Haem inter-tissue distribution and storage is facilitated by haem-binding protein ( s ) . In the cattle tick R . microplus , the most abundant haemolymph protein , named HeLp , was reported to bind haem in the haemocoel ( Maya-Monteiro et al . , 2000 ) . The genome of I . scapularis contains at least five genes related to HeLp , annotated as carrier proteins ( cp1‒5 ) . In I . ricinus , we identified and sequenced the cp3 orthologue , further referred to as ir-cp3 ( GenBank KP663716 ) . Expression profiling over I . ricinus developmental stages and tissues revealed that ir-cp3 mRNA was consistently up-regulated by blood-feeding and was predominantly expressed in the trachea-fat body complex and , to a lesser extent , in salivary glands and ovaries of adult females ( Figure 5—figure supplement 1 ) . SDS PAGE and Western blot analysis revealed that IrCP3 was most abundant in tick haemolymph ( Figure 5—figure supplement 1 ) , where its levels were not affected by the presence or absence of haemoglobin in the tick diet ( Figure 5B , C ) . Native pore-limit PAGE , followed by detection of haem via its peroxidase activity with 3 , 3’-diminobenzidine ( DAB ) , showed that haem was associated with the ~ 300 kDa band of IrCP3 only in the haemolymph from BF ticks ( Figure 5C , DAB panel ) . RNAi-mediated silencing of ir-cp3 in I . ricinus females ( ir-cp3 KD ) resulted in the disappearance of the haem Soret peak ( Figure 5D ) , a substantial ( ~80% ) reduction in IrCP3 levels on SDS PAGE ( Figure 5E ) , and the absence of IrCP3-associated DAB stained haem on the native gel ( Figure 5F ) . These results collectively demonstrate that IrCP3 is the major haem-binding protein in I . ricinus haemolymph . Extracts from I . ricinus ovaries were colourless until the 3rd day after detachment ( AD ) from the host , and then the Soret peak absorbance gradually increased , indicating an increase in haem concentration up to 8 days AD ( Figure 6A ) . SDS PAGE and Western blot analysis of ovary homogenates revealed that levels of IrVg1- and IrVg2-derived proteolytic products gradually increased after tick detachment whereas IrCP3 remained constant ( Figure 6—figure supplement 1 ) . Native pore-limit PAGE followed by DAB-based haem co-detection and Western blot analyses confirmed that the appearance of haem in tick ovaries was coincident with the occurrence of vitellins ( Figure 6B ) . I . ricinus vitellogenin genes ( ir-vg1 and ir-vg2 ) are exclusively expressed in fully engorged females , predominantly in the gut , salivary glands and trachea-fat body complex , but not in the ovaries ( Figure 6—figure supplement 2 ) . As vitellins are predominantly found in ovaries , their precursors ( vitellogenins ) must be transported from their site of synthesis to the ovaries . 10 . 7554/eLife . 12318 . 019Figure 6 . Vitellins are the major haem-binding proteins in tick ovaries . ( A-B ) Haem accumulation in tick ovaries occur concurrently with the appearance of vitellins . Ovaries were dissected from I . ricinus females at subsequent time-points after detachment ( AD ) from the host: FE - fully-engorged; 3 AD , 6 AD , 8 AD - 3 , 6 , and 8 days AD , respectively . ( A ) Absorbance spectra of ovaries homogenates show gradually increasing Soret peak following the 3rd day AD . ( B ) Native pore-limit PAGE of ovaries homogenates stained with Coomassie ( CBB ) , co-detection of haem-associated peroxidase activity with 3 , 3´-diaminobenzidine ( DAB ) , and Western blot analyses of vitellogenin 1- and vitellogenin 2- cleavage products ( αIrVg1 and αIrVg2 , respectively ) . Note that the native IrVg1- and IrVg2-specific bands correspond to the positions of the major haemoproteins in tick ovaries ( red asterisks ) . ( C-D ) RNAi-mediated silencing of I . ricinus vitellogenin 1 and 2 . Unfed I . ricinus females were pre-injected with gfp dsRNA ( control , gfp ) , ir-vg1 dsRNA ( ir-vg1 KD ) , and ir-vg2 dsRNA ( ir-vg2 KD ) , allowed to feed naturally on guinea pigs and then re-injected after detachment from the host with the same amount of dsRNA . ( C ) Effect of I . ricinus vitellogenin 1 and 2 RNAi-mediated silencing on ovaries appearance and haem levels . Bottom panels show the detailed view of ovary parts depicted by the yellow dashed squares above . Levels of haem b were determined by HPLC in three independent homogenates of ovaries dissected from each tick group 6 days after detachment . ( D ) Native pore-limit PAGE of ovaries homogenates ( 10 μg protein per lane ) dissected 6 days AD from control ( gfp ) , ir-vg1 KD and ir-vg2 KD ticks . Gels were stained with Coomassie ( CBB ) for proteins , 3 , 3´-diaminobenzidine for peroxidase activity of haem ( DAB , red asterisks ) , and Western blot analyses were performed with antibodies against vitellogenin 1 ( αIrVg1 ) and vitellogenin 2 ( αIrVg2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 01910 . 7554/eLife . 12318 . 020Figure 6—figure supplement 1 . SDS-PAGE and Western blot analyses of ovary homogenates from I . ricinus . Ovaries were dissected from I . ricinus females at subsequent time-points after detachment ( AD ) from the host: FE - fully-engorged; 3 AD , 6 AD , 8 AD - 3 , 6 , and 8 days AD , respectively . Protein profiles of ovaries homogenates were visualised using TGX Stain-Free technology ( TGX ) and corresponding Western blots of vitellogenin 1- , vitellogenin 2-derived cleavage products and IrCP3 were detected using αIrVg1 , αIrVg2 , and αIrCP3 specific antibodies , respectively . Note the gradual increase in both vitellins but a constant level of Ir-CP3 . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 02010 . 7554/eLife . 12318 . 021Figure 6—figure supplement 2 . Stage and tissue expression of I . ricinus vitellogenin 1 ( IrVg1 ) and vitellogenin 2 ( IrVg2 ) . Stage and tissue expression of I . ricinus vitellogenin 1 ( IrVg1 ) and vitellogenin 2 ( IrVg2 ) . ( A ) qPCR analyses of ir-vg1 and ir-vg2 expression in developmental stages of I . ricinus . ( B ) qPCR analyses of ir-vg1 and ir-vg2 expression in tissues dissected from ticks 4 days after detachment . Data were obtained from three independent cDNA sets , and normalized to elongation factor 1 ( ef1 ) or actin . UF - unfed; FE - fully engorged; SG - salivary glands , OVA - ovaries; TRA - trachea-fat body complex; MT - Malpighian tubules; REST - remaining tissues . ( C ) Western blot analyses of IrVg1 , and IrVg2 detected with specific antibodies ( αIrVg1 ) , and ( αIrVg2 ) , respectively . SG - salivary glands , OVA - ovaries; TRA - trachea-fat body complex; HEM - haemolymph . Gut homogenate ( 50 µg of protein ) or other tissue homogenates ( 10 µg of protein ) were loaded per lane . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 02110 . 7554/eLife . 12318 . 022Figure 6—figure supplement 3 . RNAi-mediated silencing of I . ricinus vitellogenin 1 and 2 . Unfed I . ricinus females were pre-injected with gfp dsRNA ( gfp ) , ir-vg1 dsRNA ( ir-vg1 KD ) , ir-vg2 dsRNA ( ir-vg2 KD ) , and ir-cp3 dsRNA ( ir-cp3 KD ) , allowed to feed naturally on guinea pigs and re-injected immediately after detachment ( AD ) with the same amount of dsRNA . Tissues were dissected 6 days AD . ( A ) qPCR analysis of ir-vg1 and ir-vg2 gene expression in the tick gut upon RNAi-mediated silencing . Note the mutual co-silencing of both genes . ( B ) SDS-PAGE protein profiles and corresponding Western blot analyses of ovary homogenates dissected from gfp , ir-vg1 KD , ir-vg2 KD , and ir-cp3 KD ticks . Proteins were visualized using the TGX Stain-Free technology ( TGX ) , and Western blots of IrVg1 , IrVg2 , and IrCP3 were detected with specific antibodies ( αIrVg1 ) , ( αIrVg2 ) , and ( αIrCP3 ) , respectively . Note the mutual co-silencing of IrVg1 and IrVg2 at the protein level . DOI: http://dx . doi . org/10 . 7554/eLife . 12318 . 022 RNAi-mediated silencing of ir-vg1 and ir-vg2 resulted in a substantial decrease in mRNA levels of both vitellogenin genes in gut tissues , and the same dual silencing effect was also observed at the protein level for IrVg1 and IrVg2 in tick ovary homogenates ( Figure 6—figure supplement 3 ) . This result points to a mutual regulation of both genes by an as yet unknown mechanism . More importantly , silencing of both ir-vg1 and ir-vg2 led to impaired development of tick ovaries and concomitant reduction in haem content in this tissue ( Figure 6C ) . Additionally , native pore-limit PAGE followed by DAB-staining and Western blotting ( Figure 6D ) showed that decreased levels of both IrVg1 and IrVg2 in ovary homogenates from ir-vg1 and ir-vg2 KD ticks were associated with the disappearance of DAB-stained haem . These results collectively show that vitellins are the major haem-binding proteins in I . ricinus ovaries and imply that the majority of haem is transported , along with vitellogenins , to the developing oocytes after tick detachment from the host .
The previous report on non-functional haem biosynthesis in the cattle tick R . microplus ( Braz et al . , 1999 ) prompted us to screen available mite genomes ( namely: I . scapularis , T . urticae , and M . occidentalis ) and reconstitute their gene repertoires for enzymes of the haem biosynthetic and degradative pathways . We found that 5-aminolevulinate synthase , together with the whole cytoplasmic segment of the haem biosynthetic pathway , is completely missing in hard ticks , but is present in other mites . Therefore we hypothesise that during evolution , ticks have lost most of the genes encoding haem biosynthesis as a consequence of their strict haematophagy . Only three genes encoding the vestigial mitochondrial enzymes of the haem biosynthetic pathway , namely PPOX , CPOX and FECH , have been retained in the I . scapularis genome ( Gulia-Nuss et al . , 2016 ) ( Figure 1B ) and their orthologues were also identified in midgut and salivary gland transcriptomes of I . ricinus ( Kotsyfakis et al . , 2015 ) ( Figure 1—figure supplements 1–3 ) . PPOX transcripts were also found in salivary gland transcriptomes from various species of the genus Amblyomma ( Garcia et al . , 2014; Karim and Ribeiro , 2015 ) . The same partial reduction in genomic coding for haem biosynthesis has been reported for a unicellular parasite , Leishmania major , in which the intracellular amastigote form expresses an active PPOX that likely sequesters the haem precursor coproporphyrinogen III from the macrophage cytosol to complete synthesis of its endogenous haem ( Zwerschke et al . , 2014 ) . Another haem auxotroph , the nematode Brugia malayi , was suggested to bypass its incomplete haem biosynthetic pathway using tetrapyrrole intermediates from endosymbionts ( Wu et al . , 2009 ) . Two lines of evidence suggest that PPOX , CPOX and FECH are not involved in haem biosynthesis in adult ticks: ( i ) Earlier , it was reported for R . microplus that no radioactively labelled δ-aminolevulinic acid was incorporated into haem present in haemolymph and ovaries ( Braz et al . , 1999 ) ; ( ii ) Recently , we have shown that RNAi-mediated silencing of the terminal FECH did not exert any effect on tick engorgement , oviposition , and larval hatching ( Hajdusek et al . , 2016 ) . This data suggests that these remnants of the haem biosynthetic pathway in I . ricinus do not contribute to the tissue haem pool that sustains successful reproduction . Therefore the reason for retaining genes encoding the last three enzymes of the haem biosynthetic pathway in ticks remains obscure and should undergo further investigation . The differential in vitro membrane feeding of I . ricinus females on whole blood ( BF ) or haemoglobin-free serum ( SF ) allowed us to investigate the importance of haemoglobin acquisition and inter-tissue transport of dietary haem in the hard tick I . ricinus , in an as yet unexplored way . These experiments surprisingly revealed that haemoglobin , which makes up about 70% of total blood proteins , is not a necessary source of amino acids for vitellogenesis ( Figure 2 and Figure 3 ) . Moreover , we have unambiguously demonstrated that haem in tick eggs originates entirely from host haemoglobin acquired during female feeding on hosts . Serum-fed I . ricinus were capable of full engorgement and oviposition , however embryonal development and larval hatching was aborted ( Figure 2 ) . The capability of tick embryos to develop viable progeny could be fully rescued by addition as little as about 1% of the physiological concentration of haemoglobin ( 0 . 1% in serum ) ( Figure 2—figure supplement 2 ) . In contrast to ticks , serum-fed triatomine Rhodnius prolixus were capable of laying eggs and giving rise to viable larvae ( Machado et al . , 1998 ) . As Triatominae insects possess a complete haem biosynthetic pathway ( Kanehisa and Goto , 2000 ) , they can apparently reproduce even in the absence of dietary haem . In the majority of animals studied so far ( including insect blood-feeders ) , haem degradation represents the main source of iron , and conversely , iron is mainly utilised for de novo haem biosynthesis ( Zhou et al . , 2007; Gozzelino and Soares , 2014 ) . Although it has been reported that , under certain conditions , haem can be degraded non-enzymatically ( Atamna and Ginsburg , 1995 ) , haem degradation-based on haem oxygenase ( HO ) is the most physiologically relevant ( Khan and Quigley , 2011 ) . We found that the HO gene was missing in the tick genome and correspondingly , the haem degradation product , biliverdin IX , could not be found in I . ricinus gut homogenates ( Figure 3—figure supplement 3 ) . We further noted that the absence of the HO gene is a common feature in other mite genomes ( Figure 1B ) and respective HO orthologues could not been found even in non-Acari genomes: the chelicerate genome of Stegodyphus mimosarum ( Sanggaard et al . , 2014 ) and the myriapode genome of Strigamia maritima ( Chipman et al . , 2014 ) . The apparent absence of HO transcripts in two color-polymorphic spiders of the genus Theridion is in agreement with the notion that these animals do not produce bilin pigments as haem degradation products ( Croucher et al . , 2013 ) . As HO gene is present in the genomes of Hexapoda ( Adams et al . , 2000; Holt et al . , 2002 ) and Crustacea ( Colbourne et al . , 2011 ) , we hypothesise that the loss of HO is an old ancestral trait of Chelicerata and Myriapoda that are phylogenetically supported as sister groups ( Dunn et al . , 2008 ) . Such a finding raises the question of dietary iron source for these animals , since iron is an essential electron donor/acceptor involved in vitally important physiological processes such as energy metabolism , DNA replication , and oxygen transport ( Hentze et al . , 2004; Dunn et al . , 2007 ) . Earlier , we and others reported that successful tick development and reproduction is strictly dependent on the availability of iron and maintenance of its systemic homeostasis ( Hajdusek et al . , 2009; Galay et al . , 2013 ) . Here , we demonstrate that levels of intracellular ferritin , as an indicator of bioavailable iron , as well as the concentration of elemental iron , do not significantly differ in tick tissues dissected from BF and SF females ( Figure 3C , D ) . These results further support the conclusion that bioavailable iron does not originate from haemoglobin-derived haem , but rather from serum iron-containing proteins , most likely host transferrin ( Hajdusek et al . , 2009; Galay et al . , 2014; Mori et al . , 2014 ) . However , an unequivocal identification of the source ( s ) of bioavailable iron for tick metabolic demands has to await the implementation of a chemically defined artificial tick diet , as recently reported for the mosquito Aedes aegypti ( Talyuli et al . , 2015 ) . The entire dependence of ticks on haem derived from host haemoglobin underscores the importance of a deeper understanding of haem inter-tissue transport from the site of haemoglobin digestion in the gut to ovaries and other peripheral tissues . In the triatomine bug , R . prolixus , a 15-kDa haemolymphatic haem-binding protein ( RHBP ) was reported to transport haem to pericardial cells for detoxification and to growing oocytes for yolk granules as a source of haem for embryo development ( Walter-Nuno et al . , 2013 ) . The haem transport and/or binding in ticks is mediated by HeLp/CPs and vitellins ( Maya-Monteiro et al . , 2000; Logullo et al . , 2002; Boldbaatar et al . , 2010; Smith and Kaufman , 2014 ) , that belong to the family of large lipid transfer proteins ( LLTP ) known to facilitate distribution of hydrophobic molecules across circulatory systems of vertebrates , as well as invertebrates ( Smolenaars et al . , 2007 ) . Vitellogenins are reported to be expressed only in fertilised fully-fed females , whereas HeLp/CPs are expressed ubiquitously in various stages , including adult males , and tissues ( Donohue et al . , 2008; Donohue et al . , 2009; Khalil et al . , 2011; Smith and Kaufman , 2014 ) . Based on these criteria , we clearly distinguished the I . ricinus carrier protein IrCP3 from two vitellogenins , IrVg1 and IrVg2 ( Figure 5—figure supplement 1; Figure 6—figure supplement 2 ) and demonstrated that during tick feeding , most haem in haemolymph is bound to IrCP3 . The haem is mainly transported to the developing ovaries during the off-host digestive phase , however the proportion of haem transported by IrCP3 or vitellogenins remains to be investigated . In ovaries , haem is sequestered by vitellins serving as haem-storage proteins for embryonal development . Further studies of the native arrangement and haem-binding capabilities of tick vitellins are needed to determine whether one or both vitellin apoproteins are involved in haem binding . Collectively , our results demonstrate that ticks lack functional haem biosynthesis , recycle dietary haem originating from digested haemoglobin , and the acquired haem does not contribute to the cellular iron pool . Therefore , haem and iron metabolism in ticks constitute a major departure from its canonical functioning described for other eukaryotic cells , where haem and iron homeostasis is based on balancing the flux between the opposing haem biosynthetic pathway and the HO-based degradative pathway . Further investigations of the exact molecular mechanisms involved in haem inter-tissue transport , intracellular trafficking , and compartmentation within the tick digest cells , promise to identify vulnerable targets in tick haem auxotrophy . This may lead to novel strategies for controlling ticks and the diseases that they transmit .
A pathogen-free colony of Ixodes ricinus was kept at 24°C and 95% humidity under a15:9-hr day/night regime . Twenty five females and males were placed into a rubber ring glued on the shaven back of guinea pigs and ticks were allowed to feed naturally for a specified time or until full engorgement ( 7‒9 days ) . Partially or fully engorged ticks were then either dissected or kept separately in glass vials until oviposition and larval hatching . All laboratory animals were treated in accordance with the Animal Protection Law of the Czech Republic No . 246/1992 Sb . , ethics approval No . 095/2012 . Membrane feeding of ticks in vitro was performed in feeding units manufactured according to the procedure developed by Kröber and Guerin ( Kröber and Guerin , 2007 ) . Whole bovine blood was collected in a local slaughter house , manually defibrinated and supplemented immediately with sterile glucose ( 0 . 2% w/vol ) . To obtain serum , whole blood samples were centrifuged at 2 500 × g , for 10 min at 4°C and the resulting supernatant was collected and centrifuged again at 10 000 × g , for 10 min at 4°C . Diets were then supplemented with 1 mM adenosine triphosphate ( ATP ) and gentamicin ( 5 µg/ml ) , pipetted into the feeding units and regularly exchanged at intervals of 12 hr . For feeding , fifteen females were placed in the feeding unit lined with a thin ( 80‒120 μm ) silicone membrane , previously pre-treated with a bovine hair extract in dichloromethane ( 0 . 5 mg of low volatile lipids ) as described ( Kröber and Guerin , 2007 ) . After 24 hr , unattached or dead females were removed and an equal number of males were added to the remaining attached females . For rescue experiments , pure bovine haemoglobin ( Sigma , St . Louis , MO , H2500 ) was added to the serum diet since the 5th day of feeding at a concentration of 10% , 1% , or 0 . 1% ( w/vol ) and then feeding was resumed until tick full engorgement . Naturally or in vitro fed I . ricinus females were forcibly removed from the guinea pig or membrane at a specified time of feeding , or collected at a specified time after detachment . Haemolymph was collected into a glass capillary from the cut front leg , pooled , and used for subsequent experiments . Other tissues , namely ovaries , salivary glands , gut , tracheae with adjacent fat body cells , Malpighian tubules , and the remaining tissues tagged as ‘rest’ were dissected on a paraplast-filled Petri dish under a drop of DEPC-treated PBS . Total RNA was isolated from dissected tissues using a NucleoSpinRNA II kit ( Macherey-Nagel , Germany ) and stored at –80°C prior to cDNA synthesis . Total RNA from haemolymph was isolated using TRI reagent ( Sigma ) . Single-stranded cDNA was reverse-transcribed from 0 . 5 µg of total RNA using the Transcriptor High-Fidelity cDNA Synthesis Kit ( Roche Diagnostics , Germany ) . For subsequent applications , cDNA was diluted 20 times in nuclease-free water . The search for tick genes encoding enzymes possibly involved in the haem biosynthetic and haem degradative pathways , a BLAST search using mosquito ( Anopheles gambiae ) genes was performed in the genome-wide database of Ixodes scapularis ( https://www . vectorbase . org/organisms/ixodes-scapularis ) . Genes encoding canonical haemoproteins were identified based on their genomic annotation . Other mite genomes , namely T . urticae ( Grbić et al . , 2011 ) and M . occidentalis , were mined in available databases http://metazoa . ensembl . org/Tetranychus_urticae/Info/Index/ and http://www . ncbi . nlm . nih . gov/bioproject/62309 , respectively . Additionally , transcriptomes available at the National Center for Biotechnology Information ( http://www . ncbi . nlm . nih . gov ) were screened using the BLAST® program . Metabolic pathways were reconstituted according to the Kyoto Encyclopedia of Genes and Genomes ( Kanehisa and Goto , 2000 ) . Gene products of 1806 bp , 2070 bp , 2151 bp , and 519 bp encoding fragments of I . ricinus carrier protein CP3 ( ir-cp3 ) , I . ricinus vitellogenin 1 ( ir-vg1 ) , vitellogenin 2 ( ir-vg1 ) , and complete ferritin 1 ( ir-fer1 ) , respectively , were amplified from a whole body cDNA library using primers designed according to corresponding I . scapularis orthologues or the ir-fer1 sequence ( for primer sequences , see Supplementary file 3 ) . Resulting amplicons were purified using the Gel and PCR Clean-up kit ( Macherey-Nagel ) , cloned into the pET100/D-TOPO vector of Champion pET directional TOPO expression kit ( Invitrogen , Carlsbad , CA ) , and expressed using E . coli BL 21 Star ( DE3 ) chemically competent cells . Expressed fusion proteins were purified from isolated inclusion bodies in the presence of 8M urea using a 5 ml HiTrap IMAC FF ( GE Healthcare Bio-Sciences AB , Sweden ) metal-chelating column charged with Co2+ - ions and eluted with an imidazole gradient . The recombinant proteins ( for sequences , see Supplementary file 2 ) were refolded by gradually decreasing the concentration of urea , finally dialyzed against 150 mM Tris/HCl , 150 mM NaCl , pH = 9 . 0 , and used to immunize rabbits as described previously ( Grunclová et al . , 2006 ) . The immune sera against IrCP3 , IrVg1 , IrVg2 and IrFer1 , tagged as αIrCP3 , αIrVg1 , αIrVg2 and αIrFer1 , were collected , aliquoted , and stored at –20°C until use . cDNA preparations from developmental stages and tissues were made in independent triplicates and served as templates for the following quantitative expression analyses by quantitative real-time PCR ( qPCR ) . Samples were analysed using a LightCycler 480 ( Roche ) and Fast Start Universal SYBR Green Master Kit ( Roche ) . Each primer pair ( for the list of qPCR primers , see Supplementary file 3 ) was inspected for its specificity using melting curve analysis . Relative expressions of ir-cp3 , ir-vg1 and ir-vg2 were calculated using the ΔΔCt method ( Pfaffl , 2001 ) . The expression profiles from adult I . ricinus female tick tissues were normalized to actin and the developmental stage expression profiles were normalized to elongation factor 1 ( ef1 ) ( Nijhof et al . , 2009; Urbanová et al . , 2014 ) . A 521-bp fragment of ir-cp3 ( corresponding to positions 2688–3208 bp , GenBank KP663716 ) , a 301-bp fragment of ir-vg1 ( corresponding to positions 2277–2577 bp of I . scapularis orthologue ISCW013727 ) , a 303-bp fragment of ir-vg2 ( corresponding to positions 801–1103 bp of I . scapularis orthologue ISCW021228 ) were amplified from tick gut cDNA and cloned into the pll10 vector with two T7 promoters in reverse orientations ( Levashina et al . , 2001 ) , using primer pairs CP3-F_RNAi , CP3-R_RNAi ( Supplementary file 3 ) containing the additional restriction sites ApaI and XbaI . dsRNA of ir-fer1 and ir-irp were synthesized as described ( Hajdusek et al . , 2009 ) . Purified linear plasmids served as templates for RNA synthesis using the MEGAscript T7 transcription kit ( Ambion , Lithuania ) . dsRNA ( ~1 µg in 350 nl ) was injected into the haemocoel of unfed female ticks using Nanoinject II ( Drummond Scientific Company , Broomall , PA ) . Control ticks were injected with the same volume of gfp dsRNA synthesized under the same conditions from linearized plasmid pll6 ( Levashina et al . , 2001 ) . After 24 hr of rest in a humid chamber at room temperature , ticks were allowed to feed naturally on guinea pigs . The gene silencing was verified by qPCR and/or Western blot analyses . Tissue homogenates were prepared in 1% Triton X-100 in PBS supplemented with a CompleteTM cocktail of protease inhibitors ( Roche ) using a 29G syringe , and subsequently subjected to three freeze/thaw cycles using liquid nitrogen . Proteins were then extracted for 1 hr at 4°C and 1 200 rpm using a Thermomixer comfort ( Eppendorf , Germany ) . Samples were then centrifuged 15 000 × g , for 10 min at 4°C . Protein concentrations were determined using the Bradford assay ( Bradford , 1976 ) . Electrophoretic samples for SDS-PAGE were prepared in reducing Laemmli buffer supplemented with β-mercaptoethanol . Ten micrograms of protein were applied per lane unless otherwise specified . Proteins were separated on gradient ( 4–15% ) Criterion TGX Stain-Free Precast gels ( BioRad , Hercules , CA ) in Tris-Glycine-SDS running buffer ( 25 mM Tris , 192 mM glycine , 0 . 1% ( w/vol ) SDS , pH 8 . 3 ) and visualized using TGX stain-free chemistry ( BioRad ) . Proteins were transferred onto nitrocellulose using a Trans-Blot Turbo system ( BioRad ) . For Western blot analyses , membranes were blocked in 3% ( w/vol ) non-fat skimmed milk in PBS with 0 . 05% Tween 20 ( PBS-T ) , incubated in immune serum diluted in PBS-T ( αIrFer1-1:50 , αIrVg1-1:1 000 , αIrVg2-1:1 000 , αIrCP3-1:1 000 ) , and then in the goat anti-rabbit IgG-peroxidase antibody ( Sigma ) diluted in PBS-T ( 1:50 000 ) . Signals were detected using ClarityWestern ECL substrate , visualized using a ChemiDoc MP imager , and analysed using Image Lab Software ( BioRad ) . Normalisation of Western blot analyses of gut homogenates were conducted using antibodies against IrCP3 , and homogenates of ovaries and eggs were normalised against the whole lane protein profile . Membrane stripping was carried out in a solution of 2% ( w/vol ) SDS and 0 . 5% ( vol/vol ) β-mercaptoethanol , and membranes were incubated for 1 hr at room temperature . Tissue homogenates were prepared as described above in Tris-Borate-EDTA ( TBE ) buffer ( 0 . 09M Tris , 0 . 08M boric acid , 2mM EDTA ) supplemented with a CompleteTM protease inhibitor cocktail ( Roche ) . Electrophoretic samples for pore-limit native PAGE were supplemented with 10% ( vol/vol ) glycerol and 0 . 001% ( w/vol ) bromophenol blue . Samples were run in 4‒16% Bis-Tris gel ( Invitrogen ) at 150 V in a cold room for 12 hr . Proteins were stained with Coomassie Brilliant Blue R-250 ( CBB ) . For visualisation of haem-associated peroxidase activity , the gel was rinsed in water and then incubated in 100 mM sodium acetate pH 5 . 0 with 0 . 2% ( w/vol ) 3 , 3’-diaminobenzidine ( DAB ) and 0 . 05% ( vol/vol ) hydrogen peroxide ( McDonnel and Staehelin , 1981 ) . Alternatively , proteins were transferred onto nitrocellulose using a Trans-Blot Turbo system ( BioRad ) and used for Western blot analyses as described above . Homogenates of five ovaries were prepared as described above in 400 μl TBE buffer and briefly spun down . Haemolymph samples were diluted 1:4 in TBE . Collected faeces ( 10 mg ) were homogenised in 100 µl of TBE buffer and briefly spun down . Supernatants from all samples were applied in a 2 μl-drop on a NanoQuant Plate ( Tecan , Austria ) and absorbance over the UV-VIS spectrum was scanned using the model Infinity 200 M Pro microplate reader ( Tecan ) . One dissected ovary , or 10 mg of eggs , was manually homogenised in methanol / 0 . 2% NH4OH ( vol/vol ) and centrifuged ( 15 000 × g , 10 min ) . The supernatant was discarded and haem was extracted from the pellet in 80% acetone / 2% HCl ( vol/vol ) . The extract was immediately separated by HPLC on a Nova-Pak C18 column ( 4-μm particle size , 3 . 9 × 75 mm; Waters , Milford , MA ) using a linear gradient of 25–100% ( vol/vol ) acetonitrile/0 . 1% trifluoroacetic acid at a flow rate of 1 . 0 ml/min at 40°C . Haem b was detected by a diode array detector ( Agilent 1200; Agilent Technologies , Santa Clara , CA ) and quantified using an authentic haemin standard ( Sigma , H9039 ) . Tick guts ( wet weight ~20 mg ) were dissected from naturally fed ticks 5 days after detachment from the guinea pig and homogenized individually in 100 µl of sterile PBS . For a positive control , 13 Aedes aegypti females were allowed to feed on mice and homogenized the 3rd day after feeding in 200 µl of sterile PBS . The samples were centrifuged ( 15 , 000 × g , 10 min ) , supernatants were extracted in 80% acetone / 2% HCl ( vol/vol ) and separated by HPLC on a Zorbax Eclipse plus C18 column ( 3 . 5 µm particle size 4 . 6 x 100 mm , Agilent ) . A linear gradient ( 0–100% , 20 min ) of solvent A ( methanol: acetonitrile: 0 . 01 M sodium acetate pH 3 . 65; 1:1:2 ) and solvent B ( acetonitrile / 0 . 1% TFA ) at a flow rate of 0 . 6 ml/min at 40°C was used . Biliverdin IX and haem b were detected simultaneously using an Agilent 1200 diode array detector at wavelengths of 660 nm and 375 nm , respectively . I . ricinus females were membrane fed on a blood or serum diet for 7 days until partial engorgement . Ovaries and salivary glands were dissected , taking special care to avoid contamination with gut contents , and washed in ultrapure 150 mM NaCl ( TraceSELECT , Fluka , Switzerland ) . Pools of tissues , collected from 5 females , were spun down briefly to remove excess saline , and freeze-dried . The dry tissue samples were weighed on microbalances ( with microgram precision ) and submitted for elemental analysis using graphite furnace atomic absorption spectroscopy , kindly performed by Prof . Hendrik Küpper , Institute of Plant Molecular Biology , BC CAS , České Budějovice . The iron concentrations obtained were expressed in parts per million ( ppm ) related to the dry weight of tissues . Data were analysed by GraphPad Prism 6 for Windows , version 6 . 04 and an unpaired Student’s t-test was used for evaluation of statistical significance . | Ticks are small blood-feeding parasites that transmit a range of diseases through their bites , including Lyme disease and encephalitis in humans . Like other blood-feeders , ticks acquire essential nutrients from their host in order to develop and reproduce . Iron and haem ( the iron-containing part of haemoglobin ) are essential for the metabolism of every breathing animal on Earth . Most organisms obtain iron by degrading haem and , reciprocally , most of the iron in cells is used to make haem . However , an initial search of existing genome databases revealed that ticks lack the genes required to make the proteins that make and degrade haem . Perner et al . wanted to find out if ticks can steal haem from the host and use it for their own development . To achieve this , Perner et al . exploited a method of tick membrane feeding that simulates natural feeding on a host by using a silicone imitation of a skin and cow smell extracts ( “l´odeur de vache” ) . Ticks were fed either a haemoglobin-rich ( whole blood ) or a haemoglobin-poor ( serum ) diet . This experiment revealed that ticks can develop normally without haemoglobin , but female ticks fed a haemoglobin-poor diet lay sterile eggs out of which no offspring can hatch . Further investigation showed that haemoglobin is vitally important as a source of haem but not as a source of the amino acids needed to produce the vitellin proteins that nourish embryos . As ticks are not armed with the ability to degrade haem , they do not acquire iron from the host haem but rather from a serum transferrin , a major iron transporter protein found in mammalian blood . Further experiments revealed that ticks have evolved proteins that can transport and store haem and so make the obtained haem available across the whole tick body . Overall , Perner et al . ’s findings suggest that targeting the mechanisms by which ticks metabolise haem and iron could lead to the design of new “anti-tick” strategies . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | Acquisition of exogenous haem is essential for tick reproduction |
Some anaerobic bacteria use insoluble minerals as terminal electron acceptors and discovering the ways in which electrons move through the membrane barrier to the exterior acceptor forms an active field of research with implications for both bacterial physiology and bioenergy . A previous study suggested that Shewanella oneidensis MR-1 utilizes a small , polar , redox active molecule that serves as an electron shuttle between the bacteria and insoluble acceptors , but the shuttle itself has never been identified . Through isolation and synthesis , we identify it as ACNQ ( 2-amino-3-carboxy-1 , 4-naphthoquinone ) , a soluble analog of menaquinone . ACNQ is derived from DHNA ( 1 , 4-dihydroxy-2-naphthoic acid ) in a non-enzymatic process that frustrated genetic approaches to identify the shuttle . Both ACNQ and DHNA restore reduction of AQDS under anaerobic growth in menaquinone-deficient mutants . Bioelectrochemistry analyses reveal that ACNQ ( −0 . 32 VAg/AgCl ) contributes to the extracellular electron transfer ( EET ) as an electron shuttle , without altering menaquinone generation or EET related cytochrome c expression .
Life is powered by redox reactions . It survives on the energy released as electrons move from the substrates that are oxidized to terminal electron acceptors ( TEAs ) . The redox reactions that support life embrace a huge variety of substrates , intermediates , and acceptors as different environments demand different tactics . Anaerobic bacteria with an insoluble TEA face an especially demanding challenge as they must move electrons from inside their cells where the life-sustaining redox reactions occur , to outside the cell where the TEA resides . One solution to this transfer is an ‘electron shuttle’ – a redox active , diffusible molecule that shuttles electrons between a bacterial cell and a TEA ( Brutinel and Gralnick , 2012; Gralnick and Newman , 2007 ) . Shewanella oneidensis MR-1 has been a model organism for studying extracellular electron shuttles since it was discovered by Myers and Nealson in 1988 for its ability to use an array of insoluble , extracellular TEAs ( Myers and Nealson , 1988 ) . In 2000 Newman and Kolter provided evidence for an extracellular electron shuttle in MR-1 by demonstrating that strains with a defect in menaquinone ( MK ) biosynthesis could not grow anaerobically with some TEAs , unless nearby wild type MR-1 provided a ‘quinone-like’ extracellular electron shuttle that functioned as a public good ( Newman and Kolter , 2000 ) . While their study suggested the existence of an electron shuttle , it did not identify it , and additional attempts at identification led to conflicting results . Subsequent studies by several research groups have since demonstrated that excreted flavins by MR-1 can mediate the reduction of various insoluble TEAs ( Brutinel and Gralnick , 2012; Kotloski and Gralnick , 2013; Marsili et al . , 2008; von Canstein et al . , 2008 ) , such as Fe ( III ) oxide and electrodes . Conversely , a report by Myers and Myers in 2004 used thin-layer chromatography and UV absorption spectroscopy to conclude that the excreted ‘quinone-like’ metabolite was not an electron shuttle , but rather a biosynthetic intermediate of MK that is capable of bypassing the MK-deficient mutants’ defect and thus generating small amounts of demethyl-MK ( Myers and Myers , 2004 ) . Collectively the previous attempts to identify the elusive electron shuttle demonstrated that it is small ( <300 amu ) , very polar , potent , and excreted under both anaerobic and aerobic conditions by many different bacteria . We were motivated to reinvestigate the identity of this metabolite - shuttle or MK intermediate - as it was never characterized and its identity has reemerged as a research priority through its potential role in bio-energy development ( Shi et al . , 2016 ) . In this report we provide evidence that ACNQ ( 2-amino-3-carboxy-1 , 4-naphthoquinone ) acts as MR-1’s elusive electron shuttle , describe the unusual biosynthetic pathway leading to ACNQ formation and demonstrate the ability of ACNQ to transfer electrons from MR-1 to carbon felt anodes .
We began by culturing MR-1 as previously described and assaying for shuttling activity using the original assay ( Newman and Kolter , 2000 ) - the ability to recover respiration of an isogenic MK-deficient menC mutant strain ( menC::Tn10 ) on a color-changing TEA , anthraquinone-2 , 6-disulfonate ( AQDS ) , ( Figure 1A ) . Rounds of bioassay-guided isolation ( see Materials and methods – Figure 1—figure supplement 1A , B ) produced highly active reduced complexity fractions and ultimately a single fraction containing the active molecule ( C11H8NO4 , [M + H]+ m/z 218 . 0445 , calcd . 218 . 0448 Δ 1 . 4 ppm ) . Detailed characterization was challenging as positive mode MS fragmentation did not match any metabolite in commercially available libraries ( Figure 1—figure supplement 2 ) , and our extremely mass-limited samples precluded NMR techniques . Additional MS-based experiments revealed that three of the seven protons were exchangeable ( Figure 1—figure supplement 3 ) , and negative mode MS data analyzed with the Metlin database ( Guijas et al . , 2018 ) identified a MK fragment ( Figure 1—figure supplement 4 ) . The candidate fitting the molecular composition , MK core , and other data was ACNQ ( 2-amino-3-carboxy-1 , 4-naphthoquinone , Figure 1B ) . An independent synthesis of ACNQ and LCMS comparison ( Figure 1—figure supplements 2 and 4 ) confirmed that ACNQ was the unidentified metabolite in the wild type supernatant that recovered utilization of AQDS ( −184 mV vs . SHE; Aulenta et al . , 2010 ) by the menC mutant ( menC::Tn10 ) . In a functional assay , ACNQ potently recovers utilization of AQDS by menC mutant ( menC::Tn10 ) with an EC5025 . 0 ± 6 . 0 nM ( Figure 1C ) . Newman and Kolter reported that both Vibrio cholerae and other Shewanella species could recover respiration of the MK-negative mutant whereas Pseudomonas fluorescens could not ( Newman and Kolter , 2000 ) . In order to confirm that ACNQ is responsible for the recovery phenotype , we screened other bacterial strains for ACNQ production . ACNQ production was previously reported from Propionibacterium freudenreichii ( Kaneko , 1999; Kouya et al . , 2007; Mori et al . , 1997 ) , and we confirmed production in several Gram-negative bacteria – V . cholerae , Escherichia coli , and Bacteroides fragilis – and a Gram-positive bacterium – Lactococcus lactis ( Figure 2—figure supplements 1–2 ) . Quantitative analysis showed that MR-1 produces approximately 10–100 times more ACNQ than either E . coli or V . cholerae ( Figure 2A ) . The ACNQ concentration in MR-1 supernatant under laboratory conditions is approximately 1 . 5 nM ( Figure 2—figure supplement 3 ) . Although this is lower than the calculated EC50 value , simulation of ACNQ production and diffusion in an agarose gel indicates that local concentration of ACNQ when MR-1 forms a biofilm may reach 400 nM in two days ( Figure 2—figure supplement 4 ) . Once the structure of the active metabolite was known , we addressed its biosynthesis by MR-1 and other bacteria . Based on ACNQ’s structural similarity to MK and the genetic evidence that transposon mutations in menC or menB in MR-1 ( Myers and Myers , 2004; Newman and Kolter , 2000 ) and E . coli ( Figure 2—figure supplement 5 ) ] led to loss of production of ACNQ , it seemed likely that ACNQ was produced largely by the known MK biosynthetic pathway ( Figure 2B ) . As mutations upstream of the synthesis of DHNA ( 1 , 4-dihydroxy-2-naphthoic acid ) abolish ACNQ production , we hypothesized that DHNA is used as the substrate for ACNQ production ( Figure 2C ) . menA mutants derived from MR-1 and E . coli K12 ( menA::Himar and menA789::kan , respectively ) produced 2 . 5 times higher amounts of ACNQ than their wild type parent counterparts ( Figure 2A and Figure 2—figure supplement 5 ) . This increase is consistent with the function of MenA in converting DHNA to demethyl-MK and thus reducing the levels of DHNA available for conversion to ACNQ . Additional support came from the observation that while the genes for the production of MK are not clustered , the genes for the production of DHNA are in one small region of the circular chromosome , while menA , which encodes a protein carrying out the step directly following DHNA production in the canonical pathway , is on the opposite side of the chromosome ( Figure 2B ) . To further characterize the conversion of DHNA to ACNQ we searched for the transaminase responsible for this reaction . The most likely candidate seemed to be GlmS , a glutamine:fructose-6-phosphate transaminase , which is expressed by a gene located downstream of menB . However , knocking out glmS did not result in loss of production of ACNQ ( Figure 2—figure supplement 6 ) . As there were 18 other transaminases in the NCBI-annotated MR-1 genome , we adopted a systematic approach using E . coli K12 rather than a piecemeal approach using MR-1 to identify the responsible transaminase . Strains containing disruptions in all 22 annotated genes encoding aminotransferases/transaminases found in E . coli K12 genome were acquired from The Coli Genetic Stock Center ( CGSC ) , ( Supplementary file 1B ) . All of the mutant strains were able to make ACNQ ( Figure 2D ) . The failure to identify an E . coli aminotransferase/transaminase mutant unable to make ACNQ coupled with the failure of several prior genetic screens ( transposon libraries ) ( Myers and Myers , 2004; Newman and Kolter , 2000 ) in MR-1 background to identify respiration defects in any genes other than MK biosynthetic genes suggested that DHNA might be converted to ACNQ by a non-enzymatic pathway . To investigate whether ACNQ forms spontaneously , dilute DHNA was incubated in a variety of aqueous solutions , including the minimal media used in this work to culture MR-1 , minimal media with a mixture of amino acids as the sole nitrogen source , and an ammonium phosphate buffered solution . All conditions led to the conversion of DHNA to ACNQ after 24 hr with no remaining DHNA detected ( Figure 2—figure supplements 7–9 ) . DHNA is quite labile and appeared to react with other media components as was evident by the complex LCMS chromatogram of the reaction mixtures . Furthermore , MR-1 produced ACNQ when grown on a variety of nitrogen sources , i . e . ammonium sulfate or different amino acid mixtures ( Figure 2—figure supplement 10 ) . Quantification of both the MK pool in the cell pellet and ACNQ in the supernatant , 350 nM and 1 . 5 nM of MR-1 culture , respectively , indicated that only 2% of the DHNA pool would need to be diverted to ACNQ production if ACNQ were to be generated entirely non-enzymatically ( Figure 2—figure supplement 11 ) . A likely non-enzymatic mechanism would commence with the oxidation of DHNA to the quinone then conjugate addition of ammonium at the C3 position followed by enol formation and a subsequent oxidation of the hydroquinone ( Figure 2—figure supplement 12 ) . ACNQ , DHNA , menadione , MK-4 , and riboflavin were each evaluated for their ability to recover menC mutant ( menC::Tn10 ) reduction of AQDS to AHQDS under anaerobic conditions when it is the sole TEA . These metabolites where chosen because they represent either products of the disrupted pathway in the menC mutant strain ( MK-4 and DHNA ) , were previously implicated in electron shuttling ( riboflavin ) or to define structure-activity relationships ( menadione ) . ACNQ , DHNA , and MK-4 exhibited activity with EC50 values of 25 . 0 ± 6 . 0 , 106 . 2 ± 11 . 2 , and 87 . 8 ± 4 . 7 nM , respectively ( Figure 1C ) , while neither menadione nor riboflavin allowed for AQDS reduction by menC mutant strain ( concentrations up to 50 μM ) . It should be noted that while DHNA can recover respiration nearly as efficiently as ACNQ , DHNA decomposes relatively quickly in media and was not detected in the spent supernatant of MR-1 cultures ( data not shown ) . Although ACNQ is structurally similar to the MK precursor DHNA , addition of ACNQ does not replenish the MK pool in menC mutant ( menC::Tn10 ) as we demonstrated by HR-LCMS analysis on the lipid contents of both MK-negative and MR-1 cultures grown in the presence of various concentrations of ACNQ . This analysis revealed no evidence that ACNQ was being converted to MK ( Figure 2—figure supplements 13–15 ) , demethyl-MK , or an amino-modified MK analog . The instrument limit of detection for MK and demethyl-MK under the tested conditions is approximately 60 fmol , allowing for the detection of the presence of approximately 20 pmol of MK in a 50 mL culture . The inability of ACNQ to be converted to MK or related analog contrasts with previous findings by Myers and Myers ( Myers and Myers , 2004 ) , however it is important to note that the previous work used a complex mixture of metabolites that is present in spent supernatant and used thin-layer chromatography , a much less informative analytical method . To better understand the ability of ACNQ to act as an electron shuttle , we followed the work of Marsili et al . ( Marsili et al . , 2008 ) and performed detailed bioelectrochemical experiments in which we varied the availability of ACNQ . We grew a biofilm of MR-1 using lactate as the sole electron donor , M9 medium as the electrolyte with 1 μM ACNQ as the possible electron shuttle , and a carbon felt electrode biased at −0 . 1 VAg/AgCl as the sole electron acceptor . This potential was chosen in order to restrict the redox processes close to the redox potential of ACNQ ( −0 . 32 VAg/AgCl , Figure 3—figure supplement 1A ) . The established biofilm produced a significant oxidation current , ~22 μA·cm−2 , after 25 hr ( Figure 3A ) . When the medium was removed and replaced with fresh anaerobic medium , the oxidation current dropped more than 90% [Figure 3A , curve ( 1 ) ] , showing that the major carrier of current was soluble . 1 μM ACNQ was then added into the reactor , resulted in an immediate increase in current [Figure 3A , curve ( 2 ) ] , which stabilized in 6 hr to 45% of the original biofilm current , indicating that ACNQ is responsible for the current generation . The loss of planktonic cells and secreted flavins in the original media when changing the media might be the reason that current is not fully restored . Cyclic voltammetry ( CV ) analysis [Figure 3B , curve ( 1 ) and ( 2 ) ] reveals a major catalytic wave initiated at around −0 . 34 VAg/AgCl , which matches the oxidation peak potential of ACNQ . No peaks associated with flavins were observed under these conditions . Also , the peak width at half height ( W1/2 ) obtained by differential pulse voltammetry ( DPV ) is 66 mV ( Figure 3—figure supplement 1A ) , indicating that more than one electron is transferred during the redox reaction as would be expected for the oxidation of a hydroquinone to quinone . To determine the role of ACNQ more accurately , we performed a second replacement of the media with fresh media lacking ACNQ . The current decreased by 95% [Figure 3A , curve ( 3 ) ] . Next , we re-introduced the original ACNQ-containing media after removing the planktonic cells through two rounds of centrifugation and filtration through a 0 . 22 μm membrane . The current immediately increased and recovered to 85% of the current before second media replacement [Figure 3A , curve ( 4 ) ] . CV analysis [Figure 3B , curve ( 3 ) and ( 4 ) ] shows the major catalytic wave starts at the same ACNQ-related potential . Shewanella oneidensis mutants that are unable to secrete flavins ( Δbfe ) ( Kotloski and Gralnick , 2013 ) or lack outer membrane cytochromes ( Δmtr ) ( Coursolle and Gralnick , 2012 ) show similar chronoamperometry ( Figure 3—figure supplement 2A , B ) and CV ( Figure 3—figure supplement 2C , D ) results , indicating EET by ACNQ does not require outer membrane cytochromes or flavins . Further CV analysis ( Figure 3—figure supplement 1B ) indicates that ACNQ is also acting as an absorbed redox species , much like what was observed with flavins ( Marsili et al . , 2008 ) . These data indicates that ACNQ can act as the major electron carrier to a carbon-based electrode when biased at −0 . 1 VAg/AgCl . A series of chronoamperometry experiments was performed to investigate how ACNQ influences extracellular electron transfer . In the absence of bacterial cells , ACNQ does not generate current to a biased electrode ( Figure 3—figure supplement 1C ) . Likewise , in the absence of ACNQ , both wild type MR-1 and menC::Tn10 produce very low current density ( Figure 3C ) . We suggest that this occurs because the −0 . 1 VAg/AgCl potential is too negative for electron transfer through outer membrane cytochromes , and because the bacteria were washed before injection so neither flavins nor ACNQ is present at sufficient concentration to mediate electron transfer . Addition of exogenous ACNQ to wild type MR-1 and menC mutant ( menC::Tn10 ) immediately increases current 33- and 36- fold , respectively ( Figure 3C ) . In accordance with our previous results , HR-LCMS analysis of the relative MK concentration revealed no differences in MK or amino-modified MK analogues’ levels between ACNQ and DMSO-treated ( Figure 3—figure supplement 3 ) . Furthermore , ACNQ does not affect the cytochrome c expression in menC mutant ( Figure 3—figure supplement 1D ) , which excludes the possibility that the current increase was due to any change in the cyt c . These observations strongly suggest that ACNQ can act to transfer electrons to extracellular TEAs in S . oneidensis . Although defining ACNQ’s precise role ( s ) in extracellular electron transfer will require future studies , in this work we demonstrated that it is the unidentified quinone-like metabolite whose existence was established almost 20 years ago . We confirmed that ACNQ is excreted from S . oneidensis MR-1 ( and other bacteria ) , is responsible for AQDS reduction , and is capable of increasing current to an electrode via a mechanism that does not require menaquinone or outer membrane cytochrome c . The identification of ACNQ could prove useful in future bioenergy applications and should promote an examination of the roles flavins and ACNQ have as extracellular electron shuttles under different conditions .
UV spectra were measured on an Amersham Biosciences Ultrospec 5300-pro UV/Visible spectrophotometer and IR spectra were measured on a Bruker Alpha-P spectrophotometer . NMR spectra were recorded with methanol or DMSO as an internal standard ( δH 2 . 50 and δC 39 . 5 ) on a Bruker Avance 1 500 MHz spectrometer equipped with a TXO Helium CryoProbe ( 500 and 125 MHz for 1H and 13C NMR , respectively ) . LR-LCMS data were obtained using an Agilent 1200 series HPLC system equipped with a photo-diode array detector and a 6130-quadrupole mass spectrometer . HRESIMS was carried out using an Agilent 6530 LC-q-TOF Mass Spectrometer equipped with an uHPLC system . HPLC purifications were carried out using Agilent 1100 or 1200 series HPLC systems ( Agilent Technologies ) equipped with a photo-diode array detector . All solvents were of HPLC quality . Shewanella oneidensis MR-1 , an isogenic menC mutant ( menC::Tn10 ) , and Vibrio cholerae V52 were obtained from the Kolter lab at Harvard Medical School ( Newman and Kolter , 2000 ) . S . oneidensis strain with a menA::Himar mutation derived from MR-1 was kindly provided to us from the Barstow group at Princeton University ( Baym et al . , 2016 ) , S . oneidensis bfe and mtr mutant strains ( Δbfe and Δmtr ) were provided by the Gralnick group at the University of Minnesota ( Coursolle and Gralnick , 2012; Kotloski and Gralnick , 2013 ) . glmS knockout strain in MR-1 background was constructed as part of this study ( see below ) . Keio E . coli strains were purchased from the Coli Genetic Stock Center ( CGSC ) at Yale University ( see Supplementary file 1B ) . Bacteroides fragilis ATCC 25285 and Lactococcus lactis lactis ( DSM 20481 ) were purchased from commercial sources . All overnight cultures were grown in LB media ( DB Difco ) at 30°C , 200 rpm and all experimental cultures were grown at 30°C for 72 hr in the following lactate minimal media unless specifically stated otherwise: 1 . 19 g/L ( NH4 ) 2SO4 , 0 . 99 g/L K2HPO4 , 0 . 45 g/L KH2PO4 , 0 . 168 g/L NaHCO3 , 54 mg/L CaCl2 , 4 . 8 g/L disodium fumarate , 0 . 1 g/L vitamin-free casamino acids , 0 . 247 g/L MgSO4·7H2O , 1 . 6 g/L sodium lactate , 100 mg/L l-Arg-HCl , 100 mg/L l-Glu , 100 mg/L l-Ser , and 1 mL/L of the following trace elements solution with the pH adjusted to 7 . 4 . Trace element solution: 25 . 0 g/L Na2-EDTA , 1 . 5 g/L FeSO4·7H2O , 0 . 29 g/L ZnSO4·7H2O , 0 . 22 g/L MnSO4·H2O , 3 . 50 g/L H3BO3 , 1 . 41 g/L CoSO4·7H2O , 0 . 05 g/L CuSO4·5H2O , 1 . 97 g/L Ni ( NH4 ) 2SO4·6H2O , 0 . 94 g/L Na2MoO4·2H2O , 0 . 28 g/L Na2SeO4 , and 0 . 58 g/L NaCl . For purification of the active metabolite , S . oneidensis MR-1 was grown in batches of 16 × 4 L Erlenmeyer flasks containing 1 L of lactate minimal liquid media . After 40 min sterilization cycle each Erlenmeyer flask was cooled to room temperature ( rt ) overnight and then inoculated with 1 mL of a 5 mL overnight LB culture of MR-1 ( grown at 30°C with shaking at 200 rpm ) . The Erlenmeyer flasks were then incubated at 30°C for 72 hr without shaking . Conditioned supernatant was separated from cells by centrifugation ( 14 , 000 rcf , 30 min ) , filter-sterilized ( 0 . 2 μm , Thermo Scientific , cat #: 595–4520 ) , and acidified to a pH of 6 . 5 with 12 . 1 M HCl ( aq ) . This material was then passed through a glass column containing 1 g of Strata-X-A resin ( Phenomenex , 33 μm , 85 Å ) that was prepped by washing with 15 mL of 100% MeOH and subsequently 15 mL of 100% H2O . Upon passing all of the conditioned supernatant through the column , the resin was washed with 12 mL of 25 mM ammonium acetate ( aq ) pH 6 . 5% and 100% methanol . The crude material was eluted from the column with 15 mL of 5% formic acid ( FA ) in acetonitrile . The solvent then was immediately removed under vacuum to yield the crude extract . Activity-guided isolation of the crude extract by reverse-phase ( RP ) high performance liquid chromatography ( HPLC ) yielded a potently active reduced-complex fraction . This active fraction eluted near the 33rd min using Phenomenex 4 µm Hydro semi-preparative column ( 250 × 10 mm ) , under the following conditions: hold 20% ACN +0 . 1% FA/H2O + 0 . 1% FA for 5 min then gradient to 50% ACN +0 . 1% FA/ H2O+0 . 1% FA over 35 min with a flow rate of 3 mL/min . This reduced complex fraction was then analyzed on HRLCMS using a Phenomenex Kinetex 2 . 6 μm EVO C18 100 Å ( 100 × 2 . 1 mm ) under the following LC method: hold 10% ACN +0 . 1% FA/H2O + 0 . 1% FA for 1 min followed by a gradient to 100% ACN +0 . 1% FA over 9 min then hold 100% ACN +0 . 1% FA for 3 . 5 min with a constant flow rate of 0 . 3 mL/min . For MS and MS/MS measurements the electrospray ionization ( ESI ) parameters were set to 10 L/min sheath gas flow , 3 . 0 L/min auxiliary gas flow , and 325°C gas temperature . The spray voltage was set to 3 . 5 kV , the inlet capillary was set to 300°C and a 65 V S-lens radio frequency ( RF ) level was applied . MS/MS spectra were recorded in Auto MS/MS mode using a preferred ion list only . Scans were acquired at a rate of 13 , 577 transients/spectrum with a mass range of 50–1 , 700 m/z . The collision energy was set to 10 , 20 , and 30 eV for all ions . Analysis of the active reduced-complex HPLC fraction by the same HR-LCMS described above , except exchanging D2O ( Sigma Aldrich ) for H2O in the mobile phase , allowing for the determination of the exact number of exchangeable hydrogens the active metabolite contained . An aliquot ( 5 μL ) of the reduced-complex HPLC fraction was analyzed using this modified method and a 3 Da mass shift was observed for both [M + D]+ and [M + Na]+ ion adducts . This indicated that the active metabolite possessed three exchangeable hydrogens . Ammonium chloride ( 20 g , 0 . 374 mol ) was added to a glass jar containing 2 L of deionized ( DI ) water and a magnetic stir bar . Once the solid dissolved , the pH of the mixture was adjusted to 9 . 0 using ~10 mL of ammonium hydroxide solution . Then 1 , 4-dihydroxy-2-naphthoic acid ( DHNA - 1 . 0 g , 4 . 5 mmol ) was added to the mixture . The reaction mixture was stirred for 24 hr at rt , at which point the mixture was acidified to pH five and then extracted with 2 L ethyl acetate ( EtOAc ) x 2 . The organic layer was concentrated under vacuum . The crude material was then resuspended in 10 mL 50% ACN/H2O , syringe filtered ( Corning Inc , 0 . 2 μm ) , and purified using prep-HPLC . Pure ACNQ eluted between 33–35 min using Phenomenex Luna 5 µm C8 ( 2 ) 100 Å preparative column ( 250 × 21 . 2 mm ) , with the following method: holding 20% ACN +0 . 1% FA/H2O + 0 . 1% FA for 5 min then gradient to 55% ACN +0 . 1% FA/H2O + 0 . 1% FA over 35 min at 10 mL/min flow rate . The overall reaction yield , through purification , is approximately 1% . Synthetic ACNQ was compared to the isolated material by back-to-back injection on the HRLCMS using a Phenomenex Kinetex 2 . 6 μm EVO C18 100 Å ( 100 × 2 . 1 mm ) under the following LC method: holding 2 . 0% ACN +0 . 1% FA/H2O + 0 . 1% FA for 1 min then gradient to 20% ACN +0 . 1% FA/H2O + 0 . 1% FA over 39 min at 0 . 3 mL/min . An aliquot of DHNA ( 15 μL of 10 mM solution in MeOH ) was added to 10 mL of either lactate minimal media described above , lactate minimal media with an amino acid mixture replacing the ammonium sulfate [amino acid mixture ( 10x mixture ) : L-arginine: 1 . 26 g/L , L-cystine: 0 . 24 g/L , L-histidine HCl: 0 . 42 g/L , L-isoleucine: 0 . 52 g/L , L-leucine: 0 . 52 g/L , L-lysine HCl: 0 . 73 g/L , L-methionine: 0 . 15 g/L , L-phenylalanine: 0 . 33 g/L , L-threonine: 0 . 48 g/L , L L-tryptophan: 0 . 1 g/L , L-tyrosine: 0 . 36 g/L , L-valine: 0 . 467 g/L] , and ammonium phosphate buffered aqueous solution , pH 8 . The mixtures of samples were kept under aerobic conditions at rt , with 5 μL of each mixture analyzed on the LC-q-TOFMS for the production of ACNQ and consumption of DHNA at specific time points . Amorphous yellow solid; UV ( MeOH ) λmax ( log ε ) 207 ( 3 . 82 ) , 232 ( 4 . 04 ) , 266 ( 4 . 05 ) , 334 ( 3 . 30 ) , and 416 ( 3 . 25 ) nm; IR ( KBr ) 3294 , 3173 , 1684 , 1593 , 1542 , 1479 , 1462 , 1395 , 1367 , 1319 , 1286 , 1232 , 907 . 3 , 775 , 725 , 695 , 658 , and 583 cm-1; 1H NMR ( 500 MHz , d6-DMSO ) 14 . 5 ( s ) , 10 . 0 ( bs ) , 9 . 6 ( bs ) , 8 . 12 ( d; 7 . 8 Hz ) , 8 . 06 ( d; 7 . 7 Hz ) , 7 . 93 ( t; 7 . 5 Hz ) , and 7 . 83 ( t; 7 . 83 Hz ) ppm; 13C NMR ( 75 MHz , d6-DMSO ) 184 . 1 , 178 . 9 , 169 . 3 , 155 . 8 , 135 . 8 , 133 . 8 , 132 . 3 , 130 . 3 , 126 . 7 , 126 . 5 , and 96 . 9 ppm; HRESIMS [M + H]+ m/z 218 . 0456 ( calcd . for C11H8O4N 218 . 0453 , Δ 1 . 2 ppm ) . To follow the activity through purification , we used a modified method of the AQDS assay previously described ( Newman and Kolter , 2000 ) . An aliquot ( 10 μL ) of an overnight bacterial culture grown in LB ( 30°C , with shaking at 200 rpm ) was added to 2 mL of lactate minimal media with AQDS ( 5 . 0 mM ) as the sole TEA ( no sodium fumarate added ) . Reduction of AQDS to AHQDS generates a red color . Wild type S . oneidensis MR-1 was added to the positive control culture tubes and menC mutant was added to both negative control and test substrate culture tubes . Test substrates – HPLC fractions or pure compounds – were resuspended to the desired concentration ( 1 mg/mL or 150 μL whichever was higher for HPLC fractions ) in 50% ACN/H2O and then 25 μL was added to the culture tube of a test substrate . The solvent was added to both the positive and negative controls . Upon addition of test substrates and solvent blanks the culture tubes were incubated at room temperature for 24 hr in an anaerobic chamber ( Coy Laboratory ) , at which point they were visually inspected for color changes comparing the culture tubes of the test substrates to both the positive and negative controls . To quantify the reduction of AQDS , the above method was modified to be compatible with a 96-well format in order to read the production of AHQDS with a plate reader . Each test substrate was resuspended to the desired concentration in 50% ACN/H2O and 1 μL was added to individual wells in triplicate along with solvent blanks in both the positive and negative control wells . An aliquot ( 100 μL ) of aerobic cultures of both wild type MR-1 and menC mutant that were grown overnight in LB at 30°C with 200 rpm shaking , was used to inoculate 20 mL of minimal media that was supplemented with AQDS ( 5 . 0 mM ) as the sole TEA . This mixture was vortexed and then 199 μL was added to each well – wild type MR-1-containing media to positive control wells and menC mutant-containing media to both negative control and test substrate wells . The plates were then incubated in an anaerobic chamber at room temperature for 24 hr before being analyzed on a Molecular Devices M5 plate reader by monitoring absorbance at 408 nm . To estimate the local concentration of ACNQ produced by a streak of Shewanella oneidensis MR-1 in a 1 . 5% agarose gel , we assumed that only two processes occurred: production of ACNQ by the bacteria and diffusion of ACNQ through the agarose . We estimated the production of ACNQ and its diffusion using our data and other data from the literature . To estimate the production rate of ACNQ per cell of Shewanella oneidensis ( p ) , we used the fact that the concentration of ACNQ was 1 . 5 nM after 4 days of production by a stationary phase culture with a cell density of ~1×109 cell mL−1 . This calculation yielded p=4×10−23 moles cell−1 s-1 . The diffusion coefficient of ACNQ in water has been measured to be 5 . 2 × 10−6 cm2 s−1 ( Batchelor-McAuley et al . , 2010 ) . Molecules of similar size and hydrophobicities have a diffusion coefficient in 1 . 5% agarose that is 95% that of their diffusion coefficient in water ( Fatin-Rouge et al . , 2004 ) , largely because the porosity of agarose is large relative to the size of these molecules . Therefore , we estimate the diffusion coefficient of ACNQ in 1 . 5% agarose to be D = 5×10−6 cm2 s−1 . Coupled diffusion and production of a molecule is described by the Kolmogorov-Petrovsky-Piskunov equation ( Equation 1 ) , where C is the concentration of ACNQ ( in moles L−1 ) , D is the diffusion coefficient of ACNQ ( in cm2 s−1 ) , and P is the production rate of ACNQ ( in moles s−1 ) . ( 1 ) dC ( x , t ) dt = -Dd2C ( x , t ) dx2+P We numerically solved Equation 1 using an explicit time-domain finite difference formulation ( 2nd order in space and time ) with far-field Dirchelet boundary conditions ( concentration of 0 ) over a large model domain ( 10 cm ) to minimize boundary interaction . Initial conditions were set to C = 0 over the modeling domain and the source term ( P ) was implemented as an addition to C over each timestep . The formulation was implemented in MATLAB ( Mathworks , Cambridge , MA ) . glmS knockout mutation in S . oneidensis MR-1 background was made as described previously ( Saltikov and Newman , 2003 ) . Upstream flanking region of glmS was PCR-amplified with GlmS1_fwd ( gagcgcgcgtaatacgactcactataggCGTGGCACTTGAAGCTAAG ) and GlmS1_rev ( attggctttgattACGATTCCGCACATAGTTTTTAC ) primers , downstream flanking region was PCR-amplified with GlmS2_fwd ( tgtgcggaatcgtAATCAAAGCCAATAAAAAACC ) and GlmS2_rev ( aaccctcactaaagggaacaaaagcCGCTGAAGAAGGTAAAGC ) primers . pSMV10 ( Saltikov and Newman , 2003 ) was PCR amplified with pSMV10_fwd ( GCTTTTGTTCCCTTTAGTG ) and pSMV10_rev ( CCTATAGTGAGTCGTATTACGC ) . The flanking regions were introduced into pSMV10 using NEBuilder HiFi DNA Assembly Cloning Kit and transformed into chemically competent λpir E . coli UQ950 ( Saltikov and Newman , 2003 ) . Transformants were selected on 50 μg/mL kanamycin . The construct was isolated and transformed into diaminopimelic acid ( DAP ) auxotroph E . coli donor strain WM3064 ( Saltikov and Newman , 2003 ) and plated on LB supplemented with kanamycin and 0 . 3 mM DAP . For conjugation WM3064 containing the knockout , the construct was grown overnight in LB broth supplemented with kanamycin and DAP . An aliquot ( 250 μL ) of the overnight were washed with 500 μL of LB ( Lennox ) and mixed with 250 μL of S . oneidensis MR-1 overnight . The suspension was plated on LB supplemented with DAP and incubated at 30°C for 8 hr . A streak of the lawn was then spread on LB supplemented with kanamycin , but without DAP . A colony of the trans-conjugant was grown overnight at 30°C in LB broth . glmS encodes a putative glutamine-fructose-6-phosphate aminotransferase required for N-acetylglucosamine ( GlcNAc ) biosynthesis , therefore we predicted that the mutant may be a GlcNAc auxotroph . The overnight was diluted 3000 times and 100 μL was spread on LB agar ( Lennox ) containing 10% ( w/v ) sucrose with or without 10 mM GlcNAc and grown overnight at 30°C . The next day the plate supplemented with GlcNAc contained approximately twice the number of colonies found on plates without GlcNAc . Colonies from GlcNAc plate were patched on LB , LB supplemented with GlcNAc , or LB supplemented with GlcNAc and kanamycin . Approximately half of the isolates did not proliferate in the absence of GlcNAc . Of these , all were sensitive to kanamycin and contained mutant glmS as confirmed by PCR . In order to check for production of ACNQ by different microbial strains - S . oneidensis MR-1 , its isogenic menC , menA , and glmS mutants , V . cholerae V52 , E . coli K12 , its isogenic menB , menA , and menC knockout mutants , Bacteroides fragilis ( DSM 20481 ) , Lactococcus lactis lactis ( ATCC 25285 ) - each was grown in 1 L cultures . For absolute quantification , 100 μL of overnight LB cultures of S . oneidensis MR-1 , isogenic menC , and menA mutants , E . coli , and V . cholerae were each used to inoculate 500 mL of lactate minimal media . Cultures were grown under either anaerobic ( B . fragilis and L . lactis lactis ) or aerobic ( S . oneidensis , V . cholerae , E . coli ) conditions for 96 hr . The spent supernatant was then isolated by centrifugation ( 14 , 000 rcf , 30 min ) , filter-sterilized ( 0 . 2 μm , Thermo Scientific , cat #: 595–4520 ) , and acidified to a pH of 6 . 5 with 12 . 1 M HCl ( aq ) . The cell pellets were freeze-dried and weighed to quantify total biomass of each culture . The spent supernatant was then passed through a glass column containing 1 g of Strata-X-A resin ( Phenomenex , 33 μm , 85 Å ) that was prepped by washing with 15 mL of 100% MeOH and subsequently 15 mL of 100% H2O . Upon passing all of the spent supernatant through the column , the resin was washed with 12 mL of both 25 mM ammonium acetate ( aq ) pH 6 . 5% and 100% methanol . The dried crude extract was resuspended in 200 μL of 50% ACN/H2O and 5 μL was quantified using the same HR-LCMS method described above . Integrated extracted ion chromatograms ( EIC ) for two ion adducts of ACNQ - [M + H]+ and [M-H2O + H]+ - were summed and compared to a six-point standard curve of synthetically-derived ACNQ in order to obtain the absolute production quantification . Recovery percentages of ACNQ from supernatant to crude extract were also quantified by spiking in 10 nM of 15N-ACNQ into the spent supernatant of MR-1 cultures prior to the centrifuge step . Cultures ( 100 mL ) of bacteria were grown with biological replicates under aerobic conditions in lactate minimal media . Each culture was inoculated with 200 μL of an overnight culture grown in LB and incubated at 30°C , 200 rpm for 72 hr . After which each culture was centrifuged ( 3900 rcf , 10 min ) , the supernatant decanted , and the cell pellet was placed on the lyophilizer for 24 hr . The dried cell pellet was transferred and weighed in a 40 mL glass vial , ground , and then extracted with 4 . 5 mL of 2:1 dichloromethane ( DCM ) /MeOH for 2 hr while stirring with a magnetic stir bar . The organic solvent was filtered using a glass plug containing celite and dried under vacuum . The crude material was then resuspended in 300 μL of 2:1 isopropanol ( IPA ) /MeOH and 5 μL of this mixture was analyzed on the HRLCMS where the menaquinone analogs were quantified using the Phenomenex Luna 5 μm C5 100 Å ( 50 × 4 . 6 mm ) under the following method: hold 100% solvent A for 5 min then quickly gradient to 80% solvent A/20% solvent B over 0 . 1 min , then gradient to 100% solvent B over 34 . 9 min with a flow rate of 0 . 4 mL/min ( solvent A: 95% H2O/5% MeOH +0 . 1% FA with 5 mM ammonium acetate , solvent B: 60% IPA/35% MeOH/5% H2O + 0 . 1% FA with 5 mM ammonium acetate ) . Integrated extracted ion chromatograms for two ion adducts , [M + H]+ and [M+NH4]+ , for each menaquinone analog were summed and compared to a four-point standard curve of commercially available menaquinone-4 ( Sigma Aldrich ) in order to obtain the absolute production quantification . Overnight cultures were grown in LB were centrifuged ( 3900 rcf , 10 min ) and the pellets were rinsed twice with 10 mL DI water . The pellets were then resuspended in 10 mL DI water , where the optical density ( OD ) was recorded . Larger cultures ( 3 × 50 mL ) were inoculated to a starting OD of 0 . 05 and grown under the following conditions: 1 ) aerobically with 30 nM sodium fumarate for 48 hr , 2 ) anaerobically with 30 nM sodium fumarate for 96 hr , and 3 ) anaerobically with 20 mM trimethylamine N-oxide ( TMAO ) for 96 hr in the presence and absence of ACNQ . At either 48 or 96 hr , the OD was measured , and each culture was centrifuged ( 3900 rcf , 10 min ) , decanted , rinsed with 10 mL DI water , and re-pelleted . The rinsed pellets were subsequently placed on the lyophilizer for 24 hr and processed using the same protocol described in as the absolute quantification of menaquinone . Dual-chamber electrochemical reactors were used for all the electrochemical experiments . The reactors consisted of a graphite felt working electrode ( 6 . 35 mm thick with 16 mm radius , Alfa Aesar ) and an Ag/AgCl ( 3 M KCl , CHI111 , CH Instruments ) reference electrode in the anode chamber , a Titanium wire ( Alfa Aesar ) counter electrode in the cathode chamber , chambers were separated by a cation exchange membrane ( CMI-7000 , Membranes International , Ringwood , NJ ) . To maintain anaerobic conditions , bioreactors were continuously purged with N2 gas . All the experiment were tested under 30°C . Electrolyte contained M9 growth medium ( BD ) , 50 mM D , L-lactate . Bacteria were cultured aerobically in LB broth at 30°C with 200 rpm shaking overnight . Before injection into the bioreactor , the cells were washed twice with M9 media to remove the LB media . Chronoamperometry , cyclic voltammetry , and differential pulse voltammetry were carried out using a Bio-Logic Science Instruments potentiostat model VSP-300 . For the chronoamperometry test , the applied potential was set at −0 . 1 V versus Ag/AgCl , electric currents are reported as a function of the geometric surface area of the electrode . Cyclic voltammetry measurements in the potential region of −0 . 8 to +0 . 6 V versus Ag/AgCl and a scan rate of 10 mV s−1 ( if there is no additional declaration ) . Differential pulse voltammetry measurement was performed with the same scan window , 50 mV pulses height , 500 ms pulses width , 1 mV step height and 1000 ms step time . | In order to survive , we break down food through a series of chemical reactions that release energy to power our cells . In these metabolic reactions , small electrically charged particles called electrons are removed from the food molecule , and transferred , via a series of reactions , to a terminal electron acceptor . For humans and many other organisms , oxygen is the terminal electron acceptor . Bacteria generate energy through a similar series of chemical reactions , but many species of bacteria live in environments where oxygen is absent . Some bacteria solve this problem by transferring the electrons released in their metabolic reactions to acceptor compounds in the external environment . These species must therefore employ a small molecule ‘shuttle’ to carry the electrons to the acceptor . Previous work has shown the bacterial strain Shewanella oneidensis MR-1 releases a small molecule into its surrounding environment , which serves as its electron shuttle . Despite identifying a mutant strain of MR-1 that cannot produce this shuttle , researchers have been unable to determine the exact chemical identity of this critical molecule . Now , Mevers , Su et al . have identified this elusive electron shuttle . This involved growing MR-1 and isolating the active molecule which restores the mutant bacteria’s ability to shuttle electrons . Further experiments characterizing the structure of this compound using techniques involving analytical and synthetic organic chemistry revealed it be a small molecule known as ACNQ . Mevers , Su et al . showed MR-1 produces this elusive electron shuttle by releasing a precursor structure into the environment where it spontaneously converts into ACNQ . As a result , there are no genes present in the genome of MR-1 or other bacterial strains that are required for the production of ACNQ . This genetic absence and low production levels of ACNQ has frustrated previous attempts to identify MR-1’s electron shuttle . Bacterial metabolism is studied for its applications in bioenergy ( producing renewable energy using living organisms ) and bioremediation ( detoxification of substances using the reactions of bacterial metabolism ) . A better understanding of bacterial metabolism is thus essential for the continued development of these useful technologies . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | An elusive electron shuttle from a facultative anaerobe |
Evolutionary changes in the anatomy and physiology of the female reproductive system underlie the origins and diversification of pregnancy in Eutherian ( ‘placental’ ) mammals . This developmental and evolutionary history constrains normal physiological functions and biases the ways in which dysfunction contributes to reproductive trait diseases and adverse pregnancy outcomes . Here , we show that gene expression changes in the human endometrium during pregnancy are associated with the evolution of human-specific traits and pathologies of pregnancy . We found that hundreds of genes gained or lost endometrial expression in the human lineage . Among these are genes that may contribute to human-specific maternal–fetal communication ( HTR2B ) and maternal–fetal immunotolerance ( PDCD1LG2 ) systems , as well as vascular remodeling and deep placental invasion ( CORIN ) . These data suggest that explicit evolutionary studies of anatomical systems complement traditional methods for characterizing the genetic architecture of disease . We also anticipate our results will advance the emerging synthesis of evolution and medicine ( ‘evolutionary medicine’ ) and be a starting point for more sophisticated studies of the maternal–fetal interface . Furthermore , the gene expression changes we identified may contribute to the development of diagnostics and interventions for adverse pregnancy outcomes .
Evolutionary changes in the ontogeny of anatomical systems are ultimately responsible for their functional conservation and transformation into new tissue and organ systems ( novelties ) with new physiological functions that are outside of the range of the ancestral ones ( innovations ) . These same evolutionary and developmental histories limit ( constrain ) the range of genetic and environmental perturbations those physiological functions can accommodate before leading to dysfunction and disease ( i . e . , their reaction norms ) . Evolution of the structures and functions of female reproductive system and extraembryonic fetal membranes , for example , underlie the evolution of pregnancy ( Armstrong et al . , 2017; Hou et al . , 2009; Kin et al . , 2015; Lynch et al . , 2015; Lynch et al . , 2008 ) and likely adverse pregnancy outcomes such as infertility ( Cummins , 1999 ) , recurrent spontaneous abortion ( Kosova et al . , 2015 ) , preeclampsia ( Carter , 2011; Crosley et al . , 2013; Elliot , 2017; Rosenberg and Trevathan , 2007 ) , and preterm birth ( LaBella et al . , 2020; Marinić et al . , 2021; Plunkett et al . , 2011; Swaggart et al . , 2015 ) . Thus , reconstructing the evolutionary and developmental history of the cells , tissues , and organs involved in pregnancy may elucidate the ontogenetic origins and molecular etiologies of adverse pregnancy outcomes . Extant mammals span major stages in the evolution and diversification of pregnancy , including the origins of maternal provisioning ( matrotrophy ) , placentation , and viviparity ( Behringer et al . , 2006; Freyer et al . , 2003; Freyer and Renfree , 2009; Hughes and Hall , 1998; Renfree , 1995; Renfree and Shaw , 2013 ) . Eutherian mammals have also evolved a complex suite of traits that support prolonged pregnancies such as an interrupted estrous cycle , maternal recognition of pregnancy , maternal–fetal communication , immunotolerance of the antigenically distinct fetus , and implantation of the blastocyst into maternal tissue ( Abbot and Rokas , 2017 ) . There is also considerable variation in pregnancy traits within Eutherians . Catarrhine primates , for example , have evolved spontaneous decidualization ( differentiation ) of endometrial stromal fibroblasts ( ESFs ) into decidual stromal cells ( DSCs ) under the combined action of progesterone , cyclic adenosine monophosphate ( cAMP ) , and other unknown maternal signals ( Carter and Mess , 2017; Gellersen et al . , 2007; Gellersen and Brosens , 2003; Kin et al . , 2016; Kin et al . , 2015; Mess and Carter , 2006 ) , deeply invasive interstitial hemochorial placentas ( Carter et al . , 2015; Pijnenborg et al . , 2011a; Pijnenborg et al . , 2011b; Soares et al . , 2018 ) , menstruation ( Burley , 1979; Emera et al . , 2012b; Finn , 1998; Strassmann , 1996 ) , and a unique but unknown parturition signal ( Csapo , 1956; Csapo and Pinto-Dantas , 1965 ) . Humans have evolved interstitial trophoblast invasion , in which the blastocyst is embedded and encased entirely within the uterine endometrium ( McGowen et al . , 2014; Norwitz et al . , 2001; Salamonsen , 1999 ) , shorter interbirth intervals ( Galdikas and Wood , 1990 ) , and longer pregnancy and labor ( Bourne , 1970; Keeling and Roberts , 1972 ) than other primates . Humans also are particularly susceptible to pregnancy complications such as preeclampsia ( Crosley et al . , 2013; Elliot , 2017; Marshall et al . , 2018 ) , and preterm birth ( Phillips et al . , 2015; Rokas et al . , 2020; Wildman et al . , 2011 ) than other primates . Gene expression changes ultimately underlie the evolution of anatomical structures , suggesting that gene expression changes at the maternal–fetal interface underlie these primate- and human-specific pregnancy traits . Therefore , we used comparative transcriptomics to reconstruct the evolutionary history of gene expression in the pregnant endometrium and identify genes that gained ( ‘recruited genes’ ) or lost endometrial expression in the primate and human lineages . We found genes that evolved to be expressed at the maternal–fetal interface in the human lineage were enriched for immune functions and diseases such as preterm birth and preeclampsia , as well as other pathways not previously implicated in pregnancy . We explored the function of three recruited genes in greater detail , which implicates them in a novel signaling system at the maternal–fetal interface ( HTR2B ) , maternal–fetal immunotolerance ( PDCD1LG2 ) , and remodeling of uterine spiral arteries and deep placental invasion ( CORIN ) . These data indicate that explicit evolutionary studies can identify genes and pathways essential for the normal healthy functions of cells , tissues , and organs , and that likely underlie the ( dys ) function of those tissue and organ systems .
To identify gene expression gains and losses in the endometrium that are phylogenetically associated with derived pregnancy traits in humans and catarrhine primates , we assembled a collection of transcriptomes from the pregnant or gravid endometrium of 20 Eutherian mammals , including human ( Homo sapiens ) , baboon ( Papio anubis ) , Rhesus monkey ( Macaca mulatta ) , and Pig-Tailed macaque ( Macaca nemestrina ) , three Marsupials , platypus , three birds , and six lizard species , including species that are both oviparous and viviparous ( Figure 1 and Figure 1—source data 1 ) . The complete dataset includes expression information for 21 , 750 genes and 33 species , which were collected at different gestational times from early- to midpregnancy , by multiple labs , and sequencing methods . Thus , differences in transcript abundance between samples may reflect biological differences in mRNA abundances between gestational ages or species , differences in sequencing protocols , or other technical factors unrelated to the biology of pregnancy ( i . e . , batch effects ) . Therefore , we transformed quantitative gene expression values coded as transcripts per million ( TPM ) into discrete character states such that genes with TPM ≥2 . 0 were considered expressed ( state = 1 ) , genes with TPM <2 . 0 were considered not expressed ( state = 0 ) , and missing genes coded as unknown ( ? ; Box 1 ) . Consistent with significant noise reduction , multidimensional scaling ( MDS ) of species based on gene expression levels ( TPMs ) was essentially random ( Figure 1—figure supplement 1A ) , whereas MDS of the binary encoded dataset grouped species by phylogenetic relatedness ( Figure 1—figure supplement 1B ) . Next , we used the binary encoded dataset to reconstruct ancestral transcriptomes and trace the evolution of gene expression gains ( 0 → 1 ) and losses ( 1 → 0 ) . Ancestral states were inferred with the empirical Bayesian method implemented in IQ-TREE 2 ( Minh et al . , 2020; Nguyen et al . , 2015 ) using the species phylogeny ( Figure 1A ) and the GTR2 + FO + R4 model of character ( Soubrier et al . , 2012 ) . Interested readers are referred to other publications for more information about ancestral reconstruction methods ( Joy et al . , 2016; Pauling et al . , 1963 ) . Internal branch lengths of the gene expression tree were generally very short while terminal branches were much longer , indicating pronounced species-specific divergence in endometrial gene expression ( Figure 1A ) . MDS of extant and ancestral transcriptomes ( Figure 1B ) generally grouped species by phylogenetic relationships , parity mode , and degree of placental invasiveness . For example , grouping platypus , birds , and reptiles ( cluster 1 ) , viviparous mammals with noninvasive placentas such as opossum , wallaby , and horse , pig , and cow ( clusters 2 and 3 ) , and Eutherians with placentas such as mouse , rabbit , and armadillo ( cluster 4 ) . Human , baboon , Rhesus monkey , and Pig-Tailed macaque formed a distinct group from other Eutherians ( cluster 5 ) , indicating that catarrhine primates have an endometrial gene expression profile during pregnancy that is distinct even from other Eutherians ( Figure 1B ) . We identified 923 genes that gained endometrial expression in the human lineage with Bayesian posterior probabilities ( BPPs ) ≥0 . 80 ( Figure 2—source data 2; Box 2 ) . These genes are enriched in 54 pathways , 102 biological process Gene Ontology ( GO ) terms , and 91 disease ontologies at a false discovery rate ( FDR ) ≤0 . 10 ( Figure 2 ) . Among enriched pathways were ‘GPCRs , Class A Rhodopsin-like’ , ‘Signaling by GPCR’ , ‘Cytokine–cytokine receptor interaction’ , ‘Allograft Rejection’ , and ‘Graft-versus-host disease’ . The majority of enriched GO terms were related to signaling processes , such as ‘cAMP-mediated signaling’ and ‘serotonin receptor signaling pathway’ or to the immune system , such as ‘acute inflammatory response’ and ‘regulation of immune system process’ . The majority of enriched disease ontologies were related to the immune system , such as ‘Autoimmune Diseases’ , ‘Immune System Disease’ , ‘Inflammation’ , ‘Asthma’ , ‘Rheumatic Diseases’ , ‘Dermatitis’ , ‘Celiac Diseases’ , and ‘Organ Transplantation’ , as well as ‘Pregnancy’ , ‘Pregnancy , First Trimester’ , ‘Infertility’ , ‘Habitual Abortion’ , ‘Chorioamnionitis’ , ‘Pre-Eclampsia’ , and ‘Preterm Birth’ , consistent with observations that women with systemic autoimmune diseases have an elevated risk of delivering preterm ( Kolstad et al . , 2020 ) . Seven hundred and seventy-one genes lost endometrial expression in the lineage with BPP ≥0 . 80 ( Figure 2—source data 3 ) . These genes were enriched in 48 pathways , 42 biological process GO terms , and 3 disease ontologies at FDR ≤0 . 10 ( Figure 2 ) . Enriched pathways included ‘immune system’ , ‘pregnancy’ , ‘pregnancy first trimester’ , ‘infertility’ , ‘habitual abortion’ , ‘preeclampsia’ , and ‘preterm birth’ . Unlike genes that gained endometrial expression in the human lineage , those that lost endometrial expression were enriched in disease ontologies unrelated to the immune system , but did include ‘Preterm Birth’ , as well as ‘Selenocysteine Synthesis’ and ‘Selenoamino Acid Metabolism’ , the latter two which have been previously implicated in preterm birth by genome-wide association study ( GWAS; Zhang et al . , 2017 ) . In stark contrast , genes that gained ( + ) or lost ( − ) endometrial expression during pregnancy in the stem lineage of primates ( +63/−34 ) did not include terms related to the immune system or pregnancy . Thus , genes that gained or lost endometrial expression in the human lineage are uniquely related to immune regulatory process , autoimmunity , inflammation , and allograft rejection , signaling processes such as cAMP-mediated and serotonin receptor signaling , and well as adverse pregnancy outcomes . The maternal–fetal interface is composed of numerous maternal and fetal cell types including endometrial stromal lineage cells ( perivascular , EFSs , and DSCs ) , uterine natural killer cells ( uNKs ) , decidual macrophage ( uMP ) , dendritic cells ( DCs ) , T helper cells ( Th cells ) , regulatory T cells ( Tregs ) , various innate lymphoid cells ( ILCs ) , and multiple trophoblast cell types ( Suryawanshi et al . , 2018; Vento-Tormo et al . , 2018; Wang et al . , 2020 ) . To infer if genes recruited into endometrial expression in the human lineage are enriched in specific cell types , we used a previously published single-cell RNA-Seq ( scRNA-Seq ) dataset generated from the first trimester human decidua ( Vento-Tormo et al . , 2018 ) to identify cell types at the maternal–fetal interface ( Figure 3A; Figure 3—figure supplement 1 ) . Next , we determined the observed fraction of human recruited genes expressed in each cell type compared to the expected fraction and used a two-way Fisher exact test to identify cell types that were significantly enriched in human recruited genes . Remarkably , human recruited genes were enriched in five of six endometrial stromal lineage cells , including perivascular endometrial mesenchymal stem cells ( pvEMSCs ) and four populations of DSCs , as well as plasmocytes , endothelial cells ( ECs ) , DCs , and extravillus cytotrophoblasts ( Figure 3B ) . Consistent with these findings , the expression of human recruited genes defines distinct cell types at the maternal–fetal interface ( Figure 3—figure supplement 2 ) . Our observation that human recruited genes have predominantly remodeled the transcriptome of endometrial stromal lineage cells prompted us to explore the development and gene expression evolution of these cell types in greater detail . Pseudotime single-cell trajectory analysis of endometrial stromal lineage cells identified six distinct populations corresponding to a perivascular mesenchymal stem cell like endometrial stromal population ( pvEMSC ) population and five populations of DSCs ( DSC1–5 ) , as well as cells between pvEMSCs and DSCs that likely represent nondecidualized ESFs and ESFs that have initiated decidualization ( Figure 4A and B ) . In addition , ESFs that decidualize branch into two distinct lineages , which we term lineage 1 DSCs ( DSC1–DSC3 ) and lineage 2 DSCs ( DSC4 and DSC5 ) ( Figure 4A and B ) . These cell populations differentially express human recruited genes ( Figure 4C ) , which are dynamically expressed during differentiation of perivascular cells ( PVCs ) into lineage 1 and 2 DSCs ( Figure 4D ) . Genes that were recruited into endometrial expression in the human lineage are enriched the serotonin signaling pathway ( Figure 2 ) , but a role for serotonin signaling in the endometrium has not previously been reported . Among the recruited genes in this pathway is the serotonin receptor HTR2B . To explore the history of HTR2B expression in the endometrium in greater detail , we plotted extant and ancestral gene expression probabilities on tetrapod phylogeny and found that it independently evolved endometrial expression at least seven times , including in the human lineage ( Figure 5A ) . To investigate which cell types express HTR2B , we used the scRNA-Seq dataset from the first trimester maternal–fetal interface and found that HTR2B expression was almost entirely restricted to the DSC cluster ( Figure 5B ) . We further explored the expression dynamics HTR2B during decidualization using a scRNA-Seq time-course dataset ( Lucas et al . , 2020 ) and found that its expression is transiently downregulated during the initial inflammatory decidual phase but upregulated upon the emergence of decidual cells and senescence decidual cells after 4 days of differentiation ( Figure 5—figure supplement 1 ) . HTR2B was also the only serotonin receptor expressed in either human ESFs or DSCs at TPM ≥2 ( Figure 5B and Figure 5—figure supplement 2A ) and was highly expressed in uterine tissues ( Figure 5—figure supplement 2B ) . Additionally , we found that HTR2B was only expressed in human and mouse ESFs , but not in ESFs at TPM ≥2 from other species in a previously generated multispecies ESF RNA-Seq dataset ( Figure 5C ) . Pseudotime single-cell trajectory analysis of endometrial stromal lineage cells indicates that HTR2B is expressed in most lineage 1 DSCs , which coexpress other genes such as IL15 , INSR , and PRDM1 ( Figure 5D ) ; HTR2B is also expressed by a minority of lineage 2 DSCs , ESFs , and PVCs ( Figure 5D ) . One hundred and ninety-four genes were differentially expressed between HTR2B+ and HTR2B− DSCs ( Figure 5E ) . These genes were enriched in numerous pathways including ‘Regulation of Insulin-like Growth Factor ( IGF ) transport and uptake by Insulin-like Growth Factor Binding Proteins ( IGFBPs ) ’ , ‘Complement and coagulation cascades’ , ‘BMP2–WNT4–FOXO1 Pathway in Human Primary Endometrial Stromal Cell Differentiation’ , ‘IL-18 signaling pathway’ , and disease ontologies including ‘Small-for-dates baby’ , ‘Premature Birth’ , ‘Inflammation’ , ‘Fetal Growth Retardation’ , ‘Pregnancy Complications’ , ‘Hematologic Complications’ , and ‘Spontaneous abortion’ ( Figure 5F and Figure 5—source data 1 ) . To determine if HTR2B expression was regulated by progesterone , we used previously published RNA-Seq data from human ESFs and ESFs differentiated into DSCs with cAMP/progesterone ( Mazur et al . , 2015 ) . HTR2B was highly expressed in ESFs and downregulated during differentiation ( decidualization ) by cAMP/progesterone into DSCs ( Figure 6A and Figure 6—figure supplement 1 ) . HTR2B has hallmarks of an expressed gene in DSCs , including residing in a region open chromatin assessed by previously published FAIRE-Seq data ( Figure 6B ) , an H3K4me3 and H3K27ac marked promoter and polymerase II binding , as well as a promoter that makes long-range loops to binding sites for transcription factors that orchestrate decidualization such as the progesterone receptor A isoform ( PGR-A ) , FOXO1 , FOSL2 , GATA2 , and NR2F2 ( COUP-TFII ) in previously published ChIP-Seq data ( see methods ) ( Figure 6B ) . The HTR2B promoter also makes several long-range interactions to transcription factor-bound sites as assessed by H3K27ac HiChIP data generated from a normal hTERT-immortalized endometrial cell line ( E6E7hTERT; see methods ) ( Figure 6B ) . Consistent with regulation by these transcription factors , knockdown of PGR , FOXO1 , and GATA2 upregulated HTR2B in DSCs ( Figure 6C ) . HTR2B is also differentially regulated throughout menstrual cycle ( Figure 6D ) and pregnancy ( Figure 6E ) , and is expressed in DSCs in the endometrium during the window of implantation ( Figure 6—figure supplement 2 ) . To test if human ESFs and DSCs were responsive to serotonin , we transiently transfected each cell type with reporter vectors that drive luciferase expression in response to activation the AP1 ( Ap1_pGL3-Basic[minP] ) , MAPK/ERK ( SRE_pGL3-Basic[minP] ) , RhoA GTPase ( SRF_pGL3-Basic[minP] ) , and cAMP/PKA ( CRE_pGL3-Basic[minP] ) signaling pathways , and used a Dual Luciferase Reporter assay to quantify luminescence 6 hr after treatment with either 0 , 50 , 200 , or 1000 μM serotonin . Two pathway reporters were responsive to serotonin: ( 1 ) the serum response element ( SRE ) reporter in DSCs treated with 1000 μM serotonin ( unpaired mean difference between is 1 . 35 [95 . 0% CI 0 . 624 , 2 . 69] , two-sided permutation t-test p = 0 . 00 ) ; and ( 2 ) the cAMP/PKA response element ( CRE ) reporter in ESFs treated with 1000 μM serotonin ( unpaired mean difference between is 0 . 296 [95 . 0% CI 0 . 161 , 0 . 43] , two-sided permutation t-test p = 0 . 00 ) and in DSCs treated with 50 μM ( unpaired mean difference = −10 . 1 [95% CI −13 . 8 , −6 . 28] , two-sided permutation t-test p = 0 . 001 ) , 200 μM ( unpaired mean difference = 17 . 4 [95% CI 11 . 6 , 24 . 6] , two-sided permutation t-test p = 0 . 0004 ) , and 1000 μM serotonin ( unpaired mean difference is 16 . 7 [95 %CI 7 . 67 , 26 . 8] , two-sided permutation t-test p = 0 . 006 ) ( Figure 6F and Figure 6—figure supplement 3 ) . Human recruited genes are enriched numerous immune pathway ( Figure 2 ) , among these genes are the PD-1 ligand PDCD1LG2 ( PD-L2 ) ( Figure 7A ) . We found that PDCD1LG2 was expressed by several cell types at the first trimester maternal–fetal interface , including DCs , macrophages , ESFs and DSCs , and multiple trophoblast lineages ( Figure 7B ) , and is highly expressed in uterine tissues ( Figure 7—figure supplement 1 ) . While each of these cell-type populations has individual cells with high-level PDCD1LG2 expression , only 3%–5% of DSCs , 3 % of DCs , 14 % of macrophage , and 66 % of cytotrophoblasts express PDCD1LG2 ( Figure 7C ) . Consistent with recent recruitment in the human lineage , PDCD1LG2 was highly expressed in human but either moderately or not expressed in ESFs from other species ( Figure 7D; Figure 5—figure supplement 1 ) . The human PDCD1LG2 locus has the hallmarks of an actively expressed gene , such as a promoter marked by H3K27ac , H3K4me3 , and H3K4me1 , and binding sites for several transcription factors in previously published ChIP-Seq data from DSCs ( Figure 7E ) . The PDCD1LG2 promoter also makes several long-range interactions to transcription factor-bound sites , including downstream site that is in the region of open chromatin and bound by PGR/GATA/FOXO1 ( Figure 7E ) . PDCD1LG2 was highly expressed in ESFs and DSCs ( Figure 7F ) but downregulated by cAMP/progesterone treatment ( Figure 7G ) . Knockdown of PGR and FOXO1 up- and downregulated PDCD1LG2 in DSCs , respectively ( Figure 7G ) . PDCD1LG2 introns also contain several single nucleotide polymorphisms ( SNPs ) previously associated with gestational duration and number of lifetime pregnancies as assessed by GWAS ( Aschebrook-Kilfoy et al . , 2015; Sakabe et al . , 2020; Zhang et al . , 2017 ) , albeit with marginal p values , implicating PDCD1LG2 in regulating gestation length ( Figure 7E ) . Among the human recruited genes enriched in disease ontologies related to preeclampsia ( Figure 2 ) is CORIN ( Figure 8A ) , a serine protease which promotes uterine spiral artery remodeling and trophoblast invasion ( Cui et al . , 2012; Yan et al . , 2000 ) . We found that CORIN was exclusively expressed by a subset of endometrial stromal lineage cells ( Figure 8B and C ) , dramatically upregulated in DSCs by cAMP/progesterone treatment ( Figure 8D ) , and highly expressed in uterine tissues ( Figure 8—figure supplement 1; Figure 5—figure supplement 1 ) . The CORIN locus has hallmarks of an actively expressed gene in DSCs , including a promoter in a region of open chromatin assessed by previously published ATAC- and DNase-Seq data and marked by H3K4me3 in previously published ChIP-Seq data ( Figure 8E ) . The CORIN promoter also makes long-range interactions to transcription factor-bound sites as assessed by HiChIP , including an upstream site bound by PGR , FOSL2 , GATA2 , FOXO1 , and NR2F2 in previously published ChIP-Seq data from DSCs ( Figure 8E ) . Consistent with these observations , knockdown of PGR , NR2F2 , and GATA2 downregulated CORIN expression in DSCs ( Figure 8F ) .
The maternal–fetal interface is composed of numerous maternal cell types , all which could have been equally impacted by genes that were recruited into endometrial expression in the human lineage . It is notable , therefore , that the expression of these genes is predominately enriched in endometrial stromal lineage cells , including perivascular mesenchymal stem cells and multiple populations of DSCs . These data suggest that remodeling of the transcriptome and functions of the endometrial stromal cell lineage has played a particularly important role in the evolution of human-specific pregnancy traits . It is also interesting to note that DSCs evolved in the stem lineage of Eutherian mammals ( Carter and Mess , 2017; Gellersen et al . , 2007; Gellersen and Brosens , 2003; Kin et al . , 2016; Kin et al . , 2015; Mess and Carter , 2006 ) , coincident with a wave of gene expression recruitments and losses that also dramatically remodeled their transcriptomes ( Kin et al . , 2015; Lynch et al . , 2015 ) . Thus , the endometrial stromal cell lineage has repeatedly been the target of evolutionary changes related to pregnancy , highlighting the importance of DSCs in the origins and divergence of pregnancy traits . These data also suggest that endometrial stromal lineage cells may play a dominant role in the ontogenesis of adverse pregnancy outcomes . Unexpectedly , human recruited genes are enriched in the serotonin signaling pathway , such as the serotonin receptor HTR2B . Though a role for serotonin in the endometrium has not previously been reported , we found that serotonin treatment effected RAS/MAPK ( ERK ) and cAMP/PKA signaling pathways , which are essential for decidualization , and that HTR2B is dynamically expressed during menstrual cycle and pregnancy , reaching a low at term . Previous studies have shown that the human placenta is a source of serotonin throughout gestation ( Clark et al . , 1980; Kliman et al . , 2018; Laurent et al . , 2017; Ranzil et al . , 2019; Rosenfeld , 2020 ) . Remarkably , a body of early literature suggests serotonin might trigger parturition . For example , levels of both serotonin ( 5-HT ) and 5-hydroxyindoleacetic acid ( 5-HIAA ) , the main metabolite of serotonin , are highest in amniotic fluid near term and during labor ( Jones and Pycock , 1978; Koren et al . , 1961; Loose and Paterson , 1966; Tu and Wong , 1976 ) while placental monoamine oxidase activity ( which metabolizes serotonin ) is lowest at term ( Koren et al . , 1965 ) . Furthermore , a single dose of the monoamine oxidase inhibitor paraglyline hydrochloride can induce abortion in humans and other animals ( Koren et al . , 1966 ) Consistent with a potential role in regulating gestation length and parturition , use of selective serotonin reuptake inhibitors is associated with preterm birth ( Eke et al . , 2016; Grzeskowiak et al . , 2012; Huybrechts et al . , 2014; Ross et al . , 2013; Sujan et al . , 2017; Yonkers et al . , 2012 ) . 5-HIAA also inhibits RAS/MAPK signaling , potentially by competing with serotonin for binding sites on serotonin receptors ( Chen et al . , 2011; Klein et al . , 2018; Schmid et al . , 2015 ) . Collectively , these data suggest a mechanistic connection between serotonin/5-HTAA , and the establishment , maintenance , and cessation of pregnancy . Among the genes with immune regulatory roles that evolved endometrial expression in the human lineage is the programmed cell death protein 1 ( PD-1 ) ligand PDCD1LG2 . PD-1 , a member of the immunoglobulin superfamily expressed on T cells and pro-B cells , regulates a critical immune checkpoint that plays an essential role in downregulating immune responses and promoting self-tolerance by suppressing T-cell inflammatory activity ( Patsoukis et al . , 2020 ) . PD-1 has two ligands , CD274 ( PD-L1 ) and PDCD1LG2 ( PD-L2 ) , which upon binding PD-1 promote apoptosis in antigen-specific T cells and inhibit apoptosis in anti-inflammatory Tregs ( Patsoukis et al . , 2020 ) . Unlike CD274 , which is constitutively expressed at low levels in numerous cell types and induced by IFN-gamma , PDCD1LG2 expression is generally restricted to professional antigen-presenting cells ( APCs ) such as DCs and macrophages and has a fourfold stronger affinity for PD-1 than does CD274 ( Ghiotto et al . , 2010; Latchman et al . , 2001; Sharpe et al . , 2007; Sharpe and Pauken , 2018 ) Remarkably , this higher affinity emerged in the Eutherian stem lineage ( Philips et al . , 2020 ) . These data suggest that a subpopulation of human DSCs have co-opted some of the immune regulatory functions of professional APCs , which may have been significantly augmented in the human lineage . While more mechanistic studies will help define the precise role of decidual cells in the establishment and maintenance of maternal–fetal immunotolerance , a role for decidual PDCD1LG2 in pregnancy is strongly suggested by its association with variants linked to gestational length and number of lifetime pregnancies ( parity ) ( Aschebrook-Kilfoy et al . , 2015; Sakabe et al . , 2020; Zhang et al . , 2017 ) . Placental invasiveness varies dramatically in Eutherians , but the cellular and molecular mechanisms responsible for this variation are ill defined . One of the genes that may play a role in the evolution of deeply invasive hemochorial placentation is the serine protease CORIN , which converts pro-atrial natriuretic peptide ( pro-ANP ) to biologically active ANP ( Yan et al . , 2000 ) . CORIN-mediated ANP production in the uterus during pregnancy has been shown to promote spiral artery remodeling and trophoblast invasion ( Cui et al . , 2012 ) . These data implicate co-option of CORIN into endometrial expression may have contributed to the evolution of particularly deep trophoblast invasion and extensive spiral artery remodeling in humans and other great apes ( Carter et al . , 2015; Pijnenborg et al . , 2011a; Pijnenborg et al . , 2011b; Soares et al . , 2018 ) . CORIN expression is also significantly lower in patients with preeclampsia than in normal pregnancies ( Cui et al . , 2012 ) , suggesting that the co-option of CORIN into human endometrium may predispose humans to preeclampsia . Additional evolutionary and molecular studies will be required to establish a mechanistic connection between the co-option of CORIN into the endometrium , the evolution of hemochorial placentation , and the origins of preeclampsia in the human lineage . A limitation of this study is our inability to determine with precise phylogenetic resolution the lineages in which some gene expression changes occurred . For example , we lack pregnant endometrial samples from Hominoids other than humans ( chimpanzee/bonobo , gorilla , orangutan , and gibbon/siamang ) , thus we are unable to identify truly human-specific gene expression changes . Similarly , we lack endometrial gene expression data from multiple human populations exposed to differing environmental stresses , and therefore are unable to determine the range of physiologically ‘normal’ gene expression or the reaction norms of individual and collective gene expression levels . Our functional genomic and experimental studies are also restricted to an in vitro cell culture system , which makes it difficult to assess the in vivo impact of gene expression changes on the biology of pregnancy . These limitations are not unique to our study and impact virtually all investigations of Hominoid development and disease , particularly the ones of human-specific traits . Endometrial organoids and iPSC-derived ESFs , however , are promising systems in which to study the development of these traits and disease susceptibility that circumvents the limitations of studying human biology ( Abbas et al . , 2020; Boretto et al . , 2017; Marinić et al . , 2020; Rawlings et al . , 2021; Turco et al . , 2017 ) . Our gene expression dataset also represents only a snapshot in time of gestation , rather than a comprehensive time course of endometrial gene expression throughout gestation . Interestingly however , the expression changes we identified from these early time points are enriched in disease ontology terms related to adverse pregnancy outcomes that span the length of gestation including infertility , recurrent spontaneous abortion , preeclampsia , and preterm birth . These findings suggest that atypical gene expression patterns and physiological changes at the earliest stages , perhaps even processes occurring in the endometrium before pregnancy ( e . g . , decidualization of ESFs into DSCs ) , may predispose to multiple adverse outcomes , including those at the latter stages like preterm birth ( birth before 37 weeks ) . An important focus of future studies should be collecting endometrial samples across species and from multiple stages of pregnancy , particularly close to term , when the mechanisms that maintain gestation cease and those that initiate parturition are likely to be activated . We found that hundreds of genes gained or lost endometrial expression in humans , including genes that may contribute to a previously unknown maternal–fetal communication system ( HTR2B ) , enhanced mechanisms for maternal–fetal immunotolerance ( PDCD1LG2 also known as PD-L2 ) , and deep placental invasion ( CORIN ) . These results demonstrate that gene expression changes at the maternal–fetal interface likely underlie human-specific pregnancy traits and adverse pregnancy outcomes . Our work also illustrates the importance of evolutionary studies for investigating human-specific traits and diseases . This ‘evolutionary forward genomics’ approach complements traditional forward and reverse genetics in model organisms , which may not be relevant in humans , as well as commonly used methods for characterizing the genetic architecture of disease , such as quantitative trait mapping and GWASs . Specifically , our data demonstrate the importance of evolutionary medicine for a mechanistic understanding of endometrial ( dys ) function , and suggest that similar studies of other tissue and organ systems will help identify genes underlying normal and pathological anatomy and physiology . We anticipate that our results will further the synthesis of evolution and medicine and may contribute to the development of interventions for adverse pregnancy outcomes such as preterm birth .
Anatomical terms referring to the glandular portion of the female reproductive tract ( FRT ) specialized for maternal–fetal interactions or shell formation are not standardized . Therefore , we searched the NCBI BioSample , Sequence Read Archive ( SRA ) , and Gene Expression Omnibus ( GEO ) databases using the search terms ‘uterus’ , ‘endometrium’ , ‘decidua’ , ‘oviduct’ , and ‘shell gland’ followed by manual curation to identify those datasets that included the region of the FRT specialized for maternal–fetal interaction or shell formation . Datasets that did not indicate whether samples were from pregnant or gravid females were excluded , as were those composed of multiple tissue types . For all RNA-Seq analyses , we used Kallisto ( Bray et al . , 2016 ) version 0 . 42 . 4 to pseudoalign the raw RNA-Seq reads to reference transcriptomes ( see Figure 1—source data 1 for accession numbers and reference genome assemblies ) and to generate transcript abundance estimates . We used default parameters bias correction , and 100 bootstrap replicates . Kallisto outputs consist of transcript abundance estimates in TPM , which were used to determine gene expression . To ensure that human decidua samples were free from trophoblast contamination , we compared the expression of placental enriched genes in RNA-Seq data from human placenta , a human ESF cell line , a human decidual stromal ( DSC ) cell line , and human first trimester decidua . These results suggest that there is likely no trophoblast contamination of human first trimester decidua samples ( Table 1 ) , thus inferences of gene expression gains in the human lineage are unlikely to be the result of trophoblast contamination . Next , we compared two different gene expression metrics to reconstruct the evolutionary history of endometrial gene expression: ( 1 ) TPM , a quantitative measure of gene expression that reflects the relative molar ratio of each transcript in the transcriptome; and ( 2 ) binary encoding , a discrete categorization of gene expression that classifies genes as expressed ( state = 1 ) or not expressed ( state = 0 ) . For binary encoding we transformed transcript abundance estimates into discrete character states , such that genes with TPM ≥2 . 0 were coded as expressed ( state = 1 ) , genes with TPM <2 . 0 were coded as not expressed ( state = 0 ) , and genes without data in specific species coded as missing ( state = ? ) ; see Box 1 for a detailed justification of the TPM ≥2 cutoff . The TPM coded dataset grouped species randomly ( Figure 1—figure supplement 1A ) , whereas the binary encoded endometrial gene expression dataset generally grouped species by phylogenetic relatedness ( Figure 1—figure supplement 1B ) , suggesting greater signal to noise ratio than raw transcript abundance estimates . Therefore , we used the binary encoded endometrial transcriptome dataset to reconstruct ancestral gene expression states and trace the evolution of endometrial gene expression changes across vertebrate phylogeny ( Figure 1A ) . Orthology assessment was inferred using Ensembl Compara . Ancestral states for each gene were inferred with the empirical Bayesian method implemented in IQ-TREE 2 ( Minh et al . , 2020; Nguyen et al . , 2015 ) using the species phylogeny shown in Figure 1A and the best-fitting model of character evolution determined by ModelFinder ( Kalyaanamoorthy et al . , 2017 ) . The best-fitting model was inferred to be the General Time Reversible model for binary data ( GTR2 ) , with character state frequencies optimized by maximum likelihood ( FO ) , and a FreeRate model of among site rate heterogeneity with four categories ( R4 ) ( Soubrier et al . , 2012 ) . We used ancestral transcriptome reconstructions to trace the evolution of gene expression gains ( 0 → 1 ) and losses ( 1 → 0 ) from the last common ancestor of mammals through to the Hominoid stem-lineage limiting our inferences to reconstructions with BPPs ≥0 . 80 ( Figure 1A and Figure 1—source data 2 ) . Ancestral reconstructions with BPP ≥0 . 80 were excluded from over representation analyses . We used classical MDS to explore the structure of extant and ancestral transcriptomes . MDS is a multivariate data analysis method that can be used to visualize the similarity/dissimilarity between samples by plotting data points ( in this case transcriptomes ) onto two-dimensional plots . MDS returns an optimal solution that represents the data in a two-dimensional space , with the number of dimensions ( k ) specified a priori . Classical MDS preserves the original distance metric , between data points , as well as possible . MDS was performed using the veganR package ( Oksanen et al . , 2019 ) with four reduced dimensions . Transcriptomes were grouped using K-means clustering with K = 2–6 , K = 5 optimized the number of distinct clusters and cluster memberships ( i . e . , correctly grouping species by phylogenetic relationship , parity mode , and placenta type ) . Maternal–fetus interface 10× Genomics scRNA-Seq data were retrieved from the E-MTAB-6701 entry as a processed data matrix ( Vento-Tormo et al . , 2018 ) . The RNA counts and major cell-type annotations were used as provided by the original publications . Seurat ( v3 . 1 . 1 ) ( Butler et al . , 2018 ) , implemented in R ( v3 . 6 . 0 ) , was used for filtering , normalization , and cell types clustering . The subclusters of cell types were annotated based on the known transcriptional markers from the literature survey . Briefly , we performed the following data processing steps: ( 1 ) cells were filtered based on the criteria that individual cells must be expressing at least 1000 and not more than 5000 genes with a count ≥1; ( 2 ) cells were filtered out if more than 5 % of counts mapping to mitochondrial genes; ( 3 ) data normalization was performed by dividing uniquely mapping read counts ( defined by Seurat as unique molecular identified [UMI] ) for each gene by the total number of counts in each cell and multiplying by 10 , 000 . These normalized values were then log-transformed . Cell types were clustered by using the top 2000 variable genes expressed across all samples . Clustering was performed using the ‘FindClusters’ function with essentially default parameters , except resolution was set to 0 . 1 . The first 20 PCA dimensions were used in the construction of the shared-nearest neighbor ( SNN ) graph and the generation of two-dimensional embeddings for data visualization using UMAP . Major cell types were assigned based on the original publication samples' annotations , and cell subtypes within major cell types were annotated using the subcluster markers obtained from the above parameters . We then chose the decidual and PV cells to perform the single-cell trajectory , pseudotime analysis , and cell ordering along an artificial temporal continuum analysis using Monocle2 ( Qiu et al . , 2017 ) . The top 500 differentially expressed genes were used to distinguish between the subclusters of decidua and PV cell populations on pseudotime trajectory . The transcriptome from every single cell represents a pseudo-time point along an artificial time vector that denotes decidual and PV lineages' progression , respectively . To compare the differentially expressed genes between HTR2B-positive and HTR2B-negative cells , we first divided the decidual and PV datasets into those groups of cells that either express HTR2B with a count ≥1 and those with zero counts . We then performed differentially expressed genes analysis between the mentioned two groups of cells using the bimodal test for significance . To calculate the enrichment score of human-gain genes in each cell type , we first transformed the data into a pseudobulk expression matrix by averaging all genes' expression in each cell type . We then calculated the fraction of human-gained genes expressed ( Observed ) and the proportion of the rest of the genes expressed in each cell type ( Expected ) . The enrichment ratio shown on the plot is the ratio of Observed and Expected values for each cell type . The p value was calculated using a two-way Fisher exact test followed by Bonferroni correction . Transcriptomic dynamics of human endometrium in vivo . Data mined from publicly available database reproductivecellatlas . org ( Garcia-Alonso et al . , 2021 ) . Data show a cellular map of the human endometrium from combinatorial transcriptomics ( scRNA-Seq and single-nuclei RNA sequencing [snRNA-Seq] ) alongside spatial transcriptomics methods ( 10× Genomics Visium slides and high-resolution microscopy ) representing 98 , 568 cells from fifteen individuals grouped into five main cellular types . No reuse allowed without permission . Single-cell analysis of peri-implantation endometrium . Six LH-timed endometrial biopsies were processed for Droplet generation and single-cell sequencing ( Drop-Seq ) as described in Lucas et al . , 2020 . Anonymized endometrial biopsies were obtained from women aged between 31 and 42 years with regular cycles , body mass index between 23 and 32 kg/m2 , and the absence of uterine pathology on transvaginal ultrasound examination . t-Distributed stochastic neighbour embedding ( t-SNE ) analysis assigned 2831 cells to four clusters , designated based on canonical marker genes as ECs ( n = 141 ) , epithelial cells ( EpC; n = 395 ) , immune cells ( IC; n = 352 ) , and ESFs ( n = 1943 ) . Data are available in the GEO repository GSE127918 . Vento-Tormo et al . , 2018 dataset consists of transcriptomes for ~70 , 000 individual cells of many different cell types , including: three populations of tissue resident decidual natural killer cells ( dNK1 , dNK2 , and dNK3 ) , a population of proliferating natural killer cells ( dNKp ) , type two and/or type three ILCs ( ILC2/ILC3 ) , three populations of decidual macrophages ( dM1 , dM2 , and dM3 ) , two populations of DCs ( DC1 and DC2 ) , granulocytes ( Gran ) , T cells ( TCells ) , maternal and lymphatic endothelial cells ( Endo ) , two populations of epithelial glandular cells ( Epi1 and Epi2 ) , two populations of PVCs ( PV1 and PV2 ) , two ESF populations ( ESF1 and ESF2 ) , and DSCs , placental fibroblasts ( fFB1 ) , extravillous- ( EVT ) , syncytio- ( SCT ) , and villus- ( VCT ) cytotrophoblasts ( Figure 3 and Figure 3—figure supplements 1 and 2 ) . We note that Vento-Tormo et al . identified five populations of cells in the endometrial stromal lineage , including two perivascular populations ( likely reflecting the mesenchymal stem cell-like progenitor of ESFs and DSCs ) and three cell types they call ‘decidual stromal cells’ and label ‘S1–3’ . However , based on the gene expression patterns of ‘dS1–3’ ( shown in Vento-Tormo et al . Figure 3a ) , only ‘dS3’ are decidualized , as indicated by expression of classical markers of decidualization such and PRL ( Tabanelli et al . , 1992 ) and IGFBP1/2/6 ( Tabanelli et al . , 1992; Kim et al . , 1999 ) . In stark contrast , ‘dS1’ do not express decidualization markers but highly express markers of ESFs such as TAGLN and ID2 , as well as markers of proliferating ESFs including ACTA2 ( Kim et al . , 1999 ) . ‘dS2’ also express ESFs markers ( TAGLN , ID2 , ACTA2 ) , but additionally LEFTY2 and IGFBP1/2/6 , consistent with ESFs that have initiated the process of decidualization . These data indicate that the ‘dS1’ and ‘dS2’ populations are both ESFs , but ‘dS2’ are ESFs that have initiated decidualization ( because they express IGFBPs but not PRL ) , and that ‘dS3’ are DSCs . Vento-Tormo et al . show that the differences in gene expression between ‘dS1–3’ are related to their topography in the endometrium , but degree of decidualization ( ‘dS1’/ESF1 < ‘dS2’/ESF2 < ‘dS3’/DSC ) is also linked to differential gene expression . Consistent with this , other scRNA-Seq studies have identified two ESF populations and one DSC population in the first trimester decidua , and used pseudotime analyses to show that they represent different states of differentiation from ESFs to mature DSCs ( Suryawanshi et al . , 2018 ) . Therefore , we prefer to use the perivascular/ESF/DSC nomenclature because it more accurately reflects the biology and gene expression profile of these cell types than the ‘dS1–3’ naming convention . We also note that while it is generally thought that ESFs are absent from the pregnant uterus , ESFs retain a presence in the endometrium from the first trimester until term ( Richards et al . , 1995; Suryawanshi et al . , 2018; Muñoz-Fernández et al . , 2019; Sakabe et al . , 2020 ) . We used WebGestalt v . 2019 ( Liao et al . , 2019 ) to identify enriched ontology terms using overrepresentation analysis ( ORA ) . We used ORA to identify enriched terms for three pathway databases ( KEGG , Reactome , and Wikipathway ) , three disease databases ( Disgenet , OMIM , and GLAD4U ) , and a custom database of genes implicated in preterm birth by GWAS . The preterm birth gene set was assembled from the NHGRI-EBI Catalog of published GWASs ( GWAS Catalog ) , including genes implicated in GWAS with either the ontology terms ‘Preterm Birth’ ( EFO_0003917 ) or ‘Spontaneous Preterm Birth’ ( EFO_0006917 ) , as well as two recent preterm birth and birth weight GWASs ( Warrington et al . , 2019; Sakabe et al . , 2020 ) using a genome-wide significant p value of 9 × 10–6 . The custom gmt file used to test for enrichment of preterm birth associated genes is included as a supplementary data file to ( Figure 2 , Figure 2—source data 1 ) . We used previously published ChIP-Seq data generated from human DSCs that were downloaded from NCBI SRA and processed remotely using Galaxy ( Afgan et al . , 2016 ) . ChIP-Seq reads were mapped to the human genome ( GRCh37/hg19 ) using HISAT2 ( Kim et al . , 2019; Kim et al . , 2015; Pertea et al . , 2016 ) with default parameters and peaks called with MACS2 ( Feng et al . , 2012; Zhang et al . , 2008 ) with default parameters . Samples included PLZF ( GSE75115 ) , H3K4me3 ( GSE61793 ) , H3K27ac ( GSE61793 ) , H3K4me1 ( GSE57007 ) , PGR ( GSE69539 ) , the PGR-A and -B isoforms ( GSE62475 ) , NR2F2 ( GSE52008 ) , FOSL2 ( GSE69539 ) , FOXO1 ( GSE69542 ) , PolII ( GSE69542 ) , GATA2 ( GSE108408 ) , SRC-2/NCOA2 ( GSE123246 ) , AHR ( GSE118413 ) , ATAC-Seq ( GSE104720 ) , and DNase1-Seq ( GSE61793 ) . FAIRE-Seq peaks were downloaded from the UCSC genome browser and not recalled . We also used previously published RNA-Seq and microarray gene expression data generated from human ESFs and DSCs that were downloaded from NCBI SRA and processed remotely using Galaxy platform ( https://usegalaxy . org/; Version 20 . 01 ) for RNA-Seq data and GEO2R for microarray data . RNA-Seq datasets were transferred from SRA to Galaxy using the Download and Extract Reads in FASTA/Q format from NCBI SRA tool ( version 2 . 10 . 4+ galaxy1 ) . We used HISAT2 ( version 2 . 1 . 0+ galaxy5 ) to align reads to the Human hg38 reference genome using single- or paired-end options depending on the dataset and unstranded reads , and report alignments tailored for transcript assemblers including StringTie . Transcripts were assembled and quantified using StringTie ( v1 . 3 . 6 ) ( Pertea et al . , 2016; Pertea et al . , 2015 ) , with reference file to guide assembly and the ‘reference transcripts only’ option , and output count files for differential expression with DESeq2/edgeR/limma-voom . Differentially expressed genes were identified using DESeq2 ( Love et al . , 2014 ) ( version 2 . 11 . 40 . 6+ galaxy1 ) . The reference file for StringTie guided assembly was wgEncodeGencodeBasicV33 . GEO2R performs comparisons on original submitter-supplied processed data tables using the GEOquery and limma R packages from the Bioconductor project ( https://bioconductor . org/ ) . These datasets included gene expression profiles of primary human ESFs treated for 48 hr with control nontargeting , PGR-targeting ( GSE94036 ) , FOXO1-targeting ( GSE94036 ) , or NR2F2 ( COUP-TFII ) -targeting ( GSE47052 ) siRNA prior to decidualization stimulus for 72 hr; transfection with GATA2-targeting siRNA was followed immediately by decidualization stimulus ( GSE108407 ) . Probes were 206638_at ( HTR2B ) , 220049_s_at ( PDCD1LG2 ) , and 220356_at ( CORIN ) for GSE4888 ( endometrial gene expression throughout menstrual cycle ) and for GSE5999 ( gene expression in basal plate throughout gestation ) . Multispecies RNA-Seq analysis of ESFs and DSCs is from GSE67659 . To assess chromatin looping , we utilized a previously published H3K27ac HiChIP dataset from a normal hTERT-immortalized endometrial cell line ( E6E7hTERT ) and three endometrial cancer cell lines ( ARK1 , Ishikawa , and JHUEM-14 ) ( O’Mara et al . , 2019 ) . Endometrial biopsies were fixed overnight in 10% neutral buffered formalin at 4°C and wax embedded in Surgipath Formula ‘R’ paraffin using the Shandon Excelsior ES Tissue processor ( Thermo Fisher ) . Tissues were sliced into 3 μm sections on a microtome and adhered to coverslips by overnight incubation at 60°C . Deparaffinization and rehydration were performed through xylene , 100% isopropanol , 70% isopropanol , and distilled water incubations . Following antigen retrieval , slides were washed , blocked , and incubated in primary HTR2B antibody ( 1:200; Fisher Scientific ) overnight at 4°C . After washing three times , slides were incubated with Alexa Fluor 594 ( 1:1000; Fisher Scientific ) for 2 hr , washed and mounted in ProLong Gold → Antifade Reagent with DAPI ( Cell Signaling Technology ) . Slides were visualized using the EVOS Auto system , with imaging parameters maintained throughout image acquisition . Human hTERT-immortalized endometrial stromal fibroblasts ( T-HESC; CRL-4003 , ATCC ) were grown in maintenance medium , consisting of Phenol Red-free DMEM ( 31053-028 , Thermo Fisher Scientific ) , supplemented with 10% charcoal-stripped fetal bovine serum ( 12676029 , Thermo Fisher Scientific ) , 1% L-glutamine ( 25030-081 , Thermo Fisher Scientific ) , 1% sodium pyruvate ( 11360070 , Thermo Fisher Scientific ) , and 1× insulin–transferrin–selenium ( ITS; 41400045 , Thermo Fisher Scientific ) . A total of 104 ESFs were plated per well of a 96-well plate , 18 hr later cells were transfected in Opti-MEM ( 31985070 , Thermo Fisher Scientific ) with 100 ng of luciferase reporter plasmid , 10 ng Renilla control plasmid , 0 . 25 μl of Lipofectamine LTX ( 15338100 , Thermo Fisher Scientific ) and 0 . 1 μl Plus Reagent as per the manufecturer’s protocol per well; final volume per well was 100 μl . Luciferase reporter plasmids were synthesized ( GenScript ) by cloning the response elements from the pGL4 . 29[luc2P/CRE/Hygro] , pGL4 . 44[luc2P/AP1-RE/Hygro] , pGL4 . 33[luc2P/SRE/Hygro] , and pGL4 . 34[luc2P/SRF-RE/Hygro] plasmids into pGL3-Basic[minP] luciferase reporter . Unlike the pGL4 series vectors ( Promega ) that are hormone responsive , pGL3-Basic[minP] luciferase reporter includes a minimal promoter but is not hormone responsive . Final pathway reporter plasmids are: CRE_pGL3-Basic[minP] ( cAMP/PKA ) , AP1_pGL3-Basic[minP] ( AP1 ) , SRE_pGL3-Basic[minP] ( MAPK/ERK ) , and SRF_RE_pGL3-Basic[minP] ( serum response factor ) . ESFs were incubated in the transfection mixture for 6 hr . Then , ESFs were washed with warm PBS and incubated in the maintenance medium overnight . The next day , the medium in half of the wells was exchanged for the differentiation medium consisting of DMEM with Phenol Red and GlutaMAX ( 10566-024 , Thermo Fisher Scientific ) , supplemented with 2 % fetal bovine serum ( 26140-079 , Thermo Fisher Scientific ) , 1 % sodium pyruvate ( 11360070 , Thermo Fisher Scientific ) , 1 μM medroxyprogesterone 17-acetate ( MPA; M1629 , Sigma Aldrich ) , and 0 . 5 mM 8-Bromoadenosine 3′ , 5′-cyclic monophosphate ( 8-Br-cAMP; B5386 , Sigma Aldrich ) . After 48 hr , serotonin ( 5-HT; H9523 , Sigma Aldrich ) was added to the wells with both maintenance and differentiation medium ( for each in triplicates ) in the following concentrations: 50 μM , 200 μM , and 1 mM; vehicle control ( 0 μM ) was water . After 6 hr of incubation , we used a Dual Luciferase Reporter Assay ( Promega ) to quantify luciferase and Renilla luminescence following the manufacturer’s Dual Luciferase Reporter Assay protocol . Human hTERT-immortalized endometrial stromal fibroblasts were purchased from ATCC ( T-HESC; CRL-4003 , ATCC ) . Their identity has been authenticated by ATCC , and was determined to be mycoplasma free . | Pregnancy is a complicated process . It has three phases: the body recognizes the embryo , it maintains the pregnancy , and finally , it induces labor . These stages happen in all mammals , but the details are different in humans . Human pregnancy and labor last longer . We menstruate . Our placentas invade deeper into the uterus , and the cues that signal pregnancy is done and induce labor are different than in most other mammals . We are also more likely to have pregnancy complications , including infertility , a dangerous rise in blood pressure called preeclampsia , and premature birth . The reasons for these differences are unknown . Human pregnancy relies on close communication between the placenta and the uterus . The immune system must allow the placenta to grow large enough to support the developing embryo , and blood vessels need to adapt to supply gases and nutrients and to remove waste . Understanding how the genes used by the human uterus are different to those used in other species could help explain why human pregnancies are so unusual . Mika , Marinić et al . compared the genes used by the pregnant human uterus to those used in 32 other species , including monkeys , marsupials and other mammals , birds , and reptiles . The analysis revealed that the humans use almost a thousand genes that other animals do not . These genes have roles in the invasion of the placenta , the growth of blood vessels , and control of the immune system . Several have links to the hormone serotonin , which had not been connected with the uterus before . Mika , Marinić et al . suggest that it might control the length of pregnancy , the timing of labor , and communication between parent and baby . The genes identified here provide a starting point for further investigation of human pregnancy . In the future , this may help to prevent or treat infertility , preeclampsia , or premature birth . A possible next step is to examine our closest living relatives , the great apes . Performing similar experiments using tissues or cells from chimpanzees , gorillas , and orangutans could reveal more about the genes unique to human pregnancy . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2021 | Evolutionary transcriptomics implicates new genes and pathways in human pregnancy and adverse pregnancy outcomes |
Spontaneous glutamate release-driven NMDA receptor activity exerts a strong influence on synaptic homeostasis . However , the properties of Ca2+ signals that mediate this effect remain unclear . Here , using hippocampal neurons labeled with the fluorescent Ca2+ probes Fluo-4 or GCAMP5 , we visualized action potential-independent Ca2+ transients in dendritic regions adjacent to fluorescently labeled presynaptic boutons in physiological levels of extracellular Mg2+ . These Ca2+ transients required NMDA receptor activity , and their propensity correlated with acute or genetically induced changes in spontaneous neurotransmitter release . In contrast , they were insensitive to blockers of AMPA receptors , L-type voltage-gated Ca2+ channels , or group I mGluRs . However , inhibition of Ca2+-induced Ca2+ release suppressed these transients and elicited synaptic scaling , a process which required protein translation and eukaryotic elongation factor-2 kinase activity . These results support a critical role for Ca2+-induced Ca2+ release in amplifying NMDA receptor-driven Ca2+ signals at rest for the maintenance of synaptic homeostasis .
Studies in the last decade have shown that spontaneous release events trigger biochemical signaling leading to maturation and stability of synaptic networks , local dendritic protein synthesis and control postsynaptic responsiveness during homeostatic synaptic plasticity ( Chung and Kavalali , 2006; Sutton et al . , 2006; Kavalali , 2015 ) . Most surprisingly , these studies have demonstrated that postsynaptic excitatory receptor blockade or inhibition of neurotransmitter release in addition to action potential blockade induces faster and more pronounced homeostatic synaptic potentiation ( Sutton et al . , 2006; Nosyreva et al . , 2013 ) . There is evidence that alterations in resting Ca2+ signaling , partly triggered via activation of NMDA receptors at rest , is critical for these effects ( Wang et al . , 2011; Nosyreva et al . , 2013 ) . However , to date there is no direct information on the properties of dendritic Ca2+ signals elicited by spontaneous release events under physiological circumstances . Our group's previous work as well as others used electrophysiology to show that , indeed , under physiological levels of extracellular Mg2+ spontaneous miniature excitatory postsynaptic currents ( mEPSCs ) possess a sizable NMDA receptor-mediated component , indicating that NMDA receptors signal at rest under physiological conditions without requiring local AMPAR-mediated dendritic depolarizations ( Espinosa and Kavalali , 2009; Povysheva and Johnson , 2012; Gideons et al . , 2014 ) . The existence of an NMDA component within mEPSCs agrees with earlier estimates of incomplete Mg2+ block of canonical NMDA receptors near resting membrane potentials , and therefore it does not necessarily involve NMDA receptor subunits with altered Mg2+ sensitivity ( Jahr and Stevens , 1990 ) . Nevertheless , the NMDA receptor Ca2+ influx under these conditions is estimated to be small , corresponding to approximately 20% of the full Ca2+ influx carried by unblocked NMDA receptors ( Espinosa and Kavalali , 2009 ) . Therefore , as NMDA receptor-driven Ca2+ signals at rest are expected to be relatively minor in magnitude , it remains unclear how their blockade could be critical in producing homeostatic synaptic scaling . To address this question , we visualized the resting NMDA receptor-driven Ca2+ signals and found that they are amplified by a Ca2+-induced Ca2+ release mechanism to elicit downstream signaling events . Importantly , based on this information , we also show that direct suppression of these resting Ca2+ signals is sufficient to elicit eEF2 kinase dependent postsynaptic scaling .
To detect transient Ca2+ signals that occur under resting conditions—in the absence of action potentials—we took advantage of the Ca2+ indicator dye Fluo-4 or the Ca2+ sensitive fluorescent protein GCaMP5K as reporters ( Gee et al . , 2000; Akerboom et al . , 2012 ) . To visualize synapses , both reporters were used on hippocampal neurons that were infected with lentivirus expressing the fusion protein Synaptobrevin2-mOrange ( Syb2-mOrange ) consisting of a chimera of the synaptic vesicle protein synaptobrevin2 with the pH sensitive red-shifted fluorophore mOrange ( Ramirez et al . , 2012 ) ( Figure 1A–D ) . In these experiments , the signal contribution of Syb2-mOrange during live imaging is negligible ( see Figure 2—figure supplement 1 ) . In Fluo-4 experiments , neurons were initially incubated and labeled with the membrane permeable analog of Fluo-4 ( Fluo-4 AM ) ( Figure 1A ) followed by dye removal and perfusion with a Tyrode's solution containing 2 mM Ca2+ , 1 . 25 mM Mg2+ as well as 1 μM tetrodotoxin ( TTX ) to block action potentials . Fluorescence images were collected at a frequency of 10 Hz and fluorescence intensity traces were generated for the regions of interest ( ROIs ) selected over Syb2-mOrange puncta which fluorescence was maximized at the end of each experiment using 50 mM NH4Cl ( Figure 1E ) . Under these conditions , we could detect rapid Ca2+ transients ( miniature spontaneous calcium transients or mSCTs ) with absolute values that were at least 2 standard deviations above the mean of the preceding baseline period ( 2 s ) ( Figure 1F ) . These events occurred at a frequency of 0 . 32 ± 0 . 04 min−1 per ROI , consistent with earlier estimates of the frequency of spontaneous fusion events per release site ( Leitz and Kavalali , 2014 ) . Repeating the same experimental protocol with Fluo-4 AM in the absence of Mg2+ did not yield a significantly different mSCT frequency ( Figure 1F ) suggesting that under physiological Mg2+ concentrations we could detect a majority of mSCTs . Interestingly , even though the presence of extracellular Mg2+ is expected to greatly diminish NMDAR current magnitudes ( Espinosa and Kavalali , 2009; Gideons et al . , 2014 ) , imaging experiments did not reveal a significant difference in mSCT amplitudes detected in Mg2+ ( 1 . 25 mM Mg2+ ∆F/Fo = 0 . 063 ± 0 . 001 , 0 mM Mg2+ ∆F/Fo = 0 . 067 ± 0 . 002 , p = 0 . 16 , Student's unpaired t-test , N = 825 events from 8 experiments ) . The fact that mSCT amplitude was unaffected by extracellular Mg2+ indicates mSCTs measured by Fluo-4 AM were not likely to be solely dependent on NMDA receptor activity . 10 . 7554/eLife . 09262 . 003Figure 1 . Multiple approaches to detect miniature spontaneous Ca2+ transients ( mSCTs ) in the presence of TTX and physiological Mg2+ . ( A ) Loading dissociated rat hippocampal cultures with Fluo-4 AM dye labels all cells on the coverslip ( above ) and produces the largest signal amplitudes , shown as example traces and an average with standard deviation ( below ) N = 38 experiments , 7 cultures . ( B ) Individual neurons were loaded with the salt form of Fluo-4 at the whole cell recording configuration via a pipette containing 200 μM of the dye . N = 4 experiments , 1 culture . ( C ) Low efficiency lipotransfection with the highly sensitive GCaMP5K variant produces sparse labeling of neurons across the coverslip but low signal ( ΔF/F ) amplitudes . N = 5 experiments , 1 culture . ( D ) Lentiviral mediated transfection with GCaMP5K-PSD95 targets the fluorescent construct to the postsynaptic densities of all cells on each coverslip . N = 15 experiments , 4 cultures . ( E ) Example images and trace of a mSCT visualized with Fuo-4 AM and its corresponding Syb2-mOrange puncta . Panels show baseline and peak fluorescence intensity with the arrow marking peak fluorescence intensity of the mSCT . Scale bar 5 µm . ( F ) Frequencies expressed as mSCTs per ROI per minute show the highest efficiency of mSCT detection with Fluo-4 AM . Fluo-4 AM based experiments performed with no Mg2+ in the external solution reported no significant changes in mSCT compared to the presence Mg2+ . The postsynaptically localized reporter GCaMP5K-PSD95 reports statistically lower frequencies when compared to Fluo-4 AM ( Fluo-4 AM 1 . 25 mM Mg2+ vs GCaMP5K-PSD95 1 . 25 mM Mg2+ p = 0 . 0003 , Fluo-4 AM 0 mM Mg2+ vs GCaMP5K-PSD95 1 . 25 mM Mg2+ p = 0 . 0031 , via one-way ANOVA with Holm-Sidak's multiple comparisons ) 0 mM Mg2+ Fluo-4 AM N = 16 experiments , 8 cultures . 0 mM Mg2+ GCaMP5kK-PSD95 N = 10 experiments , 4 cultures . DOI: http://dx . doi . org/10 . 7554/eLife . 09262 . 003 In parallel experiments , we delivered the salt form of Fluo-4 ( 200 µM ) with a patch pipette in the whole-cell recording configuration and performed the same imaging protocol as above ( Figure 1B ) . In this setting , we detected a lower frequency of events ( 0 . 125 ± 0 . 035 min−1 per ROI ) , indicating that some of the mSCTs may be susceptible to postsynaptic dialysis and wash out of soluble factors ( Figure 1F ) . In agreement with this premise , when the same optical recording conditions were applied to neurons expressing a soluble version of the green emission Ca2+ indicator probe GCaMP5K , we could detect a higher frequency of mSCTs ( 0 . 230 ± 0 . 04 min−1 per ROI ) . In subsequent experiments , we expressed a fusion construct of GCaMP5K with the postsynaptic scaffolding protein PSD95 ( GCAMP5K-PSD95 ) in order to target the calcium sensor specifically to the postsynaptic density ( Figure 1D ) . In the presence of extracellular Mg2+ based on the population average this setting provided the lowest estimate for the mSCT frequency ( 0 . 009 ± 0 . 004 min−1 per ROI ) ( Figure 1F ) . In contrast , removal of Mg2+ augmented the mSCT detection rate to a level comparable to the rates we observed with Fluo-4 or soluble GCaMP5K ( Figure 1E ) . This finding suggests that , in the presence of Mg2+ , postsynaptically localized GCaMP5K-PSD95 has limited ability to detect the Ca2+ signals generated in its vicinity via Ca2+ influx . However , experiments in the absence of Mg2+ indicate that this probe is functional and can in principle detect these spontaneous local Ca2+ transients as reported earlier ( Leitz and Kavalali , 2014 ) . Recording in the presence of 1 . 25 mM Mg2+ and 1 µM TTX , we could detect spontaneously generated Ca2+ transients in the dendrites of hippocampal pyramidal cells with all four techniques . Although , each probe reports a different frequency these differences are statistically insignificant except when considering the difference between Fluo-4 AM and GCaMP5K-PSD95 ( Figure 1F ) . Relatively lower detection efficiency of GCAMP5K-PSD95 compared to soluble probes illustrates that the majority of these transients are not localized to the postsynaptic density . The failure of Mg2+ to decrease mSCT amplitudes as measured with Fluo-4 AM strongly suggest that a majority of transients are generated by a signaling process downstream of Ca2+ entry rather than reporting the Ca2+ influx per se . In order to identify the nature of this signaling , in subsequent experiments , we used the Fluo-4 AM based imaging to test conditions that alter mSCTs . To characterize mSCTs , neurons were labeled with Fluo-4 AM as in Figure 1A and imaged in Tyrode's solution containing TTX ( Figure 2A ) . Ca2+ transients were detected by the slope of the rising phase as well as the peak amplitude . To ensure that these detected peaks were not noise , only mSCTs with a peak amplitude 2 standard deviations greater than the signal average of the previous 2 s were counted . Figure 2B shows the rise and decay times as well as the fluorescence amplitudes of 306 mSCTs identified from 6 experiments . In these experiments , the mean rise time was 0 . 38 s with a median of 0 . 29 s . The mean decay time was 0 . 86 s with a median of 0 . 47 s . The amplitude distribution had an average ∆F/Fo of 0 . 061 with a median of 0 . 049 ( Figure 2B ) . 10 . 7554/eLife . 09262 . 004Figure 2 . Detection and characterization of spontaneous Ca2+ transients in physiological concentrations of Mg2+ . ( A ) Events detected from Fluo-4 AM traces having rising slope greater than 350 fluorescence units/s and a peak ∆F/Fo greater than 0 . 035 were counted if the peak fluorescence value was 2 standard deviations greater than the mean of the signal 2 s previous . Gray shaded region indicates the moving average plus/minus two standard deviations and the red line indicates the 0 . 035 ∆F/Fo threshold . Red trace shows the 2 point slope with the black line as the 350 A . U . /second detection threshold . Arrows indicate peaks that satisfy these criteria . ( B ) Histograms showing rise time ( τ ) , decay time ( τ ) and amplitudes ( ∆F/Fo ) of mSCTs . N = 306 mSCTs from 6 experiments and 2 cultures . ( C , D ) Traces from cells ( C ) and Ca2+ transient frequencies ( D ) were obtained by imaging first in Tyrode's solution containing no Ca2+ , then in Tyrode's containing 2 mM Ca2+ and finally Tyrode's containing 2 mM Ca2+ and the NMDA receptor blocker AP5 . Removal of extracellular Ca2+ or block of the NMDA receptor resulted in a significant reduction in Ca2+ transient frequency ( 2 mM Ca2+ vs 0 mM Ca2+ p = 0 . 038 , 2 mM Ca2+ vs 2 mM Ca2+ + AP5 p = 0 . 038 , via 1-way ANOVA with Holm-Sidak's multiple comparisons test ) . N = 8 experiments , 2 cultures . DOI: http://dx . doi . org/10 . 7554/eLife . 09262 . 00410 . 7554/eLife . 09262 . 005Figure 2—figure supplement 1 . Syb2-mOrange fluorescence does not contaminate Fluo-4 AM signals . ( A ) Example traces from simultaneous Fluo-4 AM / synaptobrevin-mOrange imaging . Cells were imaged alternating the excitation between Fluo-4 AM and Syb2-mO for every other frame . Each frame was exposed for 100 ms as described previously and frames were collected at 5 fps per channel . Events were detected as described previously using the Fluo-4 AM channel and the data for both wavelengths was aligned to the peak value in the green channel . Green traces show example evens detected in the channel excited with 470 ± 40 nm light to excite Fluo-4 AM and the red traces show the corresponding data collected with a 548 ± 10 nm excitation filter to excite Syb2-mOrange . ( B ) Average traces made from 382 detected and aligned events . The average trace in the Fluo-4 AM channel shows a robust Ca2+ transient while the averaged Syb2-mO data shows no appreciable deviation from baseline . DOI: http://dx . doi . org/10 . 7554/eLife . 09262 . 005 Next we tested whether NMDA receptor activity is required for the generation of mSCTs . For this purpose , synaptic ROIs were imaged in three steps . First optical recordings were obtained in Tyrode's solution with nominal Ca2+ containing 1 µM TTX followed by the addition of 2 mM Ca2+ and finally in Tyrode's solution containing TTX + 2 mM Ca2+ + 50 µM AP5 ( Figure 2C ) . In the absence of Ca2+ in the bath , mSCTs were virtually undetectable suggesting that Ca2+ influx is required for their generation . Switching the Tyrode's solution to TTX + 2 mM Ca2+ brought the mSCT frequency back to normal levels , and subsequent addition of the NMDA receptor antagonist AP5 again decreased the mSCT frequency to very low levels that were not statistically different from the nominal Ca2+ condition ( Figure 2D ) . These results indicate that Ca2+ influx through the NMDA receptor is critical for the generation of mSCTs . To examine whether the NMDA receptor openings driving mSCTs were due to spontaneous glutamate release we took two complementary approaches . First , we took advantage of the fact that the acute application of 100 mM hypertonic sucrose is known to produce an increase in mEPSCs ( Fatt and Katz , 1952; Rosenmund and Stevens , 1996 ) . To measure this effect , hippocampal pyramidal cells were voltage clamped at −70 mV while a baseline AMPA mEPSC frequency was collected in Tyrode's solution containing 1 µM TTX , 50 µM PTX and 50 µM AP5 for 2 min . Perfusion was then switched to Tyrode's containing 100 mM hypertonic sucrose as the recording continued for 2 min . Quantification of these recordings revealed a 2 . 6-fold increase in mEPSC frequency upon the addition of hypertonic sucrose ( Figure 3A ) . To test whether the increase in mEPSC frequency could drive an increase in mSCT frequency the experiment was repeated in neurons loaded with Fluo-4 AM . The baseline was collected in Tyrode's solution containing only TTX before changing to a solution containing TTX + 100 mM sucrose . The addition of hypertonic sucrose produced a 2 . 3 fold increase in mSCT frequency ( Figure 3B ) , which supports the hypothesis that spontaneous glutamate release can drive the generation of postsynaptic calcium transients . 10 . 7554/eLife . 09262 . 006Figure 3 . mSCT frequency is correlated with mEPSC frequency . ( A ) Whole cell recordings from WT cells ( left ) show a ∼twofold increase in mEPSC frequency when switched to Tyrode's solution containing 100 mM sucrose ( right ) ( p = 0 . 002 , Student's paired T test , N = 8 cells , from 5 coverslips and 2 cultures ) . ( B ) Example traces ( left ) and quantification of spontaneous Ca2+ transient frequencies measured via imaging show a ∼twofold increase upon application of 100 mM sucrose ( right ) ( p = 0 . 028 , Student's paired T test , N = 9 experiments from 3 cultures ) . ( C ) Fluo-4 example traces from both control and SNAP25 KO animals before and after the application of AP5 . ( D ) Fluo-4 imaging in cultures made from SNAP25 KO and littermate control mice reveal that the KO cultures have a substantially decreased mSCT frequency . In this setting , AP5 treatment greatly decreases but does not completely abolish the remaining mSCTs . ( WT , TTX vs WT , TTX+AP5 p = 0 . 010 . WT , TTX vs KO , TTX p = 0 . 008 . KO TTX vs KO TTX+AP5 p = 0 . 003 , via 1-way ANOVA with Tukey's multiple comparisons , N = 8 experiments in WT cells and 9 experiments in KO cells from 3 cultures ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09262 . 006 Next , to assess whether a decrease in mEPSC frequency would correlate with a decrease in mSCT frequency , we utilized neurons from mice lacking the critical SNARE-mediated fusion machinery component SNAP-25 ( Washbourne et al . , 2002 ) . These mice die at birth; however , hippocampal neurons cultured from embryonic mice form synapses and manifest a virtual absence of evoked neurotransmission and highly diminished rate of spontaneous neurotransmitter release ( Bronk et al . , 2007 ) . In Fluo-4 AM imaging experiments with hippocampal cultures made from littermate control mice , the application of AP5 was able to produce a significant decrease in mSCT frequency compared to baseline recorded in TTX , as had been observed previously in wild-type rat cultures . Neurons derived from SNAP25 knock out animals had a significantly decreased baseline mSCT frequency . Also , these transients remained sensitive to AP5 , which is again consistent with mSCT generation being driven by spontaneous vesicle release ( Figure 3C , D ) . Mature glutamatergic synapses contain both AMPA and NMDA receptors ( Bekkers and Stevens , 1989; Liao et al . , 2001 ) . Therefore , in the next set of experiments we tested whether concurrent AMPA receptor activity augments NMDA receptor activity at rest through electrical means . Such synergy between the activation of the two types of receptors may be facilitated by dendritic spines that possess a high spine neck resistance that render them electrically isolated from the dendritic shaft ( Bloodgood and Sabatini , 2005; Harnett et al . , 2012 ) but see ( Popovic et al . , 2014 ) . In this way activation of AMPA receptors may result in sufficient local depolarization to facilitate relief of adjacent NMDA receptors from Mg2+ block . Additionally , AMPA receptors lacking GluA2 , which are calcium-permeable , could also contribute to these transients ( Hollmann et al . , 1991 ) . To examine the role of AMPA receptor activation on mSCTs , we performed the same analysis above in the presence of AMPA receptor antagonist 2 , 3-dihydroxy-6-nitro-7-sulfamoyl-benzo[f]quinoxaline-2 , 3-dione ( NBQX ) . In these experiments NBQX ( 5 µM ) did not affect mSCT frequency ( Figure 4A ) . This argues against a direct ( e . g . , via calcium-permeable GluA2 lacking receptors ) or indirect ( via local depolarization ) contribution of AMPA receptors to mSCTs ( Figure 4A ) . 10 . 7554/eLife . 09262 . 007Figure 4 . Spontaneous Ca2+ transient generation is decreased by blocking release of Ca2+ release from internal stores but not by blocking the AMPA receptors , L-type Ca2+ channels or group I mGluRs . ( A ) Traces ( top ) from images taken before and after treatment with the AMPA receptor blocker NBQX show no change in Ca2+ transient frequency ( bottom ) ( p = 0 . 78 with Student's paired t-test , N = 9 experiments , 2 cultures ) . ( B ) Imaging in the presence of the specific L-type calcium channel blocker nimodipine ( 5 µM ) does not affect spontaneous Ca2+ transient frequency ( bottom ) ( p = 0 . 89 with Student's paired t-test . N = 7 experiments 1 culture ) . ( C ) Imaging in the presence of mGluR1 and mGluR5 blockers YM202074 and fenobam produce no differences in spontaneous Ca2+ transient frequency ( p = 0 . 26 via Student's t-test . N = 8 cells , 2 cultures ) . ( D ) Application of internal Ca2+ store blocker dantrolene produces a significant drop in mSCT frequency in before/after experiments . ( p = 0 . 008 , Student's paired t-test , N = 9 experiments , 3 cultures ) . ( E ) Fluo-4 AM example traces and frequency quantification from cells recorded in TTX then in TTX + 30 µM Ryanodine . In before/after imaging experiments , 15 min treatment of the use dependent ER Ca2+ channel blocker ryanodine decreased the frequency of observed Ca2+ transients ( p = 0 . 004 , N = 8 via Student's paired t-test ) . ( F ) Example traces and frequency quantification of cells pre treated with ryanodine and then imaged first in Tyrode's solution containing no Mg2+ followed by Tyrode's solution containing 1 . 25 mM Mg2+ . Ca2+ transients are measured in Mg2+ free solution but not in 1 . 25 mM Mg2+ . ( p = 0 . 024 , via Student's unpaired t-test . N = 7 experiments 2 cultures ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09262 . 00710 . 7554/eLife . 09262 . 008Figure 4—figure supplement 1 . Nimodipine produces positive results in a separate assay . ( A ) Example traces showing Ca2+ influx during action potentials in the absence and presence of nimodipine . Cells were loaded with Fluo-4 AM and imaged in Tyrode's solution containing 1 . 25 mM Mg2+ , 2 mM Ca2+ , 50 µM AP5 and 20 µM NBQX . 10 action potentials were evoked via field stimulation at 0 . 2 Hz . Then , the perfusion was changed to include 5 µM of the L-type calcium channel blocker nimodipine for 1 min before repeating the stimulation . ROIs were 20 µm in diameter placed on the cell soma . ( B ) Average peak fluorescence from action potential trains before and after the addition of 5 µM nimodipine . The addition of nimodipine restricts Ca2+ influx during the action potential by ∼27% . N = 8 cells , 1 culture . p = 0 . 019 via Student's paired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 09262 . 00810 . 7554/eLife . 09262 . 009Figure 4—figure supplement 2 . YM202074 and Fenobam produce positive results in a separate assay . ( A ) Cells loaded with Fluo-4 AM were imaged in Tyrode's solution containing 1 . 25 mM Mg2+ , 2 mM Ca2+ and TTX with or without the mGluR1/5 blockers YM202074 and fenobam . The solution was then changed to include 100 µM DHPG ( a mGluR1/5 agonist ) where it elicited large slow Ca2+ influx in the cells no treated with YM202074 and fenobam . ( B ) Peak ∆F/F measurements for the experiment described in panel A shows marked inhibition of the DHPG response in cells treated with mGluR1/5 antagonists . p = 0 . 015 via Student's t-test . N = 8 cells per group from 8 coverslips , 1 culture . DOI: http://dx . doi . org/10 . 7554/eLife . 09262 . 009 Although experiments presented above showed that the NMDA receptor activity is responsible for triggering the majority of mSCTs in response to spontaneous glutamate release , it remains possible that L-type Ca2+ channels may also contribute this activity as they have been shown to open near resting membrane potentials ( Kavalali and Plummer , 1996; Magee et al . , 1996 ) . Therefore , we also tested if L-type Ca2+ channel activity contributed to the mSCT activity . In these experiments , treatment with the L-type Ca2+ blocker nimodipine ( 5 µM ) did not significantly affect mSCT frequency ( Figure 4B ) indicating that these channels do not contribute to the Ca2+ transients . However , here we should note that L-type channel activity may still be involved in setting resting Ca2+ levels and thus impact signaling ( Wang et al . , 2011 ) . Despite producing no change in mSCT frequency in this assay , nimodipine was able to decrease Ca2+ influx in a separate assay ( Figure 4—figure supplement 1 ) . In subsequent experiments we also tested the potential role of Gq-coupled metabotropic glutamate receptor subtypes 1 and 5 in maintenance of Ca2+ transients . These group I mGluRs can affect Ca2+ signaling via activation of phospholipase C and IP3 generation ( Skeberdis et al . , 2001; Topolnik et al . , 2006 ) . However , application of the mGluR1 antagonist YM202074 and mGluR5 antagonist Fenobam did not cause a significant change in mSCT frequency , indicating that activation of these receptors does not contribute to these Ca2+ transients ( Figure 4C ) . Despite producing no change in the measured mSCT frequency , these drugs were shown to be blocking their target receptors in a separate assay ( Figure 4—figure supplement 2 ) . In hippocampal pyramidal cells , NMDA receptor opening by evoked glutamate release elicits larger Ca2+ transients through a Ca2+-induced Ca2+ release mechanism ( Lei et al . , 1992; Emptage et al . , 1999 ) . In this form of signaling , the small Ca2+ transient produced by the NMDA receptor opening raises internal Ca2+ concentrations near ryanodine receptors ( RyRs ) on the endoplasmic reticulum high enough to cause their opening at which point a much larger transient is generated . To test whether this mechanism plays a role in the mSCTs we observe , we imaged with Fluo-4 AM in the presence of dantrolene , which is known to reduce the Ca2+ sensitivity of RyR1 and 3 by blocking their interaction with calmodulin ( Fruen et al . , 1997 ) . Indeed , when compared to the TTX baseline , cells imaged with TTX + dantrolene had a significantly reduced mSCT frequency ( Figure 4D ) . This finding was validated by using ryanodine , which directly blocks all three RyR isoforms in a use-dependent manner ( Hawkes et al . , 1992; Meissner and el-Hashem , 1992 ) . To facilitate use-dependent block of RyRs by ryanodine , the baseline mSCT frequency was first collected in TTX and then cells were perfused for 13 min with a solution containing 30 µM ryanodine without TTX to maximize RyR opening and ryanodine block . The cells were then perfused with Tyrode's containing TTX + ryanodine for 2 min before continuing the recording . Under these conditions , the application of ryanodine produced a ∼fivefold decrease in mSCT frequency ( Figure 4E ) . In addition , the Ca2+ transients that were detectable after ryanodine treatment were substantially decreased in amplitude suggesting that they are likely to be produced by a subpopulation of RyRs that remained unblocked or incompletely blocked ( average TTX ∆F/Fo = 0 . 052 ± 0 . 001 , TTX + Ryanodine ∆F/Fo = 0 . 045 ± 0 . 001 , p = 0 . 002 Student's unpaired t-test , n = 199 events in TTX , 110 events in TTX + ryanodine , 8 experiments , 2 cultures ) . Treatment with dantrolene or ryanodine is presumed to decrease mSCT frequency by blocking RyRs responsible for producing the Ca2+ transient . With these inhibitors present , further NMDA openings can no longer trigger an mSCT . In fact , the efficacy of ryanodine in this case allowed further investigation of the pure NMDA transient under these experimental conditions . We incubated neurons with ryanodine for 15 min to block RyRs and then loaded them with Fluo-4 AM as before . These cells were imaged in Tyrode's solution containing TTX but no extracellular Mg2+ to allow maximal NMDA currents . Under these conditions Ca2+ transients were observed , but when 1 . 25 mM Mg2+ was again added no further transients could be measured ( Figure 4F ) . These results illustrate that under physiological concentrations of Mg2+ , Fluo-4 cannot detect the NMDA Ca2+ transient without further amplification from Ca2+ induced Ca2+ release . In the next set of experiments , we aimed to examine the physiological impact of RyR-dependent mSCTs by focusing on the putative role of these Ca2+ signals in regulation of synaptic efficacy . For this purpose , we investigated the role of mSCTs in homeostatic synaptic scaling , which is a compensatory mechanism where neurons scale the strength of their synaptic inputs multiplicatively in a uniform manner in response to global increases or decreases in activity ( Turrigiano et al . , 1998 ) . This response involves the synthesis and insertion of new AMPA receptors and can be strongly induced by blocking both action potentials and NMDA receptors ( Sutton et al . , 2004 ) . Importantly , although synaptic scaling in response to activity blockade occurs within a time frame of 24–48 hr , suppression of resting synaptic activity mediated by spontaneous neurotransmitter release events results in more rapid synaptic scaling detectable within hours ( Sutton et al . , 2006; Nosyreva et al . , 2013 ) . This suggests that NMDA receptor activation at rest maintains synaptic homeostasis . However , the mechanism by which NMDA receptor activity near resting membrane potentials signals to translation machinery , in particular to eEF2 kinase , has been unclear , especially when one considers the relatively small ion conductance of NMDA receptors at rest due to Mg2+ block ( Espinosa and Kavalali , 2009 ) . To investigate the role of RyR-dependent mSCTs in homeostatic synaptic scaling , hippocampal neurons were incubated for 3 hr in culture media containing TTX + vehicle ( negative control ) , TTX + ryanodine , or TTX + AP5 as positive control . Neurons were then perfused with Tyrode's solution and whole cell voltage clamp recordings were made in 1 µM TTX , 50 µM PTX and 50 µM AP5 to isolate AMPA-mEPSCs . Under these conditions , the amplitude distributions of AMPA-mEPSCs obtained from neurons treated previously with TTX + ryanodine as well as those treated with TTX + AP5 showed a significant rightward shift towards larger amplitudes compared to the control condition ( Figure 5A , B ) . When the collected mEPSC amplitudes were plotted rank order in control vs TTX + ryanodine , a linear fit revealed a scaling factor of 1 . 28 indicating that cell-wide , mEPSC amplitudes increased uniformly 28% over 3 hr with TTX + ryanodine treatment ( Figure 5C ) . This increase in mEPSC amplitudes was not as pronounced as was found with the positive control ( TTX + AP5 ) which may correlate with the finding that ryanodine treatment does not block mSCTs as completely as AP5 ( Figures 2D , 4E ) . It is important to note that while other groups have reported an immediate decrease in mEPSC frequency with the acute application of ryanodine ( Emptage et al . , 1999 ) , in our system the mEPSC frequencies in neurons treated with ryanodine for 15 min were indistinguishable from those incubated with vehicle as control ( TTX mEPSC freq = 7 . 59 Hz ± 1 . 75 , TTX + Ryanodine mEPSC freq = 8 . 92 ± 1 . 13 , p = 0 . 54 using Student's t-test , N = 7 cells from 5 coverslips , 2 cultures ) . Since the acute application of ryanodine does not alter mEPSC frequency in this system we believe the synaptic scaling effect mainly results from ryanodine acting at the postsynapse to block mSCT activity . 10 . 7554/eLife . 09262 . 010Figure 5 . Treating cells with ryanodine + TTX produces a protein synthesis dependent increase in mEPSC frequency and amplitude indicative of homeostatic synaptic scaling . ( A ) Example voltage clamp recordings from cells treated with TTX + vehicle ( negative control , N = 9 cells from 5 coverslips , 3 cultures ) , TTX + ryanodine ( N = 8 cells from 5 coverslips , 4 cultures ) or TTX + AP5 ( positive control , N = 6 cells from 4 coverslips , 2 cultures ) for 3 hr . ( B ) Cumulative probability histogram showing significant rightward shifts ( increases ) in the amplitude of AMPA mEPSCs of cells treated with TTX + ryanodine ( red line , p = 1 . 74 × 10−17 , D = 0 . 151 ) , or TTX + AP5 ( p = 8 . 79 × 10−40 , D = 0 . 255 ) vs control via Kolmogorov–Smirnov test . ( C ) Rank order plot of TTX + vehicle mEPSC amplitudes vs TTX + ryanodine showing a multiplicative scaling factor of 1 . 28 . ( D ) Example voltage clamp recordings from cells pretreated for 30 min with the protein synthesis inhibitor anisomycin and then TTX + vehicle ( N = 6 cells from 5 coverslips , 4 cultures ) or TTX + ryanodine for 3 hr ( N = 7 cells from 5 coverslips , 2 cultures ) . ( E ) Cumulative probability histogram of mEPSC amplitudes shows no significant difference between treatment groups when pretreated with anisomycin ( p = 0 . 078 , D = 0 . 052 via Kolmogorov–Smirnov test ) . ( F ) Rank order plot of mEPSC amplitudes indicates that anisomycin pretreatment abolishes scaling between treatment groups . DOI: http://dx . doi . org/10 . 7554/eLife . 09262 . 010 In earlier experiments homeostatic synaptic scaling that occurs after blockade of resting NMDA receptor activity was shown to rely on protein synthesis , in particular synthesis of new AMPARs rather than the insertion of existing ones ( Sutton et al . , 2006 , 2007 ) . In order to test whether this is the case for RyR block-induced synaptic scaling , we repeated the experiment above with neurons that were treated with the protein synthesis inhibitor anisomycin ( 20 µM ) starting 30 min prior to their 3 hr incubation with TTX . Under these conditions , anisomycin completely abolished the increase in AMPA-mEPSC amplitudes as no significant differences were seen in their distribution after TTX + ryanodine treatment compared to treatment with TTX alone ( Figure 5D–F ) . Previous studies have also shown that a key regulator of protein synthesis , eukaryotic elongation factor 2 ( eEF2 ) , is phosphorylated and inactivated by the Ca2+-dependent eEF2 kinase thus blocking protein synthesis under resting conditions ( Sutton et al . , 2007; Autry et al . , 2011; Nosyreva et al . , 2013; Gideons et al . , 2014 ) . To test whether RyR-mediated mSCTs could be tonically activating eEF2 kinase and thus inhibiting protein synthesis in dendrites , we tested the impact of ryanodine treatment in hippocampal neuronal cultures from eEF2 kinase knockout mice . In hippocampal neurons made from wild-type littermate controls , treating with TTX + ryanodine for 3 hr produced a significant increase in mEPSC amplitudes compared to TTX + vehicle , where plotting the amplitudes in rank order revealed a 42% increase in synaptic strength ( Figure 6A–C ) . When the same experiment was performed using neurons from eEF2 kinase knockout mice , treatment with TTX + ryanodine did not produce a significant shift in mEPSC amplitudes ( Figure 6D , E ) . The rank order plot revealed only a 1% difference in synaptic strength between treatment groups ( Figure 6F ) . Taken together these results suggest that RyR-dependent mSCT-driven signaling acts through Ca2+-dependent eEF2 kinase to maintain synaptic homeostasis . 10 . 7554/eLife . 09262 . 011Figure 6 . Ryanodine treatment does not trigger homeostatic synaptic scaling in eEF2 kinase knockout neurons . ( A ) Example traces from WT littermate mice with and without 3 hr ryanodine treatment . ( B ) Cumulative probability histogram shows a significant shift in mEPSC amplitude in TTX + ryanodine treated animals ( N = 12 cells , from 7 coverslips , 3 cultures ) vs TTX + vehicle control ( N = 13 cells from 8 coverslips , 3 cultures ) ( p = 3 . 32 × 10−14 , D = 0 . 112 via Kolmogorov–Smirnov test ) . ( C ) Rank order plot shows a 1 . 42 fold increase in synaptic strength after ryanodine treatment . ( D ) Example traces from eEF2 kinase KO animals with and without ryanodine treatment . ( E ) Cumulative probability histogram shows no shift in the distribution of mEPSC amplitudes in eEF2K KO animals treated with TTX + ryanodine ( N = 9 cells , from 6 coverslips , 3 cultures ) vs TTX + vehicle ( N = 13 cells from 7 coverslips , 3 cultures ) ( p = 0 . 066 , D = 0 . 058 via Kolmogorov–Smirnov test ) . ( F ) Rank order plot shows no appreciable multiplicative change in synaptic strength in the eEF2 K KO animals with ryanodine treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 09262 . 011
In this study , we took advantage of multiple Ca2+ indicator probes to examine the properties of Ca2+ transients detected in hippocampal neurons in physiological levels of extracellular Mg2+ in the absence of action potentials . These transients are important because they are key to understanding the Ca2+ signaling events occurring at rest that result in regulation of protein translation and gene transcription leading to synaptic plasticity ( Chen et al . , 2014; Lalonde et al . , 2014 ) ( for review see Kavalali , 2015 ) . Under these conditions we detected robust NMDA receptor dependent Ca2+ transients at a rate of 0 . 32 ± 0 . 03 min−1 where previously our group was able to measure a per synapse spontaneous release rate of 0 . 76 ± 0 . 03 min−1 by imaging with presynaptic probes ( Leitz and Kavalali , 2014 ) . The relatively higher release rate measured earlier may suggest that not every release event is able to generate an mSCT . The link between these Ca2+ transients and spontaneous neurotransmitter release was verified by the parallel increase in mSCT and spontaneous neurotransmitter release frequencies in response to application of hypertonic sucrose . Furthermore , the frequency of Ca2+ transients was also significantly diminished in neurons lacking SNAP-25 , which show a substantial reduction in spontaneous release , and in cultures treated with NMDA receptor blockers . Interestingly GCaMP5K-PSD95 , a probe located near the postsynaptic density , revealed only a very small number of events that fulfilled our detection criteria while soluble probes proved to be much better indicators . This observation indicates that although mSCT generation depends on NMDA receptor driven Ca2+ influx , this does not result in strong signals at the postsynaptic density . Rather , mSCTs seem to rely on the activation of Ca2+ release from smooth endoplasmic reticulum which is present in spines or adjacent dendritic regions ( Spacek and Harris , 1997 ) . The generation of mSCTs was not dependent on AMPA receptors , L-type Ca2+ channels or postsynaptic metabotropic glutamate receptor subtypes 1 and 5 . While all of these play a role in Ca2+ dynamics under other circumstances the mSCTs we observed under resting conditions were primarily driven by the coupling of the NMDA receptor to internal Ca2+ stores through the ryanodine receptor , as the application of dantrolene or ryanodine produced a marked reduction in both mSCT frequency and amplitude . Earlier studies performed in hippocampal synapses discovered that unitary evoked EPSCs were accompanied by Ca2+ transients that were only minimally dependent on voltage gated Ca2+ channels or AMPA receptors . However , unlike the mSCTs we observe , the application of ryanodine produced only a small reduction in Ca2+ transient amplitude in these experiments ( Kovalchuk et al . , 2000 ) . This difference may suggest that spontaneous glutamate release-driven Ca2+ transients are more dependent on internal Ca2+ stores compared to Ca2+ transients elicited by evoked release . In this study , we tested a key prediction of these observations on synaptic plasticity by assessing the role of Ca2+-induced Ca2+ release in synaptic scaling triggered at rest . Our experiments showed that the synaptic scaling produced by the blockade of spontaneous NMDA-mEPSCs is also produced by blocking the Ca2+ release from internal stores indicating a strong link between the two signals . The generation of relatively large store-driven Ca2+ transients provides a critical amplification step for the relatively small NMDA-mEPSCs seen under physiological conditions ( Espinosa and Kavalali , 2009; Gideons et al . , 2014 ) . The resulting signal is delocalized and pulsatile which may allow synaptic NMDA receptors to exert signaling influence in the surrounding dendritic regions . This could be critical for local translational control as eEF2 localizes to the dendritic shaft rather than dendritic spines ( Asaki et al . , 2003 ) . The low frequency of observed mSCTs may also be a defining attribute , as the ubiquitous Ca2+ binding protein calmodulin is predicted to interact with different target kinases and enzymes based on the frequency and duration of its activation by free Ca2+ ( Saucerman and Bers , 2008; Slavov et al . , 2013 ) . Taken together these findings identify a critical missing mechanistic link between spontaneous neurotransmission and the control of dendritic signaling events that regulate synaptic efficacy .
Hippocampal cultures from Sprague–Dawley rats or eEF2 kinase knockout mice and their wild-type littermate controls were generated from postnatal day 1–3 male and female pups and plated on Matrigel ( Corning Inc , NY ) coated coverslips as described previously ( Kavalali et al . , 1999 ) . Neurons were infected with lentivirus at 4 days in vitro . Neurons were used for experiments between 14 to 18 days in vitro . Dissociated hippocampal cultures from SNAP25 knockout mice and their wild-type littermates were generated from E17-20 embryos and were plated on poly-d-lysine coated coverslips as described previously ( Bronk et al . , 2007 ) . Neurons were used for experiments 14–18 days in vitro . Dissociated hippocampal cultures aged 14–18 days in vitro were voltage clamped at −70 mV using an Axon Instruments Axopatch 200B amplifier with access resistances less than 25 MΩ for each recording . Internal pipette solution contained ( in mM ) : 120 K-Gluconate , 20 KCl , 10 NaCl , 10 HEPES , 0 . 6 EGTA , 4 Mg-ATP and 0 . 3 Na-GTP at pH 7 . 3 . To isolate AMPA-mEPSCs , the extracellular solution contained 1 µM TTX , 50 µM picrotoxin ( PTX , to block mIPSCs ) and 50 µM ( 2R ) -amino-5-phosphonovaleric acid ( AP5 ) , 2 mM Ca2+ and 1 . 25 mM Mg2+ . All whole cell patch clamp recordings were performed under continuous perfusion . Cells were perfused for 3-min prior to recording to achieve stable baselines . No more than 2 recordings were obtained per coverslip . AMPA-mEPSCs were quantified using Synaptosoft MiniAnalysis software . Frequency data was collected by quantifying 4 min per cell starting at the beginning of each recording . To ensure that high frequency cells did not skew the amplitude comparisons by being over represented , 200 mEPSC amplitudes were randomly selected from each recording to build the cumulative probability histograms and rank order plots . Kolmogorov–Smirnov test was performed using Past 3 . 02 ( http://folk . uio . no/ohammer/past/ ) . | Learning and memory is thought to rely on changes in the strength of the connections between nerve cells . When an electrical impulse travelling through a nerve cell reaches one of these connections ( called a synapse ) , it causes the cell to release chemical transmitter molecules . These bind to receptors on the cell on the other side of the synapse . This starts a series of events that ultimately leads to new receptors being inserted into the membrane of this second cell , which strengthens the connection between the two cells . The receptors involved in this process belong to two groups , called AMPA and NMDA receptors . Both groups are ion channels that regulate the flow of charged particles from one side of a cell's membrane to the other . In resting nerve cells , NMDA receptors are partially blocked by magnesium ions . However , the binding of the transmitter molecules to AMPA receptors causes these receptors to open and allow positively charged sodium ions into the cell . This changes the electrical charge across the cell membrane , which displaces the magnesium ions from the NMDA receptors so that they too open . Calcium ions then enter the cell through the NMDA receptors and activate a signaling cascade that leads to the production of new AMPA receptors . Nerve cells also release transmitter molecules in the absence of electrical impulses , and evidence suggests that individual cells can use this ‘spontaneous transmitter release’ to adjust the strength of their synapses . When these spontaneous release levels are high , AMPA receptors are removed from the membrane of the nerve after the synapse to make it less sensitive to the transmitter molecules . Conversely , when spontaneous release levels are low , additional AMPA receptors are added to the membrane to increase the sensitivity . Reese and Kavalali have now identified the mechanism behind this process by showing that spontaneously released transmitter molecules cause small amounts of calcium to enter the second nerve cell through NMDA receptors , even when these receptors are blocked by magnesium ions . This trickle of calcium triggers the release of more calcium from stores inside the cell , which amplifies the signal . The ultimate effect of the flow of calcium into the cell is to block the production of AMPA receptors , and ensure that the synapse does not become any stronger . As confirmation of this mechanism , Reese and Kavalali showed that simulating low levels of spontaneous activity by blocking the so-called ‘calcium-induced calcium release’ has the opposite effect . This led to more AMPA receptors being produced and stronger synapses . Taken together these findings indicate that spontaneous transmitter release exerts an outsized influence on communication between neurons by maintaining adequate levels of AMPA receptors via these ‘amplified’ calcium signals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Spontaneous neurotransmission signals through store-driven Ca2+ transients to maintain synaptic homeostasis |
Technological advances have enabled the analysis of cellular protein and RNA levels with unprecedented depth and sensitivity , allowing for an unbiased re-evaluation of gene regulation during fundamental biological processes . Here , we have chronicled the dynamics of protein and mRNA expression levels across a minimally perturbed cell cycle in human myeloid leukemia cells using centrifugal elutriation combined with mass spectrometry-based proteomics and RNA-Seq , avoiding artificial synchronization procedures . We identify myeloid-specific gene expression and variations in protein abundance , isoform expression and phosphorylation at different cell cycle stages . We dissect the relationship between protein and mRNA levels for both bulk gene expression and for over ∼6000 genes individually across the cell cycle , revealing complex , gene-specific patterns . This data set , one of the deepest surveys to date of gene expression in human cells , is presented in an online , searchable database , the Encyclopedia of Proteome Dynamics ( http://www . peptracker . com/epd/ ) .
Recent technological advances in both mass spectrometry and nucleic acid sequencing have created new high throughput methods for the quantitative measurement of protein ( Lamond et al . , 2012; Mann et al . , 2013 ) and RNA levels ( Nagalakshmi et al . , 2008; Wilhelm et al . , 2008 ) . This in turn has allowed studies , in both model organisms and human cells , which seek to document global proteomes and transcriptomes . For example , several laboratories have performed in-depth proteomic profiling of different human cell lines and independently concluded that the minimum number of proteins expressed in each cell line is on the order of 10 , 000 ( Beck et al . , 2011; Nagaraj et al . , 2011 ) . In a more recent study , the proteomes of 11 cell lines were profiled to a similar depth ( Geiger et al . , 2012 ) . These data enable comparison of protein expression levels between cells that differ in tissue type , developmental origin , and mode of in vitro culture . At the transcriptome level , microarray analysis and RNA-Seq have been used to document global mRNA expression across extensive panels of human cells and tissues ( Ramaswamy et al . , 2001; Su et al . , 2002 ) , and to examine features of RNA regulation , such as alternative splicing ( Braunschweig et al . , 2013 ) . Although global quantification of mRNA levels is often more convenient , several studies have reported that for human cells and model organisms RNA levels alone are not uniformly predictive of protein levels ( Gygi et al . , 1999; Vogel et al . , 2010; Maier et al . , 2011; Nagaraj et al . , 2011; Schwanhausser et al . , 2011 ) . The relationship between protein and mRNA abundance remains a topic of debate and the studies above highlight the challenges of relying on analysis of mRNA alone to measure gene expression at the protein level . High throughput methods have also been used to analyze changes in gene expression in response to specific events , including cell cycle progression . Our objective in this study was to capitalize on these technological advances to provide an in-depth characterization of gene expression in human cells , including cell cycle-associated changes in the proteome and transcriptome . We thus address the key question of how different layers of gene expression affect corresponding levels of protein and mRNA in a biologically important , dynamic system . The mitotic cell cycle is a conserved and highly regulated process in all eukaryotes , which has been categorized into four consecutive phases , that is , Gap 1 ( G1 ) , Synthesis ( S ) , Gap 2 ( G2 ) and Mitosis ( M ) . Regulation of the cell cycle is important for controlling cell growth and proliferation and for coordinating the timing of major cellular events , such as DNA replication and cell division ( Hunter and Pines 1994 ) . Regulatory pathways and checkpoints allow cells to respond quickly to DNA damage and other forms of stress that require cell cycle arrest to prevent uncontrolled cell division ( Hartwell and Kastan , 1994; King et al . , 1994; Elledge , 1996; Pines , 1999 ) . Many signaling mechanisms also impact on the control of cell cycle to allow cells to grow and divide in response to both developmental and environmental cues . Misregulation of the cell cycle machinery can lead to inappropriate cell proliferation , as often seen in neoplastic disease . There are significant technical challenges involved in the analysis of cell cycle-dependent regulation of gene expression . Most cell cycle analyzes make use of cell synchronization methods to enrich populations of cells at specific cell cycle stages in sufficient quantities for biochemical characterization . For example , multiple strategies have been used to characterize levels of mRNA expression at different cell cycle stages in the budding yeast , Saccharomyces cerevisiae , including conditional knockdown of cell cycle regulators , withdrawal of growth factors , use of chemical inhibitors and physical size separation by centrifugal elutriation ( Cho et al . , 1998; Spellman et al . , 1998 ) . Large-scale transcriptome analyzes have also been performed in mammalian cells , particularly in HeLa cells , to compare mRNA expression levels across the cell cycle ( Cho et al . , 2001; Whitfield et al . , 2002 ) . More recently , several groups have also examined cell cycle variation in the mammalian proteome and phosphoproteome in cell line models ( Ohta et al . , 2010; Olsen et al . , 2010; Pagliuca et al . , 2011; Lane et al . , 2013 ) . The established methods to achieve highly synchronized mammalian cells usually involve arresting cells , either by inducible genetic depletion of factors needed to drive cell progression , or by drug treatments that either activate checkpoints , block major metabolic pathways , or else disrupt the mitotic spindle . Inhibiting or depleting essential factors and activities needed for proper cell cycle progression , however , may in turn cause side effects that alter gene expression independent from direct , cell cycle-based regulation ( Cooper et al . , 2007 ) . An additional challenge is that several studies have suggested that there may be tissue-specific plasticity in cell cycle regulation ( Pagano and Jackson , 2004 ) . For example , studies in mice with genetic deletions of D-type cyclins have shown that the hematopoietic system is the only tissue that requires D-type cyclins for cell proliferation ( Kozar et al . , 2004 ) . In contrast to epithelial tumor cell lines , large-scale studies examining protein expression in hematopoietic cells are sparse , with the Jurkat-T and K562 cell lines being the only immune cell lines comprehensively profiled ( Geiger et al . , 2012 ) . To the best of our knowledge , there have been no previous large-scale studies on myeloid cell cycle gene expression . We have addressed these challenges by undertaking a high-resolution proteomic analysis of cell cycle gene expression in human NB4 cells . These cells are derived from the myeloid lineage and have been widely used as a model system for studying acute promyelocytic leukemia and myeloid biology ( Drexler et al . , 1995 ) , due to their ‘undifferentiated’ promyelocyte state ( Grisolano et al . , 1997; He et al . , 1997; Zhu et al . , 2005 ) . The data characterize the proteome of NB4 cells to a depth of over 10 , 000 proteins , with high average sequence coverage , including analysis of isoform expression and post-translational phosphorylation . We analyze cell cycle-regulated gene expression in NB4 cells at both the protein and mRNA level , using counterflow centrifugal elutriation ( Banfalvi , 2008 ) combined with high-throughput , label free mass spectrometry-based proteomics and RNA-Seq . We identify subsets of genes encoding proteins whose abundance is cell cycle regulated , including novel factors , isoforms , and phosphorylation sites . All of the resulting data have been incorporated into the Encyclopedia of Proteome Dynamics ( http://peptracker . com/epd/ ) , an online , free to access , searchable database .
Our goal was to obtain deep proteome coverage that would document global gene expression in the human myeloid cell lineage and to combine this with an unbiased , quantitative chronology of changes in gene expression across the mitotic cell cycle , with minimal perturbation to cellular physiology . To achieve this , we analyzed the human NB4 promyelocytic leukemia cell line using a strategy that combines centrifugal elutriation ( Banfalvi , 2008 ) , a physical method of enriching cell populations at different cell cycle stages , with high throughput analyzes of both protein and poly ( A ) + mRNA levels . This strategy is outlined schematically in Figure 1 . First , six elutriated fractions were collected from an unsynchronized population of NB4 cells grown in normal , label free medium . Second , to obtain deep proteomic information with high peptide and protein coverage , extracted proteins from each elutriated fraction were digested with trypsin and/or LysC and further fractionated by offline peptide chromatography prior to mass spectrometric analysis . As described below , this allowed quantitative profiling of the NB4 proteome to a depth of >10 , 000 proteins , with high mean sequence coverage ( ∼38% ) . Third , poly ( A ) + RNA was also isolated from each of the same elutriated NB4 cell fractions that were used to isolate proteins and analyzed by an RNA-Seq transcriptomics workflow ( Figure 1 ) . This allowed us to quantitate transcripts from 12 , 078 protein coding genes , including 9667 genes whose mRNA expression was measured separately in each of the three pooled , elutriated fractions and in asynchronous cells . 10 . 7554/eLife . 01630 . 003Figure 1 . Experimental workflow . NB4 cells were harvested and fractionated by cell size using centrifugal elutriation . Six fractions were collected and processed separately for transcriptomics and proteomics . For proteomics , cells were lysed and digested with either Lys-C or Lys-C/trypsin . Peptides were then separated by two orthogonal modes of chromatography prior to analysis using an Orbitrap mass spectrometer . Data normalization , peak picking , database searching , peptide and protein identification were performed using the MaxQuant software suite . For transcriptomics , cells from the six fractions were pooled into three ( G1 , S , and G2&M-enriched fractions ) . Total RNA was extracted , and subjected to poly ( A ) + tail selection . Poly ( A ) + transcripts were shattered , reverse transcribed to establish cDNA libraries , which were sequenced using Illumina paired-end sequencing technology . Reads were aligned to the human genome ( build hg19 ) using TopHat , and then used for quantitative gene expression analysis of known protein coding genes using Cufflinks . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 003 Cells from an asynchronous , unfractionated NB4 cell culture and from each of the six elutriated fractions were analyzed by both flow cytometry and protein blotting to characterize their cell cycle profiles ( Figure 2 ) . First , cells in the total unfractionated NB4 population and in each of the six elutriated fractions were stained with propidium iodide ( PI ) and analyzed by flow cytometry . Histograms showing the results of PI fluorescence measurements confirm that there is a pronounced differential enrichment for populations of cells in G1–G2 and M phases across elutriated fractions 1–6 ( Figure 2A ) . The proportion of cells in each cell cycle phase was calculated by fitting the data to the Watson model ( Watson et al . , 1987 ) . Fractions F1 and F2 were enriched in G1 cells , F3 and four were enriched in S phase cells and F5 and F6 were enriched in G2 and M phase cells ( G2&M ) . 10 . 7554/eLife . 01630 . 004Figure 2 . Validation of cell cycle enrichment by centrifugal elutriation . ( A ) Cells from asynchronous cells ( top left ) and each elutriation fraction ( top right ) were stained with a DNA-binding fluorescent dye and analyzed with flow cytometry . Proportions of cells in each cell cycle phase ( bottom ) were estimated using the Watson model . Fractions 1 and 2 ( F1 and F2 ) are enriched in G1 , fractions 3 and 4 ( F3 and F4 ) are enriched in S , and fractions 5 and 6 ( F5 and F6 ) are enriched in G2 and M phase ( G2&M ) . ( B ) Immunoblot analyses of the protein lysates for known cell cycle phase-specific markers ( cyclin E , phospho-Histone H3 S10 , aurora kinase B , cyclin A , and cyclin B1 ) are consistent with previous literature and the enrichment profiles in ( A ) . ( C ) Forward scatter and side scatter plots ( first column ) and cell cycle distributions ( remaining columns ) for three representative fractions post inoculation ( F1 , F4 , and F6 ) . The forward and side scatter plots for each elutriated fraction are shown in cyan , and are directly compared with the same plot for asynchronous cells , which is shown in red . Cell cycle distributions of the three fractions are measured directly after elutriation ( 0 hr ) , and 2 and 4 hr after inoculation into tissue culture medium . Note that the cell cycle distributions shown includes cells with <2N DNA content . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 004 Second , we used immunoblotting to compare the levels of selected marker proteins , whose expression across the cell cycle is known , with the enrichment for each cell cycle stage detected by flow cytometry ( Figure 2B ) . Equal amounts of total NB4 proteins from either the asynchronous cell population or from each elutriated fraction were separated by SDS-PAGE , electroblotted onto nitrocellulose and marker proteins detected using specific antibodies , as described in the ‘Materials and methods’ . This shows that , as expected , GAPDH is expressed at similar levels across all fractions , while cyclin E , which is a marker for the G1/S transition , has maximal expression in fractions F1 and F2 . In contrast , markers for G2 and M phase cells , such as cyclin B1 , aurora kinase B and phospho-Histone H3 ( S10 ) , all have maximal expression in the last fractions ( F5 and F6 ) . All of the protein blotting data are thus consistent with the profiles of cell cycle stage enrichment deduced from the flow cytometry analysis and confirm that the six elutriated fractions are differentially enriched for NB4 cells at different stages of progression through interphase and into mitosis . We separately tested for viability of the collected NB4 cells post elutriation . Re-inoculation of the fractionated NB4 cells into tissue culture medium showed that the bulk of the cells survived the elutriation procedure with high viability and minimal damage and rapidly resumed growth when returned to culture , as judged by subsequent FACS analysis of the replated cells ( Figure 2C , right ) . Elutriated NB4 cells were of similar size and granularity ( as measured by forward and side scatter , Figure 2C , left ) , as a control NB4 cells that were not exposed to elutriation . In summary , based on the combination of flow cytometry , immunoblot , and cell culture analyzes , we conclude that the elutriation strategy provides an effective physical method for fractionating unsynchronized populations of human immune NB4 cells into viable subpopulations of cells that are enriched in distinct cell cycle phases , with minimal perturbation to normal cell physiology and viability . Further , the elutriation methodology is highly reproducible , as documented below . To evaluate the biological and technical reproducibility of the elutriation strategy , we performed three rounds of elutriation on NB4 cultures harvested on different days . Proteins were isolated from each elutriated fraction from each biological replicate . Samples were processed for mass spectrometry using a Single Shot workflow and analyzed in technical triplicate by reversed phase LC-MS/MS . Label free intensities were calculated from MS peptide-extracted ion chromatograms , as previously described ( Luber et al . , 2010 ) . Comparison of the label free intensities within both the technical and biological replicates shows that the average Pearson correlation coefficients are greater than 0 . 97 , indicating that both the elutriation-based cell cycle enrichment and the label free MS-based peptide quantitation methods are highly reproducible ( Figure 3—figure supplement 1 ) . To achieve deep coverage of the total NB4 cell proteome , we used a proteomic workflow combining digestion with two proteases with extensive pre-fractionation of peptides prior to MS analysis . Thus , protein samples isolated from each elutriate from a single elutriation experiment were divided and digested with either Lys-C alone ( Lys-C ) , or double digested with Lys-C and Trypsin ( Trypsin-DD ) . The resulting peptides were separated by analytical , hydrophilic Strong Anion eXchange ( hSAX ) chromatography into 12 fractions ( Ritorto et al . , 2013 ) . Each fraction was then analyzed by LC-MS/MS . Using the proteomics workflow described above , over 150 , 000 peptides were identified ( Dataset S1 , entire file available at Dryad , Ly et al . , 2014 ) , corresponding to 10 , 929 unique protein groups , over 10 , 000 of which were detected with two or more peptides . These proteins represent expression of more than 9000 genes . As shown in Figure 3A , the quantitative data set includes proteins whose MS-measured extracted ion chromatogram intensities span eight orders of magnitude , which corresponds to at least four orders of magnitude in protein copy number ( Nagaraj et al . , 2011 ) . A wide variety of biological functions are captured , reflected by the wide range of GO annotations , from low abundance proteins involved in transcription , to very high abundance proteins , such as histones and factors involved in ribosome subunit assembly ( Figure 3A ) . 10 . 7554/eLife . 01630 . 005Figure 3 . Quantitative , in-depth characterization of a myeloid leukemia proteome . ( A ) A histogram of log-transformed protein abundance ( iBAQ-scaled protein intensities ) . Quartile regions are shown in different colors , and enriched gene ontology terms ( p<0 . 01 ) are shown above each region . ( B ) A cumulative plot of protein abundance , as estimated using iBAQ-scaled intensities . In total , 10 , 193 proteins were identified with at least two supporting peptides per protein . Protein abundances follow an exponential increase , with 90 proteins ( 0 . 9% ) constituting 50% of the bulk protein mass , and 1028 proteins ( 10% ) constituting 90% of the bulk protein mass . The remaining protein identifications ( 9075 or 89 . 1% ) comprise less than 10% of the bulk protein mass in NB4 cells . ( C and D ) Venn diagrams showing the total number of sequence-unique peptides ( 154 , 985 ) and amino acid coverage ( 1 , 976 , 427 ) split by digestion method . Lys-C increases the number of peptides identified by 44% relative to Trypsin-DD . Amino acid coverage was calculated by mapping sequences back to an assembled proteome . Over 30% of the amino acids detected using Lys-C digestion reside in sections of protein sequences that are complementary to Trypsin . In summary , complementary digestion methods substantially increase the overall sequence coverage , as shown in ( E ) . Combining data from both methods boosts the mean sequence coverage to 37 . 8% with comprehensive proteome depth of over 10 , 000 proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 00510 . 7554/eLife . 01630 . 006Figure 3—figure supplement 1 . Estimation of technical and biological variances among replicates indicates highly reproducible protein quantitation . A correlation matrix showing pairwise comparisons between biological and technical replicates of the SingleShot proteomics workflow is shown . Sample identifiers are shown along the diagonal . Log-transformed label-free quantitation ( LFQ ) intensities are shown along the bottom left corner , and the associated Pearson correlation coefficients are shown along the top right corner . All pairwise comparisons reveal high correlation ( >0 . 95 ) between replicates , indicating high biological and technical reproducibility . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 00610 . 7554/eLife . 01630 . 007Figure 3—figure supplement 2 . Comparison of expected versus observed amino acid and gene ontology frequencies reveals no major detection bias in the proteomics data set . ( A ) The amino acid frequency of identified proteins using the hSAX workflow was compared against the search database ( the UniProt Complete Human Reference Proteome ) . Cellular compartment ( B ) and biological process ( C ) gene ontology term frequencies were calculated for the identified data set and the search database . High correlation between the expected frequencies from the search database and the observed frequencies in the identified proteome suggests that the data set is not obviously biased against or for a particular cellular compartment or biological process . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 007 Given the wide range of protein expression levels , the bulk of protein abundance in NB4 cells results from the expression of a relatively small number of proteins , as is shown in a cumulative protein abundance plot ( Figure 3B ) . Thus , half of the total protein molecules expressed correspond to only ∼90 highly abundant proteins ( 0 . 9% of the proteins detected ) , which is similar to previous reports on other cell types ( Nagaraj et al . , 2011 ) . In the other extreme , 10% of the total protein abundance reflects the expression of over 9000 different proteins , which highlights the prerequisite of detecting proteins across a wide range of abundance values for comprehensive proteome characterization . We have generated here deep coverage of the NB4 proteome with a group of over 10 , 000 proteins identified that are each supported by an average of 15 separate peptides ( Supplementary file 1 ) . In total , 154 , 985 sequence-unique peptides were detected , of which 30 . 8% were identified by only Lys-C , and 50 . 5% by only Trypsin-DD , as shown in Figure 3C . These peptides map to a total of 1 , 976 , 427 unique amino acid locations in the UniProt Human Reference Proteome ( Figure 3D ) , yielding on average 139 amino acids sequenced per protein . As shown , Lys-C and Trypsin-DD peptides map to largely complementary sequence regions ( Figure 3D ) . Thus , the combined use of these two proteases significantly increases the overall sequence coverage . As shown in Figure 3E , use of complementary proteases improves the mean sequence coverage by over 8% compared with a single digestion method . This yields a combined mean sequence coverage of ∼38% for over 10 , 000 proteins , providing one of the most detailed protein expression maps so far reported . Furthermore , using the double protease strategy , many proteins with low sequence coverage based on single protease digestion ( observed as a second , broad peak to the left in Figure 3E ) , are shifted to higher sequence coverage . High sequence coverage not only improves the accuracy of protein and isoform identification , it is also important for high resolution analysis of protein regulation during the cell cycle , as highlighted below . We next evaluated the NB4 data set for possible detection bias , for example due either to differential efficiency of extracting distinct classes of protein in the lysates prepared , or to failure by MS to detect protein types featuring high levels of modification and/or unusual sequence combinations that are not cleaved efficiently into peptides . To address this , we compared the observed frequencies of GO terms and the amino acid proportions for all of the proteins detected in this NB4 cell data set with the corresponding predicted frequencies calculated from in silico translation of the entire human proteome ( Figure 3—figure supplement 2 ) . The Pearson correlation coefficients observed between these predicted and measured frequencies ( r >0 . 98 ) indicate that the sampling of proteins in the NB4 data set is highly representative of the human proteome . While inevitably some expressed proteins have not been detected , particularly in the very low abundance range , we can effectively exclude that there is a major bias , either from under-sampling specific protein classes ( e . g . , membrane proteins ) , or from an absence of lower abundance proteins in general . Next , we compared this proteome analysis of NB4 cells , a human promyelocytic leukemia cell line that grows in suspension culture , with other recent examples of in depth proteomic analysis of different human cell lines , most of which are adherent tumor cell lines , of either fibroblast or epithelial origin . This meta-analysis included protein data from 14 cell line proteomes: 3 × HeLa , 2 × U2OS , A549 , GAMG , HEK293 , K562 , LnCap , MCF7 , RKO , HepG2 , and Jurkat-T ( Lundberg et al . , 2010; Beck et al . , 2011; Nagaraj et al . , 2011; Geiger et al . , 2012 ) , which were consolidated and mapped to Ensembl Genes prior to comparison . The combined data set provides evidence of protein-level expression of over 11 , 000 human genes . Of these , a common set of ∼3000 genes are identified by protein data from all these cell lines , defining a core , shared proteome ( Supplementary file 2 ) . Interestingly , the abundance values of proteins in this core proteome span the full abundance range of the entire NB4 proteome . This suggests that the core proteome is not simply reflecting a detection bias towards abundant proteins . The core proteome is enriched in proteins associated with RNA processing , translation , cell cycle , and DNA metabolic processes , which together highlight key biological processes required for cell proliferation . In contrast , analysis of cell type-specific proteomes highlight specialized biological functions that are associated with cell lineage and mode of culture , as will be discussed below . Approximately , 10% of the expressed genes we detected in NB4 cells at the protein level are exclusive to this study and have not been reported in large-scale proteomic studies of other human cell lines ( listed in Supplementary file 2 ) . Interestingly , this NB4-specific pool is enriched in proteins that regulate cation flux in the cell , proteins involved in the innate immune response , zinc finger proteins and transcription factors ( >200 ) , including proteins known to be important to leukemic and immune cell biology , such as RARα , RXRβ , CEBPα , GFI-1 and PU . 1 ( Zhu et al . , 2001; Orkin and Zon , 2008 ) . We next focused on comparing the NB4 proteome with the most recent study describing in detail protein expression in several human cell lines ( Geiger et al . , 2012 ) , including the K562 and Jurkat-T cancer cell lines derived from the immune lineage ( myeloid and lymphoid , respectively ) , that are the most related to NB4 ( myeloid ) . The other two cell lines compared ( HeLa and MCF7 ) are derived from epithelial tumors . Pairwise comparisons were performed to determine sets of genes that are uniquely detected in each cell line . Enriched gene ontology terms for each set are shown in Figure 4A . Comparison of these cell line-specific subproteomes reveals proteins with functions that highlight not only the differences in lineage , but also distinguish mode of culture , for example suspension vs adherent culture . For example , HeLa- and MCF7-specific sets are enriched in genes involved in cell adhesion , such as cadherins and integrins , whereas the Jurkat-T-specific set is enriched in genes involved in T-cell selection and activation , such as CD1 , CD3 , and CD4 ( Figure 4A ) . 10 . 7554/eLife . 01630 . 008Figure 4 . Identification of myeloid-specific factors in the NB4 proteome . ( A ) Pairwise comparisons between the NB4 proteome ( this study , acute myeloid leukemia ) and K562 ( chronic myeloid leukemia ) , Jurkat-T ( T-cell leukemia ) , HeLa ( cervical carcinoma ) , and MCF7 ( breast carcinoma ) proteomes published by Geiger et al . ( 2012 ) . Enriched gene ontology terms and the enrichment p-values are shown for each pairwise comparison . The observed cell-line specific gene ontology enrichments are consistent with the developmental origins of the cell lines ( immune vs epithelial ) , and culturing conditions ( suspension vs adherent ) . The NB4 proteome is highly enriched in transcription factors when compared to cell lines that are not in the myeloid lineage ( Jurkat-T , HeLa , and MCF7 ) , implying that there is set of shared transcription factors between NB4 and K562 that may be myeloid-specific . ( B ) A transcriptional regulatory network analysis of proteins identified in myeloid cells ( NB4 and K562 ) . Arrows connect transcription factors with their predicted gene substrates ( MSigDB ) . JUN and SP1 appear to be regulatory hubs that can regulate the expression of numerous NB4- and K562-specific genes ( Friedman , 2002 ) . Together , these data highlight a protein group that may have important transcriptional regulatory activity in myeloid cells . Circles indicate genes that are annotated as being involved in myeloid differentiation ( red ) or transcription ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 008 For three out of the four cell lines , pairwise comparisons reveal specific transcription factors that are enriched in the NB4-specific data set , as similarly found for the broader comparison described above . In contrast , comparison of the NB4 proteome with the proteome of the myeloid K562 cells reveals that many transcriptional regulators are shared between these two myeloid cell lines . Thus , among the 87 genes that express proteins in K562 and NB4 , but which are not detected in MCF7 and HeLa ( Geiger et al . , 2012 ) , 22 are either known or putative transcription factors including SP1 and JUN ( Friedman , 2002 ) , and five have been annotated with gene ontology terms associated with myeloid differentiation ( Figure 4B ) . Further network analysis using the MSigDB transcription factor binding database ( Subramanian et al . , 2005; Matys et al . , 2006 ) , revealed evidence for cross-talk between JUN , SP1 , and many of the additional genes whose expression was shared between K562 and NB4 cells , but not observed in either MCF7 or HeLa ( Figure 4B ) . Interestingly , immunohistochemical detection of SP1 protein levels by the Human Protein Atlas project ( www . proteinatlas . org ) showed ubiquitous expression and especially high levels in hematopoetic and placental tissues ( Uhlen et al . , 2010 ) . In summary , we have identified a specific set of proteins that are preferentially detected in myeloid cells and that may be important in specifying myeloid cell function . We conclude that deep proteome analysis helps to provide a molecular characterization of cell identity by defining sets of genes uniquely expressed in specific cell types . Next , we studied how gene expression in NB4 cells varies across the cell cycle . To do this , we compared the subproteomes of the six , separate elutriated NB4 cell fractions and analyzed how protein abundance varies between the different cell cycle phases . To increase the accuracy of this analysis , we focused on a subset of the total NB4 proteomic data set for which we have protein abundance measurements in all six elutriated fractions and in the asynchronous samples . The 6505 proteins that meet this requirement are supported on average by over 22 distinct peptides per protein , and have a mean sequence coverage of >45% . An arbitrary fold-change cutoff of 2 . 0 ( 1 . 0 in the log2-transformed axis ) , was chosen here as the threshold value for cell cycle-regulated abundance change because we observed that this was sufficient to highly enrich for proteins annotated with cell cycle associated GO terms ( p<<0 . 01 ) , as shown in Figure 5A . Using these parameters , we identified a group of 358 proteins whose abundance varies across the cell cycle by twofold or more , corresponding to ∼5 . 5% of the high quality , filtered proteomic data set of 6505 proteins ( Supplementary file 3 ) . 10 . 7554/eLife . 01630 . 009Figure 5 . Identification of cell cycle-regulated proteins . ( A ) The fold change in label free intensities between any two fractions are shown as a histogram . To identify cell cycle-regulated proteins , an arbitrary fold change cutoff of 2 . 0 ( 1 . 0 in the log2-transformed axis ) was set , as indicated by the border between the orange and blue boxes . Highly significant enriched gene ontology terms ( p<<0 . 01 ) are shown above each group . A twofold change is sufficient to enrich for cell cycle related gene ontology terms , such as M-phase , nuclear division , and mitosis . ( B ) Clustering of the 358 cell cycle-regulated proteins identified in ( A ) . Scaled protein intensity profiles were clustered by the phase of maximum expression , with the exception of a small minority of proteins that peaked in multiple phases . Graph titles indicate the phase that is enriched in that fraction . ( C ) The number of cell cycle-regulated proteins split by cluster . Half of the cell cycle-regulated proteins are maximally expressed in the G2&M phase of the cell cycle . ( D ) Scaled intensities are depicted as a heat map . Each vertical line represents a cell cycle-regulated protein , and the shading indicates the intensity ( bright yellow being the most intense ) . Cell cycle regulators established in the literature are highlighted along the bottom of the heat map , and include cyclins A2 , B1 , B2 , and CDT1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 009 This group of 358 proteins whose abundance is cell cycle regulated was clustered by intensity profiles , which resulted in seven distinct clusters . Scaled protein expression profiles by cluster are shown both as line graphs ( Figure 5B ) and as a heatmap ( Figure 5D ) . Six clusters vary primarily by variations in their maximum expression in different elutriated fractions ( clusters 1 through 6 ) . In cluster 7 , proteins show a marked decrease in abundance during S-phase and peak in the G2&M and G1 fractions ( termed ‘G2&M + G1’ cluster ) . Approximately , half of the proteins whose abundance varies significantly across the cell cycle exhibit peak expression in elutriated fractions 5&6 , corresponding to late S , G2 , and M phases ( Figure 5C ) . Proteins whose abundance peaks in S phase are the second most frequent class ( 27% ) , followed by proteins peaking in G1 ( 17% ) and proteins that peak in both G2&M and G1 ( 7% ) . A large number of human proteins previously reported to be regulated during the cell cycle were identified in this unbiased data set . For example , proteins involved in origin licensing in G1 , such as ORC1 ( Origin Recognition Complex 1 ) and DNA replication factor CDT1 , peak in G1 . UNG ( Uracil-DNA glycosylase ) , which is involved in DNA repair , and Cyclin A2 peak in S phase . Aurora kinase B and cyclins B1 and B2 , which are proteins involved in mitosis , peak in G2&M ( Murray , 2004; Musacchio and Salmon , 2007; Dephoure et al . , 2008; Olsen et al . , 2010 ) . The clusters above differ primarily in peak expression across the six elutriated fractions , which can also be broadly classed as G1- , S- , or G2&M-enriched . The six clusters were grouped into these three broader classifications ( excluding cluster 7 , which peaks in both fractions 1 and 6 ) and analyzed for enrichment of gene ontology terms ( Table 1 ) . Additionally , we tested whether the cyclical regulation of protein abundance may be explained , at least in part , by changes in transcription factor activity across the cell cycle . The UCSC TF database contains the predicted promoter binding sites for many important and well-studied transcription factors and is incorporated into the DAVID gene ontology enrichment tool . Thus , we also examined whether the promoters of genes encoding these proteins are differentially enriched in transcription factor binding sites . 10 . 7554/eLife . 01630 . 010Table 1 . Enriched functional annotations among the cell cycle varying proteinsDOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 010Peak phaseFunctional annotationProteins% of totalp valueG1 ( 42 proteins ) transcription cofactor activity512%0 . 003transcription factor binding512%0 . 012S ( 110 proteins ) phosphoprotein8275%1 . 2E-07E2F*8275%0 . 002DNA metabolic process109%0 . 012positive regulation of gene expression87%0 . 037cell cycle1110%0 . 009G2/M ( 180 proteins ) M phase2614%7 . 5E-20cell cycle4123%3 . 9E-19phosphoprotein9452%1 . 8E-06NFY*9352%3 . 6E-05Complex ( 26 proteins ) STAT3*2077%0 . 003nucleotide-binding727%0 . 015Proteins were partitioned into four categories by peak phase and analyzed for functional annotation enrichment . Functional annotations include gene ontology terms and predicted transcription factor binding sites in the promoter region of the encoding gene . Enriched annotations , their enrichment p values , the number and percentage of proteins with the specified annotation are shown . *Transcription factor binding sites from the UCSC TFBS database . Table 1 lists the corresponding enriched gene ontology terms and transcription factor binding sites by category . Each category is enriched in distinct annotations , indicating that different types and different functional classes of proteins are being regulated during separate phases of the cell cycle . In general , the GO terms we observe enriched in each phase are consistent with many of the activities and processes known and expected for that stage of cell cycle progression . For example , in the category with proteins whose abundance peaks in S phase , there is a clear enrichment for GO terms associated with DNA metabolic functions , reflecting DNA replication as the major metabolic event in S phase . Additionally , the promoters of genes encoding proteins that peak in S-phase are enriched with respect to predicted binding sites for E2F ( Table 1 ) , a transcription factor that is known to play major roles in regulating entry into the cell cycle and the G1–S transition ( Mudryj et al . , 1991 ) . Interestingly , we identify members of the E2F family ( E2F6 and E2F8 ) as proteins whose abundance peaks in S and G2&M phases , which is consistent with recent reports documenting their role in transcriptional inactivation of G1/S genes ( Bertoli et al . , 2013 ) . In contrast , the category containing proteins whose abundance peaks at G2&M phase is instead highly enriched for GO terms associated with cell division , M-phase , and NF-Y transcription factor binding sites in their promoters . The category containing proteins that peak in G1 and G2&M ( cluster 7 ) , meanwhile , is enriched for genes with promoters that have STAT3 transcription factor binding sites; indeed , 20 out of the 26 genes encoding these proteins ( 77% ) , have predicted STAT3 binding sites . Notably , the GO term ‘cell cycle’ is only enriched in the S and G2&M clusters . Based on our current data ( Figure 5C ) , this may reflect the fact that more cell cycle regulated proteins have peak abundance in S , G2 , and M phase than in G1 phase . However , it may also illustrate a feature of the gene ontology annotation system used , linked with the preponderance of previous ‘cell cycle’ research that has concentrated specifically on analyzing either the entry of cells into mitosis or on studying events during chromosome segregation and mitotic progression and exit . However , our data demonstrate multiple proteins whose abundance is also ‘cell cycle regulated’ at other stages during interphase , outside of mitosis , which suggests that they should also be annotated with the GO term ‘cell cycle’ . In most cases , the protein groups identified by MS analysis correspond to groups of protein isoforms that usually originate from the same open reading frame . However , protein isoforms , even when encoded by the same gene , can have distinct biological properties and can differ in their subcellular localization patterns , interaction partners , and biological functions ( Trinkle-Mulcahy et al . , 2006; Ahmad et al . , 2012; Kirkwood et al . , 2013 ) . The high sequence coverage achieved in this NB4 data set improved our ability to discriminate between separate protein isoforms . The peptide data map to 33 , 575 separates protein isoforms in the Uniprot Human Reference Proteome . Quantitative peptide data , such as either intensities or spectral counts , are normally aggregated into protein groups by sequence similarity and shared peptide evidence . In this study , we have pooled quantitative data by protein isoform , which facilitates the analysis of isoform-specific cell cycle behavior ( Figure 6 ) . Given that current proteomics workflows achieve ∼30 to 40% mean sequence coverage at best , a comprehensive analysis of isoform-specific variation remains challenging . However , the three examples selected below illustrate the importance of examining isoform-specific protein variation across the cell cycle . 10 . 7554/eLife . 01630 . 011Figure 6 . CASC4 , PPFIBP1 , and SDCCAG8 are examples of ORFs that encode multiple splice isoforms that behave differently across the cell cycle . Protein spectral count profiles across the six elutriated fractions for three open reading frames ( ORFs ) showing protein-level isoform-specific cell cycle variation: CASC4 ( A ) , PPFIBP1 ( B ) , and SDCCAG8 ( C ) . Isoform sequences are shown schematically above each graph . Sequence regions for which direct peptide evidence was detected are shaded in blue , and sequence motifs known to be important in post-translational regulation are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 011 Peptide-level MS data were re-analyzed in the context of predicted and known splice isoforms in the UniProt database ( Kirkwood et al . , 2013 ) . To enhance data quality , we only considered ORFs that have at least two peptides with unique amino acid sequences per protein isoform that were quantified in all six elutriated fractions . These were screened for isoform-specific abundance variation across the cell cycle . Examples of cell cycle-regulated isoforms include proteins encoded by the genes CASC4 , PPFIBP1 , and SDCCAG8 . Line graphs show the spectral counts across the six elutriated fractions ( Figure 6 ) . Peptide evidences and isoform-specific sequence motifs that target proteins for cleavage and/or degradation , which can affect the activity and abundance of one isoform differentially from the remaining isoforms , are depicted schematically in Figure 6 . We observed differential behavior for CASC4 isoforms , with the abundance of CASC4-1 and CASC4-2 peptides substantially decreasing from G1 to G2&M , whereas CASC4-3 remains invariant ( Figure 6A ) . An important point is that simply aggregating the peptide intensities for all of these CASC4 isoforms produces a relatively constant expression pattern across the elutriated fractions . Therefore , conventional bottom-up MS analysis ignoring isoform variation would indicate that the CASC4 ORF generates a polypeptide that is not cell cycle regulated . In contrast , examining the data set with isoform resolution showed isoform-specific cell cycle regulation of separate polypeptides encoded by the CASC4 gene . PPFIBP1 was initially identified as a cell cycle-regulated protein using the isoform-naive methods described above . However , closer analysis shows that the PPFIBP1 gene encodes multiple isoforms that behave differentially across the cell cycle fractions . Thus , isoforms 1 through 4 vary in abundance across the cell cycle , whereas the abundance of isoform 5 remains relatively constant ( Figure 6B ) . Interestingly , several D-box motifs associated with targeted protein degradation are only found in isoforms 1–4 , suggesting a mechanism for the isoform-specific cell cycle regulation observed . The SDCCAG8 gene encodes multiple protein isoforms: three that decrease from G1 to G2&M ( isoforms 1 , 3 , and 4 ) and isoform 2 that instead peaks in S-phase ( Figure 6C ) . As shown with CASC4 above , aggregated peptide data for all isoforms of SDCCAG8 would indicate constant expression across the elutriated fractions , if interpreted as having all of the peptides belonging to a single protein species . However , as a result of the high sequence coverage obtained , it is apparent that the peptides from the SDCCAG8 gene belong to separate isoforms that are differentially regulated across the cell cycle . These three examples underline the value of high peptide sequence coverage that allows isoform-level resolution . The data also indicate that current analyzes likely underestimate the total number of cell cycle-regulated polypeptides , because with current methods many peptides are not detected and thus many isoforms still cannot be reliably discriminated . Cell cycle progression is controlled not only by changes in protein abundance , but also by other protein properties , including post-translational modifications ( PTMs ) . One of the most important and well-characterized classes of PTM is reversible phosphorylation , which modulates the activity of numerous cell cycle regulatory proteins ( Dephoure et al . , 2008; Olsen et al . , 2010 ) . In this data set a total of 2761 phosphopeptides were identified , of which 28% were detected with multiple phosphorylated residues in the same peptide ( Supplementary file 4 ) . Most of the phosphorylation sites identified were on Ser ( 64% ) , with the remaining sites evenly split between Thr ( 17% ) and Tyr ( 16% ) ( Figure 7A ) . Many of the pTyr sites were found in abundant cytoskeletal proteins , such as actin , myosin , and titin ( Supplementary file 4 ) . Among the phosphorylation sites that were independently identified by MS/MS in all six elutriated fractions , 89 phosphorylated peptides varied in abundance by more than twofold ( ∼3% of phosphorylation sites detected , Supplementary file 5 ) . These sites , which mapped to 79 different proteins , were considered to be cell cycle regulated . Interestingly , only four of these cell cycle-regulated phosphosites mapped to proteins whose abundance varied by more than twofold across the cell cycle ( Figure 7B ) . As shown in Figure 7C , most of the cell cycle-regulated phosphosites were on Ser ( 87% ) . The only cell cycle-regulated phosphotyrosine identified was mapped to Tyr15 of CDK2 and/or CDK1 ( both these proteins upon digestion with Lys-C or trypsin , produce the same Tyr15-containing peptide ) . It has been previously shown that the activities of CDK2 and CDK1 are modulated by differential phosphorylation of Tyr15 at different stages of the cell cycle ( Gu et al . , 1992 ) . 10 . 7554/eLife . 01630 . 012Figure 7 . Identification of cell cycle-regulated phosphopeptides . ( A ) A total of 2761 phosphorylation sites were identified without phosphopeptide enrichment , which are shown split by residue ( Ser , Thr , and Tyr ) . Cell cycle regulated phosphosites are shown in green . The numbers on top of each bar indicate the total number of pSer , pThr , and pTyr residues detected , respectively . The proportions of cell cycle-regulated pSer , pThr , and pTyr , relative to the total pSer , pThr , and pTyr sites detected respectively , are shown as percentages . ( B ) Overlap between proteins whose abundances are cell cycle regulated and proteins whose phosphorylation is cell cycle regulated . ( C ) A breakdown of cell cycle-regulated phosphosites by residue . The number and the percentage of phosphosites relative to the total number of cell cycle regulated phosphosites are shown . ( D ) Scaled phosphopeptide intensity profiles plotted as a heatmap . Representative cell cycle-regulated phosphorylations that are established in the literature are shown along the top of the heatmap . ( E ) Scaled phosphopeptide and summed protein intensity profiles for four cell cycle regulated phosphorylated proteins ( TOP2A , UNG , TP53 , and histones ) . The peptide intensity graphs are annotated with the mapped phosphorylation site . For histones , several phosphopeptide profiles ( light purple , light blue , and light orange ) and the average ( black ) are shown on the same graph . The total histone intensity is calculated as the sum of all histones identified . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 012 We further identified cell cycle-regulated phosphorylated peptides that peak in abundance at different stages of the cell cycle ( Figure 7D , E ) . Four examples of proteins whose phosphorylation status varies across the cell cycle are shown in Figure 7D . Several cell cycle-regulated phosphorylation sites were identified in histones ( Figure 7D , right ) , which are known to be increasingly phosphorylated from G1 to G2&M ( Sarg et al . , 2006 ) . In the case of TOP2A , which is a protein involved in modulating the topological state of DNA , levels of the peptide phosphorylated at Ser1377 peak in G1/S , whereas an increase in total TOP2A protein abundance is observed from G1 to G2&M . These data suggest two modes of TOP2A activity across the cell cycle: one that is modulated by phosphorylation at Ser1377 during G1/S and a second that may require more copies of TOP2A in the later stages of the cell cycle . In the case of UNG , levels of both the phosphopeptide and the total protein abundance vary in a similar manner across the cell cycle , suggesting that the phosphorylation stoichiometry at this site ( Ser23 ) is relatively constant . In contrast , while the total abundance of TP53 protein is relatively constant , the phosphorylation level of Ser315 on TP53 varies significantly across the cell cycle . To investigate the role of transcriptional regulation in the abundance changes observed at the protein level , we undertook a parallel large-scale RNA-Seq analysis of the NB4 cell transcriptome . This was performed both using asynchronous NB4 cells and cells enriched at different cell cycle stages by elutriation . The RNA-Seq analysis was performed both in biological and technical duplicate . The combined transcriptomics data set detected mRNA expression from 12 , 078 genes in NB4 cells ( Supplementary file 6 ) . This data set was filtered to identify a subset of 9667 genes whose mRNA expression was quantified in each of the three elutriated samples ( i . e . , combined elutriated fractions F1+F2 , F3+F4 and F5+F6 ) . Pairwise comparisons between both the technical and biological replicates show high correlation ( r >0 . 90 ) , ( Figure 8—figure supplement 1 ) , albeit with higher variance than for replicate proteomics data sets ( Figure 3—figure supplement 1 ) . Nonetheless , given the overall high degree of reproducibility observed in the biological and technical replicate proteomics and transcriptomics data , we conclude that variance observed comparing protein and mRNA levels ( as will be discussed below ) , predominantly represents biologically significant differences in protein and mRNA expression and cannot be simply explained by variability in either sample preparation and/or measurement error . Next , we compared directly the expression levels of cognate protein and mRNAs from the same genes . To do this , our detailed proteome ( 8510 proteins ) and transcriptome ( mRNA from 9667 genes ) data sets from the asynchronous NB4 cell cultures were merged by mapping the respective protein and gene identifiers to Ensembl Gene IDs . Quantitative protein abundances ( iBAQ-scaled , see ‘Materials and methods’ ) from expression variants and isoforms that originate from the same gene were aggregated by summation . Histone transcripts were removed from the analysis , due to the known under-representation of poly ( A ) - mRNAs in our data sets . In total , we could directly compare the abundances of cognate protein and mRNA for 6170 genes ( Figure 8A ) . Qualitatively , this shows that overall , as expected , the levels of protein and mRNA from the same gene are clearly positively correlated . However , the moderate value of the Spearman rank correlation coefficient ( 0 . 63 ) , indicates that this positive correlation is not strong enough to consider the level of mRNA alone as a reliable predictor of protein levels for many specific genes . The Spearman rank correlation coefficient of 0 . 63 measured here is within the upper quartile of the range of previously reported values for correlations between protein and mRNA levels in other mammalian cell types ( Tian et al . , 2004; Maier et al . , 2009; Lundberg et al . , 2010; Nagaraj et al . , 2011; Schwanhausser et al . , 2011; Vogel and Marcotte , 2012 ) . 10 . 7554/eLife . 01630 . 013Figure 8 . Correlation of protein and RNA levels across the cell cycle . Log-transformed , iBAQ-scaled protein intensities and log-transformed FPKM values ( RNA ) from asynchronous cells ( A ) and cell cycle fractions ( B ) . RNASeq data are expressed as Fragments Per Kilobase of exon per million fragments Mapped ( FPKM ) , which is a proxy for RNA copy number . Histone genes were removed from the analysis due to the absence of poly ( A ) + tails in histone-encoding messages . Each graph is annotated with the calculated Spearmen correlation coefficients ( r ) . ( C ) Correlation of the protein and mRNA abundances in asynchronous cells of the 358 proteins whose abundances are cell cycle regulated ( r = 0 . 45 ) . ( D ) The same data shown in ( C ) , but split by protein clusters as described in Figure 5 . ( E ) Correlation of the expression profiles of the 358 cell cycle regulated proteins and their associated transcripts . Genes were classified into two groups based on protein and RNA expression correlation ( Pearson’s correlation coefficient greater than or equal to 0 . 5 ) . ( F ) Cyclin A2 and Cdt1 are examples highlighted from the groups in ( E ) . Protein and RNA abundance standard errors were calculated from the variance in scaled peptide intensities and biological replicates , respectively . ( G ) Immunoblot analysis of Cdt1 and GAPDH protein expression across asynchronous and elutriated NB4 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 01310 . 7554/eLife . 01630 . 014Figure 8—figure supplement 1 . Analysis of technical and biological variance among duplicates reveals highly reproducible RNA quantitation . A correlation matrix showing pairwise comparisons between biological and technical replicates of the RNASeq transcriptomics workflow . Sample identifiers are shown along the diagonal . Log-transformed FPKM values are shown along the bottom left corner , and the associated Pearson correlation coefficients are shown along the top right corner . All pairwise comparisons reveal high correlation ( >0 . 90 ) between replicates , indicating high biological and technical reproducibility . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 01410 . 7554/eLife . 01630 . 015Figure 8—figure supplement 2 . Correlation of protein and RNA abundances of cell cycle-regulated proteins . Comparison of protein and RNA abundances in G1 , S , and G2&M phases of the cell cycle for proteins that peak in G1 , S , and G2&M , respectively . Spearman correlation coefficients , which are shown in the inset , follow the same trend as observed in Figure 8D , with G2&M-peaking proteins having the poorest protein and mRNA abundance correlation . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 015 We also compared the correlation between the expression levels of protein and mRNA from the same genes for the separate data sets derived from NB4 cell populations enriched by elutriation for G1 , S , or G2&M phases ( Figure 8B ) . This shows that there is little or no difference in the degree of correlation between protein and mRNA , with the same Spearman rank correlation coefficient ( r = 0 . 65 ) at each of the G1 , S , and G2&M phases . These correlation values are similar to what we measured in the asynchronous NB4 cell data set . We conclude that there is little or no global change in the overall relationship between bulk mRNA and protein levels in NB4 cells at different stages of interphase . Furthermore , based on our measurements and correlations of protein and mRNA levels across the elutriated fractions , the data are consistent with there being a large contribution from post-transcriptional mechanisms to controlling gene expression in NB4 cells . Having established that absolute abundances of protein and mRNA are moderately correlated , and that this correlation is independent of cell cycle phase , we next compared specifically the protein and mRNA expression of the 358 proteins whose abundances are cell cycle regulated , as identified in our proteomic data set . First , we examined whether the correlation of absolute protein and mRNA abundances in asynchronous cells is different for genes encoding proteins whose abundance is cell cycle regulated . As Figure 8C shows , the protein and mRNA correlation for these genes ( r = 0 . 47 ) is weaker than the overall correlation for all expressed genes ( 0 . 63–0 . 65 ) . These data suggest that post-transcriptional mechanisms may contribute to a larger extent in cell cycle-regulated gene expression , as compared with bulk gene expression . We next separately compared the protein and mRNA levels in asynchronous NB4 cells for the four previously determined protein clusters ( see Figure 5 ) , which differ primarily by the phase in which peak expression occurs ( i . e . , G1 , S , G2&M or G2&M+G1 ) . As Figure 8D shows , correlation of protein and mRNA abundances is moderate ( r = 0 . 57 ) for cell cycle-regulated proteins that peak in either G1 or S . In contrast , protein and mRNA abundances are more poorly correlated for proteins that peak either in G2&M ( r = 0 . 31 ) or in G1 and G2&M , that is the G2&M+G1 cluster ( r = 0 . 20 ) . This result is not limited to abundances measured in asynchronous cells , as a similar trend is observed when protein and mRNA abundances are compared in elutriated cells at specific phases of the cell cycle , as shown in Figure 8—figure supplement 2 . For example , the correlation coefficient is 0 . 42 for G2&M-peaking proteins in G2&M-phase elutriated cells , as compared with values of 0 . 67 and 0 . 73 for G1-peaking proteins in G1-phase elutriated cells and S-peaking proteins in S-phase elutriated cells , respectively . These data show that for cell cycle-regulated proteins , the extent to which protein and mRNA abundance correlates is dependent on when during the cell cycle maximum expression occurs . Thus , cell cycle-regulated proteins that peak in G2&M and G2&M+G1 are more likely to have poor protein/mRNA abundance correlation than cell cycle-regulated proteins whose abundance peaks in other phases of the cell cycle . These results indicate that mRNA levels are particularly poor predictors of protein abundance in the case of cell cycle-regulated proteins whose expression peaks in either G2&M or G2&M+G1 . Next , we examined how well the protein and mRNA expression profiles were correlated across the elutriated fractions . Mechanistically , these patterns result from the combined effects of differential transcriptional , translational , and post-translational regulation of gene expression across the cell cycle . Highly concerted protein and mRNA expression levels will typically result from synergistic transcriptional and post-transcriptional regulation . Several genes have been previously found to show highly correlated protein and mRNA expression across the cell cycle , including A-type cyclins . Consistent with previous work in other systems ( Pines and Hunter , 1990 ) , we find that Cyclin A2 protein and mRNA expression is highly correlated in NB4 cells ( Figure 8F , top ) . Detailed analyzes have documented regulatory mechanisms that underlie this correlation: namely , cell cycle-regulated transcription of the cyclin A gene , with cyclin A mRNA synthesis being highest in S and G2 ( Henglein et al . , 1994 ) , and destabilization of cyclin A mRNA and protein during early mitosis and G1 ( Pines and Hunter , 1990; Glotzer et al . , 1991; Henglein et al . , 1994; Dawson et al . , 1995; Sudakin et al . , 1995 ) . These synergistic regulatory mechanisms result in differential expression of the cyclin A gene across the cell cycle . Nearly half of the cell cycle-regulated proteins identified in our proteomic data set showed moderate to high correlation ( r > 0 . 5 ) between protein and mRNA expression patterns ( Figure 8E ) . Expression of these genes , like cyclin A , may also be controlled by concerted regulatory mechanisms . In contrast , over half of the proteins whose abundance is shown here to be regulated across the cell cycle have low correlation ( r < 0 . 5 ) between protein and mRNA expression patterns . Cdt1 gene expression is even anti-correlated in terms of protein and RNA levels ( Figure 8F , bottom ) . Cdt1 mRNA abundance peaks in S-phase , where Cdt1 protein expression is lowest . This surprising inverse relationship between protein and mRNA levels is not likely to be due to low data quality for the Cdt1 gene . A high correlation ( >0 . 75 ) is observed among the eight , independently identified peptides used to quantify Cdt1 protein expression and error bars indicate the relatively low standard error in scaled intensities from the eight supporting peptides . Furthermore , the Cdt1 protein expression pattern across the cell cycle is confirmed by immunoblot analyzes of elutriated NB4 cells ( Figure 8G ) , and is consistent with what is known in other cell types ( Wohlschlegel et al . , 2000 ) . Similarly , the Cdt1 mRNA expression pattern is reproducibly detected , as indicated by the error bars for mRNA quantitation and is consistent with recent reports showing that Cdt1 mRNA levels are high in S-phase due to positive regulation by geminin ( Ballabeni et al . , 2013 ) . In summary , this study of gene expression in NB4 cells indicates that , even though expression profiles for many genes are positively correlated at the transcript and protein level , for a surprisingly large fraction of human genes mRNA abundance alone is not a reliable predictor of the corresponding abundance of the protein encoded by that mRNA . From the data described above , we determined that protein and mRNA expression are moderately correlated with respect to absolute abundance , that the correlation for bulk gene expression is primarily cell cycle independent , but that over half of the proteins identified as cell cycle regulated have discordant mRNA expression patterns . Out of the 358 proteins whose abundances are cell cycle regulated , 31 of the cognate mRNAs also vary across the elutriated fractions by more than 1 . 5-fold . Comparison of protein and mRNA abundance profiles across the elutriated fractions for these genes reveals highly coordinated expression ( Figure 9A , B ) , as measured by the high Pearson correlation coefficient calculated for the mean protein and RNA profiles ( r = 0 . 93 ) . For each of these 31 genes , protein , and RNA abundances in asynchronous ( Figure 9C ) and elutriated G1 , S , and G2&M cells are also positively correlated ( r = 0 . 76 , 0 . 68 , 0 . 79 and 0 . 84 , respectively ) . These data show that specific subsets of genes can be highly coordinately expressed at both the protein and mRNA level . 10 . 7554/eLife . 01630 . 016Figure 9 . Protein and RNA levels are correlated for the specific subset of cell cycle-regulated proteins whose cognate mRNA change by 1 . 5-fold . Of the 358 proteins whose abundances are cell cycle regulated , the cognate mRNA of 31 proteins also changes across the elutriated fractions by more than 1 . 5-fold . Scaled protein ( A ) and mRNA expression profiles ( B ) are shown as line graphs for these 31 genes , respectively . ( C ) Comparison of protein and mRNA abundances in asynchronous cells reveals a Spearman correlation of 0 . 76 . ( D ) 27 of the 31 genes have predicted NF-Y binding sites in their promoters , and all 31 encode proteins containing a KEN or D-box sequence degron . KEN-motifs are especially enriched ( >eightfold enrichment , compared to 7% expected by random chance ) . Three genes have not been previously annotated as being cell cycle regulated: FAM125B , ZNF646 , ARHGAP11A . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 016 We note that for genes whose protein and mRNA levels covary across the cell cycle , coordinated synthesis of protein and mRNA is likely matched by coordinated protein degradation , as it is known that many proteins that are required at high levels in G2&M phase are targeted for degradation by the Anaphase Promoting Complex/Cyclosome ( APC/C ) complex during the late stages of mitosis and G1 ( Amon et al . , 1994 ) . Sequence analysis shows that all of these 31 genes contain at least one sequence motif that is known to target proteins for degradation by the APC/C ( KEN or RxxL ) , and most ( 18/31 ) contain both motifs ( Figure 9D ) . Most of these genes ( 28/31 ) have been previously annotated in the literature as either cell cycle regulated and/or critical for cell cycle progression . Additionally , we observe that 26/31 have predicted NF-Y transcription factor binding sites in their promoters ( 84% ) , which is significantly more frequent than random ( p=2 . 1 × 10−6 ) . The association with NF-Y transcription factor binding sites in the promoter region is less frequent for G2&M-peaking cell cycle-regulated proteins ( ∼50% , as shown in Table 1 , ∼49% if the 31 co-regulated genes are excluded ) , though this is also higher than random ( ∼20% across all promoters ) . Among the 31 genes identified whose protein and mRNA abundances are both coordinately cell cycle regulated are three genes ( ARHGAP11A , ZNF646 , and FAM125B ) that have not been previously annotated as being cell cycle regulated ( Figure 9D ) . We chose to characterize ARHGAP11A further , as it was the only gene coding for a protein for which validated antibodies were readily available . ARHGAP11A encodes a protein that is predicted to function as a RhoGAP , and has been recently shown to be important in regulating formation of the cytokinetic furrow ( Zanin et al . , 2013 ) . Figure 10A shows the MS and RNA-Seq quantitation for ARHGAP11A , indicating that the protein and mRNA abundances are lowest in the first elutriated fractions ( G1 ) and peak in the last elutriated fractions ( G2&M ) . These cell cycle-dependent changes in ARHGAP11A mRNA abundance are consistent with previous microarray studies performed in HeLa cells ( Whitfield et al . , 2002 ) . The variations in protein levels observed by MS were confirmed independently by immunoblot analysis of elutriated lysates ( Figure 10B ) , using a specific antibody that was validated by siRNA depletion ( Figure 10—figure supplement 1 ) . In contrast , GAPDH does not vary in abundance across the elutriated lysates . Thus , the MS and immunoblot data both show that ARHGAP11A is a cell cycle-regulated protein whose expression peaks in G2&M and is lowest in G1 . 10 . 7554/eLife . 01630 . 017Figure 10 . Identification of ARHGAP11A as a cell cycle regulated protein and a substrate of the APC/C . ( A ) MS and RNA-Seq quantitation for ARHGAP11A protein ( left ) and mRNA ( right ) , respectively . ( B ) Immunoblot analysis of ARHGAP11A ( HPA antibody ) and GAPDH protein expression across asynchronous and elutriated NB4 cells . ( C ) Lysates from U2OS cells treated with either a non-targeting control siRNA ( lane 1 ) or siRNAs targeting Cdh1 and Cdc20 ( lane 2 ) were probed for levels of ARHGAP11A , Cdh1 , Cdc20 , cyclin B1 , and GAPDH by immunoblot . ( D ) Asynchronous or serum-starved RPE-1 cells were treated with either a non-targeting control siRNA or an siRNA against Cdh1 . Lysates were then probed with antibodies against ARHGAP11A ( Bethyl antibody ) , Cdh1 , cyclin B1 , and GAPDH . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 01710 . 7554/eLife . 01630 . 018Figure 10—figure supplement 1 . Validation of anti-ARHGAP11A antibodies by siRNA-based depletion of ARHGAP11A protein . ( A ) HeLa cells were either mock treated , treated with siRNAs that target lamin A/C , or treated with different concentrations of siRNAs that target ARHGAP11A . Cells were cultured for 48 hr before harvest , lysis , and immunoblot analysis with a Human Protein Atlas ( HPA ) antibody recognizing ARHGAP11A ( top ) or an antibody recognizing lamin A/C ( bottom ) . The anti-ARHGAP11A recognizes a band at the correct molecular weight ( ∼100 kDa ) , which is significantly decreased upon siRNA depletion of ARHGAP11A protein . Note that lamin A/C levels are not significantly perturbed by treatment of siRNA targeting ARHGAP11A . ( B ) U2OS cells were either mock treated or treated with siRNAs that target ARHGAP11A , incubated for 48 hr , lysed , and immunoblotted with antibodies recognizing ARHGAP11A ( Bethyl , top ) or alpha-tubulin ( bottom ) . The anti-ARHGAP11A recognizes a band at the same molecular weight as the HPA antibody that is significantly decreased upon siRNA depletion of ARHGAP11A . Note that alpha-tubulin levels are unchanged +/− ARHGAP11A siRNA . *Non-specific bands . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 018 Given that the expression of ARHGAP11A is the lowest in G1 and that the protein sequence contains several degrons ( one KEN and two D-box motifs ) , we next tested whether levels of ARHGAP11A protein are regulated by the APC/C . The APC/C is a multimeric protein complex whose activity and specificity is cell cycle regulated by interactions with co-activating factors Cdh1 and Cdc20 ( Visintin et al . , 1997 ) . Intriguingly , Cdc20 was among the 31 genes identified as being regulated both at the protein and mRNA level ( Figure 9 ) . Disruption of the APC/C complex , for example by siRNA-mediated depletion of Cdh1 and/or Cdc20 , is expected to stabilize substrate levels . Thus , we assayed whether ARHGAP11A levels are dependent on APC/C activity by immunoblotting lysates from cells where Cdh1 and/or Cdc20 have been transiently depleted in both U2OS and RPE-1 cell lines using siRNAs . U2OS cells were treated with specific siRNAs that target either Cdh1 or Cdc20 , mRNA for degradation . Transient depletion of both Cdh1 and Cdc20 decreases APC/C activity , as evidenced by a stabilization of cyclin B1 protein ( Figure 10C ) , whose levels are known to be regulated by the APC/C ( King et al . , 1995 ) . More importantly , disruption of APC/C activity in U2OS cells increases levels of ARHGAP11A ( Figure 10C ) . Similarly , transient depletion of Cdh1 in RPE-1 cells resulted in stabilization of the ARHGAP11A protein in asynchronous cells ( Figure 10D ) . G1-arrest by serum starvation results in very low levels of ARHGAP11A . Transient depletion of Cdh1 results in stabilization of ARHGAP11A , thus showing that ARHGAP11A levels are APC/CCdh1-dependent during G1-phase ( Figure 10D , lanes 3 and 4 ) . The blots were also probed with anti-cyclin B1 antibodies to confirm disruption of APC/C activity ( Figure 10D ) . These data independently show that ARHGAP11A levels are lower in G1-phase than in asynchronous RPE-1 cells ( Figure 10D , lanes 1 and 3 ) , which is consistent with the cell cycle regulation of ARHGAP11A abundance observed in elutriated NB4 cells . Thus , we conclude that ARHGAP11A is a cell cycle-regulated protein , and that its levels are regulated by targeted degradation mediated by the APC/C . In addition to uploading the raw MS ( http://www . ebi . ac . uk/pride/archive/projects/PXD000678 ) and RNA-Seq data files ( https://www . ebi . ac . uk/ega/datasets/EGAD00001000736 ) to public repositories ( PRIDE and EGA for MS and RNA-Seq data , respectively ) , we have incorporated the entire analyzed protein and RNA data sets from this study into the Encyclopedia of Proteome Dynamics ( http://www . peptracker . com/epd/ ) , a publicly-available , searchable web resource ( Larance et al . , 2013 ) . The EPD aggregates proteomics data from this study on the myeloid cell cycle with our previous large-scale studies on protein complexes , subcellular localization and turnover in HeLa and U2OS cells . The EPD facilitates cross-correlation and analysis of protein properties across numerous , multidimensional proteomics studies . Additionally , for any specific protein , users can quickly retrieve all protein properties measured so far by providing a protein identifier , such as a Uniprot ID . In this study , we highlight as an example the EPD page for cyclin B1 ( Figure 11 ) , which displays the protein and RNA quantitation across the separate elutriated fractions from this study , and the protein abundance bins to which cyclin B1 belongs . Users can also retrieve any previously determined properties for this protein , such as putative interaction partners , turnover rate and half-life , subcellular localization and estimated abundances in other cell lines . 10 . 7554/eLife . 01630 . 019Figure 11 . The Encyclopedia of Proteome Dynamics , a fully searchable , open-access online repository of proteome data . Quantitative protein and RNA data from this study are available through the Encyclopedia of Proteome Dynamics ( EPD ) . A screenshot of the EPD is shown , which displays protein and mRNA expression profiles across the elutriated fractions , the calculated Pearson correlation coefficient between the protein and mRNA profiles , and protein and mRNA abundances in asynchronous cells for cyclin B1 ( CCNB1_HUMAN ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 019
To facilitate analysis specifically of physiologically relevant variations in gene expression across the cell cycle , we developed an experimental strategy that avoided the use of synchronization procedures to accumulate cells blocked at specific cell cycle stages . In previous studies , this is usually done using either chemical inhibitors or genetic depletion of essential factors , to either activate checkpoints or otherwise block progression through the cell cycle , thus allowing a large enough population of cells at the same cell cycle stage to be harvested for subsequent biochemical analysis . Although effective , the potential disadvantage of these approaches is that they inevitably cause a major metabolic perturbation and/or stress to achieve the cell cycle block and this in turn may affect gene expression in ways that would not occur during normal cell cycle progression . In addition , to minimize possible effects of specialized media on the normal cell cycle , we also avoided here using cell media with dialyzed serum , as frequently used for metabolic labeling of cells with heavy isotope substituted amino acids . Consequently , we focused on growing cells in normal sera and identified proteins by performing label-free MS analysis and we used centrifugal elutriation as the method to generate sub-populations of cells enriched at different cell cycle stages . Centrifugal elutriation is a simple yet effective physical method of enriching for cells at different cell cycle phases that is suitable for isolating sufficient quantities of cells for large-scale biochemical analysis ( Banfalvi , 2008 ) . We used a variety of criteria here , including FACS analysis and immunoblotting , to detect multiple known cell cycle-regulated and control proteins , thus validating the successful separation of NB4 cells into six fractions that were differentially enriched for cells at different cell cycle stages . Compared with using metabolic inhibition to block cells at different cell cycle stages , elutriation generates minimal stress or disruption of cellular metabolism and physiology . We confirmed that here by showing that the elutriated NB4 cells remained viable and continued to grow and proliferate when returned to culture post elutriation without showing evidence of changes in morphology . Because the separation principle of elutriation is based on physical size , we note that this method can also be used to separate cells of different sizes and thus could also be employed to examine changes in either gene expression or other processes , associated with cell size variation ( Bjorklund et al . , 2006; Tzur et al . , 2009; Navarro et al . , 2012 ) . In combination with centrifugal elutriation , we thus determined both protein and mRNA levels for more than 6000 human genes in each of the six , separate elutriated fractions . Analysis of these data highlighted examples of three major types of protein cell cycle regulation , that is , changes in protein abundance , isoform-specific changes in abundance and changes in protein phosphorylation . We identified ∼5% of genes as encoding proteins whose abundance varies across the cell cycle by at least twofold . These genes formed seven distinct clusters based upon the cell cycle stage where the protein showed maximum expression level . As this study represents a re-evaluation of cell cycle-regulated gene expression and was not influenced by expectations from the literature , it was reassuring that it identified and confirmed so many previously documented cell cycle-regulated factors . In addition , it detected novel cell cycle regulated genes and showed that the current gene ontology annotations are primarily associated with genes in the clusters showing peak expression at either entry or exit from mitosis . However , other clusters show collections of genes that also express proteins whose abundance is regulated at other stages of interphase and we propose that these should also be annotated in future as ‘cell cycle regulated’ and that this term should not be restricted to genes linked with mitosis . Our analysis of the genes encoding proteins whose abundance is cell cycle-regulated implicates several transcription factors as being particularly important for the control of protein expression during the cell cycle . For example , in addition to recapitulating the previously described importance of E2F transcription factor activity in S-phase ( Sherr , 1996 ) , these data also highlight the potentially important roles for the NF-Y and STAT3 transcription factors for gene regulation in G2&M and G1&G2 , respectively . The enrichment of NF-Y binding sites we observed in the promoters of genes encoding proteins that peak in the G2&M fraction is consistent with previous reports linking NF-Y transcriptional activity to G2&M phase cell cycle functions ( Hu et al . , 2006 ) . In comparing our present results with previous data on cell cycle-regulated gene expression in human cells , it is interesting that the rather low number of cell cycle-regulated proteins detected here in elutriated NB4 cells ( ∼5% ) , contrasts with a much higher estimate of ∼40% of the detectable proteome varying by at least twofold in abundance across the cell cycle in HeLa cells ( Olsen et al . , 2010 ) . This proportion was determined using thymidine- and nocodazole-synchronized HeLa cells and is significantly higher than what was reported for HeLa cells synchronized by thymidine alone ( ∼15% ) ( Lane et al . , 2013 ) . The difference in the proportion of cell cycle-regulated proteins observed between these studies may be due to technical differences , analysis criteria , growth conditions , the type of quantitation employed and/or the synchronization method . Differences between studies performed in HeLa and NB4 cells may also reflect tissue-specific and/or cell line-specific plasticity in cell cycle regulation in mammalian systems . The data in this study allowed us to examine in detail the relationship between protein and mRNA abundance , both in terms of global genomic expression and for specific sets of genes and for cells at different stages of cell cycle progression . The results showed that this relationship is remarkably complex and clearly indicate that measurements of mRNA levels alone cannot be relied upon to provide an accurate reflection of protein abundance in many situations . Indeed , as illustrated for the Cdt1 gene , we identify examples where protein and mRNA levels are even inversely correlated . The rather complex relationship we observe between protein and mRNA levels is important for interpreting previous studies that have relied on using either microarray or RNA-Seq data alone as a surrogate for reporting on the regulation of protein expression . Our data on NB4 cells show that for over 6000 genes protein and mRNA abundances are positively correlated , both in asynchronously growing cells and in cells at G1 , S , and G2&M phases , though in each case with a moderate correlation coefficient of ∼0 . 63 to 0 . 65 . However , this correlation is weaker ( 0 . 47 ) for the ∼5% of genes encoding proteins whose abundance varies across the cell cycle . Interestingly , this subset of cell cycle-regulated genes shows further heterogeneity in the relationship between protein and mRNA abundance levels when examined more closely to compare the separate clusters of genes encoding proteins whose expression peaks at different stages . Thus , we observe dramatically varying correlations in protein/mRNA abundance levels in different clusters ( Figure 8 ) . This ranges from the very low value of ∼0 . 2 ( ‘G2&M+G1’-peaking proteins , cluster 4 ) , up to the higher than average value of ∼0 . 69 for proteins with peak abundance in S phase ( cluster 2 ) . The NB4 cell data show that a subset of the genes that encode proteins whose abundance is cell cycle regulated is concordantly expressed at the protein and mRNA level . This is the case particularly for genes encoding proteins whose abundance peaks in either S or in the G2&M phases of the cell cycle . This suggests that transcriptional regulation mechanisms contribute significantly to modulating protein expression levels in these phases , but less so in G1 . A likely explanation for this observation , consistent with the literature on cell cycle regulatory mechanisms , is that post-transcriptional controls , including targeted protein degradation and regulation of translation , operate differentially across the cell cycle and cause temporal imbalances in the relation between transcript levels and proteins . It is well known that mechanisms exist for controlling the targeted degradation of specific proteins , for example , based on the substrate-specific action of E3 ligases targeting proteins for degradation by the proteasome ( King et al . , 1996 ) . A number of cell cycle-regulated proteins have been shown to be targeted for degradation at specific cell cycle stages , as exemplified by mechanisms such as the degradation of proteins by the APC/C at the end of mitosis and into G1 phase ( King et al . , 1996; Reed , 2003; Pines , 2011 ) . Indeed , recent high-throughput studies of protein turnover in HeLa and U2OS cells demonstrated that the most rapidly degraded group of proteins were strongly associated with the GO terms ‘cell cycle’ and ‘mitosis’ ( Boisvert et al . , 2012; Larance et al . , 2013 ) . In this study , we identify many known cell cycle-regulated APC/C substrates , including aurora kinases ( Honda et al . , 2000 ) and securin ( Zur and Brandeis , 2001 ) . In addition , we validated here a novel cell cycle-regulated protein ( ARHGAP11A ) , whose regulation is mediated by targeted degradation by the APC/C . In addition to the APC/C , whose activity is restricted to mitosis and G1 , there are other complexes that target proteins for degradation in other phases of the cell cycle . For example , the SCFSkp2 complex targets proteins for degradation during and at the entry of S-phase . Substrates include Cdt1 ( Wohlschlegel et al . , 2000 ) , which is critical for replication origin licensing and Cep192 , whose hydroxyproline-modified form is targeted for degradation by the SCFSkp2 complex ( Moser et al . , 2013 ) . We note that the clustering analysis identified a subset of NB4 proteins whose abundances are similarly regulated across the cell cycle as Cdt1 ( Figure 5 , cluster 7 ) . Proteins with Cdt1-like expression patterns may be similarly regulated by SCFSkp2; however , additional experiments are required to test these hypotheses . In contrast with the extensive literature on cell cycle-regulated protein degradation , studies examining translational control across the cell cycle , particularly high-throughput studies , are few in number . However , a recent study using puromycin-analogs and in vitro immunoprecipitation in thymidine-synchronized HeLa cells revealed that a subset of the proteome ( 339/4984 proteins ) is differentially translated across the cell cycle , while the translation of most proteins remains relatively constant ( Aviner et al . , 2013 ) . Interestingly , these authors propose that mRNA translation is particularly important in regulating G2&M-associated proteins . It will be interesting in future to examine the contribution of translational regulation mechanisms to the cell cycle variations in protein abundance measured here in NB4 cells . We have shown that the use of a dual protease digestion strategy combined with extensive prefractionation of isolated peptides by strong anion exchange-enabled proteomic characterization of a leukemia cell line to a depth of over 10 , 000 proteins , identified with on average 15 peptides per protein and with high sequence coverage ( mean ∼38% ) . To the best of our knowledge , this data set , together with the accompanying RNA-Seq data , provides the most comprehensive study to date of gene expression in human myeloid cells . The depth of proteome sequence coverage is comparable with , if not exceeding , the proteome coverage obtained in recent deep proteome studies performed on other human tumor cell lines , including epithelial tumor cell lines such as HeLa and U2OS . The proteomic workflow is not exclusive to either suspension cells or leukemia cell lines and , based on current technology and instrumentation , is applicable in principle to any cell type where ∼200 μg of protein can be isolated . The high protein sequence coverage obtained was shown to be particularly helpful for allowing a more detailed comparison of gene expression and regulation at the level of separate protein isoforms . Importantly , using the information from the high number of independent peptide identifications per protein group , we could show clear examples where genes encoded multiple isoforms , only a subset of which were cell cycle regulated in their expression pattern . This also showed that aggregating all of the peptide information and interpreting it in terms of the behavior of a single hypothetical polypeptide , as typically done in proteomic studies , would lead to an incorrect conclusion that the corresponding gene was not subject to cell cycle regulation because the peptide information for the cell cycle-regulated isoform was diluted by the contribution of the peptides shared with the other , unregulated isoforms encoded by the same gene . We infer that our current data underestimate the total number of polypeptides whose abundance is cell cycle regulated , not only because we lack combined protein and mRNA data across the full cell cycle for at least 4000 additional genes whose expression was detected in the asynchronous NB4 cell populations , but also because we lack sufficient numbers of peptides to be confident that we are efficiently detecting and quantifying most of the separate protein isoforms that are expressed . Our data therefore highlight the need for further technological development in proteomics methods and instrumentation because even in the deepest analyzes reported to date , as with this present study , still less than 50% of the total protein sequence is identified for the genes we can detect being expressed at the protein level . It is also difficult to detect multiple peptides consistently across every sample in a complex experiment , as is the case here with multiple elutriated cell populations . Nonetheless , we anticipate that continued improvements in instrumentation , combined with improved sample preparation , will provide additional sequencing depth and speed in the future and the present study illustrates how this can be used to produce a more comprehensive mapping of gene expression and regulation during fundamental biological processes . A meta-analysis of the data from human cell lines where the most detailed proteomic information is available revealed a set of proteins whose expression was only detected here in NB4 cells . This NB4-specific protein set was enriched for transcription factors , including proteins that are already known to be important for myeloid cell differentiation , such as PU . 1 and C/EBPα/δ ( Orkin and Zon , 2008 ) . We also identified a core set of proteins that were detected in myeloid-derived cell lines ( K562 and NB4 ) , but not in other cell lines , most of which are non-leukemic and from epithelial origin . Many genes encoding these proteins are overexpressed at the mRNA level in normal blood , leukemia , and lymphoma cells compared to other normal and tumor tissues in the Broad Global Cancer Map ( Figure 12 ) ( Subramanian et al . , 2005 ) , and are enriched in genes that are overexpressed in leukemia/lymphoblastic tissues in the Novartis GNF tissue expression database ( 25/87 genes , p=0 . 00059 ) ( Su et al . , 2002 ) . The proteomic data here provide direct evidence that many of the overexpressed mRNAs observed are translated into protein and that these proteins are likely overexpressed in cancer cells of the myeloid lineage compared to cancer cells derived from other tissues . For comparison , HeLa-specific genes were similarly analyzed and found to be overexpressed in normal uterine tissue and prostate tumors ( Figure 12 ) . Given the high variance observed between mRNA and protein abundance , we note that direct experimental evidence of protein overexpression has added benefits to clinical pathology and diagnostics . 10 . 7554/eLife . 01630 . 020Figure 12 . Many cell-line specific genes are overexpressed in tumors and normal tissues that are associated with the developmental origin of the cell line . mRNA expression heatmaps from the Broad Global Cancer Map for NB4- and K562- specific genes ( left ) and HeLa-specific genes ( right ) . Each heatmap has tissue along the horizontal axis and gene along the vertical axis . Vertical red streaks indicate that many genes are similarly overexpressed in a particular tissue . Many NB4- and K562-specific genes are overexpressed in lymphoid , leukemia , and normal hematopoietic tissues , whereas HeLa-specific genes are overexpressed in normal uterine tissues and prostate tumors . DOI: http://dx . doi . org/10 . 7554/eLife . 01630 . 020 Gene regulation that extends to the protein level , including cell cycle-dependent regulation , broadly , encompasses the modulation of any properties of proteins and protein isoforms and not just variations in protein abundance , as we have focused on in this study . A more comprehensive analysis of cell cycle regulation of gene expression should thus in future be extended to analyze also variations in the subcellular localization of the proteome , changes in protein complex formation and protein–protein interactions and a more detailed description of protein isoform expression and post-translational protein modifications . Methods are now emerging that should allow the systematic and quantitative analysis of these varied properties at a system-wide level . It will be important also to repeat such in depth studies on gene expression across a wide range of cell types , particularly also in primary cells , to evaluate what types of mechanisms are used ubiquitously and which are used in conjunction with specific needs of specialized cell types and/or modified by the process of cell transformation and influenced by oncogene expression . It will also be interesting to compare the results of cell cycle-regulated gene expression detected here using centrifugal elutriation to separate cells into distinct populations enriched at different cell cycle stages with comparable analysis of cells , where cell cycle progression is blocked with inhibitors to compare what effect such metabolic perturbations may have on gene expression , separate to the normal control of the cell cycle . We have collated all of the data from this study into the Encyclopedia of Proteome Dynamics , a searchable online database ( http://www . peptracker . com/epd ) . The data here are combined with other high-throughput studies of protein properties in HeLa and U2OS cells , including the system-wide analysis of protein turnover and protein degradation rates in separate subcellular compartments and the analysis of native , multi-protein complexes separated by size exclusion chromatography ( Ahmad et al . , 2012; Boisvert et al . , 2012; Kirkwood et al . , 2013; Larance et al . , 2013 ) . In common with this present study , all of these data show the importance of moving beyond simple protein identification to a more detailed analysis of complex proteome dynamics , including the analysis of selective regulation of distinct protein isoforms and post-translational modifications for deciphering cellular regulation mechanisms . Furthermore , the combining of highly annotated large data sets in this format not only adds value through sharing of information with the community , it also facilitates a Super-Experiment approach ( Boulon et al . , 2010 ) . Here , the value of each individual data set is enhanced by allowing the detailed cross-comparison and analysis of protein behavior and responses between different cell types and under different growth conditions . We suggest that this provides a useful model that could be extended in future to provide a resource incorporating data generated at a community wide level .
The NB4 cell line was established from long-term cultures of acute myeloid leukemia blast cells grown on bone-marrow stromal fibroblasts ( Lanotte , 1991 ) . NB4 cells were obtained from the Hay laboratory ( University of Dundee ) . The cells were cultured at 37°C in the presence of 5% CO2 as a suspension in RPMI-1640 ( Life Technologies , United Kingdom ) supplemented with 2 mM L-glutamine , 10% vol/vol fetal bovine serum ( FBS , Life Technologies ) , 100 units/ml penicillin and 100 μg/ml streptomycin ( 100X stock , Life Technologies ) . Cell cultures were maintained at densities between 1 × 105 and 1 × 106 cells/ml and harvested by centrifugation when cultures reached ∼8 × 105 cells/ml . Pellets containing ∼5 × 108 cells were resuspended in 5 ml elutriation buffer ( PBS +1% FBS ) . The resulting cell suspension was passed through a 19G-needle three times to disaggregate cell clumps and then loaded into a Beckman counterflow centrifugal elutriator ( Beckman JE-5 . 0/JE ) , equipped with a standard elutriation chamber and a Cole–Parmer MasterFlex Model 900-292 peristaltic pump . The centrifuge was operated at 1800 rpm and the flow rate was initially set to 16 . 68 ml min−1 . After cells have been loaded into the elutriation chamber , 50 ml fractions were collected at the following flow rates: 21 . 18 , 23 . 88 , 25 . 68 , 27 . 47 , 29 . 27 , and 38 . 27 ml min−1 ( i . e . , fractions 1 through 7 ) . A 2 ml aliquot containing a minimum of 5 × 105 cells from each fraction was reserved for flow cytometry . The remaining cells were harvested for protein and RNA extraction . Cell yields range from >8 × 107 cells ( Fraction 1 , G1 ) to 2 × 106 cells ( Fraction 6 , G2&M ) , which reflect the typical cell cycle phase distribution found in cultured cell lines . Fraction 6 , which yields the lowest cell number , still provides ∼1 . 0 mg of total protein , which is sufficient for in-depth proteomics analysis using current technology ( min ∼200 μg ) . For protein extraction , NB4 cells were pelleted , washed twice with cold PBS and then lysed in 0 . 3–1 . 0 ml HES lysis buffer ( 2% SDS , 10 mM HEPES pH 7 . 4 , 1 mM EDTA , 250 mM sucrose , Roche protease inhibitors , Roche PhosStop; UK ) . Lysates were heated to 95°C for 10 min and homogenized using Qiashredder ( Qiagen ) . 200 μg of the lysate was further processed for LC-MS/MS analysis using a modification of the FASP protocol ( Wisniewski et al . , 2009 ) . Briefly , lysates were loaded onto pre-equilibrated 30 kD-cutoff spin columns ( Sartorius UK ) and washed twice using denaturing urea buffer ( 8 M urea , 10 mM Tris , pH 7 . 4 ) . Proteins were reduced with TCEP ( 25 mM in denaturing urea buffer ) , for 15 min at room temperature and alkylated with iodoacetamide ( 55 mM in denaturing urea buffer ) , in the dark for 45 min at room temperature . Lysates were then buffer-exchanged into 0 . 1 M triethylammonium bicarbonate , pH 8 . 5 ( TEAB , Sigma ) and digested with trypsin ( 1:50 , Promega UK ) overnight at 37°C . Digestion efficiency was checked by SDS-PAGE analysis and protein staining with SimplyBlue SafeStain ( Life Technologies ) . After collecting the first peptide flow-through , the spin column was washed twice with 0 . 1 M TEAB , then twice with 0 . 5 M NaCl . The flow-through and washes were combined and desalted using SepPak-C18 SPE cartridges ( Waters UK ) . Peptides were then dried and resuspended in 5% formic acid for LC-MS/MS analysis . For protein extraction , NB4 cells were pelleted , washed twice with cold PBS and then lysed in 0 . 3–1 . 0 ml urea lysis buffer ( 8 M urea , 100 mM Tris pH 7 . 4 , Roche protease inhibitors , Roche PhosStop ) . Lysates were vigorously mixed for 30 min at room temperature and homogenized using a Branson Digital Sonifier ( 30% power , 30 s ) . Proteins were reduced with TCEP ( 25 mM in denaturing urea buffer ) , for 15 min at room temperature and alkylated with iodoacetamide ( 55 mM in denaturing urea buffer ) , in the dark for 45 min at room temperature . Lysates were diluted with digest buffer ( 100 mM Tris pH 8 . 0 + 1 mM CaCl2 ) to reach 4 M urea , and then digested with 1:50 Lys-C ( Wako Chemicals , Japan ) overnight at 37°C . The digest was then split into two fractions . The first was retained as the Lys-C digest , which was shown previously to produce peptides that are complementary to trypsin ( Swaney et al . , 2010 ) . The second was diluted with digest buffer to reach 0 . 8 M urea and double-digested with trypsin ( Promega; 1:50 ) . Digest efficiencies were checked by SDS-PAGE analysis and protein staining . The digests were then desalted using SepPak-C18 SPE cartridges , dried , and resuspended in 50 mM borate , pH 9 . 3 . Peptides were separated onto a Dionex Ultimate 3000 HPLC system equipped with an AS24 strong anion exchange column , using a similar protocol to the hSAX method described previously ( Ritorto et al . , 2013 ) . Peptides were chromatographed using a borate buffer system , namely 10 mM sodium borate , pH 9 . 3 ( Buffer A ) and 10 mM sodium borate , pH 9 . 3 + 0 . 5 M sodium chloride ( Buffer B ) and eluted using an exponential elution gradient into 12 × 750 μl fractions . The peptide fractions were desalted using SepPak-C18 SPE plates and then resuspended in 5% formic acid for LC-MS/MS analysis . For tryptic digests , including tryptic + Lys-C double digests , peptide chromatography was performed using a Dionex RSLCnano HPLC . Peptides were loaded onto a 0 . 3 mm id × 5 mm PepMap-C18 pre-column and chromatographed on a 75 μm × 15 cm PepMap-C18 column using the following mobile phases: 2% acetonitrile +0 . 1% formic acid ( Solvent A ) and 80% acetonitrile +0 . 1% formic acid ( Solvent B ) . The linear gradient began with 5% B to 35% B over 156 min with a constant flow of 300 nl/min . The peptide eluent flowed into a nanoelectrospray emitter at the front end of a Velos Orbitrap mass spectrometer ( Thermo Fisher , San Jose , CA ) . A typical ‘Top15’ acquisition method was used . Briefly , the primary mass spectrometry scan ( MS1 ) was performed in the Oribtrap at 60 , 000 resolution . Then , the top 10 most abundant m/z signals were chosen from the primary scan for collision-induced dissociation and MS2 analysis in the Orbitrap mass analyzer at 17 , 500 resolution . Precursor ion charge state screening was enabled and all unassigned charge states , as well as singly charged species , were rejected . For Lys-C digests , peptide chromatography was also performed using a Dionex RSLCnano HPLC . Peptides were directly injected onto a 75 μm × 50 cm PepMap-C18 column using the following mobile phases: 2% acetonitrile +0 . 1% formic acid ( Solvent A ) and 80% acetonitrile +0 . 1% formic acid ( Solvent B ) . The linear gradient began with 5% B to 35% B over 220 min with a constant flow rate of 200 nl/min . The peptide eluent flowed into a nanoelectrospray emitter at the front end of a Q-Exactive ( quadrupole Orbitrap ) mass spectrometer ( Thermo Fisher ) . A typical ‘Top10’ acquisition method was used . Briefly , the primary mass spectrometry scan ( MS1 ) was performed in the Oribtrap at 70 , 000 resolution . Then , the top 10 most abundant m/z signals were chosen from the primary scan for collision-induced dissociation in the HCD cell and MS2 analysis in the Orbitrap at 17 , 500 resolution . Precursor ion charge state screening was enabled and all unassigned charge states , as well as singly charged species , were rejected . NB4 cell pellets from elutriation were resuspended in 0 . 25 ml PBS and immediately lysed by addition of 0 . 75 ml Trizol LS ( Sigma , United Kingdom ) . RNA extraction was then performed according to manufacturer’s instructions . Extracted RNA was resuspended in nuclease-free water and quantified by fluorometry using the RNA Qubit assay ( Life Technologies ) . Fractions with similar cell cycle phase profiles were combined to produce samples enriched in G1 ( fractions 1 + 2 ) , S ( fractions 3 + 4 ) , and G2&M ( fractions 5 + 6 ) RNA . The integrity of the total RNA was assessed using an Agilent Bioanalyser . Two biological replicates were analyzed in technical duplicate by standard Illumina RNA-Seq . Briefly , mRNA was extracted using oligo dT beads , fragmented , then reverse transcribed using random hexamers . The cDNA was then sequenced using paired ends reads at a length of 75 bp . Each sample was run on a single lane of an Illumina HiSeq , to improve coverage of lower abundance transcripts . A suite of custom scripts was designed to evaluate the quality of the resultant RNA-Seq data . Briefly , the data were evaluated for standard sequencing metrics including GC content and percent of reads with a quality score either greater or equal to 30 . RNA-Seq specific effects were scrutinized including evenness of coverage across the transcriptome , absence of significant 3’ bias , successful reduction of ribosomal RNA and high complexity of the sequenced fragments ( determined by unique start and end positions of the insert ) . The paired-end RNA-Seq data were then aligned to the human genome ( build hg19 ) , using TopHat , without providing a gene reference ( to avoid forced mappings ) . Following duplicate removal using Picard’s MarkDuplicate ( http://picard . sourceforge . net ) , we quantified the gene expression of known protein coding genes using Cufflinks ( Trapnell et al . , 2013 ) . Depletion of Cdc20 and Cdh1/Fzr1 utilized pools of four siRNAs at a final , total concentration of 20 nM ( Dharmacon ) . Lipofectamine RNAiMax ( Life technologies ) transfection reagent was used according to manufacturer’s recommendations . Negative control ( firefly luciferase ) siRNA sequence is: 5′CGUACGCGGAAUACUUCGA . Cdc20 siRNA sequences are: 5′CGGAAGACCUGCCGUUACA , 5′GCAGAAACGGCUUCGAAAU , 5′GGGCCGAACUCCUGGCAAA , 5′GCACAGUUCGCGUUCGAGA . The lamin A/C siRNA sequence is 5′CUGGACUUCCAGAAGAACA . Cdh1/Fzr1 siRNA sequences are identical to those used in previous studies ( Emanuele et al . , 2011 ) . Depletion of ARHGAP11A utilized pools of four siRNAs at a final , total concentration of 20 nM , unless otherwise specified ( sequences: 5′UACAGACUCUUAUCGAUUA , 5′GUUCGAAGAUCUCUGCGUU , 5′GGUAUCAGUUCACAUCGAU , 5′AAGCGAUCAUUGCCAGUAG ) . hTERT-RPE1 cells were harvested 24 hr post transfection for asynchronous populations . For G1 populations , hTERT-RPE1 cells were changed into serum free medium 24 hr post transfection and harvested 24 hr after that ( 48 hr post transfection ) . U2OS cells were transfected with siRNA pools targeting both Cdc20 and Cdh1 and harvested 24 hr later . Lysates were separated by SDS-PAGE , transferred to nitrocellulose membranes and immunoblotted using antibodies recognizing either GAPDH ( sc-25778; Santa Cruz , USA ) , Cdh1 ( ab3242; Abcam ) or ARHGAP11A ( HPA040419; Sigma UK and A303-097A; Bethyl USA ) following standard procedures . NB4 cells ( 5 × 105 cells , minimum ) were resuspended in cold 70% ethanol and fixed at room temperature for 30 min . The fixed cells were then washed twice with PBS and resuspended in PI stain solution ( 50 μg/ml propidium iodide and 100 μg/ml ribonuclease A in PBS ) . The cells were incubated in PI stain solution for 30 min and then analyzed by flow cytometry on a FACScalibur ( BD Biosciences UK ) . An asynchronous population of cells was used as a control to adjust flow cytometer settings , which then remained constant throughout analysis of the set of elutriated fractions . The flow cytometry data were analyzed using FlowJo ( Tree Star , Inc . , OR , USA ) . Lysates for SDS-PAGE analysis were prepared in lithium dodecylsulfate sample buffer ( Life Technologies ) and 25 mM TCEP . Samples were heated to 65°C for 5 min and then loaded onto a NuPage BisTris 4–12% gradient gel ( Life Technologies ) , in either MOPS or MES buffer . Proteins were electrophoresed and then transferred to nitrocellulose membranes using program 3 ( 7 min ) on the iBlot dry blotting system ( Life Technologies ) . Membranes were then blocked in 3% milk in immunoblot wash buffer ( TBS +0 . 1% Tween-20 ) for 1 hr at room temperature . Membranes were then probed with primary antibody overnight at 4°C , washed and then re-probed with HRP-conjugated secondary antibody . Primary antibodies for cell cycle immunoblot analysis were obtained from BD Biosciences ( aurora kinase B ) , Atlas Antibodies ( ARHGAP11A , HPA040830 ) and from Cell Signaling Technology ( cyclin B1 , cyclin A2 , cyclin E , phospho-Histone H3-S10 ) . Bands were visualized using enhanced chemiluminescence ( Millipore Immobilon UK ) and CCD camera detection ( FujiFilm LAS-4000 system ) . The RAW data files produced by the mass spectrometer were analyzed using the quantitative proteomics software MaxQuant , version 1 . 3 . 0 . 5 ( Cox and Mann , 2008 ) . This version of MaxQuant includes an integrated search engine , Andromeda ( Cox et al . , 2011 ) . The database supplied to the search engine for peptide identifications was a UniProt human protein database ( ‘Human Reference Proteome’ retrieved on 19 August 2012 ) combined with a commonly observed contaminants list . The initial mass tolerance was set to 7 p . p . m . and MS/MS mass tolerance was 0 . 5 Da . Enzyme was set to trypsin/P with up to 2 missed cleavages . Deamidation , oxidation of methionine and Gln->pyro-Glu were searched as variable modifications . Identification was set to a false discovery rate of 1% . To achieve reliable identifications , all proteins were accepted based on the criteria that the number of forward hits in the database was at least 100-fold higher than the number of reverse database hits , thus resulting in a false discovery rate of less than 1% . Protein isoforms and proteins that cannot be distinguished based on the peptides identified are grouped by MaxQuant and displayed on a single line with multiple UniProt identifiers . The label free quantitation ( LFQ ) algorithm in MaxQuant was used for protein quantitation . The algorithm has been previously described ( Luber et al . , 2010 ) . The MaxQuant data analysis was repeated with searches for the following post-translational modifications: Phospho ( STY ) , Methyl/Di-Methyl ( KR ) , and Acetyl ( K ) . Protein quantitation was performed on unmodified peptides and peptides that have modifications that are known to occur during sample processing ( pyro-Glu , deamidation ) . All resulting MS data were integrated and managed using Data Manager , a laboratory information management system ( LIMS ) that is part of the PepTracker software platform ( http://www . PepTracker . com ) . The downstream data interpretation ( protein and RNA data ) of cell cycle stages in this study was performed primarily using the R language ( version 0 . 95 . 262 ) . An initial cleaning step was performed to improve the quality and value of the data set . This step involved removing proteins with less than 2 peptide identifications , those labeled as either contaminants or reverse hits and those where data were missing in any of the fractions . Proteins were further filtered using a procedure analogous to a ‘checksum’ function in computing . An algorithm was constructed to assess the self-consistency of the quantitation based on known relationships between the elutriated fractions . The intensities measured in the asynchronous NB4 cell population can thus be modeled as a linear combination of the intensities originating from the six elutriated fractions that have been normalized by the measured cell count in each elutriated fraction . For each protein , the theoretical linear combination of elutriated fraction intensities ( scaled by cell number ) should match the protein intensity measured experimentally in the asynchronous population . Similar factors were calculated between adjacent fractions ( e . g . , F1 vs F2 ) , using cell number and the proportions of cells in each phase , as determined by flow cytometry . These stringent criteria left a subset of the total proteins detected with very high data coverage across the six elutriated cell cycle fractions and high self-consistency in quantitation . Absolute protein abundances were estimated using the iBAQ algorithm , as previously described ( Schwanhausser et al . , 2011 ) . Gene ontology analysis was performed using the DAVID web resource ( Huang da et al . , 2009 ) and STRING ( Jensen et al . , 2009 ) . Predicted transcription factor binding sites were retrieved from MSigDB ( Subramanian et al . , 2005; Matys et al . , 2006 ) . Tissue mRNA expression data were obtained from the Broad Cancer Map , as implemented in MSigDB ( Ramaswamy et al . , 2001; Subramanian et al . , 2005 ) . An arbitrary twofold cutoff was implemented to identify cell cycle varying proteins . Proteins were then clustered using the Ward algorithm into 16 clusters . 15 of these clusters were then re-clustered based on the phase of maximum expression . The final cluster , which had two peaks across the cell cycle fractions , was left unchanged ( i . e . , the G2&M+G1 cluster ) . A similar clustering analysis was performed for phosphopeptide intensities . To carry out isoform analysis , the MaxQuant data were re-analyzed to produce isoform-specific cell cycle profiles . To do this the MS/MS information from the peptide evidences ( evidence . txt ) , in the MaxQuant output , was used to determine the number of unique MS/MS counts in each fraction . These MS/MS counts were then averaged for peptides belonging to the same isoform , providing an isoform specific profile of MS/MS counts across fractions . This process was carried out with the additional quality filters described above , that is removal of contaminant and reverse hits and ensuring isoforms have a minimum of two unique peptide identifications . To highlight potentially interesting isoforms displaying differential behavior , a correlation score was calculated between isoforms of the same protein . Proteins showing a poor correlation between isoforms were used to identify examples of differentially regulated isoforms across the cell cycle fractions . To compare protein and RNA data , protein identifiers were mapped to Ensembl Gene ID . Histone genes were removed from this data analysis , due to their lack of poly ( A ) tails . Absolute protein and mRNA abundances were plotted in DataShop , a data visualization tool developed as part of the PepTracker software suite ( www . PepTracker . com ) . The PepTracker app runs on both Windows and Mac OSX and is freely available for download ( www . PepTracker . com/ds/ ) . Gene expression data sets are provided in multiple forms to facilitate access for a range of end-users . MS raw files can be accessed from the EBI PRIDE database ( accession PXD000678 ) . RNA-Seq raw files will be available from EBI ( accession EGAD00001000736 ) . Peptide evidence data derived from MaxQuant have been deposited to Dryad and can be accessed using this hyperlink: http://dx . doi . org/10 . 5061/dryad . 2r79qL ( Ly et al . , 2014 ) . Protein , phosphoprotein , and RNA identifications and quantitations are available in supplementary tables to this manuscript ( e . g . , Supplementary files 1 , 4 and 6 ) . Outputs from the cell cycle gene expression analysis , cell line meta-analysis , and comparative protein and mRNA analysis are also provided in supplementary tables ( Supplementary files 2 , 3 , and 5 ) . In addition , gene-by-gene visualization of the quality-filtered data set of protein and mRNA expression for over 6000 genes analyzed across the cell cycle is provided in a searchable , online format via the Encyclopedia of Proteome Dynamics ( EPD ) ( http://www . peptracker . com/epd ) ( Larance et al . , 2013 ) . This is a web-based tool , part of the PepTracker platform , which aims to visually communicate and disseminate data from large scale , multi-dimensional proteomic experiments . The EPD is developed using Python and the Django web framework . The EPD uses an Oracle database to store raw data , including the protein and mRNA data from this study . The visualizations depicting protein and RNA data are created using the R programming language and integrated into the web tool via the RPy2 library . | Cells are complex environments: at any one time , thousands of different genes act as molecular templates to produce messenger RNA ( mRNA ) molecules , which themselves are templates used to produce proteins . However , not all genes are active at all times inside all cells: as cells grow and divide as part of the cell division cycle , genes are switched on and off on a regular basis . Similarly , the patterns of mRNA and protein production are different in , say , immune and skin cells . In recent years , the tools available for detecting mRNA molecules and proteins have become more powerful , allowing researchers to move beyond just measuring the total amounts of mRNA and protein in the cell to now measuring individual amounts of specific mRNA and protein molecules encoded by specific genes . However , it has been a challenge to make these measurements at different stages of the cell cycle . Most of the methods used to do this have involved artificially ‘arresting’ the cell cycle , which can lead to side effects that are difficult to account for . Ly et al . have now overcome these problems using a combination of three methods to measure the levels of mRNA and protein molecules associated with over 6000 genes in human cancer cells derived from myeloid leukemia . Exploiting the fact that cells change size during the cell cycle , Ly et al . used a centrifugation technique to separate cells based on their size and , therefore , the stage of the cell cycle they were at , thus avoiding the need to arrest the cell cycle . An approach called RNA-Seq was then employed to measure the levels of the different mRNA molecules in the cells , and a device called a mass spectrometer was used to identify and measure the levels of many different proteins . In addition to being able to follow the level of mRNA and protein production for a large number of genes throughout the cell division cycle , while also obtaining detailed information about how many of the proteins are modified , Ly et al . discovered that—contrary to expectations—low numbers of mRNA molecules were sometimes associated with high numbers of the corresponding protein , and vice versa . This work provides a better understanding of the complex relationship between the levels of an mRNA and its corresponding protein product , and also demonstrates how it may be possible to detect subtle but important differences between cell types and disease states , including different types of cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2014 | A proteomic chronology of gene expression through the cell cycle in human myeloid leukemia cells |
Hair follicle ( HF ) development is orchestrated by coordinated signals from adjacent epithelial and mesenchymal cells . In humans this process only occurs during embryogenesis and viable strategies to induce new HFs in adult skin are lacking . Here , we reveal that activation of Hedgehog ( Hh ) signaling in adjacent epithelial and stromal cells induces new HFs in adult , unwounded dorsal mouse skin . Formation of de novo HFs recapitulated embryonic HF development , and mature follicles produced hair co-occurring with epithelial tumors . In contrast , Hh-pathway activation in epithelial or stromal cells alone resulted in tumor formation or stromal cell condensation respectively , without induction of new HFs . Provocatively , adjacent epithelial-stromal Hh-pathway activation induced de novo HFs also in hairless paw skin , divorced from confounding effects of pre-existing niche signals in haired skin . Altogether , cell-type-specific modulation of a single pathway is sufficient to reactivate embryonic programs in adult tissues , thereby inducing complex epithelial structures even without wounding .
The number and pattern of hair follicles ( HFs ) are specified before birth in humans . In the mouse this is true for most body areas such as back skin ( Alonso and Rosenfield , 2003; Millar , 2002; Paus and Cotsarelis , 1999 ) . HF morphogenesis requires Hedgehog ( Hh ) and Wnt/β-catenin signaling and becomes first morphologically visible at embryonic day 14 . 5 ( E14 . 5 ) in mice ( Chiang et al . , 1999; Gat et al . , 1998; Lo Celso et al . , 2004; St-Jacques et al . , 1998 ) . At this stage , the embryonic hair germ has formed , consisting of an epithelial placode and a dermal condensate , whose epithelial-mesenchymal crosstalk is essential for further HF development ( Hardy , 1992; Schmidt-Ullrich and Paus , 2005 ) . De novo HF formation in adult skin has been observed in combination with wounding in rabbits , mice , and humans ( Breedis , 1954; Ito et al . , 2007; Kligman and Strauss , 1956; Lim et al . , 2018 ) , and in unwounded skin as a response to forced epithelial Wnt/β-catenin signaling in mice ( Gat et al . , 1998; Lo Celso et al . , 2004 ) . Two decades after the initial discovery that Wnt/β-catenin-pathway activation results in ectopic HFs ( Gat et al . , 1998 ) , it is still the only known approach to induce de novo HFs in adult unwounded skin . Accordingly , it remains a major clinical challenge to generate replacement skin with hair , urging the search for new ways to achieve HF formation in adult skin . In Sonic hedgehog knock out ( Shh-/- ) mice , HF morphogenesis does not progress beyond the hair germ stage ( St-Jacques et al . , 1998 ) . Notably , compared to wild type skin , Shh-/- hair germs have normal levels of Wnt/β-catenin signaling but reduced Hh-target activation in both the placode and the dermal condensate ( Chiang et al . , 1999; St-Jacques et al . , 1998 ) . Despite the well-known importance of Hh signaling for HF morphogenesis during embryonic skin development ( Botchkarev and Paus , 2003; Mill et al . , 2003; St-Jacques et al . , 1998 ) and in wound-induced HF formation ( Lim et al . , 2018 ) , little is known about its potential for de novo HF induction in adult unwounded skin . Interestingly , already decades ago it has been observed that basal cell carcinoma ( BCC ) morphologically mimics HF development until the hair germ stage . BCC develops upon supra-physiological Hh-pathway activation in epithelial cells , with the most prevalent mutations in the inhibitory Hh-receptor gene Ptch1 . Like HF formation , Hh-driven BCC is characterized by active Wnt/b-catenin signaling ( Yang et al . , 2008 ) , and early epithelial BCC buds express typical HF-lineage markers even if they don't originate from HFs ( Kasper et al . , 2011; Yang et al . , 2008; Youssef et al . , 2012 ) . Morphologically and molecularly , these early BCC buds are thus very similar to embryonic hair germs with one major difference: BCC buds lack a dermal condensate which serves as a focal point for dermal Hh signaling and is required for HF morphogenesis ( Yang et al . , 2008 ) . Based on these observations , we hypothesized that HF development past the hair germ stage in BCC requires Hh signaling at increased levels in both epithelial and stromal cells . To address this hypothesis , we used mouse models to induce supra-physiological Hh signaling via Ptch1 deletion in the epithelium and stroma , which indeed led to the induction of new HFs in adult unwounded skin .
To test whether activation of epithelial and stromal Hedgehog signaling would result in HF formation , we focused on the touch domes ( TDs ) , which are touch-receptive structures located within the interfollicular epidermis ( IFE ) ( Figure 1A , B ) . We reasoned that in adult unwounded skin , TDs were the most likely places to achieve experimentally induced de novo HFs because TDs are hotspots for BCC formation ( Peterson et al . , 2015 ) , which resembles HF development ( Yang et al . , 2008 ) . We analyzed and compared the TDs of two mouse models with induced supra-physiological Hh-pathway levels in epithelial cells ( Lgr6 mouse model ) , or combined epithelial and stromal cells of the TD ( Gli1 mouse model ) . Specifically , we used Lgr6-EGFP-IRES-CreERT2;R26tdTomato;Ptch1fl/fl mice ( hereafter: Lgr6creERT2;R26Tom;Ptch1fl/fl ) and Gli1-CreERT2;R26tdTomato;Ptch1fl/fl mice ( hereafter: Gli1creERT2;R26Tom;Ptch1fl/fl ) . TD areas were identified by the presence of K8+ Merkel cells , and their palisading epithelial cell morphology . Tamoxifen was administered at 8 weeks of age , resulting in the constitutive activation of Hh signaling via homozygous inactivation of Ptch1 and simultaneous Tomato-tracing of Lgr6- or Gli1-expressing cells ( Figure 1C ) . First , we confirmed that Lgr6 expression , and consequently Tomato-tracing , were in the TD restricted to epithelial cells ( Figure 1D , F ) , and Gli1 expression and Tomato-tracing were present in both epithelial and stromal TD cells ( Figure 1E , G ) . Next , we analyzed the phenotypes of both Lgr6 and Gli1 mouse models 5 weeks post tamoxifen , a sufficiently long time to allow possible de novo HFs to form ( Rendl et al . , 2005 ) . Homozygous Ptch1 inactivation in Lgr6-expressing cells resulted as expected in BCC-like lesions in HFs , IFE and TDs ( Figure 1H , Figure 1—figure supplement 1; Peterson et al . , 2015 ) . Strikingly , however , homozygous Ptch1 inactivation in Gli1-expressing cells resulted in addition to BCC-like lesions in HFs and TDs , in the formation of structures in TDs that resembled de novo HFs ( Figure 1I , Figure 1—figure supplement 3 ) . These structures had the appearance of typical concentric anagen HF layers and a pigmented hair bulb or mature hair shaft ( Figure 1I , Figure 1—video 1 ) . Importantly , by 9 weeks post tamoxifen , these HF-like structures occurred in every single Gli1creERT2;R26Tom;Ptch1fl/fl TD examined ( 27/27 ) but extremely rarely in TDs of Lgr6creERT2;R26Tom;Ptch1fl/fl mice ( 2/25; please also see Discussion ) ( Figure 1J ) . No such HF-like structures were observed in TDs of wild type mice ( with physiological Hh signaling in epithelial and stromal cells ) , upon heterozygous Ptch1 inactivation ( Gli1creERT2;R26Tom;Ptch1fl/wt and Lgr6creERT2;R26Tom;Ptch1fl/wt ) ( Figure 1F , G ) , or in non-tamoxifen controls ( Gli1creERT2;R26Tom;Ptch1fl/fl and Lgr6creERT2;R26Tom;Ptch1fl/fl ) ( Supplementary file 1 ) . Therefore , induction of supra-physiological Hh signaling in epithelial and stromal cells ( Gli1 model ) but not epithelial cells alone ( Lgr6 model ) was sufficient to induce HF-like structures in TDs of adult mouse skin . Next we investigated whether the observed structures were functional HFs , and indeed de novo induced . Thus , we stained the skin of Gli1creERT2;R26Tom;Ptch1fl/fl mice for Keratin 71 ( K71 ) and Keratin 6 ( K6 ) ( Figure 2A ) , which mark specific layers of the anagen HF ( Yang et al . , 2017 ) . These Keratin-staining patterns were very similar to those of hair-cycle stage-matched wild type anagen HFs ( Figure 2B ) . Also , the presence , and specific pattern of hair pigment in these structures were typical for anagen HFs , in the shown picture matching the anagen III hair-cycle stage ( Figure 2A , B ) , further supporting that the observed structures are indeed HFs and actively growing . At 5 weeks post tamoxifen all de novo HFs were in different stages of anagen ( Figures 1I and 2A ) and by 9 weeks post tamoxifen the majority of de novo HFs were in telogen ( Figure 2C , Figure 2—figure supplement 2 ) . This demonstrates that de novo HFs enter the hair cycle after their first anagen ( Paus and Cotsarelis , 1999 ) . To verify that these HFs were de novo induced , thorough examination of the Tomato lineage tracing pattern was of key importance . The Gli1-positive HF and TD populations self-renew independently with a clear Gli1-negative gap in the infundibulum ( Figure 1E; Xiao et al . , 2015 ) . Cell crossover only occurs for example in response to full-thickness wounding or TPA treatment ( Brownell et al . , 2011; Kasper et al . , 2011 ) . Importantly , Ptch1 deletion did not trigger HF cell migration towards the IFE or TD as we never observed Tomato-tracing spanning from a pre-existing HF – via the infundibulum – to the IFE or TD in any control or Gli1creERT2;R26Tom;Ptch1fl/fl skin ( Figure 1—figure supplement 3; Figure 2—figure supplement 1 ) . A non-traced infundibulum , leaving a tracing gap between the HF and the IFE/TD , is thus characteristic for pre-existing HFs . In contrast , all de novo HFs displayed continuous Tomato-tracing from the de novo HF to its originating TD , including the infundibulum and all anagen HF lineages ( Figures 1I and 2A , Figure 1—video 1 and Figure 2—figure supplement 1 ) . As such a continuous HF-to-TD-tracing pattern can only result from newly formed HFs originating from traced TD-epithelial cells , we unequivocally demonstrated that these HFs were new . An additional characteristic of de novo HFs was that their hair shafts were considerably shorter and thinner than hair shafts from both pre-existing HFs in the same mice ( Gli1creERT2;R26Tom;Ptch1fl/fl ) and HFs in control mice with wild type phenotype ( Gli1creERT2;R26Tom;Ptch1fl/wt ) ( Figure 2C and Figure 2—source data 1 ) . In conclusion , the continuous lineage-tracing from HF to TD , and the characteristic hair shaft measurements , demonstrated that combined epithelial and stromal Hh-pathway activation in the Gli1 mouse model via homozygous Ptch1 inactivation resulted in de novo HFs , within TDs . Next we characterized the stages of de novo HF development in TDs equivalent to the embryonic HF developmental stages of HF placode , hair germ , hair peg and mature follicle ( Rendl et al . , 2005 ) . We analyzed Gli1creERT2;R26Tom;Ptch1fl/fl and control skin , tamoxifen treated at 8 weeks of age , and collected samples 10 , 17 , 25 , 27 , 29 , 33 , 35 and 36 days after tamoxifen treatment . These sampling time points enabled us to map the entire stereotypical time course of de novo HF formation ( Figure 3A–D , Supplementary file 1 ) . In some TDs , the HF-placode stage could be already detected 10 days post tamoxifen administration , and was accompanied by the appearance of a dermal condensate ( Figure 3A ) . The dermal condensate and dermal papillae were delineated by denser Tomato-tracing of stromal cells compared to adjacent traced K5+ epithelial cells . Note the lack of continuous pre-existing HF-to-TD tracing at this early developmental stage ( Figure 3A , E ) , again supporting that the fully traced HFs ( Figures 1I , 2A and 3B–D ) were newly formed and induced from traced TD cells . The hair germ and hair peg stages were mostly detected between 27 and 29 days post tamoxifen administration ( Figure 3B , C ) , and mature HFs , with a clearly visible developing hair shaft , emerged mainly at or after 33 days post tamoxifen administration ( Figure 3D , D’ , F ) . Interestingly , almost all dermal condensates and dermal papillae of de novo forming HFs were fully Tomato traced ( Figure 3A–D , F ) , suggesting that either continuous high levels of Hh signaling in stromal cells ( Ptch1fl/fl ) are necessary for all stages of de novo HF development , or that stromal cells with constitutive Hh-pathway activation outcompete dermal condensate and dermal papilla cells with lower Hh-signaling levels . Importantly , Syndecan-1 ( SDC1 ) expression that marks early dermal condensates in wild type embryonic skin ( Figure 4A; Richardson et al . , 2009 ) , could already be detected 10 days post tamoxifen in dermal condensates of Gli1creERT2;R26Tom;Ptch1fl/fl skin ( Figure 4B ) , and was fully established in the dermal papilla at the hair germ stage ( Figure 4C ) . Finally , to demonstrate that these de novo HFs do indeed have active Hh signaling , we stained for Gli1 mRNA expression; as a reporter of canonical Hh-pathway activity . The placodes ( 10 days post tamoxifen ) as well as intermediate and mature developmental stages ( 5 weeks post tamoxifen ) expressed Gli1 mRNA and hence have active Hh/Gli signaling ( Figure 4F–G and Figure 4—figure supplement 1C–D ) ; which is in line with normal embryonic HF development ( Figure 4D and Figure 4—figure supplement 1A; Chiang et al . , 1999; St-Jacques et al . , 1998 ) . Wild type TD epithelial and stromal cells expressed Gli1 mRNA as expected ( Figure 4E and Figure 4—figure supplement 1B ) . This Gli1 RNA-FISH combined with antibody staining for Tomato-lineage tracing also confirmed that Tomato-tracing ( cells with Ptch1 deletion ) and Gli1 expression were highly correlated , as expected ( Figure 4F–G and Figure 4—figure supplement 1C–D ) . We conclude that de novo HF formation in TDs of adult Gli1creERT2;R26Tom;Ptch1fl/fl skin recapitulates the hallmarks of HF development during embryogenesis . De novo HF formation in the Gli1creERT2; R26Tom; Ptch1fl/fl mouse model is accompanied with BCC growth; that is BCC-like lesions appear in pre-existing HF as well as in TDs ( Figure 1—figure supplement 3 ) . In TDs , the clearly identifiable de novo HFs develop alongside epithelial tumor growth which is characterized by palisading cells and lack of HF-like structures . It has been shown previously that BCC-like lesions in dorsal skin dramatically shrink within seven days upon vismodegib treatment ( Eberl et al . , 2018 ) . Vismodegib is a Hh-pathway inhibitor acting at the level of Smoothened ( Smo ) , and using the optimized treatment scheme from Eberl et al . ( 2018 ) we tested whether established de novo HFs would persist or would diminish as the BCC-tumor-growth area does . We treated Gli1creERT2;R26Tom;Ptch1fl/fl mice with tamoxifen at 8 weeks . Five to seven weeks after tamoxifen treatment when de novo HFs were clearly established in TDs , we took a dorsal biopsy prior to vismodegib treatment ( untreated biopsy ) , and then treated the mice daily with vismodegib ( 50 mg/kg body weight i . p . ) for seven days ( Figure 5A ) . Reassuringly , in pre-existing HFs we found considerable reduction of tumor size ( Figure 5B ) as well as absence of Ki67 staining in Tomato-traced areas when comparing 7 day vismodegib treated samples with the untreated biopsies of the same mice ( Figure 5C ) . This reduction in tumor size demonstrated that vismodegib treatment worked as expected . In TDs , the tumor areas also dramatically diminished in size and Ki67 staining , however the de novo HFs persisted ( Figure 5D and Figure 5—figure supplement 1 ) . In conclusion , these experiments confirmed that de novo HFs indeed represent HFs that are independent of tumor structures as they persist upon vismodegib treatment when the surrounding BCC-tumor-growth areas are nearly gone . De novo HFs were induced by strong activation of Hh signaling ( Ptch1fl/fl ) in epithelial and adjacent stromal cells and persisted upon vismodegib treatment when the surrounding BCC-growth areas were nearly gone . As BCC growth merely depends on epithelial Hh-pathway activation , it was tempting to test whether homozygous inactivation of Ptch1 exclusively in stromal cells would be sufficient to induce de novo HFs without tumor development . To that end , we generated the Col1a2 mouse model ( Col1a2-CreER;R26tdTomato;Ptch1fl/fl , hereafter: Col1a2creER;R26Tom;Ptch1fl/fl ) ( Figure 6A , B ) , which drives supra-physiological Hh signaling in the stromal compartment only , via the collagen type I alpha two chain promoter . Non-tamoxifen controls ( Col1a2creER;R26Tom;Ptch1fl/fl mice ) showed some tracing in the skin stroma , which however did not result in an adverse skin phenotype except for earlier anagen entry ( Figure 6—figure supplement 1 ) . Administration of tamoxifen at 8 weeks of age resulted in substantial Tomato-tracing that was restricted to the stromal skin compartment ( Figure 6—figure supplement 2 ) , and importantly , the stromal cells of the TD were also traced ( Figure 6D ) . Homozygous Ptch1 inactivation in Col1a2-expressing cells resulted in increased stromal cell density in TDs ( Figure 6E ) , but did not result in de novo HF formation ( Figure 6C , E ) , nor did stromal cells stain positive for SDC1 even 9 weeks after tamoxifen treatment ( Figure 6F ) . We also detected fully traced and highly condensed dermal clusters of cells ( resembling dermal condensates ) in TD-adjacent infundibula and underneath the regular IFE ( Figure 6G , Figure 6—figure supplement 2B ) , which did not result in de novo HF induction and the dermal cell condensations were entirely negative for SDC1 expression ( Figure 6G ) . We conclude that stromal activation of Hh signaling ( Ptch1fl/fl ) leads to increased stromal cell density and formation of cell condensates , however it is not sufficient to induce HF neogenesis in TDs nor elsewhere in skin without adjacent epithelial Hh-pathway activation . To directly compare all three different mouse models ( Lgr6creERT2/ , Gli1creERT2/ , Col1a2creER;R26Tom;Ptch1fl/fl ) in their competence to initiate de novo HFs , we analyzed TDs 10 days post tamoxifen administration , the time point when epithelial proliferation became evident via , for example , increased BrdU incorporation in hair-forming TDs before appearance of morphological hair germ formation ( Figure 7 , Figure 7—figure supplement 1 ) . We investigated the stromal TD compartment using the alkaline phosphatase ( AP ) assay and SDC1 staining , which are both characteristic for HF-inducing dermal condensates ( Ito et al . , 2007; Richardson et al . , 2009 ) . Positive staining for both alkaline phosphatase ( ALPL ) and SDC1 expression was observed in the Gli1creERT2;R26Tom;Ptch1fl/fl mice , but not in Gli1creERT2;R26Tom;Ptch1fl/wt , Lgr6creERT2;R26Tom;Ptch1fl/fl , or Col1a2creER;R26Tom;Ptch1fl/fl mice ( Figure 7A , B ) . The epithelial TD compartment could not be stained for a comparable marker of early HF induction , as early BCC buds express typical HF-lineage markers . Indeed , we and others have not found a single mRNA/protein stain that would distinguish HF epithelial placode from BCC formation ( Kasper et al . , 2011; Yang et al . , 2008; Youssef et al . , 2012 ) . Altogether , staining for early signs of HF formation demonstrated that only the Gli1 ( Ptch1fl/fl ) model bears TDs that are competent for de novo HF formation . We established that de novo HF formation required close apposition of epithelial and stromal Hh signaling ( Ptch1fl/fl ) , and furthermore how to identify de novo HFs based on their morphology and continuous lineage tracing . We next asked whether de novo HFs could also form from non-TD areas with comparable adjacent epithelial-stromal Hh-pathway activation . In addition to TDs , the HF isthmus also harbors adjacent epithelial and stromal Gli1-Tomato traced cells; the latter evident through Tomato/PDGFRb co-staining of stromal cells ( Figure 8A , B ) . Thus , we re-examined the HF isthmus areas of Gli1creERT2;R26Tom;Ptch1fl/fl skin for potential de novo HFs . Indeed , at low frequency , we observed de novo HFs in the isthmus of pre-existing HFs that were most likely newly formed ( Figure 8C , D ) . Although it was not possible to unequivocally determine de novo formation through continuous Tomato-tracing of the infundibulum ( as these de novo HFs seem to merge directly into the isthmus area of pre-existing HFs ) , based on their morphology , positioning and the co-occurrence of four hair shafts instead of normally three ( in dorsal skin of 17 week-old mice ) , these HFs likely formed newly from the isthmus of pre-existing HFs ( Figure 8C ) . Such HFs have not been observed in phenotypically normal control skin ( Gli1creERT2;R26Tom;Ptch1fl/wt and Lgr6creERT2;R26Tom;Ptch1fl/wt ) or in skin with HF tumors ( Lgr6creERT2;R26Tom;Ptch1fl/fl ) ( Figure 8D ) . Reassuringly , in the HF isthmus Lgr6 expression is restricted to epithelial cells only , whereas Gli1 is expressed in epithelial and adjacent stromal cells ( Füllgrabe et al . , 2015 ) ; suggesting that adjacent epithelial-stromal Hh signaling in areas outside of the TD may form de novo HFs . The mouse hindpaw ( plantar ) epidermis is a skin region devoid of hair follicles and sweat glands , and is therefore ideal to test whether epithelial and stromal Hh-pathway activation can induce de novo HFs divorced from any confounding effects of nearby HF- or TD-niche signals ( Figure 9A ) . When we probed for Gli1 expression in the plantar skin using Gli1LacZ reporter mice , we consistently found small Gli1-BGAL expressing clusters of epithelial and adjacent stromal cells in the plantar skin ( Figure 9B , Figure 9—figure supplement 1A ) , which we confirmed with short-term lineage tracing in Gli1creERT2;R26Tom mice ( tamoxifen at P8w; sample collection 7 days later ) ( Figure 9C , Figure 9—figure supplement 1B ) . Therefore , the plantar skin was a suitable area for studying if de novo HFs can form in the Gli1 mouse model upon homozygous Ptch1 inactivation . Testing for de novo HF induction , we analyzed the hindpaws of Gli1creERT2;R26Tom;Ptch1fl/fl mice and control littermates ( Gli1creERT2;R26Tom;Ptch1fl/wt ) using the same treatment scheme and sample collection times as for dorsal skin ( tamoxifen at P8w; sample collection 5 and 9 weeks post tamoxifen; additional time points see Supplementary file 1 ) . Indeed , we found numerous de novo HFs with hair shafts in the normally hairless region . These HFs were fully Tomato-traced and showed normal inner-layer differentiation based on morphology and K6 staining ( Figure 9D–E , Figure 9—figure supplement 2 ) . Taken together , the analysis of dorsal HF-isthmus as well as hairless paw skin demonstrated that combined epithelial and stromal Hh-pathway activation can induce de novo HFs independently of TD niches .
Hitherto , hair follicle neogenesis in adult skin had only been observed under exceptional circumstances , such as upon repair of large wounds ( Breedis , 1954; Ito et al . , 2007 ) . A recent study found that it is possible to induce HFs even in small wounds upon supra-physiological Hh-pathway activation in the wound stroma ( Lim et al . , 2018 ) . This report and our present study recognize modulation of Hh signaling as a new approach to induce de novo HFs in adult skin , and define activation in the stromal skin compartment as critical . As wounding of skin initiates a major reorganization of the epithelial and mesenchymal tissue including the activation and differentiation of a large number of cell types ( Arwert et al . , 2012; Joost et al . , 2018; Schäfer and Werner , 2008 ) , and may even provide an embryonic-like environment ( Wang et al . , 2015 ) , the precise ( e . g . minimal ) molecular signals that are required for HF induction in adult skin remain elusive . Here we revealed that in unwounded skin , experimentally elevated Hh signaling in epithelial and adjacent stromal cells was sufficient to induce de novo HFs , extending our understanding of how Hh signaling can be modulated to induce HFs in adult skin . To achieve efficient de novo HF induction in unwounded skin , supra-physiological Hh signaling in both compartments , the epithelium and stroma , was necessary . Normal TD maintenance also requires active and balanced Hh signaling in adjacent epithelial and stromal cells ( Figure 4E; Xiao et al . , 2015 ) . Increased Hh-signaling levels in TD epithelial cells result in BCC-like tumors , even though stromal cells have active ( albeit physiological ) Hh signaling . Only in extremely rare cases ( twice ) did we detect a de novo HF-like structure in TDs of Lgr6creERT2;R26Tom;Ptch1fl/fl mice based on morphology ( as lineage tracing in these mice cannot provide information on de novo HF formation; Figure 1H and Figure 1—figure supplement 2 ) . Importantly therefore , to effectively induce de novo HFs in TDs , high levels of Hh/Gli signaling in both compartments were necessary ( Gli1creERT2;R26Tom;Ptch1fl/fl ) . This requirement of Hh-signal activation at precise levels and in the right compartments is in agreement with a recent study demonstrating that β-catenin-induced de novo HF formation was not only dependent on stromal Hh signaling , but also required two intact Smo alleles ( for a maximal Hh-pathway activation ) to enable efficient de novo HF induction ( Lichtenberger et al . , 2016 ) . It has been shown more than twenty years ago that the activation of epithelial β-catenin in mouse skin can induce new HFs ( Gat et al . , 1998; Lo Celso et al . , 2004 ) , and more recently that activation of epithelial Wnt/β-catenin signaling increases de novo HF formation within wounds ( Ito et al . , 2007 ) . However , Wnt signaling has to be blocked in dermal fibroblasts to allow de novo HF induction during wound regeneration ( Rognoni et al . , 2016 ) . It is known that early stage BCCs resemble early stages of HF development and both are dependent on Wnt- and Hh-pathway activation , with the major morphological difference that BCC lacks a dermal condensate ( Yang et al . , 2008 ) . Learning from abrogated embryonic HF development ( St-Jacques et al . , 1998 ) led us to hypothesize that simultaneously activating supra-physiological Hh signaling in the stroma underneath developing BCC may enable de novo HFs . Indeed , by coordinating the activation of Hh/Gli signaling ( cell type specific and high levels ) we were able to induce de novo HFs by ‘redirecting’ some of the BCC buds to HF formation without the need of wounding . Nevertheless , this induction occurred in the presence of oncogenic signal activation ( i . e . presence of a tumor environment or tumor-like cellular status of Ptch1fl/fl HF-inducing epithelial cells ) which may to some extent mimic a wounding situation ( Dvorak , 1986 ) . The molecular and cellular similarities of tumorigenesis and wound healing are still unfolding , yet whenever de novo HFs were to be found , either oncogenic signaling or a wound environment was involved . This supports the long-standing recognition of the similarity between tumor and wound healing signaling ( tumors as ‘wounds that do not heal’ ) – and raises the key question of what exactly is the relationship between tumorigenesis and signals inducing de novo HF formation ? It may indeed be the case that in order to overcome inhibitory signals , de novo HF morphogenesis in adult skin requires such major activating signals provided by oncogenesis or wounding . Interestingly however , previous literature suggests that only initial HF placode and/or dermal condensate formation may require such strong signals whereas progression to a mature HF does not require continued tumorigenic or wound signaling ( Brown et al . , 2017; Ito et al . , 2007; Lo Celso et al . , 2004; Silva-Vargas et al . , 2005 ) . For example , HF tumors require continuous Wnt/β-catenin signaling , whereas transient activation of this pathway is sufficient to induce de novo HFs in adult mouse epidermis ( Lo Celso et al . , 2004 ) . More recently , intra vital imaging from Wnt/β-catenin induced tumor outgrowths demonstrated that non-mutant cells , remaining from regressed outgrowths , could develop into new functional HFs . Most interestingly , tumor outgrowth depended on the presence of mutated cells , however the new appendages were formed from wild type cells facilitated by their ( altered ) niche environment ( Brown et al . , 2017 ) . Taken together , these are promising examples that de novo HF induction without accompanied tumor growth in unwounded adult skin may in principle be possible , if the right signals at the right time and restricted period , and in the right compartments were provided . Here , we spatiotemporally defined such productive and specific molecular signals . Lastly , and importantly , we exploited the hairless paw plantar skin to examine de novo HF morphogenesis in the absence of confounding signals from pre-existing HFs or TDs . Strikingly , we observed numerous de novo HFs throughout this nominally hairless skin in Gli1creERT2;R26Tom;Ptch1fl/fl mice . Crucially , nearly all these de novo HFs developed without attendant BCC-like lesions suggesting that de novo HF morphogenesis may indeed be successfully initiated without a tumor microenvironment; while the vismodegib experiment suggests persistence of such structures when the tumor microenvironment regresses in dorsal skin . Examining these two divergent tissues in molecular detail , one permissive ( dorsal ) and one suppressive ( paw ) to dual BCC and HF induction , could therefore be a next step of unraveling the complexity of how these heterogeneous signals interact . In sum , molecular strategies for the induction of complex epithelial structures in the adult remain a major challenge in regenerative medicine . Our study demonstrates that cell-type specific modulation of a single pathway was sufficient to induce complex epithelial structures in the adult body , a discovery aiding our understanding of adult tissue biology and regenerative medicine .
Gli1creERT2;R26Tom;Ptch1fl/fl mice and control littermates , and Lgr6creERT2;R26Tom;Ptch1fl/fl mice and control littermates were treated with tamoxifen at second telogen ( mice aged 8 weeks , 6 mg tamoxifen i . p . in corn oil , 20 mg/mL ) . Dorsal samples were taken at different time points after tamoxifen treatment ( as indicated in the text ) , and were obtained via 3–4 mm full thickness biopsies or by sacrificing the animal ( n ≥ 3 mice for each genotype; for details please see Supplementary file 1 ) . Some of the Col1a2creER;R26Tom;Ptch1fl/fl mice , when receiving the same treatment as the Gli1 and Lgr6 models as described above , developed within a few weeks a severe intestinal phenotype , precluding comparative skin analysis at 5 and 9 weeks post tamoxifen administration . Thus Col1a2creER;R26Tom;Ptch1fl/fl mice and control littermates were treated at second telogen topically or with a reduced amount of tamoxifen i . p . ( mice aged 8 weeks , 2 × 0 . 75 mg 4-OH tamoxifen topically or 3 mg tamoxifen i . p . in corn oil , 20 mg/mL; Supplementary file 1 ) , and some were treated at first telogen ( mice aged 3 weeks , 1 mg tamoxifen i . p . in corn oil , 20 mg/mL ) . Samples were harvested at different time points after tamoxifen treatment ( as indicated in the text ) . More details of analyzed mice , including control mouse experiments , treatments , and phenotypes , are given in Supplementary file 1 . For BrdU incorporation , mice were injected i . p . with BrdU ( Sigma ) solution ( 10 mg/ml ) . A dose of 0 . 1 mg/g body weight was administered 2 hr prior to sacrifice . For vismodegib treatment , the Gli1creERT2;R26Tom;Ptch1fl/fl mice and control littermates Gli1creERT2;R26Tom;Ptch1fl/wt were treated with tamoxifen at 8 weeks of age . Five to seven weeks after tamoxifen treatment , when de novo HFs were clearly established in TDs , a dorsal biopsy prior to vismodegib treatment was taken . From then , the vismodegib was given daily ( 50 mg/kg body weight i . p . ) for a week and dorsal samples were collected for further analysis ( n ≥ 3 mice ) . Embryos were collected at E15 . 5 and E16 . 5 , and wild-type control samples for dorsal tissue were collected at postnatal day 27 and week 9 . All mouse experiments were performed in accordance to Swedish legislation and approved by the Stockholm South or Linköping Animal Ethics Committees . All antibodies , β-Galactosidase , alkaline phosphatase ( AP ) , and RNA-FISH stainings were performed on either mouse dorsal skin or hindpaw samples as described below ( 1-5 ) . For nuclear stains , TO-PRO-3 , Hoechst 33342 or DAPI ( all from Invitrogen ) were used in the different applications below ( 1-4 ) . ( 1 ) Immunofluorescence ( IF ) staining on formalin-fixed , paraffin-embedded ( FFPE ) sections . After de-waxing and antigen retrieval in 10 mM citrate buffer or Diva Decloaker ( Biocare Medical ) for approximately 20 min in a pressure cooker ( 2100 Retriever ) , the sections were blocked with serum and then incubated with primary antibodies specific for K5 ( rabbit 1:1000 ) , Syndecan-1 ( 1:500 ) . Secondary antibodies were Alexa Fluor Dyes ( Invitrogen , 1:500 ) . Used in Figure 4A . ( 2 ) Horizontal whole mount ( HWM ) IF staining . Samples of dorsal skin were fixed in 4% PFA for 20 min and mounted in OCT embedding medium ( Histolab ) . Subsequently , 60–150 μm sections were cut with a cryostat , blocked with PB buffer ( 0 . 1% fish skin gelatin , 0 . 5% Triton X-100% and 0 . 5% skimmed milk powder in PBS ) and stained as described previously ( Driskell et al . , 2009 ) . The nuclear stains were applied at the same time as secondary antibodies . Primary antibodies used: K8 ( 1:1000 ) , K5 ( rabbit 1:1000 , guinea pig 1:50 ) , K6 ( 1:2000 ) , K71 ( 1:100 ) , Syndecan-1 ( 1:500 ) , GFP ( 1:500 ) , BrdU ( 1:400 ) , RFP ( 1:100 ) , Ki67 ( 1:200 ) , CD140b ( 1:100 ) . Secondary antibodies used: Alexa Fluor Dyes 488 , 546 , 647 or 680 ( Invitrogen , 1:500 ) . Used in: Figure 1D , F–I; Figure 1—figure supplements 1 , 2 and 3; Figure 2A , C; Figure 2—figure supplements 1 and 2; Figure 3A–F; Figure 4B , C; Figure 5C , D; Figure 5—figure supplement 1; Figure 6D–G; Figure 6—figure supplements 1 and 2; Figure 7A , B; Figure 7—figure supplement 1; Figure 8A , B; Figure 9C , E; Figure 9—figure supplements 1B and 2A–B . ( 3 ) RNA Fluorescent in situ hybridization ( RNA-FISH ) . RNA-FISH for Gli1 was performed using the RNAscope Multiplex Fluorescent Detection Kit v2 according to manufacturer’s instructions using TSA with Fluorescein on 4-10 μm FFPE sections . All sections were co-stained with anti-RFP ( 1:100 ) and anti-K5 ( guinea pig 1:200 ) antibodies and counterstained with DAPI ( 1 μg/mL ) . Each co-staining was performed on two different mice for each time point , with the exception of only one mouse for the T10d time point . Used in: Figure 4D–G; Figure 4—figure supplement 1 . ( 4 ) Alkaline phosphatase ( AP ) enzymatic assay . HWM skin samples were cut into 60–100 μm sections , fixed in acetone for 1 hr at 4°C , pre-incubated with NTMT solution ( 100 mM NaCl , 100 mM Tris-Cl pH9 . 5 , 50 mM MgCl2 , 0 . 1% Tween-20 ) at RT for 10 min and stained in NBT/BCIP solution ( 20 μL in 1 ml NTMT solution ) for 1–3 min at 37°C . The reaction was stopped with 20 mM EDTA in PBS . The AP-stained tissue subsequently underwent HWM IF staining . Used in: Figure 7A . ( 5 ) LacZ ( β-Galactosidase ) staining . Freshly obtained skin tissue was fixed ( 4% paraformaldehyde in PBS ) for 30 min at RT . Tissues were washed for 15 min with rinse buffer ( 2 mM MgCl2 , 0 . 01% Nonidet P-40 in PBS ) . Subsequently , the β-galactosidase substrate solution ( 1 mg/mL X-Gal , 5 mM K3Fe ( CN ) 6 , 5 mM K4Fe ( CN ) 6·3H2O in rinse buffer ) was added and the tissues were incubated for 18 hr at 37°C in the dark . The substrate was removed , and the tissues were washed twice in PBS for 10 min and kept in 70% ethanol until embedding ( maximum 48 hr ) . The stained tissues were processed into paraffin blocks according to standard procedures . Tissue sections ( 4 μm ) were prepared and counterstained with eosin or H&E . Used in Figure 1E; Figure 9B; Figure 9—figure supplement 1A . Imaging was performed using a Leica ( color bright-field images ) , LSM710-NLO confocal microscope ( Zeiss ) or Nikon A1R confocal microscope . Image analysis was performed using NIS-Elements software ( Nikon ) , Zen 2009 software ( Zeiss ) , or ImageJ , and images were occasionally optimized for brightness , contrast , and color balance . RNA-FISH images are presented as maximum intensity projections covering 6 μm of depth . Measurements of de novo and pre-existing Zig-zag HFs were performed in samples from Gli1creERT2;R26Tom;Ptch1fl/fl and Gli1creERT2;R26Tom;Ptch1fl/wt mice taken 9 weeks after tamoxifen treatment . Image J was used to analyze the images . ( 1 ) Quantification of de novo HFs in touch domes . Touch domes were identified via Tomato-tracing and the presence of Merkel cells ( K8+ ) . Subsequently , all HFs and hair shafts were visualized with bright field images . Only telogen stage hairs that were fully visible from the bulge to the IFE opening were selected and used to measure their length ( i . e . the distance from the bottom/hair club of the telogen hair shaft to the opening of the follicle in the IFE ) , and width ( which was measured where the hair shaft meets the IFE ) ; see Figure 2C . In total , de novo HFs were observed in all of the 27 analyzed touch domes ( on average 2 . 6 de novo HFs per touch dome ) in three different mice , and 34 hair shafts from 20 touch domes qualified for measurements . As controls , pre-existing Zig-zag hairs of Gli1creERT2;R26Tom;Ptch1fl/fl ( i . e . the same samples as for de novo HF measurements; n = 3 mice; 314 hairs in total ) and Zig-zag hairs of Gli1creERT2;R26Tom;Ptch1fl/wt control mice lacking BCC and de novo HFs ( n = 3 mice; 437 hairs in total ) were analyzed in the same way ( Figure 2—source data 1 ) . To statistically estimate whether the putative de novo HFs could merely represent small Zig-zag hairs , a bootstrapping approach was used: a random sample corresponding to the number of observed de novo HFs ( n = 34 ) was taken from the distribution of observed Zig-zag hairs ( n = 751 ) for 10 . 000 . 000 times , and the probability that this random sample is equal to or smaller in length and width than the de novo HFs was subsequently calculated . ( 2 ) Quantification of de novo hairs in isthmus areas of pre-existing HFs . Quantification of de novo hairs in the isthmus area of HFs ( note that we were not able to tell Zig-zag/Awl/Au HFs apart and collectively call them Zig-zag ) was performed in samples from Gli1creERT2;R26Tom;Ptch1fl/fl , Gli1creERT2;R26Tom;Ptch1fl/wt , Lgr6creERT2;R26Tom;Ptch1fl/fl and Lgr6creERT2;R26Tom;Ptch1fl/wt mice ( n = 3 for each genotype ) taken 9 weeks after tamoxifen treatment . De novo hairs were identified by their relatively smaller size , growing pattern and/or there being more than three hairs in a pre-existing HF . De novo hairs growing from the isthmus area were only observed in the Gli1creERT2;R26Tom;Ptch1fl/fl mice with a total of 16 identified de novo hairs in 668 Zig-zag HFs . Gli1creERT2;R26Tom;Ptch1fl/wt ( 335 HFs ) , Lgr6creERT2;R26Tom;Ptch1fl/fl ( 207 HFs ) and Lgr6creERT2;R26Tom;Ptch1fl/wt ( 83 HFs ) mice showed no de novo hairs from the isthmus area . | We are born with all the hair follicles that we will ever have in our life . These structures are maintained by different types of cells ( such as keratinocytes and fibroblasts ) that work together to create hair . Follicles form in the embryo thanks to complex molecular signals , which include a molecular cascade known as the Hedgehog signaling pathway . After birth however , these molecular signals are shut down to avoid conflicting messages – inappropriate activation of Hedgehog signaling in adult skin , for instance , leads to tumors . This means that our skin loses the ability to make new hair follicles , and if skin is severely damaged it cannot regrow hair or produce the associated sebaceous glands that keep skin moisturized . Being able to create new hair follicles in adult skin would be both functionally and aesthetically beneficial for patients in need , for example , burn victims . Overall , it would also help to understand if and how it is possible to reactivate developmental programs after birth . To investigate this question , Sun , Are et al . triggered Hedgehog signaling in the skin cells of genetically modified mice; this was done either in keratinocytes , in fibroblasts , or in both types of cells . The experiments showed that Hedgehog signaling could produce new hair follicles , but only when activated in keratinocytes and fibroblasts together . The process took several weeks , mirrored normal hair follicle development and resulted in new hair shafts . The follicles grew on both the back of mice , where hair normally occurs , and even in paw areas that are usually hairless . Not unexpectedly the new hair follicles were accompanied with skin tumors . But , promisingly , treatment with Hedgehog-pathway inhibitor Vismodegib restricted tumor growth while keeping the new follicles intact . This suggests that future work on improving “when and where” Hedgehog signaling is activated may allow the formation of new follicles in adult skin with fewer adverse effects . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"developmental",
"biology"
] | 2020 | Coordinated hedgehog signaling induces new hair follicles in adult skin |
Organisms possessing genetic codes with unassigned codons raise the question of how cellular machinery resolves such codons and how this could impact horizontal gene transfer . Here , we use a genomically recoded Escherichia coli to examine how organisms address translation at unassigned UAG codons , which obstruct propagation of UAG-containing viruses and plasmids . Using mass spectrometry , we show that recoded organisms resolve translation at unassigned UAG codons via near-cognate suppression , dramatic frameshifting from at least −3 to +19 nucleotides , and rescue by ssrA-encoded tmRNA , ArfA , and ArfB . We then demonstrate that deleting tmRNA restores expression of UAG-ending proteins and propagation of UAG-containing viruses and plasmids in the recoded strain , indicating that tmRNA rescue and nascent peptide degradation is the cause of impaired virus and plasmid propagation . The ubiquity of tmRNA homologs suggests that genomic recoding is a promising path for impairing horizontal gene transfer and conferring genetic isolation in diverse organisms .
The standard genetic code allows faithful translation of proteins across nearly all living organisms and enables horizontally transferred genetic elements ( HTGEs ) , such as conjugative plasmids and viruses , to exploit a host’s translational machinery ( Krakauer and Jansen , 2002 ) . Since naturally occurring exceptions to the standard genetic code exist ( Ambrogelly et al . , 2007; Knight et al . , 2001 ) , researchers have hypothesized that such alternative genetic codes might arise to escape viral predation ( Shackelton and Holmes , 2008 ) . Recent research supports this hypothesis , with modification to codon usage or the genetic code reducing the ability of viruses and conjugative plasmids to exploit their hosts ( Coleman et al . , 2008; Lajoie et al . , 2013b; Ma and Isaacs , 2016 ) . Given the medical , technological , and evolutionary importance of HTGE-mediated horizontal gene transfer ( HGT ) ( Davies , 1994; Gogarten and Townsend , 2005; Moe-Behrens et al . , 2013; Ochman et al . , 2000 ) , understanding the molecular basis for how alternative genetic codes impede HTGEs is vital . At the molecular level , an alternative genetic code arises from reassignment of one or more codons in the genetic code , which stems from a change in the ability of an aminoacyl-tRNA or release factor ( RF ) to recognize codon ( s ) during translation . One possible alteration of the genetic code is the loss of a codon assignment through the deletion or modification of an aminoacyl-tRNA or release factor , removing the cell’s ability to decode that codon ( Figure 1A ) . Such unassigned codons are found in alternative genetic codes in nature ( Knight et al . , 2001 ) and have been engineered into genomically recoded organisms ( GROs ) derived from Escherichia coli ( Isaacs et al . , 2011; Lajoie et al . , 2013b ) . We recently demonstrated that a GRO with an unassigned UAG codon ( i . e . lacking all instances of the UAG codon and release factor 1 , RF1 ) impaired the propagation of HTGEs carrying UAG-ending genes , illustrating that alternative genetic codes can obstruct HGT ( Ma and Isaacs , 2016 ) and establishing the GRO as an ideal model to study the molecular mechanisms that act at unassigned codons to impair HTGEs . Encountering an unassigned codon during translation leads to ribosomal stalling , and without resolution , to cell death ( Keiler and Feaga , 2014 ) . However , the survival of organisms engineered to lack RF1 but retaining some UAG codons in their protein-coding sequences ( Heinemann et al . , 2012; Mukai et al . , 2010 ) and the ability of GROs to resist exploitation by and continue growth in the presence of HTGEs ( Ma and Isaacs , 2016 ) indicates that E . coli can resolve translation at unassigned UAG codons . We hypothesize that three mechanisms could resolve translation at prokaryotic ribosomes encountering these unassigned codons , each resulting in peptides with different C-terminal sequences ( Figure 1B ) : ( 1 ) suppression of the codon by a near-cognate or mutated tRNA ( e . g . amber suppressor ) and continued translation , ( 2 ) frameshifting of bases along the mRNA transcript into a new reading frame and continued translation , or ( 3 ) stalling that elicits one of three ribosomal rescue pathways ( tmRNA-SmpB , ArfA , or ArfB ) in the cell ( Keiler , 2015 ) . The tmRNA-SmpB system acts as the primary rescue mechanism in prokaryotes , resolving ribosomal stalling that arises from the translation of mRNAs lacking a stop codon due to mRNA degradation , frameshifting , and stop codon read-through ( Keiler , 2015 ) . tmRNA-SmpB can also rescue ribosomes stalled on intact mRNAs for structural reasons ( Cruz-Vera et al . , 2011; Keiler , 2015; Li et al . , 2012 ) . The ssrA-encoded tmRNA associates with SmpB to form the tmRNA-SmpB complex , which adds a C-terminal degradation tag to peptides on stalled ribosomes ( Tu et al . , 1995 ) . ArfA and ArfB , the secondary ribosomal rescue systems , alleviate stalling and release the stalled ribosome’s nascent peptide without modification ( Chadani et al . , 2012; Shimizu , 2012 ) . tmRNA , ArfA , and ArfB all act on nonstop ribosomal complexes , which are stalled ribosomes that have reached the 3’ end of an mRNA because of stop-codon readthrough or because of the loss of a stop codon due to 3’ exonuclease degradation ( Keiler , 2015 ) . A possible fourth outcome identified from in vitro studies is loss of translational fidelity after the ribosome encounters rare or unassigned codons ( Gingold and Pilpel , 2011 ) , followed by untemplated termination by release factor 2 ( RF2 ) ( Zaher and Green , 2009 ) . Studies of ribosomal stalling arising at rare codons ( Hayes et al . , 2002 ) or in contexts of depleted or inefficient cognate decoding elements ( George et al . , 2016; Li et al . , 2007; Roche and Sauer , 1999 ) suggest that a number of these mechanisms could resolve translation at unassigned codons , but a lack of well-characterized model organisms with an unassigned codon has precluded direct study of this question . Here , we use the GRO as a model to demonstrate that unassigned UAG codons in mRNA transcripts ( 1 ) elicit suppression , ribosomal frameshifting , and ribosomal rescue , ( 2 ) can induce ribosomal frameshifting from at least −3 to +19 nucleotides , and ( 3 ) lead to total loss of translational fidelity . By selectively deleting ribosomal rescue pathways in the GRO , we show that the tmRNA system is primarily responsible for rescuing ribosomes stalled at unassigned codons , with deletion of the tmRNA restoring expression of UAG-ending genes and re-enabling propagation of UAG-containing plasmids and viruses in the GRO . Our work reveals mechanistic details into how cells rescue ribosomes stalled at unassigned stop codons , providing insight into how alternative genetic codes act as barriers to HTGEs and demonstrating genomic recoding as a broadly applicable strategy to obstruct HGT in engineered organisms .
In prior work , we constructed an Escherichia coli strain in which all UAG codons were mutated to UAA , permitting the deletion of release factor 1 ( RF1 ) and resulting in an organism that lacks a codon assignment of UAG . This genomically recoded organism ( GRO ) ( Isaacs et al . , 2011; Lajoie et al . , 2013b ) exhibited resistance to multiple viruses and failure to propagate conjugative plasmids ( Lajoie et al . , 2013b; Ma and Isaacs , 2016 ) attributable to the unassigned UAG codon , but the molecular mechanisms that resolve unassigned UAG codons during translation remained unknown . In this study , we conducted two main experiments to uncover these mechanisms: ( 1 ) analysis of proteins translated from UAG-ending transcripts via mass spectrometry and western blots and ( 2 ) phenotypic assays to assess whether gene deletions of specific rescue factors restored the ability of conjugative plasmids and viruses to exploit the GRO . Since we hypothesized that the tmRNA-mediated response may resolve ribosomal stalling at the UAG codon , we also mutated the degradation tag encoded by the tmRNA from AANDENYALAA ( AA-tag ) to AANDENYALDD ( DD-tag ) for protein expression for mass spectrometry experiments . This mutation increases the half-life of protein products released by tmRNA ( Keiler et al . , 1996; Roche and Sauer , 1999 ) , enabling their detection via mass spectrometry . We assembled plasmids ( pUAG-GFP and pUAA-GFP ) encoding GFP genes with C-terminal 6x-His tags positioned immediately upstream of a UAG or UAA stop codon . We then expressed GFP from pUAG-GFP and pUAA-GFP in GRO cells containing the RF1-encoding prfA gene ( GRO . DD . prfA+ ) or in GRO cells lacking prfA and consequentially without UAG assignment ( GRO . DD ) ( Figure 2A; Table 1; see also Key Resources Table for a list of plasmids used in this study ) . We then purified proteins by nickel affinity chromatography , performed trypsin digest , and used tandem mass spectrometry to collect peptide mass data as described previously ( Aerni et al . , 2015; Amiram et al . , 2015 ) . To distinguish between mechanisms of ribosomal rescue and mistranslation at the UAG codon , we searched mass spectrometry data with theoretical peptide libraries detailed in Table 2 ( see also Supplementary file 3 and 4 ) to identify evidence for suppression , ribosomal frameshifting , rescue via tmRNA tagging , and loss of translational fidelity . In the GRO lacking UAG assignment , the UAG codon elicited a combination of ribosomal rescue mechanisms and mistranslation events , including tmRNA-mediated tagging , near-cognate suppression , and frameshifting . The mutated ssrA DD-tag appended directly to the C-terminus of GFP ( LEHHHHHHAANDENYALDD ) appeared in both UAG- and UAA-ending transcripts in GRO . DD and GRO . DD . prfA+ ( Figure 2 , Supplementary file 1 – Table S1 ) , consistent with previous reports that overexpressed proteins are targeted for degradation by the tmRNA ( Baneyx and Mujacic , 2004; Li et al . , 2007; Moore and Sauer , 2005; Tu et al . , 1995 ) . Both samples also contained the unmodified C-terminus of GFP ( LEHHHHHH ) . In GRO . DD . prfA+ , this is likely due to translational termination via RF1 , while in GRO . DD this may represent rescue of nonstop ribosomes by ArfA/ArfB , release of nascent peptides undergoing translation at the time of cell lysis , or spontaneous dissociation of the ribosome , although this last event is estimated to occur fewer than once per 100 , 000 codon decoding events ( Keiler and Feaga , 2014 ) . While these were the only C-terminal fragments detected in GRO . DD expressing UAA-GFP and in GRO . DD . prfA+ expressing UAG-GFP , GRO . DD [pUAG-GFP] contained greater than 30 unique C-terminal sequences ( Supplementary file 2 ) . The peptide fragments detected from GRO . DD [pUAG-GFP] demonstrate a combination of near-cognate suppression , ribosomal frameshifting , and tmRNA tagging ( Figure 2B ) . We identified two previously known suppression events glutamine ( Q ) and tyrosine ( Y ) ( Aerni et al . , 2015; Lajoie et al . , 2013b ) , and observed two new suppressors , aspartic acid ( D ) and valine ( V ) . We detected ribosomal frameshifting of up to −3 ( LEHHHHHHH ) and +19 nucleotides ( LEHHHHHHMVR ) , as determined by the presence of fragments from all three reading frames appended to the C-terminal peptide of LEHHHHHH . Additionally , the LEHHHHHHHH peptide may indicate a −6 frameshift , although it is impossible to determine whether this peptide arises from a −6 frameshift or two −3 frameshifts between histidine incorporation . We also detected peptides encoded as far downstream as +82 nucleotides after the UAG codon , illustrating that the ribosome can continue translation after encountering the unassigned UAG codon provided that stalling at the UAG codon is resolved . Lastly , we identified the modified ssrA DD-tag at both the site of the UAG codon and downstream on multiple peptides . Prior research in vitro revealed that a mistranslation event increases the likelihood of subsequent mistranslation events and termination by release factor 2 ( RF2 ) ( Zaher and Green , 2009 ) , and we investigated whether we could detect peptides representing such mistranslation events . Given the difficulty of distinguishing such peptides from suppression or frameshifting with one or two amino acids , we created a hypothetical peptide library ( Supplementary file 1 – Table S2 ) containing all combinations of LEHHHHHHXXX , wherein X is any amino acid incorporated at the three residue positions directly downstream of the UAG codon ( Supplementary file 4 ) . The search with this library returned 23 unique peptides , 14 of which met our scoring threshold of 15 ( Aerni et al . , 2015 ) . Five of these peptides ( LEHHHHHHEKP , LEHHHHHHQLD , LEHHHHHHQQR , LEHHHHHHSLK , and LEHHHHHHYQR ) could only arise from the mRNA transcript through two or more frameshift events after stalling at the UAG codon had already resolved ( Supplementary file 1 – Table S2 ) , suggesting they instead arise from loss of translational fidelity and spontaneous termination of translation following mistranslation at the UAG codon . We also had enough resolution in the data to manually verify the amino acid sequences of LEHHHHHHQQR and LEHHHHHHYQR , noting a 35 Da shift in mass between the Q and Y in the third position from the C-terminus . Although several alternative hypotheses may explain these random tripeptides , these explanations are either incomplete or unlikely given our current understanding of prokaryotic translation . First , it is improbable that these fragments arose from routine errors in mRNA transcription because this would require at least two transcriptional errors in a nine-nucleotide span . The transcription error rate in E . coli is estimated to be ~1 in 10 , 000 bases ( Blank et al . , 1986; Rosenberger and Hilton , 1983 ) and our strains have no known mutations that would lead to greater error rates in transcription . Second , it is possible that ArfA or ArfB may have terminated translation in these peptides due to 3’ exonuclease shortening of the mRNA transcript as the ribosome is stalled at the UAG codon ( Keiler and Feaga , 2014; Yamamoto et al . , 2003 ) . However , this does not explain the non-encoded tripeptides appended to the LEHHHHHH peptide . Lastly , the peptides LEHHHHHHQQR , LEHHHHHHSLK , and LEHHHHHHYQR may have been part of longer peptides that were cleaved off during trypsin digest . In this case , translation may have continued past the C-terminal R or K observed in these peptides , but this consideration would not apply to LEHHHHHHEKP and LEHHHHHHQLD and again does not explain the non-encoded tripeptide sequence observed appended to LEHHHHHH . Given this , we hypothesize that these five peptides result from loss of translational fidelity after stalling at the UAG codon that may lead to ( 1 ) spontaneous termination of translation due to the untemplated action of RF2 following mistranslation or ( 2 ) ArfA- or ArfB-mediated release predicated on 3’ exonuclease degradation of the mRNA . The rare event of spontaneous hydrolysis of the peptide from the ribosome is also possible . Since mass spectrometry data indicated that a combination of mechanisms could resolve stalled translation at the unassigned UAG codon , we generated targeted deletions of the ribosomal rescue systems ( ssrA , arfA , and arfB ) in strains with wild-type ssrA sequence ( GRO . AA ) to determine whether protein production from UAG-ending transcripts in ΔRF1 cells could be restored to levels seen in +RF1 cells . Using recombineering ( Sharan et al . , 2009 ) , we produced single and double deletions of the ssrA , arfA , and arfB genes that encode the ribosomal rescue systems . Efforts to generate a double deletion of ssrA and arfA failed ( data not shown ) because the resulting phenotype is synthetic lethal ( Chadani et al . , 2010 ) . We transformed each deletion strain with the UAG-GFP construct under a highly expressing , inducible pLtetO promoter ( Lutz and Bujard , 1997 ) and induced GFP expression for 20 hr , measuring the effect of protein expression on cellular growth through doubling time and maximum optical density at 600 nm ( OD600 ) ( Figure 3A and B , Supplementary file 1 – Table S3 ) . To quantify protein expression , we then assayed whole-cell lysate from equal cell numbers , as determined by OD600 , for abundance of protein via anti-GFP western blot alongside GFP standards of known concentration as described previously ( Figure 3C , Figure 3—source data 6 ) ( Pirman et al . , 2015 ) . We also included as positive controls ( 1 ) a wild-type strain ( ECNR2 ) expressing the UAG-GFP construct and ( 2 ) GRO . AA expressing UAA-GFP . Expression of UAG-GFP impaired GRO growth rate and cell density , generating a 54% increase in doubling time and 8% reduction in maximum OD600 compared to cells not expressing UAG-GFP , and a 25% greater doubling time and 14% lower maximum OD600 compared to cells expressing UAA-GFP . In contrast , ECNR2 exhibited only a 7% increase in doubling time and a 5% reduction in maximum OD600 when expressing UAG-GFP . Although deletion strains experienced reduced growth rate as measured by doubling time compared to the GRO . AA , they exhibited a less pronounced increase in doubling time when expressing UAG-GFP ( increases in doubling time between 12% and 50% ) as compared to the GRO . AA ( 54% increase in doubling time ) ( Figure 3A ) . However , deletion of ssrA reduced fitness during protein expression as measured by maximum OD600 , with GRO . AA . ∆ssrA demonstrating a 34% reduction in max OD600 and GRO . AA . ∆ssrA . ∆arfB demonstrating a 61% decrease in max OD600 . This is potentially due to increased presence of misfolded or prematurely truncated peptides that are ordinarily tagged and degraded by the tmRNA . Interestingly , deletion of arfB produces a 50% increase in doubling time during protein expression , suggesting ArfB may play a role in ribosomal rescue during high levels of ribosomal stalling . We then investigated the impact of unassigned codons on protein production using western blot densitometry , and found that the GRO expressing UAG-GFP produced less than one-fourth of the protein amount than does ECNR2 expressing UAG-GFP ( Figure 3C , 8 . 0 µg/ml for the GRO versus 35 µg/ml for ECNR2 , p=0 . 0014 ) . GRO . AA expressing UAA-GFP produced nearly nine times more protein than did GRO . AA expressing UAG-GFP ( 68 µg/ml for GRO . AA [pUAA-GFP] versus 8 . 0 µg/ml for GRO . AA [pUAG-GFP] , p<0 . 0001 ) , indicating that the UAG codon in pUAG-GFP is the cause of reduced protein expression in the GRO . Deletion of ssrA in the UAG-GFP-expressing GRO partially restored protein production to levels seen in its UAA-GFP-expressing counterpart with no knockouts ( 31 µg/ml for GRO . AA . ∆ssrA [pUAG-GFP] versus 68 µg/ml for GRO . AA [pUAA-GFP] ) and deletion of both ssrA and arfB fully restored protein production ( 70 . µg/ml ) . These ssrA deletion strains likely demonstrate increased GFP expression and reduced growth rate ( Figure 3A ) and cell density ( Figure 3B ) because translation of GFP transcripts sequesters cellular resources at the expense of cellular replication , producing GFP peptides that are freed from nonstop ribosomes via ArfA or ArfB without addition of a degradation tag . A deletion of arfB leads to strikingly low- protein abundances from UAG-GFP transcripts that approach the lower limit of detection of our assay , although this apparent reduction in protein production was not statistically significant in comparison to protein production by GRO . AA [pUAG-GFP] . These ArfB deletion data , together with the fitness reduction observed in the GRO , suggest that ArfB is constitutively expressed and relieving low levels of ribosomal stalling in E . coli . These data also suggest that while deletion of ssrA partially recovers protein production from UAG-ending transcripts in the GRO , deletion of both ssrA and arfB is necessary to fully recover protein expression from UAG-ending transcripts to levels seen from the translation of UAA-ending transcripts in the GRO . To determine whether deletions of of ssrA or arfB could restore propagation of horizontally-transferred genetic elements in the GRO , we assessed conjugation efficiency and growth rate from plasmids RK2 and F on GRO strains with single and double deletions of ssrA , arfA , and arfB . Previous research indicates that the UAG stop codon in the trfA gene on RK2 leads to impaired conjugation efficiency and replication in the GRO ( Ma and Isaacs , 2016 ) , likely because the TrfA protein is required to initiate plasmid replication ( Pansegrau et al . , 1994 ) . Phenotypically , this manifests as reduced efficiency of plasmid transfer in conjugation experiments and increased doubling times for RK2+ strains in media selecting for plasmid maintenance due to loss of plasmid and concomitant antibiotic resistance genes . We found that deletion of ssrA increased the ability of the GRO to both receive ( Figure 4A , Supplementary file 1 – Table S4 ) and replicate RK2 ( Figure 4B , Supplementary file 1 – Table S5 ) . RK2 conjugation efficiency in GRO . AA . ∆ssrA improved to 99% ( compared to 87% in GRO . AA ) , and the strain showed an increase in doubling time of only 6% compared to a 28% increase for GRO . AA ( p<0 . 0001 ) . We observed similar results for GRO . AA . ∆ssrA . ∆arfB . However , single deletion of arfB halved RK2 conjugative efficiency ( Figure 4A , p=0 . 0002 ) . This strain also exhibited a 38% increase in doubling time when bearing RK2 , compared to the 28% increase in doubling time seen in the GRO with no ribosomal rescue gene deletions ( Figure 4B , p<0 . 0001 ) . For plasmid F ( Figure 4C , Supplementary file 1 – Table S6 ) , which contains UAG-ending genes traY and traL that are essential for conjugation between cells ( Ma and Isaacs , 2016 ) , we found that deletion of ssrA increased conjugation events from the GRO donor 1 , 000-fold to 3 . 56 × 107 ( p=0 . 0015 ) compared to GRO . AA ( 3 . 30 × 104 events ) , arfA deletion ( 3 . 41 × 104 events ) , and arfB deletion ( 3 . 47 × 104 events ) . GRO . AA . ∆ssrA . ∆arfB and GRO . AA . ∆arfA . ∆arfB exhibited 5 . 2- and 2 . 3-fold decrease in conjugative efficiency when compared to GRO . AA . ∆ssrA and GRO . AA . ∆arfA single deletion strains , respectively ( p<0 . 01 for each , Figure 4C ) . These reductions in RK2 and F conjugative efficiency attributable to arfB deletion indicate that ArfB likely contributes to relief of nonstop ribosomes when encoded in its native ribosomal context , supporting evidence of ArfB’s ribosomal rescue activity previously validated in vitro ( Handa et al . , 2011 ) and when over-expressed in the absence of ssrA and arfA in vivo ( Chadani et al . , 2010 ) . However , deletion of ssrA is sufficient to restore both conjugation and propagation of RK2 and F in the GRO . We next attempted infection with phage λ on our suite of deletion strains ( Figure 4D , Supplementary file 1 – Table S7 ) . Although deletion of arfA or arfB does not recover viral infection , deletion of the ssrA gene—either alone ( p=0 . 0016 ) or alongside deletion of arfB ( p<0 . 0001 ) —recovers λ infection of the GRO to levels similar to wild-type , with about 108 plaque forming units per mL ( PFU/mL ) ( Figure 4D ) . These results demonstrate that removal of ssrA has the greatest influence in restoring conjugative plasmid transfer efficiency and viral susceptibility in the GRO ( Figure 4E and F ) .
In this study , we use a genomically recoded organism ( GRO ) containing an unassigned UAG codon as a model to investigate the molecular mechanisms that obstruct the propagation of HTGEs in organisms with alternative genetic codes . We demonstrate that unassigned stop codons elicit near-cognate suppression , frameshifting , and the action of ribosomal rescue mechanisms ( Figure 2 ) . tmRNA-mediated ribosomal rescue prompted by the unassigned codon results in the degradation of nascent peptides translated from UAG-ending transcripts and obstructs the propagation of HTGEs ( Figure 3 , Figure 4 ) . Additionally , ssrA deletion strains exhibit both significantly increased UAG-GFP yields ( Figure 3C ) and recovered propagation of HTGEs ( Figure 4 ) , consistent with evidence that deletion of ssrA removes inhibition of ArfA production and releases nascent peptides from stalled ribosomes without degradation ( Chadani et al . , 2011; Garza-Sánchez et al . , 2011; Schaub et al . , 2012 ) . Our GRO model thus sheds light on the functional significance of previously described regulatory relationships while elucidating the unique mechanistic contributions of different ribosomal rescue systems in resolving translation at unassigned stop codons . These mechanistic outcomes that occur as a consequence of ribosomal stalling could be further investigated via ribosomal profiling in future work . The mass spectrometry data collected from our GRO model demonstrate the striking proclivity for the ribosome to undergo un-programmed frameshifting at unassigned stop codons and represents , to our knowledge , the first in vivo study to examine such frameshifting . Prior studies have revealed programmed ribosomal frameshifting from −4 to +50 nucleotides ( Atkins et al . , 2016; Baranov et al . , 2015; Huang et al . , 1988; Yan et al . , 2015 ) , but these studies focused on frameshifts programmed into mRNA transcripts through combinations of four mechanisms: ( 1 ) use of rare codons to slow translation speed at the skip site , ( 2 ) weak base pairing of the P-site tRNA anticodon and mRNA codon , ( 3 ) strong base pairing of the P-site tRNA anticodon to the location where the ribosome will re-bind the mRNA , and ( 4 ) a region six bases upstream of the re-binding site that mimics a Shine-Dalgarno sequence and offsets the energetic cost of frameshifting ( Pech et al . , 2010 ) . Although the UAG codon in our GFP transcript slows translation , the P-site codon-anticodon pair for the codon immediately upstream of UAG is exact ( CAC codon and GUGHis-tRNA anticodon ) ( Hsu et al . , 1984 ) and any frameshift except backward would incur greater mispairing between the P site codon and anticodon . Additionally , no Shine Dalgarno-like sequence ( AGGAGG ) ( Shine and Dalgarno , 1974; Vimberg et al . , 2007 ) exists upstream , suggesting that the GFP construct we use contains only one of the four elements required for programmed ribosomal frameshifting ( Supplementary file 1 ) . From our construct , we observed frameshifts of potentially up to −6 and +19 nucleotides in response to the unassigned UAG codon ( Figure 2 , Supplementary file 1 – Tables S1 and S2 ) . Collectively , our work uncovers a wide variety of frameshifting events that can occur in response to ribosomal stalling in vivo , highlighting the capacity of the ribosome to continue translation despite missing an essential translational component . Mass spectrometry analysis also revealed truncated mistranslation products that possibly represent loss of translational fidelity and termination by RF2 downstream of an initial mistranslation event at the UAG codon , known as post-peptidyl transfer quality control ( Petropoulos et al . , 2014; Zaher and Green , 2009 ) , a result previously only observed in vitro . Although prior studies decades ago revealed premature truncation products in vivo ( Manley , 1978 ) , they lacked the technical capability to determine whether these peptides arose from a single mistranslation event or demonstrated loss of translational fidelity after the ribosome encounters a rare or unassigned codon . The mistranslation products we detect show repeated mistranslation events that could not have been produced by suppression , ribosomal rescue , or frameshifting , unless the ribosome frameshifted multiple times after resolving stalling at the UAG codon ( Figure 2B , Supplementary file 1 ) . These events may be followed by ribosomal rescue via ArfA or ArfB , spontaneous ribosomal dissociation , or termination via release factor 2 , though our technique was not capable of distinguishing between these fates . Previous in vitro studies using purified ribosome complexes determined that a mistranslation event destabilized the P-site helix , reducing the ability of the A-site to discriminate between anticodons and resulting in further mistranslation events and rapid termination by RF2 with the assistance of release factor 3 ( Zaher and Green , 2009; Zaher and Green , 2010 ) . The researchers predicted that a single mistranslation event would also lead to prematurely truncated peptides with two or three miscoded C-terminal amino acids appended in vivo ( Zaher and Green , 2009 ) . These findings , together with our results , motivate future work to investigate the possibility of loss of translational fidelity after an initial translation error and highlight the GRO as a model for elucidating translational fidelity in vivo . The GRO demonstrates that general ribosomal rescue mechanisms resolve ribosomal stalling at unassigned stop codons . As most sequenced bacterial species contain a homolog of the tmRNA , ArfA , or ArfB ribosomal rescue systems ( Hudson et al . , 2014; Keiler , 2015 ) and eukaryotic cells contain analogous pathways that rescue stalled ribosomes ( Graille and Séraphin , 2012 ) , we anticipate that translational stalling at unassigned codons can be resolved similarly in these organisms . Accordingly , we hypothesize that organisms beyond E . coli should tolerate unassigned codons as intermediates toward codon reassignments in genomic recoding , efforts for which are underway in numerous prokaryotic and eukaryotic species ( Lau et al . , 2017; Napolitano et al . , 2016; Ostrov et al . , 2016; Richardson et al . , 2017 ) . Additional barriers to codon reassignment exist , such as regulatory roles of codons in gene expression ( Lajoie et al . , 2013a ) , but our findings indicate that unassigned codons are tolerable in the absence of specialized translational machinery to address them , both as intermediate steps towards codon reassignment and as permanent parts of the genetic code . Our findings suggest that we can use unassigned codons to engineer organisms with broad resistance to HTGEs and impart genetic isolation , increasing engineered organisms’ stability in biotechnology applications . Since tmRNA homologs are found in >99% of all sequenced bacterial genomes ( Hudson et al . , 2014; Keiler , 2015 ) , we would expect other organisms engineered to contain unassigned codons to exhibit immunity to horizontally transferred genetic elements . As researchers pursue further efforts in whole genome recoding ( Boeke et al . , 2016; Lau et al . , 2017; Napolitano et al . , 2016; Ostrov et al . , 2016; Richardson et al . , 2017 ) and engineer organisms for use in open environments , we require strategies to genetically isolate such organisms from their surrounding environment to ensure robust function , both individually ( Moe-Behrens et al . , 2013 ) and as members of microbial communities ( Grosskopf and Soyer , 2014; Hillesland and Stahl , 2010 ) . Genomically recoded organisms with unassigned codons would possess reduced susceptibility to exploitation by HTGEs , increasing their stability in open environments . Although this work demonstrates that an unassigned stop codon acts as a barrier to HGT , this current barrier can be breached by mutation or deletion of the tmRNA to produce a functional protein . In contrast , we expect that an organism with an unassigned sense codon would have even greater barriers to HGT , as premature termination at an unassigned sense codon would likely produce a nonfunctional , truncated peptide . We thus anticipate that further genomic recoding to engineer additional unassigned sense and nonsense codons may be a broadly applicable strategy to confer genetic isolation in living systems , facilitating the safe use of engineered organisms in complex open environments .
All bacteria used in this study are derived from E . coli ECNR2 , which is in turn derived from E . coli MG1655 ( GenBank ID: U00096 ) in which mutS is replaced by a zeocin resistance cassette ( Wang et al . , 2009; Lajoie et al . , 2013b ) . Additionally , the native bioAB genes found in MG1655 are replaced by the lambda red cassette in ECNR2 . This strain is designated ECNR2 . AA ( see Table 1 for full genotype ) . For experiments expressing UAG-GFP and UAA-GFP for mass spectrometry , strains with all 321 UAG codons changed to UAA ( designated ‘GRO’ strains ) were used to control for potential differences in protein expression arising from these mutations ( GenBank ID for GRO . AA: CP006698 ) . For all other experiments , control strains labeled wild-type ( WT ) are MG1655 derivatives retaining all 321 UAG codons . All deletions of ssrA , arfA , and arfB were generated with a tolC resistance cassette via recombineering ( Sharan et al . , 2009 ) . Modification of the ssrA tag from AANDENYALAA to AANDENYALDD ( AA->DD ) to increase stability of tagged proteins was performed with MAGE as described previously ( Gallagher et al . , 2014; Wang et al . , 2009 ) . All modifications to strains made in this study were validated through Sanger sequencing ( GeneWiz; South Plainfield , NJ ) . We performed all protein expression assays and conjugation assays in LB Lennox at pH 7 . 5 . We performed all phage assays in Tryptone-KCl ( TK ) media as described previously ( Jaschke et al . , 2012; Ma and Isaacs , 2016; Valentine et al . , 2002 ) . For viral relative titers , we used phage λ cI857 obtained from Dr . John Wertz at the Yale Coli Genetic Stock Center ( CGSC ) because it is obligately lytic at 37°C , preventing possible confounding factors from lysogeny . We used the conjugative plasmid RK2 described in Isaacs et al . ( 2011 ) , which is a derivative of the RK2 plasmid described in Pansegrau et al . ( 1994 ) carrying blaR instead of kanR . The complete nucleotide sequence for the plasmid is available in NCBI database , Accession L27758 . 1 and GI 508311 . We obtained the F plasmid from the Yale CGSC ( NCBI Accession AP001918 . 1 , GI: 8918823 ) and added KanR from plasmid pZE21 for antibiotic selection . To create the UAG-GFP and UAA-GFP constructs for protein expression , we cloned an eGFP construct with a C-terminal 6xHis tag downstream of pLtetO into a modified pZE21 vector with kanamycin resistance ( kanR ) carrying a copy of the tet repressor gene ( tetR ) to prevent leaked gene expression . We then modified the stop codon of the eGFP construct to end in either a UAG or UAA stop codon . To obtain GFP for analysis via mass spectrometry , we transformed UAG-GFP and UAA-GFP constructs into wild-type and GRO strains carrying the AA->DD modification in the ssrA tag to prolong the half-life of tagged peptides . Experiments in the absence of the AA->DD modification yielded no peptides with ssrA degradation tags ( data not shown ) . We then grew 50 mL cultures of each strain at 33°C in LB Lennox with 30 μg/mL kanamycin to an OD600 of 1 . 0 and induced protein expression with the addition of 30 ng/uL anhydrotetracycline ( aTC ) . After incubation overnight , we pelleted cells and resuspended them in sterile phosphate buffer solution , then lysed cells via sonication . Cell debris was then pelleted by centrifugation and GFP purified from supernatant via a nickel resin affinity column . To concentrate protein and exchange buffer for subsequent trypsin digest , we then concentrated GFP via Millipore Amicon spin columns . For whole western blots on whole cell lysates , we transformed UAG-GFP and UAA-GFP constructs into wild-type , GRO , and GRO strains with deletions of the ribosomal rescue systems . We then grew 5 mL cultures of each strain at 33°C in LB Lennox with kanamycin overnight , then diluted all cultures OD600 of 0 . 15 in fresh media containing 30 μg/mL kanamycin and 30 ng/uL aTC for 20 hr . To quantify protein expression and compare across strains , we normalized the OD600 of all cultures to 2 . 5 and pelleted 1 mL of this culture , which we placed in the −80C for 2 hr . We then re-suspended cell pellets in lysis buffer described previously ( Aerni et al . , 2015 ) , incubated for 10 min on ice , centrifuged lysate , and ran 1:10 dilutions of resulting supernatant on gels for western blot analysis . Overnight starter cultures were diluted to an OD600 of 0 . 15 into three separate culture tubes , and cells within each tube were induced in parallel for GFP expression . GFP was purified from each of these cultures in parallel . Trypsin digest , sample preparation for mass spectrometry , and liquid chromatography elution gradients were performed as described previously ( Aerni et al . , 2015 ) . Desalted peptides were injected onto a 75 μm ID PicoFrit column ( New Objective ) packed to 50 cm in length with 1 . 9 μm ReproSil-Pur 120 Å C18-AQ ( Dr . Maisch ) . Samples were eluted over a 90 min gradient using an EASY-nLC 1000 UPLC ( Thermo ) paired with a Q Exactive Plus ( Thermo ) , using the following parameters: ( MS1 ) 70 , 000 resolution , 3 × 106 AGC target , 300–1700 m/z scan range; ( MS2 ) 17 , 500 resolution , 1 × 106 AGC target , top 10 mode , 1 . 6 m/z isolation window , 27 normalized collision energy , 90 s dynamic exclusion , unassigned and +1 charge exclusion . Peptide identification from collected spectra was performed using MaxQuant v1 . 5 . 1 . 2 ( Cox and Mann , 2008 ) . Samples were searched using custom databases representing potential translational outcomes in response to the UAG codon within the GFP reporter construct ( Supplementary file 3 and 4 ) , as well as the E . coli proteome ( EcoCyc K-12 MG1655 v17 ) . The searches considered carbamidomethyl ( Cys ) as a fixed modification and the following variable modifications: acetyl ( N-terminal ) , oxidation ( Met ) , deamidation ( Asn , Gln ) , and phosphorylation ( Ser/Thr/Tyr ) . Discovered peptides had a minimum length of five amino acids and could contain up to three trypsin miscleavage events . A 1% false discovery rate was used . The mass spectrometry proteomics data and the custom search databases have been deposited to the ProteomeXchange Consortium ( http://proteomecentral . proteomexchange . org ) via the PRIDE partner repository ( Vizcaíno et al . , 2014 ) with the dataset identifier PXD009643 . Mass spectrometry spectra were manually validated by identifying all spectra with an MS/MS score over 15 and verifying the presence sufficient b- and/or y-ion series . Western blots were run as described previously using SDS-PAGE gels ( Pirman et al . , 2015 ) . We ran GFP-6xHis standards of known amount ( 1 , 10 , 50 , and 100 ng ) alongside experimental samples and used these standards to generate linear-range calibration curves to quantify protein abundance in experimental samples ( Figure 3—figure supplement 1 ) . Because the antibody signal appeared sublinear in the 0–10 ng regime when we performed linear regression using all standards , we generated separate linear fits using the 1–10 ng standards and the 10–100 ng standards . We then determined experimental sample concentrations using these linear approximations . 20 of the 24 experimental samples quantified fell within or slightly above the 10–100 ng range ( with the highest-intensity sample quantified as 136 ng ) , and 3 of the 24 samples fell within the 1–10 ng range . The one remaining sample , which had a weaker intensity than that of the 1 ng standard , was quantified through a linear approximation between the intensity of the 1 ng sample and of a blank lane with an assumed intensity of zero . We expressed GFP-6xHis as described above , normalized cell cultures to an OD600 of 2 . 5 , and lysed cells using BugBuster protein extraction reagent ( Merck , Darmstadt , Germany ) . We then ran 10 µl of 1/150 diluted lysate per lane of the SDS-PAGE gel . We obtained primary mouse anti-GFP antibody from Invitrogen ( Ref#: 332600 , Lot#: 1513862A; RRID:AB_2234927 ) and goat anti-mouse antibody from AbCam ( Ref#: ab7023 , Lot#: GR157827-1; RRID:AB_955413 ) . Western blots were developed using Bio-Rad Clarity Western ECL Blotting Substrate and Imaged on a GE Amersham Imager 600 . We performed quantification of western blot bands as described previously ( Pirman et al . , 2015 ) . We repeated three western blots in parallel for each strain induced in separate culture tubes ( i . e . biological triplicates , see Protein expression and purification ) . To quantify relative titers , we mixed 100-fold dilutions of phage with 300 µL of mid-log ( OD600 = 0 . 5 ) cells in 3 mL of TK soft agar and poured onto TK solid agar plates . Starter cultures of cells were diluted to an OD600 of 0 . 5 into three separate culture tubes , and cells within each tube were infected with phage lambda in parallel ( i . e . biological triplicate ) . Each tube was plated on a separate TK solid agar plate . We incubated plates overnight at 37°C , and counted plaques the next day . We used conjugation conditions described previously ( Ma and Isaacs , 2016; Ma et al . , 2014 ) . Briefly , we grew cultures of donor and recipient cells to late log in antibiotics selecting for plasmid or recipient and then rinsed and re-suspended in media to remove antibiotics . After concentrating cells to an OD600 of 20 , we mixed donors and recipients in 1:1 ratio and spotted onto pre-warmed LB Lennox agar plates in 2 × 20 uL and 6 × 10 uL pattern . For F , we incubated plates at 37°C for 2 hr , then rinsed cells off plate , diluted serially 10-fold , and plated serial dilutions on plates containing antibiotic selecting for conjugants and incubated overnight at 37°C . For RK2 , we incubated plates at 37°C for 1 hr , then plated on agar plates selecting for the recipient . To quantify the rate of transfer , we then picked 86 colonies from plates selecting for the recipient strain and patched them onto plates selecting for both recipient and conjugative plasmid , incubated plates overnight at 37°C , and counted the number of patched colonies that grew . After the conjugation , colonies were plated three times to generate technical triplicates . We performed all t-tests and one-way ANOVA tests for statistical significance in GraphPad Prism 7 . We calculated doubling times and maximum OD600 values from growth curve data using MATLAB ( Newton , MA ) code that we generated ( Source code 1 ) . We used the definitions for biological and technical replicates outlined in Blainey et al . , 2014 . Biological replicates consist of parallel measurements of different biological samples subjected to the same experiment , and technical replicates are parallel measurements of a single biological sample subjected to experimentation . Data represented in ( Figures 3 , 4B and D ) are biological replicates; data represented in ( Figure 4A and C ) are technical replicates . Data for all 96-well plate assays ( Figures 3A , B and 4B ) were obtained as biological replicates: One well of each sample was grown overnight as a starter culture in a 96-well plate . Starter cultures were then inoculated into three separate wells in a separate 96-well plate . | Usually , DNA passes from parent to offspring , vertically down the generations . But not always . In some cases , it can move directly from one organism to another by a process called horizontal gene transfer . In bacteria , this happens when DNA segments pass through a bacterium’s cell wall , which can then be picked up by another bacterium . Because the vast majority of organisms share the same genetic code , the bacteria can read this DNA with ease , as it is in the same biological language . Horizontal gene transfer helps bacteria adapt and evolve to their surroundings , letting them swap and share genetic information that could be useful . The process also poses a threat to human health because the DNA that bacteria share can help spread antibiotic resistance . However , some organisms use an alternative genetic code , which obstructs horizontal gene transfer . They cannot read the DNA transmitted to them , because it is in a different ‘biological language’ . The mechanism of how this language barrier works has been poorly understood until now . Ma , Hemez , Barber et al . investigated this using Escherichia coli bacteria with an artificially alternated genetic code . In this E . coli , one of the three-letter DNA ‘words’ in the sequence is a blank – it does not exist in the bacterium’s biological language . This three-letter DNA word normally corresponds to a particular protein building block . Using a technique called mass spectrometry , Ma et al . analyzed the proteins this E . coli forms . The results showed that it has several strategies to deal with DNA transmitted horizontally into the bacterium . One method is destroying the proteins that are half-created from the DNA , using molecules called tmRNAs . These are part of a rescue system that intervenes when protein translation stalls on the blank word . The tmRNAs help to add a tag to half-formed proteins , marking them for destruction . This mechanism creates a ‘genetic firewall’ that prevents horizontal gene transfer . In organisms engineered to work from an altered genetic code , this helps to isolate them from outside interference . The findings could have applications in creating engineered bacteria that are safer for use in fields such as medicine and biofuel production . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"genetics",
"and",
"genomics"
] | 2018 | Organisms with alternative genetic codes resolve unassigned codons via mistranslation and ribosomal rescue |
Each taste modality is generally encoded by a single , molecularly defined , population of sensory cells . However , salt stimulates multiple taste pathways in mammals and insects , suggesting a more complex code for salt taste . Here , we examine salt coding in Drosophila . After creating a comprehensive molecular map comprised of five discrete sensory neuron classes across the fly labellum , we find that four are activated by salt: two exhibiting characteristics of ‘low salt’ cells , and two ‘high salt’ classes . Behaviorally , low salt attraction depends primarily on ‘sweet’ neurons , with additional input from neurons expressing the ionotropic receptor IR94e . High salt avoidance is mediated by ‘bitter’ neurons and a population of glutamatergic neurons expressing Ppk23 . Interestingly , the impact of these glutamatergic neurons depends on prior salt consumption . These results support a complex model for salt coding in flies that combinatorially integrates inputs from across cell types to afford robust and flexible salt behaviors .
Sodium is essential to survival , but its intake must be carefully regulated to maintain ionic homeostasis . It is therefore unsurprising that taste systems have evolved robust mechanisms for detecting salt , and that salt palatability depends on its concentration . In general , sodium concentrations below 100 mM tend to be attractive , while any salt present at higher concentrations becomes increasingly aversive ( Chandrashekar et al . , 2010; Lindemann , 2001; Oka et al . , 2013 ) Although there is considerable debate about modes of central taste coding , there is strong evidence that most taste modalities activate a single , molecularly defined , population of peripheral taste receptor cells ( Yarmolinsky et al . , 2009 ) . However , research in both mammals and insects has favoured a dual-pathway model for salt taste: a low-threshold sodium-specific population of ‘low salt’ cells mediates attraction , which is overridden at higher concentrations by ion non-specific ‘high salt’ cells that drive avoidance ( Ishimoto and Tanimura , 2004; Lindemann , 2001; Marella et al . , 2006; Oka et al . , 2013; Zhang et al . , 2013 ) . Moreover , two distinct aversive taste receptor cell ( TRC ) types ( bitter and sour ) contribute to high salt taste in mammals ( Oka et al . , 2013 ) . Thus , peripheral coding of salt taste appears more complex than other primary taste modalities . The Drosophila labellum contains three types of gustatory sensilla , each of which harbors 2–4 gustatory receptor neurons ( GRNs ) ( Singh , 1997; Stocker , 1994 ) ( Figure 1A ) . Short ( S-type ) and long ( L-type ) sensilla have four molecularly and physiologically distinct GRNs , while intermediate ( I-type ) sensilla have only two ( Freeman and Dahanukar , 2015; Scott , 2018; Stocker , 1994 ) . Extracellular ‘tip-recordings’ of different sensilla have identified four GRN types: a water ( W ) cell that responds to low osmolarity; a sugar ( S ) cell that responds to sweet compounds; a low salt ( L1 ) cell that is sodium-specific; and a high salt ( L2 ) cell that responds to high ionic concentrations ( >250 mM ) ( Fujishiro et al . , 1984; Hiroi et al . , 2002; Ishimoto and Tanimura , 2004 ) . S- and L-type sensilla are thought to have one of each GRN type , with S-type L2 cells responding to bitter compounds in addition to high salt ( Meunier et al . , 2003 ) ( Figure 1B ) . I-type sensilla were shown to have an S/L1 hybrid cell that responds to sugars and low salt , and an L2 cell that responds to bitters and high salt ( Hiroi et al . , 2004 ) ( Figure 1B ) . The early physiological recordings have been mostly borne out by molecular characterization of GRN types ( Freeman and Dahanukar , 2015; Scott , 2018 ) ( Figure 1B ) . S- and L-type sensilla each have a single GRN that expresses the low osmolarity sensor Pickpocket28 ( Ppk28 ) and corresponds to the W cell ( Cameron et al . , 2010; Chen et al . , 2010; Inoshita and Tanimura , 2006 ) . The S cell is labelled by the sugar receptor Gr64f , along with other members of the gustatory receptor ( GR ) family ( Dahanukar et al . , 2007; Fujii et al . , 2015; Jiao et al . , 2007; Slone et al . , 2007; Thorne et al . , 2004; Wang et al . , 2004 ) . Similarly , Gr66a is co-expressed with other Grs in a single bitter responsive neuron per S-type and I-type sensillum , corresponding to the L2 cell ( Marella et al . , 2006; Thorne et al . , 2004; Wang et al . , 2004; Weiss et al . , 2011 ) . The degenerin/epithelial sodium channel ( Deg/ENaC ) family member Ppk23 , which is required for pheromone detection in leg gustatory sensilla , is known to be expressed in a labellar neuron population that partially overlaps with Gr66a/bitter GRNs ( Thistle et al . , 2012 ) . Ppk23 neurons are necessary for calcium avoidance , but details of the labellar Ppk23 expression map , as well as the physiology and function of these neurons are largely unknown ( Lee et al . , 2018 ) . In contrast to water , sweet , and bitter tastes , the principles of peripheral salt coding in flies remain unclear . Early calcium imaging experiments revealed low salt responses in Gr5a-Gal4 GRNs , suggesting that sweet neurons may mediate low salt attraction ( Marella et al . , 2006 ) . However , Gr5a-Gal4 was later shown to label additional GRNs outside the sweet class ( Fujii et al . , 2015 ) , and the Ionotropic receptor ( IR ) family member IR76b was proposed to specifically mediate low salt taste via a dedicated low salt cell distinct from sweet GRNs ( Zhang et al . , 2013 ) . This view was challenged by the recent demonstration that IR76b is also required for high salt taste , raising questions about its utility as a marker for one defined GRN population ( Lee et al . , 2017 ) . Moreover , although Gr66a GRNs showed calcium responses to high salt concentrations and electrophysiology suggested that bitter and high salt are encoded by the same sensory neurons , genetically eliminating these cells left behavioral aversion to high salt largely intact ( Marella et al . , 2006; Wang et al . , 2004 ) . Here , we probe the logic of salt coding across the labellum by systematically characterizing the physiological and behavioral roles of molecularly-defined GRN types covering the entire labellar GRN map . We find that all GRN types show dose-dependent excitation or inhibition by salt , indicating a complex model for salt coding . Of particular interest is that , like mammals , flies have two distinct high salt cells . In addition to activating canonical bitter neurons , high salt concentrations excite a glutamatergic GRN population expressing Ppk23 . Salt responses of these ‘Ppk23glut’ GRNs require IR76b , whereas those of bitter GRNs do not . Both bitter and Ppk23glut GRNs are necessary for behavioral avoidance of high salt when flies have been reared on a salt-containing diet . However , salt deprivation reduces high salt avoidance by specifically suppressing the impact of Ppk23glut neurons , suggesting that these GRNs mediate internal state-dependent modulation of salt consumption . Consistent with this idea , closed-loop optogenetic activation of Ppk23glut neurons reduces feeding by salt-fed flies , but not those that have been salt deprived . Our results support a model where the combinatorial excitation and inhibition of various taste pathways mediates the behavioral valence of salt , with one pathway conferring the ability to specifically modulate salt consumption based on internal state .
Although several studies have mapped the expression of different receptors across the labellum ( Freeman and Dahanukar , 2015 ) , a comprehensive map covering all GRN types was still lacking . We began by asking whether the vesicular glutamate transporter ( VGlut ) may define a functionally distinct population of GRNs . An enhancer trap upstream of VGlut , OK371-Gal4 , labels an uncharacterized population of putatively glutamatergic neurons in the labellum , and a subset of pheromone-responsive GRNs in the legs ( Kallman et al . , 2015; Mahr and Aberle , 2006 ) . VGlutMI04979-Gal4 , which is a gene-trap inserted within a VGlut exon , showed expression in a single cell per S-type and L-type sensillum in the labellum ( Figure 1C ) . These cells do not overlap with those expressing a similar gene trap for choline acetyltransferase ( ChAT ) , supporting the idea that VGlutMI04979-Gal4 labels a bona fide population of glutamatergic GRNs ( Figure 1D ) . Co-labelling of VGlutMI04979-Gal4 with LexA reporters for known sensory neuron populations revealed that VGlut is not expressed in water ( Ppk28 ) , sweet ( Gr64f ) , or bitter ( Gr66a ) GRNs ( Figure 1E–G ) . However , all VGlut+ cells were positive for Ppk23 ( Figure 1H ) . Further examination of Ppk23 expression revealed that Ppk23 GRNs are comprised of two distinct subsets: most S-type and all L-type sensilla contain a single GRN that expresses Ppk23 and VGlut; and six S-type sensilla , roughly corresponding to those designated as ‘S-a’ sensilla ( Freeman and Dahanukar , 2015; Weiss et al . , 2011 ) , have a second Ppk23 GRN that is positive for Gr66a and ChAT ( Figure 1I ) . We will refer to these two populations as Ppk23glut and Ppk23chat , respectively . We then used gene trap insertions of Gal80 into VGlut and ChAT to isolate Ppk23chat and Ppk23glut GRNs . While this restriction largely agreed with our co-expression data , it was not perfect: VGlut-Gal80 restricted Ppk23-Gal4 expression to four S-type GRNs instead of the expected six ( Figure 1J ) ; and ChAT-Gal80 suppressed Ppk23-Gal4 expression in most , but not all , Ppk23chat GRNs , leaving 1 – 2 s-type sensilla with two Ppk23 neurons ( Figure 1—figure supplement 1A ) . We suspect these results reflect minor differences between expression of the Gal4 , LexA , and Gal80 reporters , although we confirmed that Ppk23-Gal4 and Ppk23-LexA labelled the same population of GRNs ( Figure 1—figure supplement 1B ) . To create a more conservative representation of Ppk23glut , we constrained Ppk23-Gal4 activity using Gr66a-LexA and LexAop-Gal80 ( Figure 1K ) . This manipulation faithfully restricted expression to only Ppk23glut GRNs , but also globally reduced expression levels . Thus , we retained both methods of isolating Ppk23glut cells for functional characterization . Our analysis of Ppk23 expression nearly completed the labellar GRN map: s-type sensilla generally have one Ppk28 ( water ) , one Gr64f ( sweet ) , one Gr66a ( bitter , some of which are Ppk23chat ) , and one Ppk23glut GRN; I-type sensilla have one Gr64f and one Gr66a GRN; and L-type sensilla have one Ppk28 , one Gr64f , one Ppk23glut , and one unidentified GRN that has been proposed to express IR76b and respond to low salt concentrations ( Freeman and Dahanukar , 2015; Zhang et al . , 2013 ) . To identify a marker for the last GRN class in L-type sensilla , we first examined IR76b . However , IR76b-Gal4 is expressed in many neurons from all four known classes of labellar GRNs , limiting its utility as a marker ( Figure 1—figure supplement 1C–F ) . We therefore visually screened the Vienna Tile ( VT ) and Janelia Rubin Gal4 collections for lines that sparsely label GRN projections in the brain , and identified VT046252-Gal4 , which drives Gal4 expression under the control of the genomic region upstream of the IR94e locus . Because the labellar projections to the subesophageal zone ( SEZ ) labeled by VT046252-Gal4 ( Figure 1—figure supplement 1G ) appear identical to those of a previously published reporter for IR94e expression ( Koh et al . , 2014 ) , we will henceforth simply refer to it as IR94e-Gal4 . IR94e-Gal4 is expressed in one cell per L-type sensillum , and does not overlap with Ppk28 , Gr64f , or Ppk23 ( Figure 1L–N ) . This driver is therefore specific for the fourth GRN class found in L-type sensilla and completes our molecular map of the labellum ( Figure 1O–P ) . With a complete labellar GRN map in hand , we examined the salt responses across all identified GRN classes . We expressed GCaMP6f under the control of each GRN class-specific Gal4 line and performed imaging of GRN axon terminals in the SEZ while stimulating the labellum with a series of tastants ( Figure 2A–D ) . As expected , known GRN classes responded strongly to their cognate modality – Ppk28 to water , Gr64f to sugar ( sucrose ) , and Gr66a to bitter ( lobeline ) ( Figure 2C–D ) . As previously demonstrated , Ppk28 neurons show dose-dependent inhibition by salt , as with any osmolyte ( Cameron et al . , 2010 ) . In contrast , Gr64f and Gr66a both showed dose-dependent excitation by salt , with Gr64f GRNs activated at a lower threshold . Moreover , Gr64f responses were sodium-specific , while Gr66a also responded to potassium chloride ( Figure 2C–D ) . These results are consistent with Gr64f operating as a ‘low salt’ cell type , and Gr66a acting as a ‘high salt’ cell type . Strikingly , we found that the two relatively uncharacterized labellar GRN types – IR94e and Ppk23 – also showed salt-evoked activity ( Figure 2C–D ) . IR94e displayed weak activation by 50 mM NaCl , but no responses to higher concentrations . Further testing of different salts at 100 mM revealed sodium-selective tuning , indicating that IR94e labels a second low salt cell type ( Figure 2—figure supplement 1A–B ) . The weak responses in IR94e neurons suggest a limited role in salt coding , but it is possible that they account for the previously observed peak response to low salt in L-type sensilla ( Zhang et al . , 2013 ) . On the other hand , Ppk23 neurons showed very strong dose-dependent salt responses that were ion non-selective . In addition to sodium chloride , we observed robust activation by 1 M solutions of potassium chloride , sodium bromide , potassium bromide , cesium chloride , and calcium chloride ( Figure 2—figure supplement 2A–B ) . As confirmation that the observed activity is salt-evoked and not a response to high osmolality , we found that Ppk23 GRNs do not respond to 1 M concentrations of sucrose ( Figure 2—figure supplement 2A–B ) . Since Ppk23 neurons on the leg are known to sense pheromones , we also tested labellar Ppk23 GRN responses to male and female cuticular hydrocarbons . We observed only very weak activation of Ppk23 neurons in female flies to a mixture of two male pheromones , and no significant responses in male flies ( Figure 2—figure supplement 2C–D ) . Together , these data suggest that a primary function of labellar Ppk23 GRNs is to mediate a high salt response , and position Ppk23 and Gr66a as markers of two high salt GRN classes . Our expression mapping revealed that Ppk23 GRNs encompass two subsets based on neurotransmitter expression: Ppk23chat and Ppk23glut . Given that Ppk23chat GRNs also express Gr66a , we suspected that this subpopulation may confer bitter responses to the Ppk23 population when measured as a whole . Although we did not observe Ppk23 activation in response to 0 . 3 mM lobeline ( Figure 2C–D ) , we did see strong responses to caffeine ( Figure 3A ) . Interestingly , we observed a marked difference in the synaptic calcium signals in response to salt and bitter stimuli . While salt stimulation of Ppk23 GRNs resulted in predominantly lateral activation of Ppk23 projections , bitter stimulation activated medial ring-like projections characteristic of Gr66a ( Figure 3A ) ( Kwon et al . , 2014; Thorne et al . , 2004; Wang et al . , 2004 ) . These activation patterns matched closely with the projections of the Ppk23glut and Ppk23chat subsets , revealed by restricting Ppk23-Gal4 activity with Gr66a-LexA and LexAop-Gal80 ( Ppk23glut ) or VGlut-Gal80 ( Ppk23chat ) ( Figure 3B ) . This suggests that Ppk23glut and Ppk23chat both respond to salt , but that only Ppk23chat responds to bitter compounds . To confirm this and reveal any functional differences in salt coding between the Ppk23 subpopulations , we measured the tuning of Ppk23glut and Ppk23chat using calcium imaging . As expected , Ppk23chat , but not Ppk23glut , GRNs exhibited bitter responses; however , both subpopulations showed strong dose-dependent excitation by salt ( Figure 3C–D ) . The salt responses in Ppk23glut GRNs appeared smaller than those of Ppk23chat , but we suspected this was due to very low GCaMP6f expression in Ppk23glut , which reduced the signal-to-noise in those measurements . We therefore repeated the Ppk23glut imaging by restricting Ppk23-Gal4 expression with ChAT-Gal80 . Consistent with the imperfect restriction we observed in the labellum ( Figure 1—figure supplement 1A ) , these flies had small , although insignificant , caffeine responses ( Figure 3—figure supplement 1A–B ) . However , this primarily Ppk23glut population exhibited very strong activation by salt ( Figure 3—figure supplement 1A–B ) . Taken together , our anatomical and functional studies support a salt coding model with two functionally distinct high salt GRN populations: Gr66a and Ppk23glut . We currently lack evidence of any functional distinctions between Gr66a GRNs that are positive or negative for Ppk23 . Therefore , for the purposes of salt coding , we will consider Gr66a GRNs as a uniform population that includes Ppk23chat . A previous report suggested that IR76b is specifically required for low salt responses in an L-type GRN class distinct from Gr64f ( Zhang et al . , 2013 ) . However , more recent evidence points to a role in both high and low salt taste ( Lee et al . , 2017 ) . Since we observed widespread IR76b-Gal4 expression in many GRN classes , we sought clarity on the role of IR76b in salt taste responses across the labellum . Calcium imaging in IR76b mutants revealed that IR76b is absolutely required for salt-evoked activity in Gr64f GRNs ( Figure 4A–B ) . By contrast , the salt responses of Gr66a GRNs were only mildly decreased in the mutants , showing that these neurons have a mostly IR76b-independent mechanism for detecting high salt . Ppk23 salt responses had a much stronger dependence on IR76b , with significantly decreased peak values , compared to controls , at all concentrations tested ( Figure 4—figure supplement 1A–B ) . Given the IR76b-independent salt responses in Gr66a GRNs , it was unsurprising that IR76b mutants showed some Ppk23 GRN activity in the medial region targeted by Ppk23chat ( Gr66a-positive ) projections ( Figure 4C ) . We therefore reanalyzed the Ppk23 dataset by quantifying fluorescence change in a region-of-interest restricted to the lateral areas characteristic of Ppk23glut projections ( Figures 3B and 4C ) . IR76b mutants exhibited essentially no salt-evoked activity in this target region , suggesting that IR76b is necessary for both the sodium and potassium salt responses of the Ppk23glut population ( Figure 4A–B ) . Since IR25a is expressed in GRNs and thought to be another broadly acting co-receptor ( Ahn et al . , 2017a; Benton et al . , 2009; Cameron et al . , 2010; Chen and Amrein , 2017; Lee et al . , 2018 ) , we also tested its involvement in salt taste . We found that IR25a mutants have GRN response profiles similar to those of IR76b mutants , suggesting that perhaps IR25a and IR76b act in a complex to mediate gustatory salt responses ( Figure 4—figure supplement 2 ) . However , in contrast to IR76b and IR25a , mutations in Ppk23 and the related ENaC Ppk29 had no observable effect on the salt-evoked calcium responses of Ppk23 GRNs , consistent with previously reported behavioral tests ( Figure 4—figure supplement 3A–B , [Thistle et al . , 2012] ) . Thus , the Ppk23 gene marks a salt-responsive GRN population but does not appear to be involved in salt detection . The fact that IR76b mutants lack salt responses in the primary low salt GRN class and one of two high salt GRN classes provides an explanation for observed defects in both low salt attraction and high salt avoidance ( Lee et al . , 2017; Zhang et al . , 2013 ) . Before further dissecting the cellular contributions of different GRN classes to salt behaviors , we wanted to establish behavioral assays that replicated these phenotypes . To test low salt attraction , we used a binary choice assay where flies were given the option to feed on either 50 mM salt mixed with low sugar ( 2 mM sucrose ) , or the same concentration of sugar alone ( LeDue et al . , 2015; Tanimura et al . , 1982; Zhang et al . , 2013 ) . As previously reported , control flies are strongly attracted to the salt-containing option , while IR76b mutants lose this attraction ( Figure 4D ) . We used a similar assay to probe high salt avoidance . In this case , control flies avoid 250 mM salt mixed with 25 mM sucrose in favor of plain sucrose at a lower concentration ( 5 mM ) . Much like their defects in low salt attraction , IR76b mutants are severely impaired in high salt aversion ( Figure 4E ) . To probe the cellular basis of salt behaviors , we conditionally silenced different GRN populations using Kir2 . 1 expression temporally restricted with Gal80ts . As expected , both Ppk23glut and Ppk28 GRNs were dispensable for low salt attraction ( Figure 5A ) . Focusing on the two GRN classes with low salt tuning properties , we found that Gr64f GRN activity is necessary for attraction to 50 mM NaCl , but expression of Kir2 . 1 in IR94e GRNs had no effect ( Figure 5A ) . Further , silencing Gr64f and IR94e neurons together resulted in behavior indistinguishable from Gr64f silencing alone . Puzzled by the apparent lack of a role for IR94e GRNs in salt attraction , we expressed a different effector – tetanus toxin ( TNT ) – in these neurons without any temporal restriction with Gal80ts , and observed reduced low salt attraction ( Figure 5—figure supplement 1 ) . Moreover , attraction was virtually eliminated when TNT was expressed in both Gr64f and IR94e GRNs ( Figure 5—figure supplement 1 ) . Thus , we conclude that sweet GRNs likely mediate the bulk of low salt attraction , with additional input from the IR94e class . Interestingly , Kir2 . 1 expression in IR76b-Gal4 GRNs had a similar effect to Gr64f silencing , suggesting that Gr64f mediates the bulk of IR76b-dependent low salt attraction . However , this phenotype appears less severe than that of IR76b mutants , which display mild low salt avoidance ( Figure 4D ) . This could reflect incomplete silencing from Kir2 . 1 , as suggested by the lack of observable effects in IR94e GRNs , or weak IR76b-independent low salt responses in Gr66a ( bitter/high salt ) GRNs that further reduce salt preference in IR76b mutants . In any case , restoring IR76b selectively to Gr64 neurons rescues low salt attraction in IR76b mutants , further supporting the role of sweet neurons in salt attraction ( Figure 4D ) . Since Gr64f neurons are necessary for sugar detection , we sought verification that the Gr64f salt attraction phenotype was not from an inability to sense the low concentration of sucrose in both food options . Indeed , Gr64f silencing caused a similar reduction in salt attraction in the absence of sugar ( Figure 5B ) . We then tested the role of each high salt GRN class in high salt avoidance and found that Gr66a and Ppk23glut GRNs are both necessary for this behavior ( Figure 5C ) . To confirm the novel role for Ppk23glut in behavioral salt avoidance , we tested its impact on the Proboscis Extension Reflex ( PER ) , which is an acute measure of gustatory palatability . Consistent with our binary choice assay , silencing Ppk23glut GRNs severely impaired the inhibition of PER by high salt ( Figure 5—figure supplement 2A ) . Moreover , rescue of IR76b expression in either Gr66a or Ppk23glut GRNs partially restores high salt avoidance to IR76b mutants ( Figure 4E ) . Fly gustatory responses are frequently modulated by need for specific nutrients ( Kim et al . , 2017 ) . However , modulating salt behaviors presents a complex problem because two of the three GRN classes exhibiting strong salt-evoked activity – Gr64f ( sweet ) and Gr66a ( bitter ) – have prominent roles in the detection of other modalities . These are therefore poor candidates for need-dependent modulation of salt responses , unless plasticity is achieved by regulating a salt-specific receptor . We therefore speculated that Ppk23glut GRNs , which to our knowledge specifically respond to salt , may tune the fly’s salt behaviors based on need . The high salt assay shown in Figure 5C was performed on flies under salt fed conditions ( three days with food containing 10 mM NaCl ) to maximize salt avoidance . We subsequently repeated this experiment with flies deprived of salt for three days and observed the expected weakening of salt aversion in controls ( Figure 5D; p<0 . 0001 compared to Figure 5C ) . Strikingly , while silencing Gr66a GRNs further reduced salt avoidance , silencing Ppk23glut GRNs had no effect ( Figure 5D ) . This suggests that the aversiveness of Ppk23glut GRN activation is suppressed by salt deprivation . To verify this result , we again turned to PER and found that salt deprivation reduced high salt inhibition of PER and suppressed the role of Ppk23glut GRNs ( Figure 5—figure supplement 2A ) . Interestingly , Gr66a GRN silencing produced only weak effects on PER inhibition by high salt , which were significantly manifested only in the salt deprived state ( Figure 5—figure supplement 2B ) . We next asked whether the observed behavioral modulation by salt deprivation would be evident in the calcium responses of these neurons; however , salt deprivation led to only a very mild and statistically insignificant reduction in Ppk23glut salt responses ( Figure 5—figure supplement 3 ) . Therefore , modulation is likely to occur downstream of GRN output . To further explore this idea , we built a closed-loop system for real-time optogenetic activation of neurons during feeding behavior . Developed as an add-on to the fly Proboscis and Activity Detector ( FlyPAD; [Itskov et al . , 2014] ) , our system triggers illumination of a red LED immediately upon detecting a fly’s interaction with one of the two food sources ( Figure 6A ) . We call this system the Sip-TRiggered Optogenetic Behavior Enclosure ( STROBE ) . The STROBE is similar in concept to another recently described optogenetic FlyPAD ( Steck et al . , 2018 ) , but implements sip detection and light triggering in a different way to minimize latency and achieve illumination during sips , with LED activation tightly locked to sip onset and offset . As expected , sip-induced triggering of Gr64f GRN activation makes a tasteless food source attractive compared to the same food without light stimulation , and this effect is independent of salt deprivation ( Figure 6B ) . Similarly , Gr66a activation is strongly aversive for both salt fed and salt deprived flies . Consistent with their lack of a strong phenotype when silenced , activation of IR94e neurons did not produce a detectable phenotype in either condition . However , stimulating Ppk23glut GRNs is aversive , but only when flies have been pre-fed on a salt-containing diet ( Figure 6B ) . This supports a model where salt need modulates salt avoidance downstream of Ppk23glut GRN activation .
Electrophysiological recordings of individual labellar taste sensilla identified high salt responses in the bitter-sensing neurons of S- and I-type sensilla , and previous GRN calcium imaging confirmed that Gr66a neurons respond to 1 M NaCl and KCl ( Marella et al . , 2006; Meunier et al . , 2003 ) . However , two key results suggested that bitter GRNs did not account for all high salt taste . First , high salt neurons have been identified in L-type sensilla ( which don’t have Gr66a neurons ) via tip recordings , although this has subsequently been debated ( Hiroi et al . , 2002; Ishimoto and Tanimura , 2004; Zhang et al . , 2013 ) . Second , genetically ablating Gr66a GRNs did not block the inhibition of PER by high salt ( Wang et al . , 2004 ) . The existence of Ppk23glut high salt cells likely explains both of these observations and provides a mechanism by which flies can specifically modulate their salt behavior in response to need . In addition to the modulation of their behavioral impact , Ppk23glut neurons display some notable characteristics , the most conspicuous being they are the only GRN class to express a marker for glutamatergic , rather than cholinergic , neurons . This adds a potential new dimension to the gustotopic GRN map formed in the fly brain and may be a key mechanism by which the output of Ppk23glut neurons remains functionally distinct from other GRN classes targeting postsynaptic neurons in the same area . Indeed , the aversive nature of Ppk23glut output stands in contrast to what one would predict from their projection morphology , which looks qualitatively similar to known appetitive ( Gr64f , Ppk28 ) , rather than aversive ( Gr66a ) GRNs . It is possible that Ppk23glut aversiveness is mediated through inhibition of appetitive taste pathways , as glutamate can have excitatory or inhibitory postsynaptic effects , depending on the receptor present ( Liu and Wilson , 2013 ) . Although Ppk23glut defines a novel high salt cell , it is important to note that the Ppk23 channel is not required for its salt responsiveness . This raises questions about what Ppk23-dependent responses these cells may exhibit . Since Ppk23 is required for leg GRN pheromone-evoked activity that regulates courtship , a related function for Ppk23 labellar GRNs cannot be excluded . Indeed , weak , but significant Ppk23-dependent pheromone responses have been observed in labellar GRNs , although it’s unclear whether these were from Ppk23glut or Ppk23chat cells ( Thistle et al . , 2012 ) . Moreover , the interaction between salt taste and mating suggests that perhaps there is a need to co-modulate salt and social cues based on salt diet ( Walker et al . , 2015 ) . In contrast to the strong salt responses in Ppk23glut cells , the other uncharacterized GRN class we identified , IR94e , displayed only weak salt-evoked activity . We therefore expect that this class primarily responds to other , yet unidentified , taste ligands . Given the lack of strong effects we observe upon activation of IR94e GRNs in the STROBE , we also suspect that the behavioral impact of IR94e activation is , like Ppk23glut , state- or context-dependent . To date , IR76b has been shown to be necessary for gustatory responses to low salt , high salt , calcium , acids , amino acids , fatty acids , and polyamines ( Ahn et al . , 2017b; Chen and Amrein , 2017; Hussain et al . , 2016b; Lee et al . , 2017; Zhang et al . , 2013 ) . Consistent with these widespread roles in the taste system , we find expression of IR76b-Gal4 in every GRN type tested . Nonetheless , we felt it important to clarify the role of IR76b in salt taste , given the apparent complexities in salt responses across labellar GRN types , and the previous demonstration that IR76b can function as a sodium leak channel ( Zhang et al . , 2013 ) . We find that Gr64f salt responses are completely dependent on IR76b , consistent with its proposed role in low salt taste . Ppk23glut salt responses also require IR76b , but those in Gr66a GRNs do not , indicating two different salt transduction mechanisms in these two high salt cells . This may explain why prior reports differed on whether high salt responses remain intact in IR76b mutants ( Lee et al . , 2017; Zhang et al . , 2013 ) . Interestingly , the IR76b-dependent salt responses in Ppk23glut GRNs are not sodium specific , as we see loss of high sodium and potassium salt-evoked activity . This suggests that , although IR76b is primarily permeable to sodium when expressed in heterologous cells ( PNa: PK = 1: 0 . 4 ) , it may function in complexes with other subunits that confer different ion selectivity in different GRN classes ( Zhang et al . , 2013 ) . Recently , Ppk23 GRNs were identified as underlying IR76b-dependent calcium taste avoidance ( Lee et al . , 2018 ) . Although it isn’t clear whether Ppk23glut or Ppk23chat ( or both ) subpopulations are responsible , our results indicate that this effect is not specific to calcium , but rather a general salt avoidance mechanism . Indeed , Ppk23 GRNs respond to high concentrations of all salts tested . Moreover , we find that IR25a , which was implicated in Ppk23-mediated calcium taste ( Lee et al . , 2018 ) , is necessary for salt responses in Gr64f and Ppk23glut GRNs , similar to the requirements for IR76b . This stands in contrast to results reported by Zhang et al . ( 2013 ) , which suggested that IR25a did not play a role in sodium taste . The similar requirements for IR76b and IR25a also suggest that these two receptors may act in a complex to mediate salt taste , which is consistent with previous evidence that IR25a is a broadly expressed coreceptor ( Ahn et al . , 2017a; Benton et al . , 2009; Cameron et al . , 2010; Chen and Amrein , 2017; Lee et al . , 2018 ) . Changes in gustatory sensitivity based on internal state are a widespread feature of the fly taste system: starvation potentiates sweet GRN sensitivity and suppresses bitter GRN responses; mating increases taste peg GRN sensitivity to polyamines and behavioral sensitivity to low salt in females; and protein deprivation sensitizes taste peg GRNs to yeast and increases behavioral sensitivity to amino acids ( Hussain et al . , 2016a; Inagaki et al . , 2012; Inagaki et al . , 2014; LeDue et al . , 2016; Steck et al . , 2018; Toshima and Tanimura , 2012; Walker et al . , 2015 ) . Although modulation of salt taste has not been previously examined in flies , salt depletion in humans increases salt palatability ( Beauchamp et al . , 1990 ) . In line with all these results , we observe significant modulation of fly salt taste behavior by salt deprivation . In contrast to most taste modalities , which activate a single GRN population , modulation of salt taste presents a complicated problem , because tuning the gain of Gr64 or Gr66a GRN output would have side effects on sweet and bitter taste sensitivity that may be situationally inappropriate . Here , we have presented evidence that the fly gustatory system solves this problem by specifically modulating the effects downstream of Ppk23glut activation . Salt deprivation suppresses the aversiveness of these neurons , allowing the fly to be less repulsed ( or more attracted ) to salty foods . Thus , the fly taste system appears to encode salt as a complex mixture of attractive and repulsive sensory responses . Two GRN classes – Gr64f and Gr66a – provide a baseline level of attraction or avoidance , and this response is then adjusted to need via modulation of a third class of salt-responsive GRNs , Ppk23glut . The apparent specificity of labellar Ppk23glut GRNs to salt may also provide an important neural substrate for discrimination between salt and other taste modalities . Continued exploration of how salt , and other , taste signals are integrated higher in the brain will provide insight into how an apparently low-dimensional sensory system can successfully encode a variety of diverse chemical cues .
Figure panelGenotypeFigure 1C+/+; vGlutMI04979-Gal4/+; UAS-CD8::tdTomato/+Figure 1D+/+; vGlutMI04979-Gal4/LexAop-CD2::GFP ; UAS-CD8::tdTomato/ChATMI04508- LexA::QFADFigure 1E+/+; vGlutMI04979-Gal4/LexAop-CD2::GFP; UAS-CD8::tdTomato/Ppk28-LexAFigure 1F+/+; vGlutMI04979-Gal4/LexAop-CD2::GFP; UAS-CD8::tdTomato/Gr64fLexAFigure 1G+/+; vGlutMI04979-Gal4/LexAop-CD2::GFP; UAS-CD8::tdTomato/Gr66a-LexAFigure 1H+/+; vGlutMI04979-Gal4/LexAop-CD2::GFP; UAS-CD8::tdTomato/Ppk23-LexAFigure 1I+/+; Gr66a-Gal4/LexAop-CD2::GFP; UAS-CD8::tdTomato/Ppk23-LexAFigure 1J+/+; vGlutMI04979-Gal80/UAS-GCaMP6f; Ppk23-Gal4/+Figure 1KGr66a-LexA/+; LexAop-Gal80/UAS-GCaMP6f; Ppk23-Gal4/+Figure 1L+/+; UAS-CD8::tdTomato /LexAop-CD2::GFP; IR94e-Gal4/ppk28-LexAFigure 1M+/+; UAS-CD8::tdTomato /LexAop-CD2::GFP; IR94e-Gal4/Gr64fLexAFigure 1N+/+; UAS-CD8::tdTomato/LexAop-CD2::GFP; IR94e-Gal4/ppk23-LexAFigure 1—figure supplement 1A+/+; ChATMI04508-Gal80/UAS-GCaMP6f; Ppk23-Gal4/+Figure 1—figure supplement 1B+/+; UAS-CD8::tdTomato/LexAop-CD2::GFP; Ppk23-Gal4/Ppk23-LexAFigure 1—figure supplement 1C+/+; IR76b-Gal4/LexAop-CD2::GFP; UAS-CD8::tdTomato/Ppk28-LexAFigure 1—figure supplement 1D+/+; IR76b-Gal4/LexAop-CD2::GFP; UAS-CD8::tdTomato/Gr64fLexAFigure 1—figure supplement 1E+/+; IR76b-Gal4/LexAop-CD2::GFP; UAS-CD8::tdTomato/Gr66a-LexAFigure 1—figure supplement 1F+/+; IR76b-Gal4/LexAop-CD2::GFP; UAS-CD8::tdTomato/Ppk23-LexAFigure 1—figure supplement 1G+/+; UAS-CsChrimson/+; IR94e-Gal4/+Figure 2C and D+/+; LexAop-GCaMP6f/+; Ppk28-LexA/++/+; UAS-GCaMP6f/Gr64f-Gal4; +/++/+; UAS-GCaMP6f/+; IR94e-Gal4/++/+; UAS-GCaMP6f/Gr66a-Gal4; +/++/+; UAS-GCaMP6f/+; Ppk23-Gal4/+Figure 2—figure supplement 1A–D+/+; UAS-GCaMP6f/+; Ppk23-Gal4/+Figure 2—figure supplement 2+/+; UAS-GCaMP6f/+; IR94e-Gal4/+Figure 3A+/+; UAS-GCaMP6f/+; Ppk23-Gal4/+Figure 3B+/+; UAS-GCaMP6f/+; Ppk23-Gal4/+Gr66a-LexA/+; LexAop-Gal80/UAS-GCaMP6f; Ppk23-Gal4/++/+; vGlutMI04979-Gal80/UAS-GCaMP6f; Ppk23-Gal4/+Figure 3C and DGr66a-LexA/+; LexAop-Gal80/UAS-GCaMP6f; Ppk23-Gal4/++/+; vGlutMI04979-Gal80/UAS-GCaMP6f; Ppk23-Gal4/+Figure 3—figure supplement 1A and B+/+; ChATMI04508-Gal80/UAS-GCaMP6f; Ppk23-Gal4/+Figure 4A and B+/+; Gr64f-Gal4/UAS-GCaMP6f; IR76b2/++/+; Gr64f-Gal4/UAS-GCaMP6f; IR76b1/IR76b2+/+; Gr64f-Gal4 , UAS-GCaMP6f/UAS-IR76b; IR76b1/IR76b2+/+; Gr66a-Gal4/UAS-GCaMP6f; IR76b2/++/+; Gr66a-Gal4/UAS-GCaMP6f; IR76b1/IR76b2+/+; Gr66a-Gal4 , UAS-GCaMP6f/UAS-IR76b; IR76b1/IR76b2+/+; Ppk23-LexA/LexAop-GCaMP6f; IR76b2/++/+; Ppk23-LexA/LexAop-GCaMP6f; IR76b1/IR76b2+/+; UAS-GCaMP6f/UAS-IR76b; IR76b1 , Ppk23-Gal4/IR76b2Figure 4C+/+; Ppk23-LexA/LexAop-GCaMP6f; IR76b2/++/+; Ppk23-LexA/LexAop-GCaMP6f; IR76b1/IR76b2Figure 4D+/+; +/+; IR76b1/++/+; +/+; IR76b2/++/+; +/UAS-IR76b; IR76b1/IR76b2+/+; Gr64f-Gal4/+; IR76b1/IR76b2+/+; Gr64f-Gal4/UAS-IR76b; IR76b1/IR76b2Figure 4E+/+; +/+; IR76b1/++/+; +/+; IR76b2/++/+; +/UAS-IR76b; IR76b1/IR76b2+/+; Gr66a-Gal4/+; IR76b1/IR76b2+/+; Gr66a-Gal4/UAS-IR76b; IR76b1/IR76b2+/+; +/+; Ppk23-Gal4 , IR76b1/IR76b2+/+; +/UAS-IR76b; Ppk23-Gal4 , IR76b1/IR76b2Figure 4—figure supplement 1A and B+/+; Ppk23-LexA/LexAop -GCaMP6f; IR76b2/++/+; Ppk23-LexA/LexAop-GCaMP6f; IR76b1/IR76b2+/+; UAS-GCaMP6f/ UAS-IR76b; IR76b1 , Ppk23-Gal4/IR76b2Figure 4—figure supplement 2+/+; IR25a1/+; Gr64f- Gal4/UAS-GCaMP6f+/+; IR25a1/IR25a2; Gr64f-Gal4/UAS-GCaMP6f+/+; UAS-IR25a , IR25a1/IR25a2; Gr64f-Gal4/UAS-GCaMP6f+/+; IR25a1/+; Gr66a-Gal4/UAS-GCaMP6f+/+; IR25a1/IR25a2; Gr66a-Gal4/UAS-GCaMP6f+/+; UAS-IR25a , IR25a1/IR25a2; Gr66a-Gal4/UAS-GCaMP6f+/+; IR25a1/+; Ppk23-Gal4/UAS-GCaMP6f+/+; IR25a1/IR25a2; Ppk23-Gal4/UAS-GCaMP6f+/+; UAS-IR25a , IR25a1/IR25a2; Ppk23-Gal4/UAS-GCaMP6fFigure 4—figure supplement 3A and BΔPpk23/ΔPpk23; ΔPpk29/ΔPpk29; Ppk23-Gal4/UAS-GCaMP6fFigure 5A+/+; +/+; UAS-Kir2 . 1 , tub-Gal80ts/++/+; Gr64f-Gal4/+; +/++/+; Gr64f-Gal4/+; UAS-Kir2 . 1 , tub-Gal80ts/++/+; +/+; IR94e-Gal4/++/+; +/+; IR94e-Gal4/UAS-Kir2 . 1 , tub-Gal80tsGr66a-LexA/+; LexAop-Gal80/+; Ppk23-Gal4/+Gr66a-LexA/+; LexAop-Gal80/+; Ppk23-Gal4/UAS-Kir2 . 1 , tub-Gal80ts+/+; +/+; Ppk28-Gal4/++/+; +/+; Ppk28-Gal4/ UAS-Kir2 . 1 , tub-Gal80ts+/+; IR76b-Gal4/+; +/++/+; IR76b-Gal4/+; UAS-Kir2 . 1 , tub-Gal80ts/++/+; Gr64f-Gal4/+; IR94e-Gal4/++/+; Gr64f-Gal4/+; IR94e-Gal4/UAS-Kir2 . 1 , tub-Gal80tsFigure 5B+/+; +/+; UAS-Kir2 . 1 , tub-Gal80ts/++/+; Gr64f-Gal4/+; +/++/+; Gr64f-Gal4/+; UAS-Kir2 . 1 , tub-Gal80ts/+Figure 5C+/+; +/+; UAS-Kir2 . 1 , tub-Gal80ts/++/+; Gr66a-Gal4/+; +/++/+; Gr66a-Gal4/+; UAS-Kir2 . 1 , tub-Gal80ts/+Gr66a-LexA/+; LexAop-Gal80/+; Ppk23-Gal4/+Gr66a-LexA/+; LexAop-Gal80/+; Ppk23-Gal4/UAS-Kir2 . 1 , tub-Gal80tsFigure 5D+/+; +/+; UAS-Kir2 . 1 , tub-Gal80ts/++/+; Gr66a-Gal4/+; +/++/+; Gr66a-Gal4/+; UAS-Kir2 . 1 , tub-Gal80ts/+Gr66a-LexA/+; LexAop-Gal80/+; Ppk23-Gal4/+Gr66a-LexA/+; LexAop-Gal80/+; Ppk23-Gal4/UAS-Kir2 . 1 , tub-Gal80tsFigure 5—figure supplement 1+/+; UAS-impTNT/+; +/++/+; UAS-TNT/+; +/++/+; +/+; IR94e-Gal4/++/+; UAS-impTNT/+; IR94e-Gal4/++/+; UAS-TNT/+; IR94e-Gal4/++/+; Gr64f-Gal4/+; IR94e-Gal4/++/+; UAS-impTNT/Gr64f-Gal4; IR94e-Gal4/++/+; UAS-TNT/Gr64f-Gal4; IR94e-Gal4/+Figure 5—figure supplement 2AGr66a-LexA/+; LexAop-Gal80/+; Ppk23-Gal4/UAS-Kir2 . 1 , tub-Gal80tsGr66a-LexA/+; LexAop-Gal80/+; Ppk23-Gal4/++/+; +/+; UAS-Kir2 . 1 , tub-Gal80ts/+Figure 5—figure supplement 2B+/+; Gr66a-Gal4/+; UAS-Kir2 . 1 , tub-Gal80ts/++/+; Gr66a-Gal4/+; +/++/+; +/+; UAS-Kir2 . 1 , tub-Gal80ts/+Figure 5—figure supplement 3+/+; UAS-GCaMP6f/+; Ppk23-Gal4/+ Figure 6B+/+; UAS-CsChrimson/Gr64f-Gal4; +/++/+; UAS-CsChrimson/+; IR94e-Gal4/++/+; UAS-CsChrimson/Gr66a-Gal4; +/+Gr66a-LexA/+; LexAop-Gal80/UAS-CsChrimson; Ppk23-Gal4/+ Flies were raised on standard cornmeal fly food at 25°C in 70% humidity . The following genotypes were used: vGlutMI04979-Gal4 , ChATMI04508-Gal4 , vGlutMI04979-LexA::QFAD , ChATMI04508- LexA::QFAD , vGlutMI04979-Gal80 , ChATMI04508-Gal80 ( Diao et al . , 2015 ) ; Gr66a-LexA , ppk28-LexA , ppk23-Gal4 , UAS-CD8::tdTomato ( Thistle et al . , 2012 ) ; Gr64fLexA ( Miyamoto et al . , 2012 ) ; ppk23-LexA ( Toda et al . , 2012 ) ; IR76b-Gal4 , IR76b1 , IR76b2 , UAS-IR76b ( Zhang et al . , 2013 ) ; IR25a1 , IR25a2 ( Benton et al . , 2009 ) ; UAS-IR25a ( Abuin et al . , 2011 ) ; Gr66a-Gal4 ( Wang et al . , 2004 ) ; Gr64f-Gal4 ( Dahanukar et al . , 2007 ) ; Ppk28-Gal4 ( Cameron et al . , 2010 ) ; LexAop-CD2::GFP ( Lai and Lee , 2006 ) ; UAS-Kir2 . 1 ( Baines et al . , 2001 ) ; tub-Gal80ts ( McGuire et al . , 2004 ) ; IR94e-Gal4 ( Tirián and Dickson , 2017 ) ( Vienna Drosophila Resource Center: v207582 ) ; LexAop-Gal80 ( 32214 ) , LexAop-GCaMP6f ( 44217 ) , UAS-GCaMP6f ( 42747 and 52869 ) , UAS-CsChrimson ( 55135 ) , UAS-TNT ( 28838 ) , UAS-impTNT ( 28840 ) ( Bloomington Stock Center ) . The following tastants were used: Sucrose , NaCl , KCl , NaBr , KBr , CsCl , CaCl2 , Lobeline hydrochloride , Caffeine ( Sigma-Aldrich ) ; 7 , 11-heptacosadiene ( 7 , 11-HC ) , 7 , 11-nonacosadiene ( 7 , 11-NC ) , 7-tricosene ( 7 T ) , and cis-vaccenyl acetate ( c-VA ) ( Cayman Chemical Company , Ann Arbor , MI ) . Tastants were mostly kept as 1 M stocks and diluted as needed . Lobeline hydrochloride was kept as a 1 . 25 mM stock . 7 , 11-heptacosadiene ( 7 , 11-HC ) , 7 , 11-nonacosadiene ( 7 , 11-NC ) , and 7-tricosene ( 7 T ) were diluted in water to desired 0 . 0001 mg/ul . Cis-vaccenyl acetate ( c-VA ) was diluted to stock solution of 0 . 01 mg/ul in EtOH , and then diluted in water . All hydrocarbons stocks were kept at −20°C , diluted as needed , and stored at 4°C for up to seven days . 1% of each Ethanol and Hexanol were diluted in a mix with water and kept at 4°C as control solution for pheromone imaging . Immunofluorescence on labella was carried out as described ( Jeong et al . , 2016 ) . Labella were dissected and fixed for 25 min in 4% paraformaldehyde in PBS + 0 . 2% Triton . After washing with PBS + triton ( 0 . 2%; PBST ) , labella were blocked in 5% NGS diluted with PBST for 40 min . The following primary antibodies were applied and incubated at 4°C overnight: chicken anti-GFP ( 1:1000 , Abcam , Cambridge , UK , #13970 ) and rabbit anti-RFP ( 1:200 , Rockland Immunochemicals , Pottstown , PA , #600-401-379 ) . After washing for 1 hr , the following secondary antibodies were added for 2 hr: goat anti-chicken Alexa 488 ( 1:200 , Abcam #150169 ) and goat anti-rabbit Alexa 647 ( 1:200 , Thermo Fisher Scientific , Waltham , MA , #A21245 ) . Labella were washed again for 40 min , placed on slides in SlowFade gold ( Thermo Fisher Scientific ) , with small #1 coverslips as spacers . Brain immunofluorescence was carried out as described previously ( Chu et al . , 2014 ) . Primary antibodies used were chicken anti-GFP ( 1:1000 , Abcam #13970 ) and mouse anti-brp ( 1:50 , DSHB #nc82 ) . Secondary antibodies used were goat anti-chicken Alexa 488 ( 1:200 , Abcam #150169 ) and goat anti-rabbit Alexa 568 ( 1:200 , Thermo Fisher Scientific #A11036 ) . All images were acquired using a Leica SP5 II Confocal microscope with a 25x water immersion objective . Images were processed in ImageJ ( Schneider et al . , 2012 ) and Adobe Photoshop . To annotate the expression of different markers in the labellum , each sensillum was analyzed in 4 – 8 labella stained for each combination of markers . Confocal z-stacks were examined to identify how many neurons in each sensillum were positive for the different drivers , and which neurons overlapped with the respective co-labelled population . The most common result for each neuron in each sensillum was reported . Sensilla S0 , I0 , I9 , and I10 were the most difficult to score because of viewing difficulties . At times there were duplications of specific sensilla on a labellum , in which case both sensilla were considered . For calcium imaging experiments , female or male flies were aged from 2 to 10 days in groups of both sexes . Females were used for all experiments except where indicated ( pheromones ) . Prior to imaging , flies were briefly anesthetized using CO2 , legs amputated for full access to the proboscis , and placed in custom chamber suspended from their cervix . To ensure immobilization , a small drop of nail polish was applied to the back of the neck and the proboscis was pulled to extension and waxed out on both sides . A modified dental waxer was used to apply wax on each side of the chamber rim , making little contact with the feeding structure . Flies were left to recover in a humidified chamber for 1 hr . The antenna were removed from the fly and a small window of cuticle was removed from the top of the head , exposing the SEZ . Adult Hymolymph Like ( AHL ) buffer was immediately applied to the preparation ( 108 mM NaCl , 5 mM KCl , 4 mM NaHCO3 , 1 mM NaH2PO4 , 5 mM HEPES , 15 mM ribose , pH 7 . 5 ) . The air sacs , fat , and esophagus were clipped and removed to allow clear visualization on the SEZ . Once ready to image , AHL buffer was added that includes Mg2+ and Ca2+ ( 108 mM NaCl , 5 mM KCl , 4 mM NaHCO3 , 1 mM NaH2PO4 , 5 mM HEPES , 15 mM ribose , 2 mM Ca2+ , and 8 . 2 mM Mg2+ ) . GCaMP6f fluorescence was observed using a Leica SP5 II Confocal microscope with a 25x water immersion objective . The relevant area of the SEZ was visualized at a zoom of 4x , a line speed of 8000 Hz , a line accumulation of 2 , and resolution of 512 × 512 pixels . The pinhole was opened to 2 . 98 AU . For each taste stimulation , data was acquired during a baseline of 5 s prior to stimulation , 1 s during tastant application , and 9 s following the stimulation . Tastant stimulations were done using a pulled capillary pipette that was filed down to match the size of the proboscis and fit over all taste sensilla on both labellar palps . The pipette was filled with 1 – 2 μl of a tastant and positioned close to the proboscis labellum . At 5 s a micromanipulator was used to apply the tastant to the labellum manually . Between taste stimulations of differing solutions , the pipette was washed with water . All NaCl solutions were applied in the order of increasing concentration , finishing with 1M KCl . All other solutions were applied in random order to control for potential inhibitory effects between modalities . The maximum change in fluorescence ( ΔF/F ) was calculated using the peak intensity ( average of 3 time points ) minus the average intensity at baseline ( 10 time points ) , divided by the baseline . Quantification of fluorescence changes was performed in ImageJ and graphed in GraphPad Prism6 . For quantification of Ppk23glut projections , the caffeine response for each fly was used to create a region of interest starting below the ‘bitter ring’ and extending across to encompass the lateral projections . This same region of interest was applied to the salt responses of that fly to exclude the Ppk23chat population overlapping with Gr66a in this ‘bitter ring’ . Flies were placed in one of two conditions for 2 – 3 days: 1% agar , 5% sucrose , and 10 mM NaCl ( salt fed ) ; or 1% agar and 5% sucrose ( salt deprived ) . Binary choice preference tests were similar to those previously described ( LeDue et al . , 2015 ) . Female flies aged 2–5 days were sorted into groups of 10 and placed in conditions of either salt feeding or salt deprivation ( see above ) and shifted to 29°C for 48 hr to induce expression of Kir2 . 1 in the cells of interest . For the low salt assay , salt deprived flies were then tested directly . Flies for the high salt assay were subjected to a subsequent 12 hr on medium without sugar ( but with the same salt content ) to increase sugar attraction . For both assays , flies were then transferred into testing vials containing six 10 μL dots of agar that alternated in color . For most low salt attraction assays , the food choices were: 1% agar with both 2 mM sucrose and 50 mM NaCl ( Food 1 ) , and 1% agar with 2 mM sucrose ( Food 2 ) . The experiment in Figure 5B was done without sucrose . For the high salt avoidance assays , the food choices were: 1% agar with both 25 mM sucrose and 250 mM NaCl ( Food 1 ) , and 1% agar with 5 mM sucrose ( Food 2 ) . Each choice contained either 0 . 125 mg/mL blue ( Erioglaucine , FD and C Blue#1 ) or 0 . 5 mg/mL red ( Amaranth , FD and C Red#2 ) dye , and half the replicates for each experiment were done with the dyes swapped to control for any dye preference . Flies were allowed to feed for 2 hr in the dark at 29°C and then frozen and scored for abdomen color . Preference index ( PI ) was calculated as ( ( # of flies labeled with Food 1 color ) – ( # of flies labeled with Food 2 color ) ) / ( total number of flies that fed ) . For PER , 2 – 5 day old females were collected and treated exactly as described above for salt fed and deprived conditions . Flies were then mounted inside pipette tips that were cut to size so that only the head was exposed . The tubes were sealed at the end with tape , positioned on a glass slide with double-sided tape . After a 1 – 2 hr recovery , flies were stimulated with water and allowed to drink until satiated . Each fly was then stimulated on the labellum with increasing concentrations of salt ( 0 mM salt , 250 mM NaCl , 500 mM NaCl , 1 M NaCl , and 1 M KCl ) mixed with 100 mM sucrose using a 20 μL pipette attached to a 1 mL syringe . Stimuli were presented three times each per fly . Four groups of 10 flies for each genotype were tested over four days . The order of genotypes tested on each day was randomized . The STROBE builds on the FlyPAD system’s hardware ( Itskov et al . , 2014 ) by adding a lighting circuit and opaque curtain ( to prevent interference from outside lighting ) to each of 16 FlyPAD arenas . Thus , together , a functioning STROBE system consists of a field programmable gate array ( FPGA ) controller attached to a multiplexor board , adaptor boards , fly arenas equipped with capacitive sensors , and lighting circuits . The arrangement of the STROBE chambers mirrors the design of the FlyPAD , except each of the eight adaptor boards connects to two FlyPAD arenas and two lighting units instead of the original four FlyPAD arenas . These adaptor boards link the chambers to the FPGA , which is a Terasic DEV0-Nano mounted onto a custom multiplexor board with a FTDI module allowing data transfer over serial communications with a computer . The multiplexor board has eight 10-pin ports , each of which connects to an adaptor board that splits the 10-pin line into four 10-pin ports connecting to two fly arenas and two lighting circuits . The fly arena consists of two annulus shaped capacitive sensors and a CAPDAC chip ( AD7150BRMZ ) that the main multiplexer board communicates with to initiate and collect data ( and ultimately stop collecting data ) . The CAPDAC interprets data from the two capacitive sensors on the fly-arena ( Itskov et al . , 2014 ) . The lighting circuit consists of connectors for power from an external power supply and for signaling from the FPGA controller via the intermediate components , a 617 nm light emitting diode ( LUXEON Rebel LED – 127lm @ 700mA; Luxeon Star LEDs #LXM2-PH01-0060 ) , two power resistors ( TE Connectivity Passive Product SMW24R7JT ) for LED current protection , and two metal oxide semiconductor field effect transistors ( MOSFETs; from Infineon 634 Technologies , Neubiberg , Germany , IRLML0060TRPBF ) allowing for voltage signal switching of the LEDs . When the signal from a capacitive sensor rises during a fly sip ( or other food interaction ) , the CAPDAC on the fly arena propagates a signal through the adaptor board via the multiplexor to the FPGA controller . The FPGA processes the capacitive sensor signal using code built atop the original VHDL code from the FlyPAD ( Itskov et al . , 2018 ) . The STROBE VHDL code ( Chan , 2018b ) implements a running minima filter that operates in real-time to detect when a fly is feeding or otherwise interacting with the food . The filter determines the minimum signal value in the last 100 ms and compares the current signal value with this minimum . If the current signal value is greater than the minimum , and the difference between them is greater than a threshold set to exceed noise ( 100 units for all experiments ) , this is considered a rising edge and the filter will prompt the lighting activation system to activate the LED ( or keep it on if it is already on ) . By design , this means that the control system will send a signal to deactivate the lighting upon the falling edge of the capacitance signal , or if the capacitance signal has plateaued for 100 ms , whichever comes sooner . At this point , a low signal is sent to the MOSFET which pinches off the current flowing through the lighting circuit , turning off the light . The signal to lighting response transition times are on the order of tens of milliseconds , providing a nearly instantaneous response . The system automatically records the state of the lighting activation system ( on/off ) and transmits this information through USB to the PC , where it is received and interpreted by a custom end-user program ( built using Qt framework in C++ ) which can display both the activation state and signal measured by the STROBE system in each fly arena in real-time . All STROBE software is available for download from Github: FPGA code: https://github . com/rcwchan/STROBE-fpga All other code: https://github . com/rcwchan/STROBE_software/ Flies were place in vials for three days under ‘salt fed’ or ‘salt deprived’ conditions described above . All flies were 5 – 9 days old at the time of the assay . For retinal groups , food was supplemented with all trans-Retinal at a final concentration of 1 mM ( Sigma-Aldrich ) . Both channels of STROBE chambers were loaded with 4 μl of 1% agar ( GR64f and IR94e experiments ) or 1M sucrose mixed in 1% agar ( Gr66a and Ppk23glut experiments ) . Acquisition on the STROBE software was started and then single flies were transferred into each arena by mouth aspiration . Experiments were run for 60 min , and the preference index for each fly was calculated as: ( sips from Food 1 – sips from Food 2 ) / ( sips from Food 1 + sips from Food 2 ) . Statistical tests were performed using GraphPad Prism six software . Descriptions and results of each test are provided in the figure legends . Sample sizes are indicated in the figure legends . Sample sizes were determined prior to experimentation based on the variance and effect sizes seen in prior experiments of similar types . Whenever possible , all experimental conditions were run in parallel and therefore have the same or similar sample sizes . However , in some cases this was impossible , due to the concurrent availability of different genotypes or the size of the experiment . These situations account for instances where some control genotypes have very large sample sizes , since they were run in parallel with multiple experimental groups ( e . g . Figure 5A ) . All replicates were biological replicates using different flies . Data for all quantitative experiments were collected on at least three different days , and behavioral experiments were performed with flies from at least two independent crosses . Specific definitions of replicates are as follows . For calcium imaging , each data point represents the response of a single fly to the indicated stimulus . A given fly was stimulated with a specific tastant only once . For binary choice behavioral tests , each data point represents the calculated preference for a group of 10 flies . For PER , each replicate is composed of 10 independent flies tested in parallel . For STROBE experiments , each data point is the calculated preference of an individual fly over the course of the experiment . Outliers were occasionally observed but were not removed from the datasets . For example , in Figure 2D , two flies had strong water responses in Gr64f sweet neurons . Although these appear to be outliers , we left them in the dataset because there was no other justification for removing them . There were two conditions where data were excluded that were determined prior to experimentation and applied uniformly throughout . First , in calcium imaging experiments , all the data from a fly were removed if either: a ) there was too much movement during stimulation to reliably quantify the response; or b ) there was no response to a known , robust , positive control ( rare ) . Second , for STROBE experiments , the data from individual flies were removed if the fly did not pass a set minimum threshold of sips ( 15 ) , or the data showed hallmarks of a technical malfunction ( rare ) . A third condition for data exclusion arose during pilot experiments and was then applied subsequently: A subset of flies expressing GCaMP in IR94e neurons ( ~20% ) showed a large response to water alone . These flies were removed from the analysis . All the quantitative data used for statistical tests can be found as supplements for each figure . | Salt is essential for our survival , but too much can kill us . Our taste system has therefore evolved two different pathways to help us maintain balance . Low concentrations ( like the salt on our chips ) activate a pathway that makes us want to eat . But high concentrations ( like the salt in seawater ) activate pathways that do the opposite . The nervous system takes on the role of detecting salt and encoding the information in a way that the brain can use . One specific type of cell detects each of the four other tastes: sweet , bitter , sour , and umami . But salt , with its two sensing pathways , is the exception to this rule . Previous work has examined salt taste responses in flies , but the picture is incomplete . In flies , one type of taste neuron uses a different signaling mechanism to the others , suggesting that it might play a special role . So here , Jaeger , Stanley et al . asked how fly sensory cells encode salt information for the brain , and what those unusual neurons are for . Mapping the taste receptor neurons in the tongue-like structure of the fly , the proboscis , revealed that salt information is not restricted to one or two types of cell . In fact , all five types of neurons tested ( covering more than 90% of all the taste neurons present in flies ) responded to salt in some way . Of these , two ‘low salt’ cell types made the fly want to eat salt , and two ‘high salt’ cell types made the fly want to avoid it . One of these high salt cell types was the unusual taste neuron identified previously . Rather than always encoding high salt as 'bad' , the message from this type of cell changed depending on the diet of the fly . Salt-deprived flies ignored the activity of that cell type altogether . This complex way of encoding taste allowed the fly to change its behavior depending on how much salt it needed . This work opens new questions , like how do the fly's neuronal circuits process this complex salt code ? And how do the ‘high salt’ cells achieve their negative effect only when the need for salt is low ? Understanding more about this system could lead to a better understanding of why our own brains enjoy salty foods so much . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | A complex peripheral code for salt taste in Drosophila |
Understanding the mechanisms regulating development requires a quantitative characterization of cell divisions , rearrangements , cell size and shape changes , and apoptoses . We developed a multiscale formalism that relates the characterizations of each cell process to tissue growth and morphogenesis . Having validated the formalism on computer simulations , we quantified separately all morphogenetic events in the Drosophila dorsal thorax and wing pupal epithelia to obtain comprehensive statistical maps linking cell and tissue scale dynamics . While globally cell shape changes , rearrangements and divisions all significantly participate in tissue morphogenesis , locally , their relative participations display major variations in space and time . By blocking division we analyzed the impact of division on rearrangements , cell shape changes and tissue morphogenesis . Finally , by combining the formalism with mechanical stress measurement , we evidenced unexpected interplays between patterns of tissue elongation , cell division and stress . Our formalism provides a novel and rigorous approach to uncover mechanisms governing tissue development .
The advances in live imaging have beautifully illustrated how the growth and shaping of tissues and organs emerge from the collective dynamics of cells ( for review see [Keller , 2013] ) . In this context , the development of quantitative methods is essential to determine the role in tissue and organ development of each cell process , in particular cell divisions , cell rearrangements , cell size and shape changes and apoptoses ( Rauzi et al . , 2008; Blanchard et al . , 2009; Aigouy et al . , 2010; Bosveld et al . , 2012; Tomer et al . , 2012; Krzic et al . , 2012; Economou et al . , 2013; Heller et al . , 2014; Khan et al . , 2014; Zulueta-Coarasa et al . , 2014; Monier et al . , 2015; Lau et al . , 2015; Rozbicki et al . , 2015; Etournay et al . , 2015 ) . However , a general formalism valid in two and three dimensions and allowing to unambiguously decompose tissue growth and morphogenesis into the parts due to each cell process is still needed . Building such a formalism is critical to define the roles of each process and advance our understanding of how gene activities and mechanical forces cooperate in controlling cell dynamics to regulate the growth , shaping , repair or homeostasis of monolayered or three-dimensional cohesive tissues . In particular the lack of a general formalism has impeded a comprehensive characterization of the role of cell divisions and of their orientations during tissue development . The growth and morphogenesis of tissues typically involves cell divisions leading to the formation of organs of the correct size and shape . So far , two fundamental properties of cell division have been reported during the morphogenesis of proliferative tissues . First , the orientation of cell divisions has been shown to play a key role in tissue elongation , either by being an anisotropic source of force , or by reducing mechanical stress ( Baena-López et al . , 2005; Saburi et al . , 2008; Aigouy et al . , 2010; Quesada-Hernández et al . , 2010; Ranft et al . , 2010; Mao et al . , 2011; Gibson et al . , 2011; Aliee et al . , 2012; Mao et al . , 2013; Legoff et al . , 2013; Campinho et al . , 2013; Wyatt et al . , 2015 ) . Second , cell division orientation has been reported to be mainly oriented along the direction of mechanical stress ( Fink et al . , 2011; Mao et al . , 2013; Legoff et al . , 2013; Campinho et al . , 2013; Wyatt et al . , 2015 ) . These conclusions were mainly drawn from the observation of tissues where cell divisions are the major contributor of tissue elongation or where cell divisions , tissue deformation and mechanical stress display a common preferred orientation . Yet , embryos , as well as many other tissues or organs , are heterogeneous: cell divisions display rates and orientations varying in space and time , and they are concomitant to other morphogenetic processes such as cell rearrangements , cell size and shape changes and apoptoses . One of the broad challenges in developmental biology is therefore to test and generalize these proposed functions of cell divisions in more heterogeneous contexts .
We have previously proposed a statistical method to quantify cell rearrangements and cell size and shape changes during tissue development ( Bosveld et al . , 2012; Bardet et al . , 2013 ) . The method took advantage of the links joining the centroids of a cell and its neighbors ( Graner et al . , 2008 ) . Measuring the changes of position , size and direction of links , as well as link swapping , yielded a quantification of cell rearrangements and cell size and shape changes characterized by an amplitude , an anisotropy and an orientation . This enabled to separately measure cell rearrangements and cell size and shape changes during tissue development . Generalizing the method to incorporate the remaining biological cell processes , in particular cell divisions and apoptoses , is not straightforward since these cell processes can substantially modify the cell and link numbers . We therefore developed a novel formalism that takes into account the changes in link number and which disentangles the measurement of each cell process during tissue development ( See Appendix for detailed information and comparison with previous approaches ) . In this novel formalism , valid in two and three dimensions , the four main cell processes ( cell divisions , cell rearrangements , cell shape and size changes and apoptoses ) are unambiguously distinguished and independently quantified by four measurements ( D , R , S and A , respectively ) . These four measurements quantify the changes in link length or orientation as well as link appearances or disappearances due to their respective cell processes; they add up to the local tissue rate of deformation measuring the rate of tissue growth and morphogenesis ( G ) due to these four cell processes ( Figure 1a ) . Whenever needed , the subdivision of these main cell processes can be further refined , for instance in a mono-layered tissue to distinguish apoptoses from live cell extrusions , or to distinguish simple rearrangements through four-fold vertices from those with five-fold vertices or higher . Furthermore , introducing cell divisions and apoptoses naturally enables the addition of the other cell processes changing the number of cells such as: ( i ) the integration of new cells in epithelium sheets ( N ) ; ( ii ) the fusion ( coalescence ) of cells ( C ) , and ( iii ) the in/outward cell flux ( J ) , representing the cells entering and exiting the microscope field of view or the boundaries of the tissue of interest ( Figure 1—figure supplement 1a ) . The formalism therefore yields a complete and unified quantitative characterization of tissue deformation and of all cell processes reported to occur during epithelial tissue development and homeostasis . 10 . 7554/eLife . 08519 . 003Figure 1 . Definitions of the main formalism quantities and analysis workflow . ( a ) Characterizations of the four main elementary cell processes and of tissue deformation: D divisions ( green; and dark green for the link created between the daughter cells ) ; R , rearrangements ( magenta ) ; S , size and shape changes ( cyan ) ; A , apoptosis/delaminations ( black ) . They are defined and measured from the rates of changes in length , direction and number of cell-cell links , here on two schematized successive images . They make up the tissue deformation rate G , the measurement of which is based on geometric changes of conserved links ( dark blue links ) excluding non-conserved links ( green ) . Dots indicate cell centroids . Lines are links between neighbor cell centroids . Dashes are links on the first image ( left ) which are no longer present on the second one ( right ) . Some cells are hatched in grey to facilitate the comparison . ( b ) Measurements of the four elementary main cell processes rates and of tissue deformation rate . Same as ( a ) , this time showing cell-cell links on two actual successive segmented images extracted from experimental time-lapse movies . ( c ) Representation with circles and bars of the quantitative measurements performed on ( b ) of the deformation rates explained in ( d ) . ( d ) Deformation rate: a deformation quantifies a relative change in tissue dimensions: it is expressed without unit , e . g . as percents . A deformation rate is thus expressed as the inverse of a time , e . g . 10-2 h-1 represents a 1% change in dimension within one hour . It can be decomposed into two parts . First ( left ) : an isotropic part that relates to local changes in size . The isotropic part can either be positive or negative , reflecting a local isotropic growth or shrinkage of the tissue . The rate of dilation is represented by a circle , the diameter of which scales with the magnitude of the rate . Positive and negative dilations are represented by circles filled with white and grey , respectively . Second ( right ) : an anisotropic part that relates to local changes in shape . The anisotropic part of the deformation rate quantifies the local contraction-elongation or convergence-extension ( CE ) without change in size . It can be represented by a bar in the direction of the elongation , the length and direction of which quantify the magnitude and the orientation of the elongation . ( e ) Workflow used to quantify tissue development . Image analysis leads to characterization of cell contours ( segmentation ) , and lineages ( tracking ) in the case of movies . Our formalism yields an identification of each cell-level process and its description in terms of cell-cell links ( see a–b ) and a quantitative measurement of their associated deformation rate ( see c–d ) . Averaging over time , space and/or movies of different animals yields a map of each quantity in each region of space at each time with a good signal-to-noise ratio ( see Videos 1 , 4 , 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 00310 . 7554/eLife . 08519 . 004Figure 1—figure supplement 1 . Characterizations of the additional elementary cell processes N , C , J . ( a ) The new cell integration N ( purple ) , cell fusion C ( crimson ) , cell flux through tissue boundaries J ( grey ) on two schematized successive images . Dots indicate cell centroids . Lines are links between neighbor cell centroids . Dashes are links on the first image ( left ) which are no longer present on the second one ( right ) . Some cells are hatched in grey to facilitate the comparison . ( b ) Measurements of the three additional cell processes rates . Same as ( a ) , this time showing cell-cell links on two actual successive segmented images extracted from experimental time-lapse movies . J is defined through links which cross the boundary of the field of view . Dark grey cells are boundary cells , partly out of the field of view , and their centroids are not defined . Light grey cells touch a boundary cell : their links with dark grey cells are ill-defined and are therefore excluded from calculations . ( c ) Representation with circles and bars of the quantitative measurements performed on ( b ) of the deformation rates explained in Figure 1d . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 004 In a tissue where tissue deformation is solely associated with cell divisions , cell rearrangements , cell size and shape changes and apoptoses , this unified characterization is expressed as a balance equation where the deformation rate of a region in the tissue is decomposed into the sum of the deformation rates associated with each cell process: ( 1 ) G=D+R+S+A Note that in the absence of divisions , rearrangements and apoptoses ( i . e . D=R=A=0 ) , our formalism therefore yields an exact equality between the rates of tissue deformation G and of cell size and shape changes S . Here , we explicitly describe its practical implementation and measurements in the context of the general interest in understanding the growth and morphogenesis of epithelial sheets ( Heisenberg and Bellaïche , 2013; Guillot and Lecuit , 2013 ) . We first acquire a time-lapse movie in which cell apical contours have been labeled by E-Cadherin:GFP , images are segmented and cells tracked to determine their positions over time and their lineages . Then , the formalism is applied between successive images to separately measure the growth and morphogenesis associated to each cell process ( D , R , S and A ) and of the tissue ( G ) ( Figure 1b and Figure 1—figure supplement 1b ) . Each measurement can be represented with a circle and a bar ( Figure 1c and Figure 1—figure supplement 1c ) . The circle diameter represents the local rate of tissue isotropic dilation or tissue growth: it is positive for an increase in size ( white filled circle ) and negative for a decrease ( grey filled circle ) . The bar , which has a length and orientation , represents the local rate of tissue anisotropic deformation or tissue morphogenesis: it quantifies the local elongation rate ( and respective equal contraction rate in the orthogonal direction ) , thereafter named the contraction-elongation ( CE ) rate ( Figure 1d ) . Finally , the analysis is multi-scale , in the sense that each statistical measurement can be averaged at any supra-cellular scale over space , over time , and over several animals , thereby linking the length and time scales associated with cells , groups of cells and the whole tissue ( Figure 1e , Video 1 ) . 10 . 7554/eLife . 08519 . 005Video 1 . Workflow of measurements of tissue and cell process CE rates at the patch scale . ( a ) Detail of a movie in the scutellum region , tissue labeled with E-Cad:GFP and imaged by multi-position confocal microscopy at a 5 min time resolution , 11:25 to 27:25 hAPF . ( b ) Cell tracking: cells are colored in shades of green according to their number of divisions: light ( 1 ) , medium ( 2 ) , dark ( ≥3 ) ; black for the last five frames before a delamination; red for fused cells . ( c ) Evolution of cell-cell links: links which appear or disappear are represented with thick straight lines and colored as follows: divisions ( green ) , rearrangements ( magenta ) , delaminations ( black ) , integrations ( purple ) , fusions ( red ) , boundary flux ( orange ) . Conserved links are represented with thin dark blue lines . Links corresponding to four-fold vertices are in lighter colors . Cell contours are indicated by thin grey outlines , patch contours by thick black outlines . ( d-h ) Maps of dilation rates ( circles filled in white [positive] or grey [negative] ) and of CE rates ( orientation: bar direction; anisotropy: bar length ) , for ( d ) the tissue G ( compare bar amplitudes and orientations with the evolution of patch shapes ) , ( e ) cell divisions D , ( f ) rearrangements R , ( g ) cell size and shape changes S and ( h ) delaminations A . Patch contours are indicated by thick grey outlines . Dilation and CE rates in a given patch are calculated from the evolution of links in this patch between two successive images , then averaged with a sliding window of 2 h . The stillness at the beginning and end of the measurement movies comes from this time averaging . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 005 In order to confirm that each measurement quantifies unambiguously and accurately its associated biological process , we applied the formalism on computer simulations of cell patches undergoing known deformations . In each simulation , we imposed that the growth and morphogenesis of the cell patch was mainly driven by only one of the cell processes: cell divisions , cell rearrangements , cell size and shape changes or apoptoses . We first tested the measurements of G and S by imposing an isotropic dilation of the cell patch , followed by its CE along the horizontal axis , both patch deformations solely occurring via cell size and shape changes . We independently measured the imposed deformation rates for G and S with 0 . 3% of error , and obtained G=S as expected ( Figure 2a , Video 2a ) . Next , we tested the measurements of D , R , A by allowing deformation of the cell patch by oriented cell divisions , oriented rearrangements and apoptoses , respectively . In each simulation , the balance equation shows that the tissue deformation rate G was determined by the main process enabling the deformation of the cell patch ( Figure 2b–d , Video 2b–d; see Figure 2—figure supplement 1 and Video 2e–i for the others processes ) . This confirmed that the formalism unambiguously measures the tissue deformation rate as well as the deformation rates associated with each individual cell process . 10 . 7554/eLife . 08519 . 006Figure 2 . Computer simulations validating the quantitative characterizations of the main cell processes and tissue deformation . In ( a–d ) , upper panels: simulated deformation of a cell patch; left: initial state of the simulation; middle: intermediate state; right: final state . Lower panels: Equation 1 is visually displayed . ( a ) By direct image manipulation ( hence not followed by any cell shape relaxation ) , the initial pattern ( left ) is dilated ( middle ) then stretched ( right ) with known dilation and CE rates , thereby solely generating the same size and shape changes for the patch and for each individual cell . The patch deformation rate G and the cell size and shape change rate S are measured independantly with 0 . 3% of error , and , as expected when no topological changes occur , we find G=S . This validates the measurement of G and S , which in turn validates the other measurements in the next simulations . ( b ) Potts model simulation of oriented cell divisions . Forces are numerically implemented along the horizontal axis . They drive the elongation of the cell patch while each cell divides once along the same axis . Therefore both G and D have their anisotropic parts along the horizontal direction . The residual cell rearrangements and cell shape changes CE rates R and S are respectively due to some cell rearrangements actually occurring in the simulation , and to some cells having not completely relaxed to their initial sizes and shapes . This is not due to any entanglement between the cell process measurements in the formalism . Divided cells are in green . ( c ) Potts model simulation of oriented cell rearrangements . The same forces as in ( b ) drive the elongation of the cell patch first leading to the elongation of cells that then relax their shape by undergoing oriented rearrangements along the same axis , thereby leading to both G and R having their anisotropic parts along the horizontal direction . The cell shape relaxation is not complete as cells remain slightly elongated by the end of the simulation ( right ) , thereby giving a residual S . ( d ) Potts model simulation of cell delaminations . Delaminations were obtained by gradually decreasing the cell target areas of half the cells of the initial patch to 0 , thereby driving the isotropic shrinkage of the patch to half of its initial size . It leads to equal negative growth rates for G and A , up to residual other processes . Delaminating cells are in black . The white scale bar and circle in ( d ) both correspond respectively to CE and growth rates of 10-2 h-1 for simulation movies lasting 20 h , in all panels ( a–d ) . Only measurements with norm > 10-3 h-1 have been plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 00610 . 7554/eLife . 08519 . 007Figure 2—figure supplement 1 . Computer simulations validating the quantitative characterization of the additional cell processes N , C , J and testing rotation . In ( a–c ) , upper panels , simulated deformation of a cell patch . Left: initial state of the simulation; middle: intermediate state; right: final state . Lower panels: Equation 15 is visually displayed . ( a ) Simulation of patch growth via new cell integration N , produced by time-reversal of the delamination simulation ( Figure 2d ) , thereby leading to equal positive growth rates for G and N , up to residual other processes . ( b ) Simulation of patch undergoing cell fusion C , produced by the random removal of cell-cell junctions . In this particular example , the patch undergoes no deformation at all , while the removal of junctions lead to an artificial increase of average cell size measured by a positive cell growth rate S , and an opposite fusion rate C that cancels it , thus leading to G=0 . This validates C measurement since S has been validated in Figure 2a . ( c ) Simulation of cell outward flux J . Outer layers of cells of the patch are progressively removed , and there is again virtually no morphogenesis like in ( b ) . The small cell size and shape changes measured is due to the smaller number of cells over which it is averaged and is completely compensated by the flux term J , leading to G=0 . ( d ) By direct image manipulation using an image treatment software , an initially elongated pattern ( left , identical to the final pattern of Figure 2a ) is rotated anticlockwise by 90° ( right ) . This validates that for rigid body movements such as a rotation , the formalism does not detect any significant CE rates , as expected . ( e ) For comparison with ( d ) , by direct image manipulation using an image treatment software , the same initial elongated pattern of ( d ) is now brought to a round pattern ( middle ) by a convergence-extension , and is then stretched again by the same convergence-extension , which leads to its elongation in the perpendicular direction , resulting in a final pattern very similar to the one in ( d ) , with same aspect ratio ( right ) . This illustrates that for such a pure CE without rotation , the formalism does detect significant CE rates for G and S as expected , although the initial and final states are very similar to ( d ) . This also illustrates that all our measurements depend on the deformation path between the initial and final states . The white scale bar in ( e ) is equivalent to: ( a , d ) 10-2 h-1 , ( c ) 0 . 1 10-2 h-1 , ( b , e ) 2 10-2 h-1 . Only measurements with norm > 10-3 h-1 have been plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 00710 . 7554/eLife . 08519 . 008Video 2 . Movies of computer simulations corresponding to ( a-d ) Figure 2a–d and to ( e–i ) Figure 2—figure supplement 1a–e . ( a ) Cell size and shape changes S , ( b ) oriented cell divisions D , ( c ) oriented cell rearrangements R , ( d ) delaminations A , ( e ) integrations of new cells N , ( f ) fusions of two or more cells C , ( g ) cell flux J , ( h ) rotation , ( i ) convergence-extension with initial and final states similar to those in ( h ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 008 Having validated the formalism in silico , we illustrate its relevance to study tissue development by undertaking an analysis of the role of cell division orientation and its relationship to other cell processes and to mechanical stress during the development of a heterogeneous epithelial tissue . During pupal metamorphosis , the Drosophila dorsal thorax ( notum , yellow dashed box in Figure 3a , b ) is a monolayered cuboidal epithelial tissue . From 10 h after pupa formation ( hAPF ) , it undergoes several morphogenetic movements associated with cell divisions , cell rearrangements and cell size and shape changes as well as delaminations , which can be due to live cell extrusions or apoptoses ( Bosveld et al . , 2012; Marinari et al . , 2012 ) . An important feature of this tissue is its heterogeneity , which enables to simultaneously investigate the various mechanisms driving morphogenesis and their interplays . Furthermore , applying our formalism on this tissue will provide a valuable resource since it is a general model to uncover conserved mechanisms that regulate planar cell polarization , tissue morphogenesis , tissue homeostasis and tissue mechanics , and to perform genome-wide RNAi screen ( see for example [Mummery-Widmer et al . , 2009; Olguín et al . , 2011; Bosveld et al . , 2012; Marinari et al . , 2012; Antunes et al . , 2013] ) . 10 . 7554/eLife . 08519 . 009Figure 3 . Quantitative characterization of tissue morphogenesis of the whole Drosophila notum . ( a ) Drosophila adult fly . Yellow dotted box is the notum , circles filled in yellow are macrochaetae . ( b ) Drosophila pupa . Yellow dotted box is the region that was filmed . Black dotted line is the midline ( along the anterio-posterior direction , mirror symmetry line for the medial-lateral axis ) . ( c ) Rate of cell divisions obtained from cell tracking . Number of cell divisions color-coded on the last image of the movie ( 28 hAPF ) , light green cell: one division; medium green cell: two divisions; dark green cells: three divisions and more; purple cells: cells entering the field of view during the movie . Circles filled in yellow indicate macrochaetae . The other white cells are microchaetae . ( d ) Growth and morphogenesis of cell patches during notum development . ( Left ) Cell contours ( thin grey outlines ) at 14 hAPF . A grid made of square regions of 40 μm sides was overlayed on the notum to define patches of cells whose centers initially lied withing each region ( ∼30 cells per patch in the initial image ) . Within each patch , all cells ( and their future offspring ) were assigned a given color . The assignement of patch colors was arbitrary but nevertheless respected the symmetry with respect to the midline ( in cyan ) to make easier the pairwise comparison of patches . Each patch was then tracked as it deformed over time to visualize tissue deformations at the patch scale . ( Right ) Cell contours at 28 hAPF . The variety of patch shapes reveals the heterogeneity of deformations at the tissue scale , as well as their striking symmetry with respect to the midline . ( e ) Map of average cell division orientation ( bar direction ) and anisotropy ( bar length ) , Do ( Appendix C . 3 . 2 ) . Its determination is solely based on the links between newly appeared sister cells ( link in dark green in Figure 1a , b ) . ( f–j ) Maps of orientation ( bar direction ) and anisotropy ( bar length ) of CE rates , for ( f ) the tissue G ( compare the bar amplitude and orientation pattern with the pattern of patches in ( d ) right ) , ( g ) cell divisions D , ( h ) cell rearrangements R , ( i ) cell shape changes S and ( j ) delaminations A . In this Figure ( and Figure 3—figure supplements 1 and 2 ) , measurements over the whole notum have been averaged over 14 h of development ( between 14 and 28 hAPF ) and plotted on the last image of the movie ( for their time-evolution see Video 4 ) ; contours of cells ( thin grey outlines ) and of initially square patches ( thick grey outlines ) ; black boxes outline the posterior regions ( medial and lateral ) described in the text; patches near the tissue boundary contain less data and are plotted accordingly with higher transparency; circles filled in yellow indicate macrochaetae; dashed black line is the midline . Scale bars: ( a , b ) 250 μm , ( c , d ) 50 μm , ( e–j ) 2 10-2 h-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 00910 . 7554/eLife . 08519 . 010Figure 3—figure supplement 1 . Complete set of maps of dilation rates ( isotropic parts of measurements ) . The additivity ( Equation 15 ) also applies separately to these isotropic parts . Circle diameters are proportional to the traces of: ( a ) tissue dilation G , ( b ) cell divisions D , ( c ) cell rearrangements R , ( d ) cell size changes S , ( e ) delaminations A , ( f ) new cell integrations N , ( g ) fusions C , ( h ) boundary flux J . Scale circle diameters: 2 10-2 h-1 . Note that the isotropic part of cell divisions D is always positive and that of delaminations A is always negative , as expected . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 01010 . 7554/eLife . 08519 . 011Figure 3—figure supplement 2 . Complete set of maps of contraction-elongation ( CE ) rates ( anisotropic parts of measurements ) . The additivity ( Equation 15 ) also applies separately to these anisotropic parts . Panels already presented in the Figure 3f–j are replotted here ( a–e ) for comparison with additional measurements ( f–h ) : ( f ) new cell integrations N , ( g ) fusions C , ( h ) boundary flux J . Scale bars: 2 10-2 h-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 011 We imaged the development of this tissue by labeling cell adherens junctions with E-Cadherin:GFP and followed ~10 103 cells over several cell cycles with 5 min resolution from at least 14 to 28 hAPF . We segmented and tracked the cells of the whole movie ( ~3 106 cell contours with a relative error below 10-4 , Figure 3c , Video 3a ) . The display of cell displacements , as well as the tracking of cell patches deforming over time enable to visualize the heterogeneity of tissue growth and morphogenesis between 14 and 28 hAPF ( Figure 3d , Video 3b , c ) . Directly measuring the rate and orientation of cell divisions , we found that ~17 103 divisions take place during the development of the tissue , and that both the cell division rates and orientations display major variations in space and time ( Figure 3c , e ) . Cell division rate is higher in the posterior part of the tissue ( Figure 3c ) and many regions harbor oriented cell divisions ( Figure 3e ) . Division orientation is represented by a bar ( Do , Appendix C . 3 . 2 ) , the length and orientation of which represent respectively the cell division orientation anisotropy and main orientation in each region . In particular , in the central posterior part of the tissue the orientation of cell divisions is medial-lateral , while in a more anterior and lateral domain cell divisions are oriented at roughly 45° relative to the anterior-posterior axis ( Figure 3e , boxes ) . While these descriptions of tissue development by following patches of cells , cell division rate and orientation are essential , we now explain how the formalism enables to rigorously tackle three major steps to quantitatively study the morphogenesis of an epithelial tissue: ( i ) measure the local CE rates associated with each process for one animal; ( ii ) determine the average and the variability of cell dynamics over several animals; and ( iii ) measure the components of each cell process CE rate along tissue morphogenesis . 10 . 7554/eLife . 08519 . 012Video 3 . Movies of cell tracking , cell trajectories and patch deformation in the whole Drosophila notum . ( a ) Cell tracking displaying divisions ( see Figure 3c ) : cells are color coded in light green for the first division , medium green for two divisions , dark green for three divisions and more; black for the last five frames before a delamination; red for cells that fuse; grey for the first layer of boundary cells; purple for new cells . ( b ) Cell trajectories: as it moves , each cell leaves a trail corresponding to the successive positions of its center in the last 10 images . Same color code as in ( a ) . ( c ) Growth and morphogenesis of cell patches during development ( see Figure 3d ) . Cell contours are indicated by thin grey outlines , cells within a patch have same arbitrarily assigned colors that respect the symmetry with respect to the midline . Movie between 11:30 and 30:45 hAPF . Note that the original movie of the E-Cad:GFP tissue is visible in Supplementary Video 1 of ( Bosveld et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 012 We found that in the wild-type tissue cell divisions display various orientations with respect to the direction of tissue elongation and thus can have negative and positive components along tissue morphogenesis . These observations raise important questions regarding the role of cell divisions per se in tissue development , namely the role of proliferation and division orientation , as well as its interplay with the other cell processes during tissue morphogenesis . We illustrate here how the formalism can help analyze these central questions by allowing for a rigorous quantification of different experimental conditions . To experimentally study the role of cell divisions in morphogenesis , we overexpressed the tribbles gene ( trblup ) , an inhibitor of G2/M transition ( Grosshans and Wieschaus , 2000; Mata et al . , 2000; Seher and Leptin , 2000 ) using the Gal4/Gal80ts system to inhibit proliferation specifically at pupal stage ( McGuire et al . , 2003; Bosveld et al . , 2012 ) . We segmented and tracked five trblup hemi-thoraxes over time ( corresponding to ~3 . 7 106 cell contours ) . Both visual inspection of the movie and cell tracking revealed that a trblup hemi-notum hardly displays any division as compared to wild-type tissue: ~1 . 7 103 , i . e . only 4 . 3% of the number of wild-type divisions ( Figure 6a , b ) . We then registered and rescaled in space the five hemi-thoraxes , synchronized them in time , and applied the formalism to determine tissue morphogenesis and the respective CE rates of each cell process ( Figure 6d , f and Figure 6—figure supplement 1b , d ) . As expected , the measured division CE rate D nearly vanishes in accordance with the nearly complete disappearance of cell proliferation ( Figure 6c , d ) . Furthermore , we find that tissue CE rate G is disrupted in trblup tissue , both in direction and amplitude , suggesting that the absence of proliferation impacts tissue morphogenesis ( Figure 6e , f ) . 10 . 7554/eLife . 08519 . 019Figure 6 . Quantitative characterization of tissue development in trblup mutant notum . ( a , b ) Comparison between rate of cell divisions in ( a ) wild-type ( extracted from half of Figure 3c ) and ( b ) trblup tissues . Number of cell divisions color-coded on the last image of the movie ( 28 hAPF ) , light green cell: one division; medium green cell: two divisions; dark green cells: three divisions and more; purple cells: cells entering the field of view during the movie . ( c–j ) Comparisons of CE rates in wild-type and trblup mutant tissues . Time averages were performed between 14 and 28 hAPF . In this Figure ( and Figure 6—figure supplement 1 ) , values larger than the local biological variability are plotted in color while smaller ones are shown in grey; a local transparency is applied to weight the CE rate according to the number of cells and hemi-nota in each group of cells; black outline delineates the archetype hemi-notum; the midline is the top boundary; circles filled in yellow indicate archetype macrochaetae . ( c , d ) Comparison between cell division CE rate ( D , green ) in ( c ) wild-type ( extracted from Figure 4b ) and ( d ) trblup tissues . ( e , f ) Comparison between tissue CE rate ( G , dark blue ) in ( e ) wild-type ( reproduced from Figure 4a ) and ( f ) trblup tissues . ( g–j ) Subtraction of measurements in trblup tissue minus measurements in wild-type tissue , for ( g ) tissue CE rate ( ΔG , dark blue ) , and for components along wild-type tissue CE rate of ( h ) tissue CE rate ( ΔG// , dark blue ) , ( i ) cell rearrangements CE rate ( ΔR// , magenta ) and ( j ) cell shape changes CE rate ( ΔS// , cyan ) . ( k–m ) Simulations illustrating the impact of cell divisions on tissue elongation and on the other processes . Top: last image of Potts model simulations; bottom: measurement of CE rates ( bar ) and of their components along G ( circles ) . ( k ) As in Figure 2b , numerically implemented forces elongate the cell patch along the horizontal axis , and cell divisions are oriented along the direction of patch elongation; ( l ) same as ( k ) but with divisions now oriented orthogonally to the direction of tissue elongation , and ( m ) without any division occurring during tissue elongation . Only non-zero values are plotted . Scale bars: ( a , b ) 50 μm , ( c–g ) 2 10-2 h-1 , ( k–m ) equivalent to 10-2 h-1 for simulation movies lasting 20 h; scale circle areas: ( h–m ) 0 . 1 h-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 01910 . 7554/eLife . 08519 . 020Figure 6—figure supplement 1 . Additional quantitative characterizations of trblup tissue development . ( a–c ) Maps of overlayed CE rates of cell divisions ( D , green ) , cell rearrangements ( R , magenta ) and cell shape changes ( S , cyan ) , averaged over 5 hemi-nota , in ( a ) wild-type tissue ( reproduced from Figure 4b for comparison ) , ( b ) trblup tissue , ( c ) their difference . ( d ) Average map of delamination CE rate ( A , black ) in trblup tissue . Note that the number of delaminations decreased to ~7 102 , i . e . 13% of the number of delaminations found in wild-type tissue . ( e ) Average map of the component along wild-type tissue CE rate of the difference between division CE rate in wild-type and trblup tissues . Time averages were performed between 14 and 28 hAPF . Scale bars: ( a–d ) 2 10-2 h-1; scale circle area: ( e ) 0 . 1 h-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 020 Two complementary maps can be used to quantitatively study the effects of trbl overexpression: the difference between the tissue CE rates in trblup tissue ( Gtrbl ) and in wild-type tissue ( Gwt ) , namely ΔG ( Figure 6g ) , and its projection onto the wild-type tissue CE rate , namely ΔG// ( Figure 6h ) . ΔG represents the change brought to wild-type tissue morphogenesis by the trbl overexpression , and ΔG// measures the effective contribution of this change to wild-type tissue morphogenesis . A region where ΔG// is positive means that wild-type tissue morphogenesis has been increased by trbl overexpression , and conversely , ΔG// is negative means that it has been decreased in this region . Therefore , the ΔG// map provides a visual representation of where the tissue morphogenesis is increased or reduced and can be interpreted as the role of cell divisions ( proliferation and oriented divisions ) during tissue morphogenesis . This map reveals that in almost all regions of the tissue , and regardless of the orientation of cell divisions relative to tissue elongation in wild-type , overexpressing trbl reduces wild-type tissue elongation ( full circles in Figure 6h ) . A similar approach can be applied to each cell process to determine how it is impacted by the trbl overexpression ( Figure 6h–j and Figure 6—figure supplement 1 ) . Thus , the respective changes in each cell process due to trbl overexpression can be measured using ∆D// , ΔR// , and ΔS// ( ΔA// , not shown ) . As expected from the nearly complete absence of division in trblup tissue , the ΔD// map representing the changes in tissue morphogenesis due to the loss of cell division is almost the opposite of the D// map ( compare Figure 4e and Figure 6—figure supplement 1e ) . More interestingly , the ΔR// , and ΔS// maps directly demonstrate that both cell rearrangements and cell shape changes are significantly modified in trblup tissue and contribute to the overall changes in tissue morphogenesis due to trbl overexpression ( Figure 6i–j ) . This indicates that suppressing proliferation not only makes oriented division CE rate vanish , but also has an indirect impact on both cell rearrangements and cell shape changes . In conjunction with our results in the wild-type tissue , this suggests that both cell proliferation and the orientation of divisions determine the morphogenesis of the tissue , and that a complex interplay exists between cell divisions and the other processes such as cell shape changes and rearrangements . To better understand this last point , and more generally the effect of oriented cell divisions , we used computer simulations . We compared our previous simulation of cell divisions oriented along the horizontal axis of tissue elongation ( Figure 2b and 6k and Video 6a ) with simulations where only the pattern of divisions has been modified in two distinct ways: ( i ) we aligned all cell divisions along the vertical axis , namely orthogonally to tissue elongation , thereby leading to a negative component of D ( D// < 0 ) ( Figure 6l and Video 6b ) , thus mimicking our observation in region 3 of the wild-type tissue ( Figure 4e ) ; ( ii ) we suppressed divisions ( D// = 0 , Figure 6m and Video 6c ) , mimicking our observation in trblup tissue ( Figure 6d ) . In both cases , we found that modifying the pattern of divisions impacts simultaneously G , R and S in addition to D ( Figure 6l , m ) . When divisions are orthogonal to tissue elongation , cell rearrangements , and to a lesser extent cell shape changes , are greatly increased along the direction of deformation , but they only partly compensate the CE rate of horizontally oriented divisions in the initial simulation , thereby resulting in reduced tissue elongation ( Figure 6l ) . When divisions are suppressed , cell rearrangements and cell shape changes are moderately increased along the direction of deformation , and compensate even less horizontally oriented divisions , thereby resulting in further reduced tissue elongation ( Figure 6m ) . These two simulations therefore recapitulate two aspects of our experimental observations: ( i ) how divisions orthogonal to the tissue CE rate in wild-type have a negative component along tissue morphogenesis , as found in some regions of the wild-type tissue ( Figure 4e , region 3 ) ; ( ii ) how divisions , regardless of their orientation , can facilitate tissue elongation by indirectly impacting cells rearrangements and cell shape changes , as observed in trblup tissue where proliferation is severely reduced and tissue deformations are globally decreased ( Figure 6h ) . 10 . 7554/eLife . 08519 . 021Video 6 . Potts model simulations illustrating the impact of cell divisions on tissue elongation and on the other cell processes ( see Figure 6k–m ) . ( a ) As in Figure 2b , numerically implemented forces elongate the cell patch along the horizontal axis , and cell divisions are oriented along the direction of patch elongation . ( b ) Same as ( a ) but with divisions now oriented orthogonally to the direction of tissue elongation . ( c ) Same as ( a ) but without any division occurring during tissue elongation . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 021 Altogether , our formalism reveals the extent of the heterogeneity of division orientation in a tissue , and our analyses of simulations and trblup experimental condition show that both cell proliferation and oriented divisions can influence tissue morphogenesis . Lastly , our formalism provides a unified approach to independently quantify each cell process , thus revealing a complex interplay between cell divisions , cell rearrangements and cell shape changes and providing a rigorous framework for its future characterization using both mutant conditions and modeling . Epithelial tissue growth and morphogenesis is regulated by mechanical stress ( Heisenberg and Bellaïche , 2013 ) . To provide a complete set of methods to study tissue development , we therefore aimed to combine our formalism with the measurement of mechanical stress due to tension in adherens junctions . This ‘junctional stress' gathers all forces ( regardless of their biological origin , including cortical and cytoplasmic forces ) transmitted between cells via adherens junctions . The relevance of junctional stress quantification to understand tissue development has been demonstrated by methods such as laser ablation ( for review see [Rauzi and Lenne , 2011] ) or optical trapping of cell junction ( Bambardekar et al . , 2015 ) . However , with these methods , it is difficult to obtain spatial and temporal stress maps at the scale of the whole tissue . Others and we have previously developed force inference approaches to quantify junction stress from segmented images independently of possible external forces such as a friction of the epithelium on an outer layer ( Brodland et al . , 2010; 2014; Ishihara and Sugimura , 2012; Chiou et al . , 2012; Ishihara et al . , 2013; Sugimura and Ishihara , 2013 ) . We improved our method to make it numerically more robust and efficient , thereby enabling the determination of cell pressures , junction tension and junctional stress over the whole tissue ( see Materials and methods , Figure 7—figure supplement 1 ) . The stress has an isotropic part related to the pressure represented by a circle ( Figure 7—figure supplement 1c ) . Its anisotropic part has an amplitude and a direction of traction represented by a bar , and a direction of compression ( of equal magnitude and perpendicular , the display of which is redundant ) . Even on a single animal , the junctional stress maps vary smoothly over time and space , and are symmetric with respect to the midline , revealing the quality of the signal-to-noise ratio ( Figure 7a , Figure 7—figure supplement 1c and Video 4f ) . We then performed their ensemble average over several animals ( Figure 7b ) and compared the anisotropic part of the junctional stress maps and of the CE rate maps of the different processes measured by the formalism . Focusing here on divisions , the analysis confirms that on average cell division orientation aligns well with junctional stress orientation , even in such a heterogeneous tissue ( Figure 7—figure supplement 2 , alignment = 0 . 87 ) . Moreover , the division CE rate , which is more relevant to tissue morphogenesis , is also well correlated with junctional stress orientation ( Figure 7b , alignment = 0 . 73 ) . 10 . 7554/eLife . 08519 . 022Figure 7 . Maps of junctional stress σ and comparison with division CE rate D . ( a ) Map of the anisotropic part of local junctional stress σ covering the whole notum . Average performed between 14 and 28 hAPF , plotted on the last corresponding image: contours of cells ( thin grey outlines ) and of initially square patches ( thick grey outlines ) . ( b ) Overlay of division CE rate ( D , green ) and of anisotropic part of junctional stress ( σ , red ) . Measurements averaged over time between 14 and 28 hAPF and over 5 hemi-nota . In this Figure ( and Figure 7—figure supplement 2 ) , values larger than the local biological variability are plotted in color while smaller ones are shown in grey; a local transparency is applied to weight the CE rate according to the number of cells and hemi-nota in each group of cells; black outline delineates the archetype hemi-notum; the midline is the top boundary; circles filled in yellow are archetype macrochaetae . Black rectangular boxes outline the four regions numbered 1 to 4 described in the text , same as in Figure 4 . Stress is expressed in arbitrary unit ( A . U . ) proportional to the average junction tension ( not determined by image analysis ) . Scale bars: ( a , b ) 0 . 1 A . U . , ( b ) 2 10-2 h-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 02210 . 7554/eLife . 08519 . 023Figure 7—figure supplement 1 . Maps of cell pressures , junction tensions and junctional stress inferred on single images . Left: 14 hAPF; right: 28hAPF . No averaging over time or animals has been performed on any of these images . ( a ) Maps of cell pressure p , expressed in arbitrary unit proportional to the average junction tension ( not determined by image analysis ) . The bar on the right indicate the color code . Positive or negative values indicate pressures above or below the average pressure . ( b ) Maps of apical junction tension γ , expressed in units of the average cell junction tension . The bar on the right indicate the color code . Values larger or smaller than 1 indicate tensions above or below the average tension . ( c ) Maps of stress σ obtained from data of pressure , junction tension and cell size . Stress is expressed in arbitrary unit ( A . U . ) proportional to the average apical junction tension . By convention , the isotropic part is plotted at zero average , so positive values ( white circles ) and negative values ( grey circles ) correspond to isotropic parts above or below average . Scale bar ( anisotropic part ) : 0 . 1 A . U . Scale circle diameter ( isotropic part ) : 0 . 1 A . U . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 02310 . 7554/eLife . 08519 . 024Figure 7—figure supplement 2 . Comparison of orientations of junctional stress σ and cell division orientation Do . Overlay of division orientation CE rate ( Do , dark green ) and junctional stress ( σ , red ) anisotropic part . Measurement averaged over time between 14 and 28 hAPF and over 5 hemi-nota . Black rectangular boxes outline the four regions numbered 1 to 4 described in the text , same as in Figures 4 and 7 . Alignment coefficient is 0 . 67 in R1 , 0 . 98 in R2 , 0 . 94 in R3 , 0 . 89 in R4 , and 0 . 87 over the whole tissue . Stress is expressed in arbitrary unit ( A . U . ) proportional to the average apical junction tension ( not determined by image analysis ) . Scale bars: 0 . 1 A . U . , 2 10-2 h-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08519 . 024 Taking further advantage of averaged maps of division CE rate on the one hand , and of tissue and cell process component maps on the other hand , enables to explore more finely the alignment between cell divisions and stress . In particular , we can exclude that a positive or negative component of cell divisions would be due to distinct relationships between division CE rate and stress orientations . Indeed , cell divisions have a positive component in region 1 and 2 , while cell division CE rate D is either poorly aligned ( region 1 , alignment = 0 . 16 ) or well aligned ( region 2 , alignment = 0 . 97 ) with junctional stress orientation ( Figure 7b ) . In addition to regions where stress , division CE rate and tissue elongation are well aligned ( region 2 , Figure 7b , Figure 4e ) , we also find regions where , although cell divisions and junctional stress remain well aligned ( region 3 , alignment = 0 . 94 , Figure 7b ) , the tissue CE rate ( G ) is almost orthogonal to divisions and stress ( alignment = -0 . 88 , Figure 4e ) , mostly occurring through cell rearrangements and cell shape changes ( Figure 4f , g ) . Altogether our results illustrate how the combination of the formalism and a stress inference method enables to uncover additional interplays between cell divisions , stress and tissue elongation . This sets the stage for in-depth spatial and temporal investigations of the interactions between each cell process and mechanical stress during tissue development . We have developed a unified multiscale formalism that relates cell and tissue behaviors to characterize the growth and morphogenesis of epithelial tissues in two and three dimensions . The formalism is free from assumptions regarding biological mechanisms , modeling or external forces and it has numerous advantages . Its unified and separate measurements of the contributions of each cell process to tissue growth and morphogenesis significantly help describe and quantify the mechanisms governing tissue development . These measurements have been validated with computer simulations . They can be easily represented on spatial and temporal maps or graphs to describe the interplay between divisions , cell rearrangements , cell shape and size changes and apoptoses , as well as the interplays between cell processes and junctional stress , thus facilitating their comparison in wild-type and mutant conditions . In combination with the recent advances in light microscopy , genetics and physical approaches , our unified framework and methods provide a basis for comprehensive analyses of the mechanisms driving tissue development .
When the data were averaged over larger time or length scales , the left-right symmetry was visually better ( we have checked it quantitatively , data not shown ) , the reproducibility from one animal to another was increased , and the data maps appeared smoother . Unless stated otherwise , all results presented are computed in boxes of size 128×128 pixels2 ( at the onset of the movie ) , namely 40×40 μm2 , with 50% overlap . Time averages are over 2 h for movies or 14 h for still images . Whole notum images are measurements over one animal; archetype refers to average over 5 hemi-notum movies ( 1 whole animal and 3 half animals ) . This yielded good statistics while preserving the fine spatial variations of the data maps . Averaging different movies was made possible by defining their common space and time coordinates . We developed a general method to rescale and register movies from different animals and genotypes in time and space , as follows . | In animals , the final size and shape of each tissue is determined by the precise control of when , where and how much individual cells grow , divide , move and die . An important challenge in biology is to understand how the behaviors of each individual cell can act together to generate a large and reproducible change at the scale of entire tissues and organs . Here , Guirao et al . have developed a new approach to provide maps that reveal how much each cell process contributes to the development of tissues . A caterpillar becoming a butterfly is a famous example of insect ‘metamorphosis’ . The fruit fly offers another example of such tissue development: within five days , a rice grain-like maggot morphs into an adult fly with long antennae , legs and wings . Guirao et al . used a microscope to observe cells over a period of several hours during the metamorphosis of the adult fruit fly wings and thorax ( the region between the neck and abdomen ) . In both regions , Guirao et al . showed that all the cell processes participate in the formation of the adult tissue . Cell division , cell death , and changes in cell size affect the size of the tissue , while cell division , cell rearrangements , and changes in cell shape alter the shape of the tissue . The relative contributions of these cell processes varied a lot in both space and time . Further experiments then used mutant flies with defects in cell division to analyse the impact of cell division on the other cell processes and the eventual shape of the tissue . Finally , Guirao et al . showed that there are unexpected interactions between the patterns of tissue growth , cell division and the mechanical forces in the tissue . These findings provide a new approach to uncover how animals from different species can have such a variety of shapes and sizes , even though they each start life as a single cell . Ultimately , this may also aid efforts to understand how certain diseases affect the development of tissues . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"physics",
"of",
"living",
"systems",
"tools",
"and",
"resources"
] | 2015 | Unified quantitative characterization of epithelial tissue development |
Microglia , the resident CNS macrophages , have been implicated in the pathogenesis of Rett Syndrome ( RTT ) , an X-linked neurodevelopmental disorder . However , the mechanism by which microglia contribute to the disorder is unclear and recent data suggest that microglia do not play a causative role . Here , we use the retinogeniculate system to determine if and how microglia contribute to pathogenesis in a RTT mouse model , the Mecp2 null mouse ( Mecp2tm1 . 1Bird/y ) . We demonstrate that microglia contribute to pathogenesis by excessively engulfing , thereby eliminating , presynaptic inputs at end stages of disease ( ≥P56 Mecp2 null mice ) concomitant with synapse loss . Furthermore , loss or gain of Mecp2 expression specifically in microglia ( Cx3cr1CreER;Mecp2fl/yor Cx3cr1CreER; Mecp2LSL/y ) had little effect on excessive engulfment , synapse loss , or phenotypic abnormalities . Taken together , our data suggest that microglia contribute to end stages of disease by dismantling neural circuits rendered vulnerable by loss of Mecp2 in other CNS cell types .
Rett Syndrome ( RTT ) is a devastating , X-linked neurodevelopmental disorder marked by a developmental stagnation and regression in neurological function . Early on these neurological deficits often have autistic-like features and are accompanied by an array of somatic impairments ( Chahrour and Zoghbi , 2007; Zoghbi , 2003; Lombardi et al . , 2015 ) . Since the discovery that mutations in the transcriptional regulator Methyl-CpG-binding protein 2 ( Mecp2 ) underlie the vast majority of RTT cases , studies in mouse models of RTT have implicated virtually every resident brain cell type ( neurons and glia ) in the disorder ( Amir et al . , 1999; Guy et al . , 2011; McGann et al . , 2012; Lyst and Bird , 2015; Li , 2012 ) . However , it remains unclear which cell types primarily contribute to each phenotype and how these vastly different cell types work in concert with each other to initiate and propagate the disorder . Microglia , the brain resident myeloid-derived cell , are among the most recent cell types implicated in RTT pathogenesis ( Derecki et al . , 2012; Cronk et al . , 2015; Jin et al . , 2015; Maezawa and Jin , 2010 ) . However , the data have been a subject of increasing controversy ( Wang et al . , 2015 ) . The initial study by Derecki et al . transplanted wild-type ( WT ) bone marrow ( BM ) into an irradiated mouse model of RTT , Mecp2 null mouse ( Mecp2-/y ( Mecp2tm1 . 1Jae/y ) ) prior to phenotypic symptom onset ( ~4 weeks of age ) ( Derecki et al . , 2012 ) . When WT BM-derived microglia-like cells engrafted the CNS , many RTT-like phenotypes were arrested and lifespan was significantly increased . While data suggested that phagocytic activity of microglia may be disrupted in Mecp2 null mice , it remained unclear precisely how microglia were contributing to the disorder . In a follow-up study , these data were replicated using a more specific , tamoxifen-inducible Cre driver on a Mecp2 null background ( Cx3cr1CreER; Mecp2lox–stop/y ) ( Cronk et al . , 2015 ) . In addition , RNAseq analysis revealed abnormalities in glucocorticoid signaling , hypoxia responses , and inflammatory responses in peripheral macrophages and resident brain microglia isolated from Mecp2 null mice . While these data support a role for myeloid-derived MeCP2 in RTT phenotypes and pathology , another recent study demonstrated little to no effect of re-introducing MeCP2 into myeloid cells by BM chimerism in three different RTT mouse models ( Mecp2tm1 . 1Jae/y , Mecp2LucHyg/y and Mecp2R168X/y mice ) , or by genetic expression of MeCP2 in hematopoietic cells ( including microglia ) in a MeCP2 null background ( Vav1-Cre; Mecp2LSL/Y ) ( Wang et al . , 2015 ) . Thus , it remains unclear if and how microglia , specifically , contribute to pathogenesis . Recent work in the healthy , developing CNS has demonstrated a surprising new role for microglia in synaptic circuit remodeling and maturation ( Schafer et al . , 2013; Tremblay , 2011a; 2011b; Salter and Beggs , 2014 ) . Among the functions at developing synapses , we recently showed in the retinogeniculate system that microglia contribute to the process of removing excess synapses by phagocytosing less active or ‘weaker’ presynaptic inputs ( Schafer et al . , 2012 ) . Importantly , disrupting microglial phagocytic activity resulted in sustained increases in synapse density and connectivity into adulthood . In the current study , we hypothesized that microglia-mediated synaptic remodeling were abnormal in mouse models of neurodevelopmental disorders associated with aberrant brain wiring and chose RTT to test this hypothesis . In many different RTT mouse models , synaptic circuit dysfunction can be detected often prior to presentation of significant phenotypic abnormalities ( Zoghbi , 2003; Banerjee et al . , 2012; Dani et al . , 2005; Dani and Nelson , 2009; Nguyen et al . , 2012; Wood et al . , 2009; Noutel et al . , 2011; Moretti et al . , 2006; Medrihan et al . , 2008 ) . This includes work in the retinogeniculate system where decreases in single fiber synaptic strength are detected at early stages of disease followed by changes in structural circuits at late phenotypic stages ( Noutel et al . , 2011 ) . In addition , studies assessing synapse density in postmortem human and mouse brain tissue have identified abnormalities , including reductions in synapse number ( Nguyen et al . , 2012; Chapleau et al . , 2009; Fukuda et al . , 2005; Jiang et al . , 2013; Stuss et al . , 2012; Xu et al . , 2014; Chao et al . , 2007 ) . Here , using the retinogeniculate system , we examined the interactions between microglia and synapses before , during , and after the onset of phenotypic regression in the MeCP2 null mouse ( Mecp2tm1 . 1Bird/y ) ( Guy et al . , 2001 ) . Furthermore , we use Cre-lox technology to specifically ablate or express Mecp2 in microglia and determine whether these cells play a causative role in the structural and functional synaptic abnormalities . Our data demonstrate that microglia play a role in pathogenesis of synapses by excessively engulfing presynaptic inputs at end stages of disease in the visual system; however , this effect is largely secondary and independent of microglia-specific loss of Mecp2 expression .
The retinogeniculate system , a classic model for studying multiple waves of developmental synapse refinement , is comprised of retinal ganglion cells ( RGCs ) residing in the retina that project to relay neurons in the lateral geniculate nucleus ( LGN ) of the thalamus ( Guido , 2008; Hong and Chen , 2011; Huberman , 2007 ) . We previously established that microglia contribute to early phase synapse refinement by engulfing , thereby eliminating , presynaptic inputs at P5 ( Schafer et al . , 2012 ) . Synaptic engulfment was subsequently downregulated during later waves of refinement ( P9-P30 ) ( Guido , 2008; Hong and Chen , 2011; Huberman et al . , 2008; Torborg and Feller , 2005 ) . It was unknown whether microglia regulate presynaptic input density after P30 . Given that Mecp2 null mice begin to phenotypically regress ≥P30 and continue regression until premature death ~P60 ( Chahrour and Zoghbi , 2007; Guy et al . , 2011; Lyst and Bird , 2015; Guy et al . , 2001 ) , we first needed to establish a baseline engulfment in >P30 WT mice . Recently , a new late wave of refinement was identified between P30 and P60 in which RGC arbors decrease in size and presynaptic boutons decrease in number ( Hong et al . , 2014 ) . We hypothesized that microglia were contributing to this late phase refinement by transiently engulfing synapses between P30 and P60 . We first confirmed a reduction in retinogeniculate synapses in the late , juvenile brain of WT mice by immunolabeling P30-P60 LGN with antibodies against the RGC-specific presynaptic protein vesicular glutamate transporter 2 ( VGlut2 ) , and the postsynaptic protein Homer1 ( Figure 1—figure supplement 1 ) . There was a significant reduction in the density of RGC-specific synapses ( VGlut2/Homer1-positive ) and a reduction in VGlut2-positive terminal size between P30 and P60 ( Figure 1—figure supplement 1A–F ) . This was in contrast to VGlut1-positive corticocortical synapses , which remain unchanged ( Figure 1—figure supplement 1G–I ) . To determine whether microglia contribute to late phase synaptic remodeling in the late , juvenile brain and establish a baseline to assess whether these interactions are disrupted in phenotypic Mecp2 null mice , we next used our established assay to monitor microglia-synapse interactions in the retinogeniculate system ( Schafer et al . , 2012 , 2014 ) . One day prior to analysis , RGC inputs were labeled by injection of anterograde dye into both eyes , cholera toxin conjugated to Alexa dye 594 or 647 ( CTB-594 or CTB-647 ) , which is resistant to lysosomal degradation . Microglia were labeled by either genetic expression of EGFP ( Cx3CR1EGFP/+ mice ) or by immunohistochemistry using an antibody specific to the microglia-marker , Iba-1 . Lysosomes that are specific to and within microglia were labeled with an antibody against CD68 . Similar to previously published work ( Schafer et al . , 2012 ) , microglia preferentially engulfed RGC inputs within the LGN at P5 ( Figure 1—figure supplement 2 ) compared to older ages . However , our new data revealed a second wave of engulfment that occurred in the juvenile brain specifically at P40 , which is downregulated by P50 ( Figure 1A–B ) and accompanied by a transient increase in lysosomal content within microglia ( Figure 1C ) . Together with recent work demonstrating decreases in RGC arbor size and bouton numbers between P30 and P60 ( Hong et al . , 2014 ) , our work suggests that microglia contribute to this fine-scale refinement by engulfing RGC presynaptic inputs at P40 . 10 . 7554/eLife . 15224 . 003Figure 1 . Microglia transiently engulf retinogeniculate presynaptic inputs in the juvenile P40 brain consistent with late stage synapse refinement . ( A ) Representative surface rendered microglia ( green ) and engulfed retinogeniculate inputs ( red ) from P30 , P40 , and P60 LGN . See also Figure 1—figure supplement 2 . Grid line increments = 5 µm . ( B ) Quantification of engulfment reveals a transient and significant increase in engulfment of RGC inputs within microglia at P40 , an age consistent with late-stage synaptic refinement ( Figure 1—figure supplement 1 ) . ( C ) Accompanying increased engulfment , microglia also upregulate engulfment capacity at P40 as measured by lysosomal content within each microglia ( CD68 immunoreactivity per cell ) . *p<0 . 05 , **p<0 . 01 by one-way ANOVA , Dunnett’s post hoc test ( all ages are compared to P60 ) . All error bars represent SEM; N = 4–6 mice per age of mixed sex ( equal ratios of males and females were used across ages ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 00310 . 7554/eLife . 15224 . 004Figure 1—figure supplement 1 . Refinement of structural synapses in the late juvenile retinogeniculate system . ( A–B ) Immunohistochemistry in P30 ( A ) and P60 ( B ) dLGN for vesicular glutamate transporter 2 ( VG2 ) to label RGC-specific presynaptic terminals ( ii , green in iv ) and Homer1 to label postsynaptic densities ( iii , red in iv ) . DAPI was used to label nuclei ( i ) . Scale bar = 20 µm . ( C–F ) Quantification reveals a developmental decrease between P30 and P60 in structural VG2+ terminal size ( C ) , VG2+ terminal density ( D ) , and VG2+ synapse density ( co-localized VG2 and Homer1 ) ( F ) . ( G–I ) Quantification of immunohistochemistry for vesicular glutamate transporter 1 ( VG1 ) to label corticothalamic-specific presynaptic terminals reveals no significant difference in VG1+ terminals ( G–H ) or VG1-containing synapses ( I , colocalized VG1 and Homer1 ) in P30 vs P60 dLGN . *p<0 . 05 by Student’s t-test . All error bars represent SEM; N = 3–4 per age . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 00410 . 7554/eLife . 15224 . 005Figure 1—figure supplement 2 . Presynaptic input engulfment in early and late phases of synaptic refinement in the developing retinogeniculate system . Engulfment of retinogeniculate presynaptic inputs is significantly increased in P5 and P40 LGN compared to P50 . *p<0 . 05 , ***p<0 . 001 by one-way ANOVA , Dunnett’s post hoc test ( all ages are compared to P60 ) . All error bars represent SEM; N = 4–6 mice per age . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 005 Phenotypic regression is evident in Mecp2 null mice by P40 . Furthermore , these abnormalities occur after the onset of electrophysiological weakening of single fiber synaptic responses in the P20-P30 Mecp2 null retinogeniculate system ( Noutel et al . , 2011 ) . We hypothesized that microglia-mediated engulfment of retinogeniculate inputs in P40 , juvenile mice was enhanced in Mecp2 null mice with weakened synapses . Similar to experiments described to assess engulfment in the juvenile WT brain , RGC presynaptic inputs from both eyes were labeled with CTB-594 or CTB-647 and microglia were labeled by either genetic expression of EGFP ( Cx3CR1 EGFP/+; Mecp2-/y or Cx3CR1EGFP/+; Mecp2+/y ) or immunolabeling with anti-Iba-1 . In addition , to measure lysosomal content , microglia were labeled with anti-CD68 . Using these methods , we detected no significant difference in microglia-mediated engulfment of retinogeniculate presynaptic inputs in P5-P50 WT or Mecp2 null mice compared to WT littermate controls ( Figure 2C ) . However , in late phenotypic P56-P60 ( >P56 ) Mecp2 null mice , we found significant increases in engulfed inputs and lysosomal content within microglia processes and soma compared to WT littermates ( Figure 2A–D ) . Consistent with engulfment being specific to synaptic compartments , we observed no significant RGC death in the retina or LGN ( Figure 2—figure supplement 1 ) and we observed no changes in engulfment of other neuronal compartments including NeuN-positive somas or MAP2-positive dendrites ( Figure 2—figure supplement 2 ) . In addition and in contrast to previously published reports ( Cronk et al . , 2015; Jin et al . , 2015 ) , we observed no significant changes in morphology ( as measured by volume of the cell ) or density of microglia , indexes of the gross , overall reactive state of these cells ( Figure 2E–F ) . However , our analyses were restricted to microglia within the LGN , a region that was not analyzed previously ( Cronk et al . , 2015; Jin et al . , 2015 ) . These data demonstrate that , while microglia-mediated waves of synaptic engulfment are normal in the P5 and P40 Mecp2 null brain , engulfment is excessive in ≥P56 mice—an age corresponding to late stages of phenotypic regression ( Guy et al . , 2001 ) . Furthermore , this timing occurs after significant weakening of single fiber strength at Mecp2 null retinogeniculate synapses ( Noutel et al . , 2011 ) . These data suggest that microglia do not actively induce circuit defects in Mecp2 null mice but rather facilitate late stage circuit defects by removing previously weakened structural synapses . 10 . 7554/eLife . 15224 . 006Figure 2 . Microglia excessively engulf retinogeniculate presynaptic inputs in late phenotypic Mecp2 null mice . ( A–B ) Representative surface rendered microglia ( green ) and engulfed RGC inputs ( red ) demonstrates enhanced engulfment of presynaptic inputs in ≥P56 ( P56-P60 ) Mecp2 null dLGN ( B ) as compared to WT littermate controls ( A ) . Grid line increments = 5 µm . ( C ) Quantification of engulfment across development reveals excessive engulfment of presynaptic inputs within ≥P56 Mecp2 null dLGN as compared to WT littermate controls in the absence of significant RGC cell death ( Figure 2—figure supplement 1 ) or engulfment of other non-synaptic neuronal debris ( Figure 2—figure supplement 2 ) . *p<0 . 05 by multiple unpaired Student’s t-tests; N = 4–6 mice per age and genotype; all data are normalized to WT controls at each age . ( D ) Quantification of lysosomal content ( CD68 immunoreactivity ) within microglia in ≥P56 LGN reveals a significant increase in phagocytic capacity in Mecp2 null mice as compared to WT littermate controls . **p<0 . 01 by unpaired Student’s t-tests; N = 3 mice per genotype; data are normalized to WT control . ( E–F ) There is no significant difference in numbers or volume of microglia within the ≥P56 LGN by unpaired Student’s t-test; N = 3–5 mice per genotype; data are normalized to WT control . All error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 00610 . 7554/eLife . 15224 . 007Figure 2—figure supplement 1 . There is no significant cell death in the retinas of Mecp2 null mice . ( A–B ) Quantification of total DAPI ( A ) and NeuN ( B ) cells in the retinal ganglion cell layer of the retina from ≥P56 Mecp2 WT ( grey bars ) and null ( red bars ) mice reveals no significant changes in cell numbers . All error bars represent SEM; N = 3 mice per genotype . ( C ) Immunohistochemistry in the Mecp2 WT ( top ) or null ( bottom ) retina for TUJ1 ( red ) to label the retinal ganglion cell layer or cleaved caspase ( green ) to label dead/dying cells . Cleaved caspase channel is visualized separately in far right panels for both genotypes . Cell death was never observed in either genotype . Data are representative of 3 mice per genotype . Scale bar = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 00710 . 7554/eLife . 15224 . 008Figure 2—figure supplement 2 . Engulfment is specific to presynaptic inputs . ( A ) A representative maximum intensity projection of a microglia ( EGFP , green ) and neurons labeled with anti-NeuN ( blue ) , and anti-MAP2 ( red ) in the ≥P56 ( P56-P60 ) Mecp2 null LGN . Scale bar = 5 µm . ( C ) Surface rendering of the same microglia ( A ) and engulfed MAP2 and NeuN-positive debris ( arrows ) . Grid line increments = 5 µm . ( C–D ) Quantification of engulfed MAP2 ( C ) or NeuN ( D ) immunoreactive neuronal debris reveals no significant difference between Mecp2 WT ( grey bars ) and null animals ( red bars ) . Error bars represent SEM; N = 3 mice per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 008 We next assessed whether increased engulfment in P56-P60 ( ≥P56 ) Mecp2 null mice corresponded to loss of structural retinogeniculate synapses . We first immunolabeled retinogeniculate presynaptic terminals in P40 and P56-P60 ( ≥P56 ) Mecp2 null and WT littermate brains with an antibody directed against VGlut2 . While there was no change in the density of VGlut2 immunoreactivity in P40 Mecp2 null mice compared to WT littermate controls , there was a significant decrease at ≥P56 , a time point corresponding to late-stage phenotypic regression in Mecp2 null mice ( Figure 3A–D ) . To determine whether this reduction in VGlut2 was consistent with a loss of synapses , we further assessed P56-P60 ( ≥P56 ) Mecp2 null mice for changes in retinogeniculate synapse density defined as co-localized presynaptic VGlut2 and postsynaptic Homer1 immunoreactivity . Consistent with the reduction in VGlut2 and excessive synaptic engulfment , there was a significant decrease in retinogeniculate synapses in P56-P60 ( ≥P56 ) Mecp2 null mice as compared to WT littermate controls ( Figure 3 ) and this synapse loss was due to loss of VGlut2-positive terminals ( Figure 3C ) versus a decrease in the postsynaptic protein Homer1 ( Figure 3H ) or RGC cell death ( Figure 2—figure supplement 1 ) . Retinogeniculate synapses represent <10% of total synapses within the LGN ( Bickford et al . , 2010 ) . To assess the other more abundant excitatory synapses , we immunolabeled corticogeniculate synapses with an antibody against VGlut1 within the LGN and observed no significant difference in the density of these synapses or presynaptic terminals . ( Figure 3—figure supplement 1A–E ) . In addition , we assessed VGlut2 and VGlut1-positive synapse density in a neighboring thalamic nuclei ( medial geniculate nucleus , MGN; Figure 3—figure supplement 1F–G ) and observed no significant loss of these structural synapses . These results demonstrate a specific loss of retinogeniculate presynaptic terminals in late phenotypic ≥P56 Mecp2 null mice concomitant with increased microglia-mediated engulfment of presynaptic inputs . 10 . 7554/eLife . 15224 . 009Figure 3 . Retinogeniculate presynaptic terminals and synapses are reduced in late phenotypic ≥P56 Mecp2 null mice . ( A ) Immunohistochemistry for VGlut2 to label retinogeniculate presynaptic terminals , in the dLGN of P40 ( A ) and ≥P56 ( B , P56-P60 ) Mecp2 wild-type ( WT; left column ) and null ( right column ) littermates . Images are single planes of a confocal z-stack . Scale bar = 20 µm . ( C–D ) Quantification of RGC presynaptic terminal ( VGlut2+ puncta ) immunohistochemistry reveals a significant decrease in RGC-specific terminal density ( C ) and size ( D ) in ≥P56 Mecp2 null mice ( red bars ) as compared to WT littermate controls ( grey bars ) . No significant difference was observed at P40 . All data are normalized to WT control for each age . **p<0 . 01 unpaired Student’s t-test at each age; N = 3–4 mice per age and genotype; ( E–F ) Immunohistochemistry for VGlut2 ( green ) and the postsynaptic marker Homer1 ( in the dLGN of ≥P56 Mecp2 WT ( E ) and null ( F ) littermates . Images are single planes from confocal z-stacks . The VGlut2 and Homer1 channels are separated in panels ii–iii . Panels Eiv and Fiv are colocalized VGlut2 and Homer1 puncta . Scale bar = 10 µm . ( G–H ) Quantification reveals a a significant decrease in RGC-specific synapses ( colocalized VGlut2 and Homer ) within the LGN of >P56 Mecp2 null mice ( red bars ) as compared to WT littermate controls ( grey bars ) ( G ) and no significant change in the density of the postsynaptic protein Homer1 ( H ) . No significant difference in density was observed in corticogeniculate-specific , VGlut1-positive synapses within the LGN or VGlut2 or VGlut1-containing synapses within a neighboring thalamice nuceli ( Figure 3—figure supplement 1 ) . *p<0 . 05 by Student’s paired t-test . ; N = 5 mice per genotype; all data are normalized to WT control . All error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 00910 . 7554/eLife . 15224 . 010Figure 3—figure supplement 1 . Presynaptic terminal and synapse loss are specific to retinogeniculate synapses . ( A–B ) Immunohistochemistry for the corticogeniculate presynaptic terminal marker VGlut1 ( green ) and the postsynaptic marker Homer in the dLGN of ≥P56 Mecp2 WT ( A ) and null ( B ) littermates . Images are single planes from confocal z-stacks . The VGlut1 and Homer channels are separated in panels ii-iii . Colocalized VGlut1 and Homer puncta are visualized in panels Aiv and Biv . Scale bar = 10 µm . ( C–E ) Quantification reveals no significant difference in the density of VGlut1-positive terminals ( C ) , Homer-positive postsynaptic densities ( D ) , or colocalized VGlut1/Homer puncta ( E ) in the LGN of ≥P56 Mecp2 null mice ( red bars ) as compared to WT littermate controls ( grey bars ) . ( F–G ) Quantification of density of VGlut2 ( F ) and VGlut1 ( G ) presynaptic terminals in the medial geniculate nucleus ( MGN ) of ≥P56 Mecp2 WT ( grey bars ) and null ( red bars ) mice reveals no significant changes in terminal density in this neighboring thalamic nuclei . All error bars represent SEM; N = 4–5 mice per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 010 To address how loss of Mecp2 expression specifically affects microglia function , we crossed Mecp2fl/y mice with Cx3cr1CreER mice to conditionally ablate Mecp2 in microglia following tamoxifen injection ( Cronk et al . , 2015; Goldmann et al . , 2013; Yona et al . , 2013 ) . To achieve microglia-specific Mecp2 ablation , tamoxifen was injected at P21-P25 and mice were assessed ~2 . 5 months later ( P110-P120; Figure 4A; Figure 4—figure supplement 1 ) . Consistent with microglia performing a secondary role , we found no significant increase in retinogeniculate synapse engulfment or RGC presynaptic terminal ( VGlut2-positive ) loss when Mecp2 was ablated specifically in microglia ( Cx3cr1CeER/+;Mecp2fl/y Tam , blue hashed bars ) as compared to all control groups ( Figure 4B–C ) . 10 . 7554/eLife . 15224 . 011Figure 4 . Excessive engulfment , synapse loss , and phenotypic regression are not induced following microglia-specific loss of Mecp2 expression . ( A ) Paradigm for inducing recombination in which mice receive 2 tamoxifen or vehicle ( oil ) injections 48 hr apart between P21 and P25 ( Figure 4—figure supplement 1 ) . Behavior and postmortem analyses are subsequently performed in P110-P120 mice . ( B–C ) Quantification of engulfment ( B ) and VGlut2 terminal density ( C ) in the LGN of oil ( solid bars ) or tamoxifen ( Tam , hashed bars ) -treated mice expressing Cx3cr1CreER/+;Mecp2fl/y ( blue bars ) or Cx3cr1CreER/+;Mecp2+/y ( grey bars ) reveals no significant effect when Mecp2 expression is specifically ablated in microglia ( Cx3cr1CreER/+;Mecp2fl/y Tam , blue hashed bars ) compared to all control groups . N = 4–6 mice per genotype ( D–G ) . Quantification of neurological scores , weight loss , latency to fall from a rotarod , and behavioral visual acuity ( optometry ) in oil ( solid bars ) or tamoxifen ( Tam , hashed bars ) -treated mice expressing Cx3cr1CreER/+;Mecp2fl/y or Cx3cr1CreER/+;Mecp2+/y . There is no significant difference between mice with Mecp2-deficient microglia ( Cx3cr1CreER/+;Mecp2fl/y Tam , blue hashed bars ) versus the same genotype treated with oil ( Cx3cr1CreER/+;Mecp2fl/y Oil blue solid bars ) in any assays . However , there is a small but significant deficit in neurological score ( D ) and visual acuity ( G ) when comparing mice with Mecp2-deficient microglia ( Cx3cr1CreER/+;Mecp2fl/y Tam , blue hashed bars ) to WT controls ( Cx3cr1CreER/+;Mecp2+/y , grey bars ) , an effect likely due to the hypomorphic Mecp2fl/y allele . *p<0 . 05 , **p<0 . 01 by one-way ANOVA , Tukey’s post hoc test; N = 7–13 mice per genotype . All error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 01110 . 7554/eLife . 15224 . 012Figure 4—figure supplement 1 . Validation of Cre-mediated recombination and Mecp2 deletion in microglia . ( A–E ) Cx3CR1CreER mice were crossed with Ai9 ( RCL-tdT ) mice which harbor a targeted mutation of the Gt ( ROSA ) 26Sor locus with a loxP-flanked STOP cassette that prevents transcription tdTomato . Mice were administered tamoxifen at P21-P25 and assessed for recombination >P80 . ( A–B ) Images demonstrating colocalization between Iba-1 positive microglia ( green ) and tdTomato-positive cells ( red ) in the LGN and cortex . Scale bar = 100 µm . ( C–D ) Magnified regions from A and B . Scale bar = 100 µm . ( E ) Quantification of images reveals that nearly 100% of Iba-1 positive microglia were positive for tdTomato ( green bar ) and nearly 100% of td Tomato-positive cells were positive for Iba-1 ( red bar ) . Error bars represent SEM; N = 3 mice . ( F ) PCR for the intact ( ~500 ) and the excised mecp2fl allele ( ~400 bp ) on genomic DNA of microglia isolated from whole brains of Cx3cr1CreER/+:Mecp2fl/y mice , treated or non-treated tamoxifen ( TAM ) in vivo . ( G ) Flow cytometric analysis of microglia isolated from whole brains of TAM-treated Cx3cr1CreER/+; Mecp2+/yand Cx3cr1CreER;Mecp2fl/y mice reveals loss of Mecp2 protein in Cx3cr1CreER:Mecp2fl/y mice ( red line ) . ( H ) Immunohistochemistry in the LGN for Mecp2 ( C-terminal antibody was a generous gift from Dr . M . Greenberg ) following antigen retrieval and amplification using HRP-conjugated secondary antibodies reveals Mecp2 protein ( red ) within the nucleus ( DAPI , blue ) of Iba-1-positive microglia in WT ( top row ) animals ( yellow arrow ) and loss of Mecp2 specifically in microglia ( red arrow ) following recombination in Cx3cr1CreER;Mecp2fl/y mice ( P110-P120; bottom row ) . Merged and individual channels are shown as well as magnified region of Mecp2 channel ( far right panels ) . Scale bar =10 µm . It should be noted that several other Mecp2 antibodies and staining conditions were attempted , but only this antibody and condition enabled us to detect Mecp2 protein even in WT microglia . ( I ) Quantification of Mecp2 expression in microglia reveals a significant loss of Mecp2 protein following recombination in Cx3cr1CreER;Mecp2fl/y mice . ****p<0 . 0001 by Student’s t-test . Error bars represent SEM; N = 3 mice per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 012 We also assessed other general phenotypic abnormalities known to be significantly affected in Mecp2 null animals including overall neurological score , weight loss , rotarod performance and the optomotor task , an assessment of behavioral visual acuity previously shown to be significantly decreased in Mecp2 null mice ( Figure 4D–G ) ( Durand et al . , 2012 ) . We observed no significant defects in weight loss or rotarod performance in mice that lacked Mecp2 specifically in microglia ( Cx3cr1CeER/+;Mecp2fl/y Tam , blue hashed bars ) , as compared to all controls ( Figure 4E–F ) . Neurological score and optomotor task performance ( Figure 4D , G ) were also not significantly different between mice that lacked Mecp2 in microglia ( Cx3cr1CreER/+;Mecp2fl/y Tam , blue hashed bars ) and the same genotype treated with oil ( Cx3cr1CreER/+;Mecp2fl/y Oil , blue solid bars ) . However there was a small but significant effect when compared to WT controls ( Cx3cr1CreER/+;Mecp2+/y , gray bars ) , an effect which may be confounded by the hypomorphic Mecp2fl/y allele ( see Discussion ) ( Samaco et al . , 2008; Kerr et al . , 2008 ) . Together , these data demonstrate that loss of Mecp2 in microglia is largely insufficient to induce excessive engulfment , synapse loss or phenotypic abnormalities . In addition to assessing mice that specifically lack Mecp2 expression in microglia , we did the converse experiment using a similar tamoxifen injection paradigm to express Mecp2 specifically in microglia in an otherwise Mecp2 null background ( Cx3cr1CreER/+;Mecp2LSL/y ) ( Figure 5A , Figure 5—figure supplement 1 ) . Similar mice have been assessed by other groups and have generated differing results--one group demonstrated significant attenuation of phenotypes , while another group observed no effect ( Derecki et al . , 2012; Cronk et al . , 2015; Wang et al . , 2015 ) . We sought to assess microglia dysfunction , synapse loss , and general phenotypes in Cx3cr1CreER/+;Mecp2LSL/y and clarify these disparate results . 10 . 7554/eLife . 15224 . 013Figure 5 . Mecp2 expression in microglia is largely insufficient to attenuate excessive engulfment , synapse loss , or phenotypic regression in Mecp2 null mice . ( A ) Paradigm for inducing recombination in which mice receive 2 tamoxifen or oil injections 48 hr apart between P21 and P25 . Behavior and postmortem analyses are subsequently performed in P78-P90 mice ( Figure 5—figure supplement 1 ) . ( B–C ) Quantification of engulfment ( B ) and VGlut2 terminal density ( C ) in the LGN of oil ( solid bars ) or tamoxifen ( hashed bars ) -treated mice expressing Cx3cr1CreER/+;Mecp2LSL/y ( red bars ) or Cx3cr1CreER/+;Mecp2+/y ( grey bars ) . ( B ) There was a significant decrease in VGlut2 terminal density in mice null for Mecp2 in all cells ( red solid bars ) and this effect was attenuated when Mecp2 was expressed in microglia ( red hashed bars ) , an effect which may have resulted from tamoxifen treatment which induces a trend towards increased VGlut2 density in WT mice ( grey hashed bars ) . *p<0 . 05 , **p<0 . 01 by one-way ANOVA , Tukey’s post hoc test; N = 4–6 mice per genotype . ( C ) In addition , there is a significant increase in engulfment in mice null for Mecp2 in all cells ( red solid bars ) compared to WT , oil-treated littermates ( grey solid bars ) and this effect was not significantly attenuated when Mecp2 was expressed in microglia ( red hashed bars ) . However , expression of Mecp2 in microglia is no longer significant from controls , which suggests a modest effect . *p<0 . 05 by one-way ANOVA , Dunnett’s post hoc test ( all genotypes compared to Cx3cr1CreER/+;Mecp2+/y Oil , grey bars; data are not significant by Tukey’s post hoc test ) ; N = 3–5 mice per genotype . All error bars represent SEM . ( D–G ) Expression of Mecp2 in a null background ( red hashed bars ) has no significant effect on attenuation of deficits in neurological score ( D ) weight loss ( E ) latency to fall from a rotarod ( F ) or visual acuity ( G ) compared to mice null for Mecp2 in all cells ( red bars ) . However , rotarod performance in Cx3cr1CreER/+;Mecp2LSL/y Tam mice ( red hashed bar ) was no longer significant from controls ( grey hashed and solid bars ) , which suggests a modest effect . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 by one-way ANOVA , Tukey’s post hoc test; N = 6–11 mice per genotype . All error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 01310 . 7554/eLife . 15224 . 014Figure 5—figure supplement 1 . Validation of loss or gain of Mecp2 protein in dLGN microglia following Cre-mediated recombination . ( A ) Immunohistochemistry in the LGN for Mecp2 ( C-terminal antibody was a generous gift from Dr . M . Greenberg ) following antigen retrieval and amplification using HRP-conjugated secondary antibodies reveals Mecp2 protein ( red ) within the nucleus ( DAPI , blue ) of Iba-1-positive microglia in WT ( top row ) and Cx3CR1CreER/+;Mecp2LSL/y ( bottom row ) animals treated with tamoxifen ( yellow arrows ) and loss of Mecp2 in all cells including microglia ( red arrow ) in Cx3cr1CreER;Mecp2fl/y mice treated with oil ( middle row ) . Merged and individual channels are shown as well as magnified region of Mecp2 channel ( far right panels ) . Scale bar = 10 µm . ( B ) Quantification of Mecp2 expression in microglia reveals a significant loss of Mecp2 protein in Cx3cr1CreER;Mecp2LSL/y mice treated with oil compared to Cx3CR1CreER/+;Mecp2+/y or Cx3CR1CreER/+;Mecp2LSL/y treated with tamoxifen . ****p<0 . 0001 , ***p<0 . 001 by one-way ANOVA . Error bars represent SEM; N = 3 mice per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 15224 . 014 We first assessed whether microglia-specific Mecp2 expression was sufficient to attenuate excess engulfment and synapse loss in Mecp2 null mice . Similar to ≥P56 Mecp2 null mice , there was a significant increase in engulfment and decrease in RGC presynaptic terminal density in late phenotypic P78-P90 Mecp2LSL/y treated with oil ( Cx3cr1CreER/+;Mecp2LSL/y Oil , red solid bars ) as compared to Mecp2+/y controls ( grey bars , Figure 5B–C ) . When Mecp2 was expressed in microglia in a null background ( Cx3cr1CreER/+;Mecp2LSL/y Tam , red hashed bars ) , the excessive engulfment was not attenuated compared to the same genotype treated with oil ( Cx3cr1CreER/+;Mecp2LSL/y Oil , red solid bars ) . However , engulfment was no longer significantly different from Mecp2+/y controls ( grey solid and hashed bars ) , which suggests a modest effect . In contrast , there was a significant enhancement in RGC terminal density in Cx3cr1CreER/+;Mecp2LSL/y treated with tamoxifen ( red hashed bars ) compared to null animals ( red solid bars; Figure 5C ) . Tamoxifen administration alone ( Cx3cr1CreER/+;Mecp+/y Tam , grey hashed bars ) however , had a trend towards increased VGlut2 density in WT mice ( Figure 5C ) , which suggests this effect may result from tamoxifen treatment . We next assessed general phenotypes and behavioral visual acuity in Cx3cr1CreER/+;Mecp2LSL/y mice . We did not observe a significant improvement in neurological score , weight loss , or visual acuity and only a small improvement in rotarod performance when Mecp2 was specifically expressed in microglia ( Figure 5D–G , Cx3cr1CreER/+;Mecp2LSL/y Tam , red hashed bars ) . Together with data from Cx3cr1CreER/+;Mecp2fl/y mice , excessive engulfment , synapse loss , and phenotypic abnormalities are largely independent of microglia-specific loss or gain of Mecp2 expression . Our data are most consistent with recent reports that microglia-specific Mecp2 expression is insufficient to attenuate phenotypes in Mecp2 null mice ( Wang et al . , 2015 ) .
In the process of establishing a baseline of engulfment in the WT juvenile brain , we identified a new window of microglia-mediated presynaptic engulfment at P40 ( Figure 2 ) . This age corresponds to a newly identified window of late-stage , fine-scale structural synapse elimination in the retinogeniculate system ( Figure 1—figure supplement 1 ) ( Hong et al . , 2014 ) . One open question is what molecular mechanism underlies this late-stage engulfment and , if disrupted , are there sustained deficits in circuit structure and function . In early postnatal development ( first postnatal week ) , we previously identified that microglia engulf presynaptic inputs , in part , through complement-dependent phagocytosis ( Schafer et al . , 2012 ) . Mice deficient in the microglial phagocytic receptor , complement receptor 3 ( CR3 ) , or complement components C3 and C1q had sustained deficits in engulfment and synaptic remodeling in the retinogeniculate system ( Schafer et al . , 2012; Bialas and Stevens , 2013; Stevens et al . , 2007 ) . Furthermore , this process was dependent upon neural activity whereby microglia preferentially engulfed less active or ‘weaker’ presynaptic inputs . It is unknown whether this late phase presynaptic input engulfment is dependent upon complement or activity . Given that CR3 and C3 ( the ligand for CR3 ) decrease over development and C3 , in particular , is very low/undetectable in the juvenile brain ( Schafer et al . , 2012; Stevens et al . , 2007; Stephan et al . , 2013 ) , it is likely that another mechanism underlies microglia-synapse interactions in juvenile animals . It is also clear that this late-stage presynaptic input engulfment is independent of Mecp2 , as engulfment at P40 is indistinguishable from WT littermates ( Figure 3C ) . Future work to assess other molecular pathways underlying microglia-synapse interactions in the juvenile brain will be important going forward . While in vitro work has suggested that loss of Mecp2 in microglia can affect glutamate-mediated neurotoxicity and synapses ( Jin et al . , 2015; Maezawa and Jin , 2010 ) , it was unknown whether microglia affect synapses in Mecp2 null mice in vivo . Furthermore , while deficits in glutamatergic , glucocorticoid , hypoxia , and immune-related pathways have recently been reported in Mecp2 null microglia ( Cronk et al . , 2015; Derecki et al . , 2012; Jin et al . , 2015; Maezawa and Jin , 2010 ) , it has remained unclear precisely how microglia were contributing to disease on a mechanistic level in vivo . Our data offer significant insight into these unanswered questions . We demonstrate that microglia excessively engulf presynaptic inputs in the Mecp2 null LGN concomitant with loss of structural retinogeniculate-specific synapses in the same region . These data are in contrast to previously published work that has suggested microglial phagocytic activity is decreased compared to WT mice ( Derecki et al . , 2012 ) . However , this discrepancy can be explained by differences in experimental design used to measure phagocytic activity . Assays used to measure phagocytosis in previous work were in vitro in response to the addition of UV-irradiated neural precursor cells ( Derecki et al . , 2012 ) , a context that is very different from assessing engulfment of presynaptic inputs in the retinogeniculate system in vivo . In the same study , annexin V was administered in vivo to block phagocytic activity in Mecp2 null mice with WT BM-derived cells ( Mecp2LSL/y/LysmCre ) , which resulted in failure of WT BM-derived cells to attenuate phenotypes in Mecp2LSL/y mice . However , it is unclear where and how annexin V is acting given that phagocytosis was not directly assayed in vivo and the LysmCre induces expression in many myeloid-derived cell types besides microglia . Indeed , this same group published findings that loss of Mecp2 primarily affects peripheral myeloid-derived cell numbers and gene expression early in disease and only later affects microglia ( Cronk et al . , 2015 ) . Going forward , it will be important to understand the contribution of these peripheral cell types to disease phenotypes . To address whether microglia were primary or secondary to synapse loss , we used Cre-lox technology to specifically express or ablate Mecp2 in microglia . In doing so , we demonstrate that synapse loss and excessive engulfment in Mecp2 null mice are largely independent of microglial-specific loss of Mecp2 expression . This is inconsistent with data in which re-expression of Mecp2in myeloid-derived cells ( including microglia ) attenuates several behavioral phenotypes and cell loss in Mecp2 null mice ( Derecki et al . , 2012; Cronk et al . , 2015 ) . However , our data are consistent with data from this same group suggesting that microglia in Mecp2 null mice are abnormal only in late phenotypic mice , suggesting a secondary effect ( Cronk et al . , 2015 ) . Furthermore , our data are consistent with the newest report from a different group that re-expression of Mecp2 in myeloid cells ( including microglia ) does not attenuate phenotypes in Mecp2 null or mutant mice ( Wang et al . , 2015 ) . While our data suggest that microglia are largely secondary responders to synapses rendered vulnerable by loss of Mecp2 in other CNS cell types , the significance of excessive synaptic engulfment to disease progression is still unknown . The molecular mechanism driving secondary engulfment of synapses in Mecp2 null mice and whether modulating this excessive engulfment results in attenuation of synapse loss are also unknown . Complement-dependent phagocytic signaling is one mechanism by which microglia have been shown to engulf synapses in the healthy brain , a pathway which is also dysregulated in disease ( Schafer et al . , 2012; Stephan et al . , 2012; Chung et al . , 2015; Hong et al . , 2016; Lui et al . , 2016 ) . In addition , there are a number of other inflammatory genes that have been identified as dysregulated in Mecp2 null microglia and may contribute to increased phagocytic activity ( Cronk et al . , 2015 ) . Finally , we demonstrate that microglia are largely secondary responders in two mouse models of RTT ( Mecp2 null and Mecp2LSL/y ) . It is still possible that microglia may be primary initiators of synaptic defects in other RTT models ( Mecp2-/+ , Mecp2 duplication , Mecp2R270X etc . ) , a mechanism recently reported in mouse models of frontotemporal dementia and Alzheimer's disease ( Chahrour and Zoghbi , 2007; Chung et al . , 2015; Baker et al . , 2013; Chahrour et al . , 2008; Hong et al . , 2016; Lui et al . , 2016 ) . Assessing microglia function at synapses in these other disease-relevant models will be important future directions going forward . Ablating Mecp2 expression specifically in microglia had little effect on phenotypic regression ( Figure 4 ) . The minimal effect observed when comparing Cx3cr1CeER/+;Mecp2fl/y tamoxifen-treated mice to WT controls ( Cx3cr1CreER/+;Mecp2+/y ) is likely due to the Mecp2fl/y hypomorphic allele . The Mecp2fl/y mice have reduced Mecp2 expression and develop RTT phenotypes in the absence of Cre-mediated recombination ( Samaco et al . , 2008; Kerr et al . , 2008 ) , effects which may become apparent if the trajectory of phenotypes were assessed after P120 . Similarly , expression of Mecp2 specifically in microglia in an otherwise null animal ( Figure 5 ) had little to no effect on attenuation of any phenotype assessed . Together , our data are most consistent with the recent report that WT microglia/myeloid cells have no effect on phenotypes in Mecp2 null or mutant mice but rather phenotypes are more likely due to loss of Mecp2 expression in other CNS cell types such as neurons or astrocytes ( Wang et al . , 2015; Lioy et al . , 2011; Giacometti et al . , 2007; Luikenhuis et al . , 2004; Chao et al . , 2010; Ito-Ishida et al . , 2015 ) . Our data are in contrast to two other reports from another group that demonstrate introducing Mecp2 in microglia and other myeloid cells on a Mecp2 null background results in significant attenuation of phenotypes ( Derecki et al . , 2012; Cronk et al . , 2015 ) . The discrepancy may result from difference in paradigms used to express Mecp2 . For example , in our study , we induced recombination with tamoxifen in Cx3cr1CreER/+;Mecp2LSL/y mice at P21-P25 and assessed phenotypes at ≥P78 . This paradigm results in purely microglia-specific expression of Mecp2 due to ongoing hematopoiesis that replaces peripheral Mecp2-null cells with WT cells ( Goldmann et al . , 2013; Yona et al . , 2013 ) . In contrast , Cronk , Derecki et al . induced recombination in these same mice at 9 weeks ( ~P63 ) Cronk et al . , 2015 . This late tamoxifen administration may be necessary to observe significant effects on phenotypes and may result from expression of Mecp2 in peripheral myeloid cells . Furthermore , previous work by this same group demonstrated a significant attenuation of phenotypic regression in Mecp2 null mice after BM transplantation at P28 and engraftment with WT myeloid cells by ~P84 or with Cre mediated recombination ( LysmCre ) in myeloid cells from birth Derecki et al . , 2012 . These paradigms also affect Mecp2 expression in peripheral immune cells . Thus , we speculate that these divergent results may be due to differences in peripheral myeloid-derived cell-specific Mecp2 expression , which is intriguing and worthy of future investigation . It should also be noted that we did not measure the entire panel of phenotypic abnormalities ( breathing , open field , etc . ) or survival so it is unknown if our results differ in these contexts . Finally , it is unclear how recently published data using a similar BM chimerism strategy but a different Cre mouse ( Vav1-Cre ) resulted in contradictory results and is also worthy of follow-up investigation ( Wang et al . , 2015 ) . There have been conflicting reports regarding if and how microglia contribute to phenotypes in mouse models of RTT ( Derecki et al . , 2012; Cronk et al . , 2015; Wang et al . , 2015 ) . Our data offer significant insight into how microglia contribute to disease in Mecp2 null mice . While microglia-specific loss of Mecp2 is largely insufficient to induce synapse loss and phenotypic regression and gain of Mecp2 in expression in Mecp2 null mice is insufficient to attenuate these parameters , microglia contribute secondarily by dismantling synaptic circuits in complete Mecp2 null mice . Taken together with previously published data that single fiber strength decreases during early stages of phenotypic regression in the Mecp2 null retinogeniculate system ( Noutel et al . , 2011 ) , we propose that microglia dismantle neural circuits in the late phenotypic Mecp2 null brain by engulfing synapses previously rendered vulnerable and weakened by loss of Mecp2 expression in , most likely , neurons . Given that recent studies demonstrate the reversibility of circuit defects and phenotypes in RTT mouse models ( Derecki et al . , 2012; Lombardi et al . , 2015; Lioy et al . , 2011; Giacometti et al . , 2007; Luikenhuis et al . , 2004; Cronk et al . , 2015; Guy et al . , 2007; Jugloff et al . , 2008; Castro et al . , 2014; Garg et al . , 2013; Patrizi et al . , 2016; De Filippis et al . , 2015; Ma et al . , 2015 ) , identifying a molecular mechanism by which microglia dismantle circuits during late phenotypic stages and determining whether this is critical to end-stages of disease will be an important future directions with therapeutic potential .
Cx3cr1EGFP/+ , Ai9 ( RCL-tdT ) , MeCP2-/y ( Mecp2tm1 . 1Bird/y ) , Mecp2LSL/y ( Mecp2tm2Bird/y ) , and Mecp2fl/y ( Mecptm1Bird/y ) mice were obtained from Jackson Labs ( Bar Harbor , MA ) and Cx3cr1CreER/+ mice were obtained from Jonathan Kipnis , University of Virginia . All mice were maintained by breeding to C57BL/6J . For some engulfment experiments , MeCP2-/+ female mice were crossed with male Cx3cr1EGFP/EGFP mice . For Cre-lox experiments , Mecp2LSL/+ or Mecp2fl/+ female mice were crossed with male Cx3cr1CreER/CreER mice . All experiments using Cx3cr1EGFP/+ or Cx3cr1CreER/+ mice were performed with heterozygotes . Unless otherwise noted in figure legend , experiments were performed in male mice . For Cre-lox experiments , P21-P25 Cx3cr1CreER/+-expressing mice were injected with tamoxifen ( 20 mg/kg; Sigma Aldrich , Natick , MA ) or vehicle ( corn oil; Sigma Aldrich , Natick , MA ) subcutaneously two times , 48 hr apart , a protocol previously demonstrated to induce efficient recombination ( Goldmann et al . , 2013; Yona et al . , 2013 ) . All experiments were approved by institutional animal use and care committees and performed in accordance with all NIH guidelines for the humane treatment of animals . Analysis of engulfment was performed using previously published procedures ( Schafer et al . , 2012 , 2014 ) . Briefly , both eyes were injected with the same fluorophore-conjugated tracer ( either cholera toxin β subunit conjugated to Alexa dye 594 ( CTB-594 ) or 647 ( CTB-647 ) ( Life Technologies , Carlsbad , CA ) . Mice were then sacrificed 24 hr later . Brains were fixed in 4% paraformaldehyde ( PFA; EMS , Hatfield , PA ) for 3–4 hrs and 40 µm thick sections were prepared . Sections were further immonstained with antibodies against Iba-1 ( Wako Chemicals , Richmond , VA ) and/or CD68 ( AbD Serotec , Raleigh , NC ) to measure lysosomal content as previously described ( Schafer et al . , 2012 ) . For analysis of non-synaptic material , adjacent brain sections were immunolabeled with antibodies against Iba-1 ( Wako Chemicals , Richmond , VA ) , NeuN ( EMD Millipore , Darmstadt , Germany ) , and MAP2 ( EMD Millipore , Darmstadt , Germany ) . Sections were then imaged on a UltraView Vox spinning disk confocal microscope equipped with diode lasers ( 405 nm , 445 nm , 488 nm , 514 nm , 561 nm , and 640 nm ) and Volocity image acquisition software ( Perkin Elmer , Waltham , MA ) . Two LGN sections were imaged per animal and 4-63x fields of view were collected from the dorsal and ventral regions of each dLGN ( 8 fields of view total per animal ) . Images were subsequently processed in Image J ( NIH ) and analyzed using Imaris software ( Bitplane , Zurich , Switzerland ) as previously described ( Schafer et al . , 2012 , 2014 ) . Synapses were quantified similar to previously published work with modifications ( Schafer et al . , 2012 ) . Briefly , mice were either perfused with 4% PFA followed by a 2 hr drop fix in 4% PFA or fixed identical to those methods described for engulfment analysis ( see above ) . Tissue sections 15 or 40 µm ) were subsequently prepared and immunostained for synaptic proteins . Antibodies included anti-Homer1 ( Synaptic Systems GmbH , Goettingen , Germany ) , anti-Vesicular Glutamate Transporter 2 ( VGlut2; EMD Millipore , Darmstadt , Germany ) , and anti-VGlut1 ( EMD Millipore , Darmstadt , Germany ) followed by appropriate , species-specific secondary antibodies conjugated to Alexa dyes ( Life Technologies , Carlsbad , CA ) . Immunostained sections were imaged with a 63x Zeiss pan-Apochromat oil , 1 . 4 NA objective on a Zeiss LSM 700 Laser Scanning Confocal equipped with diode lasers ( 405 , 488 , 555 and 633 nm ) and Zen image acquisition software ( Carl Zeiss , Oberkochen , Germany ) . Alternatively , sections were imaged with a Leica SP8 X confocal ( Wetzlar , Germany ) equipped with multiple laser lines ( 405 , 458 , 488 , 496 , 514 , and 470–670 nm white light ) using a HC PL APO 63x/1 . 40 oil CS2 or a HC PL APO 40x/1 . 10 W motCORR ( only 40 µm-thick sections ) objective and LasX software . To maintain consistency across animals , the most medial dLGN sections were chosen for imaging . A total of 3 confocal z-stacks ( 1 µm spacing ) were then collected from dorsal , medial , and ventral regions of the dLGN section . For each z-stack , ( 2 confocal planes with the most robust DAPI staining were subsequently chosen and analyzed for blind analysis using Image J software ( NIH , Bethesda , MD ) . As a result , a total of 6 single confocal planes were analyzed for each animal . Fluorescent images of pre and/or postsynaptic markers were separated and thresholded blind . Density of thresholded pre and/or postsynaptic markers were calculated using the measure particles function where a puncta size was defined and maintained for all analyses across animals for each marker ( VGlut2 = 0 . 2-infinity; Homer1 = 0 . 1-infinity; VGlut1 = 0 . 1-infinity ) . The colocalization of puncta was quantified subsequently using the Image Calculator function applied to thresholded pre and postsynaptic images . The size and area of each puncta were recorded and then the total puncta area and average puncta size were calculated for each animal . The synapse or terminal densities were calculated by taking the total puncta area and dividing it by the total area of the field of view . The puncta density and puncta size were averaged across fields for each animal . Samples were prepared and imaged similar to methods described for engulfment and synapse density quantification ( see above ) . For microglia numbers , 10 x fields of vew were collected and cells were counted blind using the point tool in Image J . For cell death analysis , retinas were immunolabeled with antibodies against cleaved caspase 3 ( Cell Signaling Technology , Danvers , MA ) , TUJ1 ( BioLegend ( formerly Covance ) San Diego , CA ) or NeuN ( EMD Millipore , Darmstadt , Germany ) , mounted with media containing DAPI ( Vectashield; Vector Labs , Burlingame , CA ) , and 4 fields of view ( 20X ) were collected . Cells were counted blind for each field of view using the point tool in Image J . For validation of loss or gain of Mecp2 protein in microglia , a subset of tissue sections collected for engulfment or synapses analysis were selected and subjected to antigen retrieval using Retrievagen A ( BD Biosciences , San Jose , CA ) . Briefly , sections were microwaved ( power = 2 ) for 5 min in Retrievagen A solution . This was repeated once and then sections were washed 3 times with 0 . 1 M phosphate buffer . Sections were then immunostained with a rabbit antibody directed against the C-terminus of Mecp2 ( a generous gift from M . Greenberg Harvard Medical School ) ( Ballas et al . , 2009 ) and a chicken antibody against Iba-1 ( Abcam , Cambridge , MA ) , overnight at room temperature . Sections were then washed and HRP-conjugated rabbit and Alexa fluor-conjugated rat antibodies were added to the sections for 1–2 hr at room temperature . Sections were subsequently washed and an Alexa-fluor conjugated anti-HRP antibody was added overnight at 4 degrees . After the overnight incubation , sections were mounted and imaged . It should be noted that several other Mecp2 antibodies and staining conditions were attempted , but only this antibody and condition enabled us to detect Mecp2 protein even in WT microglia . Two 63x fields of view were collected in the lateral , medial , and ventral portions of the LGN per animal ( 3 animals per condition ) and images were assessed for Mecp2-positive microglia . Sorted cells were lysed and digested in TES buffer ( 10 mM Tris buffer , pH = 8 , 5 mM EDTA , 0 . 1 M NaCL , 0 . 5% SDS and 100 ug PK ) overnight in 56°C . DNA was precipitated in 70% ethanol for 30 mins at room temperature , centrifuged twice at top speed and the tubes were left to dry . Pellets were reconstituted with TE buffer for subsequent PCR . Genomic PCR for Mecp2 gene was performed using the following primers: 5'-TGGTAAAGA CCCATGTGACCCAAG-3' , 5'-GGCTTGCCACATGACAAGAC-3' , 5'-TCCACCTAG CCTGCCTGTACTTTG-3' . Brain samples were harvested from individual mice and tissues were homogenized and incubated with a HBSS solution containing 2% BSA ( Sigma-Aldrich ) , 1 mg/ml collagenase D ( Roche ) , and 0 . 15 mg/ml DNase1 , filtered through a 70 µm mesh . Homogenized sections were filtered through 80 μM wire mesh and resuspended in 40% Percoll , prior to density centrifugation ( 1000 x g . 15 min at 20°C with low acceleration and no brake ) . Cells were acquired on LSRFortessa systems ( BD ) and analyzed with FlowJo software ( Tree Star ) . For cell sorting , the FacsAria ( BD ) was used . Antibodies used include: CD11b ( clone M1/70; AbD Serotec , Raleigh , NC ) , CD45 ( clone 30F11; AbD Serotec , Raleigh , NC ) , and MeCP2 ( EMD Millipore , Darmstadt , Germany ) . Acuity was measured blind using methods identical to those previously described ( Durand et al . , 2012; Prusky et al . , 2004 ) . Rotarod performance was measured blind using methods similar to those previously described ( Derecki et al . , 2012; Crawley , 2008 ) . One day prior to training , mice were acclimated to a non-accelerating rotarod 5 RPM for 5–10 min . The following day , the animals were tested for performance ( latency to fall ) on an accelerating rotarod over 5 trials , which were subsequently averaged to plot an average latency to fall for each animal . Neurological scores were recorded blind using methods similar to those previously described ( Derecki et al . , 2012; Crawley , 2008 ) . Mice were scored on a scale from 0 to 2 , with ‘0’ being no phenotype , and ‘2’ being severe phenotype . For gait , mice were assessed for wide-spread hind limbs and waddling while locomoting . Hind limb clasping was assessed by suspending mice by the tail and assessing clenching of hind limbs across the ventral aspect of the body . Tremor was characterized as a visible involuntary shaking and was scored based on the severity . Appearance was scored based on the presence or lack of grooming and/or hunched posture . The scores were subsequently summed to give a neurological score . For all statistical analyses , GraphPad Prism 5 software ( La Jolla , CA ) was used . Analyses used include unpaired Student’s t-test , one-way ANOVA , or two-way ANOVA with 95% confidence and appropriate post hoc analyses ( indicated in figure legends ) . All p and N values are indicated in figure legends . All N’s represent biological replicates ( number of mice used for the study ) . Sample size was chosen based on our previous work analyzing engulfment and synapse density and work by other groups assessing phenotypic changes in Mecp2 mutant mice ( Derecki et al . , 2012; Cronk et al . , 2015; Schafer et al . , 2012; Guy et al . , 2001; Schafer et al . , 2014; Durand et al . , 2012; Lioy et al . , 2011; Chao et al . , 2010; Guy et al . , 2007; Patrizi et al . , 2016 ) . | Rett Syndrome is a neurodevelopmental disorder with symptoms that typically begin in girls between 6 and 18 months old . Those affected developmentally stagnate and regress – during which they lose some of their previously acquired skills and develop an array of physical impairments . Mutations in a gene called Mecp2 on the X chromosome cause most cases of Rett Syndrome . Mice that lack the Mecp2 gene develop symptoms similar to those seen in people with Rett Syndrome , and so such “Mecp2 null” mice are often used to study the disorder . Microglia , the resident immune cells of the central nervous system , have been implicated in the development of Rett Syndrome . Introducing microglia that carry the Mecp2 gene into Mecp2 null mice has been shown to reduce several disease-associated abnormalities . However , exactly how microglia contribute to these changes remains unknown . In addition , a more recent report failed to reproduce these findings , and instead obtained results suggesting that microglia do not affect the development of Rett syndrome . Schafer et al . now use the mouse visual system as a model to determine if and how microglia contribute to the development of Rett Syndrome . Like many other brain regions , the developing visual system initially has a surplus of connections between neurons , or synapses , which are subsequently pruned back . Schafer et al . previously showed in the developing visual system of early postnatal ( 5 days after birth ) control mice ( who express the Mecp2 gene ) that microglia contribute to this pruning by engulfing and eliminating a subset of these excessive synaptic connections . The new experiments by Schafer et al . show that another wave of microglia-mediated synaptic pruning occurs in 40-day-old juvenile control mice . Because Mecp2 null mice begin to display features of Rett Syndrome when they’re about 40 days old , Schafer et al . tested whether the microglia of these animals inappropriately prune synaptic connections . While this process occurred normally in neonatal and juvenile Mecp2 null mice , microglia began to excessively engulf cells in Mecp2 null mice when they were around 56 days old . Unexpectedly , deleting or reintroducing the Mecp2 gene solely in the microglia of these mice had little effect on pruning activity of the microglia , and failed to affect Rett-syndrome-like symptoms in the mice . Taken together , the data presented by Schafer et al . suggest how microglia contribute to the final stages of Rett Syndrome: by dismantling circuits of neurons that are rendered vulnerable by the loss of the Mecp2 gene in other cell types . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Microglia contribute to circuit defects in Mecp2 null mice independent of microglia-specific loss of Mecp2 expression |
Rapid repair of plasma membrane wounds is critical for cellular survival . Muscle fibers are particularly susceptible to injury , and defective sarcolemma resealing causes muscular dystrophy . Caveolae accumulate in dystrophic muscle fibers and caveolin and cavin mutations cause muscle pathology , but the underlying mechanism is unknown . Here we show that muscle fibers and other cell types repair membrane wounds by a mechanism involving Ca2+-triggered exocytosis of lysosomes , release of acid sphingomyelinase , and rapid lesion removal by caveolar endocytosis . Wounding or exposure to sphingomyelinase triggered endocytosis and intracellular accumulation of caveolar vesicles , which gradually merged into larger compartments . The pore-forming toxin SLO was directly visualized entering cells within caveolar vesicles , and depletion of caveolin inhibited plasma membrane resealing . Our findings directly link lesion removal by caveolar endocytosis to the maintenance of plasma membrane and muscle fiber integrity , providing a mechanistic explanation for the muscle pathology associated with mutations in caveolae proteins .
Ca2+ entry in wounded cells triggers a repair mechanism that reseals the plasma membrane ( PM ) within a few seconds ( McNeil et al . , 2003 ) . Ca2+ influx induces exocytosis of lysosomes , a process required for PM resealing ( Reddy et al . , 2001 ) . PM repair was initially suggested to be mediated by a membrane patch applied to the wound site ( Miyake and McNeil , 1995 ) , or through exocytosis-mediated reduction in PM tension ( Togo et al . , 2000 ) . However , Ca2+-dependent lysosomal exocytosis is also required for the resealing of cells injured by pore-forming toxins ( Walev et al . , 2001; Idone et al . , 2008 ) . These toxins generate stable , protein-lined transmembrane lesions that cannot be resealed by a membrane patch or simply by relieving PM tension . Recent studies clarified this issue , by showing that lysosomal exocytosis in wounded cells is followed by a rapid , cholesterol-dependent form of endocytosis that removes pores and lesions from the PM ( Idone et al . , 2008 ) and directs them to lysosomes for degradation ( Corrotte et al . , 2012 ) . Ca2+-triggered exocytosis of the lysosomal enzyme acid shingomyelinase ( ASM ) is required for the endocytic process that promotes wounded cell resealing . Transcriptional silencing of ASM abolished PM repair , and addition of exogenous ASM restored resealing ( Tam et al . , 2010 ) . These findings provided a novel conceptual framework for the mechanism of PM repair , indicating that exocytosis promotes resealing not by generating a membrane patch , but by releasing enzymes that remodel the outer leaflet of the PM and stimulate endocytosis . ASM converts the abundant PM lipid sphingomyelin into ceramide ( Schissel et al . , 1996 ) , and ceramide-enriched microdomains can trigger invagination of lipid bilayers ( Holopainen et al . , 2000; Trajkovic et al . , 2008 ) . However , the exact role of ceramide and the nature of the endocytic vesicles triggered by cell wounding are still unknown . PM repair is of paramount importance in muscle . Muscle fibers are frequently injured in vivo ( McNeil and Khakee , 1992 ) and failure to repair the sarcolemma causes muscular dystrophy ( Bansal and Campbell , 2004 ) . Intriguingly , the fragile muscle fibers from patients with Duchenne muscular dystrophy contain elevated numbers of caveolae-like vesicles ( Bonilla et al . , 1981; Repetto et al . , 1999 ) , and mutations in the muscle-specific caveolar protein caveolin-3 ( Cav3 ) cause multiple forms of muscle pathology ( Gazzerro et al . , 2010 ) . These observations , taken together with the recently uncovered role for endocytosis in PM repair ( Idone et al . , 2008 ) , raised the possibility that caveolae-derived endocytic vesicles might play a direct , heretofore unrecognized role in the mechanism responsible for resealing PM wounds . In this study we show that injury-induced internalization of caveolae-derived vesicles is a dynamic process essential for the restoration of PM integrity .
Prior studies suggested that ASM released from lysosomes during cell wounding triggers formation of ceramide-enriched endocytic vesicles ( Tam et al . , 2010 ) . By performing cryo-immuno EM assays with specific anti-ceramide antibodies ( Fernandes et al . , 2011 ) we detected anti-ceramide reactivity throughout the cytoplasm and in small clusters near the PM of NRK cells ( Figure 1A , B , control ) . An isotype control antibody showed little , if any , labeling in the same preparations ( not shown ) . Treatment with purified Bacillus cereus sphingomyelinase ( SM ) for 30 s enhanced the anti-ceramide staining along the PM . Permeabilization with the pore-forming toxin streptolysin O ( SLO ) had a similar effect , rapidly increasing the anti-ceramide reactivity at the cell periphery ( Figure 1A , B ) . These results suggested that injury with SLO or exposure to SM triggered the formation of ceramide-enriched structures that might represent PM invaginations or intracellular vesicles . 10 . 7554/eLife . 00926 . 003Figure 1 . Caveolae-like vesicles accumulate in cells exposed to SLO and sphingomyelinase . ( A ) Cryo-immuno EM with anti-ceramide in NRK cells untreated or exposed to SLO or SM for 30 s . Bars: 100 nm . Arrows: patches of ceramide staining near the PM . ( B ) Quantification of anti-ceramide label in cells treated as in ( A ) . All gold particles ( 2522–6876 ) within an area of 200 nm along the PM were counted in 14–31 cell sections . Data represent mean ± SEM of gold particles/cell section . *p=0 . 023 , ***p<0 . 001 . The results are representative of two independent experiments . ( C ) TEM of NRK cells exposed or not to SLO+Ca2+ or SM in the presence of BSA-gold . Arrows: <80 nm vesicles with BSA-gold . Arrowheads: merged vesicles . Bars: 100 nm . ( D ) Quantification of vesicles with BSA-gold in control , SLO or SM-treated cells after 30 s . All vesicles containing BSA-gold ( 191–485 ) were counted in 20 cell sections/sample . Data represent mean ± SEM of BSA-gold-containing vesicles/cell section . ***p<0 . 001 . The results are representative of two independent experiments . ( E ) Numbers of BSA-gold positive <80 nm and >80 nm vesicles over time in SLO treated cells . Data represent mean ± SEM of vesicles/cell section . *p=0 . 033 , **p=0 . 004 , ***p<0 . 001 ( comparison with <80 nm vesicles in the same time point ) . ( F ) Average area of BSA-gold positive vesicles over time . Data represent mean ± SEM of vesicle area/cell section . ***p<0 . 001 ( comparison with 30 s time point ) . ( G ) BSA-gold particles detected within <80 nm and >80 nm vesicles over time . Data represent mean ± SEM of gold particles . **p=0 . 0019 ( comparison with <80 nm vesicles in the same time point ) . From ( E ) to ( G ) , all gold-containing vesicles ( 73–142 ) were quantified in 14–47 cell sections . ( H ) TEM of NRK cells untreated ( control ) or treated with ASM in the presence of BSA-gold as an endocytic tracer . Arrows point to <80 nm vesicles containing BSA-gold; arrowheads point to vesicle fusion profiles . Bars: 100 nm . ( I ) Quantification of BSA-gold containing vesicles over time in cells treated or not with ASM . All BSA-gold carriers ( 58–309 ) were counted in 10–20 sections . Data represent mean ± SEM of BSA-gold-containing vesicles/cell section . *p=0 . 03–0 . 04 , **p=0 . 005 ( comparison with controls in each time point ) . All datasets were compared using an unpaired Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 00310 . 7554/eLife . 00926 . 004Figure 1—figure supplement 1 . Transcriptional silencing of ASM inhibits intracellular accumulation of caveolae-like vesicles after SLO injury . ( A ) TEM of control and ASM siRNA-treated HeLa cells incubated or not with SLO for 60 s . Arrows: <80 nm profiles . Bars: 100 nm . ( B ) Number of <80 nm vesicular profiles/µm in H . All vesicles ( 127–216 ) <80 nm diameter were counted in 40 random fields/sample and normalized by PM length . Data represent mean ± SEM of vesicles/cell section . *p=0 . 021; **p=0 . 004 ( comparisons with control condition or control siRNA ) , unpaired Student's t test . The results are representative of two independent blinded quantifications performed by two independent investigators . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 004 To directly visualize newly formed structures , we examined cells by transmission electron microscopy ( TEM ) at increasing periods after permeabilization with SLO or exposure to SM . Previous TEM studies detected numerous large , irregularly shaped endocytic vesicles in cells fixed 4–5 min after SLO permeabilization ( Idone et al . , 2008 ) . Surprisingly , when cells were examined just 30 s after treatment with SLO or SM , the newly formed endocytic vesicles ( identified by luminal BSA-gold added as an endocytic tracer ) appeared as homogeneously round and small ( <80 nm ) . Similar peripheral <80 nm endocytic vesicles were present in untreated cells , albeit in lower numbers ( Figure 1C ) . Quantification revealed that treatment with SLO or SM for 30 s increased the number of BSA-gold-containing vesicles relative to controls ( Figure 1D ) . Clathrin-coated vesicles in the same preparations did not contain BSA-gold , in agreement with the slower rate of formation of this class of endocytic vesicles ( results not shown ) . At later time points ( 60 and 180 s ) larger compartments suggestive of homotypic fusion of the <80 nm vesicles were increasingly observed ( Figure 1C ) . Quantification of vesicle size , area and BSA-gold content supported the conclusion that the small endocytic vesicles induced by exposure to SLO or SM increase in size over time ( Figure 1E–G ) . Notably , the number of <80 nm vesicles containing the endocytic tracer BSA-gold also increased when cells were treated with recombinant human ASM ( He et al . , 1999 ) ( Figure 1H , I ) . Furthermore , transcriptional silencing of ASM reduced the number of peripheral <80 nm vesicles seen by TEM in cells exposed to SLO+Ca2+ ( Figure 1—figure supplement 1 ) . These results reinforce the view that ASM released through lysosomal exocytosis in wounded cells can generate ceramide on the outer leaflet of the PM ( Schissel et al . , 1998 ) and promote endocytosis ( Tam et al . , 2010 ) . The newly-formed endocytic vesicles observed in SLO or SM-treated cells strongly resembled caveolae , the flask-like PM invaginations enriched in cholesterol and sphingolipids that are present in many cell types ( Palade , 1953; Parton and Simons , 2007 ) . To investigate a potential role of caveolae-derived vesicles in the internalization of SLO pores , cells were permeabilized with GFP-tagged SLO ( which retains full pore-forming activity [Idone et al . , 2008] ) and analyzed by cryo-immuno EM using antibodies against GFP or the caveolae-associated protein caveolin-1 ( Cav1 ) ( Drab et al . , 2001 ) . The amount of GFP-SLO associated with flat regions of the PM gradually decreased over time , consistent with a toxin internalization process ( Figure 2A , B ) . Importantly , during the first 60 s after injury GFP-SLO was increasingly detected on <80 nm vesicles containing Cav1 , which are properties of caveolae ( Figure 2A , C ) . By 300 s the amount of SLO co-localizing with Cav1 decreased , in agreement with the previously described traffic of internalized SLO into later compartments of the endocytic pathway ( Corrotte et al . , 2012 ) . The number of <80 nm vesicles positive for Cav1 alone or SLO alone also decreased over time , simultaneously with an increase in the number of >80 nm vesicles containing either Cav1 alone or both Cav1 and SLO ( Figure 2C , D ) . These results are fully consistent with our TEM analysis of SLO-permeabilized cells , showing an initial increase in the number of <80 nm endocytic vesicles followed by the gradual appearance of merged compartments ( Figure 1C ) . 10 . 7554/eLife . 00926 . 005Figure 2 . SLO is internalized in Cav1-positive caveolae-like vesicles that separate from the PM . ( A ) Cryo-immuno EM localization of GFP-SLO and Cav1 in NRK cells . 5 nm gold: anti-GFP ( arrowheads ) ; 10 nm gold: anti-Cav1 ( arrows ) . Bars: 100 nm . ( B–D ) Quantification of the relative amount of GFP-SLO and/or Cav1 on flat PM structures ( B ) , vesicular profiles <80 nm ( C ) or vesicular profiles >80 nm ( D ) . All labeled structures ( 17–280 ) in 80 random fields were counted and the data expressed as % of total antibody-stained structures . Data represent mean ± SEM of labeled structures/cell section . *p=0 . 039–0 . 052 , **p=0 . 006–0 . 007 , ***p<0 . 001 , unpaired Student’s t test . The results are representative of two independent experiments . ( E ) FACS analysis of NRK cells exposed to Alexa 488-SLO at 4°C for 5 min , followed by anti-Alexa Fluor 488 quenching antibodies for 2 min . The percentage of quench-protected toxin fluorescence above the endogenous cellular background level ( BG ) is indicated . ( F ) FACS analysis of NRK cells exposed to Alexa 488-SLO ± Ca2+ at 37°C for 5 min , followed by PI staining . The percentage of PI-negative cells is indicated . ( G ) FACS analysis of NRK cells exposed to Alexa 488-SLO + Ca2+ at 37°C for 5 min , followed by anti-Alexa Fluor 488 quenching antibodies for 2 min . The profile shown corresponds to the PI-negative cell population shown in ( F ) . The percentage of quench-protected toxin fluorescence above the endogenous cellular background level ( BG ) is indicated . Dashed lines , no Alexa 488-SLO background controls . The results are representative of at least five independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 005 The results described above strongly suggested that SLO pores were removed from the PM by internalization in Cav1-positive caveolar vesicles . However , the possibility still remained that caveolar vesicles containing SLO might still be connected to the PM . To directly address this issue , we developed a flow cytometry ( FACS ) assay for detection of cell-associated fluorescent SLO before and after quenching with specific antibodies . To ensure accessibility of the fluorescent moiety to the quenching antibodies , we used a single cysteine SLO mutant labeled with Alexa Fluor 488 at the N-terminus ( a region not inserted into membranes during pore-formation [Shatursky et al . , 1999] ) . When cells were incubated with the labeled toxin at 4°C , addition of anti-Alexa Fluor 488 antibodies quenched >90% of the fluorescence ( Figure 2E ) . Thus , under conditions that allow toxin binding but not endocytosis , most cell-associated toxin is accessible to quenching antibodies . When the same amount of labeled toxin was added to cells at 37°C , cells were fully permeabilized and about 30% resealed in the presence of Ca2+ under the assay conditions , as indicated by propidium iodide ( PI ) exclusion ( Figure 2F ) . By gating on the PI-negative ( resealed ) cell population we found that at least 50% of the cell-associated Alexa 488-SLO was protected from quenching ( Figure 2G ) , indicating that it entered compartments no longer in contact with the extracellular medium . Antibody-mediated quenching of extracellular Alexa 488-SLO was also observed by confocal microscopy ( Figure 3A ) , and quench-protected intracellular toxin colocalized with Cav1-positive puncta ( Figure 3B ) . 10 . 7554/eLife . 00926 . 006Figure 3 . Internalized SLO colocalizes with Cav1 . ( A ) HeLa cells were pre-incubated with 3 µg/ml Alexa 488-SLO ( green ) for 5 min at 4°C , washed and either kept at 4°C ( 0 s ) or incubated at 37°C in DME+Ca2+ ( 180 s ) , followed or not by anti-Alexa Fluor 488 quenching antibodies ( blue ) for 30 min at 4°C . After fixation cells were permeabilized , labeled with anti-Cav1 antibodies ( red ) and analyzed by confocal microscopy ( single optical sections are shown ) . Bars: 10 µm . ( B ) Higher magnification images of cells treated as in A and incubated for 180 s at 37°C with quenching antibodies . Arrows: Vesicular carriers positive for both Cav1 and Alexa 488-SLO . Bars: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 006 To obtain dynamic data on the behavior of SLO carriers , we performed live imaging of cells expressing mRFP-Cav1 and permeabilized with GFP-SLO . The transfected mRFP-Cav1 was detected on peripheral punctate structures , in a pattern indistinguishable from the distribution of endogenous Cav1 ( Figure 4A ) . A few seconds after pore formation was triggered by increasing the temperature , GFP-SLO was observed moving into cells in structures containing mRFP-Cav1 ( Figure 4B , arrowheads; Videos 1 , 2 , 3 ) . Some Cav1-positive SLO carriers merged intracellularly while moving deeper into cells ( Figure 4B arrows; Video 3 ) . Kymograph analysis detected numerous Cav1-positive SLO carriers entering a peripheral intracellular region of SLO-permeabilized cells ( Figure 4C , lower panels ) . In contrast , only a few intracellular Cav1-containing structures were detected at the periphery of untreated cells ( Figure 4C , upper panels ) . While some GFP-SLO signal was present in compartments with no detectable Cav1 , the majority of SLO carriers detected in the kymographs co-localized with the caveolae marker ( Figure 4C , lower panels ) . 10 . 7554/eLife . 00926 . 007Figure 4 . SLO enters cells associated with Cav1-positive carriers . ( A ) Confocal optical section at the bottom surface of a HeLa cell transfected with mRFP-Cav1 ( red ) and stained with anti-Cav1 antibodies ( green ) . ( B ) Time-lapse images of HeLa cells expressing mRFP-Cav1 and imaged for 300 s after incubation with GFP-SLO+Ca2+ . Dotted line: PM . Arrowheads: Cav1/SLO carrier moving rapidly into cell . Arrows: Cav1/SLO carriers that merge after internalization . Bars: 1 µm . See also Figure 2 and Videos 1 , 2 , 3 . ( C ) Kymographs of mRFP-Cav1 and GFP-SLO fluorescence along a line at the cell periphery . PDM: Positive Difference of the Mean ( co-localization index ) . Bars: 10 µm . The results are representative of seven independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 00710 . 7554/eLife . 00926 . 008Video 1 . Internalization and lateral movement of SLO/Cav1 carriers , ( related to Figure 4 ) . HeLa cells expressing mRFP-Cav1 ( red ) were pre-incubated for 5 min with 800 ng/ml of GFP-SLO ( green ) at 4°C and transferred in cold DMEM+Ca2+ to a live imaging chamber at 37°C , to allow for progressive warming , pore formation , and PM repair . Images were acquired for 5 min at 2 s/frame on a spinning disk confocal microscope . The video shows a vesicular structure positive for Cav1 and SLO that appears to separate rapidly from the PM and move laterally along the PM before disappearing into a different focal plane . Video is displayed at 6 . 67 frames/s . Dotted line: PM . Arrow indicates Cav1/SLO carrier . Bars: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 00810 . 7554/eLife . 00926 . 009Video 2 . Internalization and trafficking of SLO/Cav1 carriers ( related to Figure 4 ) . HeLa cells were treated as in video 1 . The video shows a Cav1 positive structure on the PM that accumulates SLO before being internalized and moving rapidly into the cell . Video is displayed at 6 . 67 frames/s . Dotted line: PM . Arrow indicates Cav1/SLO positive vesicle . Bars: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 00910 . 7554/eLife . 00926 . 010Video 3 . Intracellular merger and rapid internalization of SLO/Cav1 carriers ( related to Figure 4 ) . HeLa cells were treated as in video 1 . The video shows two separate SLO and Cav1 positive carriers close to the PM ( arrows ) that merge before rapidly moving further into the cell . The same cell contains a SLO/Cav1 positive structure ( arrowhead ) that suddenly separates from the PM and moves rapidly into the cell . Video is displayed at 6 . 67 frames/s . Dotted line: PM . Bars: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 010 The ATPase EHD2 recently emerged as a regulator of caveolae dynamics . By linking caveolae to actin filaments , EHD2 was proposed to retain caveolae on the PM , preventing their internalization ( Moren et al . , 2012; Stoeber et al . , 2012 ) . As an additional independent assay for detecting SLO-induced caveolar endocytosis , we examined the co-localization of EHD2 with Cav1 clusters at the PM by TIRF microscopy , after cells were permeabilized with SLO or treated with SM . Most of the Cav1 staining at the surface of untreated cells appeared as puncta that were also positive for EHD2 , as previously described for the stable population of PM-associated caveolae ( Stoeber et al . , 2012 ) . The intensity of Cav1 puncta decreased with time after SLO or SM exposure , reflecting an inward movement away from the bright TIRF field adjacent to the PM ( Figure 5A , B ) . In addition , cells permeabilized with SLO or exposed to SM showed a progressive loss in Cav1-EHD2 co-localization ( Figure 5A , C ) , as expected for caveolar endocytosis . 10 . 7554/eLife . 00926 . 011Figure 5 . EHD2-Cav1 colocalization is decreased after caveolar endocytosis triggered by SLO or sphingomyelinase . ( A ) TIRF images of Cav1 and EHD2 immunostaining in HeLa cells treated or not with SLO or SM for 30 or 60 s . Bars : 50 µm . Arrows: Cav1 positive , EHD2 negative puncta corresponding to internalized caveolar vesicles . ( B ) Quantification of Cav1 fluorescence intensity in cells treated as in C . Quantifications were performed on 53–79 cells/sample . **p=0 . 002 , ***p<0 . 001 , unpaired Student’s t test . ( C ) Colocalization of Cav1 with EHD2 in cells treated as in B . Quantifications were performed on 65–82 cells/sample . Data represent mean ± SEM of values/cell . **p=0 . 009 , ***p<0 . 001 , unpaired Student’s t test . The results in A–C are representative of four independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 011 Collectively , our assays showing co-localization of endogenous Cav1 and GFP-SLO on <80 nm vesicles ( Figure 2A–D ) , detection of internalized Alexa 488-SLO after extracellular quenching ( Figure 2E–G and 3 ) , live imaging of GFP-SLO entering cells in vesicles containing mRFP-Cav1 ( Figure 4B , C and Videos 1–3 ) and loss of Cav1 and EHD2 co-localization after exposure to SLO ( Figure 5A–C ) support the conclusion that the pore-forming toxin SLO is removed from the PM in Cav1-positive , caveolae-derived endocytic vesicles . Furthermore , all assays also indicate that SLO internalization occurs within a few seconds of pore formation , coinciding with the known kinetics of PM resealing ( Steinhardt et al . , 1994; Idone et al . , 2008 ) . Cav1 is required for caveolae assembly in many cell types ( Drab et al . , 2001; Parton and Simons , 2007 ) . To examine the requirement for caveolae in PM repair , we transcriptionally silenced Cav1 expression ( Figure 6A ) and examined the ability of cells to reseal after SLO injury using a live imaging assay that follows the influx of the lipophilic dye FM1–43 ( Idone et al . , 2008; Tam et al . , 2010 ) . FM1–43 was detected only on the PM of cells not exposed to SLO , consistent with an intact lipid bilayer . After treatment with SLO in the absence of Ca2+ ( a condition that does not allow PM repair ) there was massive FM1–43 influx , reflecting rapid PM permeabilization ( Figure 6B , Video 4 ) . Quantification of FM1–43 influx showed that cells treated with control or Cav1 siRNA were similarly susceptible to SLO permeabilization ( Figure 6C ) . In the presence of Ca2+ , SLO did not trigger FM1–43 influx in cells treated with control siRNA , as expected from the rapid Ca2+-dependent resealing process ( Idone et al . , 2008; Tam et al . , 2010 ) . In contrast , FM1–43 flowed rapidly into cells treated with Cav1 siRNA even in the presence of Ca2+ , reflecting defective PM repair ( Figure 6B , C ) . 10 . 7554/eLife . 00926 . 012Figure 6 . PM repair and endocytosis of SLO in caveolar vesicles are Cav1-dependent . ( A ) Western blot with anti-Cav1 and anti-actin ( loading control ) in NRK cell lysates treated with control or Cav1 siRNA . ( B ) Live imaging of FM1–43 influx in NRK cells treated with control or Cav1 siRNA , with and without SLO ± Ca2+ . Bars: 10 μm . See Video 4 . ( C ) Quantification of intracellular FM1–43 fluorescence in B . Data represent mean ± SEM of fluorescence intensity/cell . The results are representative of four independent experiments . ( D ) TEM of control and Cav1 siRNA-treated NRK cells incubated or not with SLO or SM for 30 s . Bars: 100 nm . ( E ) Number of <80 nm vesicular profiles/µm in cells treated as in D . All vesicles <80 nm diameter ( 1411–3912 ) were counted in 20–30 sections/sample and normalized by PM length . Data represent mean ± SEM of vesicles/cell section . **p=0 . 003 , ***p<0 . 001 , unpaired Student’s t test . The results are representative of two independent blinded quantifications performed by two independent investigators . ( F ) Cryo-immuno EM localization of Cav1 in NRK cells . All labeled structures ( 281–418 ) in 80 random fields were counted and the data expressed as % of total antibody stained structures . Data represent mean ± SEM of gold particles/cell section . ***p<0 . 001 ( comparison with 0 s condition ) , unpaired Student’s t test . The panels on the left show representative images of Cav1 staining ( arrows , 10 nm gold label ) on PM and <80 nm vesicular profiles . Bars: 100 nm . The results are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 01210 . 7554/eLife . 00926 . 013Video 4 . RNAi-mediated silencing of Cav1 expression inhibits PM repair , allowing sustained FM1–43 influx into SLO-permeabilized NRK cells ( related to Figure 6 ) . NRK cells treated with control or Cav1 siRNA were left untreated ( no SLO ) or pre-incubated with SLO at 4°C and transferred to a live imaging chamber at 37°C in the presence or absence of Ca2+ , followed by addition of FM1–43 and time-lapse imaging in a spinning disk confocal microscope for 4 min at 1 frame/3 s . Video is displayed at 10 frames/s . Bar: 18 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 013 All <80 nm vesicles with caveolae-like morphology observed along the cell periphery were quantified by TEM , and the results indicated that SLO or SM exposure increases the number of caveolae-like vesicles in control cells , but not after treatment with Cav1 siRNA ( Figure 6D , E ) . Thus , injury with SLO or exposure to the ceramide-generating enzyme SM increase the population of peripheral caveolae-like vesicles in a Cav1-dependent manner , a process that correlates with the ability of these cells to reseal their PM . Quantification by cryo-immuno EM revealed that ∼20% of the Cav1 detected on sections of NRK cells was initially associated with flat PM regions without caveolae , and this fraction was reduced to ∼9% after 60 s of exposure to SLO+Ca2+ , along with an increase in labeling of intracellular <80 nm vesicles ( Figure 6F ) . These results suggest that the intracellular accumulation of caveolae-like vesicles after exposure to SLO or SM results from internalization of pre-existing and , to a certain extent , of de novo assembled caveolae . Surprisingly , we found no requirement for dynamin-1 and -2 in the internalization of caveolar vesicles triggered by SLO permeabilization or SM exposure . Cells depleted in dynamin-2 with Dyn2 siRNA ( Figure 7A ) were strongly deficient in transferrin endocytosis ( Figure 7B ) , but fully capable of repairing their PM after permeabilization with SLO+Ca2+ ( Figure 7C ) . SLO-treated Dyn2-deficient cells contained numerous caveolae-like <80 nm vesicles that were not stained by externally added ruthenium red , consistent with a complete fission from the PM . Clathrin-coated vesicles in the same preparations were strongly stained with extracellular ruthenium red and showed elongated ‘necks’ in continuity with the PM , typical features of dynamin deficiency ( Figure 7D ) . In sharp contrast , no similar structures were observed associated with caveolae . To examine a potential compensatory role of dynamin-1 , we performed similar assays using an inducible fibroblast cell line generated from dynamin-1 and -2 double conditional knockout mice ( Ferguson et al . , 2009 ) . After tamoxifen induction these cells became strongly depleted in both dynamin-1 and dynamin-2 ( Figure 7E ) and defective in transferrin endocytosis ( Figure 7F ) . However , after exposure to SLO+Ca2+ or SM these cells were still fully capable of blocking FM1–43 and propidium iodide ( PI ) influx ( Figure 7G ) and of upregulating endocytosis of the B subunit of cholera toxin ( which is internalized by caveolar and other forms of endocytosis [Kirkham et al . , 2005; Chinnapen et al . , 2007] ) ( Figure 7H ) . These results suggest that caveolar endocytosis induced by PM injury or by exposure to the ceramide-generating enzyme SM may occur independently of dynamin function . 10 . 7554/eLife . 00926 . 014Figure 7 . Depletion in dynamin-1 and -2 does not inhibit PM repair and SLO or SM-induced internalization of caveolar vesicles . ( A ) Western blot with anti-dynamin-2 and anti-actin ( loading control ) in lysates of NRK cells treated with control or Dyn2 siRNA . ( B ) Quantification of transferrin uptake in NRK cells treated with control or Dyn2 siRNA . Data represent mean ± SEM of fluorescence intensity/microscopic field . ***p<0 . 001 . Inset images: red , transferrin; blue , DAPI-stained nuclei . ( C ) Quantification of FM1–43 influx in NRK cells treated with control or Dyn2 siRNA , and exposed or not to SLO ± Ca2+ . Data represent mean ± SEM of fluorescence intensity/cell . The results are representative of three independent experiments . ( D ) TEM of Dyn2 siRNA-treated cells exposed to SLO or ( SM ) for 60 s , fixed and stained with ruthenium red . Small arrows: unstained caveolae . Wide arrows: stained clathrin-coated vesicles and elongated endocytic structures connected to the PM . Bars: 100 nm . ( E ) Western blot with anti-dynamin-1 , anti-dynamin-2 , and anti-actin ( loading control ) in lysates of cells derived from dynamin-1 and -2 conditional double knockout mice , induced with tamoxifen ( DKO ) or not induced ( Control ) . ( F ) Quantification of transferrin uptake in control or dynamin-1-2 DKO cells . Data represent mean ± SEM of fluorescence intensity/microscopic field . **p=0 . 003 . Inset images: red , transferrin; blue , DAPI-stained nuclei . ( G ) Time-lapse imaging of FM1–43 and PI influx during permeabilization with SLO ± Ca2+ in control and dynamin-1 and -2 DKO cells , ( H ) Quantification of CTxB-A488 uptake by control and dynamin-1-2 DKO cells exposed or not to SLO or SM for 180 s . Data represent mean ± SEM of fluorescence intensity/microscopic field . *p=0 . 014 , ***p<0 . 001 ( comparison with respective control conditions ) , unpaired Student’s t test . All results in this figure are representative of three or more independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 014 We also assessed the involvement of caveolar endocytosis in the resealing of mechanical lesions . When cells were injured by exposure to glass beads , wounds were visualized by TEM as sites where ruthenium red entered the cytosol ( Figure 8A , B , Figure 8—figure supplement 1 , large arrows ) . ‘Hot spots’ of caveolae-like vesicle accumulation , often arranged in rows , were frequently seen close to wound sites . Several peripheral caveolae-like vesicles observed 30–60 s after wounding appeared disconnected from the PM , based on the lack of luminal staining with ruthenium red added during fixation ( Figure 8B , Figure 8—figure supplement 1A , small arrows ) . At later time points , more complex structures suggesting homotypic fusion of caveolar vesicles were also observed , and some of these merged compartments were still connected to the PM , as indicated by luminal staining with ruthenium red ( Figure 8B , 180 s ) . 10 . 7554/eLife . 00926 . 015Figure 8 . Caveolae accumulate at sites of mechanical wounding , and Cav1 is required for mechanical wound repair . ( A ) TEM of NRK cells wounded with glass beads and stained with ruthenium red during fixation . Numerous caveolae-like vesicles ( arrows , lower magnified image ) are visible near the wound , identified by ruthenium red influx in cell #1 ( large arrow , upper image ) . Two non-wounded cells are present in the same field ( #2 and #3 , upper image ) . Bar: 500 nm . ( B ) TEM of NRK cells wounded with glass beads and stained with ruthenium red . Small arrows: clusters of caveolae-like vesicles . Large arrows: wound sites . Arrowheads: merged caveolae-like vesicles connected to the PM . Bars: 100 nm . ( C ) NRK cells treated with control or Cav1 siRNA , wounded with glass beads ± Ca2+ and stained with PI ( red ) and ( blue ) . Bar: 50 μm . ( D ) Quantification of PI positive nuclei in D . Data represent the mean ± SEM of the %PI positive cells/field . **p=0 . 0012 ( compared to no Ca2+ conditions ) , unpaired Student’s t test . The results are representative of two independent experiments . ( E ) FACS quantification of PI staining in NRK cells treated with control or Cav1 siRNA , mechanically wounded by scraping from the dish ± Ca2+ . The results are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 01510 . 7554/eLife . 00926 . 016Figure 8—figure supplement 1 . Clusters of caveolae are observed next to sites of mechanical wounding . TEM of NRK cells injured with glass beads for 30 s ( A ) or 60 s ( B and C ) and stained with ruthenium red after fixation to label sites of PM injury . Bars: 500 nm . Large arrows: injury sites; small arrows: caveolae-like vesicular profiles . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 016 Mechanical wounding with glass beads cannot be performed simultaneously with live imaging , so instead of monitoring FM1–43 influx we performed end-point PM repair assays based on exclusion of the membrane impermeable dye PI . As expected , in the absence of Ca2+ wounds were not repaired , allowing injured cells to be identified by nuclear PI staining . In the presence of Ca2+ , a reduction in PI positive nuclei consistent with PM repair was observed in cells treated with control siRNA . In contrast , cells treated with Cav1 siRNA and injured in the presence of Ca2+ showed significantly higher PI staining when compared to controls ( Figure 8C , D ) . Similar results were obtained after cells were wounded by scraping from the dish , followed by PI staining and FACS quantification of the whole cell population ( Figure 8E ) . Thus , repair of mechanical wounds on the PM is also inhibited after depletion of the caveolar protein Cav1 . Mutations in the muscle-specific caveolin isoform Cav3 are associated with muscular dystrophy and other serious muscle abnormalities ( Gazzerro et al . , 2010 ) . However , the role of Cav3 in muscle pathology is not fully understood , and was previously attributed to indirect effects not linked to caveolae formation and/or endocytosis ( Hernández-Deviez et al . , 2008; Gazzerro et al . , 2010 ) . We investigated this issue using C2C12 myoblasts , a cell line that faithfully reproduces muscle differentiation upon serum withdrawal , forming multi-nucleated myotubes that express muscle differentiation markers ( Blau et al . , 1983; Kim et al . , 2006; Figure 9A , B ) . Both myoblast and myotube-enriched cultures were susceptible to permeabilization by SLO , and resealed in the presence of Ca2+ ( Figure 9C ) . 10 . 7554/eLife . 00926 . 017Figure 9 . Cav3 expression , Ca2+-dependent sarcolemma repair , lysosomal exocytosis and endocytosis in C2C12 myoblasts/myotubes . ( A ) Western blot with anti-Cav3 or anti-tubulin ( loading control ) antibodies showing that differentiation of C2C12 myoblasts induced by serum starvation leads to a gradual enrichment of the cultures in myotubes expressing the caveolin isoform Cav3 . ( B ) Immunofluorescence of C2C12 cultures at day 5 after serum starvation with anti-Cav3 antibodies . Green: Cav3 staining in myotubes . Blue: DAPI-stained nuclei . Arrows point to DAPI-stained nuclei of Cav3-negative myoblasts . Bar: 100 nm . ( C ) C2C12 cells at day 0 ( myoblasts ) and day 5 ( enriched in myotubes ) after serum starvation , permeabilized with 400 ng/ml SLO and stained with PI ( membrane impermeable ) and DAPI ( membrane permeable ) after 4 min with or without Ca2+ . Bar: 10 µm . The results in A–C are representative of three independent experiments . ( D ) Immunofluorescence with anti-Lamp1 in live C2C12 myotubes 15 min after permeabilization with SLO ± Ca2+ . Green: Lamp1 luminal epitope . Blue: DAPI-stained nuclei . Bar: 10 μm . The results are representative of two independent experiments . ( E ) ASM activity secreted by undifferentiated myoblasts ( day 0 ) or myotube-enriched C2C12 cultures ( day 4 ) after SLO±Ca2+ for 240 s . Data represent mean ± SEM of triplicates . **p=0 . 002 , ***p<0 . 001 , unpaired Student’s t test . ( F ) Myotubes permeabilized or not with SLO±Ca2+ for 240 s in the presence of Texas Red-dextran . Red: dextran . Blue: DAPI-stained nuclei . Bars: 10 μm . The results are representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 017 Myotubes responded to SLO permeabilization and Ca2+ influx with exocytosis of lysosomes , as previously shown in other cell types ( Rodríguez et al . , 1997 ) . A Lamp1 luminal epitope was detected on the sarcolemma surface ( Figure 9D ) , and the lysosomal enzymes ß-hexosaminidase ( data not shown ) and ASM were released by myoblasts ( undifferentiated culture , day 0 ) and myotubes ( differentiated culture , day 4 ) ( Figure 9E ) . Injury with SLO also enhanced endocytosis in muscle fibers , detected by using fluorescent dextran as a fluid phase tracer . Untreated myotubes displayed very few dextran-positive endocytic vesicles after 4 min , reflecting low levels of endocytosis during this time period . After exposure to SLO in the absence of Ca2+ dextran filled the myotubes cytoplasm , reflecting permeabilization of the sarcolemma . In the presence of Ca2+ , a condition that allows PM repair , dextran was excluded from the myotube cytosol but was detected in a punctate intracellular pattern suggestive of endocytic vesicles ( Figure 9F ) . When untreated myotubes were examined by TEM , caveolae-like <80 nm invaginations were observed associated with the sarcolemma ( Figure 10A , control ) . Similar to what we observed in other cell types , a 30 s treatment with SLO or SM markedly increased the number of <80 nm caveolae-like profiles observed in myotube sections ( Figure 10A , SLO and SM ) . At later time points these profiles appeared larger and more complex , suggesting an ongoing process of vesicle homotypic fusion ( Figure 10A , 180 s ) . Suggesting a certain degree of de novo caveolae assembly , immunofluorescence of myotubes exposed for 30 s to SM revealed an increase in the fraction of endogenous cavin1 colocalizing with Cav3 ( Figure 10B , C ) . 10 . 7554/eLife . 00926 . 018Figure 10 . Caveolae-like vesicles accumulate in C2C12 myotubes exposed to sphingomyelinase and SLO , and resealing after injury depends on Cav3 . ( A ) TEM of myotubes untreated ( control ) or exposed to SLO or SM . Arrows: caveolae-like vesicles . Arrowheads: merged caveolae-like vesicles . Wide arrow: clathrin-coated pit . Inset: higher magnification showing luminal BSA-gold ( thin arrow ) . Bars: 100 nm . The results are representative of four independent experiments . ( B ) Confocal images of immunofluorescence with anti-cavin1 ( red ) and anti-Cav3 ( green ) in C2C12 myotubes treated or not with SM for 30 s . Arrows show colocalization of cavin1 and Cav3 staining . Bar: 10 µm . The results are representative of two independent experiments . ( C ) Quantification of the fraction of all cavin1 colocalizing with Cav3 in E . Data represent mean ± SEM of the colocalization coefficient/myotube . ***p<0 . 001 , unpaired Student’s t test . ( D ) Images of myotube-enriched cultures treated with control or Cav3 siRNA , exposed to SLO±Ca2+ for 240 s and stained with PI and DAPI . PI ( red ) and DAPI ( blue ) . Bar: 50 μm . The results are representative of four independent experiments . ( E ) Western blot of Cav3 or tubulin ( loading control ) , and quantification of PI positive nuclei in an assay performed as in ( D ) . Data represent the mean of three independent experiments ± SEM . ***p<001 , unpaired Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 018 The PM repair capacity of myotubes transcriptionally silenced for Cav3 expression was also reduced . Ca2+-free exposure to SLO resulted in PI staining of most myotube nuclei , reflecting the expected high levels of permeabilization and poor resealing . In the presence of Ca2+ very few PI positive nuclei were observed , demonstrating that myotubes effectively remove SLO pores from their sarcolemma in a Ca2+-dependent fashion . In contrast , Cav3-deficient myotubes exposed to SLO+Ca2+ showed high levels of nuclear PI staining , indicating a sarcolemma repair defect ( Figure 10D , E ) . No nuclear staining with PI was detected in C2C12 myotube cultures not exposed to SLO , with or without Ca2+ ( results not shown ) . These results actually underestimate the repair defect of Cav3-deficient myotubes , since C2C12 myotube cultures also contain undifferentiated myoblasts not expressing Cav3 , which reseal their PM after SLO+Ca2+ permeabilization ( Figure 9B , C ) . Thus , myotubes respond to wounding and Ca2+ influx with lysosomal exocytosis , secretion of ASM and intracellular accumulation of caveolar vesicles , a process that seems to be required for sarcolemma repair . Primary mouse flexor digitorum brevis muscle fibers were also examined after exposure to SLO or SM after 30 s . Without Ca2+ , SLO triggered PI influx and staining of the fiber nuclei . With Ca2+ , PI influx was blocked in most fibers , reflecting robust resealing . No nuclear PI staining was detected in fibers treated with SM , demonstrating that exposure to this enzyme does not impair sarcolemma integrity ( Figure 11 ) . As seen in C2C12 myotubes , treatment with SLO or SM resulted in increased numbers of intracellular caveolae-like vesicles along the sarcolemma , when compared to untreated controls ( Figure 12A–C ) . Dissection resulted occasionally in localized fiber wounding ( Figure 11 small arrows , Figure 12D , Figure 12—figure supplement 1 ) , providing an opportunity to examine the effect of mechanical wounding in primary muscle fibers . Intact regions along the fiber perimeter contained mostly a single layer of caveolae-like profiles close to the sarcolemma ( Figure 12D panels 3 and 4—see whole data set in Figure 12—figure supplement 1 ) . In contrast , an increased density of membrane profiles strongly resembling single and merged caveolae was evident in the proximity of wounds ( Figure 12D panels 1 and 2 , Figure 12—figure supplement 1 ) . These observations are consistent with the view that primary muscle fibers , similarly to other cell types analyzed in this study , respond to injury with a rapid Ca2+-dependent resealing process that involves intracellular accumulation of caveolae-like vesicles . 10 . 7554/eLife . 00926 . 019Figure 11 . Primary muscle fibers are sensitive to SLO permeabilization and reseal in the presence of Ca2+ . Flexor digitorum brevis mouse muscle fibers treated or not with 400 ng/ml SLO or 50 mU/ml SM in the presence or absence of Ca2+ and stained with PI ( red ) after 30 s . Small arrows point to PI positive fibers that were injured during dissection and failed to reseal . Arrowheads point to PI-negative fibers that resealed after SLO+Ca2+ , or were not injured by the SM treatment . Large arrows point to a PI-positive fiber that was injured by SLO and failed to repair without Ca2+ . Bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 01910 . 7554/eLife . 00926 . 020Figure 12 . Caveolae accumulate in primary mouse muscle fibers after exposure to sphingomyelinase or sarcolemma injury . ( A–C ) TEM of flexor digitorum brevis fibers untreated ( A ) or exposed for 300 s to SLO ( B ) or SM ( C ) . Three examples are shown for each . Arrows: single or merged caveolae-like vesicles . Bars: 100 nm . The results are representative of four independent experiments . ( D ) TEM of fiber fixed immediately after dissection , showing mechanical damage ( site #1 ) . Bar: 5 μm . Panels 1–4 show enlarged images of the regions indicated in the whole fiber image . Arrows: single or merged caveolae . Bars: 100 nm . The results are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 02010 . 7554/eLife . 00926 . 021Figure 12—figure supplement 1 . Accumulation of caveolae-like vesicles in mechanically injured flexor digitorum brevis muscle fibers . ( A ) Sequential TEM images along the whole periphery of a fiber fixed shortly after dissection . Caveolae-like vesicles are more abundant in the vicinity ( #21–24 ) of the wound site ( #23 ) . Bars: 100 nm . ( B ) TEM image of the whole fiber indicating the location of each image . Bar: 5 µm . The results are representative of several injured fibers observed in the two independent experiments described in Figure 12 . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 021
The goal of this study was to investigate the abundant , cholesterol-dependent endocytic compartments that accumulate in several cell types after PM wounding ( Idone et al . , 2008; Thiery et al . , 2011 ) . Strikingly , our results revealed that a large fraction of the endocytic vesicles formed a few seconds after injury resemble caveolae , the ubiquitous flask-shaped PM invaginations that have been implicated in transcytosis , mechanosensing and signaling responses ( Parton and Simons , 2007; Lajoie and Nabi , 2010 ) . Wounded cells show increased numbers of small , homogeneously shaped vesicles with the typical morphology and markers of caveolae . Several independent lines of evidence demonstrate that internalization of these vesicles occurs rapidly , with a time-course that matches the kinetics of PM repair . Cryo-immuno EM , endocytosis and live imaging assays showed that the pore-forming toxin SLO is removed from the PM and traffics into cells in vesicular carriers containing the caveolar marker Cav1 . Furthermore , transcriptional silencing of Cav1 in non-muscle cells and Cav3 in myotubes inhibits caveolae formation and PM repair . Collectively , our study identifies caveolar vesicles as dynamic endocytic structures that play a key role in the restoration of PM integrity . Our findings have important implications for the understanding of muscular dystrophy . Mutations in the muscle-specific caveolin Cav3 cause at least five forms of muscle pathology , including limb girdle muscular dystrophy 1C , rippling muscle disease , distal myopathy , hyperCKemia , and hypertrophic cardiomyopathy ( Gazzerro et al . , 2010 ) . Mutations in PTRF/cavin1 also lead to muscular dystrophy and cardiac dysfunction ( Rajab et al . , 2010 ) , and overexpression of this caveolae-associated protein rescues membrane repair defects in dystrophic muscle ( Zhu et al . , 2011 ) . In earlier studies these effects were not attributed to caveolae assembly and internalization , but rather to an independent role of caveolin and cavin molecules in regulating the traffic of unrelated proteins involved in muscle fiber repair , such as dysferlin and MG53 ( Hernández-Deviez et al . , 2008; Hayashi et al . , 2009; Cai et al . , 2009b; Zhu et al . , 2011 ) . Our study sheds important new light on this issue , by showing that caveolar vesicles are directly involved in the endocytic mechanism by which cells remove wounds from their PM . Duchenne muscular dystrophy fibers were shown in numerous studies to contain elevated numbers of caveolae , but this important finding was also attributed to secondary effects of alterations in Cav3 expression levels ( Bonilla et al . , 1981; Repetto et al . , 1999 ) . However , it is important to note that Duchenne muscular dystrophy is caused by mutations in dystrophin , a main component of the dystroglycan complex that confers stability to the sarcolemma ( Cohn and Campbell , 2000 ) . Duchenne dystrophic fibers are very susceptible to contraction-induced wounding , and are likely to undergo repeated cycles of injury and repair—a process that we now show to involve intracellular accumulation of caveolar vesicles , in several cell types including C2C12 myotubes and primary muscle fibers . Thus , our findings provide a novel explanation for the accumulation of caveolae within fragile muscle fibers that is fully consistent with a direct role of caveolar endocytosis in sarcolemma resealing . Our results are also in agreement with the elevated number of caveolar profiles observed in cell types subject to chronic membrane stress , such as endothelial cells and adipocytes ( Parton and Simons , 2007 ) . Caveolae are thought to be dynamic structures , but the signals leading to their assembly and internalization remain a matter of debate ( Lajoie and Nabi , 2010; Sandvig et al . , 2011 ) . Caveolae are enriched in cholesterol and glycosphingolipids , and contain the abundant surface lipid sphingomyelin that generates ceramide after cleavage of its phosphorylcholine head group by SM ( Liu and Anderson , 1995; Parton and Simons , 2007 ) . Exogenously added glycosphingolipids selectively stimulate caveolar endocytosis ( Sharma et al . , 2004 ) , and the ceramide core of glycosphingolipids was identified as an important determinant of caveolae internalization ( Singh et al . , 2003 ) . However , the mechanism by which glycosphingolipids modulate caveolar endocytosis remained unclear , because a non-hydrolyzable synthetic glycosphingolipid analog was reported to have the same effect ( Sharma et al . , 2004 ) . Our study clarifies this issue , by directly demonstrating that treatment of the outer leaflet of the PM with the ceramide-generating enzymes ASM or SM is sufficient to induce internalization of vesicles with properties of caveolae . In agreement with our findings , ASM ( Opreanu et al . , 2011 ) and ceramide ( Liu and Anderson , 1995; Bilderback et al . , 1997; Czarny et al . , 2003 ) were detected in caveolae-enriched membrane fractions . Collectively , our results support a novel model proposing that injury to the PM triggers Ca2+ influx , exocytosis of lysosomes , ASM release and generation of surface-associated ceramide—an event that facilitates caveolae internalization and lesion removal/repair ( Figure 13A , B ) . We consistently observed a strong inhibition of PM repair in cells acutely depleted of Cav1 by RNAi ( this study ) or depleted in cholesterol ( Idone et al . , 2008 ) . These findings are in full agreement with our additional lines of evidence implicating caveolar vesicles as endocytic carriers responsible for lesion removal from the PM . However , it is also important to note that a complex cross-talk exists between caveolar and other cholesterol-dependent , clathrin-independent endocytic pathways . Thus , non-caveolar cholesterol/sphingolipid raft-dependent endocytic carriers may also play a role in removing lesions from the PM under some conditions , such as in Cav1-deficient cells , which were shown to upregulate non-caveolar endocytic pathways ( Le et al . , 2002; Nichols , 2003; Singh et al . , 2003 ) . 10 . 7554/eLife . 00926 . 022Figure 13 . Model for PM repair mediated by caveolar endocytosis . Permeabilization with transmembrane toxin pores ( A ) or mechanical wounding ( B ) triggers Ca2+ influx , exocytosis of lysosomes , release of ASM , and generation of ceramide at the PM outer leaflet , a process that promotes caveolae internalization and fusion . Toxin pores would be removed from the PM by caveolar endocytosis ( A ) , while larger breaches on the lipid bilayer would be gradually constricted and resealed as a results of forces generated on the PM by the intracellular clustering , fusion and internalization of merged/branched caveolar structures ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00926 . 022 The GTPase dynamin ( Guha et al . , 2003; Hansen and Nichols , 2009 ) was detected on caveolae necks ( Henley et al . , 1998; Oh et al . , 1998 ) , and proposed to be required for some forms of caveolar endocytosis ( Henley et al . , 1998; Oh et al . , 1998; Yao et al . , 2005 ) . However , more recent studies using dominant-negative forms of dynamin-2 ( van Deurs et al . , 2006; Liu et al . , 2008 ) or cells deficient in both dynamin-1 and dynamin-2 ( Ferguson et al . , 2009 ) failed to observe inhibition of caveolar endocytosis and the accumulation of PM-connected caveolar intermediates expected from a fission block . These findings led to suggestions that indirect effects on the actin cytoskeleton or sequestration of essential factors distinct from dynamin might explain the inhibition in caveolar endocytosis by dominant-negative dynamin ( Liu et al . , 2008 ) . After extensive analysis of cells depleted in dynamin-1 and dynamin-2 we did not observe any defects in caveolar endocytosis in response to PM wounding , suggesting that this process may be dynamin-independent . Not all forms of endocytosis require the GTPase dynamin ( Guha et al . , 2003; Hansen and Nichols , 2009 ) , and the existence of different forms of caveolar endocytosis has been proposed ( Le and Nabi , 2003; Singh et al . , 2003 ) . Given that ceramide accumulation is sufficient to induce membrane invagination and inward budding of membranes ( Holopainen et al . , 2000; Trajkovic et al . , 2008 ) , it is an intriguing possibility that sphingomyelin hydrolysis and ceramide generation may bypass the requirement for dynamin in the fission of caveolar vesicles associated with PM repair . When imaged several minutes after PM injury , endocytic vesicles triggered by Ca2+ influx appear as large , uncoated vesicles containing the early endosome marker EEA1 ( Keefe et al . , 2005; Idone et al . , 2008 ) . A recent study showed that internalized SLO moves to the perinuclear area in endocytic vesicles that gradually increase in size and merge with late endosomes/lysosomes , where ubiquitinated toxin is degraded in an ESCRT-dependent manner ( Corrotte et al . , 2012 ) . Strikingly , in this study we found that cell injury or exposure to purified SM initially generates homogeneous <80 nm vesicles with a morphology typical of caveolae , which then rapidly merge originating larger , irregular intracellular compartments . Interestingly , wounded cells accumulate larger and more abundant endocytic vesicles after the cortical cytoskeleton is disrupted with cytochalasin D ( Tam et al . , 2010 ) , a condition that enhances caveolae motility ( Thomsen et al . , 2002 ) . Thus , removal of obstacles presented by the cortical cytoskeleton may facilitate internalization and intracellular merging of caveolae . Supporting this view , we observed reduced levels of EHD2 , the ATPase proposed to link caveolae to actin filaments ( Moren et al . , 2012; Stoeber et al . , 2012 ) , on caveolar vesicles that move deeper into cells after injury or exposure to SM . In several instances caveolae appeared to merge intracellularly while still connected to the PM , particularly when cells were mechanically injured . We envision that the rapid intracellular merging of surface-connected caveolae , possibly enhanced by the higher levels of Ca2+ flowing through large lesions , might generate forces on the PM that constrict and ultimately reseal wounds ( Figure 13B ) . Such ‘bunch of grapes’ pattern of caveolar vesicles is also observed in cells not subjected to a specific wounding procedure ( Fra et al . , 1995 ) . However , it is important to note that lifting cells from dishes and other routine cell handling steps often results in PM wounding and Ca2+ influx . Detaching cells from their substratum was actually reported to trigger rapid and robust caveolae internalization ( Muriel et al . , 2011 ) . Our observations suggest a model where caveolae-derived branched endocytic structures still tethered to the PM would be generated as a consequence of the vigorous Ca2+ influx and localized release of lysosomal ASM triggered by large wounds . These branched caveolae-derived structures appear to accumulate around the periphery of the PM wounds , and as their deeper portions merge intracellularly , constriction forces may be generated on the PM facilitating bilayer resealing . This model predicts that when the complex caveolar structures pinch off from the PM , large intracellular vesicles would be generated next to injury sites ( Figure 13B ) . This scenario is fully consistent with previous EM observations of large vesicles close to wound sites ( McNeil and Steinhardt , 2003 ) . These large intracellular vesicles were at the time interpreted as exocytic ‘patches’ , responsible for resealing the wound . Here we propose a very different scenario for the repair of mechanical injury to the PM: After Ca2+-triggered lysosomal exocytosis and ASM release , caveolae-derived vesicles would move into cells while undergoing a homotypic fusion process , which would ultimately result in large compartments of endocytic origin ( as opposed to a large exocytic patch , as proposed in earlier models—[McNeil and Steinhardt , 2003] ) . A vesicle internalization/constriction mechanism for PM resealing predicts that significant local tension would be generated on PM sites close to the wound . This tension may be dissipated by the localized exocytosis of lysosomes , and/or by the recently demonstrated process of flattening of individual caveolae in response to membrane stress ( Sinha et al . , 2011 ) . An intriguing possibility is that mechanical wounds may actually occur preferentially in areas where caveolae are present , because membranes may be rendered more fragile by the flattening of caveolae that follows a mechanical stretch ( Sinha et al . , 2011 ) . Caveolae are extremely abundant in cells that are under mechanical stress in vivo , such as smooth muscle fibers and endothelial cells ( Parton and Simons , 2007 ) . Our findings now suggest that frequent PM injury and repair may represent the mechanism responsible for upregulation of the caveolar population in these cells . In summary , our study provides a novel framework for understanding how endocytosis functions in the maintenance and restoration of PM integrity , particularly in tissues under high mechanical stress such as skeletal muscle . Importantly , this work also identifies caveolar endocytosis as a pathway that may be directly affected in several types of muscular dystrophy . Resealing of injured muscle fibers requires muscle-specific proteins such as dysferlin and MG53 , which accumulate in vesicles close to sites of injury and have been assumed to facilitate formation of an exocytic membrane patch ( Cai et al . , 2009a ) . In light of our findings and of the newly uncovered role of endocytosis in PM repair , it will be of great interest to determine whether these muscle-specific proteins participate in wound resealing by facilitating caveolar assembly/internalization , and lesion removal by endocytosis .
NRK and HeLa cells were cultured at 37°C in 5% CO2 in high glucose DME containing 10% heat-inactivated FBS and penicillin/streptomycin ( Invitrogen , Grand Island , NY ) . The C2C12 mouse myogenic cell line ( CRL-1772 ) , a subclone of mouse skeletal muscle C2 cells ( Blau et al . , 1983 ) , was obtained from the American Type Culture Collection ( Rockville , MD ) . C2C12 were grown to confluency at 37°C and 5% CO2 in DME supplemented with 10% FBS , 100 units/ml penicillin/streptomycin . The medium was then changed to DME containing 2% horse serum and 100 U/ml penicillin/streptomycin to trigger myogenic differentiation . The cells were maintained in this medium for 4–7 days , with fresh medium added every second day . The tamoxifen-inducible DKO fibroblast line generated from dynamin-1 and 2 double conditional knockout mice ( Ferguson et al . , 2009 ) was provided by De Camilli P , Yale University and cultured at 37°C in 5% CO2 in high glucose DME containing 10% heat-inactivated FBS and penicillin/streptomycin ( Invitrogen ) . To induce dynamin-1 and 2 depletion , cells were incubated for 2 days in medium containing 3 µM tamoxifen ( Sigma , St . Louis , MO ) followed by 3–4 days of 300 nM tamoxifen before seeding for experiments . Flexor digitorum brevis muscle fibers ( Cai et al . , 2009a ) were surgically isolated from euthanized male C57Bl/6 mice in a Tyrode solution containing 140 mM NaCl , 5 mM KCl , 2 . 5 mM CaCl2 , 2 mM MgCl2 and 10 mM HEPES ( pH 7 . 2 ) , and incubated in the same solution containing 2 mg/ml type I collagenase ( Sigma ) in an orbital shaker at 100 rpm for 60 min , followed by gravity sedimentation for 5 min at 37°C . After discarding the supernatant , Tyrode solution was added and the pellet gently resuspended , followed by a second round of gravity sedimentation for 1 min to remove large tissue aggregates . The supernatant containing isolated fibers was transferred to another tube , allowed to sediment for 5 min , resuspended in DME 10% FBS and subjected to the various treatments before PI influx assays and/or fixation for TEM . Immunoblot , immunofluorescence and immuno-EM assays were performed using rabbit anti-GFP to detect GFP-SLO ( Invitrogen ) , rabbit anti-Cavin1/PTRF ( Abcam , Cambridge , MA ) , mouse anti-Cav-1 and -3 ( BD transduction Laboratories , San Jose , CA ) , rabbit anti-Cav1 ( Santa Cruz , Dallas , TX ) , rat anti-mouse Lamp1 ( 1D4B mAb , Developmental Studies Hybridoma Bank , Iowa City , IA ) , mouse anti-ceramide ( mAb 15B4; Sigma ) , mouse anti-tubulin ( Sigma ) , mouse anti-actin ( Sigma ) , rabbit anti-dynamin-2 ( Abcam ) , rabbit anti-dynamin-1 ( Epitomics , Burlingame , CA ) , and goat anti-EHD2 ( Abcam ) . NRK or HeLa cells ( 50% confluency ) and C2C12 myoblasts ( 70% confluency , 48–72 hr after FBS removal ) in reduced serum DME without penicillin/streptomycin were transfected with Lipofectamine RNAiMax ( Invitrogen ) and 960 pmol of medium-content control , Cav1 , Cav3 , ASM or dynamin-2 ( Dyn2 ) stealth siRNA duplexes , according to the manufacturer’s instructions ( Invitrogen ) . After 48 hr–72 hr cells were treated with SLO , SM or glass beads and processed for various assays . HeLa cells ( 50% confluency ) in MatTek glass-bottom dishes containing reduced serum DME without penicillin/streptomycin were transfected with Lipofectamine 2000 ( Invitrogen ) and 1 µg of mRFP-Cav1 plasmid per dish , according to manufacturer’s instructions ( Invitrogen ) . After 24 hr cells were processed for live imaging . The correct targeting of mRFP-Cav1 to caveolae was confirmed by acquiring Z stack images ( 0 . 13 µm Z steps ) in a confocal Leica SPX5 microscope with a 63 × 1 . 4 N . A . oil objective of transfected HeLa cells fixed , permeabilized and stained with anti-Cav1 antibodies . Cells were pre-treated with 200–400 ng/ml SLO ( NRK and HeLa ) or 800 ng/ml ( C2C12 myotubes and flexor digitorum brevis mouse muscle fibers ) for 5 min at 4°C and further incubated for various time points at 37°C in DME containing BSA-gold ( OD 520 nm = 200; [Ferguson et al . , 2009] ) , or incubated in DME containing 10 µg/ml human recombinant acid sphingomyelinase ( ASM ) ( He et al . , 1999 ) or 50 mU/ml of B . cereus sphingomyelinase ( SM ) ( Invitrogen ) , before being processed for TEM as previously described ( Rodríguez et al . , 1997 ) . EM images were acquired randomly or along the PM . Quantifications were performed by counting all vesicles containing BSA-gold in several cell sections/sample , or by counting all caveolae-like vesicles with a diameter of less than 80 nm ( measured with the line tool of ImageJ , NIH ) in 20–30 cell sections/sample or in 40 images/sample . The relative amount of < and >80 nm vesicles was determined in cells treated with 200 ng/ml SLO for 30 , 60 and 180 s in the presence of BSA-gold . Vesicle area was measured using the outline function of ImageJ and the number of gold particles/vesicle was determined in 14–47 cell sections . To visualize mechanical wounding with glass beads in NRK cells or to assess complete separation of vesicles from the PM in control or Dyn2 siRNA-treated NRK cells treated with SLO or SM , cells were fixed in 2% glutaraldehyde in 0 . 1M cacodylate and 0 . 05% ruthenium red for 1 hr at room temperature before washing and processing for TEM as described in Parton et al . ( 2002 ) . EM images were blinded before quantification and scored independently by two investigators . For cryo-immuno-EM , cells were treated as described above for TEM and fixed in 4% PFA , 0 . 25 M HEPES and 0 . 1% glutaraldehyde for 1 hr at room temperature and processed for immuno-gold labeling of Cav1 , GFP-SLO or ceramide as described in Czibener et al . ( 2006 ) . Ceramide staining was quantified by drawing a line along the PM using the ImageJ brush tool set to 200 nm diameter , and counting all gold particles inside the brush tool area in all membrane sections ( cell section areas ranged from 6 to 16 µm2 and the data were normalized to particles/µm2 ) . To assess localization of Cav1 and GFP-SLO over time during PM repair , Cav1 and GFP-SLO positive structures were quantified by counting all <80 nm vesicles or >80 nm vesicles positive for anti-Cav1 alone , anti-GFP alone or both , as well as flat PM areas positive for anti-GFP . All antibody-stained structures were quantified in 80 random microscopic fields for each sample . The relative Cav1 localization between flat PM stretches and vesicular profiles was quantified by counting 25–72 flat PM segments and 256–346 vesicular profiles positive for anti-Cav1 in 80 images/sample . For all antibodies , titrations were performed and specificity was assessed using isotype control antibodies before imaging and quantifications , according to standard procedures from the Yale University Center for Cell and Molecular Imaging . To analyse SLO endocytosis during PM repair by flow cytometry , a single cysteine mutant of SLO ( SLOG66C , generated by replacing Gly66 by a cysteine in a cysteine-less derivative of SLO , kindly provided by Dr . R Tweten , U Oklahoma ) was labeled at the N-terminus with a thiol-reactive AlexaFluor-488 C5 maleimide ( Life Technologies , Frederick , MD ) according to the manufacturer’s instructions . Briefly , 500 µl of a 50 µM SLO solution ( in PBS no Ca2+ ) were incubated with 10 mM of reactive dye for 2 hr at room temperature . After reaction , the labeled SLO was separated from unbound dye by gel filtration . Subconfluent NRK cells were treated or not with control or Cav1 siRNA for 48 hr , trypsinized , counted and diluted to 1 , 5 × 105 cells/250 µl for flow cytometry . Cells were incubated at 4°C for 5 min with increasing concentrations ( 0 . 9–2 . 1 µg/ml ) of Alexa 488-SLO in PBS supplemented with 5 . 5 mM D-glucose with or without Ca2+ for PM binding , and then transferred or not to 37°C for 5 min to induce PM repair , followed by transfer to 4°C to stop the process . Cells were analyzed by flow cytometry ( FACSCanto , Becton Dickinson , Sparks Glencoe , MD ) and A488 fluorescence was assessed before and after adding 10 µg/ml of rabbit anti-Alexa Fluor 488 quenching antibody ( Life Technologies ) for 2 min . PI ( 50 µg/ml ) was added to all samples at the end of the assay to assess levels of PM repair . Data were analyzed using Flo-Jo software ( Three Star , Inc , Ashland , VA ) . Subconfluent NRK cells treated with control , Cav1 or Dyn2 siRNA , or DKO MEFs induced or not with tamoxifen were plated on glass-bottom dishes ( MatTek , Ashland , MA ) , pre-incubated with 200 ng/ml of SLO for 5 min at 4°C , transferred to a LiveCell System chamber ( Pathology Devices , Westminster , MD ) at 37°C with 5% CO2 , and exposed to pre-warmed DME containing or not Ca2+ and 4 μM ( NRK cells ) or 8 µM ( dynamin-1-2 DKO cell line ) FM1–43 ( Invitrogen ) and SLO , as previously described ( Idone et al . , 2008 ) . Spinning disk confocal images were acquired for 4 min at 1 frame/3 s using an UltraVIEW VoX system ( PerkinElmer , Waltham , MA ) attached to an inverted microscope ( Eclipse Ti; Nikon Instruments , Melville , NY ) with a 40 × NA 1 . 3 objective ( Nikon ) and a CCD camera ( C9100–50; Hamamatsu Photonics , Bridgewater , NJ ) . Quantitative analysis of fluorescence in a defined intracellular area was performed using Volocity Suite ( PerkinElmer ) . For live imaging of GFP-SLO internalization , HeLa cells were cultured on glass-bottom dishes ( MatTek ) and transfected with mRFP tagged Cav1 ( mRFP-Cav1 ) for 24 hr using lipofectamine 2000 ( Invitrogen ) . Cells expressing mRFP-Cav1 were pre-incubated for 5 min on ice with 800 ng/ml of GFP-SLO ( generated as described in Idone et al . [2008] ) , and 4°C DME with Ca2+ was added to induce PM repair after warming to 37°C on a heated stage , followed by imaging for 5 min at 1 frame/2 s as previously described for FM1–43 imaging . Since the amount of SLO fluorescence that becomes associated with cells under conditions that allow PM repair is limited , to bring the signal out from the noise floor inherent to the imaging system all datasets were processed using NIS-Elements software ( Nikon Instruments ) . First , drift correction was performed on datasets to account for drift with temperature changes at high magnification . In order to improve the signal to noise ratio and minimize the effects of the spurious noise , a 3-frame rolling average was employed followed by a regional maximum detection , which compares a pixel ( or group of pixels ) to its neighboring region and determines where there is a significant difference . Routine scaling of the image resulted in the contrast viewed in the included datasets ( Videos 1 , 2 and 3 ) . Volocity Suite ( Perkin Elmer ) was used to draw lines beneath the PM to record fluorescence intensity levels of GFP-SLO and mRFP-Cav1 and create kymographs displaying fluorescence levels over time ( Y axis ) for each pixel along the line ( X axis ) . Colocalization of GFP-SLO and mRFP-Cav1 was analyzed and displayed as positive PDM ( product of the differences from the mean ) . HeLa cells were treated or not with 450 ng/ml SLO or 50 mU/ml SM for 30 or 60 s , fixed , permeabilized with 0 . 05% saponin in PBS containing 1% BSA and immunostained overnight for Cav1 ( Santa Cruz ) and EHD2 ( Abcam ) diluted 1:500 in permeabilization buffer , followed by 1 hr incubation with secondary antibodies conjugated with Alexa fluor 488 ( anti-rabbit ) or Alexa fluor 546 ( anti-goat ) . Images were acquired using a Nikon laser TIRFm system on an inverted microscope ( Nikon TE2000-PFS ) equipped with a 63 × NA 1 . 49 Apochromat TIRF objective ( Nikon Instruments ) , a Coolsnap HQ2 charge-coupled device camera ( Roper Scientific , Sarasota , FL ) , and two solid-state lasers of wavelengths 491 and 561 nm . AF488 and AF546 images were acquired sequentially and analysis of fluorescence intensity for Cav1 staining was performed on 53–79 cells per sample using Andor iQ software ( Andor Technology , Belfast , UK ) . Colocalization of Cav1 ( AF488 ) with EHD2 ( AF546 ) was quantified on 65–82 cells/sample using Volocity Suite ( PerkinElmer ) after applying thresholding to each channel . Following extraction proteins were separated on 8 , 10 or 12% SDS-PAGE gels and blotted on nitrocellulose membranes using the Trans-Blot Transfer system ( Bio-Rad Laboratories , Hercules , CA ) overnight at 30 V , or for 2 hr at 95 V . After incubation with the primary antibodies and peroxidase conjugated secondary antibodies , detection was performed using Supersignal West Pico Chemiluminescent Substrate ( Thermo Scientific , Waltham , MA ) and a Fuji LAS-3000 Imaging System with Image Reader LAS-3000 software ( Fuji , Edison , NJ ) . Dynamin-1-2 DKO MEFs were incubated or not with 400 ng/ml of SLO for 5 min at 4°C followed by incubation for 3 to 10 min at 37°C in DME with Ca2+ and 5 µg/ml of Alexa 488 CTxB ( Invitrogen ) in the presence or not of 50 mU/ml SM . Cells were then washed twice in cold DME + 20% FBS , twice in cold acid buffer ( 0 . 2 M acetic acid , 0 . 5 M NaCl , pH 2 . 8 ) to remove extracellularly bound CTxB , followed by two additional washes in cold PBS . Cells were then fixed with 4% PFA , DAPI stained and mounted on slides before imaging with a Leica SPX5 confocal system with a 63 × N . A . 1 . 4 oil objective . Z stacks ( 0 . 13 μm Z step between optical sections ) were acquired on a minimum of 5–10 random fields containing 32–66 . Stacks of individual channels were then imported to Volocity Suite ( PerkinElmer ) , the total fluorescence intensity of the channel per microscopic field was determined ( Intensity × Voxel count ) , and the values were normalized by the number of cells in each field ( determined by DAPI staining ) . Cells treated with control or Dyn2 siRNA or induced or not with tamoxifen were incubated for 10 min with 4 µg/ml of Texas Red transferrin in FBS-free DME at 37°C . Cells were then washed twice in cold PBS and twice in cold acid buffer containing 0 . 2 M acetic acid + 0 . 5 M NaCl , pH 2 . 8 , followed by two more washes in cold PBS before fixation in 4% PFA and DAPI staining . Coverslips were imaged and the fluorescence intensity quantified as described for cholera toxin B-A488 uptake assays . NRK cells treated with control or Cav1 siRNA ( Invitrogen ) were cultured to 70% confluence on 3 cm dishes with coverslips , in MatTek glassbottom dishes or 10 cm dishes , and sprinkled with 0 . 05 g ( coverslips and Mattek dishes ) or 0 . 1 g ( 10 cm dishes ) ≤106 µm acid washed glass beads ( Sigma ) in DME ± Ca2+ , followed by gentle rocking as described in Reddy et al . ( 2001 ) and processed for PI staining , cavin pulldown or TEM as described above . NRK cells or myotubes cultured at 70% confluence on six well dishes or flexor digitorum brevis mouse muscle fibers were treated with 250–800 ng/ml SLO for 5 min at 4°C or sprinkled with glass beads and incubated for 4 min in FBS-free DME ± Ca2+ at 37°C , stained for 1 min with 50 µg/ml PI ( Sigma ) , fixed in 4% PFA , DAPI stained and imaged immediately using an Axiovert 200 ( Carl Zeiss , Inc . , Jena , Germany ) equipped with a CoolSNAP HQ camera ( Roper Scientific ) and MetaMorph software ( MDS Analytical Technologies , Sunnyvale , CA ) . Quantifications were done by counting all nuclei stained with DAPI and PI in five random fields in triplicate ( images taken with a 10 × or 32 × objective ) and determining the percentage of PI positive nuclei . To induce large mechanical wounds , NRK cells treated with control or Cav1 siRNA for 48 hr were scraped from the dish in the presence or absence of Ca2+ , stained with PI after 4 min ( for 5 min at 37°C ) and analyzed by flow cytometry ( at least 10 , 000 cells per sample ) as described ( Tam et al . , 2010 ) ( Idone et al . , 2008 ) . The data were analyzed using Flowjo software ( Tree Star , Inc . ) . Live immunofluorescence of surface-exposed Lamp1 was performed as previously described ( Reddy et al . , 2001 ) in differentiated C2C12 myotubes treated with 400 ng/ml of SLO for 5 min at 4°C and exposed to DME ± Ca2+ for 15 min at 37°C before being washed with ice cold PBS and incubated with anti-mouse Lamp1 ( 1D4B ) antibodies for 30 min at 4°C . Cells were then washed and fixed in 4% PFA , DAPI stained , incubated with anti-mouse Alexa Fluor 488 and imaged with an Axiovert 200 microscope as described above . Exocytosis of ASM by C2C12 myoblasts and myotubes was assessed by assaying enzyme activity in the supernatant of cultures treated with 400 ng/ml SLO ± Ca2+ for various time points , as previously described ( Tam et al . , 2010 ) . To visualize endosomes induced by SLO permeabilization in the presence or absence of Ca2+ , C2C12 myotubes were incubated or not with 400 ng/ml SLO for 5 min at 4°C and incubated in DME ± Ca2+ at 37°C , in the presence of 2 . 5 mg/ml lysine-fixable 10 kDa Texas Red dextran ( Invitrogen ) for 4 min before fixation in 4% PFA and DAPI staining . Cells were then imaged in an Axiovert 200 microscope as described above . C2C12 myotubes cultured on coverslips were treated or not with 50 mU/ml SM for 30 s before fixation , quenched with 50 mM ammonium chloride , blocked with 5% FBS and permeabilized with 0 . 2% saponin in PBS . Cells were then incubated with rabbit anti-PTRF/Cavin1 and mouse anti-Cav3 , 1:200 and 1:100 dilution in PBS 0 . 2% saponin respectively for 1 hr , followed by 1 hr incubation with secondary antibody conjugated with Alexa fluor 488 ( anti-mouse ) or Texas Red ( anti-rabbit ) . Nuclei were stained with 10 µM DAPI . C2C12 cells differentiated for 5 days were immunolabeled for Cav3 and DAPI-stained to distinguish myoblasts ( Cav3-negative ) from myotubes ( Cav3-positive ) . Images were acquired using a confocal Leica SPX5 microscope with a 63 × 1 . 4 N . A . oil objective . For each condition Z stacks ( 0 . 13 µm Z steps ) were obtained from upper optical sections of the myotubes , and 12 fields were imaged for each condition . Colocalization coefficients were calculated using Volocity Suite ( PerkinElmer ) . Each myotube ( identified by Cav3 expression ) was outlined by hand and defined as a region of interest . A total of 61 and 70 cells for control and SM , respectively , were analyzed . | Cells must be able to rapidly repair damage to their outer membranes . This is particularly important in the case of muscle cells , which are vulnerable to damage , and the failure of these cells to repair their outer membranes leads to the muscle wastage seen in muscular dystrophy . Researchers do not fully understand how cells repair membrane , but one popular theory is that they use the membranes of specialized vesicles to ‘patch’ areas that have been damaged . A group of proteins called caveolins have also been implicated in membrane repair but , again , the details have not been worked out . These proteins are best known for their role in the formation of caveolae — small pouches formed by invaginated sections of the plasma membrane . Now , Corrotte et al . have obtained evidence that membrane repair relies not on patching , but on endocytosis ( the process by which substances are taken into the cell in small vesicles that ‘pinch’ from the plasma membrane ) of these caveolae pouches . Corrotte et al . treated cells with streptolysin O , a toxin that forms pores in the membrane that cannot be repaired using patches , and found that this led to the formation of small membrane-derived vesicles that looked just like caveolae . Further tests confirmed that these vesicles contained caveolar proteins , and that they removed the toxin from the plasma membrane by endocytosis . Similar effects were seen in response to mechanical damage caused by tiny glass beads . Moreover , blocking the expression of caveolar genes prevented cells from repairing membrane damage . Based on their findings , Corrotte et al . propose an alternative model for the repair process; namely that cellular damage triggers an influx of calcium ions , which causes vesicles called lysosomes to release chemicals that promote the formation of caveolae . These then remove the damaged area through endocytosis , restoring the integrity of the membrane . The results offer new insights into why mutations in caveolar proteins are associated with muscle disorders , including muscular dystrophy and cardiac dysfunction . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2013 | Caveolae internalization repairs wounded cells and muscle fibers |
Despite their importance in sexual differentiation and reproduction , Y chromosome genes are rarely described because they reside in repeat-rich regions that are difficult to study . Here , we show that Guy1 , a unique Y chromosome gene of a major urban malaria mosquito Anopheles stephensi , confers 100% female lethality when placed on the autosomes . We show that the small GUY1 protein ( 56 amino acids in length ) causes female lethality and that males carrying the transgene are reproductively more competitive than their non-transgenic siblings under laboratory conditions . The GUY1 protein is a primary signal from the Y chromosome that affects embryonic development in a sex-specific manner . Our results have demonstrated , for the first time in mosquitoes , the feasibility of stable transgenic manipulation of sex ratios using an endogenous gene from the male-determining chromosome . These results provide insights into the elusive M factor and suggest exciting opportunities to reduce mosquito populations and disease transmission .
Insects employ diverse sex-determination mechanisms at the chromosomal level including XX/XY , ZW/ZZ , XX/XO , and diploid/haploid chromosomal systems ( Bopp et al . , 2014; Bachtrog et al . , 2014; Biedler and Tu , 2016 ) . Similarly , the primary molecular signals that determine sex are highly divergent in insects . In honeybees which use the diploid/haploid system , heterozygosity of the complementary sex determiner ( csd ) gene is the primary signal that initiates female development ( Hasselmann et al . , 2008 ) . In silkworms a piRNA gene on the W chromosome triggers female development ( Kiuchi et al . , 2014 ) . In the fruit fly Drosophila melanogaster , which has XX/XY sex chromosomes , the collective dose of X-linked signal elements ( XSE ) functions as the signal that specifies sex in the early embryo ( Erickson and Quintero , 2007 ) . However , a dominant male-determining factor ( M ) on the Y chromosome initiates male development in many other insects that contain XX/XY sex chromosomes ( Sanchez , 2008; Biedler and Tu , 2016 ) . Instead of a well differentiated and heteromorphic Y chromosome , mosquitoes of the Culicinae subfamily , which includes the genus Aedes , contain a homomorphic sex-determining chromosome that harbors an M factor in the male-determining locus ( M-locus ) . A novel RNA-binding protein named Nix was recently shown to be an M factor in Aedes aegypti ( Hall et al . , 2015 ) . Despite rapid changes in the primary signals or master switches , two key transcription factors at the bottom of the sex-determination pathway , doublesex ( dsx ) and fruitless ( fru ) , are highly conserved in insects . Sex-specific splicing of dsx and fru pre-mRNAs leads to the production of sex-specific DSX and FRU protein isoforms , which program sexual differentiation ( Bopp et al . , 2014 ) . The alternative splicing of dsx and fru pre-mRNAs is often controlled by a protein complex that includes a fast-evolving transformer ( TRA ) and a conserved transformer 2 ( TRA2 ) , where TRA is the sex-specific protein in this TRA/TRA2 complex . Indeed , TRA often functions as an intermediate that transduces the selected sexual fate from the primary signal to the DSX and FRU effector molecules ( Bopp et al . , 2014 ) . For example , in the medfly Ceratitis capitata , which has XX/XY sex chromosomes , a functional TRA is produced in the zygotic XX embryo as a result of splicing by a maternally deposited TRA/TRA2 complex , leading to female-specific dsx and fru splicing and thus the female sex , which is then maintained by the self-sustaining loop of tra splicing and function . In males , a yet-to-be-discovered M factor interrupts this loop of tra splicing , leading to the male sex ( Pane et al . , 2002 ) . Sex-specific splicing of dsx and fru has been described in both Aedes and Anopheles mosquitoes , suggesting the presence of a TRA-like activity ( Scali et al . , 2005; Salvemini et al . , 2011 , 2013 ) . However , a tra gene or its functional homolog has not yet been found in any mosquitoes . Anopheles mosquitoes contain well-differentiated X and Y chromosomes and genetic evidence suggests that a dominant M factor on the Y chromosome controls male development in Anopheles mosquitoes ( Baker and Sakai , 1979 ) . The Anopheles Y chromosome also regulates mating behavior ( Fraccaro et al . , 1977 ) . There is tremendous interest in deciphering Y gene function in non-model insects , including Anopheles mosquitoes , to shed light on the mechanism and evolution of sexual differentiation . There has also been strong interest in Y chromosome genes in Anopheles mosquitoes for translational motivations . Only female mosquitoes transmit disease pathogens because only females feed on blood . Thus , it is ideal , if not required , to release only males when considering genetic approaches for reducing mosquito populations or for replacing competent vector populations with populations that are refractory to disease transmission ( e . g . , [Collins , 1994; Benedict and Robinson , 2003; Windbichler et al . , 2008; Fu et al . , 2010; Black et al . , 2011; Harris et al . , 2012] ) . A better understanding of the Y chromosome function in sex-determination and male reproduction may provide novel targets for interference and lead to several practical applications . For example , transgenic lines may be obtained that produce male-only mosquitoes , resulting in more cost-effective mass production and sex separation methods than current approaches ( e . g . , [Papathanos et al . , 2009] ) . Furthermore , releasing such transgenic males is theoretically much more efficient than classic sterile insect techniques in achieving population reduction and disease control because of the added benefit of male-bias in subsequent generations ( Thomas et al . , 2000; Schliekelman et al . , 2005 ) . Despite strong interest and the availability of genomic resources ( e . g . , [Holt et al . , 2002] ) , earlier systematic efforts failed to identify Y genes in Anopheles mosquitoes ( Krzywinski et al . , 2006 ) . Several Y genes were recently discovered in An . stephensi and An . gambiae ( Criscione et al . , 2013; Hall et al . , 2013 ) by sequencing males and females separately . None of these genes is homologous to the Nix gene in Ae . aegypti ( Hall et al . , 2015 ) or encodes a predicted RNA-binding protein or splicing factor . However , among the An . gambiae Y chromosome genes , gYG2 ( An . gambiae Y Gene 2 ) is the most likely candidate for the M factor because it showed early embryonic expression ( Hall et al . , 2013 ) and it is the only Y gene that is shared among all species within the An . gambiae species complex ( Hall et al . , 2016 ) . In a recent report , gYG2 ( renamed YOB by the authors ) was shown to confer female-specific lethality in a transient embryonic assay and shift doublesex ( dsx ) splicing towards the male isoform in an An . gambiae cell line , suggesting that gYG2/YOB functions as an M factor in An . gambiae ( Krzywinska et al . , 2016 ) . Among the four Y genes identified in An . stephensi , Guy1 is the best candidate for the M factor because it is transcribed the earliest among all Y genes , at the very onset of embryonic development ( Criscione et al . , 2013 ) .
Guy1 transcription is transient and its transcript tapers off approximately 8–12 hr after egg deposition ( Criscione et al . , 2013 ) . We designed a plasmid , nGuy1 ( Figure 1A ) , which contains the entire Guy1 gene with its native promoter that was shown to function in the early embryo ( Criscione et al . , 2013 ) , to first test the function of Guy1 in An . stephensi by a transient embryonic assay . A plasmid that contains an enhanced green fluorescent protein ( EGFP ) marker controlled by D . melanogaster actin 5C promoter was co-injected to select for effective embryonic injections as indicated by the EGFP signal in the larvae . The adults that developed from EGFP positive larvae showed 25:1 and 19:2 male to female ratio ( Table 1 ) . When the same experiments were performed using Guy1m ( Figure 1B ) , a plasmid that is identical to nGuy1 except for a point mutation that changed the start codon ATG to AAG , the male-bias phenotype was no longer observed ( Table 1 ) , suggesting that the GUY1 protein is the cause of the phenotype . 10 . 7554/eLife . 19281 . 003Figure 1 . Four donor plasmids were used to generate transgenic Anopheles stephensi . nGuy1 and Guy1m were also used in the transient assays described in Table 1 . All constructs shown in the figure were flanked by the piggyBac arms to facilitate piggyBac-mediated integration into the An . stephensi genome ( Horn et al . , 2000 ) . The DsRed fluorescent marker gene under the control of the 3xP3/Hsp70 promoter ( 3xP3 ) was the transformation marker . PGUY1 refers to the native Guy1 promoter ( Criscione et al . , 2013 ) . Note that the only difference between nGuy1 and Guy1m is the point mutation in the first ATG . PbZip1 refers to a promoter derived from an An . stephensi gene ( Genbank JQ266223 ) and this promoter is used to drive early zygotic expression of the transgene ( Figure 3 ) . The C-tag and N-tag refer to the eight residue Strep II tag ( Lichty et al . , 2005 ) placed at either the C- or N-terminus of the GUY1 protein . A stretch of eight glycine residues were placed between the Strep II tag and the GUY1 protein ( Supplementary file 3 ) . The number of transgenic males and females were total counts from screens performed on all lines of each construct ( Supplementary files 1 and 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19281 . 00310 . 7554/eLife . 19281 . 004Table 1 . Transient injection of the nGuy1 , but not the Guy1m , plasmid in early embryos confers strong male bias in Anopheles stephensi* . DOI: http://dx . doi . org/10 . 7554/eLife . 19281 . 004PlasmidMaleFemalenGuy1 , replicate 1251nGuy1 , replicate 2192Guy1m , replicate 12434Guy1m , replicate 2106Notes:*An EGFP reporter plasmid under the control of the Drosophila melanogaster actin 5C promoter was co-injected with either the nGuy1 or Guy1m plasmid to ensure effective embryonic injections as indicated by an EGFP signal in the larvae . The adults that developed from EGFP positive larvae were sexed according to antennae morphology . Both nGuy1and Guy1m are piggyBac donor plasmids that contain the DsRed transformation marker ( Figure 1 ) . We were able to obtain transgenic lines using these donor plasmids together with a piggyBac helper plasmid to achieve germ line transposition . Two independent nGuy1 transgenic lines and one Guy1m line were obtained . Only a single transgene was inserted in each line , as indicated by digital droplet PCR ( not shown ) , a method that quantifies gene copy number ( Hindson et al . , 2011 ) . Inverse PCR and sequencing revealed independent insertion of the transgene in the two nGuy1 lines as the two nGuy1 insertion sites mapped to two different scaffolds ( Supplementary file 3 ) . Not a single transgenic female older than the third instar has been detected since we established the nGuy1 lines , while 2661 transgenic positive males were identified over 15 generations ( Figure 1 and Supplementary file 1 ) . Similar to what was observed during the transient assay , the Guy1m line showed no male-bias ( Figure 1B and Supplementary file 1 ) . Both nGuy1 and Guy1m transgenes are expressed in the early embryos in their respective lines ( Figure 2A and B ) . Table 2 shows the numbers of transgene positive and negative males and females of the seventh generation ( G7 ) . The lack of sex-bias in the non-transgenic progeny in all three lines indicates autosomal insertion of the transgenes . 10 . 7554/eLife . 19281 . 005Figure 2 . Early embryonic expression of the Guy1 transgene in nGuy1 ( A ) , Guy1m ( B ) , bGuy1C ( C ) , and bGuy1N ( D ) lines . ( A ) RT-PCR from 5–6 hr old genotyped single embryos . Lane 1 , Wild type male; Lane 2 , Transgenic female; Lane 3 , Wild type female . Genotyping was performed according to Criscione et al . ( 2013 ) . nGuy1 transcription is detected in transgenic females where there is no Y chromosome or endogenous Guy1 . Detecting nGuy1 expression in transgenic female embryos is the only way we can show that the nGuy1 transgene is expressed because there is no sequence difference between the nGuy1 transgene transcript and the endogenous Guy1 transcript . ( B ) A pseudo-gel image from the bioanalyzer ( Agilent Technologies , Inc . , Santa Clara , CA , USA ) showing the presence of RT-PCR products from the mutant Guy1m transgene . We took advantage of the Guy1m mutation ( ATG to AAG ) that introduced an MseI site so we do not have to perform RT-PCR on genotyped single embryos . RT-PCR was done using pooled 5–6 hr embryos . The three fragments in the MseI lane are 130 bp ( uncut wildtype Guy1 ) , 90 bp ( cut Guy1m ) , and 40 bp ( cut Guy1m ) . The presence of the MseI digested 90 and 40 bp fragments indicates that there were transcripts from theGuy1m transgene . Panels C and D show RT-PCR products using a Guy1 primer and a bZip1 UTR primer , which only amplify cDNA from the bGuy1C and bGuy1N transgenes ( Figure 1 ) . Eggs of 0–2 hr , 4–6 hr and 12–14 hr post oviposition were collected from A . stephensi bGuy1C-1 and bGuy1C-2 ( panel C ) and bGuy1D-1 and bGuy1D-2 ( panel D ) lines . RPS4 , ribosomal protein subunit 4 . All primers can be found in Table 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19281 . 00510 . 7554/eLife . 19281 . 006Table 2 . Sex ratios of transgenic and non-transgenic progeny of five transgenic lines* . DOI: http://dx . doi . org/10 . 7554/eLife . 19281 . 006LineTransgenic maleTransgenic femaleNegative maleNegative femalenGuy1-1 †7105856nGuy1-2 †12909183Guy1m †69622634bGuy1C-1 ‡7807566bGuy1N-1 ‡7706665Notes:*Crosses were done between the heterozygous transgenic males and wildtype females . Screening and sexing was initially done at the L3 instar stage . The sexed negative and positive larvae were reared separately to adulthood and sex was further confirmed on the basis of antennae morphology . The total numbers of each sex for the transgenic and non-transgenic groups are listed . †The numbers for these lines are from generation 7 ( G7 ) . Note that Guy1m has a point mutation that abolished the Guy1 open reading frame . There are more transgenic individuals than non-transgenic individuals in the Guy1m line because both transgenic females and wild type females were mated with transgenic males to maintain the line . ‡Two bGuy1C lines and nine bGuy1N lines were obtained and only one of each is shown in this Table . The numbers are from generation 4 ( G4 ) . Two additional transgenic lines were obtained using bGuy1C ( Figure 1C ) , which contains the 168 bp Guy1 open reading frame ( ORF ) plus a C-terminal Strep II tag ( Lichty et al . , 2005 ) . The expression of bGuy1C is controlled by an An . stephensi early zygotic promoter derived from the bZip1 gene ( Genbank JQ266223 ) . The 5’UTR is provided from bZip1 and the 3’ UTR from SV40 . Thus , the only Guy1 sequence in bGuy1C is the ORF . Nine additional transgenic lines were obtained using bGuy1N ( Figure 1D ) , a plasmid similar to bGuy1C except that the Strep II tag is at the N-terminus and the 119bp Guy1 3’ UTR was used instead of the SV40 3’ UTR . These 11 lines were produced from different G0 pools and , thus , are likely all independent . The bGuy1C lines produced 1156 transgenic males and 0 transgenic females and all nine bGuy1N lines produced 2703 transgenic males and 0 transgenic females in G2-4 ( Figure 1 and Supplementary file 2 ) . The early zygotic transcription of the Guy1 transgene is shown in Figure 2C and D . Transgenic lines that contain a functional Guy1 gene ( nGuy1 , bGuy1C , and bGuy1N ) produced 6520 transgenic males and zero transgenic females . These results further indicate that the GUY1 protein is the cause of the lack of females because the common feature of these lines is the Guy1 ORF while the promoters , 5’ and 3’ untranslated regions differ . The observation that all transgenic mosquitoes were males could result from either lethality or sex conversion of the XX individuals . We took advantage of an existing transgenic line that has a cyan fluorescent protein ( CFP ) marker cassette inserted on the X chromosome ( XCFP ) ( Amenya et al . , 2010 ) to monitor the genotype . The presence of the nGuy1 transgene on an autosome ( AGUY1 ) is monitored by the DsRed marker . Figure 3A shows the schematic of the crosses performed to produce progeny whose genotype can be monitored by CFP and DsRed expression ( Figure 3B ) . First , AGUY1aXY males were mated with aaXCFPXCFP females to obtain the AGUY1aXCFPY F1 progeny . These AGUY1aXCFPY males were subsequently crossed with wild type females ( aaXX ) to obtain progeny with four possible genotypes that can be differentiated based on CFP and DsRed expression: ( 1 ) AGUY1aXY , DsRed positive and CFP negative; ( 2 ) AGUY1aXCFPX , DsRed positive and CFP positive; ( 3 ) aaXCFPX , DsRed negative and CFP positive; and ( 4 ) aaXY , DsRed negative and CFP negative . As shown in Figure 3C and Figure 3—source data 1 , there is a complete lack of DsRed/CFP double positives when the F2 progeny were screened at 4th instar larvae ( L4 ) , indicating the lack of AGUY1aXCFPX larvae . All DsRed positives were confirmed to be males by the presence of a pair of testes . Sexing method was confirmed by allowing the larvae to develop into adults . We then repeated these experiments by screening the L1 instars to determine the timing of the death of AGUY1aXCFPX individuals . In four replicates , only 2–4% of the L1 larvae showed the AGUY1aXCFPX genotype and they all died within 8 hr of hatching . Thus , we have seen complete lethality of transgenic XX individuals prior to , or soon after , egg hatching . 10 . 7554/eLife . 19281 . 007Figure 3 . Analysis of the transgenic nGuy1-1 line indicates 100% female lethality prior to or soon after egg hatching . ( A ) A schematic and Punnett square showing the cross performed to genotype F2 progeny based on expression of fluorescent transformation markers . The red uppercase A indicates an autosome that carries the Guy1 transgene and the dsRED transformation marker gene , or AGUY1 . The cyan uppercase X indicates an X chromosome that carries a Cyan fl protein ( CFP ) transformation marker gene , or XCFP . AGUY1aXY males and aaXCFPXCFP females were initially crossed to obtain F1 AGUY1aXCFPY progeny . AGUY1aXCFPY were then crossed with wild type females ( aaXX ) to obtain progeny of four possible genotypes: AGUY1aXY , transgenic Guy1 males; AGUY1aXCFPX , transgenic Guy1 females; aaXCFPX , wild type females; aaXY , wild type males . ( B ) Images of transgenic L3 instar showing DsRed positive and CFP positive , respectively . ( C ) Distribution of the four genotypes in the F2 progeny at L1 and L4 instar stages , respectively . Analysis of the L1 and L4 instar is from two independent experiments . Percentages of each genotype were shown as the average of four replicates with standard error . The actual count of each genotype is provided in Figure 3—source data 1 . Note that all CFP-DsRed double positive L1 instars died within 8 hr after hatching . DOI: http://dx . doi . org/10 . 7554/eLife . 19281 . 00710 . 7554/eLife . 19281 . 008Figure 3—source data 1 . Number of the four types of progeny from wild type males mated with nGuy1-1 ( DsRed ) and CFP positive males . The mating strategy and progeny genotypes are described in Figure 3A . DOI: http://dx . doi . org/10 . 7554/eLife . 19281 . 008 Figure 4 and Figure 4—source data 1 show the results of the experiments to assess the reproductive competitiveness of the nGuy1 males under laboratory conditions . To mitigate the effect of different genetic backgrounds , transgenic males and their non-transgenic sibling males were subjected to competition for wild type virgin females . Two days after emergence , 20 transgenic males and 20 of their non-transgenic male siblings were placed in a 44 oz cup to compete for 10 wild type virgin females . The progeny of these females was screened for the Guy1 transgene and sexed at the L3 larval stage . Biological triplicates were performed for both nGuy1 lines . It was experimentally determined that transgenic females do not survive to the L3 stage ( Figure 3C ) . Thus , the expected fractions for transgenic males , non-transgenic females , and non-transgenic males were 1/7 , 3/7 , and 3/7 , respectively , assuming equal reproductive capability of the nGuy1 males and their siblings ( see Figure 4—source data 1 for genotype explanation ) . In all three experiments and for both lines , there are more transgenic nGuy1 offspring than expected under the assumption of equal reproductive ability ( Figure 4 and Figure 4—source data 1 ) . The results are statistically significant for both nGuy1 transgenic lines according to one-sample proportion tests ( Z=5 . 0 and 8 . 1 , respectively; p<0 . 001 for both lines , Figure 4—source data 1 ) . The significantly larger transgenic male population in comparison to the expected value suggests that the transgenic nGuy1 males were more competitive in reproductive output than their non-transgenic male siblings in the laboratory under our assay conditions . 10 . 7554/eLife . 19281 . 009Figure 4 . Reproductive competitiveness of transgenic males compared to their non-transgenic male siblings in two independent lines , nGuy1-1 and nGuy1-2 . Sibling cohorts of 20 transgenic and 20 non-transgenic males were mated with 10 wild type females . The resulting progeny were screened for transgenes at L3 instar stage as indicated by the DsRed marker and sexed by the presence or absence of testes . Shown in the figure percentages of male transgenics in biological triplicates for both lines . The red line indicates the expected percentage ( 1/7 or 14 . 29% ) of transgenic male progeny , assuming that the DsRed positive females do not survive beyond the L1 stage ( Figure 3 ) and the male parents ( transgenics and their non-transgenic brothers ) were equally productive ( detailed calculations are shown in Figure 4—source data 1 ) . Statistical analysis was performed using one-sample proportion tests for both lines ( Z=5 . 0 and 8 . 1 , respectively; p<0 . 001 in both cases , Figure 4—source data 1 ) . The significantly larger transgenic male population in comparison to the expected value suggests that the transgenic Guy1 containing males are reproductively more competitive than non-transgenic males under these laboratory conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 19281 . 00910 . 7554/eLife . 19281 . 010Figure 4—source data 1 . Assay for male reproductive competitiveness of nGuy1-1 and nGuy1-2 lines 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19281 . 010 We have previously shown that Guy1 is the earliest transcribed of all known Y genes in An . stephensi , at the very onset of maternal-to-zygotic transition ( Criscione et al . , 2013 ) . Here , we have shown that the native Guy1 promoter is active in the early embryos of both nGuy1 and Guy1m lines ( Figure 2A and B ) . Furthermore , the Guy1 promoter is active in female embryos , according to direct evidence by RT-PCR ( Figure 2A ) and according to the fact that nGuy1 transgene affects females during the embryonic stage . These results from transgenic lines clearly indicate that the early embryonic gene expression controlled by the native Guy1 promoter does not rely on any other Y chromosome factors because Y chromosome is not present in the female embryos .
The M factor is a primary signal from the male-determining chromosome or locus that triggers male development . We have previously shown that Guy1 is the earliest expressed Y chromosome gene and it is transcribed at the onset of maternal-to-zygotic transition , prior to sex-determination ( Criscione et al . , 2013 ) . We have now shown in transgenic lines that the native Guy1 promoter is able to direct Guy1 transcription as well as GUY1 function when placed in the autosome of genetic females which lack the Y chromosome . Thus , Guy1 is a primary signal , not a secondary signal , from the Y chromosome . We have also provided evidence that the small GUY1 protein is the functional molecule by analyzing various Guy1 mutants or constructs in transgenic lines . When ectopically expressed in females , the small GUY1 protein confers 100% female lethality during embryonic and early larvae stages . As discussed in detail below , manipulations of master switches of sex-determination also cause sex-specific lethality in fruit flies and nematodes ( Thomas et al . , 2012; Schutt and Nothiger , 2000; Penalva and Sánchez , 2003 ) . Taken together , we suggest that Guy1 is a strong candidate for the M factor in An . stephensi . Unlike the newly discovered M factor Nix in Ae . aegypti ( Hall et al . , 2015 ) , ectopic expression of Guy1 did not produce masculinized XX females . However , a recently reported M factor in An . gambiae , gYG2/YOB , also confers female-specific lethality instead of sex conversion in a transient embryonic assay using in vitro synthesized YOB mRNA ( Krzywinska et al . , 2016 ) . The cause of female lethality in the transgenic An . stephensi is unknown . A few scenarios are possible , one being related to dosage compensation , which is a mechanism known to occur in Anopheles mosquitoes ( [Hahn and Lanzaro , 2005; Mank et al . , 2011; Jiang et al . , 2015] Rose et al . , 2016 ) , including An . stephensi ( Jiang et al . , 2015 ) , to ensure equal levels of X-linked gene products in females ( XX ) and males ( XY ) . It is intriguing that three known master switches of sex-determination , Sex-lethal , Fem/Masc , and xo-lethal 1 , also regulate dosage compensation in D . melanogaster , Bombyx mori , and C . elegans , respectively ( [Thomas et al . , 2012; Schutt and Nothiger , 2000; Penalva and Sánchez , 2003; Kiuchi et al . , 2014] ) . Loss of function Sex-lethal alleles cause female embryonic lethality in D . melanogaster due to mis-regulation of dosage compensation ( Cline , 1978 ) . Similarly , knockdown of Masc also casues female-specific letahlity in B . mori , most likely due to mis-regulation of dosage compensation ( Kiuchi et al . , 2014 ) . If Guy1 is also involved in the regulation of dosage compensation in An . stephensi , ectopic expression of Guy1 in XX individuals may result in higher than normal levels of expression from X-linked genes , which could be lethal . Indeed , the sex-conversion phenotypes that resulted from ectopic expression of Nix were possible , presumably because Ae . aegypti does not require dosage compensation as it contains homomorphic sex-determining chromosomes ( Hall et al . , 2015 ) . There is no significant similarity between GUY1 and gYG2/YOB in their primary amino acid sequences . Thus , it will be difficult to determine whether Guy1 and YG2/YOB have a common evolutionary origin , especially as Y chromosome genes tend to evolve rapidly and similarities between very small proteins may not be easy to detect . However , both GUY1 and gYG2/YOB proteins are 56 amino acid long and have helical secondary structures ( Criscione et al . , 2013; Krzywinska et al . , 2016 ) . In addition , four out of the first five amino acid residues are identical between GUY1 and gYG2/YOB proteins . Thus , we cannot yet rule out the possibility that GUY1 and gYG2/YOB may be evolutionarily related . Regardless , it is unlikely that Guy1 and gYG2/YOB are complete functional orthologs because Guy1 is only expressed in the early embryos in An . stephensi ( tapering off at 8–12 hr after egg laying , Criscione et al . , 2013 ) while gYG2/YOB is expressed in the early embryos as well as in later developmental stages in An . gambiae ( Hall et al . , 2013; Krzywinska et al . , 2016 ) . Our hypothesis is that Guy1 is the initial or primary signal which triggers a cascade of events downstream . This is analogous to the early Sex-lethal gene product ( SXLe ) in D . melanogaster ( reviewed in [Herpin and Schartl , 2015] ) , which is only produced during the early embryonic stage . This transient primary signal , SXLe , is sufficient to trigger the female-specific cascade of events that are then maintained by downstream factors such as the late SXL protein throughout development . The transient expression of Guy1 suggests that it may function as a trigger or initiating signal , but not as a gene that maintains sexual identity . If additional Y genes are indeed required for either initiating or maintaining sex determination in An . stephensi , ectopic expression of Guy1 alone may not be sufficient to change dsx splicing in female cells/embryos due to the lack of Y chromosome . The transient and early embryonic expression of Guy1 also made it difficult to knockdown Guy1 simply because there is not enough time for RNAi to work in the early embryos . However , further characterization of other Y genes in An . stephensi may enable the experimental demonstration of potential Guy1 targets , which could lead to the elucidation of the Guy1-dependent developmental pathway in the early embryos . Our results demonstrated , for the first time in mosquitoes , that a Y chromosome gene , namely Guy1 , confers 100% female lethality when ectopically expressed from an autosome in XX individuals and this effect can be stably inherited for many generations . Recent work showed that ectopic expression of an M factor Nix in Ae . aegypti initiated male development in genetic females ( Hall et al . , 2015 ) and embryonic injection of in vitro synthesized gYG2/YOB mRMA conferred female-specific lethality in An . gambiae ( Krzywinska et al . , 2016 ) . However , the stability and penetrance of the phenotypes conferred by the Nix or gYG2/YOB transgenes remain to be determined as Nix or gYG2/YOB transgenic lines have yet to be reported ( Hall et al . , 2015; Krzywinska et al . , 2016 ) . Here , we have proved in principle that sustained sex ratio manipulation can be achieved using a Y chromosome gene with high penetrance . Furthermore , males that carry the Guy1 transgene are more competitive in overall reproductive output than their non-transgenic male siblings under laboratory conditions . Future studies based on analysis of progeny of individual females mated with transgenic and wildtype males will give insight regarding the mating competitiveness of the transgenic individuals . Competitive assays performed in semi-field or field conditions are needed to determine whether ectopic expression of Guy1 truly makes the males reproductively more competitive than wild type mosquitoes . Nonetheless , data presented in this study suggest that the Guy1 transgene shows no obvious detrimental effects in males and may be developed as a new method to generate male-only mosquitoes for release . As mentioned earlier , transgenic lines that produce only male progeny will improve current sex-separation methods and provide new ways to reduce mosquito population and disease transmission . The Guy1 transgenic lines offer several attractive features in this regard . With regard to sex separation , 100% female lethality is ideal for both effectiveness and ethical considerations of not accidentally releasing females . In addition , released females can compete with wild-type females to mate with the released males simply due to their presence and even through assortative mating . Thus , the complete penetrance of the male-only phenotype , as seen in every Guy1 transgenic line over many generations , with the exception of the negative controls , is a highly attractive feature when developing a novel genetic sex-separation approach based on Guy1 . The current mass production of male mosquitoes relies on physical means of sex separation , which is labor-intensive , expensive , and not 100% accurate ( [Alphey , 2014; Gilles et al . , 2014] ) . Transgenic lines that express fluorescent markers in a sex-specific pattern have been developed to allow machine-based sorting of the sexes ( e . g . , [Catteruccia et al . , 2005; Marois et al . , 2012] ) . However , such an approach is expensive with a relatively low throughput ( Gilles et al . , 2014 ) . Another positive feature of the Guy1 lines is the embryonic or early larvae lethality , which negates the need to rear female larvae altogether . The Guy1 transgenic lines also offer a few important advantages to strategies for reducing mosquito populations and disease transmission . The observed 100% female lethality will result in bias towards the non-biting males in multiple generations , which is theoretically much more efficient than classic sterile insect techniques in achieving population reduction and disease control ( [Thomas et al . , 2000; Schliekelman et al . , 2005] ) . An exciting report recently showed that an artificial sex ratio distortion system based on the shredding of the An . gambiae X chromosome produced >95% male offspring ( Galizi et al . , 2014 ) . The 100% female lethality described here was accomplished by simply expressing the endogenous Guy1 gene from the autosome . Thus , the Guy1-based genetic strategy involves minimal genetic manipulation . The existing Guy1 transgenic lines are heterozygous . Because no female transgenic individuals passed the first instar , homozygous individuals cannot be produced by mating transgenic males with transgenic females . Conditional control of transgenic Guy1 expression will enable the production of transgenic females required for generating homozygous transgenic individuals ( [Phuc et al . , 2007; Marinotti et al . , 2013; Fu et al . , 2007] ) , which is necessary for field applications . Alternatively , the Guy1 transgene may be linked to recently developed CRISPR/cas9-based gene drives ( Gantz et al . , 2015; Hammond et al . , 2016 ) for rapid spread of this female-lethal gene over multiple generations . This may represent an effective yet self-limiting strategy to control mosquito-borne infectious diseases because the spread of Guy1 could lead to severe reduction or local extinction of An . stephensi populations due to the lack of females ( Adelman and Tu , 2016 ) . Research on Guy1 has shown great promise for genetic control of mosquito-borne infectious diseases and revealed the complex and sex-specific effect of a Y gene during mosquito embryonic development . The unexplored mosquito Y chromosome may in fact turn out to be a gold mine for intriguing stories about sexual dimorphism and promising leads for new ways to control mosquito-borne infectious diseases .
Wild type Indian strain ( 'type' form ) of An . stephensi and transgenic An . stephensi were reared in incubators at 27°C and 60% relative humidity on a 16 hr light/8 hr dark photoperiod . Larvae were fed Sera Micron Fry Food with brewer's yeast , and Purina Game Fish Chow . Adult mosquitoes were fed on a 10% sucrose soaked cotton pad ( Criscione et al . , 2013 ) . All donor plasmids ( Figure 1A–D ) were constructed by inserting the gene of interest into a piggyBac donor plasmid backbone that contained the piggyBac arms and the DsRed transformation marker cassette ( Horn et al . , 2000 ) . The nGuy1 plasmid ( Figure 1A ) contained the native Guy1 gene , which included the full length Guy1 cDNA sequence ( 5’ UTR , CDS and 3’ UTR ) and the previously tested Guy1 native promoter ( Criscione et al . , 2013 ) . The native Guy1 gene was PCR amplified using genomic DNA from adult male An . stephensi as a template and primers are shown in Table 3 . A second round of PCR was performed to adapt AscI and PacI sites to clone into the above mentioned donor plasmid backbone . The nGuy1 plasmid and all other plasmids constructed in this study were verified by Sanger sequencing at the Virginia Bioinformatics Institute ( VBI ) on the Virginia Tech campus . The Guy1m plasmid ( Figure 1B ) is identical to nGuy1 except for a point mutation that changed the start codon ATG to AAG . The point mutation was introduced by synthesizing and replacing the fragment between the NheI and BglII sites in nGuy1 ( Epoch Life Science , Inc . , Missouri City , TX ) . bGuy1C ( Figure 1C ) contained the 168 bp Guy1 open reading frame ( ORF ) plus a C-terminal Strep II tag ( 22 ) and the 5’UTR was provided from an An . stephensi early zygotic gene bZip1 ( Genbank JQ266223 ) and the 3’ UTR from SV40 . Cloning of bGuy1C was done by gene synthesis ( Epoch Life Science , Inc . , Missouri City , TX ) and the sequence of the synthesized fragment is provided in the supplemental file . bGuy1N ( Figure 1D ) is similar to bGuy1C except that the Strep II tag is at the N-terminus and the 119 bp Guy1 3’ UTR was used instead of the SV40 3’ UTR . Cloning of bGuy1N was also done by gene synthesis ( Epoch Life Science , Inc . , Missouri City , TX ) and the sequence of the synthesized fragment is provided in the supplemental file . 10 . 7554/eLife . 19281 . 011Table 3 . Primers and probes used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 19281 . 011GUY1_Full-F1CCTGAAATGATGCTCTGGAAAGUY1_Full-R1GAAACGTTTTTCCAACATGTGAGUY1_Full+AscI-F1CACTGGCGCGCCCCTGAAATGATGCTCTGGAAAGUY1_Full+PacI-R1TCATGCAATTAATTGAAACGTTTTTCCAACATGTGARPS4-F2GAGTCCATCAAAGGAGAAAGTCTACRPS4-R2TAGCTGGCGCATCAGGTACsYG2_F2TGCCGGACATGACATTTGsYG2_R2TCAATGCGAACAGAAGGCTAADsRed_F2CCCCGTAATGCAGAAGAAGAGUY1_R2GATCCGTTAAAAATTGACACCAGUY1_F3TTTACTCGTCAAAGCTGCCAGUY1_R3GATCCGTTAAAAATTGACACCADsRed_F4CCCCGTAATGCAGAAGAAGADsRed_R4GGTGATGTCCAGCTTGGAGTDsRed_P4FAM-TACATGGCCAAGAAGCCCGT-BHQ1AutoRef_ddP_F4ATCACCACTCGTCGTCCGTTAutoRef_ddP_R4CGAACGAACTCGATTGACCCAutoRef_ddP_P4HEX-GCAAACACCACAACAGCAGC-BHQ1GUY1_ddP_F4GTCAAAGCTGCCACGGATCTGUY1_ddP_R4TCCAATGTCACAGCAGAGTGTTTGUY1_ddP_P4FAM-TCACAAAGTAGGCGATACAAAAACA-BHQ1iPCR_PBR_F5TACGCATGATTATCTTTAACGTAiPCR_PBR_R5TGGCTCTTCAGTACTGTCATiPCR_PBRnest_F5GTCACAATATGATTATCTTTCTAiPCR_PBRnest_R5CACTTCATTTGGCAAAATATGuy1m_F6TTAGATAACAGAAAGCGTACACTGuy1m_R6ACTTGATTTATCATTCCAAGTCAbGuy1C-F7GTTTATTGCAGCTTATAATGGTTACbGuy1C-R7TCTGACTTGGAATGATAAATCAAGbGuy1N_F7GAGAAAATAGCTGAATTGAAGGbGuy1N_R7AACACACAAAGTGGTCTTATGCRPS4_F7CACGAGGATGGATGTTGGACRPS4_R7ATCAGGCGGAAGTATTCACCNotes: Primers are presented in seven sets which alternates in white and grey backgrounds . ( 1 ) Amplification of the full length Guy1 gene sequence followed by adding a 5’ adaptor with an AscI site and a 3’ adaptor with a PacI site . ( 2 ) Primers used for genotyping embryos . RPS4 , ribosomal protein subunit 4 , was used as the positive control , sYG2 was used as a Y chromosome marker , and the transgene was amplified using a DsRed and a Guy1 primer , which is only present in transgenic individuals . ( 3 ) Primer sets used for RT-PCR of Guy1 on single embryos . ( 4 ) Digital droplet PCR primer sets used to determine Guy1 copy number and number of insertions in transgenic lines . ( 5 ) Inverse PCR primer sets used to determine the site of integration near the piggyBac right hand flanking sequences . ( 6 ) RT-PCR primers to amplify a 130 bp fragment of Guy1 to allow differentiation of Guy1m transcript from endogenous Guy1 transcript by MseI digestion . ( 7 ) RT-PCR primers used to amplify cDNA made from bGuy1C and bGuy1N transcripts . A set of RPS4 primers different from set 2 was used . The preblastoderm embryo injection was performed at the University of Maryland , College Park , Institute for Bioscience and Biotechnology Research’s Insect Transformation Facility ( http://www . ibbr . umd . edu/facilities/itf ) . For the nGuy1 and Guy1m experiments , an injection solution was used that contains a 150 ng/μl donor plasmid , a 300 ng/μl piggyBac transposase ( phsp-Pbac ) helper plasmid ( Horn et al . , 2000 ) , and a 200 ng/μl actin5C-EGFP plasmid . Crossing of G0 and screening were performed at Virginia Tech as described below . The bGuy1C and bGuy1N lines were generated at the University of Maryland Facility using a similar injection method as the one described above . No actin5c-EGFP plasmid was included . Instead , a plasmid mixture was included for monitoring the quality of injections consisting of a minimal Minos vector ( pMin QC ) and a source of Minos transposase ( pHSS6hsILMi20 ) . Injected embryos for nGuy1 and Guy1m constructs were obtained from the University of Maryland Insect Transformation Facility and reared to adulthood . G0 adult males were individually mated with five wild type females while G0 females were pooled and mated with wild type males at a 1 male to 5 female ratio . Mosquitoes were blood-fed on female Hsd:ICR ( CD-1 ) mice ( Harlan Laboratories , http://www . harlan . com ) . Egg collection occurred approximately 3 days post blood-feeding . Larvae were screened for DsRed expression during the L3 stage . G1 males and females were segregated after emergence and mated with wild type at a 1 male to 5 female ratio . The nGuy1 and bGuy1 lines produced no females , therefore , subsequent generations were maintained with the addition of up to 50 virgin wild type females . Females were produced in the Guy1m line and the line was allowed to inbreed . Inverse PCR was used to identify insertion sites of the nGuy1 and Guy1m transgenes . Genomic DNA was extracted with the gDNA Isolation Kit ( Zymo Research ) from 10 adult males in each transgenic line: nGuy1-1 , nGuy1-2 , and Guy1m . Digestion of 1 μg of gDNA was performed overnight at 37°C with 80 units of restriction enzyme and 1x digestion buffer . Digestions were performed with HaeIII , PacI , and MspI and inactivated at 85°C for 5 min . The digested DNA was then allowed to self-ligate overnight at 4°C in a 200 μL ligation reaction consisting of 20 μL 10X ligation buffer , 40 μL digestion reaction , 10 μL DNA ligase , and 130 μL H2O . The ligated product was then purified with the illustra GFX Gel Band Purification Kit ( GE Healthcare ) . Purified DNA was used as the template in PCR amplification using the piggyBac right hand primer sets ( Table 3 ) and the rTaq polymerase ( Takara ) . Cycling conditions for the first round of PCR and nested PCR were as follows: 95°C for 3 min; 30 cycles of 95°C for 30 s , 56°C for 30 s , 72°C for 3 min; and a final extension time of 5 min . PCR products were run on a 1% agarose gel , excised and purified with the illustra GFX Gel Band Purification Kit ( GE Healthcare ) . The purified PCR product was cloned into the pGEM-T easy vector . JM109 cells were transformed with each reaction mix and grown on LB+Ampicilin plates overnight at 37°C . Individual colonies were cultured in 3 mL LB overnight and PCR screened for the correct insertion . Plasmids were sequenced at VBI . The transgene copy number was determined by digital droplet PCR using DsRed primers and primers for an autosomal reference gene ( Table 3 ) as described in Hindson et al . ( Hindson et al . , 2011 ) . Analysis of the Guy1 transgene in the bGuy1C and bGuy1N lines were straightforward and transgene-specific primers can be used for RT-PCR ( Figure 2 and Table 3 ) . For the Guy1m line , we were able to take advantage of the ATG to AAG mutation , which generated a MseI site that was not present in the endogenous Guy1 transcripts . Thus , RNA was isolated from 5–6 hr embryos collected from Guy1m females and used for cDNA synthesis . Specific RT-PCR primers were designed to encompass a 130 bp region , which lacks the MseI site in the endogenous Guy1 transcript . RT-PCR products were then digested with MseI to show transgene expression as indicated by the appearance of 90 and 40 bp fragments ( Figure 2 ) . The uncut band reflects RT-PCR products from the endogenous Guy1 transcripts . For nGuy1 lines , there is no difference between the transgene and endogenous Guy1 transcripts . In order to confirm transgene expression , we have to genotype individual embryos to show that the Guy1 transgene is transcribed in transgenic females , which lack endogenous Guy1 . Individual 5–6 hr old embryos were homogenized in 5 uL Lysis Buffer ( 0 . 05M DTT and 10U RNase OUT from Invitrogen ) . The lysate was flash frozen and stored at −80°C until gDNA preparation or RNA extraction was required . The lysate was split in half , one for genotyping and one for RT-PCR ( 19 ) . For gDNA isolation 2 . 5 uL of the lysate was treated with 0 . 3 mAU of proteinase K for 30 min at 28°C , and then inactivated at 95°C for 2 min . The reaction was then diluted with 6 . 5 uL of ddH2O . The gDNA of each embryo was then used as the template to amplify RPS4 , sYG2 , and the transgene in PCR reactions for genotyping purposes ( see Table 3 for primers ) . Transgene was amplified by primers encompassing DsRed and Guy1 . sYG2 is a Y chromosome gene and RPS4 is the ribosomal protein subunit 4 positive control . sYG2 and RPS4 positive and transgene negative individuals were considered wild-type male embryos . sYG2 , RPS4 , and transgene positive individuals were considered transgenic males . Transgene and RPS4 positive and sYG2 negative embryos were considered transgenic females . Finally , RPS4 positive and sYG2 and transgene negative individuals were considered wild-type females . For RNA isolation 2 . 5 uL of the lysate was treated with 130U of DNase I ( Invitrogen , Carlsbad , CA ) for 1 hr at 25°C . The reaction was stopped with the addition of 1 μL of 25 mM EDTA and incubation at 65°C for 10 min . The resulting 4 . 5 μL reaction was carried over to cDNA synthesis using the SuperScript III reverse transcriptase and random hexamers . The cycling conditions for all reactions were as follows: 95°C for 5 min; 40 cycles at 95°C for 30 s , 60°C for 30 s , and 72°C for 3 min; and final extension at 72°C for 5 min . PCR products were run on a 1% agarose gel ( Figure 2 ) . Mating competence of transgenic males was measured through competition against non-transgenic sibling males for mates with wild type females . F1 transgenic males were mated with wild type virgin females . The resulting F2 progeny was checked for expression of DsRed and for the presence of testes . Males that were DsRed positive and DsRed negative ( siblings ) were reared separately in pans with equal larval density . Two days after emerging , twenty transgenic males and twenty non-transgenic males were placed in a 44 oz Coke cup ( 44TDCB ) together . Ten wild type virgin females ( approximately three days post pupal emergence ) were introduced and allowed to mate for twenty-four hours with the mixed males . The females were then blood fed for a minimum of thirty minutes and inspected for feeding . Seventy-two hrs post blood feeding , eggs were collected from these females for 24 hrs . The subsequent F3 progeny were screened for DsRed and sexed at the L3 larval stage . Biological triplicates were performed for these sets of experiments . The expected results for transgenic males , non-transgenic females , and non-transgenic males were 1/7 , 3/7 , and 3/7 , respectively ( see note of Figure 4—source data 1 for detailed explanations ) . It has already been experimentally determined that transgenic females do not survive to the L3 stage and are , thus , not calculated into the expected progeny ratios . | Much like humans , Anopheles mosquitoes have a pair of sex chromosomes that determine whether they are male or female: females have two X chromosomes , while males have an X and a Y . Genetic evidence has indicated that there is a dominant male-determining factor on the Y chromosome that acts as a master switch to cause mosquitoes to develop into males . Mosquitoes that lack a Y chromosome , and hence the male-determining factor , therefore develop into the default female sex . Because only female mosquitoes feed on blood and transmit disease-causing microbes – including those that cause malaria – there is strong interest in identifying the male-determining factor . Introducing this gene into females could allow mosquito sex ratios to be manipulated towards the harmless non-biting males . In 2013 , a study of male Anopheles stephensi mosquitoes identified a gene called Guy1 that is only found on the Y chromosome . Criscione et al . – who were involved in the 2013 study – now show that female A . stephensi mosquitoes die when the Guy1 gene is placed on their non-sex chromosomes . Further investigation confirmed that the protein produced from the Guy1 gene kills the females . This protein is an initiating signal that affects embryonic development in a sex-specific manner , making it a strong candidate to be the male determining factor in A . stephensi . This is consistent with previous reports in which the master switches of sex determination could be manipulated to kill specific sexes in fruit flies and nematode worms . Criscione et al . also found that males that carry the inserted Guy1 gene on their non-sex chromosomes – and so could potentially pass it on to both male and female offspring – are reproductively more competitive than their non-modified siblings under laboratory conditions . As the resulting female offspring would not survive , it is thus feasible , in principle , to genetically manipulate the sex ratio of the mosquitoes . A future challenge will be to identify how the protein encoded by the Guy1 gene acts to kill female mosquitoes . This knowledge will help to investigate the feasibility of using genetically modified mosquitoes to reduce Anopheles populations in order to control malaria . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | GUY1 confers complete female lethality and is a strong candidate for a male-determining factor in Anopheles stephensi |
Osteoarthritis , a highly prevalent degenerative joint disorder , is characterized by joint pain and disability . Available treatments fail to modify osteoarthritis progression and decrease joint pain effectively . Here , we show that intermittent parathyroid hormone ( iPTH ) attenuates osteoarthritis pain by inhibiting subchondral sensory innervation , subchondral bone deterioration , and articular cartilage degeneration in a destabilized medial meniscus ( DMM ) mouse model . We found that subchondral sensory innervation for osteoarthritis pain was significantly decreased in PTH-treated DMM mice compared with vehicle-treated DMM mice . In parallel , deterioration of subchondral bone microarchitecture in DMM mice was attenuated by iPTH treatment . Increased level of prostaglandin E2 in subchondral bone of DMM mice was reduced by iPTH treatment . Furthermore , uncoupled subchondral bone remodeling caused by increased transforming growth factor β signaling was regulated by PTH-induced endocytosis of the PTH type 1 receptor–transforming growth factor β type 2 receptor complex . Notably , iPTH improved subchondral bone microarchitecture and decreased level of prostaglandin E2 and sensory innervation of subchondral bone in DMM mice by acting specifically through PTH type 1 receptor in Nestin+ mesenchymal stromal cells . Thus , iPTH could be a potential disease-modifying therapy for osteoarthritis .
Osteoarthritis is the most common degenerative joint disorder in the USA and a leading cause of disability ( Centers for Disease Control and Prevention ( CDC ) , 2009; Murray et al . , 2012 ) . Chronic pain is a prominent symptom of osteoarthritis , affecting nearly 40 million people in the USA ( Peat et al . , 2001 ) . Pain itself is also a major risk factor for the development of functional limitation and disability in patients with osteoarthritis ( Lane et al . , 2010 ) . Unfortunately , treating osteoarthritis pain is challenging and represents a substantial unmet medical need . Available therapies ( nonsteroidal anti-inflammatory drugs , analgesics , and steroids ) do not provide sustained pain relief and can have substantial adverse effects ( Geba et al . , 2002; Karlsson et al . , 2002 ) . Inadequate control of chronic osteoarthritis pain is a major reason that patients seek surgical treatment . The sources and mechanisms of osteoarthritis pain may be multifactorial . Some studies have focused on synovial inflammation and cartilage degeneration ( Mathiessen and Conaghan , 2017; Malfait and Schnitzer , 2013 ) . Yet , osteoarthritis pain can manifest during very early stages of the disease , in the absence of synovial inflammation and independently of progressive cartilage degeneration . Many patients have radiographic osteoarthritic changes but no symptoms , whereas others have osteoarthritis pain with no radiographic indications ( Dieppe and Lohmander , 2005; Hannan et al . , 2000; Bedson and Croft , 2008 ) . Some patients have bilateral radiographic evidence of knee osteoarthritis yet have unilateral knee pain . Little attention has been paid to subchondral bone to control osteoarthritis pain . Intriguingly , subchondral bone marrow edema–like lesions are highly correlated with osteoarthritis pain ( Yusuf et al . , 2011; Kwoh , 2013 ) . Zoledronic acid , which inhibits osteoclast activity , reduces the size of bone marrow edema–like lesions and concomitantly alleviates pain ( Laslett et al . , 2012 ) . Analysis of data from the National Institutes of Health Osteoarthritis Initiative reported that patients taking bisphosphonates experienced significantly reduced knee pain at years 2 and 3 ( Laslett et al . , 2014 ) . Furthermore , patients with osteoarthritis report rapid and obvious pain relief after removal of deteriorated subchondral bone with overlying cartilage through knee replacement ( Isaac et al . , 2005; Reilly et al . , 2005 ) . Articular cartilage , which is characteristically aneural and avascular ( Bhosale and Richardson , 2008 ) , is incapable of generating pain primarily , suggesting that subchondral bone is a major source of osteoarthritis pain . Aberrant subchondral bone remodeling is a critical factor in pathological changes of osteoarthritis ( Zhen and Cao , 2014; Zhen et al . , 2013 ) , and sensory innervation induced by netrin-1 from aberrant subchondral bone remodeling is responsible for osteoarthritis pain ( Zhu et al . , 2019 ) . Furthermore , locally elevated levels of prostaglandin E2 ( PGE2 ) activate sensory nerves in porous endplates , leading to sodium influx through Nav1 . 8 channels , to mediate spinal hypersensitivity ( Ni et al . , 2019 ) . We have shown that bone homeostasis is maintained by temporal-spatial activation of transforming growth factor β ( TGF-β ) to couple bone resorption and formation ( Tang et al . , 2009 ) , in which subchondral bone is maintained in a native microarchitecture with blood vessels and nerves intertwined under normal conditions . However , excessive active TGF-β in the subchondral bone induces aberrant bone remodeling at the onset of osteoarthritis to promote its progression ( Zhen et al . , 2013 ) . Specifically , high levels of active TGF-β recruit mesenchymal stromal cells ( MSCs ) and osteoprogenitors in clusters , leading to abnormal bone formation and angiogenesis ( Zhen et al . , 2013 ) . Parathyroid hormone ( PTH ) , a U . S . Food and Drug Administration–approved anabolic agent for osteoporosis , regulates bone remodeling and calcium metabolism ( Crane and Cao , 2014; Qiu et al . , 2010 ) . The parathyroid gland , the main production site of PTH , evolved in amphibians and supported the transition from aquatic to terrestrial life , allowing terrestrial locomotion in previously aquatic vertebrates ( Okabe and Graham , 2004 ) . PTH prevents cartilage degeneration and/or the deterioration of subchondral bone ( Sampson et al . , 2011; Orth et al . , 2013; Bellido et al . , 2011; Yan et al . , 2014; Chang et al . , 2009; Lugo et al . , 2012; Eswaramoorthy et al . , 2012; Cui et al . , 2020; Morita et al . , 2018; Dai et al . , 2016; Chen et al . , 2018 ) , induces cartilage regeneration or chondrocytes proliferation in osteoarthritis ( Sampson et al . , 2011; Petersson et al . , 2006 ) , and stimulates subchondral bone and articular cartilage repair of focal osteochondral defects ( Orth et al . , 2013 ) . PTH also interacts with local osteotropic factors to orchestrate the coupling of bone resorption and formation in an anabolic signaling network ( Qiu et al . , 2010; Pfeilschifter et al . , 1995 ) , and PTH and TGF-β work in concert to exert their physiological activities in bone ( Qiu et al . , 2010 ) . TGF-β elicits its cellular response through the ligand-induced formation of a heteromeric complex containing TGF-β types 1 ( TβRI ) and 2 ( TβRII ) kinase receptors ( Tang et al . , 2009; Wrana et al . , 1992 ) . We have shown that PTH induces endocytosis of PTH type one receptor ( PTH1R ) with TβRII as a complex , and both PTH and TGF-β signaling are co-regulated during endocytosis ( Qiu et al . , 2010 ) . In this study , we investigated whether intermittent parathyroid hormone ( iPTH ) could attenuate osteoarthritis pain by modifying cartilage degeneration , subchondral bone microarchitecture , and subchondral sensory innervation . We found that iPTH reduces osteoarthritis pain and attenuates progression of osteoarthritis by attenuating subchondral bone deterioration and cartilage degeneration . Furthermore , iPTH reduces sensory innervation and the level of PGE2 in the subchondral bone and improves subchondral bone remodeling specifically through PTH1R on Nestin+ MSCs .
To examine the effect of iPTH on osteoarthritis pain , we generated a DMM model of osteoarthritis in 10 week old mice , administered daily PTH subcutaneously beginning 3 days after surgery , and assessed pain behavior during the following 8 weeks . We assessed three types of pain behavior: secondary hyperalgesia , primary hyperalgesia , and gait deficit . Secondary hyperalgesia of the affected hind paw was analyzed weekly by measuring 50% paw withdrawal threshold ( 50% PWT ) . During the first 2 weeks after surgery , 50% PWT decreased uniformly in all three groups of mice: sham-operated mice ( ‘sham mice’ ) , DMM mice that received vehicle daily ( ‘vehicle mice’ ) , and DMM mice that received PTH daily ( ‘PTH mice’ ) ( Figure 1A ) . The 50% PWT of sham mice returned to baseline by week 3 after surgery , indicating that secondary hyperalgesia during the first 2 weeks is attributable to surgery alone . The 50% PWT of vehicle mice continued to worsen from week 2 onward , but the 50% PWT of PTH mice stabilized at levels significantly higher than those of vehicle mice ( Figure 1A ) , indicating that iPTH initiated early in the development of osteoarthritis reduced secondary hyperalgesia . Primary knee hyperalgesia was analyzed weekly by measuring withdrawal threshold during direct knee press using a pressure application measurement ( PAM ) force transducer as previously described ( Miller et al . , 2017 ) . At week 1 after surgery , PAM withdrawal threshold ( PAMWT ) decreased uniformly in all three groups ( Figure 1B ) . PAMWT of sham mice recovered to near-baseline by week 4 . PAM withdrawal threshold of vehicle mice also improved by week 4 , albeit to a lesser extent compared with sham mice , and gradually declined between weeks 4 and 8 . PAMWT of PTH mice and vehicle mice changed in a similar pattern over time; however , the PAMWT of PTH mice was significantly higher than that of vehicle mice at every time point between week 2 and week 8 after surgery ( Figure 1B ) , indicating that iPTH initiated early in the development of osteoarthritis reduced primary hyperalgesia . Gait deficit was analyzed by performing gait analysis biweekly , comparing the swing speed , contact area , and paw intensity between the affected and contralateral hind paws ( Figure 1C ) . At week 8 after DMM surgery ( compared with sham surgery ) , the affected hind paw intensity , contact area , and swing speed decreased significantly relative to contralateral hind paws; however , these differences were less pronounced in PTH mice ( Figure 1D ) , indicating that iPTH initiated early in the development of osteoarthritis improved gait deficits . To examine the effect of iPTH on degeneration of articular cartilage in osteoarthritis , we performed Safranin O/fast green ( SOFG ) staining ( Figure 1E ) and used the Osteoarthritis Research Society International ( OARSI ) histologic grading system to assess articular cartilage damage . OARSI score increased progressively in DMM mice over 8 weeks , but this progression was reduced and slower in PTH mice ( Figure 1F ) , indicating that early initiation of iPTH in osteoarthritis slowed the progression of articular cartilage degeneration . We also assessed the effect of iPTH on chondrocyte degeneration by performing immunohistochemical analysis to measure the percentage of type X collagen-containing ( ColX+ ) and matrix metalloproteinase 13-containing ( MMP13+ ) chondrocytes in articular cartilage ( Figure 1G ) . Percentages of ColX+ and MMP13+ chondrocytes more than doubled in DMM mice compared with sham mice , and iPTH significantly reduced the magnitude of increase ( Figure 1H ) , indicating that iPTH inhibited chondrocyte degeneration in osteoarthritis . These results indicate that iPTH initiated early in osteoarthritis reduced pain and attenuated cartilage degeneration . To determine how iPTH reduced osteoarthritis pain , we examined the effect of iPTH on nociceptive nerve innervation in subchondral bone . DMM osteoarthritis joints were harvested at different time points for immunohistologic analysis of subchondral bone . The tibial subchondral bone was immunostained for calcitonin gene-related peptide ( CGRP ) , the markers of peptidergic nociceptive C nerve fibers . The density of CGRP+ nerve endings increased significantly after DMM surgery compared with sham surgery ( Figure 2A , B ) . Although the density of CGRP+ sensory nerve fibers also increased in PTH mice , the effect was significantly diminished relative to vehicle mice , indicating that iPTH inhibited the growth of nociceptive nerve fibers in osteoarthritic subchondral bone . Based on a newly proposed classification of sensory neurons ( Usoskin et al . , 2015 ) , additional markers of nociceptive neurons including Substance P ( SP ) ( Figure 2C ) , P2 × 3 and PIEZO2 ( Figure 2—figure supplement 1A , B ) were analyzed using immunofluorescence staining . The density of SP+ , P2 × 3+ , and PIEZO2+ nociceptive fibers also increased in the subchondral bone of vehicle mice compared with sham mice , whereas iPTH treatment attenuated the increase of these nociceptive innervations ( Figure 2D , Figure 2—figure supplement 1C , D ) . Sensory nerve fibers mostly innervate tissues together with small blood vessels . We also performed co-staining of CGRP and edomucin ( EMUN ) to examine the relationship between the CGRP+ nerve fibers and EMUN+ micro-vessels for investigating the effect of iPTH on the coupling of angiogenesis and sensory innervation in the subchondral bone . Like CGRP+ nerve fibers , the density of EMUN + vessels increased in subchondral bone of vehicle mice compared with sham mice , while iPTH treatment attenuated the increase in these vessels ( Figure 2E , F ) . We immunostained for CGRP in the joint synovium to determine whether the effect of PTH on sensory innervation was specific to subchondral bone . The density of CGRP+ sensory nerve endings increased after DMM surgery compared with sham surgery , and iPTH treatment had no influence on this effect ( Figure 2G , H ) . Together , these findings suggest that iPTH reduces osteoarthritis pain by specific inhibition of sensory nerve innervation in subchondral bone . Given that osteophytes can be responsible for increased pain sensation and abnormal gait behavior , we performed additional microcomputed tomography ( μCT ) analysis specifically examining the size of osteophyte in the DMM mice . There was no significant difference in osteophyte volume between PTH and vehicle-treated DMM mice , despite both are significantly higher than sham group ( Figure 2I , J ) . To examine the effect of iPTH on the PGE2 level in subchondral bone , we measured the expression level of cyclooxygenase 2 ( COX2 ) by immunohistochemical analysis and measured the concentration of PGE2 by enzyme-linked immunosorbent assay ( ELISA ) in tibial subchondral bone . The number of COX2+ cell increased significantly after DMM surgery in vehicle mice compared with sham mice , and iPTH diminished this increase ( Figure 2K , L ) . The level of PGE2 also increased after DMM surgery in vehicle mice compared with sham mice , and iPTH inhibited this effect ( Figure 2M ) . These results indicate that early initiation of iPTH reduced the level of PGE2 in osteoarthritis subchondral bone . To examine the effect of iPTH on subchondral bone architecture in osteoarthritis , we performed morphometric analysis on subchondral bone using 3-dimensional ( 3-D ) μCT analysis ( Figure 3A ) . Trabecular bone pattern factor ( Tb . pf ) , a measure of disruption in trabecular bone connectivity ( Hahn et al . , 1992 ) , increased rapidly by week 4 and plateaued in vehicle mice after DMM surgery ( Figure 3B ) , indicating uncoupled subchondral bone remodeling . Tb . Pf in PTH mice after DMM surgery was no different from sham mice within 2 weeks postoperatively , but increased by week 8 . However , Tb . Pf of PTH mice was significantly lower than that of vehicle mice ( Figure 3B ) , which indicates that early initiation of iPTH slowed the disruption of subchondral trabecular bone connectivity in osteoarthritis . Structure model index ( SMI ) , a measure to quantify the architectural type of cancellous bone ( Hildebrand and Rüegsegger , 1997 ) , increased significantly by week 4 . This increase was sustained in vehicle mice at week 8 after DMM surgery ( Figure 3C ) , indicating a change from plate-predominant to rod-predominant cancellous bone architecture in the development of osteoarthritis . SMI in PTH mice also increased after DMM surgery compared with sham surgery , but the magnitude of increase was less than that of vehicle mice at every time point ( Figure 3C ) . This difference indicates that early initiation of iPTH slowed the architectural deterioration of subchondral bone in osteoarthritis . Total volume of pore space ( Po . V ( tot ) ) and subchondral bone plate thickness ( SBP . Th ) increased gradually in vehicle mice and PTH mice after DMM surgery compared with sham mice , but the magnitudes of increase in PTH mice were significantly lower than those in vehicle mice at every time point ( Figure 3D , E ) . These results indicate that early initiation of iPTH slowed the deterioration and hypertrophy of subchondral bone microarchitecture in osteoarthritis . To investigate how iPTH improves subchondral bone microarchitecture , we assessed the localization of osteoid formation by performing trichrome staining and Calcein-Alizarin double labeling of tibial subchondral bone sections . Trichrome staining showed that osteoids formed islets in the subchondral bone marrow of DMM mice . iPTH induced formation of osteoids on subchondral bone surface in DMM mice , similar to sham mice ( Figure 3F ) . The result was validated in fluorescent double-labeling experiment ( Figure 3G ) . These results suggest that iPTH plays an essential role in sustaining coupled subchondral bone remodeling in osteoarthritis . To investigate the mechanism by which iPTH maintains coupled subchondral bone remodeling , we analyzed the effects of iPTH on the subchondral organization of osteoprogenitors , osteoclasts , and TGF-β signaling . Immunostaining for Nestin , a marker for adult bone marrow MSCs , showed a significantly higher number of Nestin+ cells in the subchondral bone marrow of vehicle mice compared with sham mice ( Figure 4A , B ) . Once committed to the osteoblast lineage , MSCs express Osterix , a marker of osteoprogenitors . The number of Osterix+ osteoprogenitors was also greater in the subchondral bone marrow of vehicle mice compared with sham mice ( Figure 4C , D ) . Intermittent PTH attenuated the increases of Nestin+ and Osterix+ cells in subchondral bone marrow of DMM mice ( Figure 4A–D ) , suggesting that early initiation of iPTH attenuates aberrant bone formation in osteoarthritis by modulating the recruitment of osteoprogenitors in subchondral bone . Immunostaining of pSmad2/3 revealed that the number of pSmad2/3+ cells in subchondral bone increased significantly after DMM surgery compared with sham surgery , but this effect was significantly attenuated in PTH mice ( Figure 4E , F ) . Tartrate-resistant acid phosphatase+ ( TRAP+ ) staining showed that the number of osteoclasts increased significantly in subchondral bone 4 weeks post-DMM surgery; interestingly , iPTH treatment further increased the number of TRAP+ osteoclasts in subchondral bone ( Figure 4G , H ) . Furthermore , ELISA of active TGF-β1 in serum revealed that active TGF-β1 concentration increased after DMM surgery compared with sham surgery , and iPTH further increased the active TGF-β1 concentration in DMM mice compared with vehicle mice ( Figure 4I ) . These results indicate that early initiation of iPTH attenuates aberrant subchondral bone remodeling by interfering with downstream TGF-β signaling in osteoarthritis . PTH1R and TβRII form an endocytic complex in response to PTH ( Qiu et al . , 2010 ) . To explore the mechanism by which PTH interferes with TGF-β downstream signaling , we analyzed the localization of TβRII on MSCs . TβRII was localized mainly at the MSC surface membrane in the negative control group and vehicle group , and the amount of cell-surface TβRII decreased significantly after PTH stimulation ( Figure 4J ) . Moreover , with stimulation of TGF-β1 , immunostaining of pSmad2/3 showed that phosphorylation and nuclear accumulation of Smad2/3 increased dramatically in the vehicle group compared with the negative control group; however , PTH decreased TGF-β1-induced phosphorylation and the nuclear accumulation of Smad2/3 compared with vehicle treatment in MSCs ( Figure 4K ) . Collectively , these results indicate that aberrant subchondral bone formation due to elevated TGF-β1 signaling in osteoarthritis was attenuated in the setting of PTH-induced endocytosis of TβRII . We tested a delayed regimen of starting iPTH to examine the effect of iPTH in a situation that mimics the clinical situation where therapy is initiated after the diagnosis of osteoarthritis . In this delayed iPTH treatment regimen , iPTH was initiated 4 weeks after DMM surgery . Delayed iPTH attenuated the DMM-associated decreases in 50% PWT and PAMWT ( Figure 5A , B ) . Gait analysis showed that iPTH attenuated the DMM-associated decreases in affected hind paw intensity , contact area , and swing speed relative to the contralateral side ( Figure 5C ) . The immunostaining of tibial subchondral bone sections revealed that iPTH significantly ameliorated the DMM-associated increases in the relative density of CGRP+ sensory nerve fibers in subchondral bone ( Figure 5D , E ) . ELISA analysis of DMM mice showed that delayed iPTH decreased the level of PGE2 in subchondral bone compared with vehicle-treated mice ( Figure 5F ) . Delayed iPTH attenuated the DMM-associated degeneration of articular cartilage as reflected by SOFG staining and OARSI scores ( Figure 5G , H ) . Similar to the early initiation of iPTH , a delayed regimen of iPTH significantly attenuated microarchitectural deterioration of subchondral bone after DMM surgery compared with vehicle mice ( Figure 5I ) . μCT analysis showed that delayed iPTH decreased Tb . Pf , SMI , Po . V ( tot ) , and SPB . Th compared with vehicle mice ( Figure 5J ) . These results indicate that iPTH can treat osteoarthritis pain and progression in later stages of the disease by attenuating nociceptive nerve innervation in subchondral bone and by preventing articular cartilage degeneration and deterioration of the subchondral bone microstructure . To validate the role of PTH-induced remodeling of subchondral bone in the attenuation of osteoarthritis pain and osteoarthritis progression , we induced conditional knockout of PTHIR in Nestin+ MSCs of DMM mice . Nestin-CreERT2::Pth1rfl/fl ( Pth1r−/− ) mice were injected with tamoxifen to delete PTH1R in the Nestin+ MSCs . Intermittent PTH had no effect on osteoarthritis pain when PTH1R was conditionally knocked out in MSCs , as reflected by no differences in 50% PWT , PAMWT , or gait analysis between iPTH–treated and vehicle-treated Pth1r−/− mice after DMM surgery , whereas the effect of iPTH on reducing osteoarthritis pain was re-demonstrated in Pth1rfl/fl ( Pth1r+/+ ) mice ( Figure 6A–C ) . This result indicates that iPTH attenuates osteoarthritis pain by signaling through PTH1R in Nestin+ MSCs in subchondral bone . Moreover , in DMM Pth1r–/– mice , iPTH was ineffective in reduction of subchondral sensory innervation , such as CGRP+ nociceptive nerve fibers , in coupling with decreased EMUN+ vessels post-DMM surgery ( Figure 6D–F ) . Additionally , iPTH was ineffective at reducing subchondral sensory innervation , such as SP+ , P2 × 3+ , and PIEZO2+ nociceptive nerve fibers after DMM surgery ( Figure 6—figure supplement 1A–E ) . Interestingly , the relative densities of CGRP+ sensory nerve fibers in the joint synovium were unaffected in DMM Pth1r–/– mice , with or without iPTH treatment after DMM surgery ( Figure 6G , H ) . Additionally , there were no obvious difference in osteophyte volume among these four DMM groups ( Figure 6I , J ) . Furthermore , iPTH failed to decrease the number of COX2+ cells and the level of PGE2 in tibial subchondral bone in Pth1r−/− mice ( Figure 6K , L ) , indicating that iPTH modulated the level of PGE2 in subchondral bone in osteoarthritis through its effects on subchondral bone . These results indicate that iPTH attenuates osteoarthritis pain by reducing sensory innervation and PGE2 expression in subchondral bone . Similarly , iPTH failed to prevent joint degeneration in Pth1r−/− mice after DMM surgery . As reflected by SOFG staining and OARSI scores , despite less protection of proteoglycan loss of articular cartilage by iPTH in Pth1r−/− mice compared with Pth1r+/+ mice , iPTH still significantly attenuated the DMM-associated degeneration of articular cartilage in Pth1r−/− mice compared with vehicle-treated Pth1r−/− mice ( Figure 7A , B ) . In Pth1r−/− mice , iPTH was ineffective at maintaining the microarchitecture of tibial subchondral bone after DMM surgery , including Tb . Pf , SMI , Po . V ( tot ) , and SBP . Th ( Figure 7C , D ) , but the protective effect of iPTH was re-demonstrated in Pth1r+/+ mice after DMM surgery , suggesting that iPTH maintains the subchondral microarchitecture in osteoarthritis via its effect on MSCs . The effect of iPTH on decreasing the number of pSmad2/3+ cells in tibial subchondral bone after DMM was abolished in Pth1r−/− mice compared with Pth1r+/+ mice ( Figure 7E , F ) . In addition , iPTH failed to decrease the number of Nestin+ MSCs and Osterix+ osteoprogenitors in the subchondral bone marrow in Pth1r−/− mice compared with Pth1r+/+ mice after DMM surgery ( Figure 7G–J ) . These findings support the notion that PTH sustains subchondral bone remodeling through inhibition of excessive active TGF-β signaling and suggest that the roles of PTH in the attenuation of osteoarthritis pain and the sustaining of subchondral bone microarchitecture are derived mainly from its role in subchondral bone .
The current treatments for osteoarthritis pain achieve limited therapeutic effects , and they are palliative for progressive pathological joint changes ( Hochberg et al . , 2012 ) . Although analgesics and nonsteroidal anti-inflammatory drugs were recommended in the 2012 American College of Rheumatology guidelines , these treatments produce unsustained and insufficient control of osteoarthritis pain with substantial adverse effects and fail to attenuate osteoarthritis progression effectively ( Hochberg et al . , 2012 ) . Joint replacement is the only alternative for end-stage osteoarthritis ( Thomas et al . , 2009 ) . Therefore , the development of an effective disease-modifying treatment for osteoarthritis is needed . We found that PTH reduced sensory innervation and the level of PGE2 in subchondral bone through inhibition of aberrant subchondral bone remodeling , which resulted in osteoarthritis pain relief . Importantly , PTH also attenuated osteoarthritis progression by inhibiting deterioration of subchondral bone microarchitecture and cartilage degeneration ( Figure 8 ) . Osteoarthritis pain originates primarily from synovium , ligaments , menisci , subchondral bone , and muscle and joint capsules ( Mathiessen and Conaghan , 2017; Reimann and Christensen , 1977; Ashraf et al . , 2011; Kc et al . , 2016; Hirasawa et al . , 2000; Belluzzi et al . , 2019 ) . In general , PTH attenuate osteoarthritis by maintaining microarchitecture of the subchondral bone and protection of articular cartilage ( Sampson et al . , 2011; Orth et al . , 2013; Bellido et al . , 2011; Yan et al . , 2014; Chang et al . , 2009; Lugo et al . , 2012; Morita et al . , 2018; Dai et al . , 2016; Chen et al . , 2018; Petersson et al . , 2006 ) . We have previously shown that aberrant subchondral bone is a critical source of pain in osteoarthritis ( Zhu et al . , 2019 ) . Abnormal subchondral bone remodeling induces aberrant sensory innervation ( Zhu et al . , 2019 ) , and elevated PGE2 induces hypersensitivity through sensory nerves ( Ni et al . , 2019 ) . In this study , PTH improved primary and secondary knee hyperalgesia and gait deficits in DMM mice . Furthermore , PTH effectively reduced the density of sensory nerve in the subchondral bone in DMM mice . In the Pth1r−/− mice after DMM surgery , PTH failed to improve osteoarthritis pain and reduce the sensory innervation and the level of PGE2 in the subchondral bone , but PTH worked effectively in Pth1r+/+ mice . We found that the density of CGRP+ sensory nerve fibers in the synovium increased post DMM surgery . This could be the reason that the pain behavior was not completely disappeared in the PTH treatment mice . The density of CGRP+ sensory nerve fibers in the synovium remained unchanged after PTH injection . PTH has been shown to inhibit the expression of pro-inflammatory modulators , including COX2 , in the synovial membrane of the osteoarthritis animal models ( Lugo et al . , 2012 ) . Also , the inflammatory response of synovial membrane was also reported to remain unaffected with PTH treatment ( Orth et al . , 2013 ) . The pain that originates from synovium partially reflects the severity of osteoarthritis and further limits overuse of knee joint . Considering that the articular cartilage has no nerve and blood vessel ( Schaible , 2012 ) , PTH-induced decrease of osteoarthritis pain is primarily due to its effect on subchondral bone . Blood vessels and nerve fibers often course alongside each other because they share similar mechanisms of wiring ( Carmeliet and Tessier-Lavigne , 2005 ) ; therefore , nerve fibers may undergo similar remodeling pattern to vessels with iPTH treatment . Co-immunostaining showed that CGRP+ nerve fibers is largely colocalized with endomucin+ vessels in the subchondral bone , both significantly decreased with iPTH treatment . The expression of PTH1R in arteriole ( Massfelder et al . , 2002; Song et al . , 2009 ) and endothelial cells ( Funk et al . , 2002; Isales et al . , 2000 ) has been reported in previous studies . Noteworthy , endogenous PTHrP downregulates the expression of PTH1R in vascular smooth muscle cells ( Song et al . , 2009 ) . As we all know , the vessel volume is significantly increased in the subchondral bone in osteoarthritis ( Zhen et al . , 2013 ) . PTH may inhibit vascularization or angiogenesis by negatively modulating the expression of PTH1R in the subchondral bone , where the vessel formation is usually accompanied by nerve growth . Thus , PTH reduces sensory innervation in the subchondral bone in association with decrease of blood vessels . Sensory innervation promotes osteoblast bone formation and suppresses osteoclast bone resorption ( Li et al . , 2017; Ishizuka et al . , 2005 ) . We have shown that sensory nerve regulates bone homeostasis and promote regeneration ( Chen et al . , 2019 ) . The ablation of sensory innervation by genetic or pharmacological approaches consistently results in decreased bone mass in adult mice ( Chen et al . , 2019; Fukuda et al . , 2013; Brazill et al . , 2019 ) . PTH induces osteoclastic bone resorption in coupling osteoblast bone formation during bone remodeling . Increased osteoclasts secrete PDGF-BB for type H vessel formation in support of the bone remodeling ( Xie et al . , 2014 ) . Accumulatively , the daily injection of PTH increases bone formation with net decrease of both total vessel formation and nerve innervation as the porous subchondral bone was decreased with new bone formation during the dynamic process . Therefore , iPTH decreased sensory nerve and attenuated osteoarthritis pain by a decrease of porous subchondral bone . The abnormal microarchitecture of subchondral bone caused by aberrant subchondral bone remodeling plays a causal role in the pathogenesis of osteoarthritis ( Zhen et al . , 2013; Pan et al . , 2012 ) . The maintenance of active TGF-β levels in a spatial and temporally manner is essential for coupled bone remodeling ( Tang et al . , 2009 ) , but excessive active TGF-β signaling in the subchondral bone leads to aberrant bone remodeling ( Zhen et al . , 2013 ) . PTH has been shown to preserve microarchitecture of subchondral bone and improves subchondral bone reconstitution of osteochondral defects ( Orth et al . , 2013; Bellido et al . , 2011; Yan et al . , 2014; Morita et al . , 2018; Dai et al . , 2016 ) . The decreased bone density and aberrantly elevated TGF-β levels were simultaneously occurred in the subchondral bone at the early stage of osteoarthritis ( Zhen et al . , 2013 ) . PTH induces internalization of PTH1R together with TβRII ( Qiu et al . , 2010 ) . iPTH improves the pathological subchondral bone microenvironment by downregulating the TGF-β signaling through endocytic complex of PTH1R and TβRII . Aberrant mechanical stress induces subchondral bone uncoupled remodeling at the onset of osteoarthritis and subsequently generates a pathological microenvironment with significantly increased PGE2 level and other inflammatory factors when reaching a certain threshold ( Rahmati et al . , 2016 ) . Importantly , subchondral bone has been shown as a source of inflammatory mediators and osteoarthritis pain . The abnormal microarchitecture of subchondral bone in osteoarthritis dramatically changes the stress distribution on articular cartilage . iPTH modulates the mechanical stress distribution on articular cartilage by improving microarchitecture of the subchondral bone . As a result , the production of pro-inflammatory factors , such as PGE2 , in subchondral bone is reduced . iPTH also has been shown to ameliorate both hyperplasia and fibrosis in osteoarthritis preceded by osteoporosis and inhibit expression of pro-inflammatory modulators , including COX2 , in synovial membrane ( Lugo et al . , 2012 ) . We revealed that iPTH have a therapeutic effect on osteoarthritis and pain by improving subchondral bone microenvironment through remodeling . For example , unmineralized or low mineralized bony tissues in the subchondral bone marrow cavity , not on the bone surface , were the typical osteoarthritis pathological change ( Zhen et al . , 2013 ) . The newly formed osteoid islets in the subchondral bone marrow were almost diminished in osteoarthritis mice with daily PTH injection . Articular cartilage degeneration is the primary concern in osteoarthritis . PTH has been reported to inhibit cartilage degradation , terminal differentiation , and apoptosis of chondrocytes and to promote regeneration of articular cartilage in osteoarthritis ( Sampson et al . , 2011; Orth et al . , 2013; Bellido et al . , 2011; Yan et al . , 2014; Chang et al . , 2009; Lugo et al . , 2012; Chen et al . , 2018; Petersson et al . , 2006 ) . The studies on PTH effect on osteoarthritic subchondral bone showed iPTH administration resulted in improvement of subchondral bone structure ( Bellido et al . , 2011; Yan et al . , 2014; Dai et al . , 2016; Orth et al . , 2013 ) or contradict result of induction of osteoarthritis ( Orth et al . , 2014 ) . PTH effect on the subchondral bone remains unclear . The current study provides evidence for the mechanism to clarify the different results . The result of PTH induction of osteoarthritis was conducted in normal mice rather than osteoarthritis animal models . In normal mice with no joint osteoarthritis , their subchondral bones have normal bone density and appropriate microstructure , which are in balance with articular cartilage . iPTH treatment generates additional new bone on top of normal subchondral bone density and structure . As a result , the changes of subchondral bone disrupt its balanced interplay with articular cartilage and leads to cartilage degeneration . Whereas in osteoarthritis mice , the subchondral bone is pathologically changed from coupled remodeling to uncoupled remodeling with porous structure . iPTH administration stimulates osteoclast remodeling and generates new bone to improve its structure quality and its interaction with articular cartilage . More importantly , the beneficial effect of PTH on subchondral bone deterioration and pain relief was diminished in the osteoarthritis Pth1r−/− mice . The protective effects of PTH on cartilage degeneration were also impaired in Pth1r−/− mice . Thus , our data reveal that PTH protects articular cartilage from degeneration through both articular cartilage and subchondral bone .
We purchased male C57BL/6J mice ( wild-type [WT] mice ) from Charles River Laboratories ( Wilmington , MA ) . We anesthetized mice at 10 weeks of age with xylazine ( Rompun , Sedazine , AnaSed; 10 mg/kg , intraperitoneally ) and ketamine ( Vetalar , Ketaset , Ketalar; 100 mg/kg , intraperitoneally ) . Then , the DMM model was created by transecting the meniscotibial ligament that connects the lateral side of the medial meniscus with the intercondylar eminence of the tibia to induce instability of the left knee . Sham DMM operations were performed by opening the joint capsule of the left knee of independent mice . Sham mice or DMM mice were injected subcutaneously with 40 µg/kg per day of human PTH ( Centers for Disease Control and Prevention ( CDC ) , 2009; Murray et al . , 2012; Peat et al . , 2001; Lane et al . , 2010; Geba et al . , 2002; Karlsson et al . , 2002; Mathiessen and Conaghan , 2017; Malfait and Schnitzer , 2013; Dieppe and Lohmander , 2005; Hannan et al . , 2000; Bedson and Croft , 2008; Yusuf et al . , 2011; Kwoh , 2013; Laslett et al . , 2012; Laslett et al . , 2014; Isaac et al . , 2005; Reilly et al . , 2005; Bhosale and Richardson , 2008; Zhen and Cao , 2014; Zhen et al . , 2013; Zhu et al . , 2019; Ni et al . , 2019; Tang et al . , 2009; Crane and Cao , 2014; Qiu et al . , 2010; Okabe and Graham , 2004; Sampson et al . , 2011; Orth et al . , 2013; Bellido et al . , 2011; Yan et al . , 2014; Chang et al . , 2009; Lugo et al . , 2012; Eswaramoorthy et al . , 2012; Cui et al . , 2020 ) ( Sigma–Aldrich , St . Louis , MO ) or the equivalent volume of vehicle ( phosphate-buffered saline [PBS] ) . For immediate regime of administration , daily injection of PTH was initiated 3 days after DMM surgery . The operated mice were euthanized at 2 , 4 , or 8 weeks after surgery ( n = 8–12 mice per group ) . For delayed regime of administration , daily injection of PTH was initiated 4 weeks after DMM surgery ( n = 8–12 mice per group ) . We purchased Nestin-creERT2 mice from The Jackson Laboratory ( Bar Harbor , ME ) . Mice with floxed Pth1r ( Pth1rfl/fl ) were obtained from the laboratory of Dr . Henry Kronenberg ( Kobayashi et al . , 2002 ) . Heterozygous male Nestin-creERT2 mice were crossed with Pth1rfl/flmice . The offspring were intercrossed to generate the following genotypes: WT mice , Nestin-creERT2 mice , Pth1rfl/fl ( herein , Pth1r+/+ ) mice , and Nestin-creERT2::Pth1rfl/fl mice ( herein , Pth1r−/− ) , in which Cre was fused with a mutated estrogen receptor that could be activated by tamoxifen . We determined the genotype of transgenic mice by polymerase chain reaction ( PCR ) analyses of genomic DNA isolated from mouse tails . The required primers were as follows: Pth1r: Forward 5′−3′: TGGACGCAGACGATGTCTTTACCA , Reverse 5′−3′: ACATGGCCATGCCTGGGTCTGAGA; Cre: Forward 5′−3′: ACCAGAGACGGAAATCCATCGCTC , Reverse 5′−3′: TGCCACGACCAAGTGACAGCAATG . We performed DMM surgery on 10 week old male Pth1r+/+ mice and Pth1r−/− mice . Three days after surgery , we treated each group with tamoxifen ( 100 mg/kg ) daily , and the mice were injected subcutaneously with either PTH ( 40 μg/kg/day ) or the equivalent volume of vehicle ( PBS ) subcutaneously daily for 4 or 8 weeks ( n = 8 mice per treatment group ) . Mice were euthanized with an overdose of inhaled isoflurane . We obtained whole blood samples by cardiac puncture immediately after euthanasia . Serum was collected by centrifuge at 200 × g for 15 min and stored at −80°C before analysis . The mice were then flushed with PBS for 5 min , followed by 10% buffered formalin perfusion for 5 min via the left ventricle . Then , the left knees were dissected and fixed in 10% buffered formalin for 48 hr . We isolated bone marrow MSCs from male WT mice at 6 weeks of age , as described by Soleimani and Nadri , 2009 . We maintained cells ( passage 3–5 ) in Iscove’s modified Dulbecco’s medium ( Invitrogen , Carlsbad , CA ) supplemented with 10% fetal calf serum ( Atlanta Biologicals , Flowery Branch , GA ) , 10% horse serum ( Thermo Fisher Scientific , Waltham , MA ) , and 1% penicillin–streptomycin ( Mediatech , Inc , Manassas , VA ) . We cultured bone marrow MSCs in six-well plates at a density of 1 . 8 × 105 cells per well; MSCs were then starved for 12 hr , followed by human PTH ( Centers for Disease Control and Prevention ( CDC ) , 2009; Murray et al . , 2012; Peat et al . , 2001; Lane et al . , 2010; Geba et al . , 2002; Karlsson et al . , 2002; Mathiessen and Conaghan , 2017; Malfait and Schnitzer , 2013; Dieppe and Lohmander , 2005; Hannan et al . , 2000; Bedson and Croft , 2008; Yusuf et al . , 2011; Kwoh , 2013; Laslett et al . , 2012; Laslett et al . , 2014; Isaac et al . , 2005; Reilly et al . , 2005; Bhosale and Richardson , 2008; Zhen and Cao , 2014; Zhen et al . , 2013; Zhu et al . , 2019; Ni et al . , 2019; Tang et al . , 2009; Crane and Cao , 2014; Qiu et al . , 2010; Okabe and Graham , 2004; Sampson et al . , 2011; Orth et al . , 2013; Bellido et al . , 2011; Yan et al . , 2014; Chang et al . , 2009; Lugo et al . , 2012; Eswaramoorthy et al . , 2012; Cui et al . , 2020 ) ( Sigma–Aldrich , St . Louis , MO ) and/or TGF-β1 ( R and D Systems , Minneapolis , MN ) treatment as indicated . The joint samples were decalcified in 10% ethylenediamine tetraacetic acid ( EDTA , pH 7 . 4 ) for 14 days and embedded in paraffin or Optimal Cutting Temperature Compound ( Sakura Finetek , Torrance , CA ) . Four micrometer thick sagittal-oriented sections of the medial compartment of the knee were processed for Safranin O ( Sigma–Aldrich , S2255 ) and fast green ( Sigma–Aldrich , F7252 ) staining , TRAP staining ( Sigma–Aldrich , 387A-1KT ) , and immunohistochemistry staining using a standard protocol . Ten micrometer thick sagittal-oriented sections were used for Nestin immunofluorescent staining , and 30 μm thick sagittal-oriented sections were used for nerve-related immunofluorescent staining using a standard protocol . The tissue sections were incubated with primary antibodies to mouse pSmad2/3 ( 1:50 , sc-11769 , Santa Cruz ) , Osterix ( 1:600 , ab22552 , Abcam ) , TβRII ( 1:100 , ab186838 , Abcam ) , β-Actin ( 1:3000 , CST3700 , Cell Signaling Technology ) , COX2 ( 1:100 , ab15191 , Abcam ) , MMP13 ( 1:200 , ab39012 , Abcam ) , COLX ( 1:200 , ab58632 , Abcam ) , Nestin ( 1:300 , NES0407 , Aves Labs , Inc , Davis , CA ) , CGRP ( 1:200 , ab81887 , Abcam ) , Substance P ( 1:200 , sc-21715 , Santa Cruz ) , endomucin ( 1:50 , sc-65495 , Santa Cruz ) , PIEZO2 ( 1:300 , ab243416 , Abcam ) , and P2 × 3 ( 1:500 , ab10269 , Abcam ) overnight at 4°C in a humidifier chamber . For immunohistochemical staining , a horseradish peroxidase–streptavidin detection system ( Dako , Agilent Technologies , Santa Clara , CA ) was used to detect immunoactivity , followed by counterstaining with hematoxylin ( Sigma–Aldrich ) . For immunofluorescence staining , slides were incubated with corresponding secondary antibody at room temperature for 1 hr while avoiding light . Then , the sections were counterstained with 4′ , 6-diamidino-2-phenylindole ( DAPI , Vector , H-1200 , Vector Laboratories , Inc , Burlingame , CA ) . The sample image were captured with Zeiss LSM 780 confocal microscope ( Carl Zeiss Microscopy , LLC , White Plains , NY ) or an Olympus BX51 microscope ( Olympus Scientific Solutions Americas Inc , Waltham , MA ) . Quantitative histomorphometric analysis was performed using Image J software ( Media Cybernetics Inc , Rockville , MD ) in a blinded fashion . We calculated OARSI scores as previously described ( Pritzker et al . , 2006 ) . For quantification of relative intensity of nerve fiber , positive fluorescence signals of the entire subchondral bone was threshold ( threshold range: 50–255 ) and calculated using Image J ( Version 1 . 49 ) . One section at the similar sagittal location of each mouse ( three slices per mouse and eight mice per group ) was calculated and normalized to that of sham mice ( set average to 1 ) . A double-labeling procedure was performed to label mineralization deposition . Briefly , we injected the mice subcutaneously with 0 . 1% Calcein ( 10 mg/kg , Sigma–Aldrich ) and 2% Alizarin red ( 20 mg/kg , Sigma–Aldrich ) in a 2% sodium bicarbonate solution 10 days and 3 days , respectively , before sacrifice . We used a fluorescence microscope to capture labeling images of undecalcified bone slices . Behavioral tests were performed 2 days before surgery and weekly after surgery . All behavioral tests were performed by the same investigators , who were blinded to the allocation of groups . Automated gait analysis was performed on walking mice using a ‘catwalk’ system ( CatWalk XT , Noldus Information Technology , Leesburg , VA ) . All experiments were performed and analyzed as previously reported ( Hamers et al . , 2006; Hamers et al . , 2001 ) . Gait analysis was performed in a room that was dark except for the light from the computer screen . Briefly , mice were trained to cross the catwalk walkway daily for 7 days before surgery . During the test , each mouse was placed individually in the catwalk walkway and allowed to walk freely from one side to the other side of the walkway’s glass plate . Light from an encased fluorescent lamp was emitted inside the glass plate and completely internally reflected . When the mouse paws contacted the glass plate , light was reflected down , and the illuminated contact area was recorded with a high-speed color video camera positioned under the glass plate and connected to a computer running CatWalk XT software , v9 . 1 ( Noldus Information Technology ) . Comparison was made between the left hind paw and right hind paw in each run of each animal . We calculated the following gait parameters: left hind paw ( LH ) /right hind paw ( RH ) contact area , LH/RH intensity , and LH/RH swing speed . The 50% PWT was measured by von Frey hair algesiometry . Mice were habituated to elevated plexiglass chambers and wire mesh flooring before assessments of allodynia . Then , ipsilateral hind paw mechanosensitivity was measured using a modification of the Dixon up-down method ( Dixon , 1980 ) . Allodynia was evaluated by applying von Frey hairs in ascending order of bending force ( force range: 0 . 07 , 0 . 40 , 0 . 60 , 1 . 0 , 1 . 4 , 2 . 0 , 4 . 0 , or 6 . 0 g ) . The von Frey hair was applied perpendicular to the plantar surface of the hind paw ( avoiding the toe pads ) for 2–3 s . If no response , the next higher strength of hair was used , up to the maximum level of 6 g of bending force . If a withdrawal response occurred , the paw was re-tested , starting with the next descending von Frey hair until no response occurred . Four more measurements were made after the first difference was observed . The 50% PWT was determined by using the following formula:50%PWT=10Xf+kδ/10 , 000 , where Xf is the exact value ( in log units ) of the final test of von Frey hair , K is the tabular value for the pattern of the last six positive/negative responses , and δ is the mean difference ( in log units ) between stimuli . The threshold force required to elicit paw withdrawal ( median 50% withdrawal ) was determined twice on each hind paw ( and averaged ) on each testing day , with sequential measurements separated by at least 10 min . The withdrawal thresholds in response to direct pressure on the medial part of the left knee were measured as pressure hyperalgesia ( Kim et al . , 2011 ) . A 5 mm diameter sensor tip of an applied force gauge ( SMALGO algometer; Bioseb , Pinellas Park , FL ) was used for the vocalization thresholds . The pressure force on the knee was increased at 50 g/s until an audible vocalization was made while the mice were restrained gently by the investigator . The curve of the pressure force was recorded by using BIO-CIS software ( Bioseb ) to ensure the force increased gradually . A cutoff force of 500 g was used to prevent trauma to the joint . The test was performed twice at an interval of 15 min , and the mean value was recorded as the nociceptive threshold . We dissected the left knee of mice free of soft tissue and fixed the samples in 10% buffered formalin for 48 hr . The joint samples were then transferred into PBS and analyzed by high-resolution μCT ( SkyScan 1172 , Bruker-microCT , Kontich , Belgium ) . The scanner was set at a voltage of 65 kV , a current of 153 μA , and a resolution of 9 μm per pixel to measure the subchondral bone and joint . Images were reconstructed and analyzed using NRecon v1 . 6 and CTAn v1 . 9 ( Skyscan US , San Jose , CA ) , respectively . Sagittal images of the tibial subchondral bone were used to perform 3-D histomorphometric analyses . The region of interest was defined to cover the whole medial compartment of subchondral bone . We used six consecutive images from the medial tibial plateau for 3-D reconstruction and analysis using 3-D model visualization software ( CTVol v2 . 0 , Skyscan US ) . 3-D structural parameters were analyzed: Tb . Pf , SMI , Po . V ( tot ) , and SBP . Th . The concentrations of active TGF-β1 in serum were determined using the TGF-β1 ELISA Development Kit ( MB100B , R and D Systems ) , according to the manufacturer's instructions . The level of PGE2 in subchondral bone was determined using the PGE2 Parameter Assay Kit ( KGE004B , R and D Systems ) . For sample preparation , we snap-freezed and crushed the subchondral bone in liquid nitrogen , right after tissue harvest and repaid removal of surrounding tissue and cartilage . Then homogenization buffer ( 0 . 1 M phosphate , pH 7 . 4 , containing 1 mM EDTA and 10 μM indomethacin ) were added ( ratio: 5 ml buffer to 1 g tissue ) . The sample were homogenized with a Polytron-type homogenizer . The tissue homogenates were spun at 8000 × g for 10 min , and the supernatant was collected for ELISA assay , according to the manufacturer’s instructions . All analyses were performed using SPSS , v15 . 0 , software ( IBM Corp . , Armonk , NY ) . Data are presented as means ± standard deviations . The data were normally distributed , unless otherwise noted . The PAMWTs and 50% PWTs were analyzed by two-way repeated-measures ANOVA with Bonferroni’s post hoc test in immediate and delayed regime of PTH administration . All other sets of data at 2 weeks , 4 weeks , and 8 weeks were analyzed by two-way ANOVA with Bonferroni’s post hoc test . For other comparisons among multiple groups , we used one-way ANOVA with Bonferroni’s post hoc test . For all experiments , the level of significance was set at p<0 . 05 . All inclusion and exclusion criteria were pre-established . No statistical method was used to predetermine the sample size . Experiments were randomized , and the investigators were blinded to allocation during experiments and outcomes assessment . All animal experiments were performed in accordance with NIH policies on the use of laboratory animals . All experimental protocols were approved by the Animal Care and Use Committee of The Johns Hopkins University . | Over time the cartilage between our bones gets worn down , and this can lead to a painful joint disorder known as osteoarthritis . Nearly 40 million people with osteoarthritis in the United States experience chronic pain . Although there are a number of drugs available for these patients , none of them provide sustained pain relief , and some have substantial side effects when ingested over a long period of time . Bone tissue is continuously broken down into minerals , such as calcium , that can be reabsorbed into the blood . In 2013 , a group of researchers found that the tissue in the layer of bone below the cartilage – known as the subchondral bone – is reabsorbed and replaced incorrectly in patients with osteoarthritis . This irregular ‘remodeling’ stimulates nerve cells to grow into the subchondral layer , leading to increased sensitivity in the joint . A protein called parathyroid hormone , or PTH for short , plays an important role in the loss and formation of bone . A drug containing PTH is used to treat patients with another bone condition called osteoporosis , and could potentially work as a treatment for osteoarthritis pain . To investigate this , Sun et al . – including some of the researchers involved in the 2013 study – tested this drug on a mouse model that mimics the symptoms of osteoarthritis . This revealed that PTH significantly decreases the number of nerves present in the subchondral bone , which caused the mice to experience less pain . PTH also slowed down the progression of osteoarthritis , by preventing the cartilage on the subchondral layer from deteriorating as quickly . Sun et al . found that the subchondral bones of treated mice also had a more stable structure and reduced levels of a protein involved in the reabsorption of bone . The results suggest that PTH is able to correct the errors in bone remodeling caused by osteoarthritis , and that this drug could potentially alleviate patients’ chronic pain . This drug has already been approved by the US Food and Drug Administration ( FDA ) , and could be used in clinical trials to see if PTH has the same beneficial effects on patients with osteoarthritis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine"
] | 2021 | Parathyroid hormone attenuates osteoarthritis pain by remodeling subchondral bone in mice |
During infection , pathogens are starved of essential nutrients such as iron and tryptophan by host immune effectors . Without conserved global stress response regulators , how the obligate intracellular bacterium Chlamydia trachomatis arrives at a physiologically similar ‘persistent’ state in response to starvation of either nutrient remains unclear . Here , we report on the iron-dependent regulation of the trpRBA tryptophan salvage pathway in C . trachomatis . Iron starvation specifically induces trpBA expression from a novel promoter element within an intergenic region flanked by trpR and trpB . YtgR , the only known iron-dependent regulator in Chlamydia , can bind to the trpRBA intergenic region upstream of the alternative trpBA promoter to repress transcription . Simultaneously , YtgR binding promotes the termination of transcripts from the primary promoter upstream of trpR . This is the first description of an iron-dependent mechanism regulating prokaryotic tryptophan biosynthesis that may indicate the existence of novel approaches to gene regulation and stress response in Chlamydia .
Nutrient acquisition is critical for the success of pathogenic bacteria . Many pathogenic bacteria must siphon nutrients from their hosts , such as nucleotides , amino acids and biometals ( Brown et al . , 2008; Eisenreich et al . , 2010; Ray et al . , 2009; Skaar , 2010 ) . This common feature among pathogens renders them susceptible to nutrient limitation strategies associated with the host immune response ( Hood and Skaar , 2012 ) . Counteractively , bacterial pathogens have evolved sophisticated molecular mechanisms to respond to nutrient deprivation , involving increasingly complex and sophisticated nutrient-sensing regulatory networks . These stress response mechanisms are essential for pathogens to avoid clearance by the immune system . By delineating their function at the molecular level , we can better target aspects of the host-pathogen interface suitable for therapeutic manipulation . However , stress responses in the obligate intracellular bacterium Chlamydia trachomatis are relatively poorly characterized , leaving unanswered many fundamental questions about the biology of this pathogen . C . trachomatis is the leading cause of bacterial sexually transmitted infections ( STIs ) and infection-derived preventable blindness worldwide ( CDC , 2017; Newman et al . , 2015; Taylor et al . , 2014 ) . Genital infections of chlamydia disproportionately affect women and are associated with serious sequelae in the female reproductive tract such as tubal factor infertility ( Hafner , 2015 ) . Chlamydiae are Gram-negative bacterial parasites that develop within a pathogen-specified membrane-bound organelle termed the inclusion ( Moore and Ouellette , 2014 ) . Chlamydial development is uniquely characterized by a biphasic interconversion of an infectious elementary body ( EB ) with a non-infectious , but replicative reticulate body ( RB ) ( Abdelrahman and Belland , 2005 ) . An obligate intracellular lifestyle has led to reductive genome evolution across chlamydial species; Chlamydiae have retained genes uniquely required for their survival , but have become nutritionally dependent on their hosts by discarding many metabolism-related genes ( Clarke , 2011 ) . Of note , C . trachomatis does not possess genes necessary for eliciting a stringent response to nutrient starvation ( e . g . relA , spoT ) , suggesting that this pathogen may utilize novel mechanisms to respond to nutrient stress ( Stephens et al . , 1998 ) . It is well established that in response to various stressors , Chlamydiae deviate from their normal developmental program to initiate an aberrant developmental state , termed ‘persistence’ ( Wyrick , 2010 ) . This persistent state is distinguished by the presence of viable , but non-cultivable , abnormally enlarged chlamydial organisms that display dysregulated gene expression . Importantly , Chlamydia can be reactivated from persistence by abatement of the stress condition . As such , chlamydial persistence at least superficially resembles a global stress response mechanism . Yet the mechanistic underpinnings of this phenotype are poorly understood , with most published studies focusing on the molecular and metabolic character of the aberrant , persistent form . It is therefore unclear to what extent primary stress responses contribute to the global persistent phenotype in Chlamydia . The best described inducer of persistence is the pro-inflammatory cytokine interferon-gamma ( IFN-γ ) . The bacteriostatic effect of IFN-γ has been primarily attributed to host cell tryptophan ( Trp ) catabolism , an amino acid for which C . trachomatis is auxotrophic ( Byrne et al . , 1986; Fehlner-Gardiner et al . , 2002; Taylor and Feng , 1991 ) . Following IFN-γ stimulation , infected host cells up-regulate expression of indoleamine-2 , 3-dioxygenase ( IDO1 ) , which catabolizes Trp to N-formylkynurenine via cleavage of the indole ring ( Macchiarulo et al . , 2009 ) . C . trachomatis cannot recycle kynurenines , unlike some other chlamydial species ( Wood et al . , 2004 ) , and thus IFN-γ stimulation effectively results in Trp starvation to C . trachomatis . The primary regulatory response to Trp starvation in C . trachomatis is mediated by a TrpR ortholog , whose Trp-dependent binding to cognate promoter elements represses transcription ( Akers and Tan , 2006; Carlson et al . , 2006 ) . This mechanism of regulatory control is presumably limited in C . trachomatis , as homologs of genes regulated by TrpR in other bacteria ( e . g . trpF , aroH , aroL ) have not been shown to respond to Trp limitation ( Wood et al . , 2003 ) . In many Gram-negative bacteria , such as Escherichia coli , trpR is monocistronic and distal to the Trp biosynthetic operon . In C . trachomatis , TrpR is encoded in an operon , trpRBA , which also contains the Trp synthase α- and β- subunits ( TrpA and TrpB , respectively ) , and possesses a 348 base-pair ( bp ) intergenic region ( IGR ) that separates trpR from trpBA . TrpBA catalyzes the final steps of Trp biosynthesis in bacteria; TrpA converts indoleglycerol-3-phosphate ( IGP ) to indole which is then condensed with serine by TrpB to form Trp . In C . trachomatis , TrpA cannot bind IGP and thus C . trachomatis requires indole as a substrate to synthesize Trp ( Fehlner-Gardiner et al . , 2002 ) . Despite significant research on the chlamydial trpRBA operon , the functional significance of the trpRBA IGR is poorly characterized . While a putative TrpR operator sequence was identified in the IGR overlapping an alternative transcriptional origin for trpBA ( Carlson et al . , 2006 ) , TrpR binding was not observed ( Akers and Tan , 2006 ) . Based on in silico predictions , an attenuator sequence has been annotated within the trpRBA IGR ( Merino and Yanofsky , 2005 ) , but this has not been thoroughly validated experimentally . Regardless , the IGR is >99% conserved at the nucleotide sequence level across ocular , genital and lymphogranuloma venereum ( LGV ) serovars of C . trachomatis , indicating functional importance ( Carlson et al . , 2005; Seth-Smith et al . , 2009; Stephens et al . , 1998; Thomson et al . , 2008 ) . Therefore , outside of TrpR-mediated repression , the complete detail of trpRBA regulation remains poorly elucidated and previous reports have indicated the possibility of more complex mechanisms of regulation ( Brinkworth et al . , 2018 ) . The requirement of TrpBA for C . trachomatis to survive IFN-γ-mediated Trp starvation is well documented . However , IFN-γ is also known to limit iron and other essential biometals to intracellular pathogens as a component of host nutritional immunity ( Cassat and Skaar , 2013; Hood and Skaar , 2012 ) . Whether C . trachomatis has adapted to respond to these various IFN-γ-mediated insults remains unclear . Chlamydia have a strict iron dependence for normal development , evidenced by the onset of persistence following prolonged iron limitation ( Raulston , 1997 ) . Importantly , Chlamydia presumably acquire iron via vesicular interactions between the chlamydial inclusion and slow-recycling transferrin ( Tf ) -containing endosomes ( Ouellette and Carabeo , 2010 ) . IFN-γ is known to down-regulate transferrin receptor ( TfR ) expression in both monocytes and epithelial cells with replicative consequences for resident intracellular bacteria ( Byrd and Horwitz , 1993; Byrd and Horwitz , 1989; Igietseme et al . , 1998; Nairz et al . , 2008 ) . However , iron homeostasis in Chlamydia is poorly understood due to the lack of functionally characterized homologs to iron acquisition machinery that are highly conserved in other bacteria ( Pokorzynski et al . , 2017 ) . Only the ytgABCD operon , encoding a metal permease , has been clearly linked to iron acquisition ( Miller et al . , 2009 ) . Intriguingly , the YtgC ( CTL0325 ) open reading frame ( ORF ) encodes a N-terminal permease domain fused to a C-terminal DtxR-like repressor domain , annotated YtgR ( Akers et al . , 2011; Thompson et al . , 2012 ) . YtgR is cleaved from the permease domain during infection and functions as an iron-dependent transcriptional repressor to autoregulate the expression of its own operon ( Thompson et al . , 2012 ) . YtgR represents the only identified iron-dependent transcriptional regulator in Chlamydia . Whether YtgR maintains a more diverse transcriptional regulon beyond the ytgABCD operon has not yet been addressed and remains an intriguing question in the context of immune-mediated iron limitation to C . trachomatis . Consistent with the highly reduced capacity of the chlamydial genome , it is likely that C . trachomatis has a limited ability to tailor a specific response to each individual stress . In the absence of identifiable homologs for most global stress response regulators in C . trachomatis , we hypothesized that primary stress responses to pleiotropic insults may involve mechanisms of co-regulation by stress-responsive transcription factors . Here , we report on the unique iron-dependent regulation of the trpRBA operon in Chlamydia trachomatis . We propose a model of iron-dependent transcriptional regulation of trpRBA mediated by the repressor YtgR binding specifically to the IGR , which would enable C . trachomatis to respond similarly to the antimicrobial deprivation of Trp or iron mediated by IFN-γ . Such a mechanism of iron-dependent regulation of Trp biosynthesis has not been previously described in any other prokaryote and adds to the catalog of regulatory models for Trp biosynthetic operons in bacteria . Further , it reveals a highly dynamic mode of regulatory integration within the trpRBA operon , employing bipartite control at the transcription initiation and termination steps .
To identify possible instances of regulatory integration between iron and Trp starvation in C . trachomatis , we optimized a stress response condition that preceded the development of a characteristically persistent phenotype . We reasoned that in order to effectively identify regulatory integration , we would need to investigate the bacterium under stressed , but not aberrant , growth conditions such that we could distinguish primary stress responses from abnormal growth . To specifically investigate the possible contribution of iron limitation to a broader immunological ( e . g . IFN-γ-mediated ) stress , we utilized the membrane-permeable iron chelator 2 , 2-bipyridyl ( Bpdl ) , which has the advantage of rapidly and homogeneously starving C . trachomatis of iron ( Thompson and Carabeo , 2011 ) . We chose to starve C . trachomatis serovar L2 of iron starting at 12 hr post-infection ( hpi ) , or roughly at the beginning of mid-cycle growth . At this point the chlamydial organisms represent a uniform population of replicative RBs that are fully competent , both transcriptionally and translationally , to respond to stress . We treated infected HeLa cell cultures with 100 μM Bpdl or mock for either 6 or 12 hr ( hrs ) to determine a condition sufficient to limit iron to C . trachomatis without inducing hallmark persistent phenotypes . We stained infected cells seeded on glass coverslips with convalescent human sera and analyzed chlamydial inclusion morphology under both Bpdl- and mock-treated conditions by laser point-scanning confocal microscopy ( Figure 1A ) . Following 6 hr of Bpdl treatment , chlamydial inclusions were largely indistinguishable from mock-treated inclusions , containing a homogeneous population of larger organisms , consistent with RBs in mid-cycle growth . However , by 12 hr of Bpdl treatment , the inclusions began to display signs of aberrant growth: they were perceptibly smaller , more comparable in size to 18 hpi , and contained noticeably fewer organisms , perhaps indicating a defect in bacterial replication or RB-to-EB differentiation . These observations were consistent with our subsequent analysis of genome replication by quantitative PCR ( qPCR; Figure 1B ) . At 6 hr of Bpdl treatment , there was no statistically distinguishable difference in genome copy number when compared to the equivalent mock-treated time-point . However , by 12 hr of treatment , genome copy number was significantly reduced 4 . 47-fold in the Bpdl-treated group relative to mock-treatment ( p=0 . 00151 ) . We then assayed the transcript expression of two markers for persistence by reverse transcription quantitative PCR ( RT-qPCR ) : the early gene euo , encoding a transcriptional repressor of late-cycle genes ( Figure 1C ) , and the adhesin omcB , which is expressed late in the developmental cycle ( Figure 1D ) . Characteristic persistence would display elevated euo expression late into infection and suppressed omcB expression throughout development . We observed that at 6 hr of Bpdl treatment , there was no statistically distinguishable difference in either euo or omcB expression when compared to the mock-treatment . Still at 12 hr of Bpdl treatment , euo expression was unchanged . However , omcB expression was significantly up-regulated following 12 hr of Bpdl-treatment ( p=0 . 0024 ) . This was unexpected , but we note that omcB expression has been shown to vary between chlamydial serovars and species when starved for iron ( Pokorzynski et al . , 2017 ) . The decision to begin our brief iron starvation at 12 hpi may produce notable transcriptional differences from previous studies in which iron starvation was induced at the beginning of the chlamydial developmental cycle , and thereby prevented the establishment of a normal transcriptional program by Chlamydia . Collectively , these data indicated that 6 hr of Bpdl treatment was a more suitable time-point at which to monitor iron-limited stress responses . We additionally assayed these same metrics following 6 or 12 hr of Trp starvation by culturing cells in either Trp-replete or Trp-depleted DMEM-F12 media supplemented with fetal bovine serum ( FBS ) pre-dialyzed to remove amino acids . We observed no discernable change in inclusion morphology out to 12 hr of Trp starvation ( Figure 1—figure supplement 1A ) , but genome copy numbers were significantly reduced 2 . 7-fold at this time-point ( p=0 . 00612; Figure 1—figure supplement 1B ) . The transcript expression of euo ( Figure 1—figure supplement 1C ) and omcB ( Figure 1—figure supplement 1D ) did not significantly change at either treatment duration , but Trp-depletion did result in a 1 . 88-fold reduction in omcB expression ( p=0 . 0544 ) , consistent with a more characteristic persistent phenotype . These data therefore also indicated that 6 hr of treatment would be ideal to monitor non-persistent responses to Trp limitation . We next sought to determine whether our brief 6 hr Bpdl treatment was sufficient to elicit a transcriptional iron starvation phenotype . We chose to analyze the expression of three previously identified iron-regulated transcripts , ytgA ( Figure 2A ) , ahpC ( Figure 2B ) and devB ( Figure 2C ) , by RT-qPCR under Bpdl- and mock-treated conditions ( Dill et al . , 2009; Thompson and Carabeo , 2011 ) . In addition , we analyzed the expression of one non-iron-regulated transcript , dnaB ( Figure 2D ) , as a negative control ( Brinkworth et al . , 2018 ) . Following 6 hr of Bpdl treatment , we observed that the transcript expression of the periplasmic iron-binding protein ytgA was significantly elevated 1 . 75-fold relative to the equivalent mock-treated time-point ( p=0 . 0052 ) . However , we did not observe induction of ytgA transcript expression relative to the 12 hpi time-point . Here , we considered that apparent increases in transcription could be due to two factors: developmental regulation and transcriptional response to stress . Therefore , expression of genes of interest were monitored over time , for example 18 versus 12 hpi , in addition to single-timepoint comparisons , for example 18 hpi only . While we did not observe induction of ytgA over time , which would be more consistent with an iron-starved phenotype ( i . e . ‘turning on’ gene expression ) , we reason that this is a consequence of the brief treatment period . This is in agreement with the need to prolong iron chelation to observe the transcriptional induction of ytgA ( Miller et al . , 2009; Raulston et al . , 2007; Thompson and Carabeo , 2011 ) . Similarly , we observed that the transcript expression of the thioredoxin ahpC was significantly elevated 2 . 15-fold relative to the equivalent mock-treated time-point ( p=0 . 038 ) but was not induced relative to the 12 hpi time-point . The modestly elevated expression of these genes likely represents bona fide transcriptional responses to iron starvation given that the treatment condition was optimized to avoid gross changes in chlamydial development . The transcript expression of devB , encoding a 6-phosphogluconolactonase involved in the pentose phosphate pathway , was not observed to significantly respond to our brief iron limitation condition , suggesting that it is not a component of the primary iron starvation stress response in C . trachomatis . As expected , the transcript expression of dnaB , a replicative DNA helicase , was not altered by our iron starvation condition , consistent with its presumably iron-independent regulation ( Brinkworth et al . , 2018 ) . Overall , these data confirmed that our 6 hr Bpdl treatment condition was suitable to produce a mild iron starvation phenotype at the transcriptional level , facilitating our investigation of iron-dependent regulatory integration . Upon identifying an iron limitation condition that produced a relevant transcriptional phenotype while avoiding the onset of persistent development , we aimed to investigate whether the immediate response to iron starvation in C . trachomatis would result in the consistent induction of pathways unrelated to iron utilization/acquisition , but nevertheless important for surviving immunological stress . The truncated Trp biosynthetic operon , trpRBA ( Figure 3A ) , has been repeatedly linked to the ability of genital and LGV serovars ( D-K and L1-3 , respectively ) of C . trachomatis to counter IFN-γ-mediated stress . This is due to the capacity of the chlamydial Trp synthase in these serovars to catalyze the β synthase reaction , that is the condensation of indole to the amino acid serine to form Trp ( Fehlner-Gardiner et al . , 2002 ) . In the presence of exogenous indole , C . trachomatis is therefore able to biosynthesize Trp such that it can prevent the development of IFN-γ-mediated persistence . Correspondingly , the expression of trpRBA is highly induced following IFN-γ stimulation of infected cells ( Belland et al . , 2003; Østergaard et al . , 2016 ) . These data have historically implicated Trp starvation as the primary mechanism by which persistence develops in C . trachomatis following exposure to IFN-γ . However , these studies have routinely depended on prolonged treatment conditions that monitor the terminal effect of persistent development , as opposed to the immediate molecular events which may have important roles in the developmental fate of Chlamydia . As such , these studies may have missed the contribution of other IFN-γ-stimulated insults such as iron limitation . To decouple Trp limitation from iron limitation and assess their relative contribution to regulating a critical pathway for responding to IFN-γ-mediated stress , we monitored the transcript expression of the trpRBA operon under brief Trp or iron starvation by RT-qPCR . Here again , we analyzed changes in transcript levels at the 18 hpi time-point and between the 12 hpi and 12 hpi + 6 hr time-points . This allowed us to determine if differences in expression could be accounted for by reduced , maintained or induced expression relative to 12 hpi . When starved for Trp for 6 hr , we observed that the expression of trpR , trpB and trpA were all significantly induced greater than 5 . 18-fold relative to 12 hpi ( p=0 . 0040 , 0 . 020 and 0 . 0036 , respectively; Figure 3B ) . All three ORFs were also significantly elevated relative to the equivalent mock-treated time-point ( p=0 . 0039 , 0 . 019 and 0 . 0035 , respectively ) . This result demonstrated that a relatively brief duration of Trp starvation was sufficient to induce trpRBA transcription and highlights the highly attuned sensitivity of C . trachomatis to even moderate changes in Trp levels . We then performed the same RT-qPCR analysis on the expression of the trpRBA operon in response to 6 hr of iron limitation via Bpdl treatment ( Figure 3C ) . While we observed that the transcript expression of all three ORFs was significantly elevated at least 2 . 1-fold relative to the equivalent mock-treated time-point ( p=0 . 015 , 0 . 00098 and 0 . 0062 , respectively ) , we made the intriguing observation that only the expression of trpB and trpA was significantly induced relative to 12 hpi ( p=0 . 00383 and 0 . 0195 , respectively ) . The marginal elevation in trpR expression at the 18 hpi time-point was surprising given that this gene was not identified as iron-responsive in a recent genome-wide RNA-sequencing study ( Brinkworth et al . , 2018 ) . Our results suggested that while the trpRBA operon is responsive to iron limitation , trpBA may have a more complex mode of regulation given the additional induction observed relative to trpR , which only maintained expression between 12 hpi and 12 hpi +6 hr Bpdl time-points . Taken together , these findings demonstrate that an important stress response pathway , the trpRBA operon , is regulated by the availability of both Trp and iron , consistent with the notion that the pathway may be cooperatively regulated to respond to various stress conditions . Notably , iron-dependent regulation of Trp biosynthesis has not been previously documented in other prokaryotes . We hypothesized that the specific iron-related induction of trpBA expression relative to trpR expression may be attributable to an iron-regulated alternative transcriptional start site ( alt . TSS ) downstream of the trpR ORF . Indeed , a previous study reported the presence of an alt . TSS in the trpRBA IGR , located 214 nucleotides upstream of the trpB translation start position ( Carlson et al . , 2006 ) . However , a parallel study could not identify a TrpR binding site in the trpRBA IGR ( Akers and Tan , 2006 ) . We reasoned that a similar alt . TSS may exist in the IGR that controlled the iron-dependent expression of trpBA . We therefore performed Rapid Amplification of 5’-cDNA Ends ( 5’-RACE ) on RNA isolated from C . trachomatis L2-infected HeLa cells using the SMARTer 5’/3’ RACE Kit workflow ( Takara Bio ) . Given the low expression of the trpRBA operon during normal development , we utilized two sequential gene-specific amplification steps ( nested 5’-RACE ) to identify 5’ cDNA ends in the trpRBA operon . These nested RACE conditions resulted in amplification that was specific to infected-cells ( Figure 4—figure supplement 1A ) . Using this approach , we analyzed four conditions: 12 hpi , 18 hpi , 12 hpi + 6 hr of Bpdl treatment , and 12 hpi + 6 hr of Trp-depletion ( Figure 4A ) . We observed three RACE products that migrated with an apparent size of 1 . 5 , 1 . 1 and 1 . 0 kilobases ( kb ) . At 12 and 18 hpi , all three RACE products exhibited low abundance , even following the nested PCR amplification . This observation was consistent with the expectation that the expression of the trpRBA operon is very low under normal , iron and Trp-replete conditions . However , we note that the 6 hr difference in development did appear to alter the representation of the 5’ cDNA ends , which may suggest a stage-specific promoter utilization within the trpRBA operon . In our Trp starvation condition , we observed an apparent increase in the abundance of the 1 . 5 kb RACE product , which was therefore presumed to represent the primary TSS upstream of trpR , at nucleotide position 511 , 389 ( C . trachomatis L2 434/Bu ) . Interestingly , the 1 . 0 kb product displayed a very similar apparent enrichment following Bpdl treatment , suggesting that this RACE product represented a specifically iron-regulated TSS . Both the 1 . 5 and 1 . 0 kb RACE products were detectable in the Trp-depleted and iron-depleted conditions , respectively , during the primary RACE amplification , consistent with their induction under these conditions ( Figure 4—figure supplement 1B ) . If iron depletion was inducing trpBA expression independent of trpR , we reasoned that we would observe specific enrichment of trpB transcripts in our 5’-RACE cDNA samples relative to trpR transcripts . We again utilized RT-qPCR to quantify the abundance of trpB transcripts relative to trpR transcripts in the 5’-RACE total RNA samples ( Figure 4B ) . In agreement with our model , only under iron starved conditions did we observe a significant enrichment of trpB relative to trpR ( p<0 . 01 ) . Additionally , we observed that at 12 and 18 hpi in iron-replete conditions , the ratio of trpB to trpR was approximately 1 . 0 , suggesting non-preferential basal expression across the three putative TSSs . Another factor contributing to this ratio is the synthesis of the full-length trpRBA polycistron . In support of this , the trpB to trpR ratio remained near 1 . 0 under the Trp-starved condition , which would be expected during transcription read-through of the whole operon . The apparent lack of preferential promoter utilization as described above could be attributed to the relatively low basal expression of the operon at 12 and 18 hpi under Trp- and iron-replete conditions , thus precluding quantitative detection of differential promoter utilization in this assay . To determine the specific location of the 5’ cDNA ends within the trpRBA operon , we isolated the 5’-RACE products across all conditions by gel extraction and cloned the products into the pRACE vector supplied by the manufacturer . We then sequenced the ligated inserts and BLASTed the sequences against the C . trachomatis L2 434/Bu genome to identify the location of the 5’-most nucleotides ( Figure 4C ) . These data are displayed as a statistical approximation of the genomic regions most likely to be represented by the respective 5’-RACE products in both histogram ( semi-continuous ) and density plot ( continuous ) format ( See Supplementary file 1 for a description of all mapped 5’-RACE products ) . As expected , the 1 . 5 kb product mapped in a distinct and tightly grouped peak near the previously annotated trpR TSS , with the mean and modal nucleotide being 511 , 388 and 511 , 389 , respectively ( Figure 4—figure supplement 2A ) . Surprisingly , we found that neither the 1 . 1 or 1 . 0 kb RACE product mapped to the previously reported alt . TSS in the trpRBA IGR , at position 511 , 826 . Instead , we observed that the 1 . 1 kb product mapped on average to nucleotide position 511 , 878 , with the modal nucleotide being found at 511 , 898 ( Figure 4—figure supplement 2B ) . The 1 . 0 kb product mapped with a mean nucleotide position of 512 , 013 , with the modal nucleotide being 512 , 005 ( Figure 4—figure supplement 2C ) , only 35 bases upstream of the trpB coding sequence . Interestingly , the 1 . 0 kb product mapped to a region of the trpRBA IGR flanked by consensus σ66 -10 and −35 promoter elements , found at positions 512 , 020–5 and 511 , 992–7 , respectively ( Ricci et al . , 1995 ) . In C . trachomatis , σ66 is the major housekeeping sigma factor , homologous to E . coli σ70 . In silico analyses did not reveal the presence of any promoter elements near the 1 . 1 kb product , however the mean nucleotide position is 50 bp downstream of the previously identified palindrome suspected to have a TrpR operator function ( Carlson et al . , 2006 ) . These data collectively pointed toward the 1 . 0 kb 5’-RACE product representing a novel , iron-regulated alt . TSS and bona fide σ66-dependent promoter element that allows for the specific iron-dependent expression of trpBA . As the only known iron-dependent transcriptional regulator in Chlamydia , we hypothesized that YtgR may regulate the iron-dependent expression of trpBA from the putative promoter element we characterized by 5’-RACE . Using bioinformatic sequence analysis , we investigated whether the trpRBA IGR contained a candidate YtgR operator sequence . By local sequence alignment of the putative YtgR operator sequence ( Akers et al . , 2011 ) and the trpRBA IGR , we identified a high-identity alignment ( 76 . 9% identity ) covering 67% of the putative operator sequence ( Figure 5A ) . Interestingly , this alignment mapped to the previously identified palindrome suspected to have operator functionality ( Carlson et al . , 2006 ) . By global sequence alignment of the YtgR operator to the palindromic sequence , an alignment identical to the local alignment was observed , which still displayed relatively high sequence identity ( 43 . 5% identity ) . We hypothesized that this sequence functioned as a YtgR operator , despite being located 184 bp upstream of the trpBA alt . TSS . To investigate the ability of YtgR to bind and repress transcription from the putative trpBA promoter , we implemented a heterologous in vivo two-plasmid assay that reports on YtgR repressor activity as a function of β-galactosidase expression ( Thompson et al . , 2012 ) . In brief , a candidate DNA promoter element was cloned into the pCCT expression vector between an arabinose-inducible pBAD promoter and the reporter gene lacZ . This plasmid was co-transformed into BL21 ( DE3 ) E . coli along with an IPTG-inducible pET151 expression vector with ( pET151-YtgR ) or without ( pET151-EV ) the C-terminal 139 amino acid residues of CTL0325 ( YtgC ) . Note that we have previously demonstrated that this region is a functional iron-dependent repressor domain ( Thompson et al . , 2012 ) . To verify the functionality of this assay , we determined whether ectopic YtgR expression could repress pCCT reporter gene expression in the presence of three candidate DNA elements: a no-insert empty vector ( pCCT-EV ) , the putative promoter element for C . trachomatis dnaB ( pCCT-dnaB ) , and the promoter region of the ytgABCD operon ( pCCT-ytgABCD; Figure 5B ) . As expected , from the pCCT-EV reporter construct , ectopic YtgR expression did not significantly reduce the activity of β-galactosidase . Additionally , reporter gene expression from pCCT-dnaB , containing the promoter of non-iron-regulated dnaB , was not affected by ectopic expression of YtgR . In contrast , in the presence of pCCT-ytgABCD , induction of YtgR expression produced a significant decrease in β-galactosidase activity ( p=0 . 03868 ) consistent with its previously reported auto-regulation of this promoter ( Thompson et al . , 2012 ) . Using this same assay , we then inserted into the pCCT reporter plasmid 1 ) the trpR promoter element ( pCCT-trpR ) , 2 ) the putative trpBA promoter element represented by the IGR ( pCCT-trpBA ) , and 3 ) the same putative trpBA promoter element with a mutated YtgR operator sequence that was diminished for both palindromicity and A-T richness , two typical features of prokaryotic promoter elements ( pCCT-trpBAΔOp; Figure 5C ) ( Schmitt , 2002; Tao et al . , 1992 ) . Note that the trpRBA IGR is >99% conserved at the nucleotide level across urogenital , ocular and LGV serovars of C . trachomatis , and the putative YtgR operator sequence is 100% identical ( Figure 5—figure supplement 1 ) . When YtgR was ectopically expressed in the pCCT-trpR background , we observed no statistically distinguishable change in β-galactosidase activity , indicating YtgR could not regulate transcription from the trpR promoter . However , in the pCCT-trpBA background , ectopic YtgR expression significantly reduced β-galactosidase activity at levels similar to those observed in the pCCT-ytgABCD background ( p=0 . 01219 ) . This suggested that YtgR was capable of repressing transcription from the trpBA promoter element specifically . Interestingly , this repression phenotype was abrogated in the pCCT-trpBAΔOp background , where we observed no statistically meaningful difference in β-galactosidase activity , demonstrating that YtgR transcriptional repression of trpBA is operator-dependent . We subsequently addressed whether the region of the trpRBA IGR containing the YtgR operator site was sufficient to confer YtgR repression in this assay ( Figure 5—figure supplement 2 ) . We cloned three fragments of the trpRBA IGR into the pCCT reporter plasmid: the first fragment represented the 5’-end of the IGR containing the operator site at the 3’-end ( pCCT-IGR1 ) , the second fragment represented a central region of the IGR containing the operator site at the 5’-end ( pCCT-IGR2 ) , and the third fragment represented the 3’-end of the IGR and did not contain the operator site ( pCCT-IGR3 ) . Surprisingly , we observed that none of these fragments alone were capable of producing a significant repression phenotype in our reporter system . To verify that YtgR repression of the trpBA promoter element was dependent upon iron availability , we assessed repressor activity in the presence or absence of 500 µM Bpdl , mimicking previously utilized approaches for assessing iron-dependent repressor activity of the homologous DtxR ( Ding et al . , 1996 ) . YtgR is known to bind cognate DNA elements in an iron-dependent fashion ( Thompson et al . , 2012 ) , and as such any repressor activity of YtgR should be a direct consequence of its ability to bind DNA in the presence of iron . When we treated co-transformed E . coli expressing YtgR and harboring the pCCT-trpBA reporter plasmid , we observed a modest but statistically significant increase in β-galactosidase activity ( Figure 5D; p=0 . 01409 ) , consistent with the alleviation of YtgR repression at the trpBA promoter element . To demonstrate that the repression phenotype observed in this reporter system was attributable to DNA-binding of YtgR , we optimized a targeted chromatin immunoprecipitation ( ChIP ) qPCR method to detect the abundance of co-immunoprecipitated promoter fragments with the recombinant YtgR domain . As before , we co-transformed the pET151-YtgR expression vector with pCCT-trpBA , pCCT-trpBA∆Op , pCCT-trpR or pCCT-dnaB and then fixed the co-transformed cells with formaldehyde prior to immunoprecipitation of the cross-linked 6xHis-YtgR-DNA complexes . Using this system , we observed a significant enrichment of trpBA promoter relative to the dnaB negative control promoter ( p=0 . 0018; Figure 5E ) . However , enrichment of the trpR promoter was marginal and not statistically distinguishable from that of dnaB . Consistent with its requirement for repression , enrichment of trpBA∆Op was also marginal and statistically indistinguishable from trpR or dnaB , suggesting that mutation of the putative operator sequence alone is sufficient to abrogate YtgR DNA-binding to the trpBA promoter . In conjunction with the IGR fragment analysis , these findings indicated that while the operator site was necessary for YtgR DNA-binding and transcriptional repression , further unknown structural elements in the trpRBA IGR may be required for repression . Nonetheless , this demonstrated the existence of a functional YtgR binding site that conferred transcriptional regulation to trpBA , independent of the major trpR promoter . Collectively , the most parsimonious model derived from these data is that specific and direct iron-dependent DNA-binding of YtgR at the identified operator site acts to repress expression of genes downstream of the trpBA promoter element . We hypothesized that YtgR binding at the trpRBA YtgR operator site may disadvantage the processivity of RNAP reading-through the IGR from the upstream trpR promoter , possibly leading to transcript termination . Similar systems of RNAP read-through blockage have been reported; the transcription factor Reb1p ‘roadblocks’ RNAPII transcription read-through in yeast by promoting RNAP pausing and subsequent degradation ( Colin et al . , 2014 ) . To investigate this question , we first returned to RNA-Sequencing data we generated to define the immediate iron-dependent transcriptional regulon in C . trachomatis ( Brinkworth et al . , 2018 ) . Using data obtained from C . trachomatis-infected HeLa cells at 12 hpi +6 hr mock or Bpdl treatment , we mapped the sequenced reads in batch across three biological replicates to the C . trachomatis L2 434/Bu genome ( NC_010287 ) which we modified to include annotations for non-operonic IGRs . Using this coverage map , we sought to semi-quantitatively assess the mapping of reads across the trpRBA IGR to gain insight regarding possible transcription readthrough . Our analysis revealed that under Bpdl-treated conditions , there was a 2 . 21-fold increase in reads mapping to the trpRBA IGR ( IGR_trpB ) relative to mock treatment ( Figure 6A ) . This observation is consistent with the alleviation of YtgR repression under iron-starved conditions permitting readthrough of transcription from the upstream trpR promoter . However , due to high variation and the low number of reads mapping to this region , we were unable to detect a significant difference in coverage ( p=0 . 11 ) using the genomewide RNA-seq analysis toolkit in the CLC Genomics Workbench . Regardless , a comparable increase in reads mapping to the upstream trpR CDS was not observed ( 1 . 54-fold increase , p=0 . 32 ) , suggesting that under iron replete conditions , transcripts originating from the primary trpR promoter may be terminated before reading through the IGR , thereby accounting for the increase in reads mapping to IGR_trpB following Bpdl treatment . We additionally assessed the read coverage of the IGRs upstream of euo ( IGR_euo; not iron-regulated ) and lpdA ( IGR_lpdA; iron-regulated , Brinkworth et al . , 2018 ) which are similarly configured compared to the trpRBA IGR ( i . e . the IGRs are between two ORFs in the same coding orientation ) . For both IGR_euo ( 1 . 17-fold , p=0 . 58 ) and IGR_lpdA ( 1 . 40-fold , p=0 . 37 ) , we did not observe a similar increase in read coverage following Bpdl treatment indicating that the increased coverage at IGR_trpB is likely specific ( Figure 6—figure supplement 1A–B ) . Qualitatively , we note that only IGR_trpB displays relatively uniform read coverage across the defined region , while both IGR_euo and IGR_lpdA have non-uniform coverage , reinforcing the idea that the reads mapping to IGR_trpB originate upstream in the trpR ORF and readthrough the entire region , thus presenting the opportunity for premature termination . Furthermore , the upstream ORFs for IGR_euo and IGR_lpdA did not display robust fold-increases; recJ , upstream of IGR_euo , was 1 . 29-fold increased ( p=0 . 17 ) while CTL0819 , upstream of IGR_lpdA , was 1 . 13-fold increased ( p=0 . 56 ) . We extracted individual RPKM values for each of these regions from this dataset and observed the same trend in mean fold-change differences reported from the genomewide analysis: only IGR_trpB was substantially increased relative to its upstream ORF ( Figure 6—figure supplement 2 ) . Despite being informative , genomewide RNA-Seq is ultimately insufficient to elucidate particular mechanistic details of transcriptional regulation in the trpRBA IGR and therefore we turned to more sensitive and quantitative methods to investigate possible transcript termination within the trpRBA IGR . To identify transcription termination sites ( TTSs ) in the trpRBA operon in C . trachomatis , we utilized 3’-RACE to map the 3’-ends of transcripts using gene-specific primers within the trpR CDS ( Figure 6B; lower panel ) . We again utilized two RACE amplification cycles to generate distinct , specific bands suitable for isolation and sequencing ( Figure 6—figure supplement 3B–C ) . By gel electrophoresis of the 3’-RACE products , we observed the appearance of four distinct bands that migrated with an apparent size of 0 . 55 , 0 . 45 , 0 . 40 and 0 . 20 kb . In our Trp-depleted condition , we observed only a very weak amplification of the 2 . 5–3 kb full-length trpRBA message by 3’-RACE ( Figure 6—figure supplement 3A ) . However , we did observe it across all replicates . To confirm that the full-length product was specific to the Trp-depleted treatment , we amplified the trpRBA operon by RT-PCR from the 3’-RACE total RNA ( Figure 6B; upper panel ) . As expected , only in the Trp-depleted sample did we observe robust amplification of the full-length trpRBA message . We note however that image contrast adjustment reveals a very weak band present in all experimental samples . Therefore , the specific 3’-RACE analysis identified novel transcription termination events within the trpRBA operon . To identify the specific TTS locations , we gel extracted the four distinct 3’-RACE bands across all conditions and cloned them into the pRACE sequencing vector as was done for the 5’-RACE experiments . We then sequenced the inserted RACE products and mapped them to the C . trachomatis L2 434/Bu genome ( Figure 6C ) . This revealed a highly dynamic TTS landscape within the trpRBA IGR , which has not previously been investigated ( For a full description of mapped 3’-RACE products , see Supplementary file 2 ) . The 0 . 20 kb RACE product mapped to the 3’-end of the trpR CDS , with a mean nucleotide position of 511 , 665 and a modal nucleotide position of 511 , 667 ( Figure 6—figure supplement 4A ) . Contrastingly , the other three 3’-RACE products did not map in such a way so as to produce specific , unambiguous modal peaks . Instead , their distribution was broader and more even , with only a few nucleotide positions mapping more than once . Accordingly , the 0 . 45 kb product mapped with an average nucleotide position of 511 , 889 , just downstream of the 1 . 1 kb 5’-RACE product ( Figure 6—figure supplement 4C ) , while the 0 . 55 kb product mapped with an average nucleotide position of 511 , 986 , upstream of the 1 . 0 kb 5’-RACE product ( Figure 6—figure supplement 4D ) . Interestingly , the 0 . 40 kb product mapped to a region directly overlapping the putative YtgR operator site , with a mean nucleotide position of 511 , 810 ( Figure 6—figure supplement 4B ) . We therefore reasoned that this putative TTS may have an iron-dependent function controlled by YtgR . We hypothesized that transcript termination at the YtgR operator site was regulated in an iron-dependent manner by YtgR binding to the operator DNA and blocking upstream transcription readthrough . Under iron-replete conditions , YtgR would be bound to the operator DNA and transcript termination would occur at a greater frequency , preventing readthrough of the transcription machinery initiated at the upstream trpR promoter . Under iron-depleted conditions , the inactivation of YtgR DNA-binding activity would allow the transcription machinery to readthrough the YtgR operator site to the downstream sequence , including trpBA . To test this model , we utilized RT-qPCR to quantify the amount of readthrough at the YtgR operator site in iron-depleted C . trachomatis-infected HeLa cells and in our two-plasmid E . coli system where the expression of YtgR could be controlled . We tested whether nutrient availability altered the extent of readthrough in the IGR at the YtgR operator site using a RT-qPCR-based quantification of various mRNA species ( Figure 7A ) . Levels of each intermediate transcript species indicated by unique non-overlapping amplicons were reported as ratios against a common upstream amplicon . Thus , for amplicons downstream of termination sites , a higher downstream amplicon-to-common amplicon ratio would indicate increased readthrough . Note that this value has a theoretical limit of 1 . 0 , where all transcripts would be at least as long as the downstream amplicon being measured . We designed primers to amplify regions immediately 5’ and 3’ to the YtgR operator and its corresponding TTS . Additionally , we analyzed the very 3’ end of trpA , which would be expected to only monitor complete full-length transcripts as well as alternative transcription initiation from the trpBA promoter . First , we utilized our infected cell-culture model to assess the iron-dependency of transcription readthrough at the YtgR operator site . As before , we assayed four conditions: 12 hpi , 18 hpi , 12 hpi +6 hr Bpdl and 12 hpi +6 hr Trp-depletion . Under normal developmental conditions , we expected that there would not be dramatic changes in readthrough , though we note that changes in development or nutrient availability could have this affect under normal conditions . Under Trp-depleted conditions , we predicted that the inactivation of the TrpR repressor would produce a robust readthrough phenotype that , in accordance with the literature and evidence provided herein , would transcribe the full-length trpRBA message . Thus , Trp-depletion served as a positive control for our readthrough analysis . We hypothesized that under iron-depleted conditions , we would not observe a significant increase in readthrough 5’ of the YtgR operator site , but when YtgR is inactivated , we should be able to detect an increase in readthrough 3’ of the operator site as the transcription machinery is no longer blocked by the bound repressor and is permitted to continue transcription into the downstream sequence . Consistent with our model , readthrough 5’ of the YtgR operator site was not affected by Bpdl treatment , whereas Trp-depletion resulted in significantly more readthrough than all other conditions ( Figure 7B; p<0 . 05 for all comparisons ) . However , 3’ of the YtgR operator site , Bpdl treatment significantly increased readthrough compared to 12 and 18 hpi ( p<0 . 05 for all comparisons ) , suggesting that inactivation of YtgR by iron depletion alleviates transcription termination at this site . We performed this same analysis with an alternative amplicon 3’ of the YtgR operator site and observed a similar increase in readthrough , confirming that the effect was not unique to the amplicon we used ( Figure 7—figure supplement 1; p<0 . 005 for all comparisons ) . When this readthrough analysis is applied 3’ of the alternative trpB TSS , we observe that following Bpdl-treatment , the readthrough value substantially exceeds 1 . 0 , consistent with transcription of trpBA independent of trpR under this condition . Moreover , only the Bpdl-treated group is significantly different from the other treatment conditions ( p<0 . 05 for all comparisons ) . Notably , Trp-depletion did not increase the readthrough value , suggesting that Trp-depletion does not relieve YtgR repression at the alt . TSS and the levels of trpA remain relatively constant to those of trpR . It is therefore possible that under Trp-depleted , but iron-replete conditions , YtgR repression of the alternative trpBA promoter acts as a rheostat for full-length trpRBA transcription , helping to maintain a constant ratio of trpR to trpBA . Together , these analyses indicated that iron limitation resulted in transcription readthrough specifically at the YtgR operator site . To investigate whether or not this readthrough phenotype was dependent upon YtgR , we again turned to our heterologous two-plasmid reporter system . We designed a reporter vector ( pCCT-RT ) that harbored the entire trpR-IGR DNA sequence such that any transcription initiated at the upstream arabinose-inducible pBAD promoter would have to readthrough the entire trpR ORF and the IGR before reaching the reporter gene lacZ ( Figure 7C ) . Note that this is functionally similar to expression of the trpRBA operon under Trp-starved conditions: the major Trp-dependent promoter upstream of trpR would be activated and initiate readthrough independent of the presence of YtgR bound downstream in the IGR . As such , we performed the same RT-qPCR readthrough analysis on RNA harvested from BL21 ( DE3 ) E . coli co-transformed with pCCT-RT and either pET151-EV or pET151-YtgR . We observed that ectopic expression of YtgR significantly reduced readthrough into the lacZ gene by RT-qPCR ( Figure 7D; p=0 . 003561 ) , consistent with YtgR DNA-binding specifically inhibiting readthrough via a mechanism of transcript termination at the operator site . In support of this result , we additionally observed a significant decrease in β-galactosidase activity as measured by the Miller assay from the pCCT-RT vector when YtgR was ectopically expressed ( Figure 7E; p=0 . 01723 ) , indicating that YtgR-dependent inhibition of readthrough limits the expression of downstream genes . In sum , YtgR binding to the trpRBA IGR at the predicted operator site concomitantly represses transcription from the alternative trpBA promoter while blockading transcription readthrough from the upstream trpR promoter , ultimately rendering the expression of trpBA susceptible to changes in iron availability . Importantly , alleviated transcript termination at the YtgR operator site may offer a mechanistic explanation for the moderate elevation in trpR expression observed following iron starvation , as enhanced readthrough at the YtgR operator site under this condition may produce more stable mRNA species relative to normally developing C . trachomatis . Thus , YtgR may function as an iron-dependent attenuator of trpRBA expression .
In this study , we provide a mechanistic explanation for the specific iron-limited induction of trpBA expression mediated by the repressor YtgR , representing a novel instance of integrated stress adaptation in Chlamydia . Utilizing an infected-epithelial cell culture model , we identified a previously undescribed iron-regulated promoter element independent of trpR within the trpRBA IGR that is responsible for the iron-limited induction of trpBA expression . Using in silico and biochemical methods , we demonstrate that YtgR binds the trpRBA IGR to regulate iron-dependent trpBA expression . Importantly , transcriptional repression in our heterologous system was shown to be dependent on an unaltered operator sequence that bears significant homology to the previously defined operator element in the ytgA promoter . Furthermore , our infected-cell culture studies revealed that transcripts originating from the primary trpR promoter terminate within the IGR , notably at the putative YtgR operator site , and that transcription read-through at this locus is iron- and YtgR-dependent . Thus , we propose that YtgR regulates trpBA expression at two levels: repression of the trpBA promoter and premature termination of the major transcript generated from the trpR promoter ( Figure 8; a comprehensive graphic of all T ( S/T ) Ss is provided in Figure 8—figure supplement 1 ) . To our knowledge , this is the first time an iron-dependent mode of regulation has been shown to control the expression of tryptophan biosynthesis in prokaryotes , which reflects the unique nature of C . trachomatis . While we demonstrate here that iron-dependent trpBA expression originates from a novel promoter element immediately upstream of the trpB CDS , this is not the first description of an alt . TSS within the trpRBA IGR . Carlson et al . ( 2006 ) identified an alt . TSS within the IGR which they suggested was responsible for trpBA expression . In these studies , we were unable to confirm the presence of the previously identified alt . TSS by 5’-RACE . This is likely because Carlson , et al . examined the presence of transcript origins following 24 hr of Trp starvation , whereas here we monitored immediate responses to stress following only 6 hr of treatment . Prolonged Trp depletion would result in a more homogeneously stressed population of chlamydial organisms that may exhibit the same preferential utilization of the promoter identified by Carlson , et al . , the detection of which is precluded in a more heterogeneous , transiently-stressed population . This may explain the observation of multiple T ( S/T ) Ss across the trpRBA operon in our studies . However , the contribution of such a Trp-dependent alt . TSS as identified by Carlson et al . to the general stress response of C . trachomatis remains unclear given its association with presumably abnormal organisms . Does utilization of this alt . TSS indicate abnormal growth or a bona fide stress adaptation ? Moreover , Akers and Tan were unable to verify TrpR binding to the trpRBA IGR by EMSA , suggesting that some other Trp-dependent mechanism may control transcription from this site ( Akers and Tan , 2006 ) . Ultimately , our approach of investigating more immediate responses to stress revealed previously unreported mechanisms functioning to regulate Trp biosynthesis in C . trachomatis , underscoring the value of transient as opposed to sustained induction of stress . Another mechanism of regulation reported to control the chlamydial trpRBA operon is Trp-dependent transcription attenuation . Based on sequence analysis , a leader peptide has been annotated within the trpRBA IGR ( Merino and Yanofsky , 2005 ) . Presumably , this functions analogously to the attenuator in the E . coli trpEDCBA operon; Trp starvation causes ribosome stalling at sites of enriched Trp codons such that specific RNA secondary structures form to facilitate RNAP readthrough of downstream sequences – in the case of C . trachomatis , trpBA ( Yanofsky , 1981 ) . However , robust experimental evidence to support the existence of attenuation in C . trachomatis is lacking . To date , the only experimental evidence that supports this model was reported by Carlson et al . ( 2006 ) , who demonstrated that in a TrpR-mutant genetic background , an additional increase in trpBA expression could be observed following 24 hr Trp-depletion . However , this could be attributable to an alternative Trp-dependent , but TrpR-independent mechanism controlling trpBA expression at the alt . TSS identified by Carlson , et al . None of the data presented here point conclusively to the existence of a Trp-dependent attenuator . The additional termination sites identified in our 3’-RACE assay may represent termination events mediated by a Trp-dependent attenuator , but without more specific analysis utilizing mutated sequences we cannot attribute attenuator function to those termination sites . We cannot exclude the possibility that the other TTSs observed in our 3’-RACE analysis are iron-dependent , or the product of other forms of post-transcriptional regulation such as RNA processing , stability , etc . Our focus herein was to determine the YtgR-mediated mechanism of regulation , and we have defined at least one iron-dependent , YtgR-mediated termination site . Iron-dependent termination at other TTSs in the IGR would invariably produce the same effect of limiting trpBA expression under iron-replete conditions . In Bacillus subtilis , Trp-dependent attenuation of transcription takes on a form markedly different from that in E . coli . Whereas attenuation functions in cis for the E . coli trp operon , B . subtilis utilize a multimeric Tryptophan-activated RNA-binding Attenuation Protein , TRAP , which functions in trans to bind trp operon RNA under Trp-replete conditions , promoting transcription termination and inhibiting translation ( Gollnick et al . , 2005 ) . This interaction is antagonized by anti-TRAP in the absence of charged tRNATrp , leading to increased expression of TRAP regulated genes . We suggest that YtgR may represent the first instance of a separate and distinct clade of attenuation mechanisms: iron-dependent trans-attenuation . This mechanism may function independently of specific RNA secondary structure , relying instead on steric blockage of RNAP processivity , but ultimately producing a similar result . Possible regulation of translation remains to be explored . The recent development of new genetic tools to alter chromosomal sequences and conditionally knockdown gene expression in C . trachomatis should enable a more detailed analysis of trpRBA regulation , including possible trans-attenuation ( Keb et al . , 2018; Mueller et al . , 2016; Ouellette , 2018 ) . As a Trp auxotroph , what might be the biological significance of iron-dependent YtgR regulation of the trpRBA operon in C . trachomatis ? We have already noted the possibility that iron-dependent trpBA regulation in C . trachomatis may enable the induction of a similar response to both Trp and iron starvation , stimuli likely mediated by IFN-γ in vivo . This mechanism also presents the opportunity for C . trachomatis to respond similarly to distinct sequential stresses , where a particular stress may prime the pathogen to better cope with subsequent stresses . To reach the female upper genital tract ( UGT ) , where most significant pathology is identified following infection with C . trachomatis , the pathogen must first navigate the lower genital tract ( LGT ) . Chlamydia infections of the female LGT are associated with bacterial vaginosis ( BV ) , which is characterized by obligate and facultative anaerobe colonization , some of which produce indole ( Sasaki-Imamura et al . , 2011; Ziklo et al . , 2016 ) . This provides C . trachomatis with the necessary substrate to salvage tryptophan via TrpBA . Interestingly , the LGT is also likely an iron-limited environment . Pathogen colonization and BV both increase the concentration of mucosal lactoferrin ( Lf ) , an iron-binding glycoprotein , which can starve pathogens for iron ( Spear et al . , 2011; Valenti et al . , 2018 ) . Lf expression is additionally estrogen-regulated , and thus the LGT may normally experience periods of iron limitation ( Cohen et al . , 1987; Kelver et al . , 1996 ) . Intriguingly , trpB expression has been shown to be uniquely up-regulated in estradiol-supplemented infected cell cultures , perhaps indicating the involvement of estrogen-regulated mechanisms of cell-intrinsic iron starvation ( Amirshahi et al . , 2011 ) . Moreover , the expression of TfR is constrained to the basal cells of the LGT stratified squamous epithelium ( Lloyd et al . , 1984 ) , which likely restricts necessary Tf-bound iron from C . trachomatis infecting the accessible upper layers of the stratified epithelia ( Nogueira et al . , 2017; Ouellette and Carabeo , 2010 ) . For C . trachomatis , iron limitation may therefore serve as a critical signal in the LGT , inducing the expression of trpBA such that Trp is stockpiled from available indole , allowing the pathogen to counteract impending IFN-γ-mediated Trp starvation . We suggest the possibility that iron limitation in the LGT may be a significant predictor of successful pathogen colonization in the UGT and that iron-dependent regulation of trpBA may be an important virulence trait in genital serovars of C . trachomatis . Unfortunately , testing these hypotheses in cell culture models of infection presents a significant challenge . Evaluating rescue of chlamydial growth in the presence of indole to specifically assess the iron-dependent role of trpBA requires simultaneous Trp and iron depletion . The former ensures indole utilization by the bacteria , and the latter de-represses YtgR-regulated trpBA expression . In theory , this is feasible , but in practice the combined stress rapidly induces aberrant development , muddying results obtained from such studies ( data not shown ) . Ideally , genetic approaches could be employed to distinguish the regulatory effects of YtgR independent of TrpR . However , the genetic manipulation of trans-acting factors ( e . g . YtgR ) will presumably have unpredictable off-target effects . Genetically altering cis-acting factors – such as operator sequences – is more feasible , but at present we lack the information necessary to rationally mutate these sequences in C . trachomatis to interrogate these questions . The tight regulatory coordination at both the transcription initiation and termination steps would likely mean any mutation in the cis-acting sequences would affect both processes indiscriminately . Furthermore , in vivo infection models present challenges: attempting to answer these questions will likely require the use of non-human primate studies , as mouse models of Chlamydia infection do not recapitulate immune-mediated Trp starvation ( Nelson et al . , 2005 ) . Ultimately , these limitations do not undermine the biological significance of an iron-dependent mode of regulating Trp salvage , given the critical role played by this pathway during infection . Finally , and of note , the expression of the unique class Ic ribonucleotide diphosphate reductase-encoding nrdAB was also recently shown to be iron-regulated in C . trachomatis ( Brinkworth et al . , 2018 ) . The regulation of nrdAB is known to be mediated by the presumably deoxyribonucleotide-dependent transcriptional repressor NrdR , encoded distal to the nrdAB locus ( Case et al . , 2011 ) . As NrdR activity is not known to be modulated by iron availability , this raises the intriguing possibility that here too a unique iron-dependent mechanism of regulation may integrate chlamydial stress adaptations to promote a unified response across various stresses . Future studies may require more metabolomics-based approaches to thoroughly dissect the integration of these stress responses , as transcriptome analyses alone often miss broader , pathway-oriented metabolic coordination . Ultimately , these studies point towards a need to carefully re-evaluate the molecular stress response in Chlamydia , with greater emphasis on the use of targeted approaches and treatment protocols that induce stress , but not persistence . We anticipate that the rapid progress of the field in recent years will continue to catalyze exciting and important discoveries regarding the fundamental biology of Chlamydia .
Human cervical epithelial adenocarcinoma HeLa ( ATCC CCL-2; Purchased 08/2016; Last tested for Mycoplasma 07/2018 ) cells were cultured at 37° C with 5% atmospheric CO2 in Dulbecco’s Modified Eagle Medium ( DMEM ) supplemented with 10 μg/mL gentamicin , 2 mM L-glutamine , and 10% ( v/v ) filter sterilized fetal bovine serum ( FBS ) . For all experiments , HeLa cells were cultured between passage numbers 4 and 16 . Chlamydia trachomatis serovar L2 ( 434/Bu ) was originally obtained from Dr . Ted Hackstadt ( Rocky Mountain National Laboratory , NIAID ) . Chlamydial EBs were isolated from infected HeLa cells at 36–40 hr post-infection ( hpi ) and purified by density gradient centrifugation essentially as described ( Caldwell et al . , 1981 ) . For the infection of 6-well tissue culture plates , HeLa cells cultured to 80–90% confluency were first washed with pre-warmed Hanks Buffered Saline Solution ( HBSS ) prior to the monolayer being overlaid with inoculum ( un-supplemented DMEM ) at the indicated multiplicity of infection ( MOI ) . Tissue culture plates were then centrifuged at 4° C with a speed of 1000 RPM ( Eppendorf 5810 R table top centrifuge , A-4–81 rotor ) for 5 min to synchronize the infection . Inoculum was aspirated and cells were washed again with pre-warmed HBSS prior to the media being replaced with pre-warmed complete DMEM . Infected cultures were then returned to the tissue culture incubator until the indicated times post-infection . This procedure was replicated exactly for the infection of 24-well tissue culture plates . Chlamydia trachomatis L2-infected HeLa cell cultures were starved for iron by supplementation of the media with the iron chelator 2 , 2-bipyridyl ( Bpdl; Sigma Aldrich , St . Louis , MO , USA; CAS: 366-18-7 ) essentially as described ( Thompson and Carabeo , 2011 ) . Briefly , at the indicated times post-infection , infected cell cultures were washed with pre-warmed HBSS prior to the addition of complete DMEM ( mock ) or complete DMEM supplemented with 100 μM Bpdl . Infected cell cultures were returned to the incubator for the indicated treatment periods . Bpdl was prepared as a 100 mM stock solution in 100% ethanol and stored at −20° C for no longer than 6 months . Chlamydia trachomatis L2-infected HeLa cell cultures were starved for tryptophan by replacement of complete DMEM with tryptophan-depleted medium . In brief , Tryptophan-replete or –deplete DMEM-F12 ( U . S . Biological Life Sciences , Salem , MA , USA ) powder media was prepared following manufacture instructions and supplemented with 10% ( v/v ) filter-sterilized FBS which had been previously dialyzed 16–20 hr at 4° C in PBS in a 10 kDa MWCO dialysis cassette . Media was then further supplemented with 10 μg/mL gentamicin . At the indicated times post-infection , complete DMEM was aspirated and wells were washed with pre-warmed HBSS prior to the addition of tryptophan-replete or –deplete medium . Infected cell cultures were returned to the incubator for the indicated treatment periods . All constructs were cloned using standard molecular cloning techniques , e . g . restriction enzyme , homology-directed , etc . All primers and plasmids used in this study can be found in Supplementary file 5 and 6 , respectively . All pCCT constructs were cloned by amplifying the promoter region of interest with 5’ and 3’ flanking KpnI sites , which were then KpnI digested ( New England Biolabs , Ipswich , MA , USA ) along with the pCCT-ytgA vector ( to excise the ytgA promoter fragment ) . The vector was then treated with antarctic phosphatase ( New England Biolabs ) prior to having the promoter of interest ligated into the pCCT backbone . Inserted promoters were verified to be in the correct orientation and free of sequence errors by PCR and sequencing . All pET vectors were cloned following manufacturer instructions ( Invitrogen , ThermoFisher Scientific , Waltham , MA , USA ) . At the indicated times post-infection , C . trachomatis L2-infected HeLa cell cultures seeded on glass coverslips in 24-well tissue cultures plates were first washed with pre-warmed HBSS prior to fixation with 4% paraformaldehyde ( PFA ) in phosphate buffered saline ( PBS ) for 20 min at RT ° C . Fixation solution was aspirated and wells were washed with PBS prior to permeabilization with 0 . 2% Triton X-100 in PBS for 5 min at RT° C . Permeabilization solution was then decanted and cells were washed with PBS . The coverslips were blocked for 30 min with 1% bovine serum albumin ( BSA ) in PBS at RT° C . To stain for Chlamydia , coverslips were washed with PBS and PBS supplemented with 1% BSA and 1:500 convalescent human sera was added to wells and incubated at RT° C for 1 hr with rocking . Primary antibody solution was decanted and coverslips were again washed with PBS . Goat anti-human Alexa-647 ( Invitrogen , ThermoFisher Scientific ) diluted 1:1000 in PBS with 1% BSA was then added to the wells and incubated in the dark for another hour at RT° C with rocking . Secondary antibody solution was then decanted , coverslips were washed again with PBS and coverslips were either immediately mounted on microscopy slides using ImmuMount ( ThermoFisher Scientific ) or VectaShield H-1000 ( Vector Laboratories , Burlingame , CA , USA ) or stored in the dark at 4° C until mounting . All images were acquired on a Leica TCS SP8 laser scanning confocal microscope , using identical settings , in the Integrative Physiology and Neuroscience Advanced Imaging Center at Washington State University . All images are Z-projections and were processed in Fiji ( Schindelin et al . , 2012 ) and Adobe Creative Suite identically for each comparative time-point . RNA was harvested from C . trachomatis-infected HeLa cell monolayers by scraping 3 wells of a 6-well plate in ice-cold Trizol Reagent ( ThermoFisher Scientific ) . Samples were then pooled and split into two technical replicates ( RT-qPCR ) or kept as one biological replicate ( RACE ) . Trizol-extracted samples were then thoroughly vortexed with a 100 μL volume of Zirconia beads prior to chloroform extraction . 100% ethanol was added to the aqueous phase and RNA was isolated using the Ambion RiboPure RNA Purification kit for bacteria following manufacturer instructions ( ThermoFisher Scientific ) . DNA was removed from RNA samples using the Invitrogen DNA-free DNA Removal Kit following manufacturer instructions ( ThermoFisher Scientific ) . RNA was stored at −20° C until further use . For E . coli , RNA was harvested using the Ambion RiboPure RNA Purification kit for bacteria following manufacturer instructions ( ThermoFisher Scientific ) from 9 mL of bacterial culture prepared as described below for the Two-Plasmid Reporter Assay . RNA was subsequently DNased as described above . cDNA was generated using either SuperScript IV Reverse Transcriptase ( RT-qPCR; ThermoFisher Scientific ) or SMARTScribe Reverse Transcriptase ( RACE and RACE-specific qRT-PCR ) ; Takara Bio , Kusatsu , Shiga Prefecture , Japan ) essentially as described by the respective manufacturers . For cDNA generated for RT-qPCR , 650 ng of total RNA was used as a template in a 20 μL total reaction volume . For every RT reaction , a ‘no-RT’ control , generated from 350 ng of total RNA template in a 10 μL total volume , was included . For 5’-RACE , cDNA was generated from 250 ng of total RNA using random primers in a 10 μL total volume and further processed in the RACE workflow . cDNA was stored at −20° C . gDNA was harvested from C . trachomatis-infected HeLa cell monolayers by scraping 3 wells of a 6-well plate in ice-cold PBS + 10% Proteinase K ( ThermoFisher Scientific ) . Samples were then pooled and split into two technical replicates for analysis of genome copy number by qPCR . gDNA was isolated using the DNeasy Blood and Tissue Kit following manufacturer protocols ( QIAGEN , Hilden , Germany ) . gDNA was stored at −20° C until further use . Reverse Transcription Quantitative Polymerase Chain Reaction ( RT-qPCR ) cDNA ( or gDNA in qPCR ) , prepared as described above , was diluted 1:10 or 1:100 in nuclease-free H2O depending on the experimental condition being assayed ( e . g . treatment , point in development cycle , etc . ) . On ice , 3 . 3 μL of diluted sample was added to 79 μL of PowerUp SYBR Green Master Mix ( ThermoFisher Scientific ) with specific qPCR primers diluted to 500 nM . From this master mix , each experimental sample was assayed in triplicate 25 μL reactions . Assays were run on an Applied Biosystems 7300 Real Time PCR System with cycling conditions as follows: Stage 1: 50 . 0° C for 2 min , one rep . Stage 2: 95 . 0° C for 10 min , one rep . Stage 3: 95 . 0° C for 15 s , 40 reps . Stage 4: 60 . 0° C for 1 min , one rep . Primers were subjected to dissociation curve analysis to ensure that a single product was generated . For each primer set , a standard curve was generated using purified C . trachomatis L2 gDNA from EB preparations diluted from 2 × 10−3 to 2 × 100 ng per reaction . Ct values generated from each experimental reaction were then fit to standard curves ( satisfying an efficiency of 95 ± 5% ) for the respective primer pair and from the calculated ng quantities , transcript or genome copy number was calculated as follows:Genome copy number ( genome copiesng total gDNA ) =ng genome × dfng total gDNA × 892 , 000 copiesng DNATranscript copy number transcript copiesgenome copiesng total gDNA= ng transcript × dfgenome copiesng total gDNA × 892 , 000 copiesng DNA Where df = dilution factor and the number of copies/ng DNA is calculated based on the size of the C . trachomatis L2 genome assuming that the molar mass per base pair is 650 ( g/mol ) /bp ( note that this value should be the same for any single-copy ORF on the genome ) . All quantifications of genome copy number were determined using the ahpC qPCR primer set . Values from replicate assays were averaged , and values from replicate RNA/gDNA isolations were averaged to obtain the mean and standard deviation for one biological replicate . For some experiments , to account for batch effects across biological replicates , data was transformed such that the mean of all samples in each replicate was identical . In some instances , batch correcting generated negative values , and in this case data sets were scaled such that the lowest value equaled 1 . 0 . HeLa cells were infected at an MOI of 2 for all RT-qPCR studies . For analysis of transcriptional readthrough , RT-qPCR was performed as described above and the readthrough value was computed as:Readthrough= 2 ( CtNormalization- CtExperimental ) Readthrough values were then batch corrected such that the mean of each replicate was identical . All RACE studies were performed using the SMARTer RACE 5’/3’ Kit ( Takara Bio ) . To observe 5’-RACE products from the trpRBA operon , a ‘nested’ RACE protocol was used as outlined in the SMARTer RACE 5’/3’ Kit user manual . Briefly , 1 . 25–2 . 5 μL of cDNA generated for RACE was added to a 25 μL reaction volume and run in a thermal cycler for 40 cycles using the touch-down PCR conditions described by the manufacturer . In brief , five cycles were run at an annealing temperature of both 72° C and 70° C prior to 30 cycles run with an annealing temperature of 68° C . Following this primary amplification , the RACE products were diluted 1:50 in Tricine-EDTA Buffer supplied by the manufacturer , and 2 . 5 μL of diluted primary RACE product was added to a 25 μL reaction volume and subjected to another 20 cycles of nested PCR , as described by the manufacturer , using primers designed within the amplicon of the primary RACE products . Samples were electrophoresed on a 2% agarose gel for visualization and analysis . HeLa cells were infected at a MOI of 5 for all RACE studies . 3’-RACE studies were performed essentially identical to 5’-RACE with the exception that total RNA was subjected to poly ( A ) tailing with a Poly ( A ) Polymerase following manufacturer instructions ( New England Biolabs ) . In brief , at least 3 . 5 μg of total RNA was incubated at 37° C with Poly ( A ) Polymerase in reaction buffer supplemented with ATP and murine RNase Inhibitor ( New England Biolabs ) for 30 min prior to heat-inactivation at 65° C for 20 min . RNA was re-isolated through an RNA clean-up filter cartridge ( Ambion , ThermoFisher Scientific ) . A total of 125 ng of poly ( A ) -tailed total RNA was then used to generate 3’-RACE ready cDNA in a 10 μL reaction volume following manufacturer instructions . Primary and nested RACE was performed using 3’-RACE gene-specific primers following the same protocol for amplification described for 5’-RACE , with the exception that the extension time was adjusted to accommodate amplification of the full ~3 kb trpRBA polycistronic message . 5’-RACE products generated from either primary or nested RACE reactions were excised from the agarose gel and DNA was isolated using the NucleoSpin Gel and PCR Clean-up kit ( Macherey-Nagel , Takara Bio ) . The isolated RACE products were then cloned into the pRACE vector supplied in the SMARTer RACE 5’/3’ Kit using the In-Fusion HD cloning kit ( Takara Bio ) . Ligated vectors were transformed into chemically competent Stellar E . coli cells by heat shock . Transformed bacteria were plated on LB agar containing 50 μg/mL carbenicillin and incubated overnight at 37° C . Colonies were selected and screened for relevant inserts by PCR . Positive colonies were cultured overnight at 37° C in LB liquid broth containing 50 μg/mL carbenicillin and plasmids were isolated using the QIAprep Spin Miniprep kit ( QIAGEN ) . Inserts were then sequenced by Eurofins Genomics using the default M13 Reverse sequencing primer . Returned sequencing data was aligned to the C . trachomatis L2 ( 434/Bu ) genome ( NCBI Accession: NC_010287 ) by BLAST and the most 5’ aligned nucleotide was considered the 5’ end of the insert . In the case of 3’-RACE data , the reverse complement sequence was first generated prior to alignment . Grouping of individual products was determined 1 . ) by clusters being greater than 30 nucleotides apart and 2 . ) by the specific RACE band that the alignment was derived from . These two criteria were not both satisfied in all cases and in those cases criteria 1 . ) was favored . All C . trachomatis L2 434/Bu genome sequences were obtained from NCBI Accession NC_010287 . Global pairwise sequence alignments were made using the EMBOSS Needle algorithm . Alignment parameters were set as follows: Matrix: DNAfull , Gap Open: 20 , Gap Extend: 0 . 8 , Output Format: pair , End Gap Penalty: True , End Gap Open: 10 , End Gap Extend: 0 . 5 . These conditions were sufficient to replicate the previously published alignment between the putative YtgR operator sequence and the TroR operator ( Akers et al . , 2011 ) . Local pairwise sequence alignments were made using the EMBOSS Water algorithm . The putative YtgR operator was aligned to the entire 348 bp intergenic region of the trpRBA operon ( C . trachomatis L2 [434/Bu] genome position 511 , 692–512 , 039 ) . The alignment parameters were set as follows: Matrix: DNAfull , Gap Open:10 , Gap Extend: 0 . 5 , Output Format: pair . These are the default conditions and were chosen to remove bias from the alignment results . The YtgR-binding reporter assay was performed essentially as described , with minor modifications ( Thompson et al . , 2012 ) . Promoter regions of interest were amplified from the C . trachomatis L2 ( 434/Bu ) genome by PCR using the indicated primer sets , which included KpnI restriction endonuclease sites at the 5’ and 3’ ends of the promoter amplicon . The amplified fragments and the pCCT-EV plasmid were then KpnI-digested and the promoters ligated into the vector using T4 or Quick Ligase ( New England BioLabs ) . Insert directionality was confirmed by directional colony PCR and positive clones were sequence verified . pCCT-trpBAΔOperator was cloned by amplifying two fragments of the pCCT-trpBA vector with one ~ 60 mer primer containing the bases to be substituted for each fragment . Thus , the whole vector was split into two half-fragments containing the substituted bases . The two fragments were then cloned back together using In-Fusion Homology-Directed cloning ( Takara Bio ) to yield the final vector . Electrocompetent BL21 ( DE3 ) E . coli ( Sigma Aldrich ) were co-transformed by electroporation with the pCCT reporter plasmid and the pET151 expression vector ( -EV or –YtgR ) and plated on double selective LB agar containing 50 μg/mL carbenicillin and 15 μg/mL tetracycline . Prior to plating of transformed cells , 50 μL of 40 mg/mL X-Gal in DMSO ( EMD Millipore , Burlington , MA , USA ) was applied to the plate for colorimetric determination of β-galactosidase expression . Transformants were incubated overnight at 37° C . The following evening , blue colonies from each experimental condition were selected and cultured overnight in LB liquid broth containing 0 . 2% ( w/v ) D-glucose ( for catabolite repression of expression vectors ) , 50 μg/mL carbenicillin and 15 μg/mL tetracycline . Cultures were incubated overnight at 37° C . The following morning , overnight cultures were spun down to remove glucose-containing media and sub-cultured in LB liquid broth medium containing 50 μM FeSO4 , 50 μg/mL carbenicillin and 15 μg/mL tetracycline to an OD600 of 0 . 45 . Cultures were incubated for 1 hr at 37° C and sub-cultured a second time in the same media to an OD600 of 0 . 1 . Cultures were returned to the incubator for another hour prior to the addition of 500 μM isopropyl β-D-1-thiogalactopyranoside ( IPTG ) to induce pET151 expression from the lac promoter . Cultures were incubated another hour prior to the addition of 0 . 2% L-arabinose to induce lacZ expression from the araBAD promoter . Cultures were incubated a final 2 hr prior to the collection of a 0 . 1 mL volume of cells for assaying β-galactosidase activity by the Miller Assay ( Miller , 1972 ) . Cell pellets were stored at −80° C prior to being assayed . To assay β-galactosidase activity , cell pellets were first re-suspended in Z-buffer ( pH 7 . 0 , 60 mM Na2HPO4 , 40 mM NaH2PO4 , 10 mM KCl , 1 mM MgSO4 and 2 . 7 μL/mL β-mercaptoethanol ) . 50 μL of 0 . 1% SDS and 100 μL of chloroform were then added to each sample prior to thorough vortexing . Samples were equilibrated for 5 min at 30° C and 200 μL of 4 mg/mL ortho-nitrophenyl-β-galactoside ( ONPG ) prepared in Phosphate Buffer ( pH 7 . 0 , 60 mM Na2HPO4 , 40 mM NaH2PO4 ) were added to the samples to initiate the reaction . Reactions were stopped by the addition of 500 μL 1 M Na2CO3 . Absorbance was measured on a FLUOStar Optima plate reader ( BMG Labtech , Offenburg , Germany ) at 420 nm and Miller Units were calculated as:1000 × Abs420t × v × ODfinal Where t = reaction time , v = volume of cells and ODfinal = OD600 at the time of sample collection . It was empirically determined that the subtraction of absorbance at 550 nm had a negligible effect on the calculated value . A blank sample lacking cells was included in each experimental batch and used as a reference for absorbance . For each experimental condition , three independent co-transformed colonies were assayed in technical triplicate . In some instances , significantly high Miller Unit outliers were excluded by Grubb’s Test ( p<0 . 05 ) under the assumption that extreme lacZ expression may reflect plasmid copy number or reporter gene expression issues . RNA-Sequencing experiments were performed as described in their original publication ( Brinkworth et al . , 2018 ) . Raw and processed sequencing files were submitted to the NCBI Gene Expression Omnibus ( GEO ) as a Superseries and can be found at accession number GSE106763 . Coverage maps were generated by mapping all reads across three biological replicates to a single reference file in CLC Genomics Workbench v11 . To facilitate easy analysis of IGR boundaries , the C . trachomatis L2 434/Bu genome ( Accession: NC_010287 ) was modified to contain annotations for intergenic regions that fell between two genes in the same coding orientation , and this genome was used as the reference for read mapping . Read mapping and differential expression analysis was performed using default settings in CLC Genomics Workbench . Data aggregation in the Reads track was set to aggregate above 1 bp . All graphs were generated using the ggplot2 package ( Wickham , 2009 ) in R Studio , and/or in the Adobe Creative Suite . All line plots and bar graphs represent the mean ± one standard deviation unless otherwise noted . All box and whisker plots represent the distribution of data between the 1st and 3rd quartile range within the box , while the whiskers represent data within 1 . 5 interquartile ranges of the 1st or 3rd quartile . Extreme values outside this range are plotted as open circles . The 2nd quartile ( median ) is plotted as a black line within the box . Histogram plots were generated with a bin width of 20 and are plotted on a density scale . The overlaid density plots represent a statistical approximation of the data over a continuous scale . All statistical analyses were carried out in R Studio . All statistical computations were performed on the mean values of independent biological replicates calculated from the indicated number of respective technical replicates . For single pairwise comparisons , a two-sided unpaired Student’s t-test with Welch’s correction for unequal variance was used to determine statistical significance . For multiple pairwise comparisons , a One-Way Analysis of Variance ( ANOVA ) was conducted to identify significant differences within groups . If a significant difference was detected , then the indicated post-hoc pairwise test was used to identify the location of specific statistical differences . A p-value less than 0 . 05 was considered statistically significant . For all figures , *=p < 0 . 05 , **=p < 0 . 01 , and ***=p < 0 . 005 . | All forms of life must take up nutrients from their environment to survive . Chlamydia trachomatis , a bacterium that causes many sexually-transmitted infections , is no exception . These bacteria do not normally make one of the building blocks of proteins , the amino acid tryptophan , but instead scavenge it from their human host . One way that the immune system tries to fight a chlamydia infection is by cutting off the supply of tryptophan in an attempt to starve the bacteria . But the microbes have evolved to respond to these hardships and keep themselves alive . The ‘tryptophan salvage pathway’ is a set of genes that , when switched on , allows the Chlamydia bacteria to take up a molecule found in the female genital tract that they can use to make their own tryptophan . Yet , how do the bacteria know when to activate these genes ? Tryptophan starvation is not the only strategy that the immune system uses to fight chlamydia . It also restricts the supply of the essential metal iron to these bacteria . Now , using human cells grown in the laboratory and infected with Chlamydia bacteria , Pokorzynski et al . show that iron starvation switches on the tryptophan salvage pathway . Chlamydia most likely senses changes in iron levels via a protein called YtgR , and a closer look at the bacterial DNA revealed that YtgR interacts with the genes of the tryptophan salvage pathway . When iron levels were high , YtgR locked on to the DNA in the middle of this set of genes . This effectively switched off the genes on either side of the binding site . When iron levels dropped , YtgR came away from the DNA , releasing the genes and allowing the cell to use them to start making its own tryptophan . Together these findings indicate that , when the bacteria sense that iron levels have dropped , they prepare for a shortage of tryptophan too . Chlamydia is the most common bacterial sexually transmitted infection worldwide . Left untreated , it can cause infertility and blindness . This and future studies aimed at understanding how these bacteria respond to immune attack may reveal new ways to prevent or treat these infections . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | A bipartite iron-dependent transcriptional regulation of the tryptophan salvage pathway in Chlamydia trachomatis |
Hearing loss often triggers an inescapable buzz ( tinnitus ) and causes everyday sounds to become intolerably loud ( hyperacusis ) , but exactly where and how this occurs in the brain is unknown . To identify the neural substrate for these debilitating disorders , we induced both tinnitus and hyperacusis with an ototoxic drug ( salicylate ) and used behavioral , electrophysiological , and functional magnetic resonance imaging ( fMRI ) techniques to identify the tinnitus–hyperacusis network . Salicylate depressed the neural output of the cochlea , but vigorously amplified sound-evoked neural responses in the amygdala , medial geniculate , and auditory cortex . Resting-state fMRI revealed hyperactivity in an auditory network composed of inferior colliculus , medial geniculate , and auditory cortex with side branches to cerebellum , amygdala , and reticular formation . Functional connectivity revealed enhanced coupling within the auditory network and segments of the auditory network and cerebellum , reticular formation , amygdala , and hippocampus . A testable model accounting for distress , arousal , and gating of tinnitus and hyperacusis is proposed .
A third of adults over the age of 65 suffer from significant hearing loss , a condition exacerbated by two debilitating condition , subjective tinnitus , a phantom ringing or buzzing sensation , and hyperacusis , normal sounds perceived as intolerably loud or even painful . Roughly 12% of adults experience tinnitus , but the prevalence skyrockets to 50% in young combat personnel ( Leske , 1981; Andersson et al . , 2002; Cave et al . , 2007; Michikawa et al . , 2010; Hebert et al . , 2013 ) . Tinnitus is costly with more than $2 billion paid annually in veteran disability payments . Hyperacusis affects roughly 9% of adults ( Andersson et al . , 2002 ) , but its prevalence is likely higher because of the difficulty of self-diagnosis ( Gu et al . , 2010 ) . Remarkably , among those whose primary complaint is hyperacusis , 90% also suffer from tinnitus ( Baguley , 2003 ) . Since tinnitus and hyperacusis are often triggered by cochlear hearing loss , it was long assumed that these auditory distortions resulted from hyperactivity disorders in the peripheral auditory nerve . This hypothesis , however , is contradicted by studies showing that auditory nerve spontaneous and sound-evoked firing rates are depressed in subjects with cochlear damage ( Kiang et al . , 1970; Wang et al . , 1997 ) . Moreover , surgical section of the auditory nerve fails to eliminate tinnitus ( Baguley et al . , 1992; Lockwood et al . , 2001 ) . These negative results plus recent imaging studies now suggest that tinnitus and hyperacusis arise from maladaptive neuroplastic change in the central nervous system ( CNS ) provoked by cochlear pathology ( Lockwood et al . , 1998; Husain et al . , 2011; Sereda et al . , 2011 ) . Several models of tinnitus and hyperacusis have been proposed that involve increased central gain , altered functional connectivity ( FC ) , and aberrant neural oscillations in neural networks ( Weisz et al . , 2007; Sereda et al . , 2011; Henry et al . , 2014 ) . Most of these conceptual models have emerged from human imaging studies using magnetoencephalography , electroencephalography , magnetic resonance imaging ( MRI ) , and functional MRI ( fMRI ) of the blood oxygen level-dependent ( BOLD ) response ( Llinas et al . , 1999; Weisz et al . , 2005; Auer , 2008; Gu et al . , 2010; Moazami-Goudarzi et al . , 2010; Leaver et al . , 2012; Maudoux et al . , 2012; Husain and Schmidt , 2014 ) . In the context of central gain models , some human imaging data indicate that hyperacusis is associated with enhanced sound-evoked activity in multiple-auditory processing centers , namely auditory cortex ( ACx ) , medial geniculate body ( MGB ) , and inferior colliculus ( IC ) , whereas tinnitus can be triggered solely by enhanced central gain in the ACx ( Gu et al . , 2010 ) . On the other hand , active loudness models suggest that tinnitus arises entirely from increased central noise independent of gain , whereas hyperacusis results exclusively from increased nonlinear gain that results in loudness intolerance ( Zeng , 2013 ) . While cross-sectional human brain imaging studies have identified many different sites of aberrant neural activity , published results from patients have often produced diverse , inconsistent , or contradictory findings . Some discrepancies are likely due to confounding factors such as patient heterogeneity , unknown etiology , genetic diversity , social and environmental factors , and duration or severity of tinnitus and hyperacusis . Animal models could potentially overcome many of these limitations provided that tinnitus and hyperacusis can be reliably induced , behaviorally measured , and functionally imaged . While tinnitus can develop in some individuals after intense noise exposure , the percentage of affected individuals is highly variable and its duration is unpredictable ( Heffner and Harrington , 2002; Lobarinas et al . , 2006; Heffner , 2011 ) . High doses of aspirin , an anti-inflammatory drug used to treat rheumatoid arthritis , have long been known to consistently induce acute tinnitus in humans and animals ( Myers and Bernstein , 1965; Myers et al . , 1965; Mongan et al . , 1973 ) . Moreover , high-dose sodium salicylate ( SS ) , the active ingredient in aspirin , not only consistently induces tinnitus ( Jastreboff et al . , 1988; Lobarinas et al . , 2004; Stolzberg et al . , 2013 ) , but also hyperacusis ( Chen et al . , 2014; Hayes et al . , 2014 ) ; these perceptual disorders disappear a day or two post-treatment . The highly predictable time course of SS-induced tinnitus and hyperacusis makes it an extremely powerful tool for studying the neural correlates of these perceptual disturbances . Therefore , we took advantage of our unique behavioral techniques for assessing SS-induced tinnitus and hyperacusis in rats and combined this with focused electrophysiological measurements plus global fMRI assessment techniques to map out the regions of neural hyperactivity and enhanced FC that characterize the tinnitus–hyperacusis network . To identify regions of heightened or depressed spontaneous neural activity , we measured the amplitude of low-frequency fluctuations ( ALFF ) in resting-state fMRI ( Zang et al . , 2007; Zhang et al . , 2010; Yao et al . , 2012; Wen et al . , 2013 ) and combined this with resting-state FC to identify regions of increased or decreased functional coupling between regions of the auditory pathway and other parts of the CNS . This is the first animal study to use ALFF and FC combined with detailed electrophysiological measures to provide a comprehensive neurological map of the tinnitus–hyperacusis network .
SS induced a peripheral threshold shift of approximately 20 dB for the CAP ( Figure 2A ) ( Chen et al . , 2013 , 2014 ) . The same amount of threshold shift occurred at higher levels of the auditory pathway indicating that the SS-induced hearing loss originates in the cochlea and is relayed centrally . SS also reduced the CAP neural output by ∼70% at suprathreshold intensities . Paradoxically , suprathreshold LFP amplitudes in the MGB , ACx , and LA were larger than normal despite the massive reduction in the output of the cochlea ( Figure 2B–D ) . These provocative findings provide compelling evidence for an increase in central gain , a form of homeostatic plasticity implicated in tinnitus and hyperacusis ( Salvi et al . , 1990; Auerbach et al . , 2014 ) . The enhanced LFPs seen in ACx are consistent with the enhanced fMRI response observed in the ACx of tinnitus patients , whereas the enhanced LFPs seen in both ACx and MGB are consistent with the enhanced fMRI responses observed these regions in hyperacusis patients ( Gu et al . , 2010 ) . These results are consistent with previous models and data linking tinnitus and loudness intolerance to increased central gain in the central auditory pathway in particular regions from the IC to ACx ( Salvi et al . , 1990; Qiu et al . , 2000; Auerbach et al . , 2014 ) . In some models , enhanced central gain amplifies central neural noise resulting in tinnitus ( Noreña , 2010 ) . However , in other models , central neural noise increases independent of central gain ( Zeng , 2013 ) ; this could potentially explain why some patients only experience tinnitus , but not hyperacusis ( Baguley , 2003 ) . However , this distinction is clouded by the fact that many tinnitus patients are unaware of their mild hyperacusis , that is , hyperacusis may be more prevalent in tinnitus patients than currently believed because many patients are unaware of their hyperacusis ( Gu et al . , 2010 ) . Many cellular mechanisms could enhance central gain , but one likely candidate is reduced inhibition ( disinhibition ) . Considerable evidence exists for dysregulated inhibition in central gain-control models ( Auerbach et al . , 2014 ) . First , SS can suppress GABA-mediated inhibition and enhance excitability ( Xu et al . , 2005; Gong et al . , 2008 ) . Second , SS enhances sound evoked activity in ACx when given systemically or applied locally to the LA or ACx , whereas it depresses ACx responses when only applied to the cochlea ( Sun et al . , 2009; Chen et al . , 2012 ) . Third , drugs that enhance GABA-mediated inhibition , prevent SS-induced gain enhancement ( Sun et al . , 2009; Lu et al . , 2011 ) , and suppress tinnitus ( Brozoski et al . , 2007b ) . Behaviorally , hyperacusis was initially observed at 50 dB SPL; the same low intensity at which sound-evoked neural hyperactivity occurred in the ACx . In contrast , sound-evoked hyperactivity occurred at noticeably higher intensities for the LA ( ∼60 dB SPL ) , MGB ( ∼70 dB SPL ) , and acoustic-startle reflex amplitude ( ∼95 dB SPL ) . These results suggest that neural responses from the ACx may be one of the most sensitive biomarkers of hyperacusis ( Juckel et al . , 2004; Gu et al . , 2010 ) . However , since neural responses increased in magnitude from cochlea to cortex , loudness intolerance issues likely result from multiple stages of neural amplification as signals are relayed rostrally from the cochlea to the ACx ( Auerbach et al . , 2014 ) . Indeed , there is growing evidence that the neural amplification gradually develops in the auditory brainstem and serially accumulates to supernormal levels after reaching the MGB and ACx consistent with previous electrophysiological results ( Qiu et al . , 2000; Schaette and McAlpine , 2011 ) . Some models of tinnitus are based on changes in spontaneous spiking patterns such as increased firing rate or increased neural synchrony ( Eggermont , 2015 ) . SS either decreased or had no effect on spontaneous spike rate in primary ACx ( Ochi and Eggermont , 1996; Yang et al . , 2007 ) and reportedly no effect on synchrony between neuron pairs ( Eggermont , 2015 ) . Since the BOLD and LFP responses mainly represent presynaptic activity , it is difficult to directly relate our results to these spiking models . However , the increase in very low-frequency BOLD oscillations ( 0 . 01 Hz ) represented by ALFF could be interpreted as evidence for increased presynaptic synchrony , which would likely enhance spike synchrony albeit at much longer time intervals than previously studied or over much larger neuronal populations than that reflected by spike correlations between neuron pairs . SS has also been found to increase gamma-band ( 50–100 Hz ) oscillatory activity in ACx ( Stolzberg et al . , 2013 ) ; oscillations substantially higher than in ALFF . An alternative view is that the tinnitus percept is derived from coordinated activity among several auditory and nonauditory regions ( Horwitz and Braun , 2004; Husain et al . , 2006 ) . Enhanced FC between the HIP and auditory areas provides a substrate for assigning a spatial location to a phantom sound , while coordinated activity between specific auditory areas and the reticular formation and AMY may draw attention to and add emotional significance to neural activity in the auditory pathway . Thus , functionally coordinated activity within the network may be essential for bringing tinnitus into consciousness . Tinnitus and hyperacusis , like phantom limb pain and cutaneous allodynia , are triggered by peripheral damage presumably leading to widespread changes in the CNS that involve altered connections in networks that include portions of the central auditory pathway and other regions linked to emotion , memory , attention , and arousal ( Llinas et al . , 1999; Leaver et al . , 2012; Husain and Schmidt , 2014 ) . In the resting state , SS increased ALFF and FC in a broad-neural network that included core auditory structures extending from the IC to the ACx consistent with previous studies implicating these central auditory structures in tinnitus ( Paul et al . , 2009 ) . SS also enhanced sound-evoked LFP in the MGB and ACx suggesting a key role for these auditory structures in amplifying auditory information that could manifest as loudness intolerance ( Gu et al . , 2010 ) . Although the cerebellum is mainly involved in motor planning and control , some cerebellar regions such as the PFL and vermis receive inputs from auditory centers ( Petacchi et al . , 2005 ) and respond to sound ( Lockwood et al . , 1999 ) . Interestingly , the perception of tinnitus has been linked to activation of the PFL and vermis ( Brozoski et al . , 2007a ) consistent with our results . Since ablation or inactivation of the PFL eliminates the perception of noise-induced tinnitus ( Bauer et al . , 2013 ) , some have suggested that the PFL acts as a gain control mechanism comparing the afferent input from the cochlea with descending signals from the ACx ( Bauer et al . , 2013 ) . Consistent with this view , our results show that SS leads to hyperactivity in the ACx and increases the FC between the ACx and PFL and CB4 . If this cerebellar-tinnitus gating hypothesis is correct , then ablating or inactivating the PFL should suppress behavioral measures of SS-induced tinnitus and possibly hyperacusis , providing a clear test of this model . The functional role of the PFL in tinnitus–hyperacusis network could be further elucidated by inactivating the PFL and determining the effects this has on SS-induced changes we observed in our electrophysiological and fMRI measures . The AMY , which assigns emotions such as fear or anxiety to sensory events , lies outside the classical auditory pathway; however , it is linked to several auditory areas and responds robustly to sound ( Romanski and LeDoux , 1993; Stutzmann et al . , 1998; Chen et al . , 2014 ) . In the resting state , SS enhanced the ALFF in the AMY and increased FC between ACx and AMY consistent with prior results showing increased coupling between ACx and AMY in tinnitus patients ( Kim et al . , 2012 ) and increased c-fos immunolabeling in the AMY following SS treatment ( Wallhäusser-Franke et al . , 2003 ) . SS also enhanced sound-evoked activity in the AMY consistent with the increased activation seen in the AMY of hyperacusis patients ( Levitin et al . , 2003 ) . Importantly , infusion of SS directly into AMY increases sound-evoked activity in the ACx , effects that illustrate the potent independent role that the AMY can exert on central auditory function and aural perception ( Chen et al . , 2012 ) . Collectively , these results reinforce the view that the AMY contributes to the fear and anxiety experienced by many patients with tinnitus and hyperacusis ( van Veen et al . , 1998; Juris et al . , 2013; Aazh et al . , 2014 ) . Sound and cognitive therapies aimed at reducing the emotional distress of tinnitus and hyperacusis would be expected to reduce the level of activity in the AMY and/or the functional coupling between the AMY and ACx without necessarily eliminating aberrant auditory percepts ( Hazell and Jastreboff , 1990 ) . Human imaging studies employing ALFF and FC could be used to test this hypothesis and provide an objective and independent assessment of how these therapies work and their effects on the tinnitus–hyperacusis neural network . A novel finding observed during resting-state fMRI was the SS-induced enhancement of ALFF in the reticular formation ( RN , PnO , and PMr ) together with increased FC between the reticular formation and ACx . The reticular formation is an important arousal center with numerous inputs from the cochlear nucleus and IC ( Kandler and Herbert , 1991 ) . Giant neurons in pontine reticular formation control the amplitude of the acoustic startle reflex ( Koch et al . , 1992 ) , and stimulation of the AMY enhances the response of these giant neurons ( Koch et al . , 1992 ) . Thus , the SS-induced increases of ALFF observed in the AMY and reticular formation likely contribute to the enhancement of the acoustic startle reflex . SS unexpectedly increased ALFF in SSCx , VCx , and SC raising the question of whether this might be linked to phantom visual or somatosensory perceptions . However , after an extensive search , we were unable to find evidence of such aberrant somatosensory or visual phenomena . While the heightened ALFF response in these areas is novel , such changes seem reasonable , given the multisensory interactions known to exist between auditory , somatosensory , and visual areas . One possibility is that heightened ALFF activity in ACx , MGB , and IC could spill over and enhance activity in visual and somatosensory areas ( Murray et al . , 2005 ) . However , these increases may not lead to altered perception because FC was not enhanced in visual or somatosensory areas . Alternatively , the SS-induced cochlear loss could unmask pre-existing multisensory circuits in visual and somatosensory areas leading to increased activation ( Barone et al . , 2013 ) . Although SS has long been known to cause tinnitus ( Mongan et al . , 1973; Jastreboff et al . , 1988 ) , it is now clear that it also induces strong hyperacusis-like behavior ( Chen et al . , 2014; Hayes et al . , 2014 ) . Although tinnitus and hyperacusis could conceivably arise from different mechanisms ( Zeng , 2013 ) , they frequently co-occur more frequently than previously believed ( Gu et al . , 2010 ) . Tinnitus and hyperacusis do not exist in isolation but are linked to other brain regions associated with emotions , arousal , memories , spatial location , and motor activity as schematized in the tinnitus–hyperacusis network model defined by our imaging results ( Figure 5 ) . With the seed region in IC , a significant increase in FC occurred in the MGB; this increase is likely due to the SS-induced enhancement of ALFF in the IC , which is relayed rostrally to the MGB ( Figure 5 , thick black line ) resulting in a larger and more coherent MGB response . Similarly , with the seed in the MGB , increased FC occurred in the ACx; this increase is likely due to the increased ALFF and FC occurring in MGB , which is relayed rostrally to the ACx ( Figure 5 , thick black line ) . With the seed in the ACx , increased FC was seen in the same two lower auditory centers , the MGB and IC , raising the possibility of a recurrent feedback loop in this auditory subnetwork ( Figure 5 , shaded area , bidirectional dashed red lines ) . These data combined with our electrophysiological results suggest that SS enhances the FC and response magnitude in a central auditory subnetwork that extends from the IC through the MGB to the ACx . 10 . 7554/eLife . 06576 . 010Figure 5 . Tinnitus–hyperacusis network model . Schematic showing some of the major centers in the auditory pathway and areas in the CNS showing increased FC with the auditory cortex ( ACx ) . Black lines show the neural transmission path from the cochlea through the cochlear nucleus ( CN ) , inferior colliculus ( IC ) , medial geniculate body ( MGB ) , and ACx; black double-line reflects SS-induced increases in ALFF and/or increased FC . SS increased ALFF response magnitudes and FC in a central auditory subnetwork ( gray shaded area ) comprised of the IC , MGB , and AC . The AC serves as a major hub with side branches to the amygdala ( AMY ) , reticular nuclei ( RN ) , and parafloccular lobe ( PFL ) and cerebellar lobules 4 ( CB4 ) ; these side branch regions contribute to the emotional , motoric , and conscious awareness of tinnitus and/or hyperacusis . Enhanced activity in the MGB and IC combined with the increased FC of these auditory structures with the hippocampus ( HIP ) could facilitate tinnitus memory storage or retrieval or assign spatial location to phantom or real sounds . DOI: http://dx . doi . org/10 . 7554/eLife . 06576 . 010 The FC data suggest that the ACx is a major hub in the tinnitus–hyperacusis network with side branches that extend caudally to the AMY , RN , and cerebellum ( PFL , CB4 ) ; these subdivisions all show large SS-induced increases in ALFF , as well as increased FC with the ACx . The side branch connecting the ACx to the AMY provides a pathway through which emotional significance can be attached to tinnitus or hyperacusis ( Chen et al . , 2014 ) consistent with earlier studies linking anxiety , annoyance , and fear to tinnitus and hyperacusis ( Moller , 2007 ) . The ACx-RN network provides a conduit by which increased arousal can increase awareness or attention , enhance motivation , or amplify motor responses to tinnitus or suprathreshold sounds ( Carlson and Willott , 1998; Paus , 2000 ) . The ACx-cerebellar branch could serve as a gating path for tinnitus ( Boyen et al . , 2014 ) or modulate the motor responses to or perceptual salience of suprathreshold sounds thereby contributing to hyperacusis ( Mobbs et al . , 2007 ) . Increased FC between HIP-MGB and HIP-IC could facilitate the formation or stabilization of a memory trace for tinnitus , assign a spatial location to the phantom sound within or outside the head , or promote the retrieval of previously stored auditory memories ( De Ridder et al . , 2006; Ulanovsky and Moss , 2008 ) . Our network model can be explicitly tested by administering a high-dose of SS while activating or inactivating part of the network , such as the PFL , and determining if the manipulation abolishes tinnitus or hyperacusis or alternatively determining how such manipulations affects the electrophysiological and fMRI properties of the network . Since high-dose SS is one of the most predictable and reliable inducers of tinnitus and hyperacusis , SS provides researchers with a powerful tool to gain mechanistic insights into the neurophysiological conditions needed to induce these debilitating auditory perceptions . Since SS-induced tinnitus and hyperacusis are transient phenomena that begin shortly after drug treatment , some of the neuro-pathophysiological changes we observed are likely to be similar to those that occur during the early stages of tinnitus or hyperacusis induced by acoustic trauma , Meniere's disease , or sudden hearing loss . In cases of acoustic overstimulation , most individuals develop transient tinnitus immediately post-exposure , which gradually disappears; interestingly only a small percentage develops permanent tinnitus ( Gilles et al . , 2012; Degeest et al . , 2014 ) . Thus , an important question that remains to be answered is whether the neural correlates of chronic tinnitus and hyperacusis are similar or different from the immediate and acute condition . SS-induced tinnitus and hyperacusis appear to arise from enhanced central gain and increased FC in an auditory network with four side branches . The core of the network is composed of auditory structures extending from the IC , through the MGB to the ACx plus branches to the AMY , RN , cerebellum , and HIP . These side branches presumably contribute to the emotional significance , arousal , motor response , gating , and memories associated with tinnitus and hyperacusis . | One in three adults over the age of 65 will experience a significant loss of hearing . This is often worsened by related conditions , such as: tinnitus , an unexplained constant buzzing or ringing sound; and hyperacusis , whereby everyday sounds are perceived as too loud or painful . Most hearing loss is caused by damage to the sound-sensitive cells within a structure in the inner ear called the cochlea . Some studies have also identified regions of the brain that show abnormal activity in people with tinnitus and hyperacusis . However , the results from different patients have often been inconsistent and sometimes contradictory , and so it remains unclear what exactly causes these conditions . To overcome this problem , Chen et al . made use of the fact that tinnitus and hyperacusis are common short-term side effects of certain drugs and measured the brain activity in rats before and after they were given one such drug . Before receiving the drug , the rats had first been trained to expect to receive a food pellet from the left side of their cage when they heard a steady buzzing sound . The rats were also trained to expect a food pellet from their right if they heard nothing at all . Shortly after receiving the drug , the rats often failed to respond correctly in the ‘quiet tests’ and behaved like they were already experiencing a constant buzzing sound , as would be expected if they had tinnitus . Further tests confirmed that the drug also triggered behavior in the rats that is typical of people with hyperacusis . Chen et al . then discovered that the drug treatment reduced the nerve signals that are sent from a rat's cochlea . Moreover , the drug treatment greatly increased the activity in response to sound within parts of the rat's brain; these and other parts of the brain also became overactive in drug-treated rats in the absence of sound . Finally , further experiments revealed that drug-treated rats had stronger connections between these brain regions than in normal rats . Chen et al . used these results to propose a model to explain the underlying causes of tinnitus and hyperacusis . However , because the drug treatment only induces short-term hearing impairment , further studies are needed to see if this model also applies when these conditions are long-term . | [
"Abstract",
"Introduction",
"Discussion"
] | [
"neuroscience"
] | 2015 | Tinnitus and hyperacusis involve hyperactivity and enhanced connectivity in auditory-limbic-arousal-cerebellar network |
CLC channels mediate passive Cl− conduction , while CLC transporters mediate active Cl− transport coupled to H+ transport in the opposite direction . The distinction between CLC-0/1/2 channels and CLC transporters seems undetectable by amino acid sequence . To understand why they are different functionally we determined the structure of the human CLC-1 channel . Its ‘glutamate gate’ residue , known to mediate proton transfer in CLC transporters , adopts a location in the structure that appears to preclude it from its transport function . Furthermore , smaller side chains produce a wider pore near the intracellular surface , potentially reducing a kinetic barrier for Cl− conduction . When the corresponding residues are mutated in a transporter , it is converted to a channel . Finally , Cl− at key sites in the pore appear to interact with reduced affinity compared to transporters . Thus , subtle differences in glutamate gate conformation , internal pore diameter and Cl− affinity distinguish CLC channels and transporters .
Transporters – also known as pumps – and channels both mediate the transfer of ions and molecules across biological membranes . But the two are thermodynamically contrasting: transporters require the input of external energy while channels are passive , meaning the substrate simply diffuses down its electrochemical gradient . Except in rare cases , transporters and channels correspond to separate , unrelated structural families . CLC proteins are one of the exceptions . Channel-forming CLCs are passive Cl− conductors ( Jentsch et al . , 1990; Miller and White , 1984 ) , while transporter-forming CLCs exchange , with fixed stoichiometry , two Cl− ions and one proton ( H+ ) in opposite directions ( i . e . , they are Cl−/H+ antiporters ) ( Accardi and Miller , 2004; Picollo and Pusch , 2005; Scheel et al . , 2005 ) . The external energy input in CLC transporters comes from the energetic coupling of the transported ions , Cl− and H+ , such that the electrochemical gradient of one ion drives movement of the other . The puzzling aspect of this dual functionality within the CLC protein family is that at the level of amino acid sequence , the distinction between the channels and transporters is not apparent . Conceptually , the distinction between channels and transporters in general has been explained in terms of gating models that invoke one or two primary gates: channels are described as pores with one gate and transporters as pores with two gates that are never permitted to open simultaneously ( Figure 1 ) ( for review see [Gadsby , 2009] ) . While it is true that channels and transporters are most often unrelated structurally , the gating model description implies that , in principle , similar structures could give rise to both , as one can imagine that a transporter could become a channel if one or both gates are compromised . CLC channels seem to fall under this category of channels that emerged from a family of transporters ( Accardi and Picollo , 2010; Lísal and Maduke , 2008; Miller , 2006 ) . Structural and functional studies support a plausible mechanistic model for the operation of CLC transporters . CLC transporter structures show a narrow Cl− transport pathway with three consecutive Cl−-binding sites , referred to as Sext , Scen and Sint , for external ( nearest the extracellular solution ) , central and internal ( nearest the intracellular solution ) , respectively . Chloride is observed at these sites in various structures ( Dutzler et al . , 2002; Dutzler et al . , 2003; Feng et al . , 2010; Jayaram et al . , 2011 ) . In addition , the transporters all contain a glutamate residue positioned such that its side chain carboxylate group can bind either at Sext or Scen – in competition with a Cl− ion – or reside in the extracellular solution . Thus , CLC transporters are like Cl− channels with a weird feature – a glutamate side chain that clogs its own pore . This led to the idea that glutamate might not only be a competitor for the Cl− binding sites as the structures suggest , but it might also transfer a proton from inside to out ( or the reverse ) when it moves between its Scen position to its extracellular position ( Feng et al . , 2010; Feng et al . , 2012 ) . The transfer would naturally give rise to the 2:1 Cl−:H+ exchange stoichiometry characteristic of CLC transporters because 2 Cl− ions must be displaced when the glutamate gate moves between the extracellular solution and Scen . This mechanism is consistent with the demonstrated conversion of a CLC transporter into a passive ( but slow ) Cl− channel upon mutation of the glutamate , as well as the demonstrated ability of small carboxylate-containing organic acids to compete with Cl− inside the pore ( Accardi et al . , 2004; Accardi and Miller , 2004; Feng et al . , 2012 ) . But there was one important caveat to make this transporter mechanism work: there must exist a relatively high kinetic barrier to Cl− flow near the intracellular side of the pore ( Feng et al . , 2010 ) . This barrier would serve as the ‘second gate’ in the gating model conceptualization of transporters . So far , data for CLC transporters seem consistent with this mechanism: they have a channel-like pore , an external ‘glutamate gate’ that competes with Cl− binding and ( presumably ) transfers H+ across the membrane , and structurally what appears to be a relatively high resistance ( i . e . , a large kinetic barrier ) to Cl− flow near the intracellular aspect of the pore ( i . e . , the pore there is very narrow . ) Less is known about the chemistry and structure of CLC channels . Only one CLC channel structure has been determined , CLC-K from Bos taurus ( referred to as bCLC-K or shortly CLC-K ) ( Park et al . , 2017 ) . This is a special case , a rare type of CLC channel that can be distinguished from CLC transporters based on its amino acid sequence because it does not have a ‘glutamate gate’ . That difference alone renders CLC-K inert to H+ transfer . The structure of CLC-K also revealed a wider pore diameter on the intracellular side , consistent with a lowered kinetic barrier to Cl− flow . CLC-0/1/2 channels , by contrast , contain a ‘glutamate gate’ and are not distinguishable from CLC transporters by sequence . Thus , there must be an even more subtle distinction between these CLC channels and the transporters . Why does the glutamate gate in these channel CLCs not give rise to H+ transfer coupled to Cl− transfer ? Is a reduced kinetic barrier to Cl− flow near the intracellular side , suggested by the CLC-K structure , a common feature in CLC channels ? To address these questions , we have determined the structure of CLC-1 from Homo sapiens ( referred to as hCLC-1 or CLC-1 ) . We are also interested in the CLC-1 channel because it plays an important role in membrane repolarization of skeletal muscle cells following muscular contraction , and its mutation in humans causes hereditary muscle disorders known as myotonia congenita ( George et al . , 1993; Koch et al . , 1992; Lorenz et al . , 1994; Steinmeyer et al . , 1991 ) .
We purified the CLC-1 protein in mild detergent from cultured human cells and examined them by cryo-EM single particle analysis ( Figure 2 and Figure 2—figure supplements 1 and 2 ) . Despite its small molecular size ( 200 kDa ) , particles showed good contrast on micrographs under the optimized freezing and data acquisition conditions ( Figure 2A ) . Two-dimensional ( 2D ) class averages of selected particles displayed 2-fold rotational symmetry around an axis normal to the membrane ( detergent micelle ) ( Figure 2B ) , as expected from the homodimeric architecture of CLC proteins ( Dutzler et al . , 2002; Ludewig et al . , 1996; Miller and White , 1984 ) . After removing artifacts and damaged particles by 2D classification , a density map was reconstructed at 3 . 9 Å resolution with C2 symmetry imposed ( Figure 2—figure supplement 1B ) . This map showed a well-resolved transmembrane domain ( TMD ) with clearly visible α-helical features . By contrast , density for the carboxy-terminal cytosolic domain ( CTD ) was lower quality , suggesting conformational flexibility in this region . To improve the map quality , we subjected particles to a round of 3D classification ( Figure 2—figure supplement 1B ) . The results demonstrated that while the TMD is largely indistinguishable between classes , the CTDs deviate from each other by pivotal movements of varying degrees ( Figure 2—figure supplement 1C ) . Based on this , we pooled ~170 , 000 particles from the two most populated and structurally similar classes , which correspond to 50% of particles . This particle set led to an improved density map at an overall resolution of 3 . 6 Å ( data not shown ) . Using masking techniques to isolate individual regions , the resolution of the TMD was further improved to 3 . 4 Å ( Figure 2C , Figure 2—figure supplement 1B , Video 1 ) . The CTD remained poorly defined , likely due to continuous pivotal movements of its two wing-like structures ( Figure 2C , Figure 2—figure supplement 1C , Video 1 ) . The good quality TMD density map enabled building a molecular model that included nearly all side chains ( Figure 2C , Figure 2—figure supplement 2 , and Video 1 ) . The model was refined using Rosetta ( Wang et al . , 2016 ) . The CTD map did not show side chain density but we could dock with confidence the crystal structure of the CLC-0 CTD ( Figure 2D and Figure 2—figure supplement 1D ) ( Meyer and Dutzler , 2006 ) . Both CLC-1 and CLC-0 channels contain a large loop extending from the CTD’s cystathionin-β-synthase ( CBS ) domains , which was not visible in either the EM density map or the crystal structure . The function of the CTD is poorly understood; it may even be dispensable for ion transport given its high tolerance to mutation ( Estévez et al . , 2004 ) and absence in most bacterial CLC transporters . The TMD of CLC-1 exhibits the canonical dimeric architecture of a CLC protein ( Figure 2C , D ) . Each monomer is roughly a triangular prism shape and contains a complete ion transport pathway that appears structurally independent from that of the neighboring monomer . As in other CLC structures ( Dutzler et al . , 2002; Dutzler et al . , 2003; Feng et al . , 2010; Park et al . , 2017 ) , the Cl− transport pore in CLC-1 is most narrowly constricted halfway across the membrane , within the region referred to as the selectivity filter ( Figure 3A ) . Overall , the pore lining is charged positive to attract Cl− ( Figure 3B ) . In contrast to other CLC proteins the potential route for ion diffusion in CLC-1 is bifurcated on the intracellular side of the selectivity filter– one following the canonical Cl− transport pathway found in all CLC proteins and the ‘second’ pore directed toward the protomer-protomer boundary on the cytosolic surface , which is distinctive in CLC-1 ( Figure 3A , B and Figure 3—figure supplement 1 ) . Both branches of the bifurcation are potentially hydrated because the radius is greater than that of water ( 1 . 4 Å ) and the linings contain chemical groups with hydrogen bonding potential . A branch equivalent to CLC-1’s secondary pore in the CLC-K channel is sealed off by F222 and V226 ( corresponding to F288 and V292 of CLC-1 ) due to a different αH helix position ( Figure 3C ) . In transporters , only a much narrower ( ~0 . 9–1 . 0 Å radius ) pore could be detected , where stable dwelling of water molecules seems unlikely ( Figure 3—figure supplement 1 ) . In the E . coli transporter ( EcCLC ) , the pore is further capped near the cytosolic surface by E203 ( corresponding to V292 of CLC-1 ) . We note that E203 of EcCLC and the equivalent Glu of mammalian CLC-4 and CLC-5 transporters have been implicated in shuttling H+ between the intracellular solvent and the protein interior ( Lim and Miller , 2009; Lim et al . , 2012; Zdebik et al . , 2008 ) by side-chain protonation and deprotonation , although this feature does not seem to be essential for H+ transport in other cases , including the C . merolae transporter ( CmCLC ) ( Feng et al . , 2010; Feng et al . , 2012; Phillips et al . , 2012 ) . It is possible that during Cl−/H+ exchange cycles , the αH helix of transporters transiently undergoes a conformational change such that a water-accessible pore is formed similarly to the CLC-1 case , which might facilitate H+ transfer . Unlike transporter-type CLCs , the CLC-1 channel does not transport H+ in a manner tightly coupled to Cl− and thus it is unclear whether CLC-1’s second intracellular pore is utilized for ion transport . Cl− ions may move through this pore in addition to the primary Cl− pathway . The CLC-1 structure shows an anion selectivity filter largely similar to other CLC proteins but with some distinctive features ( Figure 4A and Video 2 ) . The filter is formed at the central constriction of the Cl− pathway by αN , αF , and αD helices , all of which point their N-terminal ends towards the center where Cl−-binding sites are formed . This arrangement contributes to an electrostatically positive environment at the Cl−-binding sites through α-helix end charges . Backbone nitrogen atoms from αN and αF segments are arranged to coordinate a partially dehydrated Cl− ion near the extracellular end of the constriction ( external site or Sext ) . In the CLC-1 density map we observe a clear density feature at Sext , which likely corresponds to a bound Cl− ion ( Figure 4A and Video 2 ) . Typically , CLC proteins have two additional Cl−-binding sites , namely , central ( Scen ) and internal ( Sint ) sites ( Dutzler et al . , 2003 ) . Scen has been observed to bind a Cl− ion through polar interactions with one or two backbone nitrogen atoms and the side chains of the conserved tyrosine ( denoted TyrC; Y578 of CLC-1 or Y445 of EcCLC ) and serine residues ( denoted SerC; S189 of CLC-1 or S107 of EcCLC ) ( Dutzler et al . , 2002; Dutzler et al . , 2003 ) . In the EcCLC transporter , Scen has been shown to bind Cl− relatively strongly ( Kd ~1 mM ) ( Lobet and Dutzler , 2006; Picollo et al . , 2009 ) . Sint is largely exposed to the intracellular solvent and binds Cl− with lower affinity ( Kd >20 mM ) ( Lobet and Dutzler , 2006; Picollo et al . , 2009 ) . In the CLC-1 map ( determined in the presence of 116 mM Cl− ) , Sint shows a density peak whose intensity is comparable to that of the Sext density ( Figure 4A and Video 2 ) . By contrast , we do not observe density for an ion at Scen above the noise level , suggesting that Scen of CLC-1 may have a lower Cl− occupancy than Sext and Sint . This is somewhat surprising given the conservation of structural elements for Scen , including TyrC and SerC . Perhaps subtle structural differences account for the absence of an ion at this site compared to other CLC proteins . For example , we note that the position of TyrC is shifted away from Scen by ~1 . 5 Å ( see Figure 5B ) . Like transporter-type CLC proteins and in contrast to CLC-K , the CLC-1 channel has a Glugate , but in CLC-1 it adopts a notably different conformation than previously observed in CLC transporters ( Figure 4 and Figure 4—figure supplement 1 ) . Based on previous studies on transporters ( Dutzler et al . , 2002; Dutzler et al . , 2003; Feng et al . , 2010 ) , Glugate , located in the immediate vicinity of Sext and Scen , plays a key role in ion transport: when deprotonated its side-chain carboxylic moiety resides in either the Sext or Scen Cl− binding sites , preventing the binding of a Cl− ion therein . In the CLC-1 structure , the Glugate side chain occupies neither Sext nor Scen , but instead it is oriented in a different direction . The difference in Glugate’s conformation is mainly due to changes in its side-chain rotamer , whereas the polypeptide backbone arrangement in this region is similar among the structures ( Figure 4C ) . The observed Glugate conformation is also different than the outwardly-oriented ( side chain projecting into the extracellular funnel ) conformation that has been seen in the structure of an EcCLC Glu-to-Gln ( E148Q ) mutant ( Figure 4C ) , which is hypothesized to mimic the protonated state of Glugate ( Dutzler et al . , 2003 ) . It is unclear whether the Glugate in the CLC-1 structure ( determined at pH 7 . 4 ) is protonated . The pKa of the Glugate side chain might be shifted towards a more neutral pH as it is neighbored by multiple hydrophobic amino acids ( Isom et al . , 2010 ) . Yet , Glugate at this position is more likely deprotonated because its side chain seems exposed to water molecules due to the presence of the second intracellular pore ( Figure 4B ) . In CLC transporters , this conformation would be highly unfavorable because it would produce steric clashes with neighboring side chains ( equivalent to V236 , V265 , and F279 of CLC-1; Figure 4—figure supplement 1 ) , which are moved away in CLC-1 by a shift of the αG and αH helices . In other words , this conformation of Glugate does not seem possible in CLC transporters studied so far . The observed Glugate conformation of CLC-1 was unexpected because it was never observed in other CLC protein structures , and yet it is consistent with an open CLC-1 channel , which is expected in the absence of an applied membrane potential . CLC-1 is a voltage-gated channel , which closes when the membrane potential is negative ( i . e . , at its ‘resting’ value ) ( Fahlke et al . , 1996; Pusch et al . , 1995 ) . Perhaps in the presence of an applied negative membrane potential the Glugate side chain moves into either the Sext or Scen position , as seen in CLC transporters , and prevents Cl− conduction . This possibility would account for the observation that CLC-1 and related CLC-0 conduct Cl− ions at all membrane voltages when the Glugate residue is mutated to Gln ( Dutzler et al . , 2003; Fahlke et al . , 1997 ) . The previous CLC-K channel structure has suggested that a wider pore diameter between Scen and Sint is crucial for its channel function ( Park et al . , 2017 ) . In CLC transporters , a kinetic barrier for Cl− passage ( i . e . , a narrowing of the pore ) exists on the intracellular side of the vestibule to preclude slippage of Cl− ions during the Cl−/H+ exchange cycle ( Feng et al . , 2010 ) . This barrier is due to a narrow pore width between Scen and Sint , which is created in part by SerC of the αC-D loop interposed between the two Cl− binding sites . In the CLC-K structure , the αC-D loop has a distinctly different conformation , where SerC is flipped down and thus no longer interposed between the two Cl− binding sites . Consequently , the pore diameter is wider such that Cl− ions will more readily permeate . Given that CLC-1 is also a channel , we wondered whether the αC-D loop in CLC-1 would adopt a similar ‘flipped-down’ conformation . While a different conformation of the αC-D loop is a key feature distinguishing CLC-K from transporters , a structural comparison shows that this is not the case for the CLC-1 channel ( Figure 5 ) . In contrast to CLC-K , the αC-D loop in CLC-1 adopts the loop conformation seen in CLC transporters , especially CmCLC ( Feng et al . , 2010 ) . Consequently , the SerC side chain is positioned between Sint and Scen ( Figure 5B ) . Therefore , in the case of CLC-1 the αC-D loop itself does not provide an explanation for why CLC-1 functions as a Cl− channel ( see below ) . This also suggests that the ‘flipped-down’ conformation of SerC may be unique to the CLC-K channel . To understand why CLC-1 functions as a channel we compared its Cl− pore structure to that of other CLC proteins . In both CLC-1 and CLC-K channels , a continuous Cl− pathway was evident in between the extracellular and intracellular funnels , through the selectivity filter ( Figure 6A , B ) . In the EcCLC and CmCLC transporters , a continuous pore could be detected only when the Glugate side-chain atoms ( from Cβ ) were excluded from the pore radius calculation as Glugate sits at Sext or Scen ( Figure 6C , D ) . These results would therefore reflect the pore structure when the transporter’s Glugate transiently moves away from the Cl− pathway upon protonation ( hypothetically , akin to the crystal structure of the EcCLC E148Q mutant ) . However , we note that calculated pore radii around Sext may be somewhat overestimated due to the actual presence of the Glugate side-chain atoms . The CLC-1 channel has the narrowest ( 1 . 0 Å in radius ) constriction above Sext toward the extracellular side due to the placement of the M485 side chain near the external end of the Cl− pathway ( Figure 6A ) . While the radius is significantly smaller than the Cl− radius ( ~1 . 7 Å ) , the flexibility of the M485 side chain must allow Cl− ions to pass through this region . Given the narrowness of this constriction , it is likely that M485 affects the Cl− throughput of the channel . In fact , its mutation to less flexible valine ( M485V ) causes recessive myotonia congenita and has been shown to reduce the single channel conductance of CLC-1 to about 20% of the wild type channel conductance ( Wollnik et al . , 1997 ) . From Sext to the intracellular opening the CLC-1 channel structure shows a relatively wide pore opening despite its ‘transporter-like’ αC-D loop . This suggests the absence of a large kinetic barrier in CLC-1 , but for reasons other than the αC-D loop conformation . Compared to the CLC-K channel , CLC-1 has a slightly narrower ( 1 . 5 Å vs 1 . 7 Å in radius ) opening between Scen and Sint because of SerC . This might create a kinetic barrier to some degree , but the pore is still significantly wider and more hydrophilic than the equivalent region in the EcCLC transporter ( Figure 6C ) . The difference originates mainly from two amino acids ( T475 and G483 ) lining the constriction . In EcCLC , the equivalent positions are F348 and I356 , which project their bulky , hydrophobic side chains towards the Cl− pathway between Scen and Sint . Together with proximal placement of SerC and TyrC , this narrows the opening ( 1 . 0 Å in radius ) in EcCLC . In the CmCLC transporter , the constriction at the kinetic barrier region is wider ( 1 . 6 Å in radius ) than EcCLC because of smaller side chains at the equivalent positions ( I421 and V429; Figure 6D ) and a slight downward shift ( 1 . 5 Å ) of SerC with respect to the positions in EcCLC ( Figure 5B ) . Yet , hydrophobicity provided by the I421 and V429 side chains might result in a significantly higher kinetic barrier than in CLC-1 . It is noteworthy that the CLC-1 channel shows a 1 . 5 Å outward shift of the TyrC side chain with respect to the position that is almost invariant in the other CLC structures ( Figure 5B ) . In CLC-1 , this shift contributes to pore widening in the cytosolic vestibule . At present it is unclear if this shift of TyrC is static or part of dynamic movements in CLC-1 and if it is unique in CLC-1 or a similar movement exists in other CLC proteins . Previous biophysical studies have proposed a movement of TyrC to explain alternating gate opening of the EcCLC Cl−/H+ transporter ( Basilio et al . , 2014; Jayaram et al . , 2008; Khantwal et al . , 2016 ) . On the other hand , EcCLC crystal structures obtained with a number of different variants and crystallization conditions have not yet revealed any movement of TyrC . Because the CLC-1 structure suggests that T475 and G483 ( equivalent to F348 and I356 in EcCLC , respectively ) likely contribute to lowering of the kinetic barrier , we compared amino acids lining this region among both CLC channels and transporters ( Figure 6—figure supplement 1 ) . Indeed , these two positions showed a distinctive differential pattern when comparing CLC channels and transporters , whereas other positions ( i . e . , H369 , C481 , L577 , and I581 in CLC-1 ) did not . Generally , these two positions are filled with large , hydrophobic amino acids in transporters but are replaced by a small , polar amino acid in CLC channels . One notable outlier is position 417 of the CLC-K channels ( Y425 ) . However , the CLC-K channel structure shows that its phenyl side chain is skewed off the Cl− pathway , and thus does not seem to create a kinetic barrier in CLC-K ( Figure 6B ) . In fact , it forms the Scen Cl− binding site together with TyrC and F519 through anion-quadrupole interactions ( Park et al . , 2017 ) ( Figure 6B ) . In summary , the observed amino acid pattern and structural information suggest that a lowered barrier in the Scen–Sint region of the pore is a common feature of CLC channels , but CLC-1 and CLC-K channels achieve this somewhat differently . In the CLC-1 channel , small side chains in pore-lining residues lower the kinetic barrier , whereas in CLC-K mainly the reorientation of SerC lowers it . The extent of the kinetic barrier should also be affected by the hydrophobic and electrostatic nature of the lining residues , not only the physical dimensions of the pore . Combining the new structural information and previous data , we propose a working model that channel behavior in CLC proteins arises out of the following physical conditions ( Figure 7 ) : ( 1 ) Glugate is either absent ( i . e . , in CLC-K ) or allowed to reside in an ‘open’ configuration ( i . e . , CLC-1 ) for a sufficiently extended period of time ( rather than occupying Sext or Scen ) ; ( 2 ) a lowered kinetic barrier between Scen and Sint; ( 3 ) reduced Cl−-binding affinity at Scen ( or Sext , as suggested by apparent low occupancy in the CLC-K structure ) . A reduced kinetic barrier would be an important feature to achieve fast Cl− throughput . On the other hand , a sufficient kinetic barrier would be crucial in transporters to preclude undesired slippage of Cl− ions through the transiently open pore ( Feng et al . , 2010 ) . In addition , reduction of Cl−-binding affinity at Scen and/or Sext , which is energetically related to the kinetic barrier , might also contribute to high Cl− throughput in channels . For example , relatively deep energy wells at Scen and Sext , as implied by the high occupancy of sites in the EcCLC transporter , would create a larger energy difference between the binding sites and the ‘transition states’ , which effectively raises the energy barrier . In CLC-1 the relatively low binding site occupancy implies not very deep energy wells and thus a smaller energy difference between the binding sites and ‘transition states’ . We carried out biophysical experiments to test some of these ideas using the EcCLC transporter ( Figure 8A , D ) . EcCLC mutants were produced , purified and reconstituted into lipid vesicles for assessment of Cl− and H+ transport activity ( Figure 8 and Figure 8—figure supplement 1 ) ( Feng et al . , 2012; Jayaram et al . , 2008; Walden et al . , 2007 ) . The ideas outlined above predict that if the kinetic barrier in EcCLC is lowered it should behave more like a CLC channel ( i . e . , rapid Cl− permeation with decreased H+ transport activity ) . Cl− permeation is expected to be further increased if the Glugate is rendered persistently opened . As reported previously ( Jayaram et al . , 2008 ) , opening of Glugate alone by the Glu-to-Ala mutation ( E148A ) abolishes the H+ transport activity , but it also reduced Cl− throughput by a factor of approximately 0 . 25 . We reason that this is likely because the mutant still retains the kinetic barrier deterring Cl− ions from moving between Scen and Sint . Thus , while removal of the Glugate is sufficient to convert the transporter into a Cl− channel , a reduced kinetic barrier would be key to an increased Cl− throughput , an important feature of the native CLC channels . Previous studies have shown that when the E148A mutation is combined with a TyrC mutation ( e . g . , Y445S ) , the Cl− transport rate dramatically increases ( Jayaram et al . , 2008 ) , demonstrating that efficient Cl− channel activity can be produced from EcCLC by altering its gates . However , we note that Y445S is rather unphysiological as TyrC is invariant among all CLC channels and transporters . Therefore , here we examined the effects of lowering the kinetic barrier in wild type EcCLC by mutating SerC or neighboring pore-lining amino acids , guided by the CLC-1 and CLC-K structures ( Figure 8B , E and Figure 8—figure supplement 1 ) . Trimming the side chain of SerC ( S107G ) , with the intention of mimicking the flipped SerC in the CLC-K channel , increased the Cl− transport rate by a factor of 2 , as previously reported ( Jayaram et al . , 2008 ) . At the same time , this mutation lowered H+ coupling 3-fold , as one would expect due to the slippage of uncoupled Cl− ions . Next , since CLC-K has a polar amino acid ( Thr ) at one of its pore-lining residues ( F348 of EcCLC ) , we further introduced a similar ( F348A ) mutation . This increased the Cl− throughput and almost abolished coupled H+ transport . Finally , by adding a Glugate mutation ( E148A ) to this double mutant the Cl− throughput was further increased . Compared to the E148A single mutant , the triple mutant ( S107G/F348A/E148A ) has a Cl− transport rate increased about 25-fold ( Figure 8—figure supplement 1 ) . Similar results were obtained when mutations mimicking the CLC-1 channel were introduced to EcCLC ( Figure 8C , F and Figure 8—figure supplement 1 ) . While single mutations at the pore-lining amino acids ( F348T or I356G ) did not increase the Cl− transport rate , the double mutation ( F348T/I356G ) moderately increased the Cl− throughput ( 1 . 5-fold with respect to the wildtype ) . We note that this mutant displayed no measurable H+ transport activity . When the double mutant was combined with the Glugate mutation ( E148A ) , which was used as a surrogate of the Glugate conformation observed in the CLC-1 structure , the Cl− throughput dramatically increased ( 22-fold with respect to the single E148A mutant; Figure 8—figure supplement 1 ) . Single mutations ( F348T or I356G ) in the E148A background showed intermediate increases in Cl− throughput , suggesting that the effects of these mutations are somewhat additive .
The human CLC-1 channel exhibits interesting structural differences in the Cl− transport pathway and the gates , which can explain why this protein functions as a Cl− channel instead of a Cl−/H+ antiporter . The outer gate of the channel remains open because the carboxylic side-chain Glugate is located off to the side , away from the Cl− transport pathway ( Figure 4 ) . The inner kinetic barrier seems to be substantially lowered compared to transporters owing to a wider pore diameter near the cytosolic side ( Figure 6 ) . The pore widening is subtle , but distinctive enough to reveal a pattern separating channels and transporters at the protein sequence level ( independent of the presence or absence of a Glugate ) ( Figure 6—figure supplement 1 ) . The position of the Glugate residue in CLC-1 is unique among CLC structures so far observed . The new Glugate position , where its carboxylic side chain is directed off to the side of the Cl− pathway , is enabled by a pocket that is large and hydrophilic ( owing to its bifurcated pore structure ) enough to accommodate Glugate’s side chain . This pocket may also exist in other Glugate-containing CLC channels ( i . e . , CLC-0 and CLC-2 ) but does not seem to exist in transporters because of a different arrangement of neighboring amino acids . It seems likely that this Glugate position is key to understanding why CLC-1 exhibits a stable open ( i . e . , conducting ) state . On the basis of mutagenesis studies ( Dutzler et al . , 2003; Fahlke et al . , 1997 ) , the Glugate in CLC-0 and CLC-1 has been identified as a ‘voltage sensor’ because its removal abolishes voltage-dependent gating . From this observation , we would suggest that the position of Glugate ( i . e . , whether it resides off to the side , not occluding the pore , or within the pore ) depends on the transmembrane voltage and generally dictates gating each CLC-1 monomer’s pore ( also referred to as a ‘protopore’ ) . An unresolved issue raised by the new Glugate side chain conformation is this: if this conformation corresponds to the conducting state , how is it favored by low pH outside ( Rychkov et al . , 1996 ) ? One possibility is the Glugate might be protonated in this conformation . Alternatively , low pH might stabilize a conformation of Glugate outside the pore , as in the EcCLC E148Q mutant . This conformation would also remove Glugate from the pore and permit conduction . Finally , the pH effect might be produced allosterically by protonation of an unidentified amino acid on the extracellular side . For example , both CLC-2 and CLC-K channels are inhibited by external pH <6 . 5 , but it has been shown that a His residue ( H532 of CLC-2 and H497 of CLC-K ) , which is located ~20 Å away from the pore , is responsible for this effect ( Gradogna et al . , 2010; Niemeyer et al . , 2009 ) . This issue remains unresolved for now . Functional experiments using EcCLC provide support for our model that a low kinetic barrier in the cytosolic vestibule is necessary for high Cl− transport rates , which are general characteristics of native CLC channels ( Figure 8 and also see ( Jayaram et al . , 2008 ) ) . The results indicate that a small increase in the pore diameter and a decrease in hydrophobicity of the pore lining can substantially lower the kinetic barrier . The structures , however , suggest that the extent might be somewhat less in the CLC-1 channel than in the CLC-K channel because of CLC-1’s SerC ‘transporter-like’ conformation . This is in fact consistent with the observation that CLC-1 has vestigial H+ transport activity ( Picollo and Pusch , 2005 ) and a relatively slow Cl− throughput compared to that of CLC-K channels ( 1 . 2–1 . 8 pS versus 20–30 pS of CLC-K ) ( L'Hoste et al . , 2013; Saviane et al . , 1999; Scholl et al . , 2006; Weinreich and Jentsch , 2001 ) . What then causes the SerC to adopt its flipped-down conformation in the CLC-K channel ? In CLC-K , position 425 contains a bulky amino acid ( Y425 ) , in contrast to other CLC proteins . In the canonical conformation SerC would sterically clash with Y425 ( e . g . , the center-to-center distance between the SerC-Oγ and Y425-Cε atoms would become 2 . 3 Å ) . We speculate that this steric incompatibility imposed by the unique Y425 might lead to the flipped-down conformation of SerC in the CLC-K channel . The observed low Cl− occupancy at Scen in the CLC-1 structure has a striking resemblance to previous crystallographic observations on EcCLC , wherein Scen remained unoccupied when experiments were performed with TyrC mutants lacking H+ transport activity or with pseudohalides , which permeate without coupled H+ transport ( Accardi et al . , 2006; Nguitragool and Miller , 2006 ) . It has been shown that in EcCLC , low Cl− occupancy correlates with low anion binding affinity ( Picollo et al . , 2009 ) . This comparison suggests a reduced Cl− binding affinity at Scen in CLC-1 , although further biophysical measurements will be necessary to confirm this . We speculate that this feature contributes to reduced H+ transport and increased Cl− conduction . Possible causes underlying the altered Cl− affinity include the shifted position of TyrC and subtle changes in positions and orientations of neighboring backbone nitrogen atoms coordinating the Cl− ion . For example , we note that CLC-0/1/2 channels have smaller , more flexible residues ( Gly or Ala ) at the G483 position , in contrast to Leu , Ile , or Val in CLC transporters . CLC-1 is now the second structure of a channel-forming CLC , the first being CLC-K ( Park et al . , 2017 ) . One of the major features giving rise to channel behavior is a more conductive pore . The structural differences giving rise to the higher Cl− conductivity are fairly subtle: the pore is slightly wider and the chemical properties a little different , accounting for what we propose to be a reduced kinetic barrier . We think there is a very important lesson here . Throughput rates in the range of 106 ions per second do not require a wide pore . We conclude that even if the pore in places is on average narrower than the ion , as long as the lining atoms are favorable to a conducting ion with respect to their electrostatic and chemical properties , and as long as they are sufficiently dynamic ( i . e . they can move out of the way ) , then the ion can diffuse through . We offer as an example of this idea , the selectivity filter of K+ channels ( Zhou et al . , 2001 ) . The atomic structures show us that in fact the pore’s radius between the K+ binding sites is smaller than the radius of a K+ ion . And yet some K+ channels approach throughput rates of 108 per second . It is not surprising to now understand that the radius of the pore in CLC channels and transporters is not very different . The structures of CLC-1 and CLC-K channels support the idea that CLC channels are ‘broken transporters’ ( Jayaram et al . , 2008; Lísal and Maduke , 2008; Miller , 2006 ) , where their channel function is built upon a transporter structure with modifications of the gates . The structures demonstrate that relatively small changes in the active site and ion transport pathway of a transporter gives rise to channel function .
Human CLC-1 was expressed in HEK293 GnTI− cells ( ATCC CRL-3022 ) by transduction using a modified baculovirus as described previously ( Goehring et al . , 2014; Park et al . , 2017 ) . A human CLC-1 coding sequence ( CDS ) was synthesized and inserted into a modified pFastBac vector , which contains a CMV promoter upstream of CDS . The expressed CLC-1 construct has a truncation of N-terminal 80 amino acids ( residues 2–80 ) , which were predicted to be unstructured , and its C-terminus is fused to enhanced green fluorescent protein ( eGFP ) ( it also contains a HRV 3C protease cleavage sequence between CLC-1 and eGFP ) . The vector was used for transformation of DH10Bac E . coli cells ( Invitrogen ) to generate a baculovirus bacmid . Baculoviruses were produced by transfecting Spodoptera frugiperda ( Sf9; ATCC CRL-1711 ) cells with the bacmid using Cellfectin-II ( Invitrogen ) . Viruses were then amplified twice for large-scale transduction . HEK293 GnTI− cells were grown at 37°C in suspension in Freestyle 293 medium ( Invitrogen ) supplemented 2% FBS in the presence of 8% CO2 . At a cell density of ~2 . 5 × 106 mL−1 , baculovirus was added to the culture ( 6–8% v/v ) . After incubating at 37°C for ~0 . 5 day , the culture was supplemented with 10 mM sodium butyrate , then further incubated at 30°C for 2 days before harvest . All protein purification steps were carried out at 4°C . Harvested HEK293 cells ( typically from 1 to 2 L ) were suspended in a buffer containing 50 mM Tris-HCl pH 7 . 5 , 300 mM NaCl , 1 mM dithiothreitol ( DTT ) , 1 mM ethylenediaminetetraacetic acid ( EDTA ) , and 10% v/v glycerol , and supplemented with protease inhibitors ( 50 μM leupeptin , 1 ug/mL aprotinin , 1 uM pepstatin and 1 mM phenylmethylsulfonyl fluoride ) . 1% dodecyl-β-maltoside ( DDM ) and 0 . 2% cholesteryl semisuccinate ( CHS ) were added to the cell suspension . After extraction for 1 . 5 h , the lysate was clarified by centrifugation ( Beckman Type 70Ti rotor , 40 , 000 RPM , 1 . 5 h ) . The clarified lysate was then mixed with 5 mL of CNBr-sepharose beads ( GE Healthcare ) coupled with anti-GFP nanobody for 2 . 5 h . Beads were washed on 60 mL of the buffer containing 0 . 04% DDM and 0 . 004% CHS . Bound protein was released from beads by overnight incubation with 5 mL buffer containing 0 . 04% DDM , 0 . 004% CHS , and 0 . 2 mg HRV 3C protease . The retrieved protein was concentrated to 0 . 5–1 . 0 mL using Amicon Ultra ( 100 kDa cutoff; EMD Millipore ) and applied to a Superose 6 300/10 GL column ( GE Healthcare ) equilibrated with 20 mM Tris-HCl pH 7 . 5 , 100 mM NaCl , 1 mM DTT , 0 . 5 mM EDTA , 0 . 04% DDM , and 0 . 004% CHS . The peak fractions were pooled and concentrated to ~4 mg/mL , and immediately used for cryo-EM grid preparation . The wild-type and mutant E . coli CLC transporter proteins ( EcCLC ) were expressed and purified essentially as described previously ( Dutzler et al . , 2002 ) . E . coli BL21 ( DE3 ) ( Novagen ) was transformed by pET28b vector containing EcCLC CDS , the C-terminal of which was fused to a hexa-histidine tag ( His-tag ) . E . coli cells were grown at 37°C in Luria broth ( LB ) medium containing 60 μg/mL kanamycin until they reached OD600 of 1 . 2 . The expression was induced by addition of 0 . 2 mM isopropyl β-D-1-thiogalactopyranoside ( IPTG ) . The cells were further grown at 21°C for ~16 h before harvest by centrifugation . The cell pellets were frozen with liquid N2 and stored at −80°C until purification . The frozen E . coli cells ( typically from 3 L ) were thawed and suspended in 20 mM Tris-HCl pH 7 . 5 and 150 mM NaCl . The cells were lysed by sonication ( 1 mM PMSF and 50 uM leupeptin were supplemented before the lysis ) , and 2% decylmaltoside ( DM ) and 10 mM imidazole were added . After 2-h gentle stirring at 4°C , the lysate was spun for 1 h at 15 , 000 rpm ( Beckman JA-17 rotor ) . The supernatant was mixed with 5 mL of Talon cobalt agarose beads ( Takara Bio ) for 2 h . The beads were packed in a column and washed with 25 mL of lysis buffer containing 20 mM imidazole , 12 . 5 mL of buffer containing 30 mM imidazole , and then 12 . 5 mL of buffer containing 40 mM imidazole . The protein was eluted by buffer containing 200 mM imidazole . The eluate was concentrated to ~0 . 5 mL using Amicon Ultra ( 50 kDa cutoff ) . The His-tag was removed by adding 0 . 5 U of Lys-C endopeptidase ( Roche ) and incubating the mixture at 23°C for 3 h . The eluate was applied to a Superdex 200 300/10 GL column ( GE Healthcare ) equilibrated with 25 mM Tris-HCl pH 7 . 5 , 100 mM NaCl , 1 mM DTT , 0 . 5 mM EDTA , 10% glycerol , and 0 . 3% DM . The peak fraction was collected and used for reconstitution without freezing . 3 μL of purified CLC-1 protein was applied to a glow-discharged gold ( or copper for the third dataset ) Quantifoil R 1 . 2/1 . 3 holey carbon grids ( Quantifoil ) and incubated for 15 s . Grids were then blotted for 1 . 5–2 . 0 s at 4°C and 90% humidity and plunge-frozen in liquid-nitrogen-cooled liquid using Vitrobot Mark III ( FEI ) . The data sets were collected on a Titan Krios electron microscope ( FEI ) operated at an acceleration voltage of 300 kV . Dose-fractionated images were recorded on a K2 Summit direct electron detector ( Gatan ) operated in super-resolution counting mode ( a super-resolution pixel size of 0 . 515 Å ) using SerialEM software ( Mastronarde , 2005 ) . For the first two datasets ( 2293 movies ) , the dose rate was 8 e− per pixel per s , and total exposure time was 10 s with 0 . 2 s for each frame ( total cumulative dose of ~75 e− per Å2 over 50 frames ) . For the third dataset ( 1998 movies ) , the dose rate was 5 . 33 e− per pixel per s , and total exposure time was 15 s with 0 . 15 s for each frame ( total cumulative dose of ~75 e− per Å2 over 100 frames ) . Defocus values were set from −0 . 8 μm to −2 . 4 μm . Dose-fractionated movies were corrected for gain and motion by MotionCor2 ( Zheng et al . , 2017 ) . Also the pixels were binned to 1 . 03 Å/pixel during this process . Defocus values were estimated using CTFFIND4 ( Rohou and Grigorieff , 2015 ) on the summed micrographs produced by MotionCor2 ( using the full dose ) . Particles were picked automatically by RELION2 ( Kimanius et al . , 2016; Scheres , 2012 ) , and obvious artifacts , such as ice contamination and carbon foil , were removed by manual inspection . Total 725 , 959 particles were extracted with a box size of 320 pixels and subjected to reference-free 2D classification ( performed separately per dataset ) . Based on visual inspection of quality of 2D average classes , 411 , 260 particles were pooled . This particle set was then applied to the alignpart_lmbfgs program ( Rubinstein and Brubaker , 2015 ) to perform per-particle motion correction ( particle polishing ) . The particle polishing step was done using motion-corrected ( whole-frame-only ) movie stacks , which were first produced by MotionCor2 and then 2x or 4x frame-binned by relion_image_handler ( resulting in a total of 25 frames per movie and 3 e− per Å2 per frame ) . Particles were extracted from 1 to 13 frames ( total dose of 39 e− per Å2 ) and using alignparts_lmbfgs’s exposure filter . The ‘polished’ particles were subjected to another round of clean-up by RELION 2D classification ( resulting in 350 , 750 particles ) . The initial model was generated by RELION auto-refine using particle images from the first dataset and a 50 Å lowpass-filtered model from the CLC-K channel density map ( excluding antibody fragments; ( Park et al . , 2017 ) ) . All 350 , 750 polished particle images were subjected to auto-refine ( RELION 2 . 1 ) , using the updated initial model and a soft mask surrounding the protein and detergent micelle density . This refinement step produced a 3 . 8 Å map ( Figure 2—figure supplement 1B ) . This was then followed by a RELION 3D classification procedure skipping image alignment ( sorting into five classes ) . Particles from two classes were combined ( 175 , 613 particles ) by visual inspection in UCSF Chimera ( Pettersen et al . , 2004 ) and subjected to RELION auto-refine again . During the later iterations ( upon entering the local search mode ) , the soft mask was updated to contain only the transmembrane or cytosolic domain ( focused refinement ) . The resolution of the final TMD domain map ( 3 . 36 Å ) was estimated by RELION based on gold-standard Fourier shell correlation ( FSC ) of independently refined half maps ( using the 0 . 143 cut-off criterion ) . The focused refinement of the cytosolic domain was performed by 2 iterations of local refinement using reference maps in which information at lower than 4 . 6 Å resolution were combined from the previous iteration’s two half maps . The nominal resolution of the final CTD map is 4 . 1 Å , but this is likely somewhat overestimated ( the resolution before the focused refinement is 4 . 5 Å ) . Local resolution was estimated using RELION2’s postprocess program ( Figure 2—figure supplement 1A ) . Unless stated otherwise , the TMD map shown in figures is a combined map , which was sharpened ( B-factor of −97 . 9 Å2 ) and lowpass-filtered at 3 . 36 Å by RELION’s automatic postprocess procedure using user-provided soft masks . The TMD map in Figure 2C and Videos 1 and 2 was sharpened with a B-factor of −97 . 9 Å2 and low-pass filtered at 3 . 1 Å . The CTD density map was low-pass filtered at 4 . 2 Å without B-factor sharpening . An initial model of the CLC-1’s TMD was generated by the SWISS-MODEL homology modelling webserver using the CLC-K model ( PDB ID: 5TQQ ) as a template . The output model was fit into the TMD density map using Chimera and rebuilt using Coot ( Emsley et al . , 2010 ) . Model refinement was done in real space using Rosetta 3 . 7 using a script developed for cryo-EM model refinements ( Wang et al . , 2016 ) ( Table 1 ) . The first round was performed with an asymmetric unit model , and the five best output models were selected based on Rosetta’s energy scores . A consensus model was generated by combining fragments from these models based on the fit to the density map . The subsequent two rounds of Rosetta refinement were done with two-fold symmetry imposed . To prevent overfitting , the weight between Rosetta energy scores and the fit to the experimental density map was adjusted , and test refinement was performed on one of two half maps . The output models were then compared to both half maps by calculating FSC ( Figure 2—figure supplement 2 ) . To this end , we used a weight of 25 , which gave us a good fitting to the map and negligible overfitting . While the first two rounds of refinement were done using one of the two half maps , the last round was performed on the combined map to maximize the use of experimental data in refining the model ( see Figure 2—figure supplement 2 for FSC between the final model and the combined map ) . The final model was selected among ~2000 Rosetta-generated models based on Rosetta’s total score ( top 20% ) and the fit of side chains to the map ( visual inspection ) . No further modifications were made except for Cl− ions at Sext and Sint , which were modelled in Coot ( Coot’s real-space refinement was used ) since Rosetta could not refine Cl− ions . Modelling of CTD was done similarly using Rosetta , but using a crystal structure of CLC-0 CBS domain as an initial model . A weight of 7 was used , and the refinement was limited to 4 . 5 Å resolution . As side chains were not visible in the CTD density map , we removed all side chain atoms from the final CTD model generated by Rosetta . The following segments were not modelled as they were invisible in the density maps: N–115 , 251–262 ( a cytosolic segment between αF and αG ) , 671–796 ( a loop in CTD ) , and 877–988 ( C ) . MolProbity was used for structural validation of models ( Table 1 ) ( Chen et al . , 2010 ) . Detection of pores and estimation of pore radii ( Figures 3 and 6 ) were performed using Caver ( Chovancova et al . , 2012 ) . In the case of EcCLC ( PDB ID: 1OTS ) and CmCLC ( PDB ID: 3ORG ) , Glugate ( E148 of EcCLC and E210 of CmCLC ) was mutated to Ala before estimation since its side chain is blocking the Cl− pathway . In the case of bovine CLC-K ( PDB ID: 5TTQ ) ( Figure 6B ) , we changed the rotamer conformation of V166 ( equivalent to Glugate ) from original gauche+ ( 63° ) to trans ( 175° ) . With the original rotamer , the constriction around Sext was found to be too narrow ( radius < 0 . 9 Å ) for pore detection . Because both rotamers can be fitted equally well into the cryo-EM density map , it is uncertain which is right or if both can exist in the protein . We note that trans is in general an energetically more favored rotamer than gauche+ . Water accessibility in CLC-1’s vestibules ( Figure 4B ) was probed using HOLLOW ( Ho and Gruswitz , 2008 ) using a probe radius of 1 . 4 Å . Protein electrostatics were calculated using the Adaptive Poisson-Boltzmann Solver ( Baker et al . , 2001 ) with a parameter of 150 mM monovalent salt concentration . UCSF Chimera and PyMOL ( Schrödinger ) were used to prepare structure figures . To reconstitute EcCLC mutant proteins for Cl− efflux assays , E . coli polar lipids in chloroform ( Avanti Polar Lipids ) was dried in a glass tube with an argon stream , followed by overnight incubation in a vacuum chamber . Dried lipids were suspended by sonication in buffer ( RB-Cit ) containing 25 mM sodium citrate ( pH 4 . 6 ) and 300 mM KCl and then solubilized with 35 mM ( 3- ( ( 3-cholamidopropyl ) dimethylammonio ) −1-propanesulfonate ) ( CHAPS; Anatrace ) and additional sonication . Purified EcCLC protein was added to the lipid/CHAPS mixture in a protein-to-lipid ratio of 1:5000 ( wt:wt ) . After 30 min incubation at 23°C , the mixture was dialyzed against RB-Cit buffer to remove CHAPS . The dialysis was carried out at 4–10°C for 48 h with three additional buffer changes . To reconstitute EcCLC proteins for fluorescence-based flux assays , the same procedure was used except that buffer containing 10 mM HEPES-NaOH ( pH 7 . 0 ) and 450 mM KCl instead of RB-Cit and a protein-to-lipid ratio of 1:500 ( wt:wt ) were used . After dialysis , proteoliposome vesicles were aliquoted , flash-frozen with liquid N2 , and stored at −80°C until use . The Cl− efflux ( dump ) assays were performed essentially as described previously ( Jayaram et al . , 2008; Walden et al . , 2007 ) . A frozen aliquot of vesicles was thawed and briefly sonicated in the bath sonicator ( Branson ) . Vesicles were extruded through a 0 . 4 μm polycarbonate filter 19 times ( Avanti Mini-Extruder ) . The extruded vesicles were desalted with a spin column packed with Sephadex G-50 resin ( ~2 . 5 mL bed volume ) equilibrated with buffer ( EB ) containing 25 mM sodium citrate ( pH 4 . 7 ) , 250 mM K2SO4 , and 1 mM NaCl . 100 μL of the desalted vesicles were then mixed with 900 μL EB in a chamber equipped with a magnetic stirrer and Cl−-selective electrode ( Fisher Accumet ) . Changes in extravesicular Cl− concentration was monitored over time by the Cl−-selective electrode connected to a computer through a digitizer ( DataQ ) . To calibrate the electrode , 0 . 1 mM NaCl was added before the vesicles were added . Flux was initiated by addition of 2 μg/mL valinomycin and 1 μg/mL carbonyl cyanide-p-trifluoromethoxyphenylhydrazone ( FCCP ) or 3 μg/mL valinomycin ( for E148A mutants ) . At the end of assays , 30 mM octyl β-glucoside ( Anatrace ) was added to release all Cl− content from vesicles . Calculation of Cl− transport rates were carried out as described previously ( Walden et al . , 2007 ) . Volume changes by dialysis , extrusion , and desalting steps were included in calculation . The fluorescence-based flux assays were performed as follows based on ( Feng et al . , 2012 ) . A frozen aliquot of vesicles was thawed and briefly sonicated in a bath sonicator . 3 μL of vesicles were mixed with 40 μL of assay buffer containing 20 mM HEPES-NaOH ( pH 7 . 0 ) , 450 mM K-gluconate and 4 μM ACMA in a well of a 384-well fluorescence assay . After measuring initial AMCA fluorescence intensity ( λEx=410 nm , λEm=490 nm ) , Cl−/H+ flux was initiated by addition of 1 μM valinomycin , followed by monitoring fluorescence over time ( 10 s intervals ) using a plate reader ( Tecan Infinite M1000 ) at 27°C . Note that there is a dead time for measurement between t = 80 s to t = 120 s due to handling of the plate during valinomycin addition . Valinomycin was added to the reactions at t = ~100 s . As a control , 0 . 9 μM carbonyl cyanide 3-chlorophenylhydrazone ( CCCP ) was added to the assay mixture at the end of the experiment to dissipate an accumulated H+ gradient . To measure relative H+ transport activity of each mutant , time required to fluorescence reaches 75% ( or 85% in the case of S106G/F348A double mutant ) of the initial fluorescence upon addition of valinomycin was calculated . This time value was then inversed and normalized with respect to a value obtained with wild-type EcCLC . Cryo-EM density maps of human CLC-1 have been deposited in the electron microscopy data bank under accession code EMD-7544 and 7545 . Atomic coordinates have been deposited in the protein data bank under accession code 6COY and 6COZ . | Channels and transporters are two classes of proteins that transport molecules and ions – collectively referred to as “substrates” – across cell membranes . Channels form a pore in the membrane and the substrates diffuse through passively . Transporters , on the other hand , actively pump substrates across a membrane , consuming energy in the process . Thus , channels and transporters work in distinct ways . Channels and transporters most often have unrelated structures , but there are rare examples of both existing within the same family of structurally similar proteins . CLC proteins , for example , include both chloride ion channels and transporters that pump chloride ions in one direction by harnessing the energy from hydrogen ions flowing in the other direction . It remains unclear why some CLC proteins work as channels while others are transporters , especially since the two seem indistinguishable on the basis of the order of their amino acids – the building blocks of all proteins . The conservation of the amino acid sequences implies they are structurally very similar . How then can different members perform such energetically distinct processes ? Park and MacKinnon now show that the answer to this question serves as a reminder of how subtle nature can be . Indeed , while the structure of a human CLC channel ( called CLC-1 ) is indeed similar to those of CLC transporters , one amino acid adopts a unique shape that explains why the protein cannot act as a transporter . This specific amino acid , a glutamate , is central to the exchange of chloride and hydrogen ions in CLC transporters . Park and MacKinnon show that its conformation in the CLC-1 channel stops this exchange , while leaving the pore open for the passive transport of chloride ions . Also , two other amino acids along the ion diffusion pathway in the CLC channel are smaller than their counterparts in CLC transporters , and so allow chloride ions to diffuse through more quickly . Lastly , Park and MacKinnon also note that channels do not require a wide pore: instead ions can still flow rapidly through a narrow pore if the chemical environment inside permits it . CLC proteins perform a number of important roles in humans , and mutations in CLC-encoding genes underlie numerous heritable diseases . It remains too early to know how this mechanistic study may or may not impact treatments , yet the findings will likely interest scientists working on ion conduction mechanisms and the evolution of molecular function . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2018 | Structure of the CLC-1 chloride channel from Homo sapiens |
Nanoparticles are used extensively as biomedical imaging probes and potential therapeutic agents . As new particles are developed and tested in vivo , it is critical to characterize their biodistribution profiles . We demonstrate a new method that uses adaptive algorithms for the analysis of hyperspectral dark-field images to study the interactions between tissues and administered nanoparticles . This non-destructive technique quantitatively identifies particles in ex vivo tissue sections and enables detailed observations of accumulation patterns arising from organ-specific clearance mechanisms , particle size , and the molecular specificity of nanoparticle surface coatings . Unlike nanoparticle uptake studies with electron microscopy , this method is tractable for imaging large fields of view . Adaptive hyperspectral image analysis achieves excellent detection sensitivity and specificity and is capable of identifying single nanoparticles . Using this method , we collected the first data on the sub-organ distribution of several types of gold nanoparticles in mice and observed localization patterns in tumors .
Nanoparticles ( NPs ) can be fashioned in precise shapes and sizes from a wide variety of materials . This synthetic versatility makes NPs excellent tools for wide-ranging biomedical applications including in vivo imaging ( Jokerst et al . , 2012; Durr et al . , 2007 ) , drug delivery ( Hauck et al . , 2008 ) , photothermal therapy ( Huang et al . , 2006; von Maltzahn et al . , 2009 ) , and gene transfection ( Huang et al . , 2009 ) . In particular , metal and metal oxide NPs made from gold , silver , iron , and titanium are commonly used in biomedicine owing to their unique electromagnetic properties ( Giustini et al . , 2011; Husain et al . , 2015; Lee and El-Sayed , 2006 ) . Once administered to a living subject , these NPs may exhibit vastly different pharmacokinetics and uptake profiles that are contingent on NP shape , size , surface coating , and other factors ( Owens III and Peppas , 2006; Chen et al . , 2009; Zhang et al . , 2009; He et al . , 2010; Decuzzi et al . , 2010 ) . These differences manifest not only at the scale of whole organs but also at the cellular level ( Giustini et al . , 2011; Yang et al . , 2014; Sadauskas et al . , 2009 ) . Ideally , biodistribution studies should address various scales – from whole animal to tissue and cellular interactions – in order to understand a given NP’s in vivo behavior . Current studies commonly employ inductively-coupled plasma ( ICP ) techniques ( Niidome et al . , 2006 ) or electron microscopy ( EM ) ( Giustini et al . , 2011 ) to interrogate metallic NP biodistribution . However , each of these techniques has notable disadvantages . ICP can be coupled to mass spectrometry ( MS ) or atomic/optical emission spectrometry ( AES/OES ) to quantify the presence of a metallic species in tissues of interest with high sensitivity ( ~10 parts per billion ) ; incidentally , detection of large metal NPs with ICP relies upon dissolving samples in strong acids . The need to dissolve NP-containing samples has severe drawbacks with respect to characterizing particle uptake including the complete loss of spatial insights such as NP distribution patterns within the given tissue . Moreover , the sample preparation itself can impede detection sensitivity , especially for small tissue samples and tissues with intrinsically low NP uptake , which must be diluted in acid . Conversely , EM studies provide exquisite high-resolution images of NP uptake by individual cells . Unfortunately , EM requires cumbersome sample preparation and acquires qualitative data over fields of view that are too small to be tractable for whole organ studies . Fluorescence ( Zhang et al . , 2009; He et al . , 2010; Poon et al . , 2015 ) and radioactivity ( Kreyling et al . , 2015; Collingridge et al . , 2003 ) detection can also be used to assess NP biodistribution , however these techniques typically require the addition of a labeling moiety to the NP prior to in vivo use . Aside from the potential that labels may detach or even alter NP pharmacokinetic properties , whole-organ studies with these techniques can be impeded by poor spatial resolution . Hyperspectral dark-field microscopy ( HSM ) is a technique that obtains scattered light spectra from a sample on a per-pixel basis ( Roth et al . , 2015 ) . HSM is capable of identifying individual nanoparticles in pure solutions and cell culture by their intrinsic scattering spectra without the addition of a labeling molecule ( Fairbairn , 2013; Fairbairn et al . , 2013; Patskovsky et al . , 2015 ) . This approach may be ideal for detecting metallic nanoparticles with unique visible and near-infrared ( NIR ) spectral signatures . Unlike current methods that characterize NP biodistribution , HSM simultaneously achieves diffraction-limited spatial resolution and excellent detection sensitivity without destroying the sample . HSM has been used to study NP uptake in cell culture ( Yang et al . , 2014; Fairbairn , 2013; Fairbairn et al . , 2013; Patskovsky et al . , 2015 ) and the induction of toxic effects in tissue ( Husain et al . , 2015 ) , but its use for characterizing NP biodistribution has not yet been demonstrated due to several outstanding constraints . The primary limitation that has prevented HSM from being used in evaluating the biodistribution of NPs in tissue is the inability to accurately distinguish NPs from the background of tissue scattering . To abate this limitation , we use a modified dark-field microscope that uses oblique sample illumination to enable 150-fold brightness enhancement and ~15-fold better signal to noise ratio ( SNR ) than standard dark-field optics ( Badireddy et al . , 2012; Zhang et al . , 2015 ) . Another challenge with HSM detection stems from the reality that individual NPs within a given sample do not exhibit the exact same spectrum . Furthermore , the NP uptake within tissues inevitably results in a combination of the NP spectrum with tissue scattering , which can be spectrally diverse . Current approaches such as spectral angle mapping ( Roth et al . , 2015; Kruse , 1993; De Carvalho and Meneses , 1999; Luc , 2005; Roth et al . , 2015 ) ( originally developed for non-biological applications ) and manual delineation ( Husain et al . , 2015; Roth et al . , 2015 ) cannot adapt to these conditions and may yield high false positive and false negative detection rates . It has been observed that no HSM method to date has demonstrated robust capabilities for quantifying false positive rates or other diagnostic measures ( Roth et al . , 2015 ) . Thus , HSM methods must be customized to address spectral mixing and diffraction effects as well as detection sensitivity and specificity if they are to be successfully used for microscopic analyses of complex biological samples . Here , we demonstrate Hyperspectral Microscopy with Adaptive Detection ( HSM-AD ) , the first HSM method based on adaptive clustering , as a viable alternative to current techniques for assessing whole-organ biodistribution and cellular uptake of NPs . In this study , we collected tissues of interest from mice that were injected with large gold nanorods ( LGNRs ) ( SoRelle et al . , 2015 ) , gold nanoshells ( Nanoshells ) , and silica-coated gold nanospheres ( GNS@SiO2 ) , and we developed pre-processing and adaptive algorithms to identify NPs that accumulated in tissue sections based on their spectral signatures . The implementation of an adaptive classification algorithm for spectral classification extended HSM’s single NP detection capabilities to tissue samples with negligible false-positive detection . HSM-AD was sufficiently robust for detecting NPs in images of different organ tissues and images acquired using variable illumination conditions . This approach may be preferable to conventional biodistribution assays for studies that simultaneously require quantification of relative NP uptake in various clearance organs and wide-field high-resolution images with histological detail .
LGNRs ( ~100 × 30 nm ) exhibiting a near infrared plasmonic peak ( Figure 1a ) were synthesized , biofunctionalized , and administered to healthy and tumor-bearing nude ( Foxn1nu/nu ) mice as previously reported ( SoRelle et al . , 2015; Liba et al . , 2016 ) . Mice were euthanized 24 hr post-injection , and various tissues were resected and fixed in 10% formalin . Fixed tissues were sectioned into 5 µm thick slices , mounted on glass slides , and stained with Hematoxylin and Eosin ( H&E ) as per standard histological preparation ( Figure 1b ) . H&E-stained sections were imaged at 40x or 100x magnification in conventional dark-field and hyperspectral microscopy modes ( CytoViva ) ( Figure 1—figure supplement 1 ) . Conventional dark-field images ( Figure 1c ) were used to guide anatomical feature identification . All spectral data and quantitative comparisons presented in this report were derived from the analysis of hyperspectral images . 10 . 7554/eLife . 16352 . 003Figure 1 . Overview of nanoparticle biodistribution analysis with HSM-AD . ( a ) Large gold nanorods ( LGNRs , ~100 × 30 nm ) exhibiting near infrared plasmon resonance were synthesized , functionalized , and intravenously injected into live nude mice . ( b , c ) 24 hr post-injection , the animals were euthanized and tissues were resected and prepared as normal histological sections for characterization with bright-field ( b ) and dark-field microscopy ( c ) neither of which was able to visualize the distribution of the LGNRs . ( d ) The same section was then imaged with hyperspectral microscopy , which showed clear signs of LGNRs accumulation ( denoted by red hues ) in various areas of the tissue and exhibited spectral peaks matching the LGNR plasmon resonance . ( e ) We then trained an adaptive clustering algorithm for spectral identification of LGNRs with hyperspectral images from injected mice . The algorithm identified several characteristic spectra representing the tissue and the H&E staining , as well as one unique spectrum representing the LGNRs ( depicted in orange ) , altogether representing a library of 5 spectra . Once a spectral cluster library is produced from the training dataset , images of unknown tissue samples can by analyzed for the presence of LGNRs via automated classification . ( f ) The resulting HSM-AD images depict the location of all points within the sample that exhibit the LGNR spectrum ( orange for LGNRs , grayscale for tissue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 00310 . 7554/eLife . 16352 . 004Figure 1—figure supplement 1 . Diagram of the CytoViva microscope used for dark-field and hyperspectral image acquisition . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 00410 . 7554/eLife . 16352 . 005Figure 1—figure supplement 2 . Image segmentation , including method for dynamic threshold determination . ( a ) A histogram of the peak intensities of each pixel in an image can be roughly divided into background ( noise ) , tissue scattering , and LGNR and bright tissue scattering . ( b ) Detection of minHist and peakHist , as described in Methods . ( c , d ) A characteristic hyperspectral image ( c ) and its corresponding segmentation map ( d ) showing background ( blue ) , tissue ( cyan ) , and potential LGNRs and bright tissue ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 00510 . 7554/eLife . 16352 . 006Figure 1—figure supplement 3 . Detailed flowchart of steps used in HSM-AD algorithm . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 00610 . 7554/eLife . 16352 . 007Figure 1—figure supplement 4 . Typical cluster results for pixel classification in an image of tissues with injected LGNRs . For a given image ( >250 , 000 pixels ) , each pixel is binned into one of the five spectral clusters . This plot depicts the means ( solid lines ) and standard deviations ( shaded areas ) of all classified pixel spectra . Although the adaptive clustering algorithm is agnostic with respect to defining the spectral clusters ( with the exception of chromatic aberration , which is user-defined ) , the learned clusters can be readily correlated to the major scattering components present in each sample , i . e . , hematoxylin-stained nuclei ( green ) , eosin-stained cytoplasm ( blue ) , and LGNRs ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 007 A hyperspectral camera with a detection range of 400–1000 nm was used to image scattered light from each sample in transmission mode . In the resulting images ( Figure 1d ) , each pixel contains the spectral profile of the sample at the corresponding spatial position and can be used to detect LGNRs with near diffraction-limited resolution ( 1 μm ) . While standard dark-field images did not reveal notable differences between uninjected and injected samples , the hyperspectral images , in which three bands of the spectrum ( 800 . 0 nm , 700 . 6 nm , and 526 . 2 nm ) were respectively color-coded as red , green , and blue , indicated that species with strong near infrared scattering ( putative LGNRs , depicted in orange ) were present in injected tissues but not observable in control tissues . We created HSM-AD , a method that combines pre-processing and adaptive classification algorithms to automatically detect and quantify LGNRs in hyperspectral images . The pre-processing stage is described in detail in the methods section ( Figure 1—figure supplements 2 , 3 ) . One of the first stages of processing includes vignette correction and a determination of whether each pixel in the image belongs to one of three categories—background , tissue , or potential LGNR—based on its average intensity across the measured spectrum . Only high-intensity pixels that belong to the potential LGNR group are classified by the adaptive algorithm ( Figure 1—figure supplement 2 ) . For the training of the adaptive algorithm , pre-processed images of tissue samples from LGNR injected mice were input into a standard k-means clustering algorithm ( Bishop , 1995 ) . Four initial clusters were identified using this scheme and were used to produce a spectral cluster library ( Figure 1e ) . Owing to the unique spectral profile of LGNR scattering , the particle spectrum was automatically recognized as a separate cluster by the k-means algorithm . Next , a fifth cluster was manually added to the spectral cluster library to account for edge artifacts caused by chromatic aberrations that were frequently falsely detected as LGNRs ( Figure 1—figure supplement 3 ) . The five spectra in this library were used as cluster centers for automatic detection of LGNRs in tissue sections using a nearest centroid ( or nearest-neighbor ) classifier . In this scheme , pixels that exhibited a spectrum that was closest ( in a Euclidean sense ) to the LGNR cluster center were identified as containing LGNRs ( denoted as LGNR+ ) . The rest of the pixels were classified as not containing LGNRs ( denoted as LGNR- ) . Initial validation of the algorithm shows that regions of the image that were detected as LGNR+ indeed exhibited the characteristic plasmonic peak at around 900 nm while pixels identified as LGNR- did not ( Figure 1f ) . The mean and standard deviation spectra of pixels classified into each cluster for a representative image also indicate the high fidelity of the algorithm ( Figure 1—figure supplement 4 ) . We characterized the sensitivity and specificity of HSM-AD by three methods . First , we measured the false positive rate in uninjected tissue samples to obtain a specificity of 99 . 7% ( Figure 2—figure supplements 1–4 ) . The false positives , which also have a spectral peak near 900 nm , usually appear near the edges of the tissue section . We attribute this red-shift of the spectrum to chromatic aberrations , ostensibly due to the spectral dependence of the diffraction diameter ( Lipson and Lipson , 2010 ) . Next , we measured the false negatives in an image of LGNRs in mounting media ( CytoSeal 60 , Electron Microscopy Sciences ) on a glass slide and obtained a detection sensitivity of 99 . 4% . We attributed the false negatives to LGNRs with hybridized surface plasmon resonances ( Funston et al . , 2009 ) , which resulted in spectral scattering that was different from the distinct plasmonic resonance of single LGNRs ( Figure 2—figure supplements 5 , 6 ) . Because all training and test samples were mounted using the same media , spectral shifts due to local refractive environments did not contribute to false detection ( Figure 2—figure supplement 7 ) . Independently we also calculated specificity and sensitivity by analyzing LGNR-injected tissue samples ( see Methods ) . We obtained a sensitivity of 89 . 5% and a specificity of 98 . 5% using this approach . The high sensitivity of the automated algorithm is further evident from its ability to detect single LGNRs , both on a glass slide and in injected tissue samples ( Figure 2—figure supplements 5 , 8 ) . We demonstrated HSM-AD as a potential biodistribution technique by analyzing various tissues resected from mice ( Figure 2 ) . For quantitative measurements of LGNR uptake , we analyzed kidney tissue from uninjected ( Figure 2a , Figure 2—figure supplements 2a , 3a , 4a ) and injected ( Figure 2b , Figure 2—figure supplements 9a , 10a , 11a ) mice . Our analysis found a relative LGNR signal of 4 . 8% ± 2 . 3% in injected mouse kidney tissue . By comparison , a relative LGNR signal of 0 . 08% ± 0 . 01% was measured from uninjected samples , indicative of the method’s high specificity ( Figure 2c , Figure 2—figure supplement 1 ) . Similar low false positive rates were measured in other organ tissues ( Figure 2—figure supplements 2b–e , 3b–e , 4b–e ) . In addition to the kidney , HSM-AD was used to analyze LGNR uptake in liver , lung , muscle , and spleen sections to demonstrate an alternative to common biodistribution techniques . While a conventional biodistribution study of LGNRs has not yet been reported , HSM-AD analysis indicated that LGNRs exhibited a similar uptake profile ( mostly in the liver and spleen ) as commonly-used smaller gold nanorods ( Zhang et al . , 2009; Niidome et al . , 2006 ) . The greatest relative LGNR signal ( 38 . 5% ± 4 . 5% ) was observed in the spleen . LGNRs were also concentrated in the liver ( 7 . 5% ± 1 . 5% ) . Particle uptake was minimal in lung tissue ( 0 . 5% ± 0 . 1% ) and muscle tissue ( 0 . 8% ± 0 . 5% ) sections ( Figure 2d , Figure 2—figure supplements 9–11 ) . Tissue sections without H&E staining were also analyzed and yielded results similar to those obtained for H&E stained sections ( Figure 2—figure supplements 12–14 ) . 10 . 7554/eLife . 16352 . 008Figure 2 . Sensitivity and specificity validation of HSM-AD . ( a , b ) Hematoxylin and Eosin ( H&E ) stained tissue samples ( kidney ) from uninjected ( a ) and injected ( b ) mice were imaged using dark-field and hyperspectral microscopy at 40x magnification and analyzed with HSM-AD to measure LGNR detection specificity and sensitivity . Conventional dark-field images highlight features including nuclei ( salmon-pink ) , cytoplasm ( green-brown ) , and erythrocytes ( yellow-orange ) within the tissues , but they reveal little information regarding the presence or absence of LGNRs . By comparison , putative LGNRs can be roughly identified as red-orange pixels in false-colored hyperspectral images , while nuclei and cytoplasm appear in green and indigo , respectively . ( c ) HSM-AD analysis of hyperspectral images demonstrates the absence of LGNRs in uninjected tissues and LGNR presence in injected samples ( two-tailed Student’s t-test , p=0 . 054 ) . Quantification of the relative LGNR signal from n = 4 tissue slices ( representing a total of 1 . 04 million pixels ) indicates that the false positive rate for LGNR detection ( determined from uninjected tissues ) is minimal . A detection specificity of 99 . 7% was determined from uninjected tissue sections , and a detection sensitivity of 99 . 4% was measured from samples of pure LGNRs analyzed using HSM-AD ( Figure 2—figure supplement 1 ) . ( d ) HSM-AD analysis of whole tissue sections ( n = 4 for each tissue type ) reveals quantitative differences in bulk LGNR uptake among various organs , in a manner analogous to conventional biodistribution methods . Quantitative data are presented as mean ± standard error of the mean ( s . e . m . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 00810 . 7554/eLife . 16352 . 009Figure 2—source data 1 . Data used for diagnostic and 95% CIs . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 00910 . 7554/eLife . 16352 . 010Figure 2—source data 2 . Data for whole organ uptake quantification . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01010 . 7554/eLife . 16352 . 011Figure 2—figure supplement 1 . Measured sensitivity and specificity values for HSM-AD method . ( a ) Sensitivity values were calculated from pure LGNR samples in CytoSeal on a microscope slide and from blinded manual identification of LGNR spectra from injected sections cross-referenced with algorithm results . Specificity values were calculated directly from uninjected ( LGNR- ) tissue samples and from blinded manual identification of non-LGNR spectra cross-referenced with algorithm results . 95% confidence intervals were calculated for each sensitivity and specificity measurement . ( b ) Summary of raw data ( pixel counts ) for each measurement reported in ( a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01110 . 7554/eLife . 16352 . 012Figure 2—figure supplement 2 . Dark-field images of additional uninjected H&E-stained tissue sections . ( a ) kidney , ( b ) liver , ( c ) lung , ( d ) muscle , and ( e ) spleen . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01210 . 7554/eLife . 16352 . 013Figure 2—figure supplement 3 . Hyperspectral images of additional uninjected H&E-stained tissue sections . ( a ) kidney , ( b ) liver , ( c ) lung , ( d ) muscle , and ( e ) spleen . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01310 . 7554/eLife . 16352 . 014Figure 2—figure supplement 4 . HSM-AD detection of additional uninjected H&E-stained tissue sections . ( a ) kidney , ( b ) liver , ( c ) lung , ( d ) muscle , and ( e ) spleen . Pixels identified as LGNR+ are denoted in orange . These analyzed samples were used for calculations of detection specificity . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01410 . 7554/eLife . 16352 . 015Figure 2—figure supplement 5 . Analysis of LGNRs on a glass slide . ( a ) Hyperspectral image of LGNRs embedded in CytoSeal on a glass slide . ( b ) a segmentation map showing background in blue & cyan , high-intensity LGNR- pixels in yellow , and LGNR+ pixels in red . ( c ) HSM-AD detection of the LGNR sample demonstrated high detection sensitivity . ( d ) Inspection of individual LGNR+ and LGNR- pixels validates the detection efficacy of HSM-AD . The split-peak spectrum of pixel 2 , which is not identified as LGNR+ , is possibly indicative of LGNR surface plasmon resonance hybridization . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01510 . 7554/eLife . 16352 . 016Figure 2—figure supplement 6 . Spectral hybridization in partially aggregated LGNRs . ( a ) A sample of as-synthesized LGNRs ( without additional surface functionalization ) was centrifuged and resuspended to produce a sample with partial aggregation and imaged in water with the hyperspectral dark-field microscope . ( b ) Spectra from various distances from the aggregate center . Spectra from pixels near the edges of particle aggregates displayed scattering peaks similar to disperse LGNRs , while pixels closer to the centers of aggregates exhibited both blue-shifting and multiple spectral peaks . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01610 . 7554/eLife . 16352 . 017Figure 2—figure supplement 7 . The influence of the local refractive index on the observed LGNR spectral peak . ( a ) LGNRs exhibited a spectral peak of ~800 nm when measured as a suspension in pure water with visible/near-infrared spectrometry . ( b ) Dark-field image of LGNRs in water . ( c ) The average spectrum of LGNRs in water has a peak that is slightly blue-shifted compared to the peak observed by spectrometry ( a ) . ( d ) The average spectrum of LGNRs in CytoSeal ( n = 1 . 5 , matched to microscope immersion oil ) produced a red-shift of ~80 nm in the average LGNR spectrum . ( e ) The spectrum of a single LGNR in water shows a similar peak to the average spectrum of LGNRs in water ( c ) . ( f ) The spectrum of a single LGNR in CytoSeal shows a similar peak to the average spectrum of LGNRs in CytoSeal ( d ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01710 . 7554/eLife . 16352 . 018Figure 2—figure supplement 8 . Evidence for single particle detection sensitivity with HSM-AD . ( a , b ) Hyperspectral dark-field image ( a ) and HSM-AD detection ( b ) of a tissue section which includes presumed single LGNRs , such as the point indicated by the green arrow . ( c ) A 2D plot of the scattering intensity around the LGNR intensity peak ( ~900 nm ) of pixels in the vicinity of the LGNR+ pixel shown in ( b ) . ( d , e ) 1D plots of normalized pixel intensity as a function of distance from the center pixel ( blue traces ) and theoretical intensity profiles of a point scatterer , calculated from a Gaussian point spread function ( red traces ) . The measured intensity plots correlate well with the theoretical intensities in both vertical ( d ) and horizontal ( e ) directions . Along with the retention of the LGNR spectral peak , this result suggests that the identified location likely contains a single LGNR . If more than one LGNR were present in the same area of one pixel , the spectrum would possibly change due to plasmonic hybridization and the pixel would not be detected as LGNR+ . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01810 . 7554/eLife . 16352 . 019Figure 2—figure supplement 9 . Dark-field images of additional LGNR-injected H&E-stained sections . ( a ) kidney , ( b ) liver , ( c ) lung , ( d ) muscle , and ( e ) spleen . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 01910 . 7554/eLife . 16352 . 020Figure 2—figure supplement 10 . Hyperspectral images of additional LGNR-injected H&E-stained sections . ( a ) kidney , ( b ) liver , ( c ) lung , ( d ) muscle , and ( e ) spleen . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 02010 . 7554/eLife . 16352 . 021Figure 2—figure supplement 11 . HSM-AD detection of additional LGNR-injected H&E-stained sections . ( a ) kidney , ( b ) liver , ( c ) lung , ( d ) muscle , and ( e ) spleen . LGNR+ pixels are depicted in orange . Along with those in the main figures , these analyzed samples were used for calculations of detection sensitivity and specificity as well as whole-organ LGNR uptake . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 02110 . 7554/eLife . 16352 . 022Figure 2—figure supplement 12 . Quantitative whole-organ biodistribution measured with HSM-AD on unstained tissue sections . All values are presented as the average relative LGNR signal ( % ) ± standard error of the mean ( s . e . m ) measured over four fields of view per organ . The results from unstained tissue sections are comparable to the results obtained for H&E stained sections . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 02210 . 7554/eLife . 16352 . 023Figure 2—figure supplement 13 . Dark-field images of LGNR-injected unstained tissue sections . ( a ) kidney , ( b ) liver , ( c ) lung , ( d ) muscle , and ( e ) spleen . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 02310 . 7554/eLife . 16352 . 024Figure 2—figure supplement 14 . HSM-AD detection of LGNR-injected unstained tissue sections . ( a ) kidney , ( b ) liver , ( c ) lung , ( d ) muscle , and ( e ) spleen . LGNR+ pixels are depicted in orange . These samples were used to calculate the values of LGNR uptake in unstained sections . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 02410 . 7554/eLife . 16352 . 025Figure 2—figure supplement 15 . Spectral Angle Mapper ( SAM ) detection of LGNR+ kidney tissue . ( a ) Results of SAM classification of LGNR+ pixels ( red masks ) with various user-defined angular tolerance values ( in radians , bottom-left of each panel ) . Selection of low angular tolerance results in poor detection sensitivity , while high tolerance leads to poor detection specificity . Along with other parameters , angular tolerance must be user-defined for each hyperspectral image . SAM classification was performed as described in reference 30 . ( b ) Guide image corresponding to the SAM-classified masks in ( a ) . ( c ) HSM-AD analysis of the same hyperspectral image . Diagnostic evaluation of this and related images yielded the sensitivity and specificity values reported in Figure 2—figure supplement 1 . As noted in reference 21 , such values are not readily extracted using SAM and related methods . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 025 HSM-AD imaging of histological sections ( Figure 3a ) enabled sensitive LGNR detection with sub-cellular resolution over large fields of view ( Figure 3b–c ) , which afforded more detailed characterizations of NP uptake than those achieved by typical biodistribution methods . We used these advantages to investigate and quantify the sub-organ distribution of LGNRs . This analysis revealed well-defined patterns of LGNR uptake that appeared to be largely influenced by factors including particle size , innate immunological function , and waste-filtering anatomical structures . 10 . 7554/eLife . 16352 . 026Figure 3 . HSM-AD is capable of wide-field characterization of sub-organ distribution patterns of injected nanoparticles . ( a–c ) Millimeter-scale fields of view of histological sections of kidney . Photographed ( a ) , acquired with near diffraction-limited resolution with a hyperspectral dark-field camera ( b ) , and analyzed by HSM-AD ( c ) to reveal variable nanoparticle uptake within the fine anatomical structures . ( d–f ) As in conventional histology , micro-anatomical features of the kidney including glomeruli , Bowman’s spaces , proximal convoluted tubule ( PCT ) , and distal convoluted tubule ( DCT ) networks can be clearly identified in HSM-AD images ( d ) . The ability to distinguish such histological details enables region of interest ( ROI ) analysis to quantify sub-organ accumulation of LGNRs ( e , f ) . Quantification of the relative LGNR signal in glomeruli ( red ROI ) , PCT ( yellow ROI ) , and DCT ( blue ROI ) regions ( e ) revealed that the vast majority ( ~13-fold greater than in either tubule network ) of renal LGNR uptake is localized within glomeruli ( f ) . This is likely due to the size-dependent inability of LGNRs to traverse the ultrafiltration barrier formed by endothelial cells within glomerular capillaries . All quantitative data are represented as mean ± s . e . m . for each ROI type , as calculated from 4 unique fields of view acquired at 40x magnification . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 02610 . 7554/eLife . 16352 . 027Figure 3—source data 1 . Data for kidney sub-organ ROIs . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 027 The kidneys are responsible for filtering small , low molecular weight waste products from the bloodstream and diverting those products to the bladder for elimination ( Rouiller , 2014 ) . Waste-laden blood flows into capillary-dense structures called glomeruli within the kidney . Blood plasma containing small species including ions , biomolecules , cell fragments , and ( in some cases ) nanoparticles can extravasate from glomerular capillaries and traverse Bowman’s space before being collected into an extensive network of efferent tubes called Proximal Convoluted Tubules ( PCT ) and , further downstream , Distal Convoluted Tubules ( DCT ) , which ultimately traffic waste from the kidney out to the bladder . Using our method , we observed that the vast majority of LGNRs within the kidney were concentrated in glomeruli and were virtually absent in the PCT and DCT , which are functionally downstream ( Figure 3d–f ) . Glomerular uptake was 13-fold greater than uptake within the convoluted tubule network . These results can be explained by comparing the size of an individual LGNR with the narrow width of cellular junctions and the architecture of endothelial cells that form glomerular capillaries ( Satchell and Braet , 2009 ) . For reference , the sub-organ segmentation maps of all tissue images analyzed for quantification have been provided ( Figure 4—figure supplement 1 ) . Along with the kidneys , the liver is instrumental in clearing waste from circulation . Partially because the size cutoff for hepatic filtration is notably larger than that of renal ultrafiltration ( Longmire et al . , 2008 ) , we observed 1 . 6-fold greater LGNR uptake within the liver tissue compared to the kidney . Interestingly , the hepatic distribution of LGNRs also appeared to be non-uniform . Hepatocytes , which constitute the majority of liver tissue by mass , exhibited mild LGNR signal ( 2 . 9% ± 1 . 1% ) . By contrast , 15-fold more LGNR signal ( 43 . 5% ± 12 . 0% ) in the liver was localized in a manner consistent with the shape , size , and number of Kupffer cells ( Figure 4a , Figure 2—figure supplements 9b , 10b , 11b ) . We attribute this localization to the phagocytic function of Kupffer cells ( Owens III and Peppas , 2006; Longmire et al . , 2008 ) . Thus , the variable localization of nanoparticles within the liver appears to be largely derived from the organ’s innate immunological functions . It is interesting to note that aggregation within Kupffer cells likely caused spectral hybridization of some LGNRs , an observation that is consistent with previous studies of cellular uptake of gold NPs ( Chen , 2014 ) . This spectral shifting , which was most prevalent at the centers of LGNR aggregates , caused a portion of LGNRs to remain undetected by HSM-AD . While these aggregates were undetected by algorithmic means , their manual identification as LGNRs was evident from the lack of similar morphological features in uninjected liver tissue . 10 . 7554/eLife . 16352 . 028Figure 4 . HSM-AD reveals characteristic patterns of nanoparticle microbiodistribution contingent upon tissue function . ( a ) LGNR accumulation in hepatic tissue occurs mostly in concentrated foci located within liver sinusoids . Along with the size , shape , and frequency of these foci , this pattern strongly suggests that these particles have been phagocytosed by Kupffer cells , the resident macrophages of the liver ( red ROI ) . While there is mild uptake of LGNRs within liver hepatocytes ( blue ROI ) , HSM-AD sub-organ quantification indicates that uptake by Kupffer cells is roughly 15-fold higher than hepatocytic accumulation . ( b ) The pattern of LGNR uptake in the spleen is also consistent with the physiological functions of various splenic tissues . A greater relative LGNR signal was observed in regions of splenic red pulp ( red ROI ) , which is responsible for blood filtration , than in the white pulp follicles ( blue ROI ) that house B and T lymphocytes . ( c , d ) While LGNRs were prevalent within the liver and spleen tissues , HSM-AD results indicated minimal particle accumulation within the lung ( c ) or muscle ( d ) tissue samples ( each < 1% relative LGNR signal for whole-tissue quantification ) . Interestingly , HSM-AD analysis demonstrated that the vast majority of LGNRs in muscle tissue sections were localized in blood vessels ( red ROI ) rather than within the muscle fiber tissue itself ( blue ROI ) . Quantitative data are represented as mean ± s . e . m . as described previously . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 02810 . 7554/eLife . 16352 . 029Figure 4—source data 1 . Data for liver , spleen , lung , and muscle sub-organ ROIs . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 02910 . 7554/eLife . 16352 . 030Figure 4—figure supplement 1 . Sub-organ region of interest ( ROI ) segmentation for additional tissue sections used for quantitative results presented in Figures 3 and 4 of the main text . The ROI color schemes for each sub-organ feature are identical to those listed in the legends of the relevant figures in the main text . ( a ) kidney , ( b ) liver , ( c ) lung , ( d ) muscle , and ( e ) spleen . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 03010 . 7554/eLife . 16352 . 031Figure 4—figure supplement 2 . Detail of Figure 4b: spleen tissue histology correlated with LGNR uptake . Using HSM-AD , it is possible to cross-reference nanoparticle uptake patterns and tissue microstructures with sub-cellular resolution . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 031 The spleen comprises several unique cell types arranged into tissues with diverse biological functions including blood filtration , innate immunity , and lymphocyte activation ( Mebius and Kraal , 2005 ) . The spleen is largely composed of red pulp , white pulp , and the boundary between these two tissues , commonly referred to as the marginal zone . Consistent with each tissue’s biological function , we observed 1 . 7-fold greater relative LGNR pixel coverage in splenic red pulp than white pulp ( Figure 4b , Figure 2—figure supplements 9e , 10e , 11e ) . A similar result has been previously reported for carbon-based nanomaterials ( Chen et al . , 2015 ) . We did not definitively identify marginal zone tissue , but the radial distribution of LGNRs around white pulp follicles indicated that a significant portion of white pulp uptake may in fact be within the marginal zone ( Figure 4—figure supplement 2 ) . Despite its dense network of alveolar capillaries , lung tissue exhibited minimal LGNR accumulation relative to the organs described above ( Figure 4c , Figure 2—figure supplement 9c , 10c , 11c ) . This finding was consistent with existing biodistribution data for smaller particles , which can be explained by the lungs’ major functions of gas exchange to and from the blood rather than biomolecule or particle filtration and clearance . While whole-organ analysis indicated the presence of LGNRs in muscle tissue , HSM-AD revealed that muscle tissue itself ( which consists largely of myocytes and dense networks of extracellular collagen ) was virtually devoid of LGNRs ( Figure 4d , Figure 2—figure supplement 9d , 10d , 11d ) . Rather , the apparently high LGNR presence was traced to blood vessels found in between muscle fiber bundles . As with the accumulation patterns described for other organs within this study , this distinction would not have been possible through conventional biodistribution methods . HSM-AD images acquired at higher objective magnification ( 100x ) offered further insights into the cellular nature of LGNR uptake within the kidney and liver tissue ( Figure 5 ) . Within the kidney , LGNRs were observed mostly within or in close proximity to glomerular capillaries ( Figure 5a , b ) . HSM-AD also revealed patterns of LGNR uptake within individual Kupffer cells resident in liver sinusoids ( Figure 5c , d ) . LGNR signal was detected within the Kupffer cell cytoplasm , but not within the region of the cell nucleus . This pattern is consistent with the phagocytic function of Kupffer cells in clearing particulate matter from circulation . Interestingly , several bright regions within the Kupffer cell were not detected as LGNRs . We expect that these regions resulted from spectral hybridization of LGNRs , possibly due to aggregation induced by lysosomal acidification following particle phagocytosis . 10 . 7554/eLife . 16352 . 032Figure 5 . HSM-AD reveals the sub-cellular localization of intravenously administered nanoparticles with histological precision . ( a , b ) Hyperspectral ( a ) and HSM-AD ( b ) images of a renal glomerulus acquired at 100x magnification . A majority of LGNRS are found within or in close proximity to glomerular capillaries . Trace levels of LGNRs are observed in the kidney tissue outside of Bowman's capsule . ( c , d ) Zoomed views of Hyperspectral ( c ) and HSM-AD ( d ) images of liver tissue acquired at 100x magnification . Several erythrocytes and a Kupffer cell ( dashed white line ) can be observed residing within a liver sinusoidal vessel . Within the Kupffer cell , the nucleus ( dashed red line ) can be distinguished . HSM-AD analysis indicated the prevalence of LGNRs within the Kupffer cell relative to surrounding hepatocytes . The minimal LGNR signal was detected in the region identified as the nucleus , consistent with cytoplasmic LGNR localization . Several bright regions within the cell were not identified as LGNRs; these regions likely result from particle aggregation within acidic lysosomes following uptake by the Kupffer cell . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 032 We also injected mice with Nanoshells , which are morphologically distinct from LGNRs ( Figure 6a ) . While Nanoshells and LGNRs both exhibit near-infrared plasmonic peaks , the Nanoshell spectrum is substantially broader than the LGNR spectrum ( Figure 6b ) . A spectral cluster library was developed for H&E-stained Nanoshell+ tissues and was then used to quantify Nanoshell uptake as described for LGNRs ( Figure 6c ) . While Nanoshells and LGNR displayed related uptake patterns , several differences including negligible Nanoshell uptake in kidney tissue and Nanoshell concentration within the splenic white pulp were observed ( Figure 6d , Figure 6—figure supplements 1 , 2 , Nanoshell+ pixels are shown in cyan ) . 10 . 7554/eLife . 16352 . 033Figure 6 . Characterization of gold nanoshell uptake after intravenous administration . ( a , b ) Nanoshells ( 119 nm silica core with 14 nm-thick gold coating ) exhibit distinct particle morphology and composition ( a ) that ( like LGNRs ) yield a near-infrared ( ~800 nm ) spectral peak ( b ) . However , the Nanoshell spectrum is markedly broader than the resonance observed for LGNRs . ( c , d ) HSM-AD revealed that Nanoshell uptake displays inter-organ distribution patterns somewhat similar to those observed for LGNRs , with maximal accumulation in the spleen ( c ) . Values are represented as mean ± s . e . m . from four FOVs per tissue . However , there are notable distinctions including minimal Nanoshell uptake within kidney tissue and concentration of Nanoshells within splenic white pulp ( d ) ( Nanoshell+ pixels are depicted in cyan ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 03310 . 7554/eLife . 16352 . 034Figure 6—source data 1 . Data for Nanoshell uptake in organs . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 03410 . 7554/eLife . 16352 . 035Figure 6—figure supplement 1 . Additional HSM-AD images of Nanoshell uptake used for quantification . Quantitative data from these FOVs were used to produce the bar graph in Figure 6c . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 03510 . 7554/eLife . 16352 . 036Figure 6—figure supplement 2 . Detail of Nanoshell uptake in spleen tissue . ( a ) Conventional dark-field image of H&E-stained spleen tissue . The dashed white line approximately demarcates the marginal zone separating the red and white pulp . ( b ) HSM-AD indicates that , unlike LGNRs , Nanoshells are localized within white pulp follicles 24 hr after intravenous injection . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 03610 . 7554/eLife . 16352 . 037Figure 6—figure supplement 3 . HSM-AD spectral unmixing of samples containing gold nanoshells and LGNRs . HSM-AD was trained on a hyperspectral image of a sample containing a mixture of Nanoshells and LGNRs . Using a target of two clusters , this training yielded one spectral cluster corresponding to the spectrum of the Nanoshells and another spectral cluster corresponding to LGNRs ( far-left column ) . These clusters were then used to map images of Nanoshells + LGNRs , Nanoshells only , and LGNRs only . The presence of Nanoshells and LGNRs are marked using cyan and orange masks , respectively . HSM-AD classification using the Nanoshell cluster ( top row ) for all three sample types achieved 96 . 68% sensitivity and 99 . 16% specificity . HSM-AD classification using the LGNR cluster ( middle row ) for the samples achieved 99 . 16% specificity and 96 . 68% specificity . The bottom row depicts the merge of these two cluster maps . As a note , the reciprocal nature of the sensitivity and specificity values for the two different particle types results from the fact that , for pure particle solutions , the false positives for one particle type are false negatives for the other particle type and vice versa . The same reciprocal nature holds for true positives and true negatives as well . This relationship can be seen in the raw pixel counts used for diagnostic evaluation ( far-right column ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 037 HSM-AD was separately trained on a sample consisting of a mixture of pure Nanoshells and pure LGNRs . The spectral clusters identified during this training corresponded well to the spectra of each particle type and enabled high-specificity and high-sensitivity identification in samples of Nanoshells + LGNRs , Nanoshells-only , and LGNRs-only ( Figure 6—figure supplement 3 ) . These results demonstrate that HSM-AD can spectrally resolve plasmonic particles despite similarities in composition , although this capability was not tested in ex vivo tissues . We used HSM-AD to characterize the tissue uptake of a third particle type , GNS@SiO2 ( Figure 7 , GNS@SiO2+ pixels are shown in green ) . Notably , GNS@SiO2 are distinct from LGNRs and Nanoshells in terms of shape , size , composition , particle surface , and plasmonic resonance ( Figure 7—figure supplement 1a , b ) . Because GNS@SiO2 exhibit a visible regime plasmonic peak ( ~550 nm ) , HSM-AD analysis was performed on unstained tissue sections . First , a spectral cluster library was developed for GNS@SiO2 classification as described for LGNRs ( Figure 7—figure supplement 1c ) . Control tissues classified with this library displayed negligible false positives ( Figure 7a ) . GNS@SiO2 uptake in the liver and spleen was observed at 2 and 24 hr post-IV injection ( Figure 7b , c ) . Interestingly , GNS@SiO2 uptake appeared to be even more localized to Kupffer cells than LGNR accumulation in the liver . Furthermore , GNS@SiO2 in the spleen are consistently found in the marginal zone , and presence within the red pulp and white pulp is minimal ( Figure 7—figure supplement 2 ) . Quantitative results from HSM-AD correlate well with those obtained using ICP-MS ( Figure 7—figure supplements 3 , 4 ) , although it should be noted that HSM-AD measurements are more relative rather than absolute with respect to the amount of gold present in each tissue . As for LGNR quantification , four FOVs per sample were analyzed ( Figure 7—figure supplements 5 , 6 ) . 10 . 7554/eLife . 16352 . 038Figure 7 . HSM-AD analysis of GNS@SiO2 . ( a ) Classification of unstained control tissues yields negligible false-positive detection for GNS@SiO2 ( green ) , which exhibit a peak plasmonic resonance of ~550 nm . ( b , c ) Intravenously-administered GNS@SiO2 accumulate in the Kupffer cells of the liver and the marginal zone of the spleen within 2 hr ( b ) and persist up to 24 hr ( c ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 03810 . 7554/eLife . 16352 . 039Figure 7—source data 1 . Data for GNS@SiO2 uptake in organs . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 03910 . 7554/eLife . 16352 . 040Figure 7—figure supplement 1 . Structural and spectral characterization of GNS@SiO2 . ( a ) TEM of GNS@SiO2 . ( b ) Vis-NIR absorbance spectrum of GNS@SiO2 in water ( SPR ~550 nm ) . ( c ) Spectral library clusters identified from training on images of unstained tissues ( in CytoSeal ) resected from mice injected with GNS@SiO2 . The target number of clusters was set to three rather than five ( used for LGNR detection ) due to the absence of H&E staining . Cluster 2 ( green ) corresponds to the GNS@SiO2 and exhibits a red-shifted plasmonic peak relative to particles in water , as expected due to the difference in refractive environment . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 04010 . 7554/eLife . 16352 . 041Figure 7—figure supplement 2 . Detail of GNS@SiO2 in spleen . HSM-AD reveals that GNS@SiO2 accumulate mostly in the marginal zone tissue between red and white pulp . The approximate boundaries between red pulp and white pulp follicles are marked by dashed white lines . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 04110 . 7554/eLife . 16352 . 042Figure 7—figure supplement 3 . HSM-AD quantification of GNS@SiO2 uptake in liver and spleen tissue . Relative GNS@SiO2 uptake was measured using the positive ratio approach described in the methods section . All values represent the mean GNS@SiO2 uptake from 4 FOVs for each injection/tissue combination . Error bars represent standard error of the mean ( s . e . m . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 04210 . 7554/eLife . 16352 . 043Figure 7—figure supplement 4 . Inductively-Coupled Plasma Mass Spectrometry ( ICP-MS ) quantification of GNS@SiO2 uptake in liver and spleen tissue . Quantitative measurements of atomic gold present in tissue samples prepared through microwave digestion . Counts are given in parts per billion ( ppb ) of Au . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 04310 . 7554/eLife . 16352 . 044Figure 7—figure supplement 5 . Additional HSM-AD images of GNS@SiO2 uptake in liver tissue . Quantitative data from these FOVs were used to produce the bar graph in Figure 7—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 04410 . 7554/eLife . 16352 . 045Figure 7—figure supplement 6 . Additional HSM-AD images of GNS@SiO2 uptake in spleen tissue . Quantitative data from these FOVs were used to produce the bar graph in Figure 7—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 045 One hallmark of tumor growth is angiogenesis , the stimulated development of new blood vessels to provide nutrients to rapidly dividing cancer cells . This newly-formed vasculature is composed of endothelial cells that express high levels of cell adhesion receptors including αVβ3 integrin ( Avraamides et al . , 2008 ) . Thus , αVβ3 is commonly used as a target biomolecule for tumor imaging ( Sipkins et al . , 1998 ) . Such studies have demonstrated that NPs targeted to αVβ3 exhibit greater accumulation in tumors in vivo than NPs coated with non-specific antibodies or small molecules . We hypothesized that the presence or absence of specific molecular targeting moieties would influence tissue-NP interactions beyond simply the degree of accumulation in target tissues . To test this , we used HSM-AD to observe the spatial patterns of targeted and non-targeted LGNR uptake within U87MG ( human glioblastoma cells , αVβ3+ ) tumor xenografts . We observed 7 . 4-fold greater relative LGNR signal of anti-αVβ3 LGNRs than isotype LGNRs in tumor tissue ( Figure 8a–d ) . However , the most striking differences were in the localization patterns of each LGNR type . Anti-αVβ3 LGNRs were present in high density around the edges of small blood vessels within the tumor while isotype LGNRs showed no such association ( Figure 8c–f , Figure 8—figure supplement 1 ) . The prevalence of anti-αVβ3 LGNRs around the edges of tumor capillaries is highly consistent with the expression pattern of αVβ3 in angiogenic vessels . Moreover , isotype LGNRs found outside of the vasculature were notably dispersed compared to extravascular anti-αVβ3 LGNRs , which often appeared in small clusters . While NPs are known to accumulate in tumors regardless of molecular specificity due to leaky vasculature , these results indicated that the enhanced extravascular accumulation of anti-αVβ3 LGNRs may have originated from specific binding of αVβ3 integrins present on the U87MG cells themselves . 10 . 7554/eLife . 16352 . 046Figure 8 . Active molecular functionalization affects nanoparticle uptake quantitatively and spatially within target tissues . ( a , b ) HSM-AD images of sub-dermal U87MG tumor xenografts from mice injected with LGNRs display distinct accumulation patterns depending on the molecular specificity of the LGNR surface coating . Anti-αVβ3 LGNRs exhibit 7 . 5-fold greater accumulation in tumor tissue ( a ) than spectrally-identical LGNRs with non-specific IgG antibody coating ( b ) ( n = 4 FOVs for Anti-αVβ3 LGNRs , n = 5 FOVs for IgG-LGNRs , two-tailed Student’s t-test , p=0 . 0041 ) . The greater uptake of anti-αVβ3 LGNRs may result in part from specific LGNR binding to αVβ3 integrin , which is over-expressed by U87MG cells . ( c–f ) Validation of HSM-AD images with dark-field images of slightly higher spatial resolution further indicates that a substantial portion of anti-αVβ3 LGNRs are located along the edges of small capillaries within the tumor tissue ( c , e ) while no such association is observed for IgG-LGNRs ( d , f ) . This observation is consistent with the nature of angiogenic tumor vasculature , which is also characterized by high expression levels of αVβ3 integrin in the vascular endothelium . Individual erythrocytes within angiogenic capillaries are denoted by white arrows , and capillary edges are approximately outlined by dashed blue ovals ( e , f ) . Discrete regions of anti-αVβ3 LGNRs were also observed outside of the tumor vasculature , presumably due to either ( 1 ) specific binding to αVβ3integrin expressed by U87MG cells and/or ( 2 ) non-specific accumulation via the enhanced permeability and retention ( EPR ) effect characteristic of tumors . The absence of IgG-LGNR extravascular accumulation suggests the former of these mechanisms as the predominant source of anti-αVβ3 LGNR uptake in tumor tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 04610 . 7554/eLife . 16352 . 047Figure 8—source data 1 . Data for tumor uptake of targeted and untargeted LGNRs . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 04710 . 7554/eLife . 16352 . 048Figure 8—figure supplement 1 . HSM-AD images of additional tumor tissue sections resected after targeted and untargeted LGNR injections . LGNR uptake in U87MG tumors is consistently greater and more localized to blood vessel endothelial cells ( αVβ3+ ) when particles are conjugated with anti-αVβ3 antibodies ( top row ) rather than nonspecific IgG antibodies ( bottom row ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16352 . 048
The necessity of sample digestion with strong acids for ICP quantification effectively reduces an entire organ ( a remarkably rich dataset by any measure ) down to a single number representative of bulk NP accumulation . While the quantification offered by ICP is certainly valuable , it provides minimal insight into the patterns and mechanisms of NP uptake within individual cells or tissues . Unlike ICP methods , HSM-AD provides additional dimensions of anatomical detail at optical resolution to facilitate better understanding of the biology behind quantitative measurements of NP uptake . The primary solution for dealing with the limitations of ICP has been to use EM , which provides excellent spatial resolution ( at the nanometer scale ) and particle sensitivity ( down to individual nanoparticles ) . However , EM can only scan minimal fields of view—a typical transmission EM ( TEM ) image for studying NP uptake covers ~1 × 1 µm . For comparison , TEM scanning of the same area depicted in Figure 3c would require ~460 , 000 TEM images , which is infeasible for single tissue studies and virtually unrealistic for multiple-organ studies . The necessity of thin samples ( ~10 nm ) for TEM imaging compared to samples analyzed using HSM-AD ( ~1 µm optical focus ) would further multiply the number of TEM scans ( >46 million ) required for equivalent volumetric imaging . Other biodistribution techniques based on radioactivity ( Kreyling et al . , 2015; Collingridge et al . , 2003 ) , photoacoustic ( Poon et al . , 2015 ) , and fluorescence ( He et al . , 2010 ) detection have been used previously as alternatives to ICP and TEM . By comparison , HSM-AD offers roughly 100-fold higher spatial resolution ( ~1 µm vs ~100 µm ) than current fluorescence and photoacoustic biodistribution methods . Fluorescence-based methods may also suffer from high false positive detection arising from tissue autofluorescence , as has been observed for renal capsule tissue ( Poon et al . , 2015 ) . While HSM-AD was excellently suited for exploring the sub-organ localization of NPs , it has been observed that radiolabeling approaches may be poorly-equipped for accurately determining particle distribution within organs ( Kreyling et al . , 2015 ) . Moreover , single-particle detection sensitivity was not demonstrated by any of these alternatives to ICP and TEM . Biodistribution methods based on imaging mass spectrometry were recently demonstrated to enable sub-organ quantification of carbon nanomaterials ( Chen et al . , 2015 ) . This impressive approach can obtain images of full mouse tissue sections ( cm in scale ) , but the limited spatial resolution ( 50 µm ) precludes the study of NP uptake within individual cells . Because detection relies upon particle fragmentation and ionization , it is unclear whether imaging mass spectrometry can achieve single particle sensitivity—calculations based upon the reported data indicate that ~10 ( von Maltzahn et al . , 2009 ) particles per pixel are required for detection . However , the cited spatial resolution negates many of the potential advantages of single-particle sensitivity such as direct observation of NP uptake by cells through endocytosis or adhesion to the cell membrane . More generally , while mass spectrometry provides an approach to biodistribution studies of certain materials , its use for identifying gold NPs has been constrained to NPs smaller than 10 nm and typically requires the inclusion of 'mass barcode' molecules as capping agents ( such as alkanethiols ) on the NP surface ( Zhu et al . , 2008; Harkness et al . , 2010 ) . Incidentally , gold NPs have previously been demonstrated as assisting matrices to improve mass spectrometry detection of biomolecules ( Su and Tseng , 2007; Huang and Chang , 2007 ) . Notably , many metallic NPs are compositionally similar yet spectrally distinct ( for example , gold nanospheres , nanorods , nanoshells , etc . ) , which may confound results in mass spectrometry-based analysis of samples containing more than one NP species . We have demonstrated that HSM-AD can successfully identify gold nanoparticles with different spectra , shape , size , and composition in tissues . Furthermore , Nanoshells and LGNRs were discernible from each other in particle mixtures . While not directly tested in this work , HSM-AD may thus be capable of distinguishing such NPs from each other in tissues , enabling biodistribution studies of multiplexed NPs . Thus , we expect that HSM-AD will extend the advantages of high sensitivity and resolution to the analysis of a variety of metallic NPs with unique spectral properties . As reported in previous studies ( Chen , 2014; Okamoto et al . , 2000 ) , we observed that the scattering spectrum from gold NPs is heavily influenced by the local refractive index ( n ) . Spectral shifts of ~80 nm were evident between preparations of LGNRs in water ( n = 1 ) relative to the same particles prepared in CytoSeal ( n = 1 . 5 ) ( Figure 2—figure supplement 7 ) , and similar shifts were observed for Nanoshells and GNS@SiO2 . Thus , in order to generate reliable spectral cluster libraries , it is critical to train the adaptive algorithm using images prepared in similar fashion to the samples being studied . The orientation of anisotropic NPs within a sample has also been shown to influence the observed spectrum ( Biswas et al . , 2012 ) . Because LGNRs are anisotropic , variation in particle orientation may affect detection sensitivity in our case , but this effect appears to be negligible in light of empirically measured sensitivity values . Because it relies on spectral identification of NPs , HSM-AD cannot detect NPs that have shifted their spectrum markedly , such as concentrated NP aggregation within cells . Such spectral hybridization is most prominent in Kupffer cells within the liver , which likely results in an artificially low measure of LGNR uptake in that organ . However , the absence of these bright aggregates in hyperspectral images of uninjected control tissues confirms their identity as LGNRs . Because of plasmon hybridization , LGNR aggregates produce shifted spectra that can resemble the scattering from H&E-stained tissue , which can impede automated detection . Thus , future efforts to detect such aggregates should rely on the analysis of unstained tissues . We tested alternate machine learning approaches including support vector machine ( SVM ) and logistic regression for nanoparticle detection . We found that unsupervised k-means outperformed these other methods when trained on images of tissues containing nanoparticles ( Liba and Shaviv , 2014 ) . Further advantages of unsupervised k-means include its ease of use and no need to pre-label samples for analysis . While HSM-AD based on k-means clustering provides a robust general platform for nanoparticle detection , it is conceivable that certain studies may benefit from learning methods tailored to address specific applications . HSM-AD imaging simultaneously achieves excellent sensitivity and specificity for detecting NPs in tissues with sub-cellular resolution . In addition to improved diagnostic capabilities , the automated and adaptive features of HSM-AD enable standardized high-throughput analysis previously absent from biomedical HSM studies ( Roth et al . , 2015 ) . Unlike Spectral Angle Mapping ( the current gold standard for HSM image analysis ) , HSM-AD does not require the manual steps typically needed to create target spectral libraries , define particle intensity and size thresholds , filter false positives from libraries , and calibrate angular tolerance for accurate classification on an individual image basis ( Figure 2—figure supplement 15 ) ( Roth et al . , 2015 ) . Along with an ability to image millimeter-scale fields of view on reasonable timescales ( <30 min ) and simple sample preparation , these properties make HSM-AD a favorable alternative to existing methods for characterizing NP biodistribution . Beyond biodistribution , this work demonstrates that HSM-AD can be used for post-injection validation of NP localization in target tissues as a function of surface modifications . Because HSM-AD is non-destructive , samples can be further analyzed by a variety of conventional microscopy techniques including immunohistochemistry to provide additional molecular detail . Collectively , the results presented herein indicate that HSM-AD provides a new approach for studying interactions of cells and whole tissues with spectrally unique NPs commonly used in biomedical imaging and therapeutic studies .
LGNRs were synthesized using methods adapted from Ye et al ( Ye et al . , 2013 ) . LGNRs were characterized using Transmission Electron Microscopy ( TEM ) , visible/near-infrared spectrometry , and dark-field hyperspectral microscopy . As-synthesized LGNRs were prepared for biological use by removing excess CTAB from solution and coating the particles with poly ( sodium 4-styrenesulfonate ) ( PSS , MW 70 kDa ) as previously reported ( SoRelle et al . , 2015 ) . PSS-coated LGNRs were then conjugated with IgG isotype antibody ( clone eB149/10H5 , eBioscience ) for use in sub-organ biodistribution experiments . We also prepared LGNRs targeted to αVβ3 integrin ( a cell-surface receptor that is overexpressed in angiogenic vasculature within tumors ) by conjugating LGNRs with anti-αVβ3 antibody ( clone 23C6 , eBioscience ) . Healthy female nude ( Foxn1nu/nu ) mice ( 6–8 weeks old , Charles River Labs ) were anesthetized with 2% isoflurane by inhalation and intravenously injected with 250 µL of IgG isotype-coated LGNRs at optical density ( OD ) 470 . In separate experiments , mice bearing U87MG tumors in the right ear pinna were injected with either IgG isotype-coated LGNRs or anti-αVβ3-coated LGNRs . Additional details of these experimental protocols can be found in the literature ( SoRelle et al . , 2015; Liba et al . , 2016 ) . In nanoshell experiments , nude mice were injected with 200 µL of OD 50 ( 2 . 5 mg/mL ) nanoshells composed of 119 nm silica cores and 14 nm-thick gold shells with PEG coating ( Nano Composix , San Diego , CA ) . In GNS@SiO2 experiments , healthy female Balb/C mice were anesthetized as described previously and injected intravenously with 150 µL of 0 . 8 nM GNS@SiO2 particles composed of 60 nm gold cores and 30 nm-thick SiO2 shells ( Oxonica , Mountain View , CA ) . For all experiments , mice were euthanized 24 hr ( or 2 hr , for GNS@SiO2 ) post-injection , and tissues including kidney , liver , lung , spleen , thigh muscle , and ( when applicable ) tumor were immediately resected and preserved in 10% formalin . Tissues were also resected from uninjected mice for control imaging experiments . These tissues were subsequently embedded in paraffin and sectioned into 5 µm thick samples . Sections were prepared with and without H&E stains and mounted on microscope slides using CytoSeal 60 ( Electron Microscopy Sciences ) as the mounting medium . All animal experiments were performed in compliance with IACUC guidelines and with the Stanford University Animal Studies Committee’s Guidelines for the Care and Use of Research Animals . Experimental protocols ( APLAC #s 27499 and 29179 ) were approved by Stanford University’s Animal Studies Committee . All tissue samples were imaged with a modified dark-field microscopy setup as shown in Figure 1—figure supplement 1 . Light from a broadband halogen lamp was coupled via an optical fiber into a custom dark-field condenser ( CytoViva , Auburn , AL ) , which produced a light cone for sample illumination . Light scattered from the sample was collected using either a 40x magnification dark-field air objective lens ( Olympus UPlanFLN 40x , 0 . 75 NA ) or a 100x magnification oil immersion objective lens ( Olympus UPlanFLN 100x , 1 . 3 NA ) and directed to one of two cameras depending on detection mode . Conventional dark-field and hyperspectral images were collected for all samples in this study . Conventional dark-field images were collected with a Dagexcel-M cooled camera ( Dage-MTI , Michigan City , IN ) . Hyperspectral images were collected with a hyperspectral camera ( iXon3 , Andor , Belfast , UK ) . Each image has 509 × 512 pixels . With a 40x lens , the sampling resolution is 410 × 408 nm , which produces a 209 × 209 µm field of view . With a 100x lens , the sampling resolution is 163 × 160 nm . Only Figure 5 shows images acquired at 100x magnification . The spectrum from each pixel was acquired with 361 uniform samples at wavelengths ranging from 400 nm to 1000 nm . The acquired raw spectra of each pixel were lamp-normalized using the Cytoviva software package ( ENVI 4 . 8 ) and exported after normalization . Processing of the hyperspectral images was done with Matlab ( Mathworks , Natick , MA ) . Hyperspectral images were created by color-coding the spectrum by integrating over three bands . The band centers were 800 . 0 nm , 700 . 6 nm and 526 . 2 nm for the red , green and blue channels , respectively . The integration was weighted by a Gaussian window with a width of 80 spectrum samples . Each channel was scaled separately for optimal viewing . Automatic detection of NPs required preprocessing the spectra prior to training and classification . First , due to noise at lower wavelengths , the spectra were truncated to disregard values below a cutoff of 566 nm . As part of HSM-AD , we initially segmented the image into background , tissue , or potential NPs based on each pixel’s average intensity across its spectrum . This segmentation allowed a more accurate calculation of the biodistribution by measuring the number of pixels that correspond to tissue inside the field of view . The segmentation of potential NPs helped to avoid classifying low intensity edges that may be falsely detected as NPs due to chromatic aberrations . Before measuring the intensity of each pixel , we also applied a correction for vignetting . Vignetting is a common artifact in photography and microscopy in which image brightness is reduced at the periphery compared to the image center . We assumed a natural illumination fall-off that follows the 'cosine fourth' law , in which the light fall-off is proportional to the fourth power of the cosine of the angle at which the light impinges on the sensor . We measured the radial falloff of several images and found that it can be approximated by cos4 ( θ ) , in which θ=tan−1 ( R/d ) , where R is the calculated distance from the center of the image and d was found by fitting to be 2 mm . In order to correct the vignetting , we divided the intensities of each field of view by cos4 ( θ ) . Next , we calculated the segmentation thresholds adaptively for each image ( to account for different exposure times and variations in tissue scattering ) . The thresholds were obtained by analyzing the histogram of pixel intensities of each image ( Figure 1—figure supplement 2 ) . The histogram was calculated with 510 bins and then re-sampled ( using interpolation ) every 5 intensity units . Pixels with the lowest intensities were segmented as background . The threshold for segmenting the background was calculated as the first minimum of the histogram ( minHist ) multiplied by a user-defined parameter ( which is slightly larger than 1 ) to allow fine tuning of the background threshold . Next , we assumed that pixels representing tissue without NPs have a relatively consistent intensity lower than that of NPs and therefore correspond to a peak in the histogram . The threshold for differentiating between pixels that correspond to tissue and those that can be potential NPs can be determined by the peak of the histogram ( peakHist ) multiplied by a pre-defined parameter ( larger than 1 ) which allows tuning of the threshold . This intensity-based segmentation was confirmed to be effective by comparing results over 20 separate fields of view from all analyzed tissue types . Pixels segmented as potential NPs were preprocessed by smoothing their spectra using a Savitzky-Golay algorithm ( Orfanidis , 1995 ) implemented by a Matlab’s built-in function . Next , we normalized the spectrum of each pixel by the maximal intensity across its spectrum . Training and classification were performed only on pixels which were segmented as potential NPs . Training of the k-means algorithm ( Bishop , 1995 ) was initially done with 3 , 4 , and 5 clusters on 6 images of injected tissue sections , from which ~500 , 000 pixels were binned into the potential LGNR group . The 4 clusters found by the k-means algorithm matched the expected spectra of the injected and stained tissue . One of the spectra automatically matched the spectrum of LGNRs , owing to their distinct spectrum compared to tissue and staining dyes , two of the spectra represent the Hematoxylin and Eosin stains , and the fourth is an intermediate cluster representing the sum of both stains ( see Figure 1—figure supplement 4 ) . Testing the algorithm on uninjected tissue samples showed false detections of LGNRs near the edges of the tissue . We attributed these false positives to spectral red-shifting caused by chromatic aberrations due to the larger point spread function of longer wavelengths compared to the smaller point spread function of shorter wavelengths . To minimize the number of these false positives , we manually added a cluster representing the spectrum of the chromatic aberrations by averaging the spectra of falsely detected pixels from uninjected samples . Indeed , this cluster showed a spectral peak near the resonance of the LGNRs , albeit much broader . The initial clusters found by k-means with the added cluster representing chromatic aberrations were used for classification of pixels as LGNR+ or LGNR- with a nearest centroid ( or nearest neighbor ) classifier , based on the Euclidean distance to the cluster centers . We then measured the sensitivity and specificity and also qualitatively assessed the results on injected tissue section using the algorithm with different numbers of clusters . We chose to use the cluster library which includes 5 clusters ( 4 obtained through k-means and the 5th added manually ) because it produced the best results and clusters that were more consistent with the actual spectra present in the samples . Several other machine learning algorithms were also explored for this purpose , including support vector machine ( SVM ) and logistic regression , but k-means yielded better results ( Liba and Shaviv , 2014 ) . The classification results are presented as detection maps in which the average intensity at each pixel is displayed in grayscale and the pixels that are LGNR+ are shown in orange . Similar training and classification steps were performed for samples containing GNS@SiO2 ( 3 clusters used ) and Nanoshells ( 5 clusters for images of ex vivo tissues , 2 clusters for Nanoshells + LGNRs ) . We characterized the sensitivity and specificity of HSM-AD by three methods . First , we measured the false positive and true negative rates in uninjected tissue samples to obtain the specificity . Note that false positives and true negative rates can be measured reliably from tissue-only samples due to the absence of LGNRs . Next , we measured the false negatives and true positives in an image of pure LGNRs in mounting media on a glass slide and obtained the detection sensitivity . Note that false negatives and true positives can be measured reliably from LGNR-only samples due to the absence of tissue and scattering media other than the particles themselves . In order to calculate the sensitivity and specificity for detecting LGNRs in tissue samples , we created a 'ground truth' data set by randomly choosing > 200 pixels , half of which were detected by the algorithm to contain LGNRs . We then manually ( and in a manner blind to the results from the algorithmic classification ) determined whether pixels were LGNR+ or LGNR- based purely on observing their raw spectra and looking for the unique plasmonic peak of LGNRs . Low-intensity pixels that were considered as the background or tissue were not considered in this calculation . By comparing the ground truth to the results of the automated algorithm we obtained the number of true positives , true negatives , false positives and false negatives , and used them to calculate additional measures of sensitivity and specificity . Confidence intervals for these measurements , presented in Figure 2—figure supplement 1 , were calculated using the 'log method' ( Altman et al . , 2000 ) with the aid of a statistical calculator ( MedCalc Software , 2016 ) . The biodistribution in each field of view was measured as the relative LGNR signal in an image . This measurement takes into account the amount of tissue versus background in a frame and also the signal of the detected LGNRs . For this calculation , we refer to LGNR signal as the average intensity around the plasmonic peak of the LGNRs ( 833–988 nm ) . The relative LGNR signal is the sum of the LGNR signal over LGNR+ pixels divided by the number of tissue pixels ( i . e . , the number of all pixels minus the number of background pixels ) and divided by the median LGNR signal over the LGNR+ pixels in the field of view . For whole-organ analysis , we measured the relative LGNR signal in four fields of view for each organ ( taken from the same mouse ) and calculated the mean and standard deviation for each organ . For the sub-organ calculation , the measurement of relative LGNR signal yielded results with substantial variability due to a small number of pixels and high variability of intensity within several regions of interest . Therefore , a simpler pixel ratio ( termed pixel coverage ) was calculated by dividing the number of LGNR+ pixels by the number of tissue pixels for each region of interest ( ROI ) . ROI maps are presented for each field of view in Figure 4 as well as for the additional images used for quantification ( Figure 4—figure supplement 1 ) . The same calculations also apply for quantification of other particle types . In order to determine whether HSM-AD is able to detect single LGNRs , we compared the theoretical point spread function of the microscope’s 40x lens with the shape of the increased intensity caused by an isolated LGNR+ pixel . The diffraction-limited point spread function of the microscope can be approximated by a Gaussian with a standard deviation of 0 . 25 μm ( based on a numerical aperture of 0 . 75 and wavelength of 910 nm ) ( Lipson and Lipson , 2010 ) , which may increase in the case of defocusing . The cross section of intensity in the wavelengths around the plasmonic peak of the LGNRs ( 833–988 nm ) showed a close resemblance to the theoretical spot size ( Figure 2—figure supplement 8 ) . This analysis supports the capability to detect single LGNRs in tissue samples . | Metallic elements like gold and silver can be made into particles that are one thousand times smaller than the width of a human hair . Researchers can create these “nanoparticles” in different sizes and shapes that exhibit unique properties . For example , gold can be made into rod-shaped particles that interact with infrared light . Other nanoparticles can be loaded with drug molecules and designed to bind to cancer cells . As a result , nanoparticles have been explored for use in a variety of biomedical imaging and therapy applications . However , we must fully understand how the nanoparticles bind to the cancer cells and how the body tolerates these nanoparticles before they can be used in humans . Experiments that explore where nanoparticles accumulate in the body are typically called biodistribution studies . However , current techniques for studying biodistribution cannot simultaneously measure the uptake of particles into organs and reveal the fine structures inside the organs that interact with the particles . SoRelle , Liba et al . aimed to address this problem by developing a new biodistribution technique called HSM-AD ( short for hyperspectral microscopy with adaptive detection ) . This new technique combines a relatively recent method called hyperspectral dark-field microscopy , which can identify nanoparticles from their unique optical signatures , with versatile computer algorithms to detect nanoparticles . HSM-AD is more sensitive than previously developed biodistribution techniques , and SoRelle , Liba et al . used it to produce highly detailed maps of nanoparticle uptake patterns in the organs of mice . These maps provide new insights into how cells and tissues in the body handle different nanoparticles . Moreover , HSM-AD was able to distinguish nanoparticles with unique shapes by their distinct optical signatures . Further experiments show that HSM-AD can reveal interactions between human tumor cells and nanoparticles specifically designed to target those cells . HSM-AD will be a useful resource for researchers studying the effect of nanoparticles on the human body . Future studies will use this technique to explore which nanoparticles have the potential to be developed for medical uses . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"tools",
"and",
"resources",
"cancer",
"biology"
] | 2016 | A hyperspectral method to assay the microphysiological fates of nanomaterials in histological samples |
Loss of polarity correlates with progression of epithelial cancers , but how plasma membrane misorganization drives oncogenic transcriptional events remains unclear . The polarity regulators of the Drosophila Scribble ( Scrib ) module are potent tumor suppressors and provide a model for mechanistic investigation . RNA profiling of Scrib mutant tumors reveals multiple signatures of neoplasia , including altered metabolism and dedifferentiation . Prominent among these is upregulation of cytokine-like Unpaired ( Upd ) ligands , which drive tumor overgrowth . We identified a polarity-responsive enhancer in upd3 , which is activated in a coincident manner by both JNK-dependent Fos and aPKC-mediated Yki transcription . This enhancer , and Scrib mutant overgrowth in general , are also sensitive to activity of the Polycomb Group ( PcG ) , suggesting that PcG attenuation upon polarity loss potentiates select targets for activation by JNK and Yki . Our results link epithelial organization to signaling and epigenetic regulators that control tissue repair programs , and provide insight into why epithelial polarity is tumor-suppressive .
The diagnosis of carcinomas–malignant tumors of epithelial origin—has long involved evaluating tissue architecture . Pronounced disorganization of biopsied epithelia is well-established to correlate with tumor malignancy and lethality . However , whether there exists a causative relationship between epithelial organization and tumor progression , as well as what the underlying mechanism might be , has been mysterious . Recent years have shed important light on the former question , identifying contexts where altered activity of proteins that regulate epithelial cell polarity can promote oncogenic phenotypes . For instance , the apical determinant atypical protein kinase C ( aPKC ) is amplified and over-expressed in multiple cancers ( Huang and Muthuswamy , 2010; Parker et al . , 2014 ) , while basolateral regulators are altered in several tumor types and degraded by viral oncoproteins ( Huang and Muthuswamy , 2010; Elsum et al . , 2012 ) ; cancer stem cell activity may also be promoted by transition from an epithelial state ( Martin-Belmonte and Perez-Moreno , 2011; Scheel and Weinberg , 2012 ) . Mouse models continue to support key roles for polarity regulators in cancer progression ( Pearson et al . , 2011; Muthuswamy and Xue , 2012; Elsum et al . , 2014; Feigin et al . , 2014 ) , but the mechanisms linking epithelial organization to tissue homeostasis , as well as the cellular targets that promote oncogenic growth upon polarity loss , remain unclear . Early evidence for causative links emerged from Drosophila , where mutations in single polarity-regulating genes can induce dramatic tumorous growths . These polarity regulators–scribble ( scrib ) , discs-large ( dlg ) , and lethal giant larvae ( lgl ) - cooperatively distinguish the basolateral domain from the apical by antagonizing aPKC activity ( St Johnston and Ahringer , 2010; Tepass , 2012 ) . This conserved ‘Scrib module’ functions in both vertebrates and invertebrates , not only in epithelia but also other polarized cell types . Conservation of these and other core polarity regulators allows Drosophila to be used as a model to study the coupling between epithelial architecture and growth control . When Scrib module function is lost from fly epithelia , mutant cells round up and become multilayered . In the imaginal discs , epithelial organs which normally have a precise intrinsic size-control mechanism , mutant tissue continuously proliferates to more than five times the WT cell number before it kills the animal . Small portions of the tumorous mass , when transplanted into adults , continue to grow uncontrollably and kill the host; such allografts can be repeated indefinitely . This disorganized , lethal and transplantable growth has been termed ‘neoplastic’ , and includes several additional features ( Gateff and Schneiderman , 1969; Bilder , 2004 ) . Neoplastic fly tissue is prone to dissemination and degrades basement membrane; in cooperation with oncogenic Ras it can migrate away from its primary site and invade other organs ( Pagliarini and Xu , 2003 ) . It is compromised in its differentiation potential , and cannot form adult structures ( Gateff and Schneiderman , 1969 ) . It can be recognized by the host innate immune system , whose cellular activities impede its growth ( Pastor-Pareja et al . , 2008; Cordero et al . , 2010 ) . Finally , it produces long-range signals that induce detrimental responses in fly hosts , including cachexia-like tissue wasting ( Figueroa-Clarevega and Bilder , 2015 ) . This suite of phenotypes , which echo those found in mammalian malignancies , suggest that elucidating mechanisms linking epithelial organization to tumor suppression in flies may provide novel insight into human cancer as well . What are the genes that induce the multiple aspects of the neoplastic phenotype , and how does loss of a single polarity regulator at the plasma membrane lead to their nuclear misregulation ? Here we define the global transcriptional changes associated with tumorigenic epithelial disorganization . By focusing on a single polarity-regulated enhancer of a gene involved in overgrowth , we then untangle signaling , transcription factor , and epigenetic activities that mediate activation upon polarity loss . Our results suggest that epithelia monitor their integrity via a coincidence detection mechanism , and respond to its loss by activating a damage-responsive gene expression program that cannot be turned off in mispolarized tumors .
The many malignant-like phenotypes observed upon loss of a single polarity regulator must be driven by altered gene expression . To identify such genes , we carried out RNA-Seq analysis of WT and mutant wing imaginal discs . We focused on changes common to neoplasm by sequencing cDNA libraries generated from both scrib and dlg tumors , which phenocopy each other ( Figure 1A–C ) ( Bilder et al . , 2000 ) . Analysis revealed 574 genes misregulated at least twofold in both mutant tissues ( FDR <5% ) , with 311 and 263 up- and downregulated respectively ( Figure 1D and Suplementary files 1–2 ) . Differentially expressed genes include several previously identified neoplastic effectors , such as the pro-invasion factors Matrix metalloprotease 1 ( Mmp1 ) and cheerio ( cher ) as well as the pupation regulator insulin-like peptide 8 ( Ilp8 ) ( Uhlirova and Bohmann , 2006; Colombani et al . , 2012; Garelli et al . , 2012; Külshammer and Uhlirova , 2013 ) ( Figure 1E ) . qRT-PCR analysis of these and other genes shows close agreement with RNA-Seq data ( R2 = 0 . 8844 ) . The transcriptome dataset therefore accurately captures the expression profile of neoplastic tissues , and contains genes that promote tumorigenesis upon polarity loss . 10 . 7554/eLife . 03189 . 003Figure 1 . Transcriptome analysis of neoplastic tumors . ( A–C ) F-actin staining reveals dramatic overgrowth and architecture defects of neoplastic dlg and scrib wing discs relative to WT . ( D ) Overlap of genes upregulated ( left ) or downregulated ( right ) in scrib and dlg tissues . ( E ) Genes previously implicated in neoplastic characteristics are differentially expressed . ( F and G ) Functional categories enriched in the upregulated and downregulated genes include markers of stress response and JAK/STAT pathway activation , and de-differentiation respectively . Selected overexpressed ( H ) and underexpressed ( I ) genes are shown . ( J–M ) Overexpression of Vg suppresses dlgRNAi-driven overgrowth and architecture defects . Dlg staining ( green ) demonstrates survival of Dlg-depleted wing cells . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 00310 . 7554/eLife . 03189 . 004Figure 1—figure supplement 1 . Decreasing oxidative stress or reexpressing eyelesss does not suppress neoplasia . ( A ) 19 genes activated in response to oxidative stress are significantly upregulated upon polarity loss . ( B–C ) Loss of dlg leads to higher superoxide levels , as evidenced by increased DHE staining , relative to WT . ( D–F ) Expression of the anti-oxidant enzymes Cat or Sod2 has no effect on dlgRNAi-mediated overgrowth . ( G–I ) Ectopic expression of eyeless ( eye ) , a master regulator of eye differentiation , induces photoreceptor formation in a small portion of wild-type and dlgRNAi ( 84 ) -expressing discs ( arrows ) , and is unable to block overgrowth upon polarity disruption . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 004 Amongst upregulated genes , Gene Ontology ( GO ) highlights factors involved in Response to Stimulus ( Figure 1F , H ) . Several in this category are immune-related factors , and may be due to the recruitment of hemocytes to neoplastic tumors ( Lebestky et al . , 2000; Pastor-Pareja et al . , 2008; Cordero et al . , 2010 ) . Others , including Glutathione S transferase E1 ( GstE1 ) and the chaperone CG7130 , are regulated by oxidative stress , and overall 19 polarity-sensitive targets are also elevated in hyperoxic conditions ( Figure 1—figure supplement 1A ) ( Landis et al . , 2004 ) . Dihydroethidium ( DHE ) , a fluorescent probe for superoxide anions , readily demonstrated elevation upon depletion of dlg ( Figure 1—figure supplement 1B–C ) . Co-overexpression of Catalase , Superoxide dismutase 2 , or rat Glutathione Peroxidase 1 , which suppress other Drosophila ROS dependent phenotypes ( Owusu-Ansah and Banerjee , 2009; Ohsawa et al . , 2012; Lim et al . , 2014 ) failed to alter the neoplastic phenotype induced by dlg knockdown ( Figure 1—figure supplement 1D–F ) , although we were unable to detect a consistent reduction of DHE in these contexts . Several metabolic regulators are also misexpressed in polarity-deficient tissues , including Drosophila Lactate Dehydrogenase ( ImpL3 ) , which contributes to a Warburg-like metabolic shift in human tumors ( Cairns et al . , 2011 ) ; however , ImpL3 knockdown also did not obviously alter neoplastic growth ( data not shown ) . Primary GO categories among downregulated genes likely reflect the failure of neoplastic tumors to differentiate ( Figure 1E , G , I ) . We investigated the functional role by ectopically expressing fate-specifying transcription factors in dlg-depleted tissue . Strikingly , co-expression of vestigial ( vg ) , a distal wing pouch selector that is downregulated in mutant discs , suppressed overgrowth and architecture defects ( Figure 1J–M ) . Though vg overexpression eliminates polarity-deficient clones through apoptosis ( Khan et al . , 2013 ) , we recovered an intact wing pouch consisting of dlgRNAi/vg co-expressing cells ( Figure 1M ) . We also tested ectopic expression of an eye-specifying transcription factor in wing and eye tissue . eyeless was incapable of suppressing dlg knockdown in either context , but was also incapable of inducing broad photoreceptor differentiation in WT or dlg-depleted tissue ( Figure 1—figure supplement 1G–I , data not shown ) . Together , these data suggest that restoring expression of differentiation-promoting transcription factors can , in some contexts , block neoplastic transformation . The only cell signaling pathway among the top GO categories is the JAK/STAT cascade . Upregulated genes include STAT targets such as chinmo and Socs36E , and a JAK/STAT activity reporter is strongly expressed in dlg and scrib discs ( Figure 2A–B ) ( Wu et al . , 2010 ) . Remarkably , each of the three unpaired ( upd ) genes , which encode the ligands for the JAK/STAT pathway , were transcriptionally elevated between ∼3- and ∼50-fold , while genes encoding other signal transduction components were unaltered ( Figure 2C ) . To assess a functional role , we used engrailed-GAL4 to express Socs36E , a negative regulator of JAK/STAT intracellular signaling ( Callus and Mathey-Prevot , 2002 ) , in the posterior compartment of wing discs carrying a hypomorphic allele of dlg , and then counted cell numbers on a cell sorter . Strikingly , Socs36E decreased proliferation of dlghypo cells by 40% , while having no significant effect on growth or viability of WT discs ( Figure 2D–H ) . Expression of Socs36E or a dominant-negative form of the JAK-STAT receptor Domeless ( DomeDN ) also suppressed the growth of scrib-depleted discs ( Figure 4—figure supplement 3A–C ) . We therefore conclude that in imaginal discs , as in RasV12-expressing clones ( Wu et al . , 2010 ) , the Scrib module regulates JAK-STAT ligand expression to suppress tissue overgrowth . 10 . 7554/eLife . 03189 . 005Figure 2 . JAK/STAT activation drives overgrowth upon polarity loss . ( A and B ) A JAK/STAT pathway reporter ( green ) is highly elevated throughout dlg as compared to WT discs , indicating strong pathway activation . ( C ) The ligand-encoding upd genes , but not other JAK/STAT pathway components , are transcriptionally upregulated in neoplastic tissues . ( D–G ) Reduction of JAK/STAT pathway activity via SOCS36E overexpression has no significant effect on WT growth , but suppresses overgrowth of dlghypo tissue . Actin ( red ) highlights cell outlines , while GFP ( green ) marks the engrailed-expressing domain . FACS-based quantification is shown in H ( **p < 0 . 001 ) . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 00510 . 7554/eLife . 03189 . 006Figure 2—figure supplement 1 . upd3 knockdown is not sufficient to prevent neoplastic tumors . Eye imaginal discs expressing upd3 RNAi alone ( A ) , dlg RNAi alone ( B ) , and dlg RNAi + upd3 RNAi ( C ) . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 006 To elucidate links between polarity and transcriptional control of growth , we focused on a single mitogenic gene: upd3 . We cloned 3 kilobases ( kb ) of genomic DNA surrounding the upd3 ATG into a lacZ reporter ( ‘upd3lacZ’ ) and found that this reporter was not expressed in WT discs . However , like the overlapping upd3 > GFP reporter , it was distinctly upregulated in neoplastic discs ( Figure 3A–C ) ( Pastor-Pareja et al . , 2008 ) . We then identified a minimal polarity-responsive region within this enhancer , using fragments previously analyzed in the adult gut ( Jiang et al . , 2011 ) . Although reporters including upd3 . 1LacZ , which is activated by perturbations in the gut epithelium , remain silent , a 1-kb element within the first intron ( upd3 . 3LacZ ) was expressed in a patchy manner throughout dlg discs ( Figure 3D–I ) . Expression of upd3 . 3lacZ , like that of upd3lacZ , was in cells of the disc proper , not in the peripodium or hemocytes ( Figure 3—figure supplement 1A–B ) ; this patchy expression resembled that seen with several other upregulated neoplastic effectors , ( Figure 4B′ , Figure 3—figure supplement 1C–H ) . Upd3 . 3LacZ was similarly activated in scrib discs , demonstrating that this enhancer is generally responsive to disruption of epithelial polarity ( Figure 4—figure supplement 3E ) and identifying a polarity-sensitive cis-regulatory region . 10 . 7554/eLife . 03189 . 007Figure 3 . Identification of a polarity-responsive enhancer in upd3 . ( A ) Schematic of upd3 reporter constructs in relation to the corresponding genomic region . ( B and C ) 3 kb upd3LacZ is not expressed in WT , but is upregulated in dlg discs . ( D and E ) upd3 . 3LacZ sub-fragment is also silent in WT , but is upregulated in dlg like upd3LacZ . ( F–I ) Other sub-fragments are not significantly expressed in either WT or dlg . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 00710 . 7554/eLife . 03189 . 008Figure 3—figure supplement 1 . Imaginal expression of polarity-responsive target genes in neoplasia . ( A–B ) The upd3LacZ and upd3 . 3LacZ reporters are expressed primarily in the disc proper , and not the hemocytes or the peripodial membrane . ( C–H ) The JNK pathway reporter AP-1-GFP , and transcriptional reporters for the polarity-sensitive targets ImpL2 and dilp8 are relatively silent in WT tissue , but active in a patchy pattern in dlg discs . Scale bars: A–B: 10 μm , C–H: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 00810 . 7554/eLife . 03189 . 009Figure 3—figure supplement 2 . Conserved AP-1 and Sd binding sites in genes upregulated in neoplasia . ( A ) The upd3 . 3 enhancer contains two evolutionarily conserved ( between D . melanogaster , D . yakuba and D . erecta ) AP-1 binding sites ( green boxes ) , and one semi-conserved Sd binding site ( red box ) . Conserved AP-1 and Sd binding sites are also evident in several neoplasia-induced genes that are also upregulated during wounding , including Ets21C ( B ) , Pvf1 ( C ) , ImpL2 ( D ) , ple ( E ) , and Ilp8 ( F ) . Exons are denoted in orange and green arrows in Ets21C and Ilp8 mark the transcription start site . ( G and H ) The conservation of the AP-1 and Sd binding sites in upd3 . 3 is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 00910 . 7554/eLife . 03189 . 010Figure 4 . JNK-Dependent transcription is necessary for overgrowth and upd3 . 3 activation upon polarity loss . WT wing discs ( A ) do not express either the JNK target Mmp1 or upd3 . 3LacZ ( A′ ) . Expression of dlgRNAi promotes overgrowth and disorganization ( B ) , as well as Mmp1 and upd3 . 3LacZ upregulation ( B′ ) . Inhibiting AP-1 transcription with either JNKDN or FosDN restores normal disc size and architecture ( C and D ) , and abrogates Mmp1 and upd3 . 3LacZ expression ( C′ and D′ ) . WT discs segregate apical aPKC and basolateral Scrib ( E ) . dlgRNAi expression leads to apical domain expansion and co-localization of aPKC and Scrib ( F , arrowheads ) . Co-expressing JNKDN and dlgRNAi restores the separation of aPKC and Scrib ( G ) . Activation of JNK is sufficient , when apoptosis is blocked with miRGH , to drive upd3 . 3LacZ , Mmp1 and overgrowth but not to alter polarity ( H and I ) . Scale bars: A–D , H: 100 μm , E–G , I: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 01010 . 7554/eLife . 03189 . 011Figure 4—figure supplement 1 . Inhibitor constructs do not significantly affect WT tissue growth and viability . ( A–B ) Blocking JNK activity with JNKDN or FosDN ( C ) has no effect on normal growth or tissue architecture , relative to wild-type . Expression of miRGH does not affect normal tissue architecture or growth ( D ) . Knockdown of Yki promotes mild architecture defects ( E ) , while BrmDN expression has no phenotype ( F ) . For all panels , transgenes were expressed in the dorsal wing pouch with the ms1096-GAL4 driver . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 01110 . 7554/eLife . 03189 . 012Figure 4—figure supplement 2 . Quantification of upd3 . 3LacZ staining . ( A ) Expression of dlgRNAi increases upd3 . 3LacZ fluorescence , which is suppressed by blocking JNK or Trx activity . ( B ) Expression of aPKCact stimulates upd3 . 3LacZ in a JNK-independent , but Yki-dependent manner . ( C ) . Hyperactivation of Yki or JNK activity upregulates upd3 . 3LacZ expression . ( D ) Alone , expression of aPKCmild , ph-pRNAi , or JNKKWT does not activate upd3 . 3LacZ; however , co-expression of aPKCmild with ph-pRNAi or JNKKWT drives upd3 . 3LacZ . ( **p < 0 . 001; n . s . = not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 01210 . 7554/eLife . 03189 . 013Figure 4—figure supplement 3 . Neoplasia induced by scrib loss is also dependent on JAK-STAT , JNK , and Yki pathway activity . ( A–C ) Reducing JAK-STAT activity with DomeDN or Socs36E attenuates scribIR-mediated overgrowth . ( D–H ) Blocking JNK pathway activation by depletion of the JNK kinase hep or overexpression of JNKDN suppresses the overproliferation , architecture defects and upd3 . 3LacZ activation induced by scrib loss . ( I and J ) Yki is necessary for neoplastic overgrowth of scrib tissue . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 013 We next sought to identify molecular pathways linking epithelial polarity to upd3 expression . Motif scanning of the upd3 . 3 enhancer detected two evolutionarily-conserved binding sites for AP-1 , the Jun kinase ( JNK ) pathway transcription factor ( Figure 3—figure supplement 2A , G ) . We tested whether JNK signaling is required for upd3 . 3LacZ activation . Expression of a dominant-negative form of Drosophila JNK ( Flybase: Basket ) , ( JNKDN ) has been shown to block neoplastic overgrowth , as well as polarity and architecture defects ( Figure 4A–C , E–G; Figure 4—figure supplement 1A–B ) ( Robinson and Moberg , 2011; Sun and Irvine , 2011 ) . Notably , JNKDN also completely abrogated dlgRNAi-induced upd3 . 3LacZ expression ( Figure 4B′ , C′; Figure 4—figure supplement 2A ) , as well as that induced by scribRNAi ( Figure 4—figure supplement 3G–H ) . Mutation of the JNK kinase , hemipterous ( hep ) also prevented upd3 . 3LacZ levels in scrib tissue ( Figure 4—figure supplement 3D–F ) , confirming that canonical JNK signaling acts downstream of polarity disruption to regulate upd3 . The mechanism by which JNK promotes neoplasia is unclear . Phosphorylation of Ajuba LIM protein ( Jub ) has been proposed to be key ( Sun and Irvine , 2013 ) ; however , the presence of AP-1 binding sites within upd3 . 3 suggests a direct transcription-mediated mechanism . To test the latter mechanism , we assayed discs co-expressing dlgRNAi and fosDN , which prevents activity of the AP-1 transcription factor ( Ciapponi et al . , 2001 ) . Strikingly , fosDN fully phenocopied the effects of JNKDN: it prevented both upd3 . 3LacZ expression and dlgRNAi-mediated neoplasia ( Figure 4D; Figure 4—figure supplement 1C; Figure 4—figure supplement 2A ) . Taken together , these experiments demonstrate that maintenance of epithelial polarity prevents transcription of oncogenic JNK-dependent target genes . Given that elevated JNK signaling is necessary for upd3 . 3LacZ expression and neoplastic overgrowth , is it sufficient ? Ectopic JNK activity in WT tissue leads to apoptosis ( Igaki et al . , 2002 ) , so we co-expressed the JNK-activating ligand Eiger with a microRNA targeting the pro-apoptotic genes reaper , grim , and head involution defective ( miRGH ) to block both cell death and caspase activation ( Siegrist et al . , 2010 ) . In this context , JNK activation alone induced upd3 . 3LacZ ( Figure 4H; Figure 4—figure supplement 1D; Figure 4—figure supplement 2C ) and increased tissue size ( Pérez-Garijo et al . , 2009 ) . However , upd3 . 3LacZ induction was low compared to the canonical JNK target Mmp1 , while dlg knockdown activated both comparably ( Figure 4B′ , H ) . Further , apical and basolateral proteins remained properly localized , indicating that JNK activation alone does not disrupt polarity ( Figure 4I ) ( Sun and Irvine , 2011 ) . Therefore , JNK signaling is sufficient for partial upd3 . 3 activation and overgrowth , but it is unable to induce full neoplasia . The inability of JNK activation to fully recapitulate dlg loss suggests that polarity regulators modulate additional factors to prevent upd3 . 3 transcription and neoplasm . One candidate is aPKC , which is strongly mislocalized upon loss of Scrib module function but not JNK activation ( Figure 4F , I ) ( Bilder and Perrimon , 2000 ) . We expressed a constitutively active form ( aPKCact ) that can drive neoplasia and found that it was sufficient to potently trigger upd3 . 3LacZ transcription ( Figure 5A; Figure 4—figure supplement 2B ) . aPKCact can also activate JNK targets ( Figure 5A′ ) , raising the possibility that aPKC regulates upd3 through JNK . However , inhibiting JNK did not prevent aPKCact-mediated upd3 . 3LacZ activation or overgrowth , while it was effective at preventing expression of Mmp1 ( Figure 5B; Figure 4—figure supplement 2B ) . Similar results were seen when membrane-bound WT aPKC ( aPKCmild ) was co-expressed with its partner Par-6 , demonstrating that the results are not transgene-specific ( Figure 5—figure supplement 1 ) and thus showing that aPKC is capable of stimulating tumorigenic transcription independently of JNK . 10 . 7554/eLife . 03189 . 014Figure 5 . aPKC activity drives upd3 . 3LacZ activation in a yki-dependent manner . ( A ) Expression of constitutively active aPKC ( aPKCact ) induces upd3 . 3LacZ and Mmp1 upregulation and neoplasia . ( B ) Expressing JNKDN suppresses Mmp1 , but does not prevent aPKCact-mediated upd3 . 3LacZ activation or overgrowth . ( C ) Knockdown of yki blocks upd3 . 3LacZ and overgrowth upon ectopic aPKC activity , while constitutively active Yki drives upd3 . 3LacZ expression and tissue overgrowth relative to WT ( D and E ) . Expression of a mildly-active form of aPKC ( F ) or JNK ( G ) alone cannot activate upd3 . 3 , but together are sufficient for expression ( H ) . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 01410 . 7554/eLife . 03189 . 015Figure 5—figure supplement 1 . Ectopic aPKC activity drives upd3 . 3LacZ in a JNK-independent manner . ( A ) Expression of aPKCmild and its partner Par6 drives upd3 . 3LacZ as well as strong overgrowth and Mmp1 expression in the wing pouch . ( B ) Co-expression of JNKDN does not suppress upd3 . 3LacZ activation or overgrowth though Mmp1 upregulation is abrogated . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 01510 . 7554/eLife . 03189 . 016Figure 5—figure supplement 2 . Scrib module and wts mutant expression profiles display limited overlap . ( A ) Comparison of the Scrib module and wts mutant transcriptomes ( see ‘Materials and methods’ ) reveals a limited degree of overlap . ( B ) Most canonical Yki growth targets are not upregulated in neoplastic tissues . ( C–F ) upd3 . 3LacZ and STAT signaling are not upregulated in wts discs . The transgenic Hpo pathway reporters Diap1-GFP3 . 5 ( G–H ) and HREX-GFP ( I–J ) are strongly upregulated in neoplastic tissue relative to WT; however , Diap1-LacZ ( K–L ) , which is inserted into the endogenous locus , is only slightly increased . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 01610 . 7554/eLife . 03189 . 017Figure 5—figure supplement 3 . Co-activation of JNK and Yki are not sufficient to drive neoplasia . ( A–B ) Ectopic expression of wild-type JNKK causes only slight morphological defects and upregulates Mmp1 , but cannot activate upd3 . 3LacZ . ( C ) Co-expression of JNKK and Ykiact activates both upd3 . 3LacZ and Mmp1 , but does not recapitulate polarity defects or neoplastic-like overgrowth . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 017 To determine how aPKC activity at the cell cortex regulates transcriptional targets , we returned to our analysis of upd3 . 3 sequences . The enhancer contains a partially evolutionarily conserved binding site for Scalloped ( Sd ) , a DNA-binding protein that recruits activated Yorkie ( Yki ) to target genes ( Figure 3—figure supplement 2A , H ) ( Wu et al . , 2008 ) . Intriguingly , conserved Sd and AP-1 binding sites are also found together in ∼1 kb regulatory regions of other upregulated genes ( Figure 3—figure supplement 2B–F ) . To determine if Yki acts downstream of aPKC , we assessed discs co-expressing aPKCact and a moderate strength RNAi against yki ( ykiRNAi ) . While yki knockdown under these conditions had a minimal effect on WT growth , it completely abrogated ectopic aPKC-driven upd3 . 3LacZ upregulation ( Figure 5C; Figure 4—figure supplement 1E; Figure 4—figure supplement 2B ) . Similarly , depletion of yki suppressed the overgrowth of scrib tissue ( Figure 4—figure supplement 3I–J ) . We then analyzed discs overexpressing constitutively active Yki ( Ykiact ) , which display massive overgrowth without affecting epithelial polarity ( Dong et al . , 2007; Oh and Irvine , 2008 ) . Upd3 . 3LacZ expression was highly elevated in Ykiact-expressing tissues ( Figure 5D–E; Figure 4—figure supplement 2C ) , indicating that Yki can also be sufficient to activate the polarity-responsive enhancer . Though hyperactivation of either JNK or Yki through overexpression of activated proteins can drive upd3 . 3 transcription , we found that only the highest levels of signaling could do so . For instance , neither upd3 . 3lacZ nor JAK/STAT signaling was active in hyperproliferating hippo ( hpo ) pathway mutant tumors ( Figure 5—figure supplement 2C–F ) . Moreover , overexpression of either WT JNK kinase , or a membrane-targeted form of WT aPKC ( aPKCmild ) , activated Mmp1 but does not cause substantial overgrowth; neither activates upd3 . 3lacZ ( Figures 5F–G , 7J , Figure 5—figure supplement 3B ) . Since loss of polarity activates aPKC and JNK signaling in parallel , we tested whether the two pathways converge upon the enhancer . Strikingly , coexpression of JNK kinase and aPKCmild induced upd3 . 3lacZ upregulation ( Figure 5H; Figure 4—figure supplement 2D ) , along with moderate overgrowth and polarity defects . These data support a model in which upd3 . 3 works as a ‘coincidence detector’ , responding to simultaneous aPKC-mediated Yki activation and JNK-dependent Fos activation upon polarity loss . The above results suggest that transcription from enhancers like upd3 . 3 is kept in check when either JNK or Yki are activated at physiological rather than manipulated experimental levels . We therefore investigated additional regulators of upd transcription . Our previous work identified the upd genes as targets of direct repression by the Polycomb Group ( PcG ) , and showed that mutations in PcG can result in tumorous growth ( Classen et al . , 2009 ) . These data suggest the hypothesis that epithelial polarity also acts through PcG to influence mitogenic gene expression . To test this hypothesis , we first asked whether PcG regulates the polarity-responsive enhancer . Imaginal discs mutant for the paralogous PcGs Psc and Su ( z ) 2 show dramatic overgrowth , in which apicobasal polarity is often intact ( Classen et al . , 2009 ) . Strikingly , they also upregulated upd3 . 3LacZ , but not other upd3LacZ subfragments ( Figure 6B and data not shown ) . This response is identical to that observed in polarity-deficient tissues . 10 . 7554/eLife . 03189 . 019Figure 6 . The Scrib module and PcGs regulate common targets . ( A and B ) Loss of the paralogous PcGs Psc and Su ( z ) 2 leads to activation of upd3 . 3lacZ , along with dramatic overgrowth and architecture defects . Activation is observed in areas of epithelial ( arrows ) and disrupted ( arrowheads ) organization . Comparison of all genes ( C ) and direct PcG targets ( D ) upregulated in Psc/Su ( z ) 2 and Scrib module mutant tissues reveals statistically significant overlaps . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 01910 . 7554/eLife . 03189 . 020Figure 6—figure supplement 1 . PcG depletion does not cause widespread loss of polarity . ( A ) Depletion of the paralogous PcGs ph-p and ph-d leads to overgrowth and upd3 . 3LacZ activation , including in areas with mild architecture defects . Arrows show areas of upd3 . 3LacZ expression in areas with epithelial organization; arrowheads indicate reporter activation in regions with disrupted architecture . ( B ) Regions of Psc/Su ( Z ) 2 mutant discs have normal polarity . ( C and D ) Most apical and basolateral regulatory genes are minimally changed in Psc/Su ( Z ) 2 tissue . Scale bars: A: 100 μm , B: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 020 The common response of upd3 to polarity regulators and PcG could be a unique case , or alternatively could reflect a larger role for PcG in polarity-sensitive growth control . To determine if the Scrib module and PcGs co-regulate additional loci , we carried out a global transcriptional analysis of PcG mutant wing disc tumors ( Supplementary file 3 ) . Comparison of Scrib module and PcG mutant RNA-Seq datasets revealed that nearly half of the genes upregulated upon polarity loss are also upregulated in PcG mutant tissues , a highly significant enrichment ( p < 6 . 98e-121 , Figure 6C ) . This degree of similarity does not reflect a general overgrowth signature , as comparison with the transcriptome of warts tumors ( Oh et al . , 2013 ) gives a much less substantial overlap ( Figure 5—figure supplement 2A ) . Further analysis of Scrib module transcriptomes revealed that nearly 25% of direct Pc-bound targets ( Kwong et al . , 2008 ) that are upregulated upon PcG loss are also upregulated in polarity-deficient tissues ( Figure 6D ) . This strong enrichment supports a model whereby the Scrib module and PcG act in concert at certain common downstream genes . The above data are consistent with a scenario whereby polarity loss weakens PcG-mediated repression of select targets that promote tumorigenesis . An alternate possibility is that PcG mutant tissue itself is polarity-defective; however , it often maintains polarized organization including areas that upregulate upd3 . 3lacZ , it does not show transcriptional changes of polarity regulators , and unlike neoplastic tissue it is not suppressed by aPKC inhibition ( [Classen et al . , 2009] , Figure 6—figure supplement 1 , data not shown ) . To assess the functional significance of PcG in neoplastic tissues , we used the genetic interaction assay of Figure 2 . Knockdown of the PcG gene polyhomeotic-proximal ( ph-p ) alone has no effect on growth of WT discs , due to the presence of its paralog polyhomeotic-distal ( ph-d ) . However , when ph-p is knocked down in hypomorphic dlg discs , it significantly increased growth and cell proliferation ( Figure 7A–E ) . Similar results were observed upon knockdown of a second PcG component , Su ( Z ) 2 ( data not shown ) . If reduced PcG function contributes to overgrowth upon polarity loss , then preventing target derepression should suppress neoplastic growth . We inhibited Brahma ( Brm ) , which suppresses PcG-mediated homeotic transformation and often opposes PcG activity at target genes ( Tamkun et al . , 1992 ) . Expression of dominant-negative Brm reduced both the growth of dlgRNAi-expressing tissue and upd3 . 3LacZ expression ( Figure 7F–G , Figure 4—figure supplement 2A ) . An analogous experiment with scrib RNAi could not be performed due to synthetic lethality with the Brm-DN transgene . These data support a role for epithelial polarity in promoting PcG-mediated repression of mitogenic target genes to suppress tumorigenesis . 10 . 7554/eLife . 03189 . 018Figure 7 . PcGs cooperate with Scrib module proteins to regulate growth . ( A–D ) Knockdown of ph-p has little effect on WT growth but increases the growth of dlghypo tissue . Quantification is in E ( ***p < 0 . 0001 ) . ( F–G ) BrmDN expression in dlgRNAi tissue decreases both upd3 . 3LacZ activation and overgrowth . ( H ) PcG components ph-p and Psc are downregulated in scrib and dlg mutant tissue ( average in green ) , similar to levels observed upon JNK activation ( blue ) . ( **p < 0 . 005 ✝FDR < 0 . 05 in one genotype; ✝✝FDR < 0 . 05 in both genotypes ) ( I–K ) Knockdown of ph-p or expression of a moderately active form of aPKC ( aPKCmild ) does not induce upd3 . 3LacZ , and aPKCmild induces only slight overgrowth . However , co-expression of these transgenes leads to strong overgrowth and upd3 . 3LacZ expression . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03189 . 018 The above analyses suggest diminished PcG activity in Scrib module mutant tissues , but do not point to a molecular mechanism . Intriguingly , using a wounding paradigm , Lee et al . found that JNK signaling can partially downregulate PcG expression , facilitating dedifferentiation and regeneration ( Lee et al . , 2005 ) . Because JNK is activated upon polarity loss , we evaluated PcG transcript levels in Scrib module mutant tissues . Expression of the core PcG components ph-p and Psc is reduced in neoplastic tumors to an extent similar to that seen upon strong JNK activation ( Figure 7H ) , suggesting that JNK signaling upon polarity loss compromises PcG function . Finally , we tested whether compromised PcG function would promote polarity-responsive enhancer activation under moderate signaling conditions . Mild activation of aPKC drove polarity alterations and a limited degree of neoplasia , along with mild JNK signaling that can activate Mmp1; at these levels , both kinases together were incapable of activating upd3 . 3 ( Figure 7J ) . However , upon knockdown of ph-p , which does not activate JNK , mild aPKC signaling not only drove robust overgrowth but also upd3 . 3LacZ upregulation ( Figure 7I–K , Figure 4—figure supplement 2D ) . From these data , we conclude that epithelial polarity normally suppresses neoplasia through PcG in cooperation with JNK and aPKC/Yki pathways .
Our data build on those of others showing that JNK , aPKC and Yki are key players in fly neoplasia ( Leong et al . , 2009; Grzeschik et al . , 2010; Robinson et al . , 2010; Zhu et al . , 2010; Doggett et al . , 2011; Sun and Irvine , 2011; Verghese et al . , 2012 ) . By focusing on a single enhancer element of a gene involved in tumorous growth , we clarify the role of implicated regulating kinases and define how proliferation can be triggered by each pathway . Inhibition of Fos can suppress upd3 upregulation and neoplasia , indicating that this transcription factor itself is the major target of JNK in this context . Yet a polarity-sensitive enhancer is not fully activated by JNK alone , even when apoptosis is blocked . aPKC is an additional regulator of this enhancer , and as previously suggested ( Doggett et al . , 2011 ) , can activate Yki independent of , rather than through , JNK . Inhibiting either the JNK or Hpo pathways , including depletion of the downstream transcription factors , prevents expression of the polarity-sensitive enhancer; our analysis predicts that mutating transcription factor binding sites would give the same effect . Knockdown of upd3 alone in neoplastic tumors does not prevent overgrowth ( Figure 2—figure supplement 1 ) ; upd1 and upd2 are also regulated by JNK , Yki , and PcG ( Pastor-Pareja et al . , 2008; Classen et al . , 2009; Jiang et al . , 2009; Staley and Irvine , 2010; Wu et al . , 2010 ) and may act through analogous enhancers to cooperatively drive tumor formation . Loss of polarity thus induces two separate signaling pathways . An unknown mechanism triggers JNK to induce Fos-dependent transcription , while at the same time mispolarization of aPKC drives Yki-dependent transcription . Under mild signaling conditions , both pathways are required simultaneously to trigger enhancer expression or overgrowth , while inhibition of either is sufficient to suppress neoplasia . We suggest that polarity-responsive enhancers like upd3 . 3 work as ‘coincidence detectors’ that during normal physiology require inputs from both JNK/Fos and aPKC/Yki . In this way , neither stress nor developmental growth signals alone run the risk of triggering malignant transformation . However , upon severe tissue damage that disrupts the epithelium , both stress and polarity signals are initiated to effect repair pathways ( see below ) . Our results also emphasize the unexpectedly central role of transcription in mediating cell polarity loss . Inhibition of Fos can revert not only growth defects but also polarity defects of neoplastic tumors . This surprising result suggests that polarity regulation by the Scribble module involves not only antagonistic interactions with the Par module at the cell cortex , but also an important transcriptional component that may be regulated similarly to the mitogenic upd3 enhancer studied here . Nevertheless , activation of JNK , Yki , or both together is insufficient to elicit polarity defects ( Figure 5—figure supplement 3 ) , while aPKC activation alone is . Thus , aPKC must have additional effectors through which it regulates transformation; further analysis of the neoplastic transcriptome will shed light on this . Yki is clearly a major regulator of neoplastic transformation , providing a link between the primary Drosophila TSG pathways ( Grzeschik et al . , 2010; Robinson et al . , 2010; Chen et al . , 2012; Verghese et al . , 2012 ) . However , our transcriptional data highlight a major puzzle . Many Hpo pathway targets , including direct growth regulators such as cycE , diap1 , and Myc , are expressed at near-normal levels in Scrib module mutants , and comparison of Scrib module and Hpo mutant transcriptomes reveals limited overlap ( Figure 5—figure supplement 2A–B ) . If Yki is activated in both types of tumorous tissue , why do they behave so differently ? Our data help to rule out several models for altered Yki target selection . It is unlikely to be driven by simultaneous activation of JNK upon polarity loss , since co-activation of Yki and JNK does not recapitulate neoplastic growth phenotypes ( Figure 5—figure supplement 3 ) . It is also unlikely to be explained by a model in which Yki activation through aPKC differs from Yki activation through canonical Hpo pathway regulators , since a transgenic 3 . 5 kb diap1 fragment is strongly upregulated in neoplastic tissue , paralleling upregulation of a minimal Yki-responsive element ( Figure 5—figure supplement 2G–J ) . Interestingly , an enhancer trap inserted at the same 3 . 5 kb sequence in the endogenous diap1 locus ( Zhang et al . , 2008 ) is only slightly upregulated by comparison ( Figure 5—figure supplement 2K–L ) , hinting that the native chromatin environment at certain Yki targets might influence target response . Our data point to PcG as a new player in the transcriptional response to polarity loss . Three pieces of evidence support a close relationship between the Scrib module and PcGs: ( 1 ) their related mutant phenotypes , ( 2 ) the extensive and highly significant overlap of their mutant gene expression profiles , and ( 3 ) the sensitivity of Scrib module mutant overgrowth to changes in PcG activity . However , since canonical PcG targets including Hox genes are not upregulated in neoplastic tissues ( Supplementary files 1–2 ) , and overall Histone H3K27me3 levels are not altered ( data not shown ) , the data rule out a global inactivation of PcG . Instead , they suggest that decreased PcG-mediated repression ‘primes’ select targets for activation by polarity-responsive effector pathways . Mild activation of either JNK or aPKC alone is insufficient to stimulate enhancers such as upd3 . 3 . However , at these targets , reduced PcG activity upon Scrib module loss synergizes with JNK and aPKC signaling , perhaps by providing a permissive chromatin environment for Fos- and Yki-stimulated transcription . More generally , the link to epigenetic regulators that control many targets provides a mechanism by which loss of a single polarity regulator can induce the widespread transcriptional changes that drive the multifaceted neoplastic phenotype . Our primary analysis focuses on overgrowth , but the transcriptome identifies further features of human cancer found in neoplastic Drosophila cells . In addition to oxidative stress , fly homologs of metabolic genes that fuel human cancer growth are elevated , including fatty acid synthase ( FASN ) which facilitates de novo lipogenesis , and LDH which promotes aerobic glycolysis in the Warburg effect ( Cairns et al . , 2011; Baenke et al . , 2013; Gorrini et al . , 2013 ) . However , glycolytic enzyme transcription in fly neoplastic tumors remains relatively unchanged , suggesting that metabolic changes may be more complex . Dedifferentiation is considered another key feature of human tumor malignancy ( Friedmann-Morvinski and Verma , 2014 ) , and the major signature evident from genes downregulated in neoplastic tissues reflects a failure to differentiate . Khan et al . recently reported that forcing differentiation can cause elimination of neoplastic clones ( Khan et al . , 2013 ) ; by contrast , our experiments show that restoring expression of the wing-fate regulator Vg suppresses tumorous overgrowth without inducing cell death . Thus , promoting tissue differentiation may be a tumor suppressive function of epithelial organization . Why might loss of polarity drive this particular constellation of events that result in tumorous overgrowth ? Our global analysis reveals that apicobasal polarity disruption elicits responses with striking parallels to those seen in epithelial wounds in both Drosophila and humans ( Schäfer and Werner , 2008; Lee and Miura , 2014 ) . These parallels , which are both thematic and extend to regulation of specific genes , include activation of stress signaling , reactive oxygen species production , upregulation of matrix remodeling enzymes , de-differentiation , recruitment of immune cells , and transcription of growth-promoting cytokines that stimulate cell proliferation . Intriguingly , several upregulated neoplastic effectors that contain conserved AP-1 and Sd binding sites are also upregulated during wound-healing ( Pastor-Pareja et al . , 2008; Wu et al . , 2009; Garelli et al . , 2012; Patterson et al . , 2013 ) ( Figure 3—figure supplement 2 ) . An attractive model is that linking transcriptional control of such targets to polarity regulators , via polarity-regulated aPKC , cell architecture-regulated Yki and stress-regulated JNK activity on both downstream transcription factors and PcG epigenetic regulators , allows the tissue to connect disturbances in its integrity to the activation of broad gene expression programs that promote repair . Following tissue damage , restoration of tissue architecture and integrity would abrogate wound-response signals . In contrast , in polarity-deficient tissues , architecture can never be restored , and these pro-growth , de-differentiation cues remain active , leading to the formation of malignant tumors that kill the organism . Our data linking apicobasal polarity to neoplastic gene expression thus suggest an evolutionarily ancient genesis for cancers as ‘wounds that never heal’ ( Dvorak , 1986 ) .
The following alleles were used in this study: white [1118] ( WT ) , dlg [40-2] , dlg [hf321] ( dlghypo ) scrib [1] , hep [r75] ( JNKK ) , Psc/Su ( Z ) 2 [XL26] ( Li et al . , 2010 ) , ykiB5 . The following additional strains were used: engrailed GAL4 , UAS-GFP , ms1096 GAL4 , eyFLP; act>>GAL4 , UAS-GFP , 10XStat92E-GFP , upd3 . 1LacZ , upd3 . 2LacZ , and upd3 . 3LacZ ( Jiang et al . , 2011 ) , thj5c8 ( Ryoo et al . , 2002 ) , diap1-GFP3 . 5 ( Zhang et al . , 2008 ) , HREX-GFP ( Wu et al . , 2008 ) , UAS-Socs36E , UAS-Dome∆cyt ( UAS-DomeDN ) , UAS-BskK53R ( UAS-JNKDN ) , UAS-fospanAla ( UAS-FosDN ) , UAS-miRNAreapergrimhid ( UAS-miRGH ) ( Siegrist et al . , 2010 ) , UAS-GFP , UAS-hippo , UAS-eiger , UAS-aPKCΔN ( UAS-aPKCact ) , UAS-ykiS168A ( UAS-ykiact ) , UAS-BrmK804R ( UAS-brmDN ) , UAS-aPKCCAAX ( UAS-aPKCmild ) , UAS-Sod2 , UAS-Catalase , UAS-Ey , UAS-vg , and UAS-hepWT ( UAS-JNKKWT ) , AP-1-GFP , ImpL2-GFP , dilp8-GFP , EcadRNAi . UAS-aPKCCAAX UAS-Par6 was a kind gift from T Harris . UAS-dlgRNAi ( 39035 ) , UAS-dlgRNAi ( 34854 ) were obtained from the Bloomington Stock Center; UAS-yki RNAi ( 104523 ) , UAS-ph-p RNAi ( 10679 ) , UAS-ph RNAi ( 50028 ) , and UAS-Su ( Z ) 2 RNAi ( 100096 ) were obtained from the Vienna Drosophila RNAi Center . Unless otherwise noted , all transgenes were driven in the wing pouch by ms1096-GAL4 . WT controls were outcrosses to w . Crosses were reared at 25°C , except for the crosses to assess upd3 . 3LacZ expression in scribIR and scribIR;BskDN tissue , which were raised at 29°C . Imaginal discs were fixed and stained ( Bilder et al . , 2000 ) with TRITC-phalloidin ( Sigma-Adrich , St . Louis , MO ) and primary antibodies against the following antigens: β-gal ( Abcam , San Francisco , CA ) , Mmp1 , Dlg , Scrib ( all from Developmental Studies Hybridoma Bank , Iowa City , IA ) and aPKC ( Santa Cruz Biotechnology , Santa Cruz , CA ) . DAPI ( Molecular Probes , Eugene , OR ) was used to visualize nuclei . Secondary antibodies were from Invitrogen ( Carlsbad , CA ) . DHE staining was performed on live tissue as previously described ( Owusu-Ansah et al . , 2008 ) . Mutant and WT discs were stained in the same tube and imaged under identical confocal settings . Images are single cross-sections obtained on either a Leica TCS or a Zeiss LSM 700 and processed with Adobe Photoshop CS2 12 . 0 . 1 . Bgal staining was quantified as the percentage of pixels above background and normalized to WT levels . At least 50 wing imaginal discs were dissected from white1118 , scrib1 , and dlg40-2/Y larvae for each biological replicate , and at least two biological replicates were sequenced per genotype . Psc/Su ( Z ) 2 [XL26] FRT42 and control isogenized FRT42 wing discs were generated using UbxFLP; cell-lethal as described ( Newsome et al . , 2000 ) . Control tissue was isolated 5–6 days after egg lay ( AEL ) , while tumorous discs was isolated 7–8 days AEL to account for the developmental delay of tumor-bearing larvae . Poly-A transcripts were purified via two rounds of extraction using the MicroPolyAPurist kit ( Ambion , Austin , TX ) . mRNA was subsequently prepared for sequencing ( Dalton et al . , 2013 ) . Libraries were sequenced by 50-bp single-end reads on either the GAIIX Genome Analyzer or HighSeq2000 platform ( Illumina , San Diego , CA ) . Reads were aligned to the Drosophila melanogaster reference genome ( version 5 . 43 ) using TopHat run under default parameters ( Langmead et al . , 2009 ) . The number of reads from each replicate falling on each exon was counted using HTSeq ( Anders et al . , 2015 ) in the UNION mode , and the differential expression levels across all of samples were calculated using DESeq ( Anders and Huber , 2010 ) . Normalized value for gene expression is reported in a single ‘reads per kilobase gene length per million total reads’ ( RPKM ) value for each gene . Supplementary file 4 contains the sequencing and mapping statistics for each replicate , and Supplementary file 5 contains the number of differentially expressed genes for each genotype . For binding profile comparison , genes associated with Pc binding ( peak_hit , peak_near , gray_hit , gray_near ) in thoracic imaginal discs ( Kwong et al . , 2008 ) were defined as PcG targets . Genes upregulated at least twofold and having an RPKM value of at least 10 . 0 in wts mutant tissue were used to assess the overlap of the Scrib module and Hippo pathway mutant transcriptome profiles ( Oh et al . , 2013 ) . p-values for significance of overlap of transcriptome profiles was found using hypergeometric probability . Gene Ontology analysis was performed using GoStat ( Beissbarth and Speed , 2004 ) . Total RNA was isolated from at least 20 wing discs co-expressing eiger and miRGH with ms1096 GAL4 , along with outcrossed controls , using the RNeasy Mini Kit ( Qiagen , Valencia , CA ) , and cDNA was generated from 500 μg of RNA using Superscript II Reverse Transcriptase ( Life Technologies , Carlsbad , CA ) . Quantitative real-time PCR was performed using SYBR GreenER qPCR SuperMix ( Invitrogen , Carlsbad , CA ) on a StepOnePlus ABI Machine ( Applied Biosystems , Foster City , CA ) . Relative gene expression levels were quantified using the ΔΔCT method , after normalization to three endogenous control genes ( GAPDH , CG12703 , Cp1 ) . Average fold expression of at least four biological replicates is shown . Primer sequences are listed in Supplementary file 6 . Genomic DNA was isolated from adult flies using standard procedures . The upd3 fragment was amplified using Phusion High Fidelity Polymerase ( NEB ) and the following primers: 5′-GGTGGTACCTCGTACAATGGTTTAAAAATAGCTCGGCCAA-3′ and 5′-GGAAGGCCTCTCCTACACATCGAGCAGCATGGTCAACGAA-3′ . The 3-kb fragment was ligated into a pH-Pelican-attB vector and sequence was confirmed . Transformation into the attP2 landing site was performed by BestGene , Inc ( Chino Hills , CA ) . At least 10 wing discs were dissected and disassociated as described ( de la Cruz and Edgar , 2008 ) . Cells were counted using an EPICS XL flow cytometer ( Beckman–Coulter , Brea , CA ) . GFP+ and GFP− gates were generated based on a white1118 negative control sample . To calculate Relative Posterior Compartment Size , the number of GFP+ cells was divided by the total number of live cells and normalized to control discs . A two-tailed Student's t-test was used to calculate the p-values based on at least three biological replicates for each genotype . | The cavities and organs within our body are lined with epithelial cells , which connect to each other to form continuous barriers . These cells have a highly polarized structure in which different components are found at the top and bottom of cells . In the fruit fly and most other animals , three genes known as the Scribble module control the polarity of epithelial cells . If these genes are faulty , the cells lose their polarity , break the epithelial barrier , and grow rapidly to form a tumor . Most malignant tumors that form from epithelial cells have lost normal cell polarity , so understanding how the organization and growth of epithelial cells are linked is a critical question . It is not clear how the loss of cell polarity can drive tumor formation . Here , Bunker et al . used a technique called RNA sequencing to study the expression of genes in tumor cells that have mutations in the Scribble module . Hundreds of genes in the tumor cells had different levels of expression from the levels seen in normal fly cells . One of these is a gene called upd3 , which was expressed much more highly in tumor cells than in normal cells . This gene activates a signaling pathway—called the JAK/STAT pathway—that promotes cell growth and division in many animals . Bunker et al . found that experimentally lowering the activity of the JAK/STAT pathway reduced the growth of the tumor cells that had lost normal polarity . Further experiments show that disrupting the layer of epithelial cells activates two other signaling pathways that work together to switch on the upd3 gene when cell polarity is lost . Proteins belonging to the Polycomb Group also control the expression of upd3 and other genes involved in cell growth by altering how genetic material is packaged in cells . The similarities between this response and the response to tissue damage suggest that the loss of polarity drives tumor formation through an unstoppable wound-healing reaction . Therefore , Bunker et al . 's findings link the formation of epithelial tumors to the signaling pathways that control the repair of damaged tissues . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2015 | The transcriptional response to tumorigenic polarity loss in Drosophila |
We generated two new genetic tools to efficiently tag genes in Drosophila . The first , Double Header ( DH ) utilizes intronic MiMIC/CRIMIC insertions to generate artificial exons for GFP mediated protein trapping or T2A-GAL4 gene trapping in vivo based on Cre recombinase to avoid embryo injections . DH significantly increases integration efficiency compared to previous strategies and faithfully reports the expression pattern of genes and proteins . The second technique targets genes lacking coding introns using a two-step cassette exchange . First , we replace the endogenous gene with an excisable compact dominant marker using CRISPR making a null allele . Second , the insertion is replaced with a protein::tag cassette . This sequential manipulation allows the generation of numerous tagged alleles or insertion of other DNA fragments that facilitates multiple downstream applications . Both techniques allow precise gene manipulation and facilitate detection of gene expression , protein localization and assessment of protein function , as well as numerous other applications .
Comprehensive gene annotation is a central challenge in the post-genomic era . Drosophila melanogaster offers more sophisticated genetic approaches and tools to assess gene function and expression than other multicellular model organisms ( Bier et al . , 2018; Cox et al . , 2017; Germani et al . , 2018; Heigwer et al . , 2018; Kanca et al . , 2017; Kaufman , 2017; Simpson and Looger , 2018 ) . A versatile tool used for functional gene annotation in Drosophila is MiMIC , a Minos-based transposon that integrates a Swappable Integration Cassette ( SIC ) in the genome ( Venken et al . , 2011a ) . MiMIC SICs contain a cassette nested between two attP sites that can be exchanged with any DNA sequence flanked with attB sites through Recombinase Mediated Cassette Exchange ( RMCE ) by ΦC31 integrase . When a MiMIC is integrated in an intron of a gene flanked on both sides by coding exons ( hereafter referred to as a coding intron ) , the SIC can easily be exchanged with an artificial exon that encodes Splice Acceptor ( SA ) - ( GGS ) 4 linker-EGFP-FIAsH tag-StrepII tag-TEV protease cleavage site-3XFlag- ( GGS ) 4 linker-Splice Donor ( SD ) ( abbreviated as GFP tag ) ( Venken et al . , 2011a; Nagarkar-Jaiswal et al . , 2015a; Nagarkar-Jaiswal et al . , 2015b ) . The GFP-tagged endogenous proteins report the subcellular localization of the gene product and are functional in ~75% of tested genes ( Nagarkar-Jaiswal et al . , 2015a ) . Importantly functional GFP-protein traps can be used for multiple assays . These include chromatin Immunoprecipitation ( ChIP ) of transcription factors ( Nègre et al . , 2011 ) , Immunoprecipitation ( IP ) -Mass Spectroscopy ( MS ) ( David-Morrison et al . , 2016; Neumüller et al . , 2012; Yoon et al . , 2017 ) , rapid conditional removal of gene products ( Caussinus et al . , 2011; Lee et al . , 2018b; Nagarkar-Jaiswal et al . , 2015a; Neumüller et al . , 2012; Wissel et al . , 2016 ) and sequestration of tagged proteins ( Harmansa et al . , 2017; Harmansa et al . , 2015 ) . Hence , tagging an endogenous gene with GFP enables numerous applications to dissect gene function . The SIC in MiMICs can also be replaced by an artificial exon that encodes SA-T2A-GAL4-polyA signal ( abbreviated as T2A-GAL4 ) ( Diao et al . , 2015; Gnerer et al . , 2015; Lee et al . , 2018a ) . T2A-GAL4 creates a mutant allele by truncating the protein at the insertion site but also expresses GAL4 with the spatial-temporal dynamics of the targeted gene . Hence , T2A-GAL4 facilitates the replacement of the gene of interest with fly or human UAS-cDNAs ( Bellen and Yamamoto , 2015; Şentürk and Bellen , 2018; Wangler et al . , 2017; Lee et al . , 2018a ) , allowing one to assess putative disease-associated variants and permitting structure-function analysis of the protein of interest . Moreover , these gene-specific GAL4 stocks can be used to drive a variety of UAS constructs to further identify and probe the function of the cells expressing the gene using UAS-Fluorescent proteins or numerous other UAS constructs ( Venken et al . , 2011b ) . This is especially useful for genes that are not abundantly expressed , providing a means to amplify the signal , as GAL4 drives overexpression of the UAS transgenes ( Diao et al . , 2015; Lee et al . , 2018a ) . In summary , MiMIC applications allow the acquisition of valuable data about the function of the gene as well as the cells in which the gene is expressed . Given the usefulness of MiMICs , the Drosophila Gene Disruption Project ( GDP ) ( http://flypush . imgen . bcm . tmc . edu/pscreen ) has generated and mapped 17 , 500 MiMIC insertion stocks ( Nagarkar-Jaiswal et al . , 2015a; Venken et al . , 2011a ) . This collection includes insertions within introns for ~1860 genes , each of which can be converted to a GFP-tagged protein trap and/or a T2A-GAL4 gene trap ( Nagarkar-Jaiswal et al . , 2015a; Nagarkar-Jaiswal et al . , 2015b; Lee et al . , 2018a ) . However , we needed to develop a complementary strategy to generate resources for genes that do not have a MiMIC randomly inserted within a coding intron . To that end , we recently developed CRIMIC ( CRISPR mediated Integration Cassette ) ; a Cas9/CRISPR Homology Directed Repair ( HDR ) mediated approach that integrates a modified SIC ( attP-FRT-SA-T2A-GAL4-polyA-FRT-attP ) in a coding intron of choice . This approach greatly expands the number of genes that can be tagged using MiMIC-like technology from 1860 to ~6000 ( Lee et al . , 2018a ) allowing about forty percent of Drosophila protein coding genes to be targeted with SICs . RMCE cassettes can either be injected into embryos as part of a circular plasmid or can be circularized in vivo from an initial insertion locus in the genome through Cre/loxP or Flp/FRT mediated recombination ( Diao et al . , 2015; Nagarkar-Jaiswal et al . , 2015b ) . Importantly , RMCE cassettes can replace a SIC in either orientation with equal probability due to inverted symmetric attP sequences . Therefore 50% of the insertions are inserted in the opposite orientation of transcription and will not be included in the mature mRNA . Hence , only half of all successful exchange events will result in protein or gene trap lines . Here , we show that by combining GFP-protein traps and T2A-GAL4 gene traps in a single RMCE construct , named Double Header ( DH ) , we significantly increased the number of productive RMCE events for MiMIC/CRIMIC containing genes to generate protein or gene trap alleles . Importantly , we expand the ability to target SICs into genes regardless of the presence of introns to allow access to virtually any gene in the fly genome based on CRISPR/Cas9-mediated HDR . This provides a means to create robust null alleles with simple screening , and to convert the SIC insertion using any DNA , creating scarless modifications to facilitate numerous downstream applications .
SICs in coding introns can be converted into GFP-protein traps or T2A-GAL4 gene traps through RMCE . However , because RMCE of SICs in MiMICs and CRIMICs can occur in either orientation , only one out of two events produces a tag that is incorporated in the gene product . Moreover , each RMCE experiment generates only a protein trap or a gene trap , requiring two independent injections or crosses to generate both reagents . In order to reduce the effort to generate these genetic tools we engineered Double Header ( DH ) , a construct that combines the two key RMCE cassettes to replace intronic MiMICs and CRIMICs: ( 1 ) SA-T2A-GAL4-polyA ( DHT2A-GAL4 ) which generates a gene trap that expresses GAL4 in the expression domain of the gene while typically inactivating gene function , in one orientation and ( 2 ) SA-GFP-SD ( DHGFP ) which generates a GFP-protein trap , in the opposite orientation . Hence , insertion of DH via RMCE in a MiMIC in either orientation should result in two valuable reagents . This compound RMCE cassette is flanked by two inverted ϕC31 attB sites in a vector backbone that contains other features including mini-white as shown in Figure 1A . The presence of white in the plasmid backbone allows for a counter selection against the integration of the whole plasmid when incorporated into a white- background , ensuring that only the DNA between the attB sites integrates . The artificial exons that are integrated into MiMIC or CRIMIC sites need to be in frame with the preceding Splice Donor ( SD ) to create a functional tag . Because exon/intron boundaries can occur at any one of three positions in a codon ( phase 0 , +1 , +2 ) , we generated three different DH plasmids ( Sequences can be found in Supplementary file 1 ) . Each construct contains the same codon phase for the two modules . Given that the RMCE cassette is about 4 . 8 kB , or about 1 . 5 times the size of the single T2A-GAL4 cassette , we anticipated a lower integration rate . We tested the integration efficacy by injecting 30 strains carrying a coding intronic MiMIC insertion . On average , ~400 embryos were injected for each MiMIC line together with a plasmid that encodes the ΦC31 integrase . We screened for the loss of yellow+ in the progeny of the injected animals , as MiMICs carry the yellow+ marker ( Figure 1—figure supplement 1A ) . We did not observe the integration of white indicating that a single RMCE event leading to the integration of the entire plasmid is rare . For 16 out of 30 MiMICs , we were able to isolate yellow- flies that carry a DH integration ( Figure 1B; Figure 1—figure supplement 1B ) . We determined the orientation of DH inserts through single fly PCR ( Figure 1—figure supplement 2 ) and determined the orientation of the cassette in 47 out of 72 DH RMCE events . For 6 MiMICs we obtained both orientations and for the other 10 MiMICs we obtained one or the other orientation ( Figure 1B; Figure 1—figure supplement 1B ) . Hence , we generated a total of 22 new reagents , increasing the overall rate of productive RMCE events by injection ( Nagarkar-Jaiswal et al . , 2015a ) . Note that the integration efficiency of DH construct by injection is lower than the smaller SICs:~50% versus~66% ( Nagarkar-Jaiswal et al . , 2015a ) . However , the number of productive events increases tagging efficacy somewhat as every successful event produces a useful line: 74% versus 66% . In an effort to avoid embryo injections and to increase integration efficiency of DH , we developed an in vivo RMCE strategy using genetic crosses similar to Trojan exons developed for T2A-GAL4 ( Diao et al . , 2015 ) . We integrated the same constructs as in Figure 1A , one for each reading frame , in the genome through P-element mediated transformation . These insertions serve as jump-starter constructs because the RMCE cassettes in these transgenes can be excised from their initial landing sites by expressing Cre recombinase in germ line cells ( Figure 2 ) . The crossing scheme is outlined in Figure 2-figure supplement 1 . Figure 3A . We generated jump-starter insertions in second and third chromosomes for all three possible phases of DH and generated double balanced stocks for subsequent crosses . We tested the efficacy of integration by crosses for DH for third chromosome MiMICs . For 12 out of 13 MiMICs tested we obtained integration of DH . We determined the orientation of DH for 48 out of 102 insertions by PCR . The inconclusive insertions either showed no PCR amplification in one end or both ends of the MiMIC ( 48/102 ) or in rare cases conflicting PCR amplification that indicates integration in both orientations ( 6/102 ) . Interestingly 44/54 inconclusive inserts happened in only two of the MiMICs , indicating locus specific issues . For 6 out of 12 MiMICs we obtained both orientations and for six we obtained one or the other , resulting in 18 tagged genes ( Figure 2—figure supplement 1 ) . Hence , the genetic strategy is about twice as efficient ( 18 constructs for 13 crossed MiMICs versus 22 constructs for 30 injected MiMICs ) as the injection strategy in generating RMCE events and requires significantly less effort . We proceeded to test whether DH functions as expected . We determined the expression patterns of genes tagged in both orientations in third instar larval brain and adult brains for MI01487 ( kibra ) , MI05208 [5-HT2B ( 5-hydroxytryptamine receptor 2B ) ] , MI06794 [Lgr4 ( Leucine-rich repeat-containing G protein-coupled receptor 4 ) ] , MI06872 ( CG34383 ) , MI08614 [Dgk ( Diacyl glycerol kinase ) ] , MI11741 ( CG12206 ) and MI15073 ( CG9132 ) ( Figure 3 ) . As we selected a few MiMICs that were previously tagged with T2A-GAL4 by the Gene Disruption Project as positive controls ( Lee et al . , 2018a; Diao et al . , 2015 ) ( MI01487 ( kibra ) , MI06794 ( Lgr4 ) , MI06872 ( CG34383 ) , MI08614 ( Dgk ) , and MI11741 ( CG12206 ) , we were able to compare expression patterns obtained by DHT2A-GAL4 to expression patterns obtained by single T2A-GAL4 ( http://flypush . imgen . bcm . tmc . edu/pscreen/rmce/ ) . In all cases the expression pattern was very similar to what was previously reported ( Figure 3 ) ( Lee et al . , 2018a ) . In addition , tracheal expression of CG12206 is consistent with a previous report ( Chandran et al . , 2014 ) and the 5HT2BT2A-GAL4 expression pattern in the adult brain matches an independently generated T2A-GAL4 ( Gnerer et al . , 2015 ) ( Figure 3—figure supplement 1 ) . In all cases , the DHGFP insertions show consistent patterns of expression in third instar larval brains , albeit at much lower levels than the DHT2A-GAL4 insertions at the same MiMIC site ( Figure 3 ) . Note that in adult brains almost no signal of DHGFP was detected , in agreement with previous observations , with the exception of MI15073 which shows ubiquitous expression , ( Diao et al . , 2015; Lee et al . , 2018a ) ( Figure 3—figure supplement 1 ) . These results indicate that neither the size nor the design of DH alters the functionality or expression patterns of the tagged genes and proteins . As the GFP protein traps should be able to report the subcellular localization of the tagged protein we turned to tissues where subcellular localization and specific cell expression is easily assessed . We therefore dissected and stained egg chambers with anti-GFP . We were easily able to visualize the GFP tagged proteins in the seven protein traps previously examined ( Figure 4 ) . The tagged proteins are shown in green and the nuclei are stained with DAPI in red . Kibra is detected in somatic follicle cell cytoplasm , including some migratory border cells . 5HT2B is expressed in both somatic and germline cells including the oocyte . Lgr4 is localized to germline nurse cell nuclei and is enriched in the oocyte anterior-dorsal and ventral cortex . CG34383 is mostly present in follicle cells , especially in their apical domain . Dgk is observed in nurse cell and follicle cell nuclei as well as their cytoplasm . CG12206 is quite enriched in the cytoplasm of centripetal cells and CG9131 is present in both germ cells and follicle cells and enriched in polar cells . In summary , GFP protein tagging with DH can be used to determine the cellular and subcellular localization of tagged proteins . To tag a gene containing a MiMIC/CRIMIC , the SIC should be integrated within a suitable coding intron , leaving about 50–60% of all Drosophila genes that encode proteins inaccessible . Targeting genes without introns by directly fusing tags in the proper reading frame has been very difficult because HDR is much less efficient than non-homologous end joining ( NHEJ ) ( Gratz et al . , 2014 ) . Hence , expanding the range of targetable genes requires precise , seamless genome editing . Gene editing using HDR is well suited to modifying genes without introducing unwanted changes ( Bier et al . , 2018 ) . HDR repairs double stranded DNA breaks using a donor template that contains two homology regions , each typically about 1000 nucleotides , which flank the desired changes . Recombination on either side of the break replaces the regions with the template , precisely modifying the locus . We therefore developed a novel SIC compatible with HDR ( Figure 5;Figure 5—figure supplements 1 and 2 ) that could be targeted to loci regardless of the presence of introns . To make a SIC that is HDR compatible , three features are important: ( 1 ) a dominant marker for screening that is compact for ease of insertion via HDR ( Li et al . , 2014 ) , ( 2 ) a method to insert the SIC in the desired location , and ( 3 ) a strategy to remove the SIC for replacement with the desired end product . To design a compact marker that is compatible with Golden Gate cloning we focused on the yellow gene which has well characterized enhancers ( Geyer and Corces , 1987 ) . We identified a 575 nucleotide regulatory region that when fused to the promoter and yellow coding sequence creates a 2 . 9 kilobase cassette that reliably drives expression only in the wing ( Figure 5—figure supplement 2 ) . We refer to this cassette as ywing2+ . To enable targeting ywing2+ into precise locations in the genome , we first define a region ( or gene ) of interest ( ROI ) flanked by two Cas9 target sites comprised of a 20 nucleotide guide sequence and an NGG PAM ( Jinek et al . , 2012; Sternberg et al . , 2014 ) ( Figure 5A ) . We then design a HDR donor template with ywing2+ flanked by homology regions . As the donor template removes part of the Cas9 target sequences neither the donor cassette nor the final product are cleaved upon HDR . Injecting the donor template along with sgRNA expression plasmids into embryos carrying a germline-specific source of Cas9 , followed by screening offspring for yellow+ wings provides a straightforward method to generate robust null alleles for the gene . Lastly , to make the cassette removable , we flanked the SIC with the nucleotides ‘GG’ and ‘CC’ upstream and downstream of the ywing2+ marker , respectively . Upon insertion , this creates two novel Cas9 target sites that are not present in the endogenous sequence ( box inset of Figure 5A ) , which can be used to remove the inserted cassette for final replacement via a second round of HDR . Finally , to facilitate cloning , the ywing2+ cassette was made compatible with Golden Gate assembly ( Engler et al . , 2009 ) . We also generated templates for designing replacement HDR constructs containing GFP and mCherry for protein tags or T2A-Gal4 that are compatible with Golden Gate assembly ( Figure 5—figure supplement 1 ) . We tested the efficacy of replacing the coding sequence of 10 loci with ywing2+ ( Table 1 ) . Nine out of ten injections led to successful integration of the cassette . We injected an average of ~500 embryos for each gene and recovered 1 to 6 independent founder lines for a total of 22 insertion events or ~2 founder animals/gene . Sanger sequencing confirmed the correct insertion of all but one HDR event , in which the entire plasmid backbone had been integrated . As shown in Table 1 , seven insertions are homozygous lethal . To test whether the lethality was specific to the removal of the targeted gene , we attempted rescue of the lethality with a genomic duplication of the locus for four genes ( Nmnat ( Nicotinamide mononucleotide adenylyltransferase ) , CG13390 , Med27 ( Mediator complex subunit 27 ) , and CG11679 , and tested for failure to complement molecularly defined deletions for two genes ( ubiquilin and Nmnat ) ( Zhai et al . , 2006 ) . In all cases , lethality mapped to the targeted locus showing that no second-site lethal mutations were induced in these lines . For the gene almondex ( amx ) , the ywing2+ insertion produced flies that were female sterile . Female sterility was previously observed for amx and a genomic fragment previously reported to rescue female sterility likewise rescued this phenotype in amxΔCDS , ywing2+ ( Cohorts for Heart and Aging Research in Genomic Epidemiology consortium et al . , 2016 ) . For the gene Stub1 ( STIP1 homology and U-box containing protein 1 ) , four positive lines were recovered; two are homozygous lethal while two are viable and fertile . Sanger sequencing confirmed the correct insertion of the cassette in all four lines , suggesting that the gene is not essential . Hence , the lethality is either caused by off-target cleavage events or the presence of a floating lethal mutation in the original strain . Thus while off-target cleavage may have occurred , this evidence suggests that it is not common , in agreement with what we have observed when we use CRIMIC ( Lee et al . , 2018a ) . In summary , we created null alleles for nine genes and show that the ywing2+ knock in cassette inserted precisely based on Sanger sequencing . The ywing2+ cassette is designed to introduce two new gRNA target sites upon replacing the endogenous locus . These newly introduced gRNA target sites can now be used for the removal of the cassette via CRISPR/Cas9 mediated HDR and replacement with the desired DNA sequence . We attempted to incorporate protein tags for five genes ( Table 2 ) . We successfully incorporated tags for Nmnat , Stub1 , CG11679 and Med27 but failed for amx . We tagged Nmnat and Stub1 with GFP , and CG11679 and Med27 with Flag tags ( see Mat . and Meth . ) . Internally GFP-tagged Nmnat ( Nmnat::GFP::Nmnat ) and C-terminally Flag tagged CG11679 ( CG11679::Flag ) reverted the lethality of the ywing2+ knock in allele and hence produced functional proteins . However , the C-terminal Flag-tagged Med27 ( Med27::Flag ) is recessive pupal lethal , similar to the ywing2+ knock in allele , suggesting that the C-terminal Flag tag disrupts protein function . Because the loss of Stub1 ( Table 1 ) does not result in an overt phenotype , we cannot determine if Stub1::GFP is functional but Sanger sequencing showed that the replacement of ywing2+ with Stub1::GFP happened precisely . Taken together , the data indicate that Cas9 mediated cassette replacement occurred correctly for four out of five genes . To highlight the utility of the ywing2+ scarless replacement strategy , we performed a structure-function analysis of Nmnat . Nmnat is an enzyme with Nicotinamide adenine dinucleotide ( NAD ) synthase activity that also functions as a molecular chaperone ( Zhai et al . , 2006; 2008 ) . Additionally , the Nmnat family is highly conserved , required for neuronal survival and protects neurons from a variety of neurodegenerative insults ( Ali et al . , 2013; Brazill et al . , 2017; Lau et al . , 2009 ) . A previous report generated a Nmnat null allele and determined that its loss causes lethality in first instar larva ( Zhai et al . , 2006 ) . For our structure-function analysis , we first created a null allele of Nmnat by replacing the entire coding sequence ( CDS ) with ywing2+ , NmnatΔCDS , ywing2+ ( see Table 1 ) . The resulting flies are homozygous first instar larval lethal , consistent with a known protein null , and can be rescued by a 3 kb Nmnat transgene known to rescue the lethality associated with the loss of Nmnat ( Zhai et al . , 2006 ) . We then replaced the ywing2+ SIC with internally GFP-tagged versions of wild-type and three variants of Nmnat ( Table 2; Figure 6A , B ) . These variants of Nmnat are known or predicted to affect specific molecular functions of Nmnat in vitro ( Figure 6B ) : ( 1 ) Nmnat::GFP::NmnatW129G reduces NAD synthase activity ( Zhai et al . , 2006 ) ( 2 ) Nmnat::GFP::NmnatΔ251…257 disrupts the ATP binding motif required for chaperone function ( Zhai et al . , 2006; Ali et al . , 2016 ) , and ( 3 ) Nmnat::GFP::NmnatC344S , C345S is predicted to disrupt critical palmitoylation sites that , in vertebrates , are required for membrane association and protein turnover ( Lau et al . , 2010; Mayer et al . , 2010 ) . All cassettes correctly replaced the ywing2+ SIC based on Sanger sequencing . We chose three independent lines of wild-type and each variant for further analysis . Homozygous Nmnat::GFP::NmnatWT flies are viable and fertile , suggesting that the internal GFP tag does not overtly affect Nmnat function . In contrast , homozygous Nmnat::GFP::NmnatΔ251…257 or Nmnat::GFP::NmnatΔ251…257/NmnatΔ4790-1 animals die as 1st instar larvae similar to NmnatΔCDS , ywing2 , suggesting that Nmnat::GFP::NmnatΔ251…257 behaves as a null allele ( Zhai et al . , 2006 ) . Homozygous Nmnat::GFP::NmnatW129G flies grow slowly and die prior to pupariation at late 3rd instar larval stage , suggesting that it is a hypomorphic allele . Finally , the Nmnat::GFP::NmnatC344S , C345S flies that lack the putative palmitoylation sites are viable and fertile , but exhibit reduced lifespan relative to Nmnat::GFP::NmnatWT controls , indicating that it behaves as a weak hypomorph . To determine protein levels and localization , we stained the brains of heterozygous GFP-tagged animals ( Figure 6C ) . Antibody staining against GFP showed robust staining for wild-type Nmnat::GFP::NmnatWT ( Figure 6C ) . Most immunofluorescence signal is confined to the nucleus and cell body of neurons , but low levels of signal are observed in axons ( Figure 6C ) . In contrast , Nmnat::GFP::NmnatΔ251…257 surprisingly produced no detectable protein ( Figure 6C ) , suggesting that the removal of the ATP-binding domain severely affects the stability of the mRNA or protein , consistent with the observation that is a genetic null allele . On the other hand , Nmnat::GFP::NmnatW129G showed a mildly reduced signal when compared to wild-type , and the GFP localization was also seen mainly in the nucleus and cell body . Finally , Nmnat::GFP::NmnatC344S , C345S flies show an obvious increase in GFP levels consistent with the hypothesis that this site is required for protein degradation as previously shown in vertebrate NMNAT ( Mayer et al . , 2010; Milde et al . , 2013; Lau et al . , 2010 ) . In summary , our data document the ability to perform structure-function analyses in the endogenously GFP tagged locus .
Here , we describe two methods to facilitate endogenous tagging and functional annotation of genes in Drosophila . DH doubles the success rate of RMCE using readily available MiMIC/CRIMIC lines to generate GFP protein traps or T2A-GAL4 gene traps . On the other hand , the ywing2+ SIC-mediated two-step scarless gene tagging strategy offers a means to manipulate more than 6000 genes that cannot be targeted with the artificial exon approaches . Together these technologies facilitate the use of MiMICs and expand the capabilities of cassette swapping to include virtually all genes in Drosophila . Recently , two other compound RMCE cassettes that encode two different modules in opposing orientations have been reported ( Fisher et al . , 2017; Nagarkar-Jaiswal et al . , 2017 ) . Flip-Flop contains a protein trap that can be inverted by Flp-FRT to a mutant allele , conditionally inactivating the gene in mitotic and postmitotic cells . The mutagenic module encodes SA-T2A-mCherry-polyA ( Nagarkar-Jaiswal et al . , 2017 ) . In FlpStop on the other hand the non-mutagenic orientation does not encode a protein , but inversion by Flp leads to a gene trap SA-stop cassette-polyA ( Fisher et al . , 2017 ) . Hence , these methods create conditional alleles . In contrast , DH creates alleles that are final and cannot be inverted or altered by Flp expression . This allows the use of Flp/FRT for independent manipulations in the DH background . Previously , Nagarkar-Jaiswal et al . , 2015b used a Flp-mediated in vivo mobilization strategy to create GFP-protein traps for intronic MiMIC containing genes whereas Diao et al . ( 2015 ) used a Cre-mediated in vivo mobilization scheme to create T2A-GAL4 gene traps . Both the GFP tag and T2A-GAL4 provide complementary means to assess gene function and expression . By combining the two in a single vector , DH greatly improves the rate of RMCE , the breadth of applications , and the amount of labor involved . We observed that although T2A-GAL4 is highly successful in marking the gene expression domains in the adult brain , for most genes the corresponding GFP-tagged protein signal in adult brains is often weak , consistent with previous results ( Lee et al . , 2018a; Diao et al . , 2015 ) . However , all the lines tested in the brain allow us to rapidly and reliably determine the cellular and subcellular localization of the GFP tagged proteins in egg chambers . Moreover , even when these GFP-tagged proteins cannot be used to detect the gene product , they can still be very useful for biochemical applications or to knock down the gene through a variety of methods to create conditional alleles ( Caussinus et al . , 2011; Harmansa et al . , 2017; Harmansa et al . , 2015; Nagarkar-Jaiswal et al . , 2015a; Neumüller et al . , 2012; Lee et al . , 2018b ) . Interestingly for three out of 28 MiMICs , we detected numerous DH RMCE inserts , judged by loss of the yellow marker , but for many of these events we could not determine conclusively the orientation or presence of DH by PCR . However , these false positives are easily identified by single fly PCR ( Figure 1—figure supplement 2 ) . Moreover , we could identify positive events in the MiMICs where the false positive rate was high , showing that the high false positive rate for these MiMICs does not limit the technique . Finally , the GAL4 >UAS system can also be used to assess the function of cells , particularly neurons . Multiple features of neurons , including electrophysiological properties , can be modulated or assessed using established UAS constructs ( Venken et al . , 2011b ) . These include the UAS-Tetanus Toxin , UAS-Kir2 . 1 or UAS-Shibirets to silence neurons ( Sweeney et al . , 1995; Baines et al . , 2001; Kitamoto , 2001 ) ; UAS-TRPM8 or UAS-ChannelRhodopsin 2 ( ChR2 ) to activate neurons ( Peabody et al . , 2009; Schroll et al . , 2006 ) ; and UAS-GCaMP to assess changes in Calcium concentrations ( Chen et al . , 2013 ) or UAS-ASAP2 that acts as voltage sensor ( Yang et al . , 2016 ) to assess neuronal activity . Hence , the cells that express T2A-Gal4 associated with a specific gene can be manipulated in numerous ways . Given that 50–60% of the protein coding genes do not contain suitable introns , numerous genes are not amenable for tagging based on our approaches . Inserting tags in genes that lack large ( >150 bp ) introns creates two main challenges: ( 1 ) screening for a precise rare gene editing event is very time consuming and ( 2 ) inserting extraneous sequences for RMCE within or near coding regions often create mutations and indels which disrupt protein function . Scarless gene editing offers obvious advantages for manipulating these loci , however , few options currently exist . Scarless gene editing can be achieved through the use of single-stranded DNA donors ( Gratz et al . , 2014; Xue et al . , 2014 ) . However , they are limited in size to ~200 nucleotides by current synthesis methods ( Korona et al . , 2017 ) . Accordingly , sgRNA sites must be close to the specific nucleotides to be edited and because they cannot carry visible markers they require laborious screening methods to find flies carrying the correct gene editing event . Two strategies have been proposed to integrate fluorescent markers in fly genes using double stranded DNA donor plasmids and remove them to perform scarless genome editing ( reviewed in Bier et al . , 2018 ) . One method , the Scarless-dsRed system ( http://flycrispr . molbio . wisc . edu/scarless ) , relies on the piggyBac transposase to remove a dominant marker by precise excision after the gene editing event has been confirmed . However , as of yet no data have been reported to determine its efficiency or efficacy . While we were developing and testing our approach , a similar method , pGEM-wingGFP-tan , was reported ( Lamb et al . , 2017 ) . This method integrates two Cas9 target sequences on either end of a GFP marker driven by a wing promotor to replace a locus . Unlike ywing2+ which is readily visible in adults , the reported wing-GFP marker needs to be scored within a narrow developmental stage in pupae with a fluorescent microscope ( Lamb et al . , 2017 ) . Moreover , Lamb et al . ( 2017 ) report a high rate of backbone insertion , where we observed only a single case out of 11 genes with our ywing2+ approach . Since the methodology was only applied to a single gene , we cannot compare our data with Lamb et al . ( 2017 ) . For 9 out of 10 genes that we targeted with ywing2+ , we obtained at least one correctly inserted SIC from an injection of ~500 embryos . The ywing2+ marker is easy to score in adult flies and is compatible with Golden Gate and Gibson assembly ( Engler et al . , 2009; Gibson et al . , 2009 ) , greatly facilitating its application to virtually any locus within the genome . The first step creates a null allele which provides an essential reference point for all subsequent genetic and molecular manipulations . Given that most genes that lack introns are rather small , they are poor targets for chemical or transposon mediated mutagenesis , although they can be targeted with CRISPR based on NHEJ . The ywing2+ SIC offers an easy way to completely remove these small genes and is not labor intensive . The major advantage of the ywing2+ SIC is that it creates a highly versatile line . Multiple manipulations within the region of interest can be performed in parallel in an essentially isogenic background using this SIC . Although we did not observe widespread off-target mutations , we suggest the use of multiple lines where possible . We have shown that the cassette swapping via the ywing2+ SIC occurs precisely with a high success rate and demonstrated its usefulness both for gene tagging and for structure function analysis . In summary , the two methodologies and accompanying tool kits presented here complement and expand existing MiMIC and CRIMIC approaches . The combination of these methodologies should enable endogenous tagging and manipulation of most fly genes , an invaluable resource for the fly research community .
Sequence of the DH plasmids can be found in Supp . file . Briefly , SA- EGFP-FlAsH-StrepII-TEVcs-3xFlag-SD cassette was PCR amplified from pBS-KS-attB1-2-PT-SA-SD-pX ( corresponding to the codon phase ) -EGFP-FIAsH-StrepII-TEV-3xFlag ( DGRC # 1298 , [Venken et al . , 2011b] ) with tags_for_BsiWI , tags_rev_AvrII primers . This fragment is cloned in , a modified pTGEM plasmid of pX ( corresponding to the codon phase ) where the loxP site before 3XP3RFP is deleted and a BsiWI site is integrated after an AvrII site using BsiWI and AvrII restriction sites . Resulting vector is cut with XbaI-BsiWI and cloned into pC- ( loxP2-attB2-SA ( 1 ) -T2A-Gal4-Hsp70 ) ( Addgene # 62955 , [Diao et al . , 2015] ) modified to include a BsiWI site after SD , generating DH pX . A sub-region of the yellow dominant marker from P{EPgy2} ( Bellen et al . , 2011 ) that contains the promoter , coding sequence and UTRs was subcloned into the plasmid pattB ( Accession # KC896839 [Bischof et al . , 2013] ) using oligos DLK0048 and DLK0049 ( see Supplementary file 1 for table showing oligonucleotides sequences used ) flanked by XhoI and XbaI to make the vector pBS II SK-attB yMP w+ . Either the full sequence of the yellow enhancers ( oligos DLK0054 and DLK0056 ) ( Geyer and Corces , 1987 ) or sub-fragments ( oligos DLK0054 and DLK0057 or DLK0055 and DLK0056 - see Figure 5—figure supplement 2 ) were then subcloned into pattB yMP w+ to make pattB expression constructs . Once expressions of the markers were verified , miniwhite+ and the loxP sites were removed by digestion with ApaI and NotI , the ends blunted with DNA Polymerase I , Large ( Klenow ) Fragment ( NEB ) and ligated using T4 Ligase ( NEB ) . As only the full body , full wing and wing2 gave positive expression , wing2 was chosen for use as the most compact construct and a multiple cloning site was then added using annealed oligos DLK0022 and DLK0023 cut with SpeI and XbaI into XbaI digested pattB ywing2+ , and sequence verified for insert direction to make functional plasmids for ϕC31 mediated transgenesis . The sequence of pattB ywing2+ can be found in Supplementary file 1 . Three versions of p{ywing2+} were cloned: for BsmBI , BsaI , and BbsI . First , annealed oligos ( DLK320 and DLK338 , DLK322 and DLK339 , and DLK324 and DLK340 for BsmBI , BbsI and BsaI , respectively ) containing the appropriate SacI overhang and ends-in TypeIIS restriction sites with a short intervening random nucleotide spacer inserted into the pM14 plasmid backbone ( Lee et al . , 2018a ) digested with enzymes SacI and EcoRV . The ywing2+ dominant reporter was then subcloned by PCR using oligos DLK326 and DLK327 from p-attB ywing2+ into the pM14 + spacer backbone digested with BbsI , BsaI or BsmBI . The sequence of p{ywing2+} can be found in Supplementary file 1 . Three versions of p{GFP} were cloned: for BsmBI , BsaI , and BbsI . Annealed oligos ( DLK461 and DLK462 , DLK463 and DLK464 , and DLK465 and DLK466 for BsmBI , BbsI and BsaI , respectively ) containing the appropriate SacI overhang and ends-in TypeIIS restriction sites were inserted into the pM14 plasmid backbone ( Lee et al . , 2018a ) digested with enzymes SacI and EcoRV to make p{spacer-L-L} where L denotes ( GGS ) 4 linker . The linker-GFP sequence was generated by PCR from pM14 ( Lee et al . , 2018b ) and using oligos DLK225 and DLK300 and subcloned into p{spacer-L-GFP-L} . The second linker was then added from annealed oligos DLK554 and DLK555 to make the BbsI version of p{spacerL--L} . Finally , the linker-GFP-linker was subcloned into versions of p{spacer-L -L} for BsaI and BsmBI to make plasmids compatible with each enzyme . Additional template vectors were produced for mCherry and T2A-GAL4 inserts which can also be found in Supp . file . mCherry was subcloned into p{spacer-L-L} from the Flip-Flop cassette which was designed with silent mutations to remove BsaI sites ( Nagarkar-Jaiswal et al . , 2017 ) and T2A-GAL4 inserts were subcloned from pM14 ( Lee et al . , 2018b ) into p{spacer-T2A} . Sequences can be found in Supplementary file 1 . Donor constructs were generated as previously described ( Housden and Perrimon , 2016 ) . Briefly , homology arms were PCR amplified from genomic DNA using Q5 polymerase ( NEB ) , run on an agarose gel and purified with the QIAquick Gel Extraction Kit ( Qiagen ) . The homology arms , pBH donor vector and either p{ywing2} or p{L-GFP-L} cassette were combined by Golden Gate assembly ( Engler et al . , 2009 ) using the appropriate type IIS restriction enzyme ( BbsI , BsaI , or BsmBI ) . The resulting reaction products were transformed into Stbl2 Chemically Competent Cells ( ThermoFisher ) , and plated overnight under kanamycin selection . Colonies were cultured for 24 hr at 30°C and DNA was prepared by miniprep ( QIAGEN ) . The entire homology arm and partial sequences of the adjacent cassette were verified prior to injection . Additional cloning information and sequences for all HDR donor plasmids can be found in Supp . file . sgRNA expression constructs were cloned into the vector pCFD3-dU6:3gRNA ( Addgene plasmid #49410 , Port et al . , 2014 ) using established protocols ( http://www . crisprflydesign . org/wp-content/uploads/2014/05/Cloning-with-pCFD3 . pdf ) . Sequences of sgRNAs can be found in Supplementary file 1 . Genomic fragments were cloned into pattB ( Ascension # KC896839 ( Bischof et al . , 2013 ) by PCR from wild-type genomic DNA and inserted BamHI/NotI for CG13390 ( oligos DLK823 and DLK824 ) and XhoI/AvrII ( into the XbaI site of pattB ) for Med27 ( oligos DLK966 and DLK969 ) MiMIC stocks are obtained from Bloomington Drosophila Stock Center . The fly lines to conduct RMCE of DH with crosses are: y[1] M{vas-int . Dm}ZH-2A w[*] , P{y[+mDint2]=Crey}1b;; Sb[1]/TM2 , Ubx[130] e[s]y[1] M{vas-int . Dm}ZH-2A w[*] , P{y[+mDint2]=Crey}1b; sna[Sco]/CyOy[1] w[*]; P{DHp0-7 [w+]}; TM2 Ubx[130] e[s]/TM6 , Tb[1] e[s]y[1] w[*]; P{DHp1-7 [w+]}; TM2 Ubx[130] e[s]/TM6 , Tb[1] e[s]y[1] w[*]; P{DHp2-4 [w+]}; TM2 Ubx[130] e[s]/TM6 , Tb[1] e[s]y[1] w[*]; Kr[If1] wg[Sp1]/CyO; P{DHp0-8 [w+]}y[1] w[*]; Kr[If1] wg[Sp1]/CyO; P{DHp1-6 [w+]}y[1] w[*]; Kr[If1] wg[Sp1]/CyO; P{DHp2-6A [w+]} Cas9 stocks for CRISPR experiments carried an isogenic chromosome on either X , II or III derived from y w and were either y1 M{nos-Cas9 . P}ZH-2A w* ( from Bloomington Drosophila Stock Center ) or y1 ( iso X ) w*; +/+; attP2 ( y+ ) {nos-Cas9 ( y+ ) } ( Ren et al . , 2013 ) in which the y + marker was mutagenized by injecting sgRNA expression plasmids ( pCFD3-y1 and pCFD3-y2 ) against the yellow coding sequence . The isogenic chromosomes ( X , II , III ) were sequenced using whole genome sequencing ( Human Genome Sequencing Center at Baylor College of Medicine ) . The sequence ( . BAM ) files are available on Zenodo ( https://zenodo . org/record/1341241 ) . " RMCE to generate DH insertions by injections is depicted in Figure 1—figure supplement 1A and in ( Nagarkar-Jaiswal et al . , 2017 ) . Briefly , the DH plasmid of the correct phase ( 500 ng/ul final concentration ) is mixed with ΦC31 integrase helper plasmid ( 400 ng/ul final concentration ) and injected in embryos of MiMIC stocks . The crossing scheme for generating DH insertions is depicted in Figure 2—figure supplement 1 . 5–7 crosses with 10–15 virgins of individual MiMIC lines are crossed with 7–10 males from phase appropriate DH transgenics on second chromosome , balanced for third chromosome . The vials are flipped every second day to prevent overcrowding . 5–10 crosses are set in the subsequent generations . The resulting individual y- flies are selected to set up stocks . Note that SICs are flanked by FRT sites in the newly generated CRIMIC alleles . Hence , the Flp/FRT mobilization schemes described by Nagarkar-Jaiswal et al . , 2015b cannot be used to generate GFP protein traps in these CRIMICs , given that expression of the Flp would excise the SIC . For the same reason , Flp/FRT cannot be used in combination with the CRIMIC T2A-GAL4 alleles for experiments that require both T2A-GAL4 and Flp . For each DH insertion stock four PCRs are set . Primer pairs are MiMIC_5’_for-GFP_DH_for , MiMIC_3’_rev-T2A_GAL4_rev , MiMIC_5’_for-T2A_GAL4_rev , and MiMIC_3’_rev-GFP_DH_for . Correct RMCE events result in 2 out of 4 successful amplicons ( Figure 1—figure supplement 2 ) . We isolated a number of lines where only one out of two amplicons was detected or gave conflicting results . These lines are not included in the analysis in this manuscript . Embryos ( designated G0 ) at less than one hour post-egg laying carrying the appropriate Cas9 allele were injected with a mixture of 150–200 ng/µl of donor plasmid and 25 ng/µl of each sgRNA expression construct , transferred to standard media after 24–48 hr , and crossed to y w flies . For experiments to insert the ywing2+ cassette , offspring were screened for presence of yellow+ wings several days after eclosion and crossed to appropriate balancers . For experiments to replace the ywing2+ cassette , G0 flies were crossed to y w flies carrying appropriate balancers and F1 offspring screened for loss of yellow+ wings . Individual F1 founders were backcrossed to y w flies carrying appropriate balancers , allowed to establish larvae , and screened for presence of the insert via PCR ( see Supplementary file 1 ) . Confocal imaging was conducted as in the previous study ( Lee et al . , 2018a ) . In brief , dissected adult brains were fixed in 4% paraformaldehyde/1xPBS overnight , then penetrated with 0 . 2% Triton X-100/1xPBS at 4°C overnight . The larval brains or other tissues were fixed in 4% paraformaldehyde/1 xPBS at 4°C for at least 2 hr , transferred to 0 . 5% Triton X-100/1xPBS for overnight 4°C incubation . For adult brains , the samples were vacuumed for 1 hr at room temperature , and left overnight in the same solution for penetration at 4°C . For immunostaining of GFP , the samples were incubated with anti-GFP antibody conjugated with FITC ( 1:500 ) ( Abcam , RRID: AB_305635 ) in 1xPBS with 0 . 5% Triton X-100 overnight . To increase signal , some samples used anti-GFP antibody ( 1:1000 ) ( Invitrogen , A11122 ) followed by incubation with secondary antibody ( Alexa Fluor 488-conjugated goat anti-rabbit IgG ) . Samples were cleared and mounted in RapiClear ( SunJin Lab Co . ) and imaged with a Zeiss LSM 880 under a Plan-Apochromat 20x/0 . 8 M27 objective with numerical aperture of 0 . 8 or Leica Sp8 Confocal Microscope under a HC PL APO 20x objective with numerical aperture 0 . 7 . Laser intensity and detector gains were adjusted as needed to increase signal-to-noise ratio and prevent signal saturation . | Organisms have tens of thousands of genes , but finding out exactly what they all do is one of the greatest challenges of modern genetics . To understand a gene’s job , it’s necessary to find out what gene is active in which tissue , where their proteins are located within the cell , and what happens when the sequence of a gene is altered or removed . This multi-step process of ‘annotating’ genes can be challenging in practice . One common approach is to make use of a DNA pattern called a MiMIC and insert it in a specific part of the gene called an intron . A tag for a protein that glows green under the microscope can then be added to a MiMIC to help visualize where and when the protein is being expressed . MiMICs can also be used to integrate a system called T2A-GAL4 , which typically creates a severe mutation in the gene and allows to track the timing of when and where the gene is expressed . This helps to discover the role of the gene in cells and tissues . However , a problem with this approach is that when either the protein tag or the T2A-GAL4 system is added , half of the time they point into the wrong direction . This is because each DNA strand is read in one direction only . Now , Li-Kroeger et al . created a so-called ‘Double Header’ system , which includes T2A-GAL4 coding in one direction and the protein tag in the other . Therefore , when the system integrates , there will always be one tag pointing in the correct direction . This makes the system twice as efficient . Not all genes have introns though . To access genes that do not contain introns , Li-Kroeger et al . developed another system , which uses the genome editing tool CRISPR-Cas9 to introduce a different kind of visible marker . Here , the whole gene is typically removed and replaced by a visible marker , which can then be replaced by any DNA , including protein tags and the T2A-GAL4 system . With these approaches , all genes in the fruit fly can now be targeted . The systems perform several tasks , including detecting gene activity and the location of proteins in the cell , and analyzing the role of the protein . The findings will be relevant to researchers interested in fruit fly genetics and cell function . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"tools",
"and",
"resources",
"genetics",
"and",
"genomics"
] | 2018 | An expanded toolkit for gene tagging based on MiMIC and scarless CRISPR tagging in Drosophila |
Cerebellar granule cells ( GCs ) make up the majority of all neurons in the vertebrate brain , but heterogeneities among GCs and potential functional consequences are poorly understood . Here , we identified unexpected gradients in the biophysical properties of GCs in mice . GCs closer to the white matter ( inner-zone GCs ) had higher firing thresholds and could sustain firing with larger current inputs than GCs closer to the Purkinje cell layer ( outer-zone GCs ) . Dynamic Clamp experiments showed that inner- and outer-zone GCs preferentially respond to high- and low-frequency mossy fiber inputs , respectively , enabling dispersion of the mossy fiber input into its frequency components as performed by a Fourier transformation . Furthermore , inner-zone GCs have faster axonal conduction velocity and elicit faster synaptic potentials in Purkinje cells . Neuronal network modeling revealed that these gradients improve spike-timing precision of Purkinje cells and decrease the number of GCs required to learn spike-sequences . Thus , our study uncovers biophysical gradients in the cerebellar cortex enabling a Fourier-like transformation of mossy fiber inputs .
Digital audio compression ( e . g . ‘MP3’; Jayant et al . , 1993 ) and image compression ( e . g . ‘JPEG’; Wallace , 1992 ) rely on Fourier transformations , which decompose a signal ( e . g . sound amplitude as a function of time or image intensity as a function of space ) into its frequency components ( power as a function of frequency ) . By storing these frequency components with different precision depending on psychophysical demands of hearing and seeing , the overall storage capacity can be increased dramatically . In principle , neuronal networks consisting of neurons with varied electrophysiological properties could be suitable for Fourier-like transformations of information . This could benefit processing in neuronal circuits by increasing the signal-to-noise ratio of input signals or by selecting only relevant spectral components of a signal . Interestingly , there are indications that for example pyramidal neurons in the visual cortex and in the hippocampus are tuned to different inputs or different input strengths ( Cembrowski and Spruston , 2019; Fletcher and Williams , 2019; Soltesz and Losonczy , 2018 ) . However , whether these neuronal networks perform a Fourier-like transform of their inputs remains unknown . Controlling the timing and precision of movements is considered to be one of the main functions of the cerebellum . In the cerebellum , the firing frequency of Purkinje cells ( PCs ) ( Heiney et al . , 2014; Herzfeld et al . , 2015; Hewitt et al . , 2011; Medina and Lisberger , 2007; Payne et al . , 2019; Sarnaik and Raman , 2018; Witter et al . , 2013 ) or the timing of spikes ( Brown and Raman , 2018; Sarnaik and Raman , 2018 ) have been shown to be closely related to movement . Indeed , cerebellar pathology impairs precision in motor learning tasks ( Gibo et al . , 2013; Martin et al . , 1996 ) and timing of rhythmic learning tasks ( Keele and Ivry , 1990 ) . These functions are executed by a remarkably simple neuronal network architecture . Inputs from mossy fibers ( MFs ) are processed by GCs and transmitted via their parallel fiber ( PF ) axons to PCs , which provide the sole output from the cerebellar cortex . GCs represent the first stage in cerebellar processing and have been proposed to provide pattern separation and conversion of the MF input into a sparser representation ( recently reviewed by Cayco-Gajic and Silver , 2019 ) . These MF inputs show a wide variety of signaling frequencies , ranging from slow modulating activity to kilohertz bursts of activity ( Arenz et al . , 2008; Rancz et al . , 2007; Ritzau-Jost et al . , 2014; van Kan et al . , 1993 ) . Interestingly , in most cellular models of the cerebellum , each MF is considered to be either active or inactive with little consideration for this wide range of frequencies ( Albus , 1971; Marr , 1969 ) . Furthermore , in these models , GCs are generally considered as a uniform population of neurons . Here , we show that the biophysical properties of GCs differ according to their vertical position in the GC layer . GCs located close to the white matter ( inner-zone ) preferentially transmit high-frequency MF inputs , have shorter action potentials , and a higher voltage threshold to fire an action potential compared with GCs close to the PC layer ( outer-zone ) . These gradients in properties of GCs enable a Fourier-like transformation of the MF input , where inner-zone GCs convey the high-frequency , and outer-zone GCs the low-frequency components of the MF input . The different Fourier-like components are sent to PCs by specialized downstream signaling pathways , which differ in PF axon diameters , action potential conduction velocity , and PC excitatory postsynaptic potential ( EPSP ) kinetics . Computational simulations show that the biophysical gradients in the GC and molecular layer significantly reduce the number of GCs required to learn a sequence of firing frequencies and reduce the time needed to switch between firing frequencies .
To investigate whether GCs are tuned for different frequencies , we first investigated the intrinsic membrane properties of GCs from different depths within the GC layer in lobule V of the cerebellum of P21-30 mice . We divided the GC layer into three zones and performed whole-cell current-clamp recordings from inner- ( closest to the white matter ) , middle- and outer-zone GCs ( closest to PCs ) ( Figure 1A , B ) . Upon current injection , inner-zone GCs were less excitable compared with outer-zone GCs ( Figure 1C ) . On average , the relationship between the frequency of action potentials and the injected current was surprisingly different for inner- and outer-zone GCs ( Figure 1D ) : inner-zone GCs needed higher current injections to fire an action potential ( inner: 56 . 8 ± 2 . 6 pA vs . middle: 51 . 2 ± 2 . 0 pA vs . outer: 39 . 4 ± 2 . 0 pA; n = 38 , 25 , and 37 , respectively; PKruskal-Wallis <0 . 0001; Figure 1E ) and to achieve the maximum firing rate compared with middle- and outer-zone GCs ( inner: 224 . 6 ± 9 . 8 pA vs . middle: 190 . 8 ± 9 . 6 pA vs . outer: 174 . 3 ± 9 . 0 pA , respectively; PKruskal-Wallis = 0 . 002; Figure 1F ) . Consistently , inner-zone GCs had a more depolarized threshold for action potential generation compared with middle- and outer-zone GCs ( −38 . 0 ± 0 . 7 mV vs . −38 . 2 ± 0 . 8 mV vs . −41 . 4 ± 0 . 6 mV; PKruskal-Wallis = 0 . 003; Figure 1G ) and a lower input resistance ( 486 ± 27 MΩ vs . 494 ± 27 MΩ vs . 791 ± 63 MΩ; PKruskal-Wallis = <0 . 0001; Figure 1H ) . Furthermore , the capacitance of inner-zone GCs was significantly larger compared with outer-zone GCs ( inner: 5 . 8 ± 0 . 2 pF vs . middle: 5 . 8 ± 0 . 2 pF vs . outer: 4 . 6 ± 0 . 1 pF; PKruskal-Wallis = <0 . 0001 Figure 1I ) . In agreement with these findings , we observed depolarization block in inner-zone GCs at higher current inputs than for outer-zone GCs ( Figure 1C , D ) . Furthermore , a larger delay of the first spike was observed in inner- compared with outer-zone GCs ( Figure 1J; PKruskal-Wallis = 0 . 0001; Figure 1K ) . The delay with 60 pA current injection was 48 ± 6 ms for inner- , 38 ± 4 ms for middle- , and 23 ± 2 ms for outer-zone GCs ( n = 32 , 25 , and 37 , respectively; note that 6 out of 38 inner-zone GCs did not fire an action potential at 60 pA ) . Finally , the action potential half-width of GCs differed significantly between the three zones ( inner: 122 ± 2 µs vs . middle: 137 ± 4 µs vs . outer: 143 ± 4 µs; PKruskal-Wallis = 0 . 0001; Figure 1L ) . The distribution of the raw data ( Figure 1—figure supplement 1 ) suggests a gradual change in the average cell parameters along the depth axis of the GC layer , but two populations of neurons ( salt and pepper distribution ) , or three populations of neurons ( inner- , middle- , and outer-zone ) cannot fully be ruled out . To test whether these gradients are specific to lobule V , we investigated GCs in lobule IX . Here , we observed very similar gradients to lobule V ( Figure 1—figure supplement 2 ) . In short , outer-zone GCs were more excitable and had broader spikes compared with inner-zone GCs . Interestingly , the absolute values between lobule V and IX differed ( Figure 1—figure supplement 2 ) , consistent with previously described differences in , for example the firing frequency in vivo between these two lobules ( Witter and De Zeeuw , 2015a; Zhou et al . , 2014 ) and in the differential density of Kv4 and Cav3 channel expression in GCs across different lobules ( Heath et al . , 2014; Rizwan et al . , 2016; Serôdio and Rudy , 1998 ) . Taking the large functional difference between spino- and vestibulo-cerebellum into account ( Witter and De Zeeuw , 2015b ) , these data suggest that different biophysical properties of GCs are likely a conserved mechanism throughout the entire cerebellar cortex , potentially tuning GCs to different frequencies . Development can have large effects on the physiology of neurons , and GCs in particular undergo profound changes during development ( Dhar et al . , 2018; Lackey et al . , 2018 ) . To exclude confounding effects of the developmental stage , we tested whether these gradients were also present in more adult mice . Recordings obtained from GCs in lobule V in animals between 80 and 100 days of age revealed very similar gradients to those observed in young animals ( Figure 1—figure supplement 3 ) . Together , these data show prominent gradients in the electrophysiological properties of GCs over the depth of the GC layer , and that these gradients can consistently be found across different lobules and ages . To investigate possible biophysical causes for the gradients in the biophysical properties , we investigated voltage-gated potassium ( Kv ) currents by performing voltage-clamp recordings in outside-out patches from somata of inner- and outer-zone GCs in lobule V ( Figure 2A ) . The maximum Kv current at -60 mV was significantly higher in inner-zone GCs ( 282 ± 29 pA , n = 48 ) compared with outer-zone GCs ( 221 ± 28 pA , n = 54 , PMann-Whitney = 0 . 04; Figure 2B; Figure 2—figure supplement 1 ) . Neither the steady-state activation curve ( Figure 2C ) nor the degree of inactivation ( Figure 2D ) was different between the two GC populations . Furthermore , steady-state inactivation , which was investigated with different holding potentials , was similar between inner- and outer-zone GCs ( Figure 2—figure supplement 2 ) . These data suggest that inner- and outer-zone GCs have a similar composition of Kv channels , but inner-zone GCs have a higher Kv channel density . The larger Kv currents in inner-zone GCs are consistent with the short action potential duration of inner-zone GCs ( cf . Figure 1 ) . Thus , our data provide a biophysical explanation for the observed gradients in GC properties . The gradients within the GC layer create an optimal range of input strengths for each GC . To test how these gradients impact the processing of synaptic MF inputs , we performed Dynamic Clamp experiments ( Desai et al . , 2017 ) and investigated whether different MF input frequencies differentially affect spiking in inner- and outer-zone GCs ( Figure 3A and B ) . We first recorded excitatory postsynaptic currents ( EPSC ) from GCs in inner- and outer-zones of lobule V after single MF stimulation . We found no significant differences in the amplitude or kinetics of EPSCs in inner- and outer-zone GCs ( Figure 3—figure supplement 1 ) . Individual MFs span the entire depth of the GC layer , contacting both inner- and outer-zone GCs ( Krieger et al . , 1985; Palay and Chan-Palay , 1974 ) . Furthermore , GCs are electronically extremely compact neurons and can be considered as a single compartment ( D'Angelo et al . , 1993; Delvendahl et al . , 2015; Silver et al . , 1992 ) . Therefore , we could use the Dynamic Clamp technique to implement the conductance of identical MF signals in inner- and outer-zone GCs based on the measured EPSC kinetics . We first applied input of a single MF with Poisson-distributed firing-frequencies ranging between 30 and 500 Hz for 300 ms duration while changing the resting membrane potential to simulate the large variability of membrane potential observed in GCs in vivo ( Chadderton et al . , 2004 ) . In line with the gradients in the electrophysiological properties of GCs , inner-zone GCs fired fewer action potentials compared with outer-zone GCs in response to low-frequency MF inputs at a membrane potential of approximately –90 mV ( Figure 3C and D; Figure 3—figure supplement 2 ) . In contrast , inner-zone GCs fired more action potentials compared with outer-zone GCs in response to high-frequency MF inputs at a membrane potential of approximately –70 mV . In vivo , such a depolarization would be caused by reduced inhibition and/or simultaneous activation of multiple MF inputs . These data suggest that outer- and inner-zone GCs are specialized to process low- and high-frequency MF inputs , respectively . To further test whether inner- and outer-zone GCs can extract different frequency components from a MF input signal , which would resemble a Fourier-transformation , we varied the MF input frequency sinusoidally between 30 and 300 Hz , representing a range of in-vivo-like tonic firing behaviour ( Figure 4A; Arenz et al . , 2008; van Kan et al . , 1993 ) . At a holding potential of −70 mV , commonly occurring in vivo ( Chadderton et al . , 2004 ) , inner-zone GCs responded preferentially to high-frequency MF inputs up to 300 Hz , while outer-zone GCs responded preferentially to low-frequency inputs up to 100 Hz ( Figure 4B; Figure 4—figure supplement 1 ) . To estimate the optimal frequency at which inner- and outer-zone GCs preferentially fire action potentials , we calculated the phase angle ( see Materials and methods , Equation 3 ) . The mean phase angle , at which GC preferentially fired , was 162 ± 8° for inner-zone ( n = 10 ) and 100 ± 20° for outer-zone GCs ( n = 7; PMann-Whitney = 0 . 02; Figure 4C ) , representing an average firing frequency of 284 and 116 Hz for inner- and outer-zone GCs , respectively ( cf . Equation 2 ) . Thus , the gradients in the biophysical properties enable the cerebellar GC layer to split incoming MF signals into different frequency bands and thereby to perform a Fourier-like transformation of the compound MF input signal . A Fourier-like transformation in the GC layer ( i . e . a separation of the spectral components of MF signals ) could be particularly relevant if downstream pathways are specialized for these spectral components . Early silver-stainings and drawings from Ramón y Cajal indicate that inner-zone GCs give rise to PFs close to the PC layer and outer-zone GCs give rise to PFs close to the pia ( Eccles et al . , 1967; Cajal , 1911 but see Espinosa and Luo , 2008; Wilms and Häusser , 2015 ) . To test this possibility , we examined the axons of GCs . First , we investigated whether there is a correlation between the relative positions of the PF in the molecular layer and the GC somata in the GC layer . DiI was injected in vivo into the GC layer to label GCs and their axons . Several GCs were clearly stained 24 hr after DiI injection ( Figure 5A ) , and the position of their soma and PF in the cerebellar cortex could be measured ( Figure 5B–D ) . Even though the length of the ascending GC axon showed considerable variation ( 196 ± 5 . 5 μm , range: 144 to 291 μm , n = 39 axons in n = 6 mice ) , after normalization for the thickness of the molecular and GC layers , the GC soma position was significantly correlated with the position of the bifurcation in the GC axon ( Figure 5C , D; R = –0 . 86 , p<0 . 001 ) . These data show that inner-zone GCs preferentially give rise to PF located near the PC layer ( inner-zone PFs ) and outer-zone GC give rise to PF near the surface of the cerebellar cortex ( outer-zone PFs ) . Next , we tested whether PFs , like GCs , have different properties depending on their position within the molecular layer . First , we compared the PF diameters in electron microscopic images of parasagittal sections of lobule V of mouse cerebellum and found significantly larger diameters for inner-zone PFs compared with middle- and outer-zone PFs ( 182 ± 2 . 6 nm , n = 703 vs . 159 ± 2 . 0 nm , n = 819 vs . 145 ± 1 . 7 nm , n = 1024 Figure 6A–C; PKruskal-Wallis <0 . 0001 ) , which is in agreement with previous investigations reported in cat ( Eccles et al . , 1967 ) , monkey ( Fox and Barnard , 1957 ) , rat ( Pichitpornchai et al . , 1994 ) , and mouse ( Wyatt et al . , 2005 ) . The axonal diameter is usually correlated with conduction velocity ( Jack et al . , 1983 ) . We therefore recorded compound action potentials of PFs in lobule V and compared their conduction velocity in the inner- , middle- , and outer-zone of the molecular layer ( Figure 6D–F ) . We detected a significantly higher velocity in inner-zone PFs compared with middle- or outer-zone PFs ( 0 . 334 ± 0 . 003 m*s−1 , n = 8 vs . 0 . 303 ± 0 . 004 m*s−1 , n = 6 vs . 0 . 287 ± 0 . 007 m*s−1 , n = 8; Figure 6F; PKruskal-Wallis <0 . 0001 ) . The absolute velocity and the gradient in the velocity from inner- to outer-zone PFs agree well with previous studies ( Baginskas et al . , 2009; Vranesic et al . , 1994 ) . These results suggest that inner-zone PFs are specialized for fast signaling , which is consistent with the concept that inner-zone GCs are tuned for high-frequency inputs ( cf . Figures 1 and 2 ) . In addition to the above results obtained from lobule V , similar gradients in both axon diameter and axon conduction velocity were found in lobule IX ( Figure 6—figure supplement 1 ) . This suggests that gradients in axon diameter and axon conduction speed are general features of the cerebellar cortex . A possible confounder of our results could be an over-representation of large-diameter Lugaro cell axons within inner-zone PFs ( Dieudonné and Dumoulin , 2000 ) . However , this would predict that the histograms of the axon diameters show two peaks with varying amplitude . Instead , we observed a single bell-shaped distribution in each zone ( Figure 6—figure supplement 2 ) , arguing that the measured differences between axon diameters were not due to varying contributions from Lugaro cell axons , but reflect the differences between inner- , middle- , and outer-zone PFs . Our data thus far indicate that GCs and PFs are adapted to different MF input frequencies and conduction velocities , respectively . This arrangement could in principle provide PFs with functionally segregated information streams that are differentially processed in PCs . To investigate this possibility , we made whole-cell current-clamp recordings from PCs in sagittal slices of the cerebellar vermis . PCs were held at a hyperpolarized voltage to prevent spiking and to isolate excitatory inputs . Electrical stimulation of PFs was performed at inner- , middle- , and outer-zones of the molecular layer and the stimulation intensity was adjusted to obtain similar EPSP amplitudes in all zones ( Figure 7A , B ) . Stimulation of inner-zone PFs resulted in EPSPs ( Barbour , 1993; Roth and Häusser , 2001 ) with shorter rise and decay times compared with EPSPs obtained from stimulating outer-zone PFs ( rise20-80: inner: 0 . 57 ± 0 . 04 ms , n = 12; middle: 0 . 93 ± 0 . 17 ms , n = 4; outer: 1 . 83 ± 0 . 33 ms , n = 12; PKruskal-Wallis = 0 . 0001; decay: inner: 21 . 9 ± 1 . 5 ms , middle: 39 . 7 ± 1 . 1 ms outer: 40 . 8 ± 4 . 1 ms; PKruskal-Wallis = 0 . 0004 , Figure 7C; Figure 7—figure supplement 1 ) . These results suggest that inner-zone PF inputs undergo less dendritic filtering in PCs compared with outer-zone PF inputs ( De Schutter and Bower , 1994a; Roth and Häusser , 2001 but see De Schutter and Bower , 1994b ) . To investigate high-frequency inputs to PCs , we elicited five EPSPs at 100 Hz and 500 Hz ( Figure 7D , E ) . Individual EPSPs evoked from inner-zone PFs showed clear individual rising phases and peaks between each stimulus and less summation compared with outer-zone PFs ( Figure 7D–F; Figure 7—figure supplement 1 ) . These results suggest that inner-zone PFs can transmit timing information more faithfully compared with outer-zone PFs and thus control spike timing of PCs more precisely . Thus far we have described prominent gradients in the electrophysiological properties of GCs over the depth of the GC layer that enable inner- and outer-zone GCs to preferentially respond to high- and low-frequency inputs , respectively . The different frequency components are transferred via specialized PFs , which enable PCs to interpret high-frequency signals rapidly at the base of their dendritic trees and low-frequency signals slowly at more distal parts of their dendritic trees ( Figure 8A ) . To address the functional implications of these gradients in the GC and molecular layer , we performed computational modeling of a neuronal network of the cerebellar cortex with integrate-and-fire neurons . The model consisted of one PC and a varying number of GCs and MFs ( Figure 8A ) . GCs received randomly determined MF inputs with either tonic ( Arenz et al . , 2008; van Kan et al . , 1993 ) or bursting ( Rancz et al . , 2007 ) in-vivo-like spiking sequences . MF inputs were randomly distributed across layers , consistent with MFs having rosettes throughout the depth of the granule cell layer ( Krieger et al . , 1985; Palay and Chan-Palay , 1974 ) . By changing the synaptic weights of the GC to PC synapses , the PC had to acquire a target spiking sequence with regular 80- , 40- and 120 Hz firing ( Figure 8B ) . The algorithm for changing the synaptic weights was a combination of a learning algorithm based on climbing-fiber-like punishments and an unbiased minimization algorithm ( see Materials and methods ) . We first compared a model without gradients , where the parameters were set at the average of the experimentally determined values , with a model including all experimentally determined gradients ( black and red , respectively , throughout Figure 8 ) . To measure the difference between the final PC spiking and the target sequence , we calculated van Rossum errors using a time constant of 30 ms ( van Rossum , 2001; Figure 8C–E ) . With an increasing number of GCs , the final PC spiking sequence resembled the target sequences increasingly better , as illustrated by an average spiking histogram from many repetitions with different random sets of MF inputs for models consisting of 100 and 1000 GCs ( Figure 8B ) . As expected , the average minimal van Rossum error ( for many repetitions with different random sets of MF inputs ) decreased with increasing number of GCs ( Figure 8C ) . For all sizes of the GC population , the average minimal van Rossum error was significantly smaller in the model containing all the experimentally determined gradients compared with the model without any gradients . For example , to obtain the spiking precision of the model containing 400 GCs with all gradients , the model without gradients required 800 GCs ( cf . red arrows in Figure 8C ) . This indicates that for a cerebellum exploiting gradients , the number of GCs can at least be halved while obtaining a certain temporal precision compared with a cerebellum containing no gradients . To investigate the relative contribution of each of the gradients , we tested models containing single gradients in isolation , resulting in intermediate van Rossum errors ( blue , yellow , and green in Figure 8C , D ) . The average relative differences between the models across all sizes of the GC populations suggest an almost additive behavior of the individual gradients to the overall performance ( Figure 8E ) . To further investigate the interplay of the different gradients , we investigated a model containing all gradients , but the connectivity between GCs , PF action potential conduction velocity , and PC EPSP kinetics were randomly intermixed ( red dashed lines in Figure 8C–E ) . The network benefits from these intermixed gradients , but maximum optimization can only be obtained with correct connectivity ( Figure 8E ) . The time constant to calculate the van Rossum error can be decreased or increased to investigate spike timing or slower changes in firing rate , respectively . The impact of the gradients increased with increasing time constant ( Figure 8—figure supplement 1A , B ) , indicating that rate coded signaling especially benefits from the here described gradients . To specifically test the effect of gradients on the cerebellum’s ability to switch between firing frequencies , we made sigmoid fits around the times of firing rate changes . The transition time ( tT; see Materials and methods ) from these fits showed that models with all gradients showed on average about 30% faster ‘frequency-switching’ than models without any gradients ( Figure 8—figure supplement 1C–F ) . Finally , we repeated the modeling experiments but used a target sequence with a firing pause ( i . e . 80 , 0 , and 120 Hz instead of 80 , 40 , and 120 Hz ) resulting in similar conclusions regarding the van Rossum errors and transition times ( Figure 8—figure supplement 1G–M ) . A pause in firing enabled us to quantify the temporal error at the beginning and the end of the pause ( Figure 8—figure supplement 1N–Q ) . These spike times have been proposed to be of particular relevance for behavior ( Hong et al . , 2016 ) . Analysis of the temporal error in the beginning and the end of the pause revealed similar results compared with the van Rossum error and the transition time . Thus , our modeling results show that the experimentally determined gradients improve the spiking precision , accelerate ‘frequency-switching’ , and increase the storing capacity of the cerebellar cortex .
Our data demonstrate that outer-zone GCs preferentially fire during MF input with low frequency ( ‘low-frequency’ GCs , magenta in Figure 9A ) , whereas inner-zone GCs preferentially fire during MF input with high frequency ( ‘high-frequency’ GCs , green in Figure 9A ) . The separation of a signal into its frequency components resembles a Fourier transformation ( Figure 9B ) . The analogy with a Fourier transformation has the limitations that ( 1 ) the separation is only partial with overlapping ranges of preferred frequency , ( 2 ) a single MF cannot transmit two frequencies simultaneously but only separated in time ( as illustrated in Figure 9A ) and ( 3 ) concurrent inputs from two MFs with different frequencies synapsing onto a single GC cannot be separated . Yet , our data indicate that the entire GC layer with several MFs sending various frequencies to numerous GCs can execute a Fourier-like transformation . In analogy to the dispersion of white light into its spectral components by an optical prism , the broadband MF signal is separated into its spectral components with inner- to outer-zone GCs preferentially transmitting the high- to low-frequency components , respectively . Such a partial separation offers the chance to differentially process high- and low-frequency components . Indeed , in the molecular layer , the high-frequency components of the MF signal are sent via rapidly conducting axons to proximal parts of the PC dendritic tree . This allows fast ( phasic ) signals to have a strong and rapid impact on PC firing . On the other hand , low-frequency components of the MF signal are conducted more slowly and elicit slower EPSPs , allowing slow ( tonic ) signals to have a modulatory impact on PC firing . Our data indicate that , in analogy to the increased storing capacity of digital audio and image compression ( Jayant et al . , 1993; Wallace , 1992 ) , the combination of a Fourier-like transformation in the GC layer and specialized downstream signaling pathways in the molecular layer dramatically reduce the number of required GCs for precise PC spiking ( Figure 8 ) . Furthermore , our data support the ‘adaptive filter’ theory of the cerebellum , where broadband MF input is differentially filtered by GCs ( Dean et al . , 2010; Fujita , 1982; Singla et al . , 2017 ) . Within this framework , our data indicate gradients in the band-pass filtering properties of GCs . Furthermore , our data could provide an additional explanation for the improvement in motor learning when elevating background activity of MFs ( Albergaria et al . , 2018 ) : the elevated MF activity will help to overcome the high threshold of inner-zone GCs , which rapidly and effectively impact PCs via fast conducting PFs at the proximal dendrite . There are at least two axes of heterogeneity in the cerebellar cortex . First , Zebrin stripes can be observed as parasagittal zones ( ‘medio-lateral’ axis ) in cerebellar cortex ( Apps et al . , 2018 ) . Firing rate , firing regularity , synaptic connectivity and synaptic plasticity seems to differ between PCs in zebrin positive and negative zones ( Valera et al . , 2016; Wadiche and Jahr , 2005; Xiao et al . , 2014; Zhou et al . , 2014 ) . Second , there is a lobular organization ( ‘rostro-caudal’ axis ) as shown here by the functional differences between lobules V and IX ( Figure 1—figure supplement 1 ) . GCs in lobule IX are tuned to lower frequencies than GCs in lobule V . These findings are largely in line with previous investigations ( Heath et al . , 2014; Witter and De Zeeuw , 2015a; Zhou et al . , 2014 ) , where the anterior cerebellum was identified to process high-frequency or bursting signals , while the vestibulo-cerebellum mainly processed lower frequency or slowly-modulating inputs . Furthermore , the optimal time intervals for introduction of spike timing dependent plasticity differ between the vermis and the flocculus ( Suvrathan et al . , 2016 ) . In addition to these two known axes of heterogeneity , we described an axis that is orthogonal to the surface of the cerebellar cortex . This ‘depth’ axis causes inner-zone GCs to be tuned to higher frequencies than outer-zone GCs . The frequency gradients along the ‘depth’-axes are in line with recently described connections of nucleo-cortical MFs and PC , which specifically target GCs close to the PC layer ( Gao et al . , 2016; Guo et al . , 2016 ) . These connections send slow feedback signals to the outer-zone GCs , which –– according to our framework –– are ideally suited to process such slow modulatory signals . Independent of these specialized feedback pathways , MFs exhibit heterogeneity ( Chabrol et al . , 2015; Bengtsson and Jörntell , 2009 ) . Consistent with MFs having rosettes throughout the depth of the granule cell layer ( Krieger et al . , 1985; Palay and Chan-Palay , 1974 ) , our data indicate that each type of the heterogeneous MF inputs is split into its frequency components along the depth axis . A preference of some MFs to specific zones could furthermore contributes to the frequency separation ( Quy et al . , 2011; Jörntell and Ekerot , 2006 ) . Our results predict that superficial GCs , such as the ones imaged recently in the investigation of eye-blink conditioning and reward representation in the cerebellar cortex ( Giovannucci et al . , 2017; Wagner et al . , 2017 ) , would preferentially convey low-frequency signals to PCs and might not be representative for the full range of frequencies present over the depth of the GC layer . Recently , diverse adaptation of GCs to 2-s-lasting current injections has been described ( Masoli et al . , 2019 ) , but it remains unknown to which extent this form of adaptation exhibits a gradient along the depth axis . The genetic reasons for the here-observed gradients in cerebellar cortex are currently not known . Due to a large variability within each zone , our data cannot rule out a salt and pepper distribution of two populations of neurons ( Espinosa and Luo , 2008 ) . However , neurons in the medial vestibular nucleus exhibit a graded tuning of the capacity for fast-spiking by expression levels of specific ion channels ( Kodama et al . , 2020 ) . Thus , including this new ‘depth’ axis , there are three orthogonal axes along which the cerebellar cortex is tuned for preferred frequency , indicating the importance of proper frequency tuning of the circuitry . In the current study we did not investigate molecular layer interneurons , which can have a large impact on PC spiking ( Blot et al . , 2016; Dizon and Khodakhah , 2011; Gaffield and Christie , 2017; Mittmann et al . , 2005; Sudhakar et al . , 2017 ) . However , the spatial arrangement of stellate and basket cell interneurons is consistent with our framework . Although the dendrites of molecular layer interneurons can span the entire molecular layer , the dendrites of basket cells seem to be preferentially located at the inner-zone of the molecular layer ( Palkovits et al . , 1971; Rakic , 1972 ) , which positions them ideally to receive rapid high-frequency signals of inner-zone PFs . Consistently , they impact PC firing rapidly and efficiently via their pinceaus ( Blot and Barbour , 2014 ) . Furthermore , the dendrites of a subset of stellate cells ( with their somata located in the outer-zone molecular layer ) are preferentially located at the outer-zone molecular layer ( Palkovits et al . , 1971; Rakic , 1972 ) , which positions them ideally to receive modulatory low-frequency signals and elicit slow IPSPs in PCs . Furthermore , molecular layer interneurons seem to represent a continuum along the vertical axis , with a correlation between the vertical location of the soma , axonal boutons , and dendrite location ( Sultan and Bower , 1998 ) , which is consistent with the here-described continuum of biophysical properties along the vertical axis of the cortex . Incorporating molecular layer interneurons , their synaptic plasticity and their potential gradients into the frequency-dispersion framework may show a further increase in the dynamic range of frequency separation within the cerebellar cortex c what we have described here ( Gao et al . , 2012 ) . MF firing frequencies range from <1 to ~1000 Hz ( Arenz et al . , 2008; Chadderton et al . , 2004; Jörntell and Ekerot , 2006; Rancz et al . , 2007; van Kan et al . , 1993 ) . Many previous modeling studies investigating cerebellar function considered the activity of each MF as a constant digital value ( Albus , 1971; Babadi and Sompolinsky , 2014; Brunel et al . , 2004; Clopath et al . , 2012; Marr , 1969 ) , a constant analog value ( Chabrol et al . , 2015; Clopath and Brunel , 2013 ) , or spike sequences with constant frequency ( Billings et al . , 2014; Cayco-Gajic et al . , 2017; Steuber et al . , 2007 ) . We focused on the time-varying aspects of MF integration in GCs , and therefore implemented a model with a corresponding large range of MF input frequencies that could change over time . It would be interesting to elucidate whether models with more uniform MF inputs , such as those found in many previous models , would benefit from the here-observed biophysical gradients . To implement these gradients in a model , we used a simplified cerebellar circuitry that does not consider active dendrites ( Llinás and Sugimori , 1980 ) or the tonic activity of PCs ( Raman and Bean , 1997 ) . It will therefore be interesting to investigate if the here-observed gradients in the GC and molecular layer improve the performance of more complex models of the cerebellar cortex ( De Schutter and Bower , 1994a; Garrido et al . , 2013; Masoli et al . , 2015; Medina et al . , 2000; Rössert et al . , 2015; Spanne and Jörntell , 2013; Steuber et al . , 2007; Sudhakar et al . , 2017; Walter and Khodakhah , 2009; Yamazaki and Tanaka , 2007 ) . Furthermore , it remains to be investigated whether gradients in the GC layer also improve models that aim to explain tasks such as eye-blink conditioning ( Mauk and Buonomano , 2004 ) and vestibulo-ocular reflexes ( Lac et al . , 1995 ) . Our model simulated the learning that PCs undergo to acquire specific firing frequencies in response to GC input . PC firing rate and spiking precision have been shown to be closely related to movement ( Brown and Raman , 2018; Sarnaik and Raman , 2018 ) . Our results show that the same temporal spiking precision or the same frequency switching speed can be obtained with approximately half the number of GCs when GC gradients are implemented ( Figure 8 ) . Taking into account the large number of cerebellar GCs in the brain ( Herculano-Houzel , 2009; Williams and Herrup , 1988 ) , a significant reduction in the number of GCs could represent an evolutionary advantage to minimize neuronal maintenance energy ( Howarth et al . , 2012; Isler and van Schaik , 2006 ) . Therefore , the dramatic increase in storing capacity for precise PC spiking provides an evolutionary explanation for the emergence of gradients in the neuronal properties . Based on the described advantages of the Fourier transformation for rapid and storing-efficient information processing , we hypothesize that other neural networks also perform Fourier-like transformations and use segregated frequency-specific signaling pathways . To our knowledge , this has rarely been shown explicitly , but similar mechanisms might operate , for example , in the spinal cord network: descending motor commands from the pyramidal tract send broadband signals to motoneurons with different input resistances resulting from differences in size . This enables small motoneurons to fire during low-frequency inputs and large motoneurons only during high-frequency inputs ( Henneman et al . , 1965 ) . Furthermore , specialized efferent down-stream signaling pathways innervate specific types of muscles with specialized short-term plasticity of the corresponding neuromuscular junctions ( Wang and Brehm , 2017 ) . In the hippocampus , frequency preferences of hippocampal neurons are well established in enabling segregation of compound oscillatory input into distinct frequency components ( Pike et al . , 2000 ) . Furthermore , there is increasing evidence that what has been considered a homogeneous population of neurons exhibit gradients in the neuronal properties ( Cembrowski and Spruston , 2019 ) , such as the intrinsic electrical properties and synaptic connectivity in CA3 pyramidal neurons ( Galliano et al . , 2013 ) . The here reported heterogeneity furthermore enables functional segregation of information streams for example in CA1 pyramidal neurons ( Soltesz and Losonczy , 2018 ) . Additionally , gradients in biophysical properties of neurons in the entorhinal cortex might serve to generate functional outcomes relevant for the generation of grid cell sizes ( Giocomo et al . , 2007; Schmidt-Hieber and Nolan , 2017; Orchard et al . , 2013 ) . Finally , in the neocortex , gradients in anatomical and biophysical properties were recently uncovered ( Fletcher and Williams , 2019 ) . In summary , our findings contribute to the growing body of evidence that the neurons of a cell layer can exhibit systematic functional heterogeneities with differential tuning of neurons along gradients . Our data furthermore suggest that such gradients facilitate complex transformation of information , such as Fourier-like transformations , to cope with a broad temporal diversity of signals in the central nervous system .
Parasagittal 300-µm-thick cerebellar slices were prepared from P21–P30 ( young animals ) or from P80–P100 ( old animals ) C57BL/6 mice of either sex as described previously ( Ritzau-Jost et al . , 2014; Delvendahl et al . , 2015 ) . Animals were treated in accordance with the German and French Protection of Animals Act and with the guidelines for the welfare of experimental animals issued by the European Communities Council Directive . The extracellular solution for the whole-cell measurements contained ( in mM ) : NaCl 125 , NaHCO3 25 , glucose 20 , KCl 2 . 5 , CaCl2 2 , NaH2PO4 1 . 25 , MgCl21 ( 310 mOsm , pH 7 . 3 when bubbled with Carbogen ( 95%O2/5%CO2 ) ) . For outside-out measurements of potassium currents ( Figure 2 ) , 150 µM CdCl2 and 1 µM TTX were added to the external solution to block voltage-gated calcium channels and sodium channels , respectively . The intracellular solution contained in mM: K-Gluconate 150 , NaCl 10 , K-Hepes 10 , Mg-ATP 3 , Na-GTP 0 . 3 , EGTA 0 . 05 ( 305 mOsm , pH 7 . 3 ) . A liquid junction potential of +13 mV was corrected for . All electrophysiological measurements were performed with a HEKA EPC10 amplifier ( HEKA Elektronik , Lambrecht/Pfalz , Germany ) under control of the Patchmaster software . All measurements were performed at 34–37°C . In order to analyze the response of GCs on in vivo-like MF inputs , we used a Dynamic Clamp implemented with the microcontroller Teensy 3 . 6 ( https://www . pjrc . com ) as described by Desai et al . ( 2017 ) . The Teensy was programmed using the Arduino integrated development environment with the code provided by Desai et al . ( 2017 ) and modified for our need as described in the following . The time course of MF conductance was ( 1 ) GEPSCt=Gmax Anorm-e-tτr+∑i=13aie-tτiwhere the exponential rise time ( τr ) was 0 . 1 ms , the decay time constants ( τ1 , τ2 , and τ3 ) were 0 . 3 , 8 , and 40 ms , respectively , and the relative amplitude of the decay components ( a1 , a2 , and a3 ) were 0 . 7 , 0 . 26 , and 0 . 04 , respectively . The peak conductance ( Gmax ) was 1 nS ( Hallermann et al . , 2010 ) and the normalization factor ( Anorm ) was 0 . 518 , which was numerically calculated to obtain a peak amplitude of 1 . The kinetics of the MF conductance were chosen to reproduce the measured mixed AMPA and NMDA EPSC kinetics of single EPSCs ( Figure 3—figure supplement 1 ) and trains of EPSCs ( Baade et al . , 2016 ) . The short-term plasticity during Poisson sequence of spikes was implemented by changing Gmax according to a simple phenomenological model ( Tsodyks and Markram , 1997 ) assuming a release probability pr0 of 0 . 4 ( Ritzau-Jost et al . , 2014 ) . Facilitation was implemented as an increase in the release probability according to pr = pr + 0 . 2* ( 1- pr ) and decaying back to pr0 with a time constant of 12 ms ( Saviane and Silver , 2006 ) . Depression was implemented according to a recovery process with a time constant of 25 ms , which approximates a biexponential recovery process of 12 ms and 2 s ( Hallermann et al . , 2010; Saviane and Silver , 2006 ) . The resulting short-term plasticity reproduced previously obtained data with regular spiking ranging from 20 to 1000 Hz ( Baade et al . , 2016; Hallermann et al . , 2010; Ritzau-Jost et al . , 2014 ) . The microcontroller was programmed to implement the MF conductance and its short-term plasticity with Poisson distributed spike times with a constant frequency ranging from 30 to 500 Hz for 300 ms ( Figure 3 ) . In each cell , each frequency was applied five times . To investigate the response to sinusoidally varying input frequencies ( Figure 4 ) , the target frequency of the Poisson process ( F ) was varied on a logarithmic scale according to: ( 2 ) Ft=explogFmin+logFmax-logFmin0 . 5-0 . 5 cos2πt/Twhere the minimal and maximal frequency ( Fmin and Fmax ) were 30 and 300 Hz , respectively , and the duration of the sine wave cycle ( T ) was 1 s . In each cell , 10 cycles were applied consecutively for at least four times ( interval >30 s ) . The histogram of the spike times ( Figure 4B ) was averaged across the last four cycles of all cells . The vector strength and phase angle ( van Kan et al . , 1993 ) were calculated as the absolute value and the argument of the complex number ρ ( i=-1 ) : ( 3 ) ρ=1N∑n=1Nei2πtnTwhere tn are the spike times of all N spikes per experiment and T the cycle duration ( 1 s ) . To increase statistical validity , only those cells that fired more than 100 action potentials during the analyzed cycles were included in the analysis . This criterion resulted in the exclusion of 3 out of 13 and 2 out of 9 cells for inner- and outer-zone GCs , respectively . However , inclusion of these cells in the analysis resulted in similar preference for MF firing frequency [phase angle along the cycle: 146 ± 10° for inner-zone ( n = 13 ) and 103 ± 18° for outer-zone GCs ( n = 9; PMann-Whitney= 0 . 06 ) , representing an average firing frequency of 246 and 123 Hz for inner- and outer-zone GCs , respectively] . Four C57BL/6 mice of either sex with an age between P23–P28 were sacrificed , followed by transcardial perfusion with saline and consecutively a fixative containing 4% paraformaldehyde and 2% glutaraldehyde in phosphate-buffered saline ( PBS ) . After removal of the brain , the tissue was allowed to post-fix over night at 4°C and sagittal sections of the cerebellum were prepared at a thickness of 60 µm using a Leica microtome ( Leica Microsystems , Wetzlar , Germany ) . The sections were stained in 0 . 5% osmium tetroxide in PBS for 30 min followed by dehydration in graded alcohol and another staining step with 1% uranyl acetate in 70% ethanol . After further dehydration , the tissue was embedded in durcupan ( Sigma-Aldrich ) , which was allowed to polymerize for 48 hr at 56°C between coated microscope slides and cover glasses . Regions of interest were identified by light microscopy , cut and transferred onto blocks of durcupan to obtain ultra-thin sections using an Ultramicrotome ( Leica Microsystems ) . Ultra-thin sections were transferred onto formvar-coated copper grids and stained with lead citrate . Ultrastructural analysis was performed using a Zeiss SIGMA electron microscope ( Zeiss NTS , Oberkochen , Germany ) equipped with a STEM detector and ATLAS software . Electron micrographs were manually analyzed in a blind manner ( numbered by masked randomization ) and each micrograph was divided into eight identically sized fields . The diameter of each parallel-fiber axon was measured as the longest chord in one or two of these fields . Cross sections with visible active zones or mitochondria were excluded from analysis . Six P20 CD1 mice were anesthetized with isoflurane ( 4% ) . An incision of the skin to expose the skull was made and a hole was manually drilled using a 25G needle above the desired injection site . Injections of small amounts of DiI ( 1 , 1-dioctadecyl-3 , 3 , 3 , 3 tetramethylindocarbocyanine perchlorate , ThermoFisher Scientific , 10% in N , N-dimethylformamide ) were performed using a broken glass pipette connected to a picospritzer II ( Parker Instrumentation ) . 24 hr after injection , animals were sacrificed and transcardially perfused with 4% paraformaldehyde in PBS . The cerebellum was dissected , fixed overnight , and embedded in 4% agarose in PBS . 150 µm thick sections were then cut in the transverse or sagittal plane using a vibratome ( VT1000 , Leica microsystems ) . Z-Stacks ( 1 µm steps ) were acquired using a confocal microscope ( Leica SP5 II , 63x objective ) . GCs were traced from their soma to the axonal bifurcation of PFs . ( Average stack depth: 84 ± 20 µm ) . GC axons were reconstructed using the ‘Simple Neurite Tracer’ plugin ( Longair et al . , 2011 ) in Fiji ( ImageJ , NIH , USA ) . This plugin allowed us to assess the continuity of axons between several cross-sections . GC ascending axons were then fully traced and measured within the Z-limits of image sections . The size of the different layers of cerebellar cortex was reconstructed in each Z-stack . To avoid variability , all distances were normalized to the corresponding molecular layer height . Current-clamp data were analyzed using custom-written procedures in Igor Pro software ( WaveMetrics , Oregon , USA ) as previously described ( Eshra et al . , 2019 ) . Intrinsic properties of GCs were determined from the injected currents that elicited the largest number of action potentials . The action potential threshold was defined as the membrane voltage at which the first derivative exceeded 100 V s−1 , the minimal action potential peak was set at −20 mV and the minimal amplitude at 20 mV . All action potentials with a half-width shorter than 50 µs and longer than 500 µs were excluded . Action potential voltage threshold and half-width were calculated from the average of the first five action potentials . If a trace contained less than five action potentials , only the first action potential was considered . The action potential frequency was determined by dividing the number of action potentials during the 300-ms-lasting current injection by 300 ms . Membrane capacitance , resting membrane potential and series resistance were read from the amplifier software ( HEKA ) after achieving the whole-cell configuration . Input resistance ( Rin ) was analyzed from alternating subthreshold current injections from −20 to 20 pA ( 2 pA steps ) . The resulting voltage was plotted against the injected current and a spline interpolation was performed to obtain the slope at the holding membrane potential ( 0 pA current injection ) . The peak-current from outside-out patches was determined from voltage steps ( −90 to +60 mV ) with Fitmaster software ( HEKA ) . Steady-state inactivation was determined from the last 2 ms of the respective sweep . Cells were only included if 50 pA <Imax < 1 nA to exclude potential whole-cell measurements and membrane-vesicles . EPSP measurements from PCs and EPSC measurements from GCs were analyzed with the Fitmaster software ( HEKA ) . For PC EPSPs , 20–80% rise time and time to peak were determined from the average of 30 individual single EPSPs . GCs EPSCs were averaged from 25 traces . To obtain the decay kinetics , single EPSPs/EPSCs were fitted with either one or two exponentials . The weighted time constant was calculated as: ( 4 ) τw=Aslow τslow +Afast τfastAslow +Afast Paired-pulse ratio was determined between the first and the 5th EPSP after stimulation with 100 Hz trains . Single EPSCs from inner- and outer-zone GCs were averaged and fitted with two exponentials . The decay kinetics and amplitude of the grand-average was used to implement the MF EPSCs for the Dynamic Clamp . The neuronal network consisted of varying numbers of MF inputs , GCs , and one PC and was implemented in Matlab ( The MathWorks , Inc , Natick , Massachusetts , R2017a ) . For each simulation , a random set of MF inputs was generated . This input was then fed to a layer of integrate-and-fire GCs . An integrate-and-fire PC received the output of the GCs as EPSPs with delays based on PF conduction velocity . The PF-to-PC synaptic weights were optimized with the aim to make the PC spiking sequence similar to the target sequence . In the following , each component of the model is explained in detail . The Matlab scripts used to reproduce the model results in Figure 8 are available at: https://github . com/HallermannLab/2019_GC_heterogen ( Straub , 2019; copy archived at https://github . com/elifesciences-publications/2019_GC_heterogen/settings ) . Data are expressed as mean ± SEM or as box plots with median and interquartile range . The number of analyzed cells is indicated in the figures . To test for statistically significant differences , we performed Kruskal-Wallis ( for three groups ) or Mann-Whitney U tests ( for two groups ) and provide the p values ( PKr-Wa , or PMann-Whitney ) above the bar-graphs . In case of three groups , we performed non-parametric Dunn’s multiple comparisons post-hoc tests and provide the p values in the figure legends ( PDunn ) . Results were considered statistically significant if p<0 . 05 . | The timing of movements such as posture , balance and speech are coordinated by a region of the brain called the cerebellum . Although this part of the brain is small , it contains a huge number of tiny nerve cells known as granule cells . These cells make up more than half the nerve cells in the human brain . But why there are so many is not well understood . The cerebellum receives signals from sensory organs , such as the ears and eyes , which are passed on as electrical pulses from nerve to nerve until they reach the granule cells . These electrical pulses can have very different repetition rates , ranging from one pulse to a thousand pulses per second . Previous studies have suggested that granule cells are a uniform population that can detect specific patterns within these electrical pulses . However , this would require granule cells to identify patterns in signals that have a range of different repetition rates , which is difficult for individual nerve cells to do . To investigate if granule cells are indeed a uniform population , Straub , Witter , Eshra , Hoidis et al . measured the electrical properties of granule cells from the cerebellum of mice . This revealed that granule cells have different electrical properties depending on how deep they are within the cerebellum . These differences enabled the granule cells to detect sensory signals that had specific repetition rates: signals that contained lots of repeats per second were relayed by granule cells in the lower layers of the cerebellum , while signals that contained fewer repeats were relayed by granule cells in the outer layers . This ability to separate signals based on their rate of repetition is similar to how digital audio files are compressed into an MP3 . Computer simulations suggested that having granule cells that can detect specific rates of repetition improves the storage capacity of the brain . These findings further our understanding of how the cerebellum works and the cellular mechanisms that underlie how humans learn and memorize the timing of movement . This mechanism of separating signals to improve storage capacity may apply to other regions of the brain , such as the hippocampus , where differences between nerve cells have also recently been reported . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2020 | Gradients in the mammalian cerebellar cortex enable Fourier-like transformation and improve storing capacity |
Chromatin boundaries subdivide eukaryotic chromosomes into functionally autonomous domains of genetic activity . This subdivision insulates genes and/or regulatory elements within a domain from promiscuous interactions with nearby domains . While it was previously assumed that the chromosomal domain landscape is fixed , there is now growing evidence that the landscape may be subject to tissue and stage specific regulation . Here we report the isolation and characterization of a novel developmentally restricted boundary factor , Elba . We show that Elba is an unusual hetero-tripartite protein complex that requires all three proteins for DNA binding and insulator activity .
Eukaryotic chromosomes are subdivided into functionally autonomous domains by special elements called chromatin boundaries ( Gaszner and Felsenfeld , 2006 ) . Boundaries define units of independent genetic activity by shielding genes and/or regulatory elements within a domain from adventitious interactions with regulatory elements/genes in nearby domains . The specific functions that can be ascribed to most boundary elements include an enhancer-blocking or insulator activity ( Holdridge and Dorsett , 1991; Kellum and Schedl , 1991; Chung et al . , 1993 ) , a silencer-blocking or barrier activity ( Kellum and Schedl , 1991; Chung et al . , 1993; Burgess–Beusse et al . , 2002 ) and , when combined in specific pairwise combinations an ability to bring distant chromosomal DNA segments in close proximity ( Cai and Shen , 2001; Muravyova et al . , 2001 ) . Boundary elements have a diverse array of functions . In the yeast S . pombe , boundaries help restrict and maintain the heterochromatin state of the silenced mating type locus ( Noma et al . , 2001 ) . In Drosophila , boundary elements play a critical role in the Bithorax complex ( BX-C ) , helping to ensure that the three BX-C homeotic genes properly specify segmental identity ( Maeda and Karch , 2006 ) . In vertebrates , boundaries have been implicated in controlling the expression of mRNAs encoding different neuronal Protocadherin-alpha isoforms and in regulating recombination and expression of immunoglobulin light and heavy chain genes ( Guo et al . , 2011; Ribeiro de Almeida et al . , 2011; Monahan et al . , 2012 ) . The first boundaries identified were able to block adventitious regulatory interactions irrespective of cell type or developmental stage ( Gyurkovics et al . , 1990; Holdridge and Dorsett , 1991; Kellum and Schedl , 1991; Chung et al . , 1993 ) . Moreover , the proteins conferring insulating activity in flies and vertebrates like Su ( Hw ) , BEAF , Zw5 , GAGA factor and CTCF are ubiquitously expressed and seemingly functional in all cell types . This led to the idea that boundaries are static structures and consequently that the regulatory domain landscape of eukaryotic chromosomes is largely invariant from one cell to the next . However recent studies argue that the domain landscape is more dynamic than previously imagined and that genes can be differentially regulated during normal development and differentiation and in the progression of disease states like cancer by redefining their domain organization ( Bell and Felsenfeld , 2000; Witcher and Emerson , 2009; Gomes and Espinosa , 2010 ) . For example , regulation of the imprinted Igf2 and H19 genes pivots on parent of origin differences in the DNA methylation pattern and consequently the activity of a CTCF-dependent boundary element upstream of the H19 gene ( Bell and Felsenfeld , 2000; Hark et al . , 2000 ) . In this and also other examples , regulatory domains are defined or redefined locally by modulating how a ubiquitous insulator protein functions at a specific boundary element . However , one mechanism for defining domains that has yet to be described are developmental stage or cell type specific changes in the available repertoire of boundary proteins . In previous studies we discovered that the insulating activity of one of the Drosophila Bithorax ( BX-C ) complex boundaries , Fab-7 , depends upon distinct , stage specific boundary factors ( Schweinsberg and Schedl , 2004 ) . Like other boundaries in BX-C , Fab-7 is required to ensure the functional autonomy of its flanking parasegment specific cis-regulatory domains , iab-6 ( parasegment 11 ) and iab-7 ( parasegment 12 ) ( Gyurkovics et al . , 1990; Figure 1 ) . When Fab-7 is deleted the iab-6 and iab-7 domains no longer function independently and instead fuse into a single domain that incorrectly specifies parasegment 11 . While the primary function of Fab-7 in BX-C is to prevent inappropriate crosstalk between regulatory elements in iab-6 and iab-7 , it can also block interactions between enhancers/silencers and promoters both in the context of the endogenous BX-C ( Mihaly et al . , 1997 ) and in reporter gene assays ( Hagstrom et al . , 1996; see Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 00171 . 003Figure 1 . The Bithorax Complex , the Fab-7 boundary and the Elba recognition element . ( A ) Drosophila Bithorax complex ( BX-C ) . BX-C spans ∼300 kb and includes three Hox-family genes Ultrabithorax , abdominal-A and Abdominal-B . Parasegment specific expression of these three homeotic genes is generated by a series of functionally autonomous cis-regulatory domains: abx/bx , bxd/pbx and iab-2-iab8 , 9 . Functionally autonomy depends upon boundary elements that lie between each cis-regulatory domain ( Maeda and Karch , 2010 ) . One of these boundary elements is Fab-7 , which is located in between the iab-6 and iab-7 cis-regulatory domains . Both in the context of BX-C and in transgene assays , the Fab-7 boundary can block the action of enhancers/silencers at all stages of development , apparently irrespective of tissue or cell type ( Galloni et al . , 1993; Hagstrom et al . , 1996; Mihaly et al . , 1997; Schweinsberg et al . , 2004 ) . ( B ) The Fab-7 boundary spans a sequence of 1 . 2 kb and consists of two prominent and one minor ( * ) chromatin specific nuclease hypersensitive regions ( shown as yellow boxes ) . There is a third prominent nuclease hypersensitive region ( blue ) just distal to the boundary , which corresponds to a Polycomb Response Element ( PRE ) for the iab-7 cis-regulatory domain ( Maeda and Karch , 2010 ) . The orange box is a ∼100 base pair ( bp ) high-homology region which is conserved among Drosophila species ( >90% ) ( Aoki et al . , 2008 ) . The ovals are binding sites for Trithorax-like ( GAGA factor ) . ( C ) pHS1 is a 236-bp fragment from the proximal side of HS1 which has enhancer-blocking activity only in early embryos ( Schweinsberg and Schedl , 2004 ) . pHS1 includes the high-homology region and two GAGA-binding sites . These two GAGA sites are important for the early boundary activity of Fab-7 , while GAGA sites elsewhere in Fab-7 are needed later in development ( Schweinsberg et al . , 2004 ) . In addition to the GAGA sites , the enhancer-blocking activity of pHS1 in early embryos also depends upon an 8-bp sequence , CCAATAAG , called Elba ( Early boundary activity ) . Mutations in this sequence compromise the blocking activity of a 4×pHS1 multimer , while multimerization of a 27-bp oligo spanning the Elba sequence ( 8×Elba ) [see ( D ) ] is sufficient to confer early blocking activity . The Elba sequence is recognized by the stage-specific Elba DNA-binding factor . Elba factor binding is detected in 0–6 hr nuclear extracts , but it is absent in 6–12 hr ( and 6–18 hr ) nuclear extracts ( Aoki et al . , 2008 ) . ( D ) Sequence of the 27-bp oligo used as the Elba probe in the EMSA experiments shown in Figures 3A , 4 , 5B , and 6 . The Elba factor in 0–6 hr nuclear extracts recognizes the 8-bp Elba sequence ( shaded by yellow ) and requires an additional 5 bp both upstream and downstream for full binding activity ( shaded by light yellow ) . The bases underlined were altered as indicated in the mutant oligos , M1–M6 . These mutant oligos were used as cold competitors in Figures 4C and 6C as indicated . For the DNA affinity beads , a 27-bp oligo containing the mutation M3 was used as the mutant Elba sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 00310 . 7554/eLife . 00171 . 004Figure 1—figure supplement 1 . Enhancer blocking activity of Fab-7 , pHS1×4 and the Elba×8 multimer . In this enhancer blocking assay , putative boundary elements are placed in between two fushi tarazu ( ftz ) enhancers ( UPS and NE ) and an hsp70:LacZ reporter . The UPS enhancer is active in early embryos and drives LacZ expression a seven stripe pair-rule pattern . The NE enhancer is active later in development and drives LacZ expression in the central nervous system . ( A ) When a ‘non-specific’ DNA sequence is placed between the ftz enhancers and the hsp70:LacZ reporter , no blocking is observed . In early embryos ( stage 10–11 in this assay ) , the ftz UPS enhancers drive LacZ expression in seven stripes . In older embryos ( stage 13–14 ) the NE enhancer drives LacZ expression in the CNS . ( B ) When Fab-7 is place between the ftz enhancers and the reporter , it blocks both enhancers from activating the hsp70 promoter and there is little if any LacZ expression in either early or late embryos . ( C ) When pHS1×4 ( four copies of the 236-bp pHS1: see Figure 1 ) is placed between the enhancers and the reporter , it blocks the UPS enhancer from activating stripe expression . However , boundary activity is stage specific and pHS1×4 does not block the NE from activating LacZ expression in the CNS in older embryos ( Aoki et al . , 2008 ) . ( D ) When mutations ( M3 in Figure 1 ) that disrupt Elba binding in nuclear extracts are introduced into the Elba sequence of pHS1 , the enhancer blocking activity of the mutated pHS1×4 is compromised . In this case , both UPS stripe and NE CNS LacZ expression is observed ( Aoki et al . , 2008 ) . ( E ) When a 27-bp oligo containing the Elba binding sequence is multimerized ( Elba×8 ) and placed in between the ftz enhancers and the reporter , it blocks UPS driven LacZ stripe expression . It also causes a reduction in the extent of NE activation . This effect is position dependent . The blocking activity seen at this stage with the 8× multimer is likely due to the presence of some residual Elba activity in older embryos as seen in EMSA experiments with 6–12 hr nuclear extracts ( Aoki et al . , 2008 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 004 Transgene assays and deletion mutations within BX-C have shown that the Fab-7 boundary spans a DNA segment of 1 . 2 kb that contains two major nuclease hypersensitive sites , HS1 and HS2 , and a minor hypersensitive site ‘*’ ( see Figure 1B ) . Like other well-characterized Drosophila boundaries , it has insulating activity throughout development apparently irrespective of tissue or cell type both in the context of BX-C and in transgene assays . However , Fab-7 appears unusual in that its constitutive boundary function is generated by sub-elements whose activity is developmentally restricted . The first evidence for this developmental restriction came from the effects of mutations in binding sites for the GAGA factor in enhancer blocking assays ( Schweinsberg et al . , 2004 ) . As shown in Figure 1B , there are three pairs of GAGA sites in HS1 . Mutations in the proximal pair weaken boundary activity in early embryos , but have little effect from mid-embryogenesis onwards . In contrast , mutations in the central pair weaken boundary activity during mid-embryogenesis and in adults , but not in early embryos . Further evidence for sub-elements with developmentally restricted boundary activity came from experiments in which small fragments from HS1 were multimerized . Multimers of small 2- to 400-bp fragments from the distal half of HS1 were found to block enhancer-promoter interactions from mid-embryogenesis onwards even more efficiently that the intact Fab-7 boundary . However , these fragments have less blocking activity than Fab-7 in early embryos . Conversely , a multimerized 236-bp fragment , pHS1 , containing the proximal pair of GAGA sites has enhancer blocking activity during early embryogenesis , but not thereafter . This is illustrated in Figure 1—figure supplement 1 which shows that pHS1×4 blocks the fushi tarazu ( ftz ) UPS stripe enhancer in early embryos , but doesn't block the ftz NE enhancer during mid-embryogenesis . Consistent with these transgene results , two partial Fab-7 deletions in BX-C that remove the distal half of HS1 plus H2 , but retain the proximal pHS1 sequence including pair of GAGA sites function as boundaries during early embryogenesis but not later in development ( Schweinsberg and Schedl , 2004 ) . In addition , the early boundary activity of these partial deletions depends upon the GAGA factor ( Schweinsberg et al . , 2004 ) . With the aim of identifying factors in addition to GAGA that confer the early boundary activity of pHS1 , we used probes from pHS1 for EMSA ( electrophoresis mobility shift assay ) experiments with staged nuclear extracts . Only one DNA binding activity , Elba , had a stage-specificity consistent with the insulating activity of pHS1 ( Aoki et al . , 2008 ) . It is present in extracts from early 0–6 hr embryos , a period when pHS1 boundary activity is high . In contrast , in extracts from older 6–12 hr embryos , where pHS1 boundary activity in vivo is largely absent , only little Elba is detected . Elba recognizes an asymmetric 8-bp sequence , CCAATAAG ( see Figure 1D ) , which is conserved in the Fab-7 elements of other Drosophila species including the distant melanogaster relative D . virilis . Moreover , two lines of evidence indicate that the Elba factor is important for boundary activity in early embryos . We found that mutations in the recognition sequence that disrupt Elba binding in nuclear extracts compromise pHS1 insulator activity in vivo . Conversely , multimerizing the Elba recognition sequence is sufficient to confer early insulating activity ( Aoki et al . , 2008; see Figure 1—figure supplement 1 ) . Understanding the role of Elba in the context of BX-C and more generally in the establishment of chromatin domains during early embryogenesis requires the identification and characterization of this novel boundary factor . We describe here a general cross-affinity purification strategy for identifying components of multi-protein DNA binding complexes by mass spectrometry . Using this purification strategy , we show that the Elba boundary factor is an unusual hetero-tripartite protein complex . Two of the proteins , Elba1 and Elba2 , share a conserved C-terminal ‘BEN domain’ ( Abhiman et al . , 2008 ) . The third protein , Elba3 , has no obvious conserved protein domains , but is encoded by a gene that is closely linked to Elba1 . All three Elba proteins are required to reconstitute DNA binding activity in vitro . DNA binding depends upon the BEN domains in Elba1 and Elba2 , while Elba3 mediates the assembly of an active Elba complex by interacting with the N-terminal domains of Elba1 and Elba2 . All three Elba proteins are present in the Elba complex detected in 0–6 hr nuclear extracts and all three are required for chromatin domain boundary function in vivo . Finally , because elba1 and elba3 are ‘mid-blastula transition’ genes , Elba only binds to its' target sequences in Fab-7 and confers insulator activity during early embryogenesis .
We tried to isolate Elba directly from nuclear extracts by DNA-affinity purification . However , excess non-specific DNA-binding activity in the extracts inhibited Elba binding to the affinity beads and necessitated the use of a pre-purification procedure ( see Figure 2 , ‘Materials and methods’ ) . Following S-Sepharose chromatography , the active fraction was split into two aliquots and fractionated on either wild type ( WT ) or mutant ( M ) DNA affinity beads . Since Elba was substantially enriched in the 1 . 0 M KCl fraction from WT beads ( WE1 ) , while there was little in the corresponding fraction from mutant beads ( WE2 ) ( not shown ) , we analyzed the protein composition by mass spectrometry . However , in two independent single DNA-affinity purifications it wasn't possible to identify the Elba factor as far too many proteins were detected in ME1 ( and also ME1 ) and none stood out from the rest ( see ‘Materials and methods’ ) . 10 . 7554/eLife . 00171 . 005Figure 2 . Elba purification scheme . Details provided in ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 005 To reduce the number of remaining non-specific binding proteins and increase the differences in Elba yield between WT and M affinity beads , we devised the cross affinity purification scheme shown at the bottom of Figure 2 . As before , the S-Sepharose faction was incubated with either WT or M affinity beads . However , instead of isolating Elba from the 1 . 0 M KCl eluates , we recovered it from the supernatant of WT ( WS ) and M ( MS ) affinity beads , respectively . As expected Elba was substantially depleted from WS , but enriched in MS ( compare WS Input and MS Input in Figure 3A ) . We then fractionated WS on mutant affinity beads , and MS on WT affinity beads . Figure 3A shows that substantial amounts of Elba were recovered in the 1 . 0 M KCl fraction ( MSW1 ) from WT beads , but not from mutant beads ( WSM1 ) . 10 . 7554/eLife . 00171 . 006Figure 3 . EMSA and Elba factor proteins . ( A ) EMSA of fractions from the cross-affinity purification . The 32P-labeled Elba probe was incubated with fractions as indicated and subjected to 4% acrylamide-gel electrophoresis . El: Elba shift . P: probe . ( B ) Schematic of the Elba factors and BEN domain-containing ( green ) orthologs . BTB/POZ ( yellow ) domain is absent from Elba factors but is present in related proteins . ( C ) Sequence alignment of the C-terminal half of the Drosophila Elba1 , Elba2 and a third Drosophila Ben protein Insv ( Insensitive ) . The sequences of the three proteins were aligned according to the results of NCBI ( National Center for Biotechnology Information ) blast search ( bl2seq ) . The amino acid residues conserved in more than two proteins are shaded with red . The residues that have similarities with each other are shaded with yellow . The predicted BEN domain region is boxed with pale blue . ( D ) Sequence alignment of Drosophila and mammalian ( human ) orthologs of BEN domain proteins . The C-terminal sequences of Elba1 , Elba2 and Insv were subjected to blast search with human databases . Within the BEN domain sequences , the closest human ortholog of Elba1 is BEND7 ( BEN domain-containing 7 ) /C10orf30 ( 29% identical , 46% positive ) , whereas for Elba2 the closest ortholog is BEND6/C6orf65 ( 29% identical , 53% positive ) . The BEN domain of Insv is most similar to three proteins: NAC2 ( Nucleus accumbens-associated protein 2 , NACC2 ) /BEND9 , BEND6 and NAC1 ( Nucleus accumbens-associated protein 1 , NACC1 ) /BEND9 . The amino acid residues conserved with each Drosophila protein are shaded with red . The residues that have similarities with each Drosophila protein are shaded with yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 006 The mass spectrometry dataset from the cross-affinity purification fractions was quite different from that of the conventional single-affinity purifications . First , the total number of proteins detected was greatly reduced , suggesting that many non-specific binding proteins were pre-cleared by the first step in the cross-affinity purification . Second , while there were still 176 unique proteins in the MSW1 fraction , three of these were substantially enriched compared to all others ( Table 1 ) . Each had 16 or more confirmed peptides with high spectral counts and their sequence coverage was over 25% . One protein , Elba1: CG12205 , is encoded by a previously described mid-blastula transition gene , Bsg25A , of unknown function ( Singer and Lengyel , 1997 ) while the other two , Elba2: CG9883 and Elba3: CG15634 , are products of uncharacterized genes . 10 . 7554/eLife . 00171 . 007Table 1 . List of proteins unique to the 1 . 0 M KCl fraction from wild type DNA affinity beads ( MSW1 ) in the third , cross-affinity , Elba purificationDOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 007Hit proteinSequence countSpectrum countSequence coverageMol . wt . 1CG12205 ( Bsg25A ) -PA/gi|7295685*2337256 . 5%/49%*41 , 583/47 , 368*2CG15634-PA1615032 . 20%43 , 0083CG9883-PA176326 . 80%43 , 3104CG12052 ( lola ) 17 subtypes32110 . 8–5 . 2%49 , 320–98 , 1625CG12052 ( lola ) -PY3218 . 70%62 , 7466CG12052 ( lola ) , unknown subtype3215 . 10%107 , 3037CG12052 ( lola ) -PD , -PE3216 . 70%79 , 4398CG14339-PA2172 . 00%117 , 9089CG2368 ( psq ) -PD , -PE , -PF , PG , -PH/-PB , -PC61512 . 7%/7 . 7%70 , 298/114 , 98410CG1249 ( snRNP2 ) -PA31331 . 90%13 , 50411CG6944 ( Lam ) -PA , -PB , -PC81219 . 10%71 , 24912CG16973 ( msn ) -PE4114 . 50%115 , 38613CG16973 ( msn ) -PC , -PD/-PB /-PA4114 . 3%/3 . 9%/3 . 1%120 , 610/130 , 341/162 , 37814CG11700 ( CR11700 ) -PA21110 . 60%34 , 33515Reverse_CG312842113 . 10%110 , 29816CG3561 ( KH1 ) -PA81015 . 30%59 , 59717CG10067 ( Act57B ) -PA/CG7478 ( Act79B ) -PA/CG18290 ( Act87E ) -PA , -PB51014 . 40%41 , 835/41 , 787/41 , 80218CG5178 ( Act88F ) -PA51014 . 40%41 , 70019CG1759 ( cad ) -PA , -PB494 . 40%51 , 30620CG12154 ( oc ) -PA395 . 20%69 , 66621CG6801 ( l ( 3 ) j2D3 ) -PA8819 . 20%44 , 83022CG18124 ( mTTF ) -PA7816 . 30%48 , 28123CG7583 ( CtBP ) -PD , -PB , -PC , -PA6820 . 50%42 , 25224CG13634-PA284 . 30%52 , 59425CG3143 ( foxo ) -PC , -PB579 . 60%67 , 413*An alternative transcript of CG12205 ( gi:7295685 ) that encoded a slightly larger protein was listed in the Genbank database at the time of this experiment . However , that sequence was subsequently removed from the CG12205 sequence list . In the third , cross-affinity purification , there were 176 proteins in the MSW1 fraction ( see Figure 2 ) which were not present in the WSM1 fraction . These 176 proteins were sorted according to the descending order of ‘Spectrum Count’ numbers . The 25 proteins with the highest ‘Spectrum Count’ are shown here . The top three proteins have a high spectrum count , and also have a high sequence count . Although the three proteins have no previously known DNA binding domains , there are several intriguing connections . First , all three are ∼40 kDa , which is close in size to the protein species in nuclear extracts that is UV cross-linked to probes containing the Elba sequence ( Aoki et al . , 2008 ) . Second , the C-terminal ∼130 amino acids of Elba1 and Elba2 show extensive sequence similarities ( Figure 3C ) . Moreover , included within this region of similarity is a conserved ∼90 amino acids BEN domain ( Figure 3B ) that has been implicated in protein:protein interactions and transcriptional regulation ( Cha et al . , 1997; Mackler et al . , 2000; Wang et al . , 2006; Korutla et al . , 2009; Duan et al . , 2011 ) . Two other fly proteins have BEN domains . One is a predicted isoform of the Mod ( mdg4 ) C boundary factor , while the other is Insensitive ( Insv ) ( Figure 3C ) , which functions in neurogenesis and Notch signaling ( Duan et al . , 2011 ) . Interestingly , the elba2 and Insv transcription units are paired with each other ( Figure 7—figure supplement 1 ) . Mammals have a large family of BEND proteins . Of the mammalian proteins , Elba1 is most closely related to BEND7 , while Elba2 is most closely related to BEND6 ( Figure 3D ) . The BEN domain of Insv is most closely related to BEN9/hNAC2 and BEND8/hNAC1 . Finally , though Elba3 differs from Elba1 and Elba2 in that it has no distinctive domains , its transcription unit is located next to elba1 ( Figure 7—figure supplement 1 ) just like the elba2 and insv pair . To determine if one of these highly enriched proteins corresponds to the Elba factor , each was synthesized by in vitro translation of the corresponding full length mRNAs and tested for DNA binding activity . Figure 4A shows that none shifted the Elba probe . We next translated all three mRNAs together . Strikingly , this combination generated a prominent shift that co-migrates with the Elba shift produced by 0–6 hr nuclear extracts ( Figure 4A ) . To ascertain which proteins must be present to generate the Elba shift we in vitro translated each mRNA separately , and then mixed the translation reactions in all pairwise combinations . Figure 4B shows that none of the pairwise combinations had DNA binding activity; however , it was possible to reconstitute the Elba shift by combining all three translation products . 10 . 7554/eLife . 00171 . 008Figure 4 . Elba is a hetero-tripartite complex . ( A ) and ( B ) All three Elba proteins are required to reconstitute DNA binding activity . In vitro translated proteins either singly or in combination as indicated were incubated with the Elba probe . El: Elba shift . P: probe . ( C ) Reconstituted Elba has the same sequence specificity as nuclear Elba . Reconstituted and nuclear extract shifts with or without ( minus ) a 100-fold excess of competitor ( Comp ) as indicated above each lane . WT: wild-type probe . M1–M6: mutant probes ( see Figure 1D ) . Ctl: no-RNA control . ( D ) Nuclear Elba factor has all three Elba proteins . Nuclear extracts ( NE ) incubated with preimmune ( P ) or immune ( I ) polyclonal rabbit serum as indicated . #1 , #2: serum from different rabbits . SS: antibody supershifts . El: Elba shift . P: free probe . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 008 These findings demonstrate that all three Elba proteins are required to reconstitute an Elba-like DNA binding activity in vitro . To confirm that the properties of the reconstituted Elba factor match the endogenous factor in 0–6 hr nuclear extracts , we tested for sequence specificity . We previously identified the core recognition sequence by introducing a series of 3-bp point mutations into the Elba probe ( Aoki et al . , 2008; see Figure 1D ) . We used the same set of competitors to compare the in vitro reconstituted Elba factor with the endogenous factor in 0–6 hr nuclear extracts . Figure 4C shows that sequence specificity of reconstituted Elba is the same as nuclear extract Elba . Like the endogenous factor , the binding activity of reconstituted Elba is competed by excess amounts of wild type and M1 , M5 and M6 DNA , whereas M2 , M3 and M4 compete poorly or not at all . The in vitro reconstitution experiments indicate that Elba is a complex composed of three distinct proteins , Elba1 , Elba2 and Elba3 . To determine if these three proteins are also components of the endogenous nuclear factor , we generated two independent polyclonal antibodies against each Elba protein . We then tested the effects of the immune and corresponding pre-immune sera on the Elba shift in nuclear extracts . Figure 4D shows that all six immune sera give an Elba supershift , while the Elba shift is unchanged by the corresponding pre-immune sera . As an additional control for specificity we tested whether these sera would supershift probes from elsewhere in Fab-7 that are recognized by factors which are present in 6–12 hr but not 0–6 hr nuclear extracts . We found that they did not . These findings indicate that the Elba factor in 0–6 hr nuclear extracts must also contain all three Elba proteins . While these results do not exclude the possibility that the Elba factor in nuclear extracts contains additional proteins besides Elba1–3 , such proteins do not remain stably associated with Elba during its purification nor are they required for its DNA binding activity in vitro . As hetero-tripartite DNA binding factors are unusual and there are no previously known DNA binding domains in the Elba proteins , we sought to learn more about the organization and DNA binding activities of the Elba complex . In preliminary experiments , we found very weak , but detectable DNA binding activity when we mixed bacterially expressed N-terminal GST ( Glutahione-S-transferase ) fusions of the two BEN domain proteins , Elba1 and Elba2 , in the absence of Elba3 . Since GST is known to form stable dimers , we reasoned that the GST moieties might be substituting for Elba3 by promoting the formation of GST-Elba1:GST-Elba2 heterodimers ( Figure 6—figure supplement 2 ) . This finding together with the asymmetric Elba binding motif , CCAATAAG , suggested a model in which the BEN domains in Elba1 and Elba2 are responsible for DNA recognition , while the function of Elba3 is to bring Elba1 and Elba2 together so that they can form the DNA binding pocket ( Figure 5C ) . 10 . 7554/eLife . 00171 . 009Figure 5 . Elba protein deletions and functional organization of the Elba complex . ( A ) Elba proteins and deletion mutants . The bars under each diagram indicate the sequences retained in the mutant protein . The numbers correspond to the amino acid residues at the N and C terminal ends of the protein . The letter in parentheses is used to designate proteins added to the lanes in panel B and in Figure 6 . Each protein was expressed either with an N-terminal FLAG tag or with an N-terminal FLAG tag plus a GST tag . The FLAG tag was used to determine the relative amount of each protein so that the input of the translated proteins in the gel shift experiments could be adjusted . ( B ) BEN domain is required for the DNA-binding activity of Elba . EMSA experiments were performed as described in Figure 4 by translating the full-length ( F ) or ∆BEN ( ∆B ) mutant Elba subunits in vitro and by mixing each as indicated above the lanes . ( C ) A schematic structure of proposed model for the Elba complex . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 009 To test this model , we deleted the BEN domain from the Elba1 and Elba2 cDNAs ( Figure 5A ) and determined whether the resulting in vitro translated ∆BEN proteins can reconstitute DNA binding activity . Figure 5B shows that mixing the three full-length proteins ( which have N-terminal FLAG tags as do all of the other proteins in Figure 5A ) generates the Elba shift , while the shift is eliminated whenever a BEN domain deletion is used instead for reconstitution . We next determined whether the combination of Elba1 and Elba2 is sufficient to reconstitute DNA binding if they are provided with heterologous GST dimerization domains . To maximize the formation of heterodimers , we co-translated mRNAs encoding the GST-Elba1F and GST-Elba2F fusions and assayed for DNA binding activity . As shown in lane 6 of Figure 6A , a prominent shift is generated when these two GST fusions are co-translated . In contrast , the individual GST fusions fail to shift the Elba probe on their own even though they are expected to form homodimers ( Figure 6A , lanes 7 and 8 ) . Consistent with the idea that only the heterodimer is able to bind to the Elba probe , we found that there is little DNA binding activity when the two GST fusions are translated independently and then mixed together just before assaying ( not shown ) . In this case , DNA binding activity is likely limited by a slow rate of dissociation of the GST-ElbaF homodimers . The fact that a heterologous dimerization domain can support what appears to be near full DNA binding activity would be consistent with the idea that the role of Elba3 is to link Elba1 and Elba2 together ( Figure 6—figure supplement 1 ) . While GST can substitute for Elba3 , the supershift generated by the addition of Elba3 ( Figure 6A , lane 5 ) indicates that the GST-Elba1F:GST-Elba2F heterodimer is still able to form a trimeric complex with Elba3 . 10 . 7554/eLife . 00171 . 010Figure 6 . Characterization of Elba complex . ( A ) DNA binding activity of different Elba protein variants . The proteins were translated in vitro from the mixed mRNAs shown above the lanes . All proteins used in this figure are FLAG tagged and approximately the same amounts of the translated proteins were added to each lane . The identities of the proteins added to each lane are indicated above the lane . For example , lane 3 has all three full length Elba proteins , while in lane 5 full length Elba3 is combined with full length Elba1 and Elba2 proteins that have an N-terminal GST tag . Lane 13 has GST fused to the C-terminal half of Elba1 and Elba2 , while in lane 16 the C terminal halves of Elba1 and Elba2 lack the GST moiety . ( B ) GST-fused Elba1F:Elba2F ( F ) or Elba1C:Elba2C ( C ) binds to the Elba probe while Elba1N:Elba2N ( N ) does not . ( C ) Sequence specificity of the GST-Elba1C:GST-Elba2C dimer is the same as the nuclear/reconstituted hetero-tripartite complex ( Figure 4 ) . The EMSA experiment with GST-Elba1C:GST-Elba2C were performed in the absence ( lanes 1–3 ) or presence of 100-fold excess of non-labeled competitors as indicated above the lanes . The sequences of competitor DNAs are shown in Figure 1 . ( D ) Elba1C and Elba2C proteins lacking the N-terminal GST moiety have a weak DNA binding activity . Position of Elba1C:Elba2C shift is indicated by closed arrowhead . Position of the Elba1C:Elba2C FLAG supershift is indicated by open arrowhead . Proteins added to each lane including FLAG and GST antibodies are indicated above the lane . Note that the same shift was detected in lane 16 of ( A ) when the X-ray film was exposed for a longer period of time . ( E ) Supershifts of the GST-Elba1F:GST-Elba2F generated by the addition of Elba3 . Proteins corresponding to the full length Elba3 ( F ) , or the N-terminal ( N ) and C-terminal ( C ) halves of Elba3 were added to an EMSA reaction mix containing the GST-Elba1:GST-Elba2 dimer . Proposed structures of the native Elba complex , the artificial GST-dependent dimer by GST-Elba1:GST-Elba2 and GST-Elba1C:GST-Elba2C are shown in Figure 6—figure supplement 1–3 , respectively . DOI: http:dx . doi . org/10 . 7554/eLife . 00171 . 01010 . 7554/eLife . 00171 . 011Figure 6—figure supplement 1 . A schematic model for the tripartite Elba complex . Elba3 links Elba1 and Elba2 through sequences in their N-terminus . This brings the C-terminal halves of Elba1 and Elba2 together to form a ‘DNA binding pocket’ which binds to the asymmetric Elba recognition sequence . DOI: http:dx . doi . org/10 . 7554/eLife . 00171 . 01110 . 7554/eLife . 00171 . 012Figure 6—figure supplement 2 . The artificial GST-Elba1:GST-Elba2 heterodimer binds to the asymmetric Elba recognition sequence . The N-terminal GST moiety mediates dimerization of Elba1 and Elba2 , mimicking the coupling of these two proteins by Elba3 . DOI: http:dx . doi . org/10 . 7554/eLife . 00171 . 01210 . 7554/eLife . 00171 . 013Figure 6—figure supplement 3 . The artificial heterodimer of GST-Elba1C:GST-Elba2C binds to the asymmetric Elba recognition sequence . The C-terminal 165–175 residues of Elba1 and Elba2 retain DNA-binding activity with the same sequence-specificity as hetero-tripartite Elba complex when they are dimerized by GST moieties . DOI: http:dx . doi . org/10 . 7554/eLife . 00171 . 013 We entertained the possibility that the BEN deletions don't remove DNA binding domains , but instead eliminate sequences used by Elba3 for linking Elba1 and Elba2 . In this case , the GST moiety should substitute for the BEN domain . However , this is not the case . When the BEN domain was deleted from either one or both of the GST-fused Elba1 and Elba2 proteins ( ∆BEN ) , DNA-binding activity was completely abolished ( Figure 6A , lanes 9–11 ) . To further characterize the functional domains in the three Elba proteins we subdivided them into N-terminal and C-terminal halves ( Figure 5A ) . As expected from the experiments described above , the N-terminal halves of Elba1 and Elba2 had no DNA binding activity either with ( Figure 6B ) or without the GST moiety . For the C-terminal halves of Elba1 and Elba2 we first generated cDNAs that span the entire 130 amino acids homology region ( including the 90 amino acids BEN domain ) either with or without the N-terminal GST ( Figure 5A ) . Like the full-length GST fusions , the GST-ElbaC homodimers have no detectable DNA binding activity on their own ( Figure 6A , lanes 14 and 15 ) , while a very prominent shift is generated when the RNAs encoding GST-Elba1C and GST-Elba2C are co-translated ( lane 13 ) . In addition to having roughly equivalent activity to the ‘native’ hetero-tripartite Elba complex generated by the three full length Elba proteins , the GST-Elba1C:GST-Elba2C heterodimer exhibits a similar sequence specificity ( Figures 4C and 6C ) . The two ElbaC proteins also resemble the corresponding full-length Elba1 and Elba2 proteins in that full DNA binding activity requires the dimerization activity of the GST ( compare Figure 6A , lanes 13 16 ) . On the other hand , the two ElbaC proteins differ from the full length Elba1 and Elba2 proteins in two respects . Most importantly , the addition of Elba3 to the two ElbaC proteins ( without the GST moiety ) does not reconstitute full DNA binding activity ( Figure 6D , lane 7 ) . This finding suggests that Elba3 orchestrates the formation of the hetero-tripartite complex by interacting with sequences in the N-terminal domains of Elba1 and Elba2 ( Figure 5C ) . Consistent with this idea , the shift generated by the active GST-Elba1C:GST-Elba2C heterodimer differs from the full-length heterodimer in that it is not supershifted by the addition of Elba3 ( Figure 6A , lane 12 ) . A second potentially interesting difference is that the mixture of Elba1C and Elba2C proteins ( lacking GST ) gives a faint shift that can be detected in long ( Figure 6D , lane 8 filled arrowhead ) but not short ( Figure 6A , lane 16 ) exposures . Like the shift generated by the corresponding GST proteins , this shift is not enhanced or supershifted by the addition of Elba3 ( lane 7 ) and it requires both proteins ( Figure 6D , lanes 9 and 10 ) . Interestingly , it can be stabilized as well as supershifted by the addition of FLAG antibodies ( but not GST antibodies: lanes 11 and 12 ) . Although further studies will be required , it would appear that there are weakly active hetero-dimerization motifs in the C-terminal halves of Elba1 and Elba2 that become more readily accessible when the N-terminus of the two proteins is removed . We also generated cDNAs that encode the 90 amino acids BEN domain and the remaining C-terminal amino acids . As shown in Figure 6A , we failed to detect a shift with the GST-heterodimers between these two ‘BEN domain’ proteins ( lane 17 ) . It is possible that the close proximity of the GST moiety in these proteins precludes formation of a dimer that can form a stable DNA–protein complex . Alternatively , sequences in the extended 130 amino acid homology region might be important in facilitating DNA binding ( e . g . , aligning the two BEN domains ) . As for the N- and C-terminal halves of Elba3 , neither supported complex formation when mixed with the two other Elba proteins ( not shown ) nor did they supershift the shift generated by the GST-Elba1F:GST-Elba2F dimer ( see Figure 6E ) . One important question is the basis for the developmentally restricted activity of the Elba factor . As all three proteins co-migrate with yolk protein they are difficult to visualize or quantitate by Western blots . For this reason we used Northerns to examine the expression of the three elba genes during development . These experiments indicate that the DNA binding/boundary activity of the Elba factor is developmentally restricted because of the temporal expression patterns of the paired elba1 and elba3 genes . Figure 7 and Figure 7—figure supplement 1 show that neither is expressed during oogenesis , and that their mRNAs are largely absent in 0–2 hr embryos . Consistent with the previous identification of elba1 as a mid-blastula transition gene ( Singer and Lengyel , 1997 ) , expression of elba1 and also elba3 peaks at the blastoderm stage ( 2–4 hr ) , and then disappears . The disappearance of the elba1 and elba3 mRNAs occurs in the same time frame as the loss of boundary activity in transgene embryos and DNA binding activity in nuclear extracts . In much older embryos elba3 mRNAs are detected , but are present at a lower level than in the blastoderm stage ( see also Figure 7—figure supplement 1 ) . While expression of elba1 and elba3 is developmentally restricted , elba2 and its neighboring ortholog , insv , appear to be expressed throughout much embryogenesis . Moreover , both are expressed during oogenesis and likely are maternally deposited , as high levels are present in 0–2 hr embryo before zygotic transcription is activated . 10 . 7554/eLife . 00171 . 014Figure 7 . Expression of mRNAs encoding Elba factors and Insv . Total RNA from ovaries and staged embryos as indicated were probed with cDNAs encoding each protein . Shown on the bottom is the ethidium bromide staining of a gel for the Northern blotting as a loading control . The positions of the RNA markers are indicated on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 01410 . 7554/eLife . 00171 . 015Figure 7—figure supplement 1 . Loci of Elba proteins and the expression patterns of their mRNAs during the development . ( A ) modEncode expression profiles of mRNAs encoding Elba1 and Elba3 . As indicated in the diagram , elba1 ( CG12205 ) and elba3 ( CG15634 ) are transcribed in opposite directions and are separated from each other by sequences encoding much of the CG11929 open reading frame http://flybase . org/reports/FBgn0000227 . html/ , http://flybase . org/reports/FBgn0031621 . html . As seen in the Northerns , there is little or no maternal deposition of elba1 or elba3 mRNAs . These two genes are not expressed until the mid-blastula transition and the highest levels of elba1 and elba3 mRNAs are found in 2–4 hr embryos . Relatively high levels of both mRNAs are found in 4–6 hr embryos and then both largely but not completely disappear during the remainder of embryogenesis . In our Northern experiments transcript levels also drop dramatically after 6 hr , but in older embryos both elba1 and elba3 are detected again . However , it is likely that the amounts of elba1 and elba3 in these later stages are overestimated in our experiments . Our gels were overloaded , and transfer of the two elba mRNAs ( which migrate very near the bottom of the rRNA band ) in the 2–4 hr samples does not appear to be as efficient as it is at later stages when their levels are greatly reduced . The presence of only quite low levels of elba1 and elba3 mRNAs at these later stages would be consistent with the modEncode data . ( B ) modEncode expression profiles of mRNAs encoding Elba2 and Insv . Like elba1 and elba3 , elba2/CG9883 and insensitive/CG3227 are closely linked and transcribed in opposite orientations . As seen in our Northern experiments , the modEncode temporal expression patterns of the mRNAs encoding these two proteins are quite different from that of elba1 and elba3 http://flybase . org/reports/FBgn0031434 . html/ , http://flybase . org/reports/FBgn0031435 . html . Both are maternally deposited and high levels are present throughout embryogenesis . Also unlike elba1 and elba3 , both transcripts are expressed at later stages of development . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 015 Taken together , the findings described above suggest a plausible model for why the insulator activity of boundaries depending on the Elba recognition sequence is restricted to early embryos . A functional hetero-tripartite Elba factor would be assembled in early embryos and would bind to boundary elements containing the Elba recognition sequence ( Figure 8C ) . Later in development , after the elba1 and elba3 mRNAs disappear , the hetero-tripartite Elba factor would disappear as well . In this case , Elba2 wouldn't be able to bind to boundaries containing the Elba recognition sequence on its own . We used chromatin immunoprecipitation ( ChIP ) of staged 2–5 hr and 9–12 hr embryos to test this model . As standard formaldehyde based ChIP protocols gave inconsistent enrichment of Fab-7 sequences , we adapted the cross-linking procedure of Nowak et al . ( 2005 ) for Drosophila embryos . Figure 8A , B shows that as predicted Elba1 and Elba2 are bound to the Fab-7 Elba sequence in 2–5 hr embryos , but not to control sequences from the twine ( twe ) and Sex-lethal ( Sxl ) genes . In contrast , neither protein appears to be bound to Fab-7 in 9–12 hr embryos . These findings dovetail nicely with the developmental profiles of Elba blocking activity in transgene assays and Elba binding activity in nuclear extracts . In addition , even though elba2 is expressed in 9–12 hr embryos , the Elba2 protein isn't found associated with the Fab-7 Elba sequence in the absence of the two other Elba proteins . 10 . 7554/eLife . 00171 . 016Figure 8 . Elba proteins are bound to Fab-7 in early but not late embryos . ( A ) and ( B ) Elba1 and 2 ChIPs . Early ( 2–5 hr ) or late ( 9–12 hr ) embryos were cross-linked and after processing immunoprecipitated with Elba1 or Elba2 antibodies , or pre-immune serum . Sequences from the pHS1 region of Fab-7 or control twine and Sex-lethal sequences were detected by qPCR . The y-axis shows the average immune/preimmune ratio . *p<0 . 05 . ( C ) Model showing binding and enhancer blocking by the Elba complex in early but not late embryos . GAF: GAGA factor and its binding sites . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 016 We sought evidence connecting Elba binding in vitro and in vivo to Fab-7 and its boundary activity during early embryogenesis . Previous studies indicate that a combination of functionally redundant factors , not just Elba , generate Fab-7 boundary activity during early embryogenesis and at other stages of development ( Mihaly et al . , 1997 ) . Consequently , the activity of the intact Fab-7 boundary in the context of BX-C or in transgene assays is not expected to be disrupted by knocking down the Elba factor or even by mutations in the Elba binding site . On the other hand , unlike the intact boundary , Elba seems to play a more critical role in the insulating activity of the pHS1 multimer ( Aoki et al . 2008 ) . For this reason , we injected double-stranded RNA ( dsRNA ) specific for each Elba protein into embryos transgenic for the pHS1×4 pCfhl boundary reporter . As a control for specificity , we also injected dsRNA specific for the BEN domain gene insv , the close relative of elba1 and elba2 . pChfl has two ftz enhancers and a LacZ reporter . The UPS enhancer drives LacZ expression in a stripe pattern during early embryogenesis , while the NE enhancer drives expression in the CNS during mid-embryogenesis . Figure 9A shows that both enhancers activate expression at the appropriate stage when a random ( or DNA ) is interposed between them and the reporter . When pHS1×4 is placed between the enhancers and the reporter , it blocks the UPS stripe enhancer from activating LacZ expression in early embryos , but unlike the full length Fab-7 boundary ( see Figure 1—figure supplement 1 ) it doesn't block the NE enhancer in mid-embryogenesis . Note that like Fab-7 , pHS1×4 must be interposed between the enhancers and the reporter to block activation ( Schweinsberg and Schedl , 2004 ) . Buffer injection had no apparent effect on pHS1×4 blocking of the UPS stripe enhancer . As observed for uninjected embryos , more than 50% of the buffer injected controls in each experiment were Class 1 and did not express LacZ ( Figure 9B ) . In contrast , UPS blocking activity was disrupted when embryos were injected with dsRNA specific for one of the Elba proteins . For both elba1 and elba3 , most fell into Class 3–5 , while for elba2 most were Class 2–4 . The loss of boundary activity is specific to constituents of the Elba complex , as injection of dsRNA for the closely related insv did not disrupt boundary activity . Similar results were obtained in two other sets of pHS1×4 dsRNA injection experiments , and in experiments using embryos transgenic for the Elba multimer ( Figure 9—figure supplement 1 ) . Like the set shown here , the disruption of boundary activity in these experiments was greatest for elba1 and elba3 , while smaller effects were observed for elba2 . The differences in expression pattern of the elba mRNAs most likely accounts for this differences in sensitivity to dsRNA . While expression of elba1 and elba3 would be turned on in most instances after injection of the dsRNA , there is a substantial pool of maternally derived elba2 mRNAs that could already be translationally engaged at the time of injection . Likewise the variability in the loss of boundary activity in elba injections most probably reflects differences in age of the embryos when they were injected , as well as variations in the amount and location of the injected dsRNAs . 10 . 7554/eLife . 00171 . 017Figure 9 . Hetero-tripartite Elba complex mediates early boundary activity . ( A ) Top: fushi tarazu ( ftz ) enhancers drives stripe expression ( UPS ) in stage 10–11 embryos and central nervous system expression ( NE ) in stage 13–14 embryos . Bottom: Four copies of Fab-7 pHS1 ( which contains the Elba sequence ) blocks the UPS , but not the NE enhancer . elba1-3 and insv dsRNAs or buffer alone were injected into embryos transgenic for the ftz-4×pHS1-LacZ enhancer blocking reporter and blocking activity examined . In each experiment ∼100 embryos were photographed and categorized into Class 1–5 as indicated . Graphs show the percentage of embryos in each class for RNAi and the buffer control in that experiment . Three independent injection experiments were done for each protein . All gave similar results and only one is shown here . p values from the t-tests are shown in the graphs . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 01710 . 7554/eLife . 00171 . 018Figure 9—figure supplement 1 . Double-stranded RNA injection into embryos transgenic for the Elba×8 multimer reporter and a reporter containing non-specific DNA in the blocking position . ( A ) Embryos transgenic for the ftz-Elba×8 ( multimer ) -Hsp70:LacZ reporter ( see Figure 1—figure supplement 1 ) were injected with dsRNAs specific for each of the elba genes . The injected embryos were allowed to develop until stage 10–11 and then stained for LacZ expression . In each experiment , buffer-injected control embryos were prepared and processed in parallel with the dsRNA injected embryos . Embryos were photographed and then categorized into Class 1–5 according to the intensity of ftz UPS-enhancer stripes . ( Figure 9 shows representative examples of each class . ) The graphs show the percentage of embryos ( y-axis ) that fell into each class ( x-axis ) in the elba dsRNAi and the corresponding control buffer injection . As was observed in the pHS1×4 injection experiments , the blocking activity of the Elba×8 multimer is compromised by injection of dsRNA for each of elba gene . The loss of blocking activity can be seen by comparing the class distribution for the dsRNA injection with the class distribution for the buffer control . ( Note that the elba2 and elba3 dsRNA injections shown here were part of the same injection experiment and have the same buffer injected control . ) The dsRNA injections for the two mid-blastula transition genes elba1 and elba3 have a greater effect on the blocking activity of the Elba×8 multimer than the dsRNA injections for elba2 . This was observed in other Elba×8 multimer injection experiments and was also seen for the pHS1×4 reporter ( see Figure 9 ) . The smaller reduction in blocking activity observed with elba2 dsRNA injections is most likely due to the presence of substantial amounts of maternal elba2 mRNA . ( B ) elba and insv dsRNA injections do not affect LacZ expression from a control reporter which has a non-specific DNA sequence in the blocking position . The expressions of the UPS LacZ stripes in dsRNA and in the buffer injected or no injection controls give a similar class distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 00171 . 018
The constitutive insulator activity of the Fab-7 boundary is generated by a series of functionally redundant sub-elements whose activity is developmentally restricted . Here we report the identification and characterization of a factor , Elba , which confers the insulator activity of one of these sub-elements in early embryos . Elba is an unusual hetero-tripartite complex . It consists of two ∼40 kDa proteins , Elba1 and Elba2 , that have a C-terminal 90 amino acid BEN domain embedded in a larger 130 amino acid region of homology . The BEN domain is found in insect and vertebrate nuclear proteins and has been implicated in protein:protein interactions and transcriptional regulation ( Abhiman et al . , 2008 ) . The third protein , Elba3 has no distinctive domains and seems to be limited to the Drosophilids . All three proteins are present in the Elba complex detected in nuclear extracts and are required to reconstitute Elba DNA binding in vitro . Our functional studies indicate that DNA binding is mediated by the C-terminal domains of Elba1 and Elba2; however , in order to form a DNA binding ‘pocket’ that recognizes the asymmetric Elba sequence , CCAATAAG , Elba1 and Elba2 must be linked in a ‘heterodimer’ ( Figure 5C ) . In the tripartite Elba complex , this seems to be the role of Elba3 . It brings Elba1 and Elba2 together by interacting with sequences in the N-terminal half of the two proteins . These conclusions are supported by a number of findings . First , it is possible to circumvent the requirement for Elba3 by fusing a heterologous dimerization domain , in this case GST , to the N-terminus of full length Elba1 and Elba2 . When co-translated , these two fusion proteins shift the Elba probe without Elba3 . On the other hand , they can still interact with Elba3 as the shift generated by the two proteins can be supershifted by the addition of Elba3 . Second , the Elba1C and Elba2C proteins , which span the 130 amino acid C-terminal homology region , can reconstitute full DNA binding activity when fused to GST . Since GST is known to dimerize , but not form higher order complexes , this would argue that the Elba DNA binding pocket is likely formed by a GST-Elba1C:GST-Elba2C dimer rather than by some other more complicated combination of the two proteins . Moreover , this binary complex has the same sequence specificity as the native hetero-tripartite complex . On the other hand , the GST-Elba1C:GST-Elba2C proteins differ from the full length GST fusions in that they don't interact with Elba3 . While these findings are consistent with the model for the Elba complex shown in Figure 5C , there are several unresolved but important issues . For one , though our experiments indicate that the BEN domain is essential for DNA binding , we were unable to demonstrate that it is sufficient . It is possible that steric hindrance from the closely linked GST moiety prevents formation of the DNA binding pocket; alternatively , the extended homology region may contain elements that are important for DNA binding . Another apparent anomaly is that the homodimers formed by the full length or the C-terminal GST fusions don't appear to bind to the Elba probe . Since we've found that the Elba factor can bind to variants of the Elba sequence , albeit with reduced affinity , we would have expected that the homodimers would exhibit at least some evidence of DNA binding activity . Further studies will be required to resolve these issues . As would be the case for many other boundaries , the presence of functionally redundant elements in the full length Fab-7 precludes a direct demonstration that a single factor like Elba is needed for Fab-7 insulating activity either in the context of BX-C or in transgenes assays . However , several lines of evidence argue that Elba does in fact have such a function for the endogenous Fab-7 boundary . To begin with , previous studies showed that the insulating activity of the 236-bp pHS1 Fab-7 sub-element in early embryos is compromised when the Elba recognition sequence is mutated . Moreover , when multimerized , the Elba sequence is sufficient on its own to confer insulating activity . That the Elba factor is responsible for this insulating activity is supported by the effects of RNAi knockdowns in embryos carrying either the pHS1×4 transgene or the Elba sequence multimer . For all three Elba proteins , RNAi knockdowns compromises the insulating activity of both pHS1×4 and the Elba multimer . In contrast , knockdowns of the closely related BEN domain protein , Insv , have no effect . Our findings would also explain why the insulating activity of the Elba factor/Elba sequence is developmentally restricted . Two of the three Elba proteins , Elba1 and Elba3 , are encoded by genes that are active during the mid-blastula transition , but not later in development . Since the hetero-tripartite complex is required to reconstitute DNA binding activity , Elba insulating activity would be expected to peak during the blastoderm/early gastrula stage when high levels of elba1 and elba3 mRNA are expressed . However , it should gradually dissipate after transcription of elba1 and elba3 ceases . While we don't know precisely when Elba1 and Elba3 disappear , there is little Elba DNA binding activity in nuclear extracts of 6–12 hr embryos . Moreover , even though the elba2 gene is expressed throughout development , it is not found associated with Fab-7 in 9–12 hr embryos . Importantly , the results of the EMSA and ChIP experiments provide strong support for the idea that the requirement for Elba activity evident in the RNAi knockdowns reflects a direct role in insulating activity rather than indirect role . One important question is why does a constitutive insulator like Fab-7 utilized developmentally limited factors like Elba ? A plausible explanation comes from the finding that boundaries are not autonomous entities , but rather function in combination ( likely pairwise ) with other boundaries ( Cai and Shen , 2001; Muravyova et al . , 2001 ) . Both boundary competition ( Gohl et al . , 2011 ) and boundary bypass experiments ( Kyrchanova et al . , 2008a; Maksimenko et al . , 2008 ) indicate that some combinations are functional while others are not . Moreover , how different boundary combinations work together depends upon developmental stage and tissue ( Gohl et al . , 2011 ) . Thus , the use of stage specific factors at Fab-7 could reflect a need to optimize combinations with other nearby BX-C boundaries or the Abdominal-B promoter ( Kyrchanova et al . , 2008b , 2011 ) . A number of findings are consistent with this idea . When heterologous boundaries are used to replace Fab-7 in BX-C , they are unable to provide bypass activity , while their insulating activity can be lost in a stage and tissue specific fashion ( Hogga et al . , 2001 ) . In contrast , Fab-7 is able to replace Fab-8 ( Iampietro et al . , 2008 ) . Likewise , the boundary activity of Fab-7 in a foreign environment is not only context dependent , but also varies during development ( Gohl et al . , 2011 ) . Probably the most direct evidence that boundaries in BX-C like Fab-7 are closely matched with their flanking neighbors comes from the boundary bypass experiments of Kyrchanova et al . ( 2011 ) . They found that bypass is observed when Fab-7 is combined with the neighboring boundaries Fab-6 and Fab-8 , while bypass interactions with boundaries from elsewhere in BX-C are only weak at best . Interestingly , in the cases that have been analyzed most thoroughly , boundary bypass is mediated by homologous protein:protein interactions ( Kyrchanova et al . , 2008a ) . Thus , bypass is observed when multimerized binding sites for the dCTCF boundary factor are paired , but not when dCTCF binding sites are paired with sites for example Zw5 . Though Fab-8 differs from Fab-7 in that it utilizes dCTCF ( Holohan et al . , 2007; Kyrchanova et al . , 2011 ) , EMSA experiments with 0–6 hr and 6–12 hr nuclear extracts indicate that they also have binding sites for several of the same stage specific factors ( Aoki et al . , 2008; Wolle et al . , unpublished data ) . These include sites for the Elba factor , and sites for two different late ( 6–12 hr ) stage specific factors . Conceivably this could also be true for Fab-6 . Another question is the fate of Elba2 after Elba1 and Elba3 disappear . While Elba2 can't bind to the Elba sequence on its own , it could function as an insulator in other contexts using different protein partners and presumably also recognition sequences . A plausible partner would be the Notch signaling pathway protein Insv , which is encoded by a tightly linked gene that has a similar expression pattern to elba2 . Insv and its closest mammalian relatives , the NAC proteins ( which are thought to function as stem cell pluripotency genes , as well as in cocaine addiction and cancer ) , are believed to regulate transcription by acting as co-repressors ( Cha et al . , 1997; Mackler et al . , 2000; Wang et al . , 2006; Korutla et al . , 2009; Duan et al . , 2011 ) . However , since our findings indicate that the BEN domains of Elba1 and Elba2 are essential for DNA binding , it would be reasonable to suppose that Insv as well as the mammalian counterparts are not co-repressors , but are rather sequence specific DNA binding proteins . Interestingly , like members of the BEN domain protein family , CTCF was initially thought to be a transcriptional repressor ( Filippova et al . , 1996 ) . Thus , a seemingly plausible speculation would be that other members of the BEN family besides the two Elba proteins function as insulators restricting the action of nearby enhancer elements instead of directly repressing transcription . Potentially supporting this idea is the finding that one of the mammalian BEN domain proteins , SMAR1 , is a component of the nuclear matrix and associates with Matrix Attachment Regions ( MARS ) ( Chattopadhyay et al . , 2000 ) . If this were the case , this would mean that CTCF is not the only vertebrate insulator protein . Also as is case for the Elba complex , modulating the expression or activity of Elba2 , Insv or NAC1 in different tissues or cell types could change the regulatory domain landscapes , and potentially alter global patterns of gene expression .
Oregon R population cages were used to collect embryos of the appropriate stage on apple juice plates . The embryos were washed off the plates , dechorionated with 50% bleach ( 2 . 6% sodium hypochloride ) for 3 min , rinsed with 0 . 12 M NaCl/0 . 04% Triton X-100 and then 0 . 12 M NaCl , and frozen in liquid nitrogen . Embryos were stored at −80°C until the extraction . The Canton S embryos ( generous gifts from Dr . James T Kadonaga and Dr . Rock Pulak ) of various ages were also used for testing different steps in the purification procedure . Small-scale embryonic nuclear extracts were prepared from 10 g of 0- to 6-hr-old Oregon R embryos as described previously with small modifications ( Aoki et al . , 2008 ) . The frozen embryos were suspended in 30 ml homogenization buffer ( HB; 3 . 75 mM Tris–HCl pH7 . 4/0 . 5 mM EDTA–KOH pH7 . 4/20 mM KCl/0 . 05 mM spermine/0 . 125 mM spermidine/0 . 5% 2 , 2′-thiodiethanol/2 μg/ml aprotinin/0 . 1 mM phenylmethylsulfonyl fluoride: PMSF/0 . 1% digitonin ) and disrupted using a motorized teflon-glass homogenizer for 10 strokes and then with a glass–glass Potter homogenizer for an additional 10 strokes . The lysate was filtered through two layers of Miracloth ( EMD Millipore , Billerica , MA ) to remove debris and centrifuged at 1900×g for 5 min in a swinging bucket rotor . The resulting nuclear pellet was resuspended and washed four times with 40 ml of HB and finally suspended in 2 ml of Nuclear Extraction Buffer 20 ( NEB20: 10 mM HEPES–KOH pH7 . 4/20 mM KCl/3 mM MgCl2 0 . 1 mM EDTA 10% glycerol/1 mM dithiothreitol: DTT/0 . 2 mM PMSF/2 μg/ml Aprotinin ) . The nuclear suspension was transferred to a polyallomer ultracentrifuge tube ( Sarstedt 65-90219 ) and an equal volume of NEB 700 ( same as NEB 20 except that the KCl concentration was 700 mM ) was added to the suspension to give a final concentration of 360 mM KCl . After 30 min incubation at 4°C , the sample was centrifuged at 150 , 000×g for 1 hr at 4°C in a Beckman SW50 . 1 swinging bucket rotor . The resulting supernatant ( ∼ 4 ml ) was divided into aliquots for storage at −80°C . For the purification of the Elba factor nuclear extracts were prepared from 60 g ( first procedure ) or 100 g ( second and third procedures ) of Oregon R 0–12 hr ( first and second ) or 0–6 hr ( third ) embryos . The embryos were divided into aliquots of 10 g each and the extract was prepared as described above except that 1 . 0% ‘Nonidet P-40 substitute’ ( Sigma-Aldrich , currently designated as ‘Igepal CA-630’ ) was used instead of 0 . 1% digitonin in HB . Because Elba activity is destabilized by freezing–thawing and dilution , the extracts were processed through the phenyl-TSK hydrophobic chromatography ( see ‘Pre-purification’ ) on the same day as the extraction . The Elba probe ( Figure 1D ) was labeled with 32P and used for EMSA assay under the same conditions as described previously ( Aoki et al . , 2008 ) except for the concentration of the non-specific competitor poly ( dI-dC ) :poly ( dI-dC ) in the binding reaction . The final concentration of poly ( dI-dC ) :poly ( dI-dC ) was varied between 12 . 5 and 250 μg/ml depending on the relative purity of Elba factor . The salt concentrations in the samples were adjusted so that the final concentration of KCl would be about 0 . 1 M in the binding reaction . For the EMSA experiments shown in Figure 4: ( A ) Control ‘nuclear extracts’ correspond to 1 μl of 0–6 hr nuclear extracts ( corresponding to about 2 . 5 mg of embryos ) from the small scale preparation described above . ( B ) In the competition experiments shown in Figures 4C and 6C , the indicated cold competitor DNA was present in 100-fold excess over the labeled probe . ( C ) For the EMSA experiments using the three in vitro-translated proteins: In Figure 4A , 6 μl of the in vitro translated proteins were used in each lane . In Figure 4B , C , 1 μl of each in vitro translated protein was used by itself or in combination as indicated . The total amount of rabbit reticulocyte lysate was adjusted between the lanes so that the non-specific DNA-binding activities would be the same . ( D ) In the ‘super-shift’ experiment in Figure 4D , 1 μl each of pre-immune or immune serum was added to the incubation mix containing 1 μl of the small-scale nuclear extract . Trichloroacetic acid ( TCA; Sigma-Aldrich ) was added to the 1 . 0 M KCl fractions from the affinity beads so that the final concentration of TCA was 25% . After an overnight incubation at 4°C , the proteins were collected by centrifugation at 20 , 000×g for 30 min . The protein pellets were washed two times with ice-cold 30% TCA and then two times with ice-cold acetone . The dried protein samples were subjected to trypsin digestion followed by electrospray ionization mass-spectrometry ( ESI-MS ) as described previously ( Washburn et al . , 2001; Bern et al . , 2004 ) . The software DTASelect and Contrast were used for peptide data analyses and protein predictions ( Tabb et al . , 2002 ) . Approximately 5 μg of plasmids encoding Elba cDNA clones were digested with appropriate restriction enzymes and used as templates for in vitro transcription . The capped mRNAs were transcribed using T3 RNA polymerase ( for Elba2; Promega ) or T7 RNA polymerase ( for Elba1 and Elba3; USB ) in the presence of ribonucleotide-triphosphates ( NTPs ) and the cap-analog m7GpppG ( Promega ) at 40°C for 1 hr . After removing the plasmids with RNase-free DNase ( RQ DNase; Promega ) , the mRNAs were extracted with phenol/chloroform and precipitated with ethanol . The mRNA pellets were suspended in 50 μl of RNase-free water and 3 . 5 μl each was used for one translation reaction of 25 μl . ( In the mixed translation of three proteins in Figure 4A , 1 . 17 μl of each mRNA was used . ) A rabbit retiulocyte lysate ( Promega ) was used for in vitro-translation reactions . To confirm that the mRNA directs the synthesis of a protein of the appropriate molecular weight , radio-labeled proteins were synthesized in parallel reactions containing 35S-methionine/cystine ( Tran35S-label; MP Biomedicals ) . In the in vitro translations of tagged Elba mutant proteins in Figures 5 and 6 , all the plasmids were digested with Asp718 ( Roche Diagnostics ) and transcribed with T7 RNA polymerase . The protein products were detected with anti-FLAG antibodies ( M2 from Sigma-Aldrich or anti-DYKDDDDK 1E6 from Wako ) in the Western blotting . Rabbit polyclonal antibodies against the three Elba proteins were generated by injecting bacterially-expressed proteins into rabbits . The Elba proteins were expressed using the 6×His-T7 tag pET28 vector or the 6×His-HA tag from a modified pET15 vector . Because most of the bacterially expressed proteins from these pET vectors were insoluble , the recombinant proteins were isolated from the whole bacterial lysate by SDS-PAGE gels . The proteins were eluted from the PAGE gels with the Bio-Rad Electro-eluter 422 . The recovered proteins were mixed with TiterMax Gold adjuvant ( Sigma-Aldrich ) and injected into two rabbits each . The 6×His-T7 tagged proteins ( 200–650 μg per rabbit ) were used for initial immunization while the 6×His-HA tagged proteins were used for all of the boosting injections . For the boosting injections , 180–500 μg protein were injected every 4 weeks . 15–25 ml of blood were collected every 2 weeks , and the titers of the sera were checked by Western blotting . The reactive sera were stored at −80°C . cDNAs encoding the Elba factors were obtained from the Drosophila Genomics Resource Center ( DGRC ) . The clone names: RE24665 ( pFlc-1-CG12205/Elba1 ) , LD10908 ( pBluescriptSK ( − ) -CG9883/Elba2 ) , LD42284 ( pOT2-CG15634/Elba3 ) . For in vitro transcription experiments , these plasmids were prepared using cesium chloride ultracentrifugation . The protein-coding region of CG3227/insensitive ( insv ) cDNA was obtained by RT-PCR using embryo RNA and subcloned into pBluescriptKS ( + ) ( Stratagene ) . For antibody production , the plasmids for 6×His-T7-tagged proteins were constructed by amplifying the protein-coding regions of Elba2 or Elba3 cDNAs with PCR and by introducing them into the Bam HI/Xho I sites of the pET28c vector ( Novagen ) . The upstream PCR primers included a Bam HI site , whereas the downstream PCR primers had either Sal I or Xho I sites . In the case of the Elba1 cDNA , which has an internal Bam HI site , the full coding sequence was amplified using a 5′ PCR primer that had a Bam H1 site and 3′ primer that had a Sal I site . The PCR product was digested with Bam H1 and introduced into pET28c . The resulting plasmid was digested with Hind III ( which cuts in the Elba1 cDNA ) and Xho I ( which cuts in the vector ) . It was then ligated to the Hind III-Sal I fragment generated by digestion of the full length Elba1 PCR product . The plasmids for 6×His-HA-tagged proteins were constructed using a two-step procedure . First , a modified pBluescriptKS ( + ) vector that contains the influenza HA tag at the Xba I–Bam HI sites ( pKS ( + ) HA; Aoki , unpublished data ) was used to attach the HA tag to 5′ end of each cDNA for the Elba proteins . The Elba cDNA fragments were generated by PCR and then introduced into at Bam HI/Sal I sites or Bam HI/Xho I sites of pKS ( + ) HA . The resulting HA-fused cDNAs were PCR amplified using an Xho I-HA-upstream primer and downstream T3 primer . The PCR products were digested with Xho I and introduced into the Xho I site of pET15b vector ( Novagen ) . A series of plasmids encoding protein tags were constructed based on the pBluescriptKS ( + ) ( Stratagene ) for the in vitro transcription of tagged Elba proteins in Figures 5 and 6 . First a pKS ( + ) FLAG vector was generated by introducing a phosphorylated double-strand DNA fragment encoding a methionine followed by the FLAG tag at the Xba I and Bam HI sites . The resulting plasmid was cut at Bam HI and additional tags , such as GST , were introduced . For the FLAG-GST-fused proteins , pKS ( + ) FLAG-GST was constructed by introducing a PCR-amplified and Bgl II/Bam HI-digested cDNA fragment encoding GST into pKS ( + ) FLAG . The non-GST-tagged Elba proteins also had HA , c-myc or Streptavidin-binding peptide ( Keefe et al . , 2001 ) tags at their N-terminus . The partial cDNA fragments of Elba proteins were amplified by PCR and introduced into the protein tag-encoding vectors described above . Further details upon request . The total RNA samples were prepared from about 200–300 mg of appropriately aged embryos and dissected ovaries using the QuickPrep Total RNA Extraction Kit from Amersham-Pharmacia ( #27-9271-01; currently discontinued by GE Healthcare ) . In the Northern blotting experiment 30 μg per lane of total RNA was applied to a 1 . 0% agarose gel ( cast in 0 . 66 M formaldehyde/1× MOPS buffer [0 . 2 M 3- ( N-morpholino ) propanesulfonic acid/50 mM sodium acetate/10 mM EDTA] ) and electrophoresis was performed in 1× MOPS buffer for 3 hr at 100 V . The separated RNA was transferred to a Zeta-Probe membrane ( Bio-rad ) by capillary blotting and cross-linked with a Stratalinker ( Stratagene ) UV cross-linker . For probes , DNAs corresponding to the protein-coding sequences of Elba1-3 and Insv were prepared by PCR amplification or by digesting plasmids with appropriate restriction enzymes . They were 32P-labeled using α-32P dCTP and Ready-To-Go DNA Labeling beads ( -dCTP; GE Healthcare ) . After denaturating in boiling water , the probes were mixed with hybridization buffer ( 0 . 2 M sodium phosphate pH7 . 2/1% BSA/7% SDS ) and hybridized with membranes overnight . The following day , the membranes were washed several times with 0 . 2× SSC/0 . 1% SDS at 65°C and exposed to the X-ray film . Staged 2–5 hr ( early ) and 9–12 hr ( late ) Oregon R embryos were collected , dechorionated and weighed . The embryos were cross-linked using a modification of the procedure of Nowak et al . ( 2005 ) . Chromatin IP steps were performed as previously described by Kappes et al . ( 2011 ) except that the washing of the immunoprecipitated beads was simplified to five successive washes with RIPA buffer only . One IP sample corresponds to approximately 100 mg of embryos . In each experiment a pair of immune-IP and pre-immune-IP samples were processed in parallel . 5 μl of immune or pre-immune serum was used for Elba1 ( anti-Elba1 #1 in Figure 4D ) and 2 5–10 μl each of serum was used for Elba2 ( anti-Elba2 #1 in Figure 4D ) . The precipitated DNA samples were suspended in 30 μl of distilled water . Supplementary file 1 shows the primer pairs that were used to detect the target locus Fab-7 HS1 ( Fab-7 ) as well as two control loci Sex-lethal ( Sxl ) and twine ( twe ) . The quantitative PCR ( qPCR ) was performed with triplicated 25 μl reactions using Power SYBR Green ( Applied Biosystems ) and 0 . 5 μl of sample DNA in Agilent M×3000P qPCR system . The ratio of immune/pre-immune precipitation was calculated by the comparative Ct method ( ∆∆CT Method ) . The chromatin IPs were done four times for Elba1 and five times for Elba2 and the significance was determined using the unpaired t-test . We used in vitro synthesized double strand RNAs for the RNAi injection experiments ( Kennerdell and Carthew , 1998 ) . A short 350- to 420-bp cDNA fragment was amplified from the elba1-3 and insv cDNAs using complementary PCR primers that also contained T7 promoter sequences at their 5′ ends . The T7 promoter-cDNA fragments were purified from agarose gels and used as templates for T7 RNA polymerase ( MEGA script T7; Ambion ) . The transcribed RNA strands were annealed by heating and then chilling , treated with DNase and purified by phenol/chloroform extraction and ethanol precipitation . The dsRNAs were suspended in injection buffer so that the final concentration was approximately 1 μg/μl ( 3 . 6–4 . 3 μM ) . Transgenic embryos were collected on 10-cm apple juice agar plates from fly cups incubated at 18°C for 30 min . They were lined up on a piece of apple juice agar and then transferred onto double-stick tape attached to a cover glass . The embryos were covered with Halocarbon oil 27 and dsRNA was injected on the ventral side at the middle of the embryo . Each set of injected embryos was incubated on cover glasses at different temperatures ( 18°C , room temperature , or 25°C ) so that they would develop to stage 10–11 at approximately the same time . The staged embryos were washed with PBS from the cover slips into a cup and then stained with X-gal as described previously ( Aoki et al . , 2008 ) . For each experiment , buffer-injected embryos were also prepared and processed in parallel . Approximately 100 embryos from each dsRNA/buffer injection were photographed and the photographed embryos classified into Classes 1–5 according to the intensity of the ftz UPS enhancer driven LacZ stripe expression . | If all of the DNA in a human cell was stretched out , it would be about 2 m long . The nucleus of a human cell , on the other hand , has a diameter of just 6 μm , so the DNA molecules that carry all the genetic information in the cell need to be carefully folded to fit inside the nucleus . Cells meet this challenge by combining their DNA molecules with proteins to form a compact and highly organized structure called chromatin . Packaging DNA into chromatin also reduces damage to it . But what happens when the cell needs to express the genes carried by the DNA as proteins or other gene products ? The answer is that the compact structure of chromatin relaxes and opens up , which allows the DNA to be transcribed into messenger RNA . Indeed , packing DNA into chromatin makes this process more reliable , thus ensuring that the cell only produces proteins and other gene products when it needs them . However , because cross-talk between neighboring genes could potentially disrupt or change gene expression patterns , cells evolved special elements called boundaries or insulators to stop this from happening . These elements subdivide eukaryotic chromosomes into functionally autonomous chromatin domains . Since the protein factors implicated in boundary function seemed to be active in all tissues and cell types , it was assumed for many years that these boundaries and the resulting chromatin domains were fixed . However , a number of recent studies have shown that boundary activity can be subject to regulation , and thus chromatin domains are dynamic structures that can be defined and redefined during development to alter patterns of gene expression . Aoki et al . report the isolation and characterization of a new fruit fly boundary factor that , unlike previously characterized factors , is active only during a specific stage of development . The Elba factor is also unusual in that it is made of three different proteins , known as Elba1 , Elba2 , and Elba3 , and all three must be present for it to bind to DNA . While Elba2 is present during most stages of development , the other two Elba proteins are only present during early embryonic development , so the boundary factor is only active in early embryos . In addition to revealing a new mechanism for controlling boundary activity as an organism develops , the studies of Aoki et al . provide further evidence that chromatin domains can be dynamic . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"biochemistry",
"and",
"chemical",
"biology"
] | 2012 | Elba, a novel developmentally regulated chromatin boundary factor is a hetero-tripartite DNA binding complex |
Axon guidance is proposed to act through a combination of long- and short-range attractive and repulsive cues . The ligand-receptor pair , Netrin ( Net ) and Frazzled ( Fra ) ( DCC , Deleted in Colorectal Cancer , in vertebrates ) , is recognized as the prototypical effector of chemoattraction , with roles in both long- and short-range guidance . In the Drosophila visual system , R8 photoreceptor growth cones were shown to require Net-Fra to reach their target , the peak of a Net gradient . Using live imaging , we show , however , that R8 growth cones reach and recognize their target without Net , Fra , or Trim9 , a conserved binding partner of Fra , but do not remain attached to it . Thus , despite the graded ligand distribution along the guidance path , Net-Fra is not used for chemoattraction . Based on findings in other systems , we propose that adhesion to substrate-bound Net underlies both long- and short-range Net-Fra-dependent guidance in vivo , thereby eroding the distinction between them .
Net , a secreted protein , and its receptor DCC together play a critical role in wiring the brain in both vertebrates and invertebrates . Net-DCC mediated axon guidance has been well characterized in the developing vertebrate spinal cord . In brief , Net expressed at the floor plate is proposed to diffuse to establish a decreasing ventral-to-dorsal gradient within the spinal cord ( Kennedy et al . , 2006; Serafini et al . , 1996 ) . This gradient , in turn , is proposed to promote the guidance of commissural neuron growth cones ventrally to the floor plate ( Fazeli et al . , 1997; Serafini et al . , 1996 ) . Classic in vitro studies using purified proteins and explant cultures also support a role for Net as a chemoattractant ( de la Torre et al . , 1997; Kennedy et al . , 1994 ) . Net-DCC based axon guidance to the midline is an evolutionarily conserved mechanism in nervous system development . In C . elegans , loss of UNC-6 - UNC-40 ( the C . elegans homologs of Net-DCC ) signaling leads to dorsal displacement of axon tracts that are positioned ventrally in wild-type ( Chan et al . , 1996; Hedgecock et al . , 1990; Ishii et al . , 1992 ) . And in Drosophila , Net and Frazzled ( Fra , the Drosophila homolog of DCC ) are required for axons to cross the midline of the embryonic ventral nerve cord ( Harris et al . , 1996; Kolodziej et al . , 1996 ) . Here , a number of neurons from each side of the midline send their axons contralaterally , creating the commissures of the ventral nerve cord . Net is expressed at the midline by glia and Net protein is found as a gradient that peaks at the midline . In Net or fra mutants , the commissures of the ventral nerve cord are missing or severely reduced , consistent with a loss of chemoattraction to the ligand source . The spatial relationship between a Net-responsive growth cone and the source of Net in the fly visual system is similar to both the fly midline and vertebrate spinal cord . The fly visual system is modular , comprising some 750 columns . Different neuronal cell types , including R8 , extend axons within each column , where they terminate in different layers . R8 growth cones reach their target , the M3 layer of the medulla neuropil , in two steps ( Ting et al . , 2005 ) ( Figure 1a ) . They ‘park’ at a temporary target at the outer surface of the medulla , referred to as M0 , and then , after a delay , extend to and terminate within the M3 layer . Salecker and colleagues demonstrated that R8 targeting to M3 requires Net-Fra signaling ( Timofeev et al . , 2012 ) . R8s express Fra; Net is expressed by L3 growth cones within the M3 layer and is seen as a shallow gradient that stretches from M0 to M3 , peaking sharply at M3 ( Timofeev et al . , 2012 ) . R8 growth cones extend from M0 to M3 within a dense neuropil containing the processes of many different cell types . Without Fra or Net , many R8 terminals remain in M0 while others are stranded between M0 and M3 ( Timofeev et al . , 2012 ) . 10 . 7554/eLife . 20762 . 003Figure 1 . Live imaging of R8 growth cones in the developing fly brain . ( a ) Schematic of R8 targeting . ( b ) , Top panel: Confocal micrograph of the medulla at 45 hAPF in a fly expressing a membrane-tethered variant of NetB ( NetB::TM ) from the NetB genomic locus . R8s ( magenta , myr::tdTOM ) and NetB ( green , myc ) are shown . Bottom panels: Higher magnification view of the boxed region in the top panel; the R8 and NetB::TM channels are displayed separately . Arrowheads bracketing the NetB::TM image mark the position of the linescan plotted to the right . Graph: Linescan of the fluorescence intensity in NetB::TM channel . Gray region marks background values . ( c ) Sample setup . Brain is shown in coronal section , viewed head-on . Body axes , ( D ) orsal- ( V ) entral and ( R ) ight- ( L ) eft , are marked . ( d ) Detail from ( b ) illustrates the imaged volume . ( e ) Three growth cones from the same WT brain . Panels were individually contrast enhanced to reveal dimmer features . See Figure 1—figure supplement 1 for a description of the image processing work-flow . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 00310 . 7554/eLife . 20762 . 004Figure 1—source data 1 . Contains numerical data plotted in Figure 1b . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 00410 . 7554/eLife . 20762 . 005Figure 1—figure supplement 1 . Processing 2-photon time series . ( a ) Major steps of the image processing work-flow: Medulla Registration: The R8 array is treated as a rigid body to correct for developmental movement . Segmentation: One stack is used to perform the only segmentation step of the work-flow . Medulla column orientations are also determined at this step . Tracking: Growth cones are tracked through the time series . Alignment: Growth cones are aligned to themselves in 3D over the time series . ( b ) The principal movement during imaging is the roll of the medulla about the horizontal axis of the imaging plane ( i . e . X ) , corresponding to the ventral-right to dorsal-left body axis . Mean ( top ) and standard deviation ( bottom ) of the 3-axis rotations are plotted . ( c ) Left: Medulla roll over 12 hr . Images are displayed with cyan-hot and yellow-hot look-up-tables ( LUTs ) to capture more of the dynamic range with limited saturation . Red rectangles mark the same group of growth cones . Gaps in the R8 array are due to a stochastic element in our labeling system . Right: The 36 hr image , after medulla registration . ( d ) Segmented growth cones are outlined in magenta in a maximum intensity projection of the medulla . Image is displayed with a cyan-hot LUT . ( e ) Outlined growth cones in ( d ) plotted on the outer medulla surface . Medulla column vectors are shown as descending gray lines . ( f ) Growth cone tracks on the outer medulla surface plot their movement . Deeper shades of magenta denotes time progression . ( g ) Tracking output for two growth cones from ( d ) . Images are displayed with a cyan-hot LUT . Data are shown at 1/6 th of the full time resolution ( 10 min/frame ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 00510 . 7554/eLife . 20762 . 006Figure 1—figure supplement 1—source data 1 . Contains numerical data plotted in Figure 1—figure supplement 1b . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 006 The findings in the R8 system are consistent with Net acting as a chemoattractant , a secreted ligand diffusing to form a gradient , which promotes extension of growth cones to the source of the ligand . However , a different interpretation was favored , one that proposed a local function for Net-DCC at M3 in target layer recognition ( Timofeev et al . , 2012 ) . Central to this conclusion was the observation that a membrane-tethered variant of Net , expressed from the endogenous locus , can support wild-type targeting . The same molecular strategy had been used in the Drosophila embryonic midline to establish that diffusible Net is not required for the Fra-dependent guidance of commissural axons ( Brankatschk and Dickson , 2006 ) . This seminal work solidified the short-range paradigm of Net-DCC mediated axon guidance in which a soluble gradient of the ligand is not required for function . Notably , at both the embryonic midline and visual system , tethered Net is expressed in a graded fashion comparable to the secreted form . At the midline , the boundaries of Net expression largely reflect the extended lateral morphology of cells expressing Net , the midline glia ( Brankatschk and Dickson , 2006 ) . Similarly , although Net is prominently expressed in L3 growth cones within the M3 layer , it is also visible as a gradient along the R8 targeting path between M0 and M3 ( see Figure 1b ) . Thus , while these studies are consistent with a Net requirement that is limited to the extent of the source cells , it is not known whether the observed graded distributions are important for axon guidance to the target—the peak of the Net signal . Given the penetrance and expressivity of the fra and Net phenotypes in R8 ( see below ) and the genetic tools available in Drosophila to explore mechanisms of axon guidance , we used live imaging to explore R8 targeting to M3 in more detail . Here we report , through detailed quantitative analysis of hundreds of mutant and wild type growth cones in intact developing animals that R8 growth cones in Net mutants or R8 growth cones lacking Fra target from M0 to M3 in a fashion indistinguishable from wild type . That is , Net does not act as a chemoattractant nor does Fra act as a chemoattractant receptor . In addition , we present evidence suggesting that R8 growth cones can recognize the target layer without Net-Fra . Instead , Netrin , DCC , and TRIM9 , a signaling component directly downstream from DCC ( Alexander et al . , 2010; Hao et al . , 2010; Morikawa et al . , 2011; Song et al . , 2011; Winkle et al . , 2016 , 2014 ) , are essential for attachment of a single leading process extended from R8 growth cones to the target layer . We propose that R8 growth cones reach and recognize the target layer independent of Fra , adhere to the target layer in a Fra-dependent step , and this adhesion is consolidated by a TRIM9-dependent step . These findings favor the notion that in Drosophila Net mediates adhesion to neuronal processes or the extracellular matrix ( ECM ) at the target layer rather than promoting directed outgrowth to or recognition of the target layer .
To compare wild-type and mutant R8 targeting , we devised a live imaging protocol to follow growth cones in intact pupae as they extend from M0 to M3 ( Figure 1c–e , Figure 1—figure supplement 1 , Video 1 ) . This system is similar to one developed by Hiesinger and colleagues to study the cellular mechanism of neural superposition , the choreographed re-distribution of R1-6 growth cones from ommaditial bundles to lamina cartridges ( Langen et al . , 2015 ) . In the medulla neuropil , Hiesinger and colleagues used ex vivo live imaging to study R7 targeting , specifically characterizing the role of the Ca2+-dependent cell adhesion molecule , N-cadherin , in mediating adhesive interaction between growth cones and the developing neuropil ( Özel et al . , 2015 ) . 10 . 7554/eLife . 20762 . 007Video 1 . Medulla Registration . The developmental roll of the medulla is corrected in preparation for growth cone segmentation and tracking . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 007 Through analysis of hundreds of wild type R8 growth cones , we characterize four distinct steps of targeting ( Figure 2a , Video 2 ) . The first , extension , begins with the polarization of growth cones in M0; a single thin process appears at the medial side of each growth cone ( ~38 hAPF ) . For many growth cones , this process probes into the medulla with multiple excursions and retractions , at rates reaching 1 μm/min ( Figure 2c ) . Extension ends at 46 . 5 ± 1 . 5 hAPF with stabilization , as the tips of the R8 processes settle at 10 . 5 ± 1 . 5 μm from M0 ( Figure 2a , b ) . In the third step , elongation , the tips of R8 projections continue to move away from M0 at the much slower rate of ~2 . 2 μm/hr ( Figure 2a , b ) . 10 . 7554/eLife . 20762 . 008Figure 2 . Wild-type R8 targeting . ( a ) Steps of WT targeting . Orange arrowhead marks the onset of transformation . Dashed yellow line marks R8 depth through elongation . See also Figure 3—figure supplement 1b for an illustration of transient excursions from the target layer after stabilization . ( b ) Average reach of the R8 tip into the medulla ( 2 animals ) . Error bars are standard deviation . Dashed magenta line and band mark stabilization . ( c ) Counts of frame-to-frame ( ∆t = 10 min ) tip movements equal to or greater than ±5 µm during extension ( 3 animals ) . Inset is the full distribution of steps , the tails of which are plotted in the parent graph . ( d ) Onset of transformation ( 3 animals ) . ( e ) Transformation proceeds from both ends . Orange arrowheads mark the first frames in which the proximal length of the thin process begins to expand . Yellow asterisks mark the expansion of the tip; see also ( f ) . ( f ) Brp , a marker for presynaptic differentiation , accumulation follows anterograde expansion ( yellow arrows ) during transformation . Panels show confocal images taken at 47–49 hAPF . R8s are labeled with myr::GFP . R8s express V5-tagged Brp using the STaR system ( Chen et al . , 2014 ) . Overlay of the Brp channel with a mask of the GFP channel highlights R8-localized puncta in magenta . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 00810 . 7554/eLife . 20762 . 009Figure 2—source data 1 . Contains numerical data plotted in Figure 2b , c , d . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 00910 . 7554/eLife . 20762 . 010Video 2 . Aligned WT Growth Cones . R8 growth cones are aligned through the time series and extracted from the full image volumes . myr::tdTOM expressing WT growth cones from one brain are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 010 Elongation may represent the final step of active growth toward the target layer . Alternatively , R8s could be increasing their lengths in concert with the growing medulla . To distinguish between these possibilities , we compared the reach of R8 projections to a fiducial marker for an outer medulla layer , the Dm3 neurons ( Figure 1a ) . In the adult , Dm3 processes project across columns to weave a meshwork at the M2/M3 layer boundary , a position that corresponds to 67 ± 5% of the depth of R8 projections ( Figure 3a ) . Between 45 and 60 hAPF , the Dm3-defined layer is gradually displaced from M0 at an average rate of 0 . 9 μm/hr ( Video 3 ) . Since the projections of these neurons extend orthogonally to medulla columns , the measured movement captures the expansion of the neuropil . During elongation , the ratio of the distance between M0 and the Dm3 layer and the reach of individual R8 projections is 61 ± 4% ( Figure 3d and Figure 3—figure supplement 1 ) , suggesting that the R8s are increasing in length in concert with the growing neuropil . This result indicates that the target layer is reached at stabilization . It is possible that R8 axons actively and independently elongate to match the growth of the tissue . However , both elongation and medulla expansion continue through pre-synaptic differentiation ( Chen et al . , 2014 and below ) , arguing that the concerted growth is supported by attachments to neighboring cells and the target layer . We conclude that R8 projections recognize and become attached to their targets before elongation , at the stabilization step , and are stretched in length as the target layer moves away from M0 . 10 . 7554/eLife . 20762 . 011Figure 3 . Analysis of elongation . ( a ) Panel: Confocal micrograph of the outer medulla in the adult brain . R8s ( red , myr::tdTOM ) and Dm3s ( green , myr::GFP ) are shown . White arcs are fits to M0 and to the Dm3 processes . Arrows , yellow and blue , mark the distance from M0 to the Dm3 layer and to the R8 tips . Graph: Ratio of Dm3 and R8 depths in the brain from panel ( n = 407 ) . ( b ) Live imaging of R8 and Dm3 . Panels from live imaging of the adult brain in ( a ) . The view presented , matching the confocal micrograph in ( a ) , was generated post-processing . Panels were individually contrast enhanced to reveal dimmer features . Note that Dm3 processes complete their expansion into layer M2-3 during the window of observation . Despite the incomplete coverage at early time-points , the representation of Dm3 throughout the time-series is sufficient to calculate a surface fit to this fiducial layer marker ( see Materials and methods ) . ( c ) Time series of an R8 and underlying Dm3 process , a fiducial marker for the M2/M3 boundary . Red and green arrows illustrate the measurements plotted in ( d ) . ( d ) Average reach of the R8 tip ( red arrow in ( c ) ) and the Dm3 distance from M0 ( green arrow in ( c ) ) , measured in one brain . Error bars are standard deviation . Inset: Ratio of the Dm3 distance to M0 and the R8 tip reach . Mean ratio between 50–60 hAPF is 0 . 65 ± 0 . 06 . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 01110 . 7554/eLife . 20762 . 012Figure 3—source data 1 . Contains numerical data plotted in Figure 3a , d . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 01210 . 7554/eLife . 20762 . 013Figure 3—figure supplement 1 . Analysis of elongation . ( a ) Data from two additional brains analyzed as in Figure 3d . ( b ) , Intensity saturated time series highlights transient projections from the target layer during elongation . Tip trace is plotted below; blue connectors map data points to source image panels . The target layer is traced in black . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 01310 . 7554/eLife . 20762 . 014Figure 3—figure supplement 1—source data 1 . Contains numerical data plotted in Figure 3—figure supplement 1a , b . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 01410 . 7554/eLife . 20762 . 015Video 3 . Two Channel Imaging of R8 and Dm3 . Live imaging of R8 ( myr::tdTOM ) and Dm3 ( myr::GFP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 015 The first three steps of targeting are defined by the position and dynamics of the R8 tip; the fourth step , transformation , is a morphological change that begins after stabilization and overlaps with elongation ( Figure 2a , d ) . At the onset of transformation , the original growth cone volume at M0 shrinks while the thin process fills out . The progress of transformation is not purely anterograde . In many cases , the apparent flow of material from the growth cone into the proximal length of the process and the expansion at the distal tip are distinguishable ( Figure 2e ) . In the following ~5 hr , the R8 terminal takes on its mature form , which will eventually contain ~50 en passant synapses ( Chen et al . , 2014; Takemura et al . , 2013 ) . To ask how transformation relates to pre-synaptic differentiation , we studied the distribution of Brp in R8 projections using the STaR system ( Chen et al . , 2014 ) , which enables cell-type specific tagging of proteins expressed at endogenous levels . We found that this marker for pre-synaptic active zones begins to populate the length of the R8 projection with the onset of transformation , following the anterograde expansion to eventually reach the expanded tip at the target layer ( Figure 2f ) . Thus , while transformation is linked to pre-synaptic differentiation , the expansion of the R8 tip at the target layer is an earlier and distinct sub-step . One potential challenge to examining the role of Net-Fra in R8 targeting is the discrepancy between franull ( fra3 [Kolodziej et al . , 1996] , referred to as franull in the main text ) and Net phenotypes reported by Salecker and colleagues . In genetically mosaic animals where fra is removed specifically from R8s , ~90% fail to reach the M3 layer ( Timofeev et al . , 2012 ) . In contrast , the failure rate in whole animal Net mutants ( NetA and NetB double mutant , referred to as Net null in the main text ) ( Newquist et al . , 2013 ) is ~50% ( Timofeev et al . , 2012 ) . These data raise the possibility of a second Fra ligand acting in parallel with Net . To clarify the interpretation of the live imaging data in the context of this ambiguity , we re-visited the mutant analysis through quantitative comparison of adult phenotypes . In the adult brain , wild-type R8 projections reach 16 . 9 ± 1 . 7 μm into the medulla ( Figure 4a , top graph ) . The spread in this distribution is partly due to the 3D organization of this tissue; the absolute distance between layers changes along the two principal axes of the neuropil . Wild-type R8 projections , when expressed as a fraction of the distance between M0 and M6 , reveal a more narrowly defined ( 0 . 75 ± 0 . 05 ) target layer ( Figure 4a , bottom graph ) . In addition to reducing variation within one brain , this normalization approach also enables more reliable quantitative comparisons to be made between R8s from different animals and different genetic backgrounds . For franull R8 terminals , the normalized depth is well-described as a mixture of three distributions ( Figure 4b and see Materials and methods ) . All three components fall short of the target layer . Thus , removing Fra from R8s yields a phenotype with nearly full penetrance ( i . e . the number of R8s affected ) and variable expressivity ( i . e . the severity of the R8 phenotype ) . 10 . 7554/eLife . 20762 . 016Figure 4 . Fra and Net are in the same genetic pathway for R8 targeting . ( a ) Panel: Confocal micrograph of outer medulla . R8s ( green , ~70% expressing myr::GFP ) and all R cells ( red , labeled with Mab24B10 ) are shown . Graphs: Absolute ( top ) and normalized ( bottom ) distance between M0 and the R8 tips measured in medulla . ( b ) Panel: Confocal micrograph of outer medulla in a MARCM brain . franull R8s ( green , expressing UtrnCH::GFP ) and all R cells ( red , labeled with Mab24B10 ) are shown . Projection depths of franull R8s were measured using a membrane-targeted marker ( myr::tdTOM , not shown ) . Graph: Normalized franull R8s projection depths . Components ( C1-3 ) for the Gaussian Mixture Model ( GMM ) fit to the distribution are plotted in red; black trace is the sum of the components . Dashed blue line marks the mean of the WT distribution from ( a ) . Table: GMM parameters . Frac . : fractional contribution of each component to the fit . ( c ) Panel: Adult brain in which fra expression was knocked down in R8s using a cell-type specific driver ( i . e . fraRNAi ) . Cell labeling as in ( a ) . Graph: Normalized projection depths for fraRNAi R8s . Table: GMM parameters . ( d ) Panel: Net adult brain . Cell labeling as in ( a ) . Graph: Normalized R8 projection depths in Net animals . Table: GMM parameters . ( e ) Panel: Net adult brain in which fra expression was knocked down in R8s using a cell-type specific driver ( i . e . fra RNAi ) . Cell labeling as in ( a ) . Graph: Normalized R8 projection depths in the Net+fra RNAi background . Table: GMM parameters . ( f ) Cumulative distribution of data in ( d ) and ( e ) . The distributions are not distinguishable by the two-sample Kolmogorov-Smirnov test at significance level α < 0 . 01 . ( g ) Comparison of the normalized distance from M0 to the midline of the Dm4 processes near M3 in WT ( top , blue ) and Net ( bottom , gray ) male adult brains . ( h ) Confocal micrographs of the outer medulla in WT ( left ) and Net ( right ) adult brains . R8s ( green , myr::GFP ) and Dm4s ( red , myr::tdTOM ) are shown . White curves are fits to the Dm4 midline . ( i ) Absolute distance between R8 tips and the Dm4 midline ( dashed red line at 0 ) measured in WT ( blue ) and Net ( gray ) adult brains . Positive values indicate that the R8 tip is past the Dm4 midline . ( j ) Two additional cell types with altered morphologies in the Net background . Dm1 and Dm6 are both multi-columnar amacrine cells with processes at M1 in the WT ( top row ) . Both cell types generate extra arborizations at M4-M5 ( bottom row , yellow arrows ) in Net animals . Dm1s and Dm6s expressing myr::tdTOM ( white ) were visualized with immunohistochemistry . Mab24B10 was used to stain for all photoreceptors ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 01610 . 7554/eLife . 20762 . 017Figure 4—source data 1 . Contains numerical data plotted Figure 4a , b , c , d , e , f , g , iDOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 017 Visual inspection of R8s in Net null animals suggests that a large fraction of R8s project to the M3 layer ( Figure 4d , panel ) . However , the normalized depth for R8s in this genetic background can also be described as a mixture of three distributions , and the deepest reaching component ( Figure 4d , graph and table ) is nearly identical to that found in fra R8s ( Figure 4b , graph and table ) in terms of mean and standard deviation . In other words , the R8 depth distribution in the Net null background is well-described without a wild-type component ( Figure 4a , bottom graph ) . To directly measure the distance between the R8 tips and the M3 layer in Net null animals , we used a Net-insensitive ( Figure 4g ) M3 marker , the Dm4 neuron ( Figure 1a and Figure 4h ) . In wildtype , 95% of R8 tips reach past the Dm4 midline ( Figure 4i ) . In contrast , only 19% of R8s in Net null animals extend this far . The majority of this 19% would belong to a sub-population distinct from wild type ( i . e . tail of a mutant distribution ) ; a much smaller fraction would be drawn from a purely wild-type component . We conclude that the penetrance of the Net mutation for R8 targeting is much higher than originally reported and that the difference between franull and Net R8 targeting defects is one of expressivity . To study the fra-Net genetic interaction , we performed epistasis analysis by reducing Fra levels in R8 cells with RNAi in the Net null background ( Figure 4e ) . The phenotype of the double mutant was indistinguishable from Net null ( Figure 4f ) . This result indicates that Fra and Net , as expected from a receptor-ligand pair , act in the same genetic pathway for R8 targeting . It also rules out the possibility of a second Fra ligand , acting in parallel with Net . However , the epistasis analysis does not rule out possible cell-autonomous contribution of other Net receptors . Indeed , the activity of such receptors in R8s would be consistent with the expressivity difference observed between the Net and franull genetic backgrounds . We assessed this possibility for Dscam1 and UNC-5 with genetic mosaic analysis and did not find any obvious defects in R8 targeting . These results are consistent with the findings of a recent a cell-type specific RNA sequencing study ( Tan et al . , 2015 ) , which showed that , at the onset of targeting , Fra is the only Net receptor expressed in R8 above 1–2 RPKM , the common threshold for noise in these types of studies . Together , the epistasis analysis and the lack of evidence for the R8-specific activity of other Net receptors indicate that the difference in expressivity between the franull and Net phenotypes is due to a non-cell autonomous , pleiotropic effect of the Net mutation . Consistent with this finding , we found two other cell types with altered morphologies in the Net background ( Figure 4j ) . Why is the expressivity of the Net whole animal phenotype on R8 growth cones less severe than the removal of Fra selectively from R8s in mosaic animals ? What is surprising is not that they are different , but rather that the pleiotropy is associated with weaker R8 expressivity . Removing Net from the whole animal alters the development of multiple neuronal classes in the medulla . Some , like R8 , may be affected through the Fra pathway , while loss of Net signaling through the UNC-5 receptor may be important in other cases . We suspect that some aspect of the micro-environment in which R8 growth cones develop becomes more permissive to the progress of transformation in the absence of Net ( see below ) . Regardless of the specific causes of pleitropy , the altered medulla environment in Net animals confounds mechanistic interpretation of R8 growth cone defects . For a more refined approach to studying the role of Net in R8 targeting , we pursued two strategies involving genetic manipulation of the L3 lamina monopolar neuron , the principal source of Net in the target layer ( Pecot et al . , 2014; Timofeev et al . , 2012 ) . First , we ablated L3s through cell-type specific RNAi-mediated knock-down of a neurotrophic receptor . This experiment was originally published by our group where we showed that removal of all L3s led to a loss of Net from the M3 layer ( Pecot et al . , 2014 ) . To compare the development of R8s with and without a home-column L3 in the same brain , we reduced the efficiency of the knock-down to achieve a partial ablation of the L3 array ( ~80% loss with ~20% surviving through eclosion ) . In this setup , all R8s with an L3 in the home column target as wild type . With R8s lacking a home column L3 , we did observe targeting defects , but the phenotype was too variable to discern general patterns of growth cone behavior and draw mechanistic conclusions . We suspect that ablating L3 , a major resident of the M3 layer , is too blunt a perturbation with pleiotropic consequences that go beyond removing a single secreted ligand . As a second approach , we followed the targeting of wild-type R8s in columns with Net null L3s in genetic mosaics ( i . e . MARCM ) using live imaging . We did not observe any overt R8 phenotypes in this genetic background . The principal caveat here is that with MARCM in the lamina , mutant neurons appear as singlets or in small patches in an array of wild-type cells . We could not achieve mutant L3 patches large enough to ensure that any R8 sharing a column with a Net null L3 did not also neighbor at least one wildtype L3 . Indeed , at 40 hAPF , L3 growth cones are very large along the dorso-ventral axis and extend into adjacent columns . Thus , it is likely that Net is contributed from wild-type L3 growth cones in neighboring columns . In summary , the quantitative re-assessment of the mutant phenotypes and the epistasis analysis establishes that , for R8 targeting , removing either Net or Fra are equivalent perturbations . Thus , while it has not been possible to directly study the cell-type specific effect of removing Net on R8 targeting , we sidestepped the complications of Net genetics by focusing our efforts to characterize the role of Net-Fra signaling on comparing the dynamics of wild-type and franull-mutant growth cones in genetically mosaic animals . To study the role of Fra in R8 targeting , we combined live imaging with MARCM , a genetic strategy that relies on mitotic recombination to generate positively marked mutant cells in otherwise wild type tissue ( Lee and Luo , 1999 ) . In this experimental design , we also incorporated a direct labeling scheme , to mark all R8s independent of genotype . This enabled us to compare franull and wild-type R8 growth cones projecting into the same wild type medulla ( Figure 5 and Video 4 ) . Due to variations in the M0-to-M3 distance between different columns in the same wild type animal and between different samples , this internally controlled setup was essential for the detailed comparisons between wild type and mutant growth cones described below . 10 . 7554/eLife . 20762 . 018Figure 5 . franull R8 targeting . ( a ) Wild-type and franull growth cones from the same mosaic brain . ( b ) Steps of franull targeting . Orange arrowhead marks the onset of transformation . ( c ) Data for WT and franull R8s from the same brain presented as in Figure 2b . Note that the apparent difference between WT and franull reach prior to retraction is due to the opposing effects on this population average metric of ( 1 ) transient extensions beyond the target layer in the WT ( Figure 3—figure supplement 1b ) and ( 2 ) the retractions of mutant growth cones during tracking ( see text ) . For other datasets in which the difference in average reach is less pronounced , see Figure 5—figure supplement 1c , d . ( d ) Data for franull R8s presented as in Figure 2c . ( e ) Data for franull R8s presented as in Figure 2d . ( f ) Brp accumulation follows anterograde expansion ( yellow arrows ) during transformation in franull growth cones . Panels show confocal images of franull R8 growth cones in MARCM brains at 45 and 50 hAPF . R8s are labeled with myr::GFP and myr::tdTOM ( MARCM label , not shown ) . R8s express V5-tagged Brp using the STaR system ( 19 ) . Overlay of the Brp channel with a mask of the GFP channel highlights R8-localized puncta in magenta . ( g ) franull dynamics during tracking; growth cone in ( b ) reproduced at full time resolution . Magenta arrowhead marks a transient retraction . See Figure 5—figure supplement 2 for more examples . ( h ) Tip trace of the WT growth cone in ( a ) , plotted at 10 min resolution . The target layer , in red , is calculated as described in Materials and methods . ( i ) Tip trace of the franull growth cone in ( b ) and ( g ) , plotted at 10 min resolution . Magenta arrowhead marks the transient retraction shown in ( g ) . The track line , in red , is calculated as described in Materials and methods . ( j ) Scatter plot of slopes and positions of target layer and tracking trendlines at 47 hAPF , for the WT and franull growth cones in ( c ) . The 47 hAPF position is a surrogate for the y-intercept of the trendlines . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 01810 . 7554/eLife . 20762 . 019Figure 5—source data 1 . Contains numerical data plotted Figure 5c , d , e , h , i , j . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 01910 . 7554/eLife . 20762 . 020Figure 5—figure supplement 1 . franull R8 targeting . ( a , b ) Extension dynamics are not altered in franullR8s . ( a ) shows counts of frame-to-frame ( ∆t = 10 min ) tip movements equal to or greater than ±5 µm for WT R8s from the same brain as the franull R8s in Figure 5d . Insets are the full distributions of steps , the tails of which are plotted in the parent graphs . Bar graphs in ( b ) display rate metrics for individual extension and retraction events for wild-type ( n = 12 ) and franull ( n = 13 ) growth cones . ‘Fast step’ is the largest tip displacement observed between consecutive frames ( ∆t = 10 min ) presented as rate of movement; ‘Event average’ is the ratio of the total displacement to the total time for a single extension or retraction event . Embedded numbers count the events scored; only extension or retraction events with the largest step ≥ 2 . 5 µm ( i . e . ≥ 0 . 25 µm/min ) were considered . Error bars are standard deviation . Wild-type and franull distributions for any of the 4 metrics presented are not distinguishable by the two-sample Kolmogorov-Smirnov test at significance level α = 0 . 05 . ( c , d ) Average reach into the medulla ( see Figure 5c ) and target layer and tracking trendline metrics ( see Figure 5j ) from two additional MARCM experiments . ( e ) Tracking metrics ( 3 animals ) . Track Start and Stop are the times of initial arrival ( light gray ) at and final retraction ( dark gray ) from the track line ( i . e . target layer ) , respectively . Tracking duration is calculated for each growth cone . On-Track residence ( blue ) is the fraction of time each growth cone spends above the track line . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 02010 . 7554/eLife . 20762 . 021Figure 5—figure supplement 1—source data 1 . Contains numerical data plotted in Figure 5—figure supplement 1a , b , c , d , e . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 02110 . 7554/eLife . 20762 . 022Figure 5—figure supplement 2 . franull dynamics during trackingSingle franullgrowth cones presented at full time resolution . Images are shown with a cyan-hot look-up table to increase displayed dynamic range . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 02210 . 7554/eLife . 20762 . 023Video 4 . Two Channel Imaging of WT and fra R8s with MARCM . Dual labeling of all ( myr::tdTOM , red ) and franull ( UtrnCH::GFP , green ) R8s reveals dynamics of wild-type and mutant growth cones in the same MARCM brain . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 023 R8 growth cones reach the target layer without fra in a fashion indistinguishable from wild type . Extension is unaffected in mutant growth cones; a single thin process appears on the medial side and reaches into the medulla as in wild type ( Figure 5a , b , d and Figure 5—figure supplement 1a , b ) . At ~48 hr , the tip of the thin process from these growth cones stalls at ~10 μm from M0 ( Figure 5a , b ) , the approximate position of the target layer recognized by wild-type growth cones . At this point , franull development diverges from wild type in two respects . First , the distal tips of franull processes do not expand ( Figure 5a , b , g and Figure 5—figure supplement 2 ) . That is , while subtle dilations or other distinct structures may appear at this active site , these are invariably transient and are remodeled within 20–30 min to restore the thin morphology of the targeting process . Second , mutant processes do not stably adhere to the target layer . Instead , for the next ~10 hr ( Figure 5—figure supplement 1e ) , as wild-type R8s passively elongate ( Figure 5h ) , franull R8s actively follow a slowly advancing front in the medulla ( Figure 5g , i and Figure 5—figure supplement 2 ) , before they ultimately retract . That is , mutant R8s continue to extend and retract their thin processes , as their reach into the medulla increases over time . We term this behavior tracking . We sought to assess whether franull growth cones track the moving target layer . For each wild type and franull growth cone in mosaic brains , we calculated the trendlines followed by the advancing R8 projections during elongation and tracking ( see Materials and methods ) . Greater than 95% of franull R8s yielded tracking trendlines that are consistent with target layer trendlines of their wild-type counterparts ( Figure 5j and Figure 5—figure supplement 1c , d ) . Thus , franull R8s track the target layer . This indicates that mutant R8 growth cones recognize one or more determinant of target layer specificity . The observation of tracking uncouples target layer recognition and attachment . In R8s , target layer recognition does not require the Net-Fra pathway . The onset of transformation occurs on schedule in franull mutants ( Figure 5e ) ; however , in the absence of tip expansion , the progress of this morphological change is now only anterograde . As in wild type , Brp puncta enter mutant R8 projections following this anterograde expansion ( Figure 5f ) . The progress of transformation is slow and fails to reach the target layer in the vast majority of franull growth cones before the final retraction after tracking ( Video 5 ) . That is , while synaptogenesis is affected in franull mutants , this is a downstream consequence of the earliest observable phenotypes—lack of tip expansion and stable attachment . The penetrance of the loss of target layer adhesion in franull growth cones is complete , consistent with our quantitative analysis of the adult phenotype . Live imaging of R8 targeting in Net null animals revealed similar behavior ( Figure 6 ) . In summary , the Fra-Net pathway is required for the attachment of the R8 terminal to M3 ( Figure 8a ) . The marked expansion at the tip in wild type is consistent with Net-Fra signaling coupling a cytoskeletal response to substrate adhesion . 10 . 7554/eLife . 20762 . 024Video 5 . Aligned franullGrowth ConesR8 growth cones are aligned through the time series and extracted from the full image volumes . myr::tdTOM expressing franullgrowth cones from one brain are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 02410 . 7554/eLife . 20762 . 025Figure 6 . R8 targeting in Net mutants . ( a ) Four growth cones from the same Net null mutant brain . Orange arrowheads mark the extent of transformation at the end of the time series in this data set; this determines the final R8 depth after retraction of the thin process . ( b ) Data for R8s in a Net mutant brain presented as in Figure 2b . The corresponding trace for franull R8s from Figure 5c is reproduced in orange . ( c ) Data presented as in Figure 2c . ( d ) Data for presented as in Figure 2d . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 02510 . 7554/eLife . 20762 . 026Figure 6—source data 1 . Contains numerical data plotted in Figure 6b , c , d . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 026 We sought to further test whether the Net-Fra pathway mediates adhesion . To do this , we assessed the function of another gene in the Net-DCC pathway that has cell-autonomous function . Trim9 has been identified as a component of Net-DCC signaling in neural development in both vertebrates and invertebrates ( Alexander et al . , 2010; Hao et al . , 2010; Morikawa et al . , 2011; Song et al . , 2011; Winkle et al . , 2016 , 2014 ) . Trim9 interacts through its SPRY domain with the Fra C-terminal cytoplasmic tail ( Morikawa et al . , 2011 ) . Trim9 is required in R8 targeting , as assessed in fixed preparations in the adult . The vast majority of Trim9null ( Trim991 [Morikawa et al . , 2011] , referred to as Trim9null in the main text ) R8s analyzed in genetically mosaic animals fall short of the target layer ( Figure 7b ) . The small ( ~1% ) fraction of mutant R8s that do overlap the wild-type depth distribution ( Figure 7a ) may represent a sub-population that targets as wild-type , or these may be the outliers of the mutant distribution . We conclude that , like franull , the Trim9 targeting phenotype exhibits near-complete penetrance . To explore the genetic interaction between Trim9 and Fra , we performed epistasis analysis by reducing Fra levels in R8 cells with RNAi in the Trim9 MARCM setup ( Figure 7c , d ) . The phenotypes of double mutant and Trim9null single mutant R8s were indistinguishable ( Figure 7e ) . Thus , Trim9 and Fra are in the same pathway for R8 targeting . 10 . 7554/eLife . 20762 . 027Figure 7 . Trim9null R8 targeting . ( a ) Panel: Confocal micrograph of outer medulla in a WT MARCM brain . MARCM labeled WT R8s ( green , expressing UtrnCH::GFP ) and all R cells ( red , labeled with Mab24B10 ) are shown . Graph: Normalized R8s projection depths . Dashed pink line marks the mean of the distribution . ( b ) Panel: Confocal micrograph of outer medulla in a Trim9null MARCM brain . Cell labeling as in ( a ) . Graph: Normalized R8s projection depths . Dashed pink line marks the mean of the WT distribution from ( a ) . ( c ) Panel: Adult brain in which Fra expression was knocked down in R8s using a cell-type specific driver ( i . e . fraRNAi ) in the WT MARCM setup . Graph: Normalized R8s projection depths . Dashed pink line marks the mean of the WT distribution from ( a ) . ( d ) Panel: Adult brain in which Fra expression was knocked down in R8s using a cell-type specific driver ( i . e . fraRNAi ) in the Trim9null MARCM setup . Graph: Normalized R8s projection depths . Dashed pink line marks the mean of the WT distribution from ( a ) . ( e ) Cumulative distribution of data in ( b ) and ( d ) . The distributions are not distinguishable by the two-sample Kolmogorov-Smirnov test at significance level α < 0 . 01 . ( f ) Reach of the tip into the medulla for WT and Trim9null R8s , compiled from 2 brains . Error bars are standard deviation for WT . Dashed magenta line and band mark stabilization . Table shows number and percentage of mutant R8s in each class and for tip collapse events . As virtually all R8 terminals in the adult fall short of the target layer , the elongating class of R8 terminals must retract after 65 hr APF . ( g ) Wild-type and Trim9null growth cones from the same mosaic brain . Images are shown with a cyan-hot look-up table to increase displayed dynamic range . Orange arrowheads mark the onset of transformation . Orange barbells mark prominent expanded tips ( closed end ) that collapse ( open end ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 02710 . 7554/eLife . 20762 . 028Figure 7—source data 1 . Contains numerical data plotted Figure 7a , b , c , d , e , f . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 028 We analyzed how Trim9null targeting defects arise using live imaging . Like fra , Trim9null R8s exhibit wild-type extension dynamics , their thin targeting processes reach depths comparable to their wild-type counterparts at the stabilization step ( Figure 7f ) , and the onset of transformation occurs on schedule ( Figure 7g , orange arrowheads ) . By contrast to franull , we observe tip expansion in ~90% of Trim9null R8s ( Figure 7g , orange barbells , closed ends ) . The expanded tips of Trim9null R8s frequently collapse ( 82%; Figure 7g , orange barbells , open ends ) —an event never seen in the wildtype after stabilization . Some of these tips exhibit multiple rounds of expansion and collapse . Tip collapse was also commonly observed in mutant R8s that had apparently completed transformation ( Figure 7g , bottom panel ) . In franull R8s , tip expansion was never observed and this correlated with a lack of stable target layer adhesion . This , together with the observation that the tips of most Trim9null R8s expand but nearly all subsequently retract from the target layer ( Figure 7b ) , suggest that Trim9 acts downstream of Fra to consolidate the attachment of R8 growth cones to the target layer . During the time interval corresponding to elongation in wildtype , Trim9null R8s exhibit one of three types of behavior ( Figure 7f ) . Retracting Trim9null R8s ( 30% ) are distinguished from fra mutants during this period only by their unstable expanded tips ( Figure 7g , second panel from top ) ; this class tracks the target layer up to ~55 hAPF and retracts within the time of observation . Stalled Trim9null R8s ( 30% ) are similar to the retracting class except that they maintain the depth they have achieved by ~55 hAPF ( Figure 7g , third panel from top ) . The last class , elongating Trim9null R8s ( 40% ) , appear to complete transformation and elongate along with their wild type counterparts ( Figure 7g , bottom panel ) . As virtually none of the mutant R8s are found at the appropriate target depth in the adult ( Figure 7b ) and tip collapse was observed in 78% of the terminals during development , this last class of Trim9null R8s must retract from the target layer sometime between 65 hAPF and eclosion ( 100 hAPF ) . We note that , similar to what is seen in the Net null background ( Figure 6a ) , the efficiency of anterograde progression of transformation correlates with the three classes—that is , retracting Trim9null R8s show minimal transformation while in all elongating Trim9null R8s the process appears complete . We do not know the factors that influence transformation efficiency . While the onset of transformation is independent of Fra , Net , and Trim9 , it is possible that the anterograde progression of transformation is linked to events regulated by these genes , such as tip expansion and target layer adhesion . We suspect , given that transformation represents pre-synaptic differentiation , that the extracellular contacts made during this process prevent complete retraction of the mutant R8 processes back to M0 , producing the expressivity spectrum observed in all three genes examined . The analysis of Trim9null supports the conclusion that the Net-Fra pathway regulates adhesion of R8 growth cones to their targets ( Figure 8a ) . 10 . 7554/eLife . 20762 . 029Figure 8 . Axon guidance through Net-DCC-mediated adhesion . ( a ) Net-Fra signaling in the second step of R8 targeting: ( 1 ) Extension of thin process into medulla , Net-Fra independent . ( 2 ) Target layer recognition , Net-Fra independent . ( 3a ) Onset of transformation , proximal expansion ( orange arrowhead ) , Net-Fra independent . ( 3b ) Tip expansion in the target layer , Net-Fra dependent ( cyan highlight ) . ( 4 ) Consolidation of adhesion to the target layer , Net-Fra-Trim9 dependent ( blue highlight ) . ( 5 ) Progress of transformation ( orange arrows ) . The extent of transformation completed before the end of tracking in the mutant backgrounds may underlie the difference in expressivity between the Net , franull , and Trim9null adult phenotypes . ( 6 ) Elongation and maturation . ( b ) Net-DCC mediated guidance works through target adhesion , not chemoattraction in a gradient . First Column: In the classical view of chemoattraction , DCC-laden filopodia sense the gradient of extracellular Net ( pink ) and direct the movement of the growth cone toward the source . Second Column: In the alternate model , the Net source is detected independent of the gradient ( gray ) , through filopodial search and capture ( t1-t3 ) or directed extension of a targeting filopodia guided by a separate mechanism ( not depicted ) . Net-DCC signaling at the ligand peak promotes attachment to the target ( t4 ) ; the growth cone proper then reaches the target with the aid of this initial anchor ( t5-t6 ) . Third Column: The generalized view of axon guidance failure in DCC-Net mutants shows that reaching the target does not require the receptor or the ligand; the final phenotype is due to retraction instead of a loss of attraction . DOI: http://dx . doi . org/10 . 7554/eLife . 20762 . 029
Live imaging coupled with the unique features of R8 targeting and comparative analysis of wild-type and mutant growth cones in mosaic animals were essential to pinpoint the requirement for the Net-Fra ligand receptor pair for target adhesion . Specifically , the identification and analysis of the tracking behavior of mutant R8s would not have been possible in fixed preparations . Standard fixation techniques do not preserve tracking filopodia at their full lengths; the shortcomings of fixation have been highlighted previously for cytonemes ( Kornberg and Roy , 2014 ) in the context of the response of cells to morphogens . In addition , the continuous , extended observation of individual growth cones was critical to recognizing general trends over transient dynamics . For example , even with perfect preservation , only half of the tracking R8 processes in fra mutants would appear to be at the target layer at any given time ( Figure 5—figure supplement 1e ) . The simplest interpretation of such data would support the erroneous conclusion that half the R8 population required Net-Fra for chemoattraction . Thus , live imaging allowed us to discriminate between chemoattraction , target recognition , and retraction after reaching the target . The canonical role of Net-DCC in neuronal development is axon guidance through chemoattraction: a gradient of soluble Net steers DCC-expressing growth cones to their targets . Adhesion is also a recognized output of Net signaling , particularly in non-neuronal systems ( Srinivasan et al . , 2003; Yebra et al . , 2003 ) . Biochemical and cell biological studies using dissociated vertebrate neurons or neural explants suggest that these different functions may share a fundamental mechanistic similarity . Structurally , Net can be described in three domains . The two N-terminal domains interact with cognate receptors , including DCC ( Keino-Masu et al . , 1996 ) . The C-terminal domain of Net is highly charged and mediates non-specific adsorption to cell surfaces , ECM components , and to tissue culture substrates ( Kappler et al . , 2000; Moore et al . , 2012; Yebra et al . , 2003 ) . In vitro , substrate-bound Net promotes adsorption of DCC-expressing cells ( Shekarabi et al . , 2005 ) and the Net-DCC interaction can withstand measurable pulling forces ( Moore et al . , 2009 ) . Net’s ‘stickiness’ can be reduced by masking the charge with heparin or removing the C-terminal domain altogether ( Moore et al . , 2012 ) . These manipulations do not alter the ligand’s interaction with DCC , but significantly reduce the neuronal response to Net in classic in vitro assays of Net-DCC function , including growth cone expansion , axon outgrowth , and turning ( Moore et al . , 2012 ) . For example , in the explant turning assay , a cluster of Net-secreting cells abutting a dissected embryonic spinal cord can attract extending commissural axons over a distance of some 200 μm ( Kennedy et al . , 1994 ) . Without the C-terminal domain , the range over which Net can attract axons is markedly reduced ( Moore et al . , 2012 ) . Consistent with the importance of substrate binding to the function of externally supplied Net , most of the endogenous Net in the developing mouse spinal cord is cell- or ECM-bound ( Kennedy et al . , 2006 ) . Together , these observations support the notion that “growth cones pull directly on the cues that guide them” ( Moore et al . , 2012 ) , thereby unifying the chemoattraction and adhesion roles of Net-DCC in the concept of haptotaxis , or motility through traction . The adhesion function of Net-DCC in vivo is well characterized in the anchor cell ( AC ) , a non-neuronal cell in C . elegans . During development , the AC polarizes toward and invades the basement membrane , initiating the attachment of the uterus to the vulva . UNC-6 – UNC-40 signaling is required for the efficient completion of this process . In a series of studies using live imaging , Sherwood and colleagues demonstrated that UNC-6 localized to the basement membrane stabilizes UNC-40 clustering in the AC , which , in turn , directs polarized F-actin production to the basal surface of this cell ( Hagedorn et al . , 2013; Wang et al . , 2014; Ziel et al . , 2009 ) . Without UNC-6 , UNC-40 can still form clusters and promote actin assembly , but these patches of activity are transient and randomly positioned around the cell periphery ( Wang et al . , 2014 ) . MADD-2 , the C . elegans homolog of Trim9 , is also required in the AC to maintain stable and appropriately polarized F-actin patches; without MADD-2 , basal patches can be transient , mis-localized actin assembly is observed , and , ultimately , basement membrane invasion is compromised ( Wang et al . , 2014 ) . The parallels between the use of Net-DCC-Trim9 in development by R8 and the non-neuronal AC—to effect a spatially constrained adhesion or polarization response at an acute presentation of Net ( i . e . gradient peak for R8 and basement membrane enrichment for the AC ) —raises the possibility that adhesion may be a widely conserved cell biological output of this signaling module . In this context , R8 targeting and midline crossing in Drosophila may function in a fundamentally similar way through an adhesion-based mechanism . Brankatschk and Dickson demonstrated that the growth cones of commissural neurons , like those of R8 , respond to a membrane-tethered form of Net in a fashion indistinguishable from the secreted form of Net ( 2006 ) . If commissural axons can , indeed , reach the midline without this signaling module—that is , if the mutant phenotypes , as in the R8 system , arise due to retraction rather than a lack of attraction , the role of Net-Fra would be better described as enabling the axons to traverse the dense midline neuropil . Alternatively , commissural growth cones may extend towards the midline via haptotaxis , growth via traction ( and hence adhesion ) along a pathway of increasing levels of Net bound to the surface of midline glia or associated ECM . Thus , given our current state of knowledge of midline crossing , it is possible that Net-Fra act through an adhesive mechanism in this system . Evidence consistent with this view comes from the work of Emoto and colleagues , who carried out a detailed analysis of Trim9 function in the C4da neurons of Drosophila ( Morikawa et al . , 2011 ) . The branched axon terminals of these sensory neurons send projections that cross the midline of the ventral nerve cord ( VNC ) , the larval counterpart of the embryonic midline . In Net , fra , and Trim9 mutants , these projections are severely reduced or missing . There are three different sub-classes of C4da neurons and each subclass elaborates a characteristic number of contralateral projections . Neurons with more contralateral projections express higher levels of Trim9 . Furthermore , overexpression of Trim9 produces ectopic projections in a dose-dependent manner . This overexpression phenotype is completely suppressed in a franull background ( Morikawa et al . , 2011 ) . Given what we know of Trim9 function in stabilizing adhesion downstream of Net-Fra in R8s and polarization in the AC , the C4da results are consistent with Trim9 controlling the fine-tuning of midline crossing probability . The Net-Fra initiated midline contacts would then be either stabilized or lost depending on Trim9 levels , resulting in successful or retracted contralateral projections , respectively . These observations in Drosophila as well as in vitro experiments with vertebrate growth cones raise the interesting possibility that Net-DCC-mediated adhesion may be a common mechanism used by growth cones in developing invertebrate and vertebrate brains . Tessier-Lavigne and Goodman ( 1996 ) divided guidance signals into four categories . Short-range signals would act through direct contact between growth cones and cells or growth cones and the ECM to attract or repel growth cones . Conversely , long-range signals , secreted by cells , were proposed to act at a distance in a graded fashion to attract or repel growth cones . In Drosophila , both at the midline and in the visual system , a membrane-tethered form of Net is sufficient to rescue guidance phenotypes ( Brankatschk and Dickson , 2006; Timofeev et al . , 2012 ) . These results have been interpreted as evidence for a local or short-range function for Net-DCC . This is by contrast to the diffusible gradient view supported by the original conception of long-range chemoattraction and the results of turning assays of vertebrate growth cones in vitro . However , both at the midline and in the medulla , tethered Net is present as a gradient that peaks at the targets of guidance , and in the absence of studies testing the relevance of this cell-bound gradient to guidance , the distinction between the short- and long- range functions of Net-DCC remained unclear . Net is a secreted molecule of the laminin superfamily and , in principle , can diffuse to form a gradient . As noted above , however , data from in vitro studies argue in favor of the relevance of immobilized Net over that of a soluble form . Together , these results raise the possibility that Net-DCC guidance relies principally on immobilized Net , thereby eroding the mechanistic distinction between the short- and long-range guidance in a gradient . In the R8 system , the gradient is not required . Our results show that the R8 growth cone can reach and recognize its target without the ligand gradient , or the receptor , in a fashion indistinguishable from wild type , and that Net-DCC act to promote adhesion at the source of Netrin . Despite the presence of Net along the path to M3 , we cannot detect an effect of the ligand-receptor interaction while the tip of the R8 growth cone is moving up the Net gradient ( Figure 5—figure supplement 1b ) . Such spatial specificity in signaling output may be achieved through fine-tuning the response of the system to the ligand concentration , or by inhibiting the activity of the receptor until the target is reached . How R8 uses Net-DCC may be idiosyncratic to this neuron . That is , the evidence against gradient-based chemoattraction in R8 targeting does not rule out a role for this mechanism in other in vivo contexts . If Net does act in a graded fashion in other systems , we suggest , in agreement with the work from the Sheetz lab , that it does so not as a soluble factor , but rather as a substrate-bound molecule that influences growth cone behavior through contact . The distribution of the ligand may be determined by localized expression at a single source followed by diffusion and adsorption . Alternatively , a track of cells may form a gradient by expressing and presenting different levels of Net on their surfaces . Thus at a mechanistic level , we propose that , in vivo , DCC responds to immobilized , not soluble , Net . This initial adhesive interaction may stabilize attachment as in the case of R8 or , alternatively , this may promote traction for growth cone motility along a surface ( i . e . haptotaxis ) . This and other studies of Net-DCC signaling in invertebrates and vertebrates raise the possibility that Net and other secreted signals within the developing CNS may act principally in close proximity to the source cells , either associated with the surface of cells or the ECM , to elicit discrete and localized responses . In the absence of action at a distance , neural circuit assembly would proceed in a stepwise fashion where neuronal processes sample a complex environment through direct contact ( i . e . via ‘touch’ rather than ‘smell’ ) , integrate these signals and transform them into morphological and biochemical specification .
Histology was performed as described previously ( Chen et al . , 2014 ) with minor modifications . After antibody incubations , brains were washed into PBS with 0 . 5% Triton X-100 ( PBT ) . To minimize tissue shrinkage , the brains were moved from PBT to mounting medium ( EverBrite , Biotium ) through a series of mixtures with increasing concentrations of the latter . The following primary antibodies were used: chicken Pab α-GFP ( abcam , ab13970 , RRID:AB_300798 , 1:1 , 000 ) , Mab24B10 ( Van Vactor et al . , 1988 ) ( DSHB , RRID:AB_528161 , 1:20 ) , rabbit Pab α-DsRed ( Clontech , Cat# 632496 , RRID:AB_10013483 , 1:200 ) , mouse α-V5 ( Serotec , Cat# MCA1360 , RRID:AB_322378 1:200 ) . The following secondary antibodies were used: Alexa Fluor 488 goat α-chicken , Alexa Fluor 568 goat α-rabbit , Alexa Fluor 647 goat α-mouse ( ThermoFisher , Cat# A-11039 , RRID:AB_2534096 , Cat# A11036 , RRID:AB_10563566 , Cat# A-21235 , RRID:AB_2535804 , 1:500 ) . Confocal images were acquired with a Zeiss LSM780 system . To carry out 3D measurements in the medulla , imaged volumes are deconstructed into oriented medulla columns bounded by computed surfaces for M0 and M6 . A typical multi-channel confocal stack of the medulla measures 150 x 150 x 180 µm and has voxel dimensions of 0 . 29 x 0 . 29 x 0 . 4 µm . In a pre-processing step , all channels are scaled in the z-dimension to achieve unit voxel aspect ratio . The Mab24B10 channel ( i . e . all R cells ) is used in the deconstruction . R7 axon terminals are the deepest-reaching visible features; a mask of the R7 axon tips is generated by manually cleaning up the image . Local intensity maxima in the original image are selected with this mask and the resulting point cloud is used to define the continuous 3D surface of the M6 layer . The bundles of R cell axons that stretch across the outer medulla surface increases the complexity of the image at M0; a more involved approach is required to define this layer . The area bounding the footprint of the R7 projections in M6 is divided into 50–72 regions , and , for each region , columnar volumes orthogonal to the M6 surface are extracted from the image . Intensity profiles along the column axis of these volumes are used to find local peaks , which are classified according to magnitude and their distance from the M6 surface . These two parameters are used to identify the peaks that reside in the M0 layer; the M0 surface is built from a point cloud derived from the selected peaks . A similar approach is used to compute the Dm4 surface . To define the medulla columns , a sequence of masks laminating the space between the M0 and M6 surfaces are used to generate maximum intensity projections ( MIPs ) of the Mab24B10 image . The cross-sections of R7-R8 projections are identified as local intensity maxima in these MIPs . Across the MIP sequence , the intensity maxima are grouped into individual tracks that define the position and orientation of the medulla columns . This information is used to create single panel MIPs of individual R8 axons from the R8 channel . The tips of R8 projections are marked manually on the MIPs and the input is used to calculate 3D distances to the M0 and M6 surfaces . Up to 500 R8s per medulla were scored with this approach . This analysis was written in Matlab ( Mathworks ) with a critical script sourced form the Mathworks File Exchange repository ( D’Errico , 2005 ) . Fiji ( ImageJ ) ( Schindelin et al . , 2012 ) was used for user-assisted tasks . Each data set in Figure 4b–e was modeled as mixtures of 2–7 Gaussian distributions using a built-in Matlab ( Mathworks ) function . For all cases , the Akaike information criterion , a fitting evaluation metric that weighs the goodness-of-fit against the number of free parameters , decreased monotonically until the minimum was reached at 4 or 5 components . While more complex models are supported by the data , we presented the fits with 3 components , the minimum number required to represent the major sub-populations . When modeled with >3 Gaussians , only the fraRNAi ( Figure 4c ) data supports a wild-type component with a mean at 0 . 75 . For each growth cone tip trace , a family of candidate trendlines was generated using data points within a moving window of 10–12 hr . For WT growth cones , a subset of the data in each window was selected by fitting a lower-bound cubic spline to the data . The complexity of the spline ( i . e . number of cubic functions used ) was increased until 30% of the data points were within 0 . 5 µm of the curve . These spline-proximal points were used to define a line using the Thiel-Sen estimator method . The intercept of the line was adjusted so as to give 90% of all the data in the time window positive residuals . For franull growth cones , an upper-bound spline was fit to the data in each time window and , again , the complexity of the spline was increased until 30% of the data points were within 0 . 5 µm of the curve . The best-fit line to the spline-proximal points was refined using a 30% subset with the smallest standard deviation in their residuals to the line . For both classes of growth cones , the optimum trendline from among the candidates was selected using the product of three weight functions . The first two of these are normal distributions , which rank the slope and intercept of the trendlines based on parameters derived from the average tip trace curves of WT growth cones ( e . g . blue mean and errors in Figure 5c ) . The third is an exponential decay function that ranks the standard deviation of the residuals to each candidate trendline . This analysis was written in Matlab ( Mathworks ) with a critical script sourced form the Mathworks File Exchange repository ( D’Errico , 2009 ) . Flies were reared at 25°C on standard cornmeal/molasses medium . Pupal development was staged relative to white pre-pupa formation ( 0 hAPF ) or head eversion ( 12 hAPF ) . | The brain of the fruit fly contains hundreds of thousands of neurons , while the human brain contains more than 80 billion . Each of these consists of a cell body that bears an array of branches called dendrites , plus a single cable-like axon . During development , the neurons organize themselves into complex networks by forming connections with one another via their axons and dendrites . But it is not clear exactly how the correct connections form in the correct places . As they grow out , axons rely on specialized moving structures at their tips – known as growth cones – to probe their environment in search of attractive and repulsive chemical signals released by other cells . When sensors on the surface of growth cones detect a target signal , they initiate processes that cause the growth cone to expand or collapse . This enables the axons to move towards or away from the signal , as appropriate . In all animals studied , proteins called DCC and Netrin form one of the best-known sensor-signal pairs . Growth cones bearing DCC sensors are thought to detect ‘wafting plumes’ or gradients of Netrin and then grow towards the Netrin source . However , nobody had directly watched neurons respond to Netrin in a living intact animal . Using a type of microscope that can look deep into the developing fly brain , Akin and Zipursky have now followed the movement of growth cones on cells called R8 neurons in fruit fly pupae . Unexpectedly , Akin and Zipursky found that the growth cones of mutant flies that lack Netrin or Frazzled ( the fruit fly version of DCC ) navigate successfully to their intended destinations . Once there , however , the mutant growth cones were unable to attach to their targets . Akin and Zipursky’s work is consistent with other observations in a number of animal and insect systems that suggest that Netrin may not attract growth cones via wafting plumes of signal . Instead , Netrin may form a sticky trail that helps growth cones to gain traction as they crawl towards or stick to their destinations . Further experiments are now needed to test whether other neurons in fruit flies and in different animals use Netrin in this way . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2016 | Frazzled promotes growth cone attachment at the source of a Netrin gradient in the Drosophila visual system |
Acoustic communication is fundamental to social interactions among animals , including humans . In fact , deficits in voice impair the quality of life for a large and diverse population of patients . Understanding the molecular genetic mechanisms of development and function in the vocal apparatus is thus an important challenge with relevance both to the basic biology of animal communication and to biomedicine . However , surprisingly little is known about the developmental biology of the mammalian larynx . Here , we used genetic fate mapping to chart the embryological origins of the tissues in the mouse larynx , and we describe the developmental etiology of laryngeal defects in mice with disruptions in cilia-mediated Hedgehog signaling . In addition , we show that mild laryngeal defects correlate with changes in the acoustic structure of vocalizations . Together , these data provide key new insights into the molecular genetics of form and function in the mammalian vocal apparatus .
Vocal communication is fundamental to social interaction . Indeed , the voice is so crucial to our quality of life that the neurobiology of speech and language has been hotly studied for decades , as has the developmental biology of the ear . These bodies of work stand in surprising contrast to our still rudimentary understanding of the developmental biology of the organs of vocalization , the larynx and vocal folds . This is true despite the fact that most animal vocalizations , including human speech , are critically dependent upon the careful control of airflow though the larynx . In fact , larynx and vocal fold morphology and elasticity are key factors influencing vocalization even in animals with widely divergent mechanisms of sound production ( e . g . audible vocalizations in humans , ultrasound in rodents ) . This deficit in our understanding of laryngeal and vocal fold development is significant , because many people who are capable of normal speech still cannot communicate due to defects in voice ( e . g . problems with pitch , loudness , etc . ) . Some voice defects arise from acute insults , such as insufficient hydration of the vocal folds in laryngitis sicca or vocal fold hemorrhages resulting from blood vessel ruptures ( Aronson and Bless , 2009 ) . Other conditions are hereditary and chronic , such as those arising from mutations in genes encoding the extracellular matrix protein Elastin ( Vaux et al . , 2003; Watts et al . , 2008 ) . All of these conditions impact the voice , thereby impacting patients’ well-being . A wide array of human birth defect syndromes also involve voice defects , and prominent among these are disorders stemming from failure of the Hedgehog ( HH ) signaling pathway , an evolutionarily conserved mechanism for cell-cell communication ( Briscoe and Thérond , 2013 ) . For example , Pallister-Hall Syndrome is caused by mutations in Gli3 , a key transducer of HH signals . These patients have hoarse and/or growling voices , and they frequently exhibit laryngeal clefts and bifid epiglottis ( Hall et al . , 1980; Tyler , 1985 ) . Pallister-Hall Syndrome is known for its variable expressivity , and accordingly , this disorder is also associated with milder laryngeal anomalies ( Ondrey et al . , 2000 ) . Importantly , laryngeal and voice defects are not limited to Gli3 mutation , but have also been associated with mutation in the related factor Gli2 ( França et al . , 2010 ) , in the Shh transducer Kif7 ( Putoux et al . , 2012; Walsh et al . , 2013 ) , and in Shh itself ( Cohen , 2004 ) . Cilia are essential organelles for transduction of HH signals ( Goetz and Anderson , 2010 ) , and as a result , voice defects are also commonly associated with ciliopathies , human diseases that share an etiology of defective cilia structure or function ( Hildebrandt et al . , 2011 ) . For example , a breathy , high-pitched voice is a diagnostic criterion for Bardet-Biedl and Oral-Facial-Digital syndromes , while hoarse voices are diagnostic for Joubert Syndrome ( Beales et al . , 1999; Garstecki et al . , 1972; Hayes et al . , 2008; Maria et al . , 1999; Rimoin and Edgerton , 1967 ) . Laryngeal defects such as laryngeal stenosis and bifid epiglottis are also common features of other ciliopathies ( Carron , 2006; Hayes et al . , 2008; Silengo et al . , 1987; Steichen-Gersdorf et al . , 1994; Stevens and Ledbetter , 2005 ) . Understanding the molecular genetic basis for voice disorders in human birth defect patients is not the only factor motivating a deeper study of laryngeal developmental biology . Indeed , vocal communication is ubiquitous in tetrapod animals , impacting a wide array of behaviors . For example , the Panamanian Tungara frog creates a complex , multi-tonal call that critically influences female mate choice , and this call requires a sexually dimporphic elaboration of the male larynx , the developmental basis of which is entirely unknown ( Griddi-Papp et al . , 2006; Ryan and Drewes , 1990 ) . So too is the morphology of the songbird syrinx central to sound production , yet almost nothing is known of the developmental biology of this functional cognate of the larynx , despite the key role of bird song as a model for the study of acoustic communication . Likewise , the larynx of mice is central to their production of ultrasonic vocalizations throughout life . Despite the widespread use of mice for studies of developmental biology , the molecular genetics of mouse laryngeal development remain only cursorily poorly defined ( e . g . [Böse et al . , 2002; Lungova et al . , 2015] ) . Clearly , a deeper understanding of the molecular genetic basis of laryngeal patterning and morphogenesis will inform our understanding of vertebrate animal behaviors involving acoustic communication . In mammals , the larynx and vocal folds are comprised of an elaborate mixture of cartilages , muscles , nerves , and connective tissue ( Harrison , 1995; Henick , 1993; Lungova et al . , 2015 ) . The flanged circle of the cricoid cartilage , along with the C-shaped thyroid cartilage and intervening paired arytenoid cartilages provide the core of the laryngeal skeleton ( Figure 1 , blue , yellow , purple ) . Anchored to these are the vocal folds , which are in turn comprised of paired cricoarytenoid , thyroarytenoid , cricothyroid and vocalis muscles ( Figure 1 , pink , magenta , grey ) , as well as paired vocal ligaments ( Figure 1 , dark blue ) and associated loose mesenchyme which we designate as the thyroglottal connective tissue ( Figure 1 , green ) . The general laryngeal structure is similar across the mammals ( Harrison , 1995; Kaufman , 1992; Roberts , 1975a; Thomas et al . , 2009 ) , though rodents communicate most commonly in the ultrasonic range , using a mechanism for sound production that is distinct from that generating audible sound ( Mahrt et al . , 2016; Roberts , 1975b ) . Importantly however , diverse aspects of rodent ultrasound production parallel those of audible vocalizations in other mammals , including tight control of laryngeal muscle activity and mechanical properties of the vocal folds ( Riede , 2011 , 2013 ) . 10 . 7554/eLife . 19153 . 003Figure 1 . Anatomy of the mouse larynx . ( A ) Diagram representing ventral view of mouse laryngeal anatomy . Dashed lines indicate sectional plane represented in panels C–F . ( B ) Ventral view of an excised adult larynx stained with alcian blue marking cartilage . ( C–E ) H&E staining of horizontal sections of E18 . 5 mouse larynx . Sectional plane is indicated in A . Diagrams indicate anatomy observed in sections . ( F ) H&E staining of sagittal section of E18 . 5 mouse larynx . Diagram indicates anatomy represented in section . Scale bar indicates 500 μm . Abbreviations: ( AC ) Arytenoid Cartilage , ( CC ) Cricoid Cartilage , ( CT ) Cricothyroid muscle , ( E ) Esophagus , ( G ) Glottis , ( L ) Larynx , ( LCA ) Lateral Cricoarytenoid muscle , ( PCA ) Posterior Cricoarytenoid muscle , ( T ) Tongue , ( TAM ) Thyroarytenoid Muscle , ( TC ) Thryoid Cartilage , ( TgCT ) Thyroglottal connective tissue , ( Tr ) Trachea , ( VL ) Vocal Ligament , ( VM ) Vocalis Muscle , ( VF ) Vocal fold . ( V ) and ( D ) indicate dorso-ventral axes . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 003 Here , we have characterized severe defects in laryngeal and vocal fold development in ciliopathic Fuz mutant mice , as well as similar though less severe defects in Gli3 mutant mice . To understand the developmental trajectory of these defects , we first performed directed genetic fate mapping of the mouse larynx . We defined the embryonic origins for diverse tissues in the larynx , and also show that that laryngeal defects in both Fuz and Gli3 mutants stem from an invasion of excess neural crest . Finally , we show that viable heterozygous Gli3 mutant mice display quantitative changes in the morphology of the vocal apparatus accompanied by significant changes in the acoustic structure of their ultrasonic vocalizations . Together , these findings provide an improved foundation for molecular genetic studies of development in the mammalian vocal apparatus , establish a genetic animal model for understanding human congenital laryngeal and voice defects , and demonstrate that excess neural crest is a common etiology underlying diverse Hedgehog-related craniofacial defects .
Voice and laryngeal defects are common in ciliopathies , including Oral-facial-digital Syndrome Type 6 ( Hayes et al . , 2008 ) . Recently , we showed that mutation of genes encoding the Ciliogenesis and Planar Cell Polarity effector ( CPLANE ) proteins results in OFD phenotypes in mice ( Tabler et al . , 2013; Toriyama et al . , 2016 ) . We therefore examined the larynx of mice lacking the CPLANE component Fuz as a first step towards understanding the developmental basis for ciliopathic larynx and voice defects . We observed severe malformation of the laryngeal cartilages in Fuz mice , as well as severely disorganized and hypoplastic vocal fold musculature ( Figure 2A , B , Figure 2—figure supplement 1B–C ) . No glottic space could be identified in the mutants , and the entire larynx was instead filled with an accumulation of loose connective tissue ( Figure 2B , B’ , Figure 2—figure supplement 1C–C ) . This severe derangement of the larynx in Fuz mutants prevented identification of specific laryngeal cartilages , making interpretation of these sections challenging . However , in frontal sections of control mice , we can identify four distinct cartilage elements ( Figure 2—figure supplement 1A , B , B' ) , including the three laryngeal cartilages and the hyoid cartilage ( Kaufman , 1992 ) , while by contrast , we observe only what appears to be a single severely disordered cartilage element in similar frontal sections of Fuz mutants ( Figure 2—figure supplement 1C , C' ) . 10 . 7554/eLife . 19153 . 004Figure 2 . Laryngeal anatomy is disrupted in Fuz and Gli3 mutants . ( A–C ) H and E staining of horizontal sections of E18 . 5 larynges . ( A’–C’ ) Diagrams of anatomy shown in ( A–C ) . Fuz mutant larynges ( B–B’ ) are significantly altered compared to controls , ( A–A’ ) . Connective tissue in mutants ( light green , ( C’ ) is increased in mutants compared to controls ( A–A’ ) , while cartilage and muscle are irregularly patterned . Gli3-/- larynges are less altered than Fuz-/- embryos compared to controls ( A–A’ ) . Thyroglottal Connective tissue appears increased in Gli3 mutants ( light green , red arrow , ( C’ ) . Scale bars indicate 500 μm . Abbreviations: ( AC ) Arytenoid Cartilage , ( CC ) Cricoid Cartilage , ( CT ) Cricothyroid muscle , ( G ) Glottis , ( L ) Larynx , ( LCA ) Lateral Cricoarytenoid muscle , ( PCA ) Posterior Cricoarytenoid muscle , ( T ) Tongue , ( TAM ) Thyroarytenoid Muscle , ( TC ) Thryoid Cartilage , ( TgCT ) Thyroglottal connective tissue , ( Tr ) Trachea , ( VL ) Vocal Ligament , ( VM ) Vocalis Muscle , ( VF ) Vocal fold . ( V ) and ( D ) indicate dorso-ventral axes . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 00410 . 7554/eLife . 19153 . 005Figure 2—figure supplement 1 . Fgf8 reduction in Fuz mutants partially rescues laryngeal phenotypes . ( A ) Diagram indicating sectional plane of E16 . 5 embryos . ( B–C ) Trichrome staining of frontal E16 . 5 Fuz+/+; Fgf8Lacz/+ ( B ) , Fuz+/+ ( C ) , and Fuz-/-; Fgf8Lacz/+ embryos . ( B’–C’ ) Diagrams illustrating anatomy observed in ( B–C ) . Hyoid bone ( dark blue ) , thyroid ( light blue ) , arytenoid ( purple ) and cricoid cartilages are present in Fuz+/+; Fgf8Lacz/+and Fuz-/-; Fgf8Lacz/+embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 00510 . 7554/eLife . 19153 . 006Figure 2—figure supplement 2 . Wnt1Cre driven deletion of Fuz does not affect laryngeal morphology . ( A–B ) H&E staining of horizontal sections E16 . 5 larynges in FuzFl/+; Wnt1Cre/+ ( A ) and FuzFl/+; Wnt1Cre/+ ( B ) embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 006 Cilia-mediated Hedgehog signals influence the processing of both the Gli2 and Gli3 transcription factors ( Haycraft et al . , 2005 ) , so we reasoned that mutation of either one of those two factors may generate milder , more interpretable laryngeal phenotypes . Gli3 mutant mice provide useful models for Gli-related human birth defects ( Böse et al . , 2002; Hui and Joyner , 1993 ) , so we examined the Gli3xt-J mice . Gli3 homozygous mutant mice developed with overt laryngeal defects , and as predicted , these were far milder than those in Fuz mice ( Figure 2B , C ) . Unlike Fuz mutants , the laryngeal cartilages appeared normal in Gli3 mutants and the glottis was evident . However , Gli3-/- mice consistently developed with an aberrant accumulation of Thyroglottal Connective Tissue ( TgCT ) around the vocal folds and particularly between the ventral limit of the glottis and the thyroid cartilage ( Figure 2C , C’ , arrow ) , which was only 2–4 cells wide in normal mice , but was substantially expanded in Gli3-/- mice . In addition , we observed a decrease in the ventral extension of at least the thyroarytenoid muscles with a concomitant expansion of loose mesenchyme between these muscles and the thyroid cartilage ( Figure 2C , C’ ) . Interestingly , the accumulated mesenchyme in Gli3 mutants appeared histologically similar to that seen in the more severely deranged larynx of Fuz mutants . Together , these data suggest a potential role for cilia-mediated Gli signaling in the patterning of the mammalian larynx . We next sought to understand the developmental trajectory of laryngeal defects in our mouse models , but this goal was hampered by the paucity of fate mapping data for the larynx . Indeed , there have as yet been only tangential reports of the developmental origins of tissues in the larynx , and even these results are not entirely consistent . For example , one study reports that the major laryngeal cartilages are of a neural crest origin ( Matsuoka et al . , 2005 ) , but that mapping is surprising in light of other mouse genetic studies that suggest a mixed lineage ( e . g . Jeong et al . , 2004; Mori-Akiyama et al . , 2003 ) . Moreover , at least some laryngeal cartilages have a mesodermal origin in birds ( Evans and Noden , 2006; Noden , 1986a ) . Recent studies using clonal or lineage analysis in mice suggest a relationship between some laryngeal muscles and the branchiomeric neck muscles ( Gopalakrishnan et al . , 2015; Lescroart et al . , 2015 ) , but information is lacking on the origin of the muscles and ligaments that comprise the vocal folds themselves . We first used genetic fate mapping with Wnt1Cre:R26mT/mGto map the descendants of neural crest cells in the larynx ( Chai et al . , 2000 ) . Histological sections revealed distinct lineages for the three laryngeal cartilages . While the thyroid cartilage was prominently labeled by Wnt1Cre::mGFP , the arytenoid and cricoid cartilages were unlabeled , suggesting they are not crest-derived ( Figure 3B ) . Strikingly , we observed a mixed lineage even within the single thyroid cartilage element; the medial caudal-most portion of the thyroid cartilage was consistently unlabeled by Wnt1Cre::mGFP ( Figure 3D , D’ , H , H’ ) . This result was unexpected , so we confirmed it using an alternative promoter to label neural crest and an alternative reporter allele ( Li et al . , 2000 ) . Pax3Cre:R26Tomato mice also displayed strong label throughout most of the thyroid cartilage , but not in the caudal ventral midline; label was also absent from the cricoid and arytenoid cartilages ( Figure 3F–F’ ) . 10 . 7554/eLife . 19153 . 007Figure 3 . Thyroid cartilage and vocal ligament are mostly neural crest derived . ( A ) Diagram representing anatomy in ( B–B’ ) . ( B–B’ ) Horizontal section of rostral E18 . 5 Wnt1Cre/+; R26mTmGlarynx . Neural crest derivatives are labeled in green while other tissues are labeled with Magenta . ( B–C ) Scale bar indicates 200 μm . ( C ) Diagram representing anatomy in ( D–D’ ) Horizontal section of caudal E18 . 5 Wnt1Cre/+; R26mTmGlarynx . ( E ) Diagram representing anatomy in ( F–F’ ) . ( F–F’ ) Horizontal caudal section of E18 . 5 Pax3Cre/+; R26tomato larynx that is also immunostained for Desmin which marks muscle ( Green and yellow in cells that have also expressed Pax3 ) and nuclei ( Blue ) . ( G ) Diagram representing anatomy in ( H–H’ ) . ( H–H’ ) Sagittal section of E18 . 5 Wnt1Cre/+; R26mTmGlarynx . Thyroglottal Insets are indicated with white dotted box . ( F–F’ ) Scale bars indicate 100 μm . Abbreviations: ( AC ) Arytenoid Cartilage , ( CC ) Cricoid Cartilage , ( CT ) Cricothyroid muscle , ( E ) Esophagus , ( G ) Glottis , ( L ) Larynx , ( LCA ) Lateral Cricoarytenoid muscle , ( PCA ) Posterior Cricoarytenoid muscle , ( TAM ) Thyroarytenoid Muscle , ( TC ) Thryoid Cartilage , ( TgCT ) Thyroglottal connective tissue , ( Tr ) Trachea , ( VL ) Vocal Ligament , ( VM ) Vocalis Muscle , ( VF ) Vocal fold . ( V ) and ( D ) indicate dorso-ventral axes . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 007 We then performed a similar analysis using Mesp1Cre;R26mT/mG and Mesp1Cre;R26Tomato to map mesoderm-derived tissues ( Saga et al . , 1999 ) ( Figure 4 ) . Mesp1Cre clearly labeled the ventral midline of the caudal thyroid cartilage , the region that was unlabeled by Wnt1Cr; R26mT/mG ( Figure 4C , D ) , arguing that this single cartilage arises from a mixture of neural crest and mesoderm . Mesp1Cre;R26mT/mG and Mesp1Cre;R26Tomato lineage analysis also revealed a mesodermal origin for the cricoid cartilage and arytenoid cartilages ( Figure 4A , C ) . 10 . 7554/eLife . 19153 . 008Figure 4 . Vocal fold muscles are from cranial mesodermal origin . ( A ) Horizontal section of rostral E18 . 5 Mesp1Cre; R26Tomatolarynx showing that the arytenoid and cricoid cartilages , and Desmin-positive vocal fold muscles are derived from mesoderm . ( B ) Horizontal section of rostral E18 . 5 Islet1Cre; R26mTmGlarynx indicating that all the vocal fold muscles are of cranial mesoderm origin . ( C ) Horizontal section of E18 . 5 larynx of Mesp1Cre; R26mTmG mouse showing the ventral part of the thyroid cartilage derived from mesoderm . ( D ) Diagram of anatomy represented in ( A–C ) . The mesoderm derivatives are labeled in light green while the specific muscular cranial mesoderm derivatives are labeled in dark green . Scale bars indicate 100 μm . Abbreviations: ( AC ) Arytenoid Cartilage , ( CC ) Cricoid Cartilage , ( CT ) Cricothyroid muscle , ( LCA ) Lateral Cricoarytenoid muscle , ( PCA ) Posterior Cricoarytenoid muscle , ( TAM ) Thyroarytenoid Muscle , ( TC ) Thryoid Cartilage , ( Tr ) Trachea , , ( VM ) Vocalis Muscle , ( VF ) Vocal fold . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 008 In the vocal folds themselves , Wnt1Cre drove GFP expression in the vocal ligaments , which are key elements for vocal fold apposition during sound production ( Figure 3A , B , B’ ) . Not surprisingly , the muscles of the vocal fold ( vocalis , thyroarytenoid ) were not labeled by Wnt1Cre::mGFP , but were robustly labeled by Mesp1Cre lineage ( Figure 3A , B; Figure 4A , D ) . We also observed threads of Wnt1Cre labeled , crest-derived cells interspersed within the vocal fold muscles ( Figure 3A , B , B’ ) . Some of these cells were identified as neurons by acetylated tubulin immunostaining ( not shown ) , consistent with reports of vocal fold paralysis in neurocristopathies such as CHARGE syndrome ( Jongmans et al . , 2006; Siebert et al . , 1985 ) . Other crest-derived cells in the vocal fold likely represent the fascia separating the vocal muscles . Finally , Wnt1Cre;R26mT/mG labeled the TgCT , the thin layer of connective tissue separating the ventral aspect of the glottis from the thyroid cartilage ( Figure 3A , B ) . Finally , because cranial and axial muscles develop via distinct genetic programs and originate from different mesodermal populations ( Sambasivan et al . , 2011 ) , we sought to determine which of the mesoderm lineages contributes to the muscles of the vocal folds . In cranial mesoderm , Isl1-positive myogenic progenitors contribute to the formation of head muscles ( Harel et al . , 2009; Nathan et al . , 2008 ) , while Pax3-positive cells in the somitic mesoderm give rise to trunk and limb musculature . Analysis of the Pax3;R26Tomatolineage suggested that the muscles of the vocal folds , marked by Desmin immunostaining , were not Pax3-derived and , thus , not of somitic origin ( Figure 3F ) . In contrast , Isl1Cre;R26mT/mG mice showed that all muscles of the vocal folds labeled by the Desmin immunostaining ( Figure 4A ) were derived from the Islet1; R26mT/mG lineage ( Figure 4B ) , demonstrating their cranial mesoderm origin . Together , these data complement previous lineage analyses of the larynx in other species such as birds ( Evans and Noden , 2006; Noden , 1986a ) and provide the first comprehensive description of the developmental origins of tissues in mammalian larynx . Our fate mapping of the normal mouse larynx provided us with a platform from which to explore the developmental basis for laryngeal defects in our mutant mice . To this end , we performed Wnt1Cre lineage labeling on Fuz-/- and Gli3-/- mice , focusing on cell lineages during initial morphogenesis of the larynx . Between E11 . 5 and E14 . 5 , laryngeal morphogenesis proceeds in a surprisingly convoluted manner , with the previously patent lumen of the developing trachea becoming occluded by the formation of a structure known as the epithelial lamina . This epithelial lamina and tissues surrounding it then remodel into the vocal folds , and a new lumen forms that will ultimately constitute the glottis ( Henick , 1993; Lungova et al . , 2015; Sañudo and Domenech-Mateu , 1990 ) . At E14 . 25 , when the re-canalized glottis is already apparent in control mice , we found that condensing Wnt1Cre labeled neural crest cells were present at the site of the future thyroid cartilage and also in dorsally projecting streams presaging the neural crest-derived tissues in the vocal folds , such as ligaments , fascia and neurons , while other structures such as precursors of arytenoid and cricoid cartilage and future vocal fold muscles were unlabeled ( Figure 5A ) . 10 . 7554/eLife . 19153 . 009Figure 5 . Neural crest is expanded in Fuz and Gli3 mutant larynges . ( A–C ) Horizontal section of E14 . 25 larynges . ( A ) Wild Type Wnt1Cre::mGFP labeled larynx ( B ) Fuz-/-; Wnt1Cre/+; R26mTmG larynx . ( C ) Gli3-/-;Wnt1Cre/+; R26mTmG . Neural crest is labeled in green and other tissues in magenta . ( A’–C’ ) Diagrams representing anatomy found in ( A–C ) . Abbreviations: ( AC ) Arytenoid Cartilage , ( CC ) Cricoid Cartilage , ( CT ) Cricothyroid , ( E ) Esophagus , ( G ) Glottis , ( L ) Larynx , ( LCA ) Lateral Cricoarytenoid , ( PCA ) Posterior Cricoarytenoid muscle , ( TAM ) Thyroarytenoid Muscle , ( TC ) Thryoid Cartilage , ( T ) Tongue , ( Tr ) Trachea , ( VL ) Vocal Ligament , ( VM ) Vocalis Muscle , ( VF ) Vocal fold . ( V ) and ( D ) indicate dorsal ventral axes . Scale bars indicates 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 009 At these stages , the entire region of the forming larynx in Fuz-/- mice was filled with Wnt1Cre; R26mT/mG positive , neural crest-derived cells; only scattered , individual unlabeled cells were observed ( Figure 5B ) , consistent with the very severe phenotype observed at later stages by H&E staining ( Figure 2B ) . Gli3-/- mice displayed less severe phenotypes , with a substantial reduction of the glottic space accompanied by a striking excess of Wnt1Cre; R26mT/mG positive cells ventral to the glottis , in the area of the thyroid cartilage and TgCT ( Figure 5C ) . Fuz mutant mice rarely survive to E17 . 5 , but at this stage in Gli3+/- and Gli3-/- mice , Wnt1Cre lineage mapping revealed a continued excess of neural crest-derived cells that were concentrated around the periphery of the glottis compared to controls ( Figure 6 ) . The accumulation of Wnt1Cre; R26mT/mG positive cells was especially pronounced in the TgCT separating the ventral edge of the glottis from the thyroid cartilage in Gli3-/- embryos ( Figure 6A , C ) . An excess of neural crest-derived cells was also observed to disrupt the normal close association of the vocal muscles with the thyroid cartilage ( Figure 6C , C’ ) . These excess neural crest cells were found precisely in the position occupied by the aberrant mesenchyme observed by H and E staining in Gli3-/- mutants ( Figure 2C; Figure 6D’ , F’ ) , indicating that the cells are neural crest-derived . 10 . 7554/eLife . 19153 . 010Figure 6 . Expanded Thyroglottal connective tissue in Gli3 mutants is neural crest derived . ( A–C ) Horizontal sections of the E18 . 5 ventral larynx in Gli3+/+;Wnt1Cre/+; R26mTmG ( A ) Gli3+/-;Wnt1Cre/+; R26mTmG ( B ) Gli3-/-;Wnt1Cre/+; R26mTmG ( C ) embryos . ( A’–C’ ) Diagrams representing anatomy observed in ( A–C ) . Black dotted line indicates sectional plane for ( D–F’ ) . ( D–F ) H&E staining of midline sagittal sections of E18 . 5 larynges in Gli3+/+ ( D ) , Gli3+/- ( E ) ; Gli3-/- ( F ) embryos . ( D’–E’ ) Magnified view of Thyroglottal Connective tissue ( green arrows ) of the vocal pouch from sections in ( D–F ) . Abbreviations: ( E ) Esophagus , ( G ) Glottis , ( VL ) Vocal Ligament , ( VM ) Vocalis Muscle , ( TAM ) Thyroarytenoid Muscle . Scale bars indicate 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 010 These findings suggest that expanded neural crest disrupts laryngeal morphogenesis , which is of interested because we previously showed that an expansion of neural crest underlies palate defects in Fuz mutant mice and skull defects in both Fuz and Gli3 mutant mice ( Tabler et al . , 2013 , 2016 ) . In those instances , the defects can be rescued by genetic reduction of Fgf8 gene dosage ( Tabler et al . , 2013 , 2016 ) . To ask if a similar mechanism acts in the larynx , we reduced the genetic dosage of Fgf8 in Fuz mutants using the Fgf8LacZ knockin allele ( Ilagan et al . , 2006 ) . Analysis of frontal sections revealed a partial rescue of Fuz mutant phenotype when Fgf8 gene dosage is reduced; while the glottis remained absent in Fuz-/-Fgf8+/LacZ mice , overall anatomy was improved , as cricoid and arytenoid cartilage elements could be identified ( Figure 2—figure supplement 1D , D' ) . We also previously found that the high arched palate phenotype of Fuz mutant mice results from effects prior to neural crest specification , because mice with specific deletion of Fuz using a conditional allele driven in neural crest by Wnt1cre do not display high arched palate ( Tabler et al . , 2013 ) . Likewise , we find here that laryngeal morphogenesis is largely normal in Fuzflox/-;WntCre/+ mice ( Figure 2—figure supplement 2 ) . Together , …… The morphology of the larynx and vocal folds is complex , so in order to assess even subtle phenotypes in our mutant mice , we adopted a strategy of laryngeal morphometrics previously applied to human larynges ( Eckel and Sittel , 1995 ) . First , we quantified the morphology of the vocal folds themselves by measuring the cross sectional area occupied by the larynx , vocal muscles , and the Wnt1Cre labeled vocal ligaments ( Figure 7A , B , C ) . Consistent with the observed excess neural crest discussed above , we detected a significant increase in the area occupied by the vocal ligament ( Figure 7C ) . We observed no corresponding increase in the vocal muscle area , resulting in a significant change in the ratio of the area occupied by vocal ligament to that occupied by vocal muscles ( Figure 7D ) . Strikingly , these phenotypes were dose dependent , with heterozygotes being significantly different from both wild-type and homozygotes ( Figure 7C , D ) . 10 . 7554/eLife . 19153 . 011Figure 7 . Gli3 mutant laryngeal morphology is significantly altered . ( A ) Diagram representing laryngeal measurement presented in ( B–F ) . ( B ) Quantification of total laryngeal cross sectional area excluding extrinsic muscles in E18 . 5 Gli3+/+ ( n = 5 ) , Gli3+/- ( n = 5 ) , Gli3-/- ( n = 5 ) embryos ( orange , ( A ) . ( C ) Quantification of Vocal ligament area E18 . 5 Gli3+/+ , Gli3+/- , Gli3-/- embryos ( blue , ( A ) . ( D ) Quantification of vocal fold muscle area in E18 . 5 Gli3+/+ , Gli3+/- , Gli3-/- embryos . ( E ) Quantification of Thyroglottal connective tissue is in E18 . 5 Gli3+/+ , Gli3+/- , Gli3-/- embryos . ( F ) Quantification of Glottic space in E18 . 5 Gli3+/+ , Gli3+/- , Gli3-/- embryos . P values , * = 0 . 05 , ** = 0 . 01 , *** >0 . 001 , **** >0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 011 Next , we quantified the expansion of the TgCT by measuring the maximum distance between the dorsal edge of the thyroid cartilage and the ventral epithelial lining of the glottis ( Figure 7A , E ) . Again , we observed a dose-dependent increase in this metric from wild-type to Gli3xt-J heterozygote to Gli3xt-J homozygotes ( Figure 7E ) . The increase in connective tissue was observed along the length of the AP axis of the larynx in mutants , as evident in H&E stained sagittal sections of the larynx ( Figure 6D’–F’ ) . Finally , we measured the area of the glottic opening , which was significantly reduced in homozygous animals compared to controls , but was not changed in heterozygous animals ( Figure 7F ) . Together , these data demonstrate that heterozygous Gli3xt-j mutant mice display a milder version of the same laryngeal phenotype observed in the homozygotes . Ultimately , vocalizations are the functional output from the larynx and vocal folds , which manifests as audible speech in humans and as audible and ultrasonic cries in mice . Throughout life , mice use a variety of ultrasonic vocalizations ( USV ) , from pup isolation calls to adult courtship displays ( Holy and Guo , 2005; Neunuebel et al . , 2015; Noirot , 1966; Sewell , 1970; Zippelius and Schleidt , 1956 ) . Because Gli3xt/+ heterozygous mice are viable and display mild defects in laryngeal morphology ( above ) , we examined recordings of pup isolation calls for evidence of altered vocalization . We analyzed over 9000 vocalizations from 5 wild type ( 4718 ) and 6 heterozygous ( 4295 ) mouse pups ( see Figure 8A and B for example spectrograms ) , finding no significant differences in vocalization duration ( F1 , 9 = 3 . 88 , p=0 . 08 ) and mean frequency ( F1 , 9 = 0 . 12 , p=0 . 73 ) , but a significant difference in bandwidth ( F1 , 9 = 12 . 22 , p=0 . 007 ) ( Figure 8C–E ) . 10 . 7554/eLife . 19153 . 012Figure 8 . WT and HT vocalizations do not differ on simple acoustic measures . ( A ) Examples of Gli3+/- vocalizations with ( top panel ) and without ( bottom panel ) frequency steps . ( B ) Examples of Gli3+/- vocalizations with ( top panel ) and without ( bottom panel ) frequency steps . ( C ) Duration of Gli3+/- ( blue ) and Gli3+/- ( red ) pup vocalizations . Average values for each individual ( left panel ) and summary histogram of all vocalizations ( right panel ) . ( D ) Bandwidth of Gli3+/+ ( blue ) and Gli3+/- ( red ) pup vocalizations . Average values for each individual ( left panel ) and summary histogram of all vocalizations ( right panel ) . ( E ) Mean frequency of Gli3+/- ( blue ) and Gli3+/- ( red ) pup vocalizations . Average values for each individual ( left panel ) and summary histogram of all vocalizations ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 012 Mouse USVs are known to be highly variable ( Heckman et al . , 2016 ) , and it was unclear a priori what features of vocalizations might be modified , therefore in addition to measures of basic acoustic properties , we used an analysis method that takes into account the entire structure of the vocal repertoire to ask if subtler differences in vocal phenotype might be present in our mutant mice . From our 9000+ calls , we constructed a map of the vocal repertoire space in which vocalizations with similar frequency contours occupy adjacent regions in the map ( Figure 9A–C; see Materials and methods for details ) . In this map , simple calls lacking abrupt and discontinuous changes in frequency ( frequency steps ) cluster in the central body of the map , while more complex vocalizations with obvious frequency steps are distributed in ‘islands’ or ‘peninsulas’ surrounding the central body ( Figure 9C ) . 10 . 7554/eLife . 19153 . 013Figure 9 . Map of pup vocal repertoire reveals differences in acoustic structure . ( A ) Position of each vocalization within the vocal repertoire map ( Individual vocalizations ) . To generate the map , we defined the difference between two vocalizations to be the dynamically time-warped ( Sakoe and Chiba , 1978 ) mean squared error between them . Low-dimensional structure is then extracted from these distances using t-Distributed Stochastic Neighbor Embedding ( t-SNE ) [van der Maaten and Hinton ( 2008 ) and Berman et al . ( 2014 ) , resulting in the points seen in ( A ) . This embedding results in a clustered structure , This two dimensional non-linear embedding preserves local neighbor relationships in the original high dimensional space . Because this embedding could be equivalently presented at any angle , the vertical and horizontal axes here are arbitrarily chosen and do not represent , for example , the leading directions of variation within the data set . ( B ) Estimated density of the vocalizations within the map ( total vocalization density map ) . ( C ) Overlapped frequency contours for regions across the vocal repertoire map , showing the distribution of syllable types ( Frequency Contour Distribution ) . ( D ) Gli3+/+ vocalizations in the map . ( E ) Gli3+/- vocalizations in the map . ( F ) Difference between the maps with significance regions outlined . ( F1 , 9 = 0 . 04 , p=0 . 85 ) were observed , and only non-significant changes we observed in frequency bandwidth ( F1 , 9 = 1 . 26 , p=0 . 29 ) and average frequency ( F1 , 9 = 0 . 28 , p=0 . 61 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19153 . 013 This analysis revealed that the vocal repertoires of the two genotypes differed , and this difference was significant at multiple map locations ( Figure 9D–F ) . While heterozygous Gli3 mutant pups were capable of producing the full range of vocalizations made by control mice , they were significantly less likely to produce vocalizations with abrupt frequency steps ( Figure 9E , F , warm colors ) , which is consistent with the observed reduction in overall vocal bandwidth in the mutant mice ( Figure 8D ) . In our map , vocalizations in the same map area are acoustically similar , and so differences in the map between genotypes represent differences in acoustic structure between the genotypes; however , relative distance in the map beyond local relationships cannot be interpreted . For example , the vocalizations in the map area with n = 549 vocalizations differs from the vocalizations in the map area with n = 763 only by the presence or absence of a small initial high frequency component ( present in n = 549 , absent in n = 763 ) . Whether this acoustic distinction is behaviorally relevant is as yet unknown . To test more directly for differences in the number of step and non-step vocalizations between WT and Gli3xt/+ each vocalization was labeled as a step or non-step based on their position in the vocal repertoire map ( see Figure 9C , vocalizations in green regions were labeled step , vocalizations in the pink region were labeled non-step ) . We found that 69% of control vocalizations had steps ( 3257/4718 ) versus only 48% of mutant vocalizations ( 2073/4295 ) . Conversely , mutant pups produced a far higher proportion of calls without steps ( Figure 9E , warm colors ) . This difference in the proportion of step vocalizations was highly significant ( chi2 = 400 . 42 , df = 1 , p<0 . 0001 ) . In sum , Gli3xt/+ heterozygous mice , display both morphological defects in the larynx and changes in their patterns of vocalization acoustics . While the acoustic structure of vocalizations is governed not only by the larynx , but also by the structure of the palate and pharynx , as well as neural inputs , our data nonetheless suggest that the mouse can provide a model for studying the links between laryngeal and voice defects in cilia and Gli-related craniofacial syndromes .
Fate mapping is a critical prerequisite for understanding the etiology of defects in development , so our understanding of laryngeal development has been hindered by a lack of directed fate mapping of this organ . In fact , our knowledge of lineage relationships in the mouse larynx comes only from tangential findings in studies focused on other topics . The fate mapping data presented here therefore provides substantial insight and serves as a useful complement to the thorough fate maps of pharyngeal regions in birds and amphibians . We consider several notable findings: First , our data demonstrate a mixed origin for laryngeal cartilages . Indeed , we find a mixed lineage even within a single cartilage element , with the thyroid cartilage being predominantly , but not completely , derived from Wnt1-cre-labeled neural crest descendants ( Figure 3 ) . Specifically , the caudal ventral midline of this cartilage was of mesodermal origin , labeled by Mesp1-cre ( Figure 4 ) . This result adds the thyroid cartilage to the roster of individual skeletal elements assembled by fusion of crest and mesoderm-derived mesenchymal precursors ( e . g . Le Lièvre , 1978; Noden , 1988 ) . In addition , we found no evidence for a neural crest contribution to the arytenoid or cricoid cartilages , and instead our Mesp1-cre lineage data suggest a mesodermal origin for these elements ( Figures 3 and 4 ) . These findings contradict a previous report suggesting a neural crest origin for all three laryngeal cartilages ( thyroid , cricoid and arythenoid ) with the anterior mesoderm boundary at the tracheal level ( Matsuoka et al . , 2005 ) . However , several independent lines of evidence support our conclusion of distinct lineages for the laryngeal cartilages . First , we found no evidence for a neural crest contribution to these cartilages using either Wnt1 or Pax3 promoters for lineage labeling . Second , conditional deletion of Smoothened or Sox9 using Wnt1-cre results in specific loss of the thyroid cartilage but leaves the cricoid and arytenoids intact ( Jeong et al . , 2004; Mori-Akiyama et al . , 2003 ) . A third line of evidence comes from whole animal loss-of-function for R-spondin2 , which functions as a modulator of Wnt signaling throughout development ( de Lau et al . , 2014 ) . In both R-spondin2 mutants and Rspondin2/Lrp6 double mutants , both cricoid and arytenoid cartilages are absent , while the thyroid cartilage remains unaffected ( Bell et al . , 2008; Yamada et al . , 2009 ) . Because neural crest and mesoderm-derived craniofacial structures are known to respond differently to Wnt signaling ( e . g . Barrell et al . , 2012 , Li et al . , 2013 , Quarto et al . , 2009 ) , these data suggest distinct embryological origins for these cartilages . Fourth , our fate mapping data are consistent with avian fate maps in which the arytenoid and cricoid cartilages have a mesodermal origin , demonstrated both by transplantation and clonal analysis after retroviral labeling ( Evans and Noden , 2006; Noden , 1986a ) . Finally , the amphibian pharyngeal skeleton , while considered to retain a more ancestral form , is nonetheless derived from a combination of neural crest and mesoderm ( Sefton et al . , 2015 ) . A second interesting finding concerns the cranial mesodermal origin for the vocal fold muscles ( Figure 4 ) . This finding is in contrast to what has been previously described; lineage analysis suggested a somitic origin of laryngeal muscles in both birds ( Couly et al . , 1992; Huang et al . , 1997; Noden , 1983 , 1986b ) and amphibians ( Piekarski and Olsson , 2007 ) . Our combined Pax3Cre and Isl1Cre lineage data indicate that the mammalian vocal fold muscles are not of somitic origin but derived from cranial mesoderm , consistent with recent studies performed in mice ( Gopalakrishnan et al . , 2015; Lescroart et al . , 2015 ) . Finally , the data argue for a neural crest origin for diverse connective tissues in the larynx , including the vocal ligaments and thyroglottal connective tissue ( Figure 3 ) . These findings are significant because the viscoelastic properties of such connective tissues play an important role in sound production in mammals ( see below ) . Thus , our findings extend previous work highlighting the intricate interrelationship between migratory neural crest- and mesoderm-derived muscles during craniofacial morphogenesis ( Noden and Trainor , 2005 ) . Indeed , interactions between cranial mesoderm and cranial neural crest cells are essential for the normal patterning of the complex musculature of the head ( Grenier et al . , 2009; Heude et al . , 2010; Rinon et al . , 2007 ) , perhaps explaining the lack of differentiated muscle in the crest-infused larynx of Fuz mutant mice ( Figure 2 ) . As such , our new data from the mammalian larynx complement existing work in other regions of the vertebrate head and highlight the key role of neural crest in the evolution of craniofacial morphology in general and the vocal apparatus specifically . Human ciliopathies commonly involve craniofacial defects as well as laryngeal and voice defects , including breathy voices in Bardet-Biedl Syndrome and a hoarse voice in Joubert and Oral-Facial-Digital Syndromes . Moreover , laryngeal stenosis or narrowing is observed in ciliopathies ( Hayes et al . , 2008; Silengo et al . , 1987 ) , consistent with glottic narrowing in Gli3 mutants . Interestingly , Barnes Syndrome is a clinical entity that very closely overlaps the spectrum of defects in the known ciliopathy Jeune syndrome , but with the addition of severe laryngeal defects ( Barnes et al . , 1969; Burn et al . , 1986 ) . The genetic basis for Barnes syndrome is unknown , but it is possible that mutations in Fuz or its interacting CPLANE proteins may be involved ( Toriyama et al . , 2016 ) . In all cases , the embryological basis for human laryngeal defects remains only very poorly understood , but data here and elsewhere argue that excess neural crest may be a central causative agent . For example , the severely deranged Fuz larynx was found to be filled with Wnt1-cre labeled crest-derived mesenchymal cells ( Figure 5 ) , which we interpret as a more severe version of the defect observed in Gli3 mutants . Likewise , the high arched palate that characterizes diverse ciliopathies is also present in Fuz mutant mice , where it is accompanied by an excess of the neural crest ( Tabler et al . , 2013 ) . Moreover , we have also recently described a novel skull defect in Fuz mutant mice in which mesoderm derived parietal bones of the skull fail to form at the expense of expanded neural crest-derived frontal bones ( Tabler et al . , 2016 ) . Importantly , we also find that a milder version of that phenotype is present in gli3 mutant mice ( Tabler et al . , 2016 ) . We conclude then that laryngeal , palatal , and skull defects arising from defective cilia-mediated Gli signaling share a common etiology rooted in the excessive neural crest . Our work here focuses on a tractable model organism with well-developed genetic tools to explore the developmental biology of the mammalian larynx , revealing a key role for neural crest . An important implication of the work , however , is that similar studies in non-model animals could substantially advance our understanding of animal communication . For example , the Panamanian tungara frog is a deeply studied model for evolution by sexual selection ( Ryan , 1985 ) . The complex mating call of the tungara frog is generated by a remarkable , sexually dimorphic elaboration of larynx called the fibrous mass ( Griddi-Papp et al . , 2006 ) . Strikingly , the embryonic origins of the fibrous mass and the molecular genetic systems underlying its development are entirely unknown . However , understanding its morphogenesis is important , as the final size and shape of the fibrous mass differs between related species in the Physaleamus genus , as do the calls produced by these species ( Ryan and Drewes , 1990 ) . Reptiles provide another interesting context for future study . While alligators have a vocal folds relatively similar to that of mammals ( Riede et al . , 2015 ) , snakes and tortoises have highly derived larynges , in which novel vibrating structures take the place of vocal folds . In bull snakes , defensive hissing sounds are generated by a flexible horizontal shelf in the larynx ( Young et al . , 1995 ) , while in tortoises , sound appears to be generated by elastic bands on the lateral walls of the larynx ( Sacchi et al . , 2004 ) . The embryonic tissue origins of these structures , as well as the molecular controls that guide their development , will be of interest . In all three cases above , these laryngeal specializations are not muscular , but rather resemble connective tissue . Our finding of a neural crest origin for connective tissue in the mouse larynx suggests that novel vocalization structures in other animals may be crest derived . In light of the importance of neural crest in the diversification of vertebrate craniofacial structures ( Frisdal and Trainor , 2014; Le Douarin and Dupin , 2012 ) , we propose that a broader study of laryngeal developmental biology will shed light on the evolutionary diversification of vertebrate vocalization mechanisms . The ultrasonic calls of rodents have emerged as a useful model for studies of mammalian vocalization ( Arriaga et al . , 2012; Fischer and Hammerschmidt , 2011 ) ( Heckman et al . , 2016; Portfors and Perkel , 2014 ) . Unlike the audible vocalizations generated by vibrations due to the pressure differential across the apposed vocal folds , rodent USVs are generated by a planar impinging air jet ( Mahrt et al . , 2016 ) . Nonetheless , USVs are generated by the larynx and vocal fold adduction is an important factor both for sound production generally and for frequency modulation ( Johnson et al . , 2010; Riede , 2013 ) . Indeed , direct imaging during USV production revealed a tight apposition of the vocal folds but an absence of vibrations normally observed during audible vocalization ( Sanders et al . , 2001 ) . Moreover , many physiological parameters of mouse USV production parallel those of audible vocalization in other mammals ( Riede , 2011 , 2013 ) . Because so little is known about the etiology of human laryngeal birth defects and their relationship to voice dysfunction , we suggest that studies in mouse models will be informative . We focused on pup isolation vocalizations , which are acoustically distinct from -and simpler than- adult ultrasonic vocalizations ( Liu et al . , 2003 ) . Pup calls are also processed preferentially in mothers ( Elyada and Mizrahi , 2015; Liu and Schreiner , 2007 ) and elicit maternal approach , retrieval and care ( Sales and Pye , 1974 ) . We found that Gli3xt/+ pups produce vocalizations with durations and average frequencies that are not significantly different from those of their control littermates , however the bandwidth of these vocalizations is significantly different . Moreover , our more fine-grained analysis of vocalization shapes revealed that control and mutant mice differed in the proportion of specific vocalization types produced . In particular , Gli3xt/+ mutation decreased the propensity of mice to make step vocalizations with abrupt frequency discontinuities , also known as ‘punctuated’ ( Panksepp et al . , 2007 ) or ‘jump’ ( Hanson and Hurley , 2012 ) syllables . The mechanisms by which such step syllables are generated remain unclear , but we consider two possible explanations for this phenotype . First , vocalization requires exquisite neural control ( Arriaga et al . , 2012; Van Daele and Cassell , 2009 ) , and HH signaling is known to control neural patterning ( Briscoe and Thérond , 2013 ) . It may be , then , that alterations in neural pattern in these mice result in imperfect neural control of the larynx . However , our data on the overall acoustic structure of cells in the mutant mice argue against this explanation . For example , call duration in rats , another rodent with similar ultrasonic vocalizations ( Sales and Pye , 1974 ) , is precisely correlated with EMG activity of laryngeal muscles , and the activity patterns of these muscles during mouse ultrasonic vocalization reflect patterns seen in other mammals during audible phonation ( Riede , 2011 , 2013 ) . However , we found that duration of vocalizations was the same between control and mutant mice ( Figure 8 ) . In addition , disruption of the vocal center of the adult mouse cortex leads to changes in the distribution of mean frequencies of vocalizations ( Arriaga et al . , 2012 ) , a parameter that was not affected in our Gli3xt/+ mice ( Figure 8 ) . Conversely , manipulation of the vocal center did not alter the distribution of syllables produced ( Arriaga et al . , 2012 ) , while Gli3 mutation did ( Figure 9 ) . Finally , step vocalizations like those affected in our mutant mice are not correlated with either thyroarytenoid muscle EMG activity or sub-glottal pressure ( Riede , 2011 , 2013 ) ; and in fact , such step vocalizations can be produced independently of muscle or neural activity in excised bat larynges ( Kobayasi et al . , 2012 ) . These data suggest that such steps may result from a passive biomechanical effect in the larynx itself , leading us to prefer the alternative explanation that defective vocalization in Gli3xt/+ mice results from defects in the larynx . By disrupting the normally tight connection of vocal fold muscles to the thyroid cartilage , we propose that the excess neural crest-derived connective tissue observed in the larynges of Gli3xt/+ heterozygous mice may disrupt the biomechanics of the vocal folds and thereby impair normal sound production . Ultimately , further studies will be required to better define the source of vocalization defects in Gli3 mutant mice . Importantly , however , the data here demonstrate that mouse models can both inform our understanding of mammalian vocalization and could also provide insights into the etiology of human laryngeal and voice defects .
The following mouse lines were used: Gli3xt-j ( Hui and Joyner , 1993; Johnson , 1967 ) ; Wnt1-cre: Tg: ( Wnt1-cre ) 11Rth ( Danielian et al . , 1998 PMID: 9636087 ) ; Mesp1Cre ( Saga et al . , 1999 ) ; Isl1Cre/+ ( Srinivas et al . , 2001 ) Pax3Cre/+ ( Engleka et al . , 2005 ) and reporter line R26tdTomato ( Ai9; Madisen et al . , 2010 ) , R26mT/mG: GT ( Rosa ) 26Sortm4 ( ACTB-tdTomato-EGFP ) Luo ( Muzumdar et al . , 2007 ) , Fuz mutants: Fuzgt ( neo ) ( Gray et al . , 2009 ) . Genotyping was performed as described in original publications . All animal work was performed in accordance with approved IACUC protocols at the University of Texas at Austin . All immunohistochemistry , skeletal and histological staining were performed according to standard protocols . All embryos were collected in cold PBS and fixed in 4% paraformaldehyde . All embryos were sectioned horizontally at 18 µm for cryosections and 4 µm for paraffin sections . R26RmT/mG cryosections were stained with DAPI ( 1:1000 ) and then coverslipped with Vectashield ( Vector Labs ) . Primary antibodies used for immunohistochemistry on cryosections: anti-human Desmin ( D33 , Dako ) . Secondary antibodies used were Alexafluor 488 ( Biotechnology Company , Austin , TX , USA ) at 1:500 . Hematoxylin and eosin staining and trichrome staining ( HT25A , Sigma ) were performed at the Dell Pediatric Research Institute Tissue Processing Core . Area and width of morphological features were determined in Fiji using the freehand selection and straight line tools , respectively . Two-four representative sections were measured from each biological replicate . All histological sections were imaged with a 20X lens on a Scanscope ( Aperio , Leica ) and processed via ImageScope ( Aperio , Leica ) and Adobe Photoshop . Male Gli3/xt/+ mice were mated with Swiss Webster females ( Charles River Laboratories ) . 5 days post-natal pups were separated from the mother and isolated on bedding in a recording chamber . Mouse vocalizations were recorded with an Ultrasoundgate 416 hr ( Avisoft ) sound recording system with a CM16/CMPA microphone ( Avisoft ) at a 250 kHz sampling rate and 16 bit resolution using Avisoft-RECORDER software , with the microphone suspended 5 cm from the pup . The start and stop times of ultrasonic vocalizations were automatically detected and frequency contours extracted using Ax ( Seagraves et al . , 2016; https://github . com/JaneliaSciComp/Ax ) . Briefly , time overlapped segments were Fourier transformed using multiple discrete prolate spheroidal sequences as windowing functions , followed by an F-test to identify time–frequency points with intensity significantly above noise ( p<0 . 01 ) . Acoustic segmentation was verified , and , if necessary , corrected manually . Signals that had exceeded the amplitude limit of the recording system ( had ‘clipped’ ) were excluded from analysis . In order to compare the vocalizations of the two genotypes we create a single high-dimensional space that fully captures all the acoustic structure in the frequency contours . We then visualize that high dimensional space using dimensionality reduction to create a two-dimensional map of the vocal repertoire . In this way , we are able to look at the vocal similarity across the vocal repertoire in the same reference frame for both genotypes . Frequency contours were mean frequency subtracted , and then all pairs of frequency contours were compared using dynamic time warping ( Sakoe and Chiba , 1978 ) to create an all-to-all distance matrix ( ( 9013 × 9013 ) /2 comparisons ) . The data in this high dimensional distance matrix were then embedded into two dimensions using t-SNE ( van der Maaten and Hinton , 2008 ) ( transition entropy = 5 , relative convergence of the cost function to 0 . 0001 ) . t-Distributed Stochastic Neighbor embedding ( t-SNE ) is a nonlinear embedding method that aims to preserve the local structure within a data set . This is achieved through placing points into a low-dimensional space such that points that were nearby in a higher-dimensional representation remain nearby in the new representation . Specifically , this embedding is calculated through optimally matching local similarity measures obtained in both the high and low dimensional spaces . Unlike other non-linear embedding approaches , this technique preserves clusters within a data set , but will allow for larger length scale distortions in order to obtain the desired dimensionality reduction . This is precisely the opposite of PCA , multi-dimensional scaling , or Isomap ( Tenenbaum et al . , 2000 ) , which aim to preserve the global structure at the expense of local distortions . Because t-SNE preserves local neighbor relationships from the full-dimensional space of the frequency contour , regions in the map can be thought of as rough categories of vocalizations based on acoustic similarity . However , only local relationships are preserved , long-length-scale relationships are distorted , which means that the axes of the map are inherently arbitrary . A detailed description of applying t-SNE to a behavioral data set can be found in ( Berman et al . , 2014 ) . Simple measures of acoustic structure ( duration , bandwidth , and average frequency were calculated from the automatically extracted contours ) . Differences between WT and HT in these simple measures were tested using single factor ANOVAs on average values for each individual . To test for differences in the number of step and non-step vocalizations between WT and HT each vocalization was labeled as a step or non-step based on their position in the vocal repertoire map . This labeling was automatic and blind to genotype . Regions of significant difference between the HT and WT maps were determined using bootstrapping , where we estimated the variation in the measured probability density functions due having a finite number of vocalizations in the data set . This was achieved through separately resampling the 2-D embeddings of the vocalizations for each case ( WT and HT ) with replacement 10 , 000 times and convolving each of these resampled data sets with a Gaussian of width 4 to create distributions , qHT ( ρ|x , y ) and qWT ( ρ|x , y ) for each of the PDFs at every point in space ( qHT ( ρ|x , y ) ≡Prob ( ρHT ( x , y ) =ρ ) and qWT ( ρ|x , y ) ≡Prob ( ρWT ( x , y ) =ρ ) . These spatially-varying PDFs were obtained by fitting a Gaussian mixture model to the sampled PDFs ( up to three peaks , chosen at each point by maximizing the Akaike Information Criterion ) . As we assume that the two populations are sampled independently , the probability that ρHT ( x , y ) is greater than ρHT ( x , y ) , defined here as PHT ( x , y ) , is thus given by numerically integrating PHT ( x , y ) =∫0∞∫0ρ_HTqWT ( ρWT|x , y ) qHT ( ρHT|x , y ) dρWT dρHT . Regions of significant difference are those where PHT ( x , y ) <α or PHT ( x , y ) >1−α . Here , we used α=0 . 05 , but corrected for multiple comparisons using the Šidák correction . We conservatively assume the number of comparisons to be 2∧H , where H is the entropy of the original 2D embedding of our data set ( H=−∫∫ρ ( x , y ) log2 ρ ( x , y ) dx dy , where ρ ( x , y ) =12 ( ρHT ( x , y ) +ρWT ( x , y ) ) ) . To test for differences in the number of step and non-step vocalizations between WT and HT each vocalization was labeled as a step or non-step based on their position in the vocal repertoire map ( see Figure 9C , vocalizations in green regions were labeled step , vocalizations in the pink region were labeled non-step ) . Because the map was generated using vocalizations from both genotypes , this labeling was automatic and blind to genotype . We then compared the number of step and non-step vocalizations in the two genotypes using the χ2 test . | Nearly all animals communicate using sound . In many cases these sounds are in the form of a voice , which in mammals is generated by a specialized organ in the throat called the larynx . Millions of people throughout the world have voice defects that make it difficult for them to communicate . Such defects are distinct from speech defects such as stuttering , and instead result from an inability to control the pitch or volume of the voice . This has a huge impact because our voice is so central to our quality of life . A wide range of human birth defects that are caused by genetic mutations are known to result in voice problems . These include disorders in which the Hedgehog signaling pathway , which allows cells to exchange information , is defective . Projections called cilia that are found on the outside of many cells transmit Hedgehog signals , and birth defects that affect the cilia ( called ciliopathies ) also often result in voice problems . Although the shape of the larynx has a crucial effect on voice , relatively little is known about how it develops in embryos . Mice are often studied to investigate how human embryos develop . By studying mouse embryos that had genetic mutations similar to those seen in humans with ciliopathies , Tabler , Rigney et al . now show that many different tissues interact in complex ways to form the larynx . A specific group of cells known as the neural crest was particularly important . The neural crest helps to form the face and skull and an excess of these cells causes face and skull defects in individuals with ciliopathies . Tabler , Rigney et al . show that having too many neural crest cells can also contribute towards defects in the larynx of mice with ciliopathies , despite the larynx being in the neck . Further investigation showed that the Hedgehog signaling pathway was required for the larynx to develop properly . Furthermore , recordings of the vocalizations of the mutant mice showed that they had defective voices , thus linking the defects in the shape of the larynx with changes in the vocalizations that the mice made . Overall , Tabler , Rigney et al . show that mice can be used to investigate how the genes that control the shape of the larynx affect the voice . The next step will be to use mice to investigate other genetic defects that cause voice defects in humans . Further research in other animals could also help us to understand how the larynx has evolved . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2017 | Cilia-mediated Hedgehog signaling controls form and function in the mammalian larynx |
Targeting conserved and essential processes is a successful strategy to combat enemies . Remarkably , the clinically important Staphylococcus aureus pathogenicity islands ( SaPIs ) use this tactic to spread in nature . SaPIs reside passively in the host chromosome , under the control of the SaPI-encoded master repressor , Stl . It has been assumed that SaPI de-repression is effected by specific phage proteins that bind to Stl , initiating the SaPI cycle . Different SaPIs encode different Stl repressors , so each targets a specific phage protein for its de-repression . Broadening this narrow vision , we report here that SaPIs ensure their promiscuous transfer by targeting conserved phage mechanisms . This is accomplished because the SaPI Stl repressors have acquired different domains to interact with unrelated proteins , encoded by different phages , but in all cases performing the same conserved function . This elegant strategy allows intra- and inter-generic SaPI transfer , highlighting these elements as one of nature’s most fascinating subcellular parasites .
The Staphylococcus aureus pathogenicity islands ( SaPIs ) are the prototypical members of an extremely common and recently identified family of mobile genetic elements , the phage-inducible chromosomal islands ( PICIs ) ( Martínez-Rubio et al . , 2017; Penadés and Christie , 2015 ) . The SaPIs are clinically relevant because they carry and disseminate superantigen genes , especially those for toxic shock toxin and enterotoxin B . They are very widespread among the staphylococci and are exclusively responsible for menstrual toxic shock , a rare but important human disease . In the absence of a helper phage they reside passively in the host chromosome , under the control of a global SaPI-coded repressor , Stl , a DNA-binding protein whose sequence is rather poorly conserved among the different members of the SaPI family ( Tormo-Más et al . , 2010 ) . Following infection by a helper phage or induction of a helper prophage , they excise , replicate extensively , and are packaged in phage-like particles composed of phage virion proteins , leading to very high frequencies of inter- as well as intrageneric transfer ( Novick et al . , 2010; Penadés and Christie , 2015 ) . In previous work we demonstrated that SaPI de-repression is effected by specific phage proteins that bind to Stl , disrupting the Stl-DNA complex and thereby initiating the excision-replication-packaging ( ERP ) cycle of the islands ( Tormo-Más et al . , 2010 ) . Different SaPIs encode different Stl repressors , so each SaPI targets a different phage protein for its de-repression . Thus , the inducers for SaPIbov1 , SaPIbov2 and SaPI1 correspond to the phage trimeric dUTPase ( Dut ) , 80α ORF15 and Sri , respectively ( Tormo-Más et al . , 2010 , 2013 ) . Since SaPIs require phage proteins to be packaged , this strategy couples the SaPI and phage cycles , but imposes a significant transmission cost on the helper phages ( Frígols et al . , 2015 ) . Importantly , although phages carrying mutations in the genes encoding the aforementioned SaPI inducers can be propagated in the lab , these mutations have a fitness cost when the mutant phages compete with the wild-type phages in the same conditions ( Frígols et al . , 2015 ) , which indicates that the phage coded SaPI inducers provide an important function for the phages in nature . We recently proposed that phages could easily overcome this SaPI imposed cost using two complementary strategies that result in phages with reduced or null capacity to induce the islands ( Frígols et al . , 2015 ) . On the one hand , phages can encode allelic variants of the SaPI inducers with reduced affinity for the SaPI coded Stl repressor . On the other hand , some phages seem to overcome SaPI induction by replacing the phage-encoded SaPI inducing gene by another one encoding an analogous protein ( an unrelated protein that performs the same biological function ) . Although experiments performed in the laboratory suggest that in response to these strategies SaPIs can antagonistically coevolve by inactivating their Stl repressors , this strategy superimposes a high cost for the bacteria , associated with an uncontrolled SaPI replication ( Frígols et al . , 2015 ) , so it is unlikely that this occurs in nature . A recent study , however , questioned the idea that phages could overcome the SaPI tyranny by replacing the SaPI inducing gene by another one encoding a functionally related protein . While all the staphylococcal S . aureus phages encode Duts; some encode dimeric and others trimeric Duts , never both ( Frígols et al . , 2015 ) . Importantly , dimeric and trimeric Duts are completely unrelated both in sequence and structure , representing a nice example of convergent evolution ( Penadés et al . , 2013 ) . While the 80α and ϕ11 phage-encoded trimeric Duts were initially described as the SaPIbov1 inducers ( Tormo-Más et al . , 2010; 2013 ) , the dimeric Dut from phage ϕNM1 also induces SaPIbov1 ( Hill and Dokland , 2016; Hill et al . , 2017 ) . The fact that both dimeric and trimeric Duts induce SaPIbov1 raised the interesting possibility that the Stl repressors could target different phage proteins , significantly increasing the capacity of the SaPIs to be induced and transferred . This result also raised other interesting questions about the SaPIs: is this phenomenon exclusive of SaPIbov1 or are other SaPIs also induced by unrelated proteins ? If that was the case for specific SaPIs , are these unrelated proteins always performing the same function for the phages or conversely it is possible that a specific SaPI repressor interacts with proteins performing unrelated functions ? And finally , what is the molecular mechanism by which the SaPI-encoded Stl repressors interact with different proteins ? Here we set out to answer all these questions and have demonstrated that it is more complicated than expected for the phages to overcome the SaPIs superimposed tyranny . Our results provide evidence of inter-species PICI transfer in nature . We have also deciphered the molecular mechanism used by the SaPIs to hijack the helper phage machinery in order to get high intra- and inter-generic transference: instead of interacting with specific partners , SaPIs have evolved a fascinating strategy that promotes their high transfer by pirating conserved phage mechanisms .
What is the mechanism by which the SaPIbov1 repressor interacts with apparently unrelated proteins ? Obviously , and since the trimeric and dimeric Duts perform the same biological function , the most likely scenario would be the existence of a conserved domain in the phage-encoded proteins that would be recognised by the SaPIbov1 coded Stl repressor . The structure of the phage 80α and ϕ11 coded Duts has recently been solved ( Leveles et al . , 2013; Tormo-Más et al . , 2013 ) . Moreover , in-depth structural , genetic and biochemical studies have demonstrated that the trimeric Dut domains IV , V and VI are involved in SaPIbov1 Stl recognition ( Maiques et al . , 2016; Tormo-Más et al . , 2010 , 2013 ) . To know whether similar domains are present in the dimeric Duts , we initially addressed the following question: does the SaPIbov1 Stl interact just with the ϕNM1 dimeric Dut or can it interact with other phage coded dimeric Duts ? To solve this question , we analysed the ϕO11 dimeric Dut . As occurred with the trimeric Duts ( Tormo-Más et al . , 2010 ) , the dimeric ϕNM1 and ϕO11 Duts are basically identical except in a divergent central region ( Figure 1—figure supplement 1 ) . Interestingly , the ϕO11 dimeric Dut also induces the SaPIbov1 and SaPIbov5 cycles ( Figure 1A ) . Note that SaPIbov5 was also included in these studies because it encodes the same Stl repressor as SaPIbov1 , with both islands being induced by the same helper phages ( Carpena et al . , 2016; Viana et al . , 2010 ) . Expression of the ϕO11 dimeric Dut ( from the Pcad promoter in expression vector pCN51 ) in a SaPIbov1 or SaPIbov5 positive strain demonstrated that this protein is sufficient to induce the SaPI cycles . Thus , when overexpressed , the cloned ϕO11 dimeric dut induced SaPIbov1 and SaPIbov5 excision and replication ( Figure 1A ) . In all 3 characterised SaPIs ( SaPI1 , SaPIbov1 and SaPIbov2 ) , Stl blocks SaPI induction by binding to the SaPI stl-str divergent region , blocking transcription of most of the SaPI genes . SaPI de-repression occurs after a direct protein-protein interaction between the cognate phage inducer and the SaPI coded Stl repressor ( Tormo-Más et al . , 2010 ) . To test if the mechanism involving the ϕO11 dimeric Dut in SaPIbov1 induction matches with that previously reported for the other SaPIs , we first demonstrated that ϕO11 Dut induces xis expression , which normally is repressed by Stl . This was confirmed using plasmid pJP674 , which carries a β-lactamase reporter gene fused to xis , downstream of str and the StlSaPIbov1-repressed str promoter , and also encodes StlSaPIbov1 ( see Figure 1B ) . The cloned ϕO11 dut gene was introduced on vector pCN51 and expression was tested in the presence or absence of an inducing concentration of CdCl2 . Induction of ϕO11 dut strongly increased β-lactamase expression from the str promoter ( Figure 1B ) . Moreover , the predicted protein–protein interaction between the ϕO11 Dut and the StlSaPIbov1 repressor was confirmed by co-expression and affinity purification of His6-StlSaPIbov1 and untagged ϕO11 Dut proteins . It was possible to co-purify a complex between His6-StlSaPIbov1 and ϕO11 Dut ( Figure 1C ) . The identity of each of these bands was confirmed by amino acid sequencing and mass spectrometry . We conclude from these results that dimeric ϕO11 Dut induces the SaPI cycle using the same mechanism described for the unrelated trimeric Dut proteins . Moreover , and although this is not the scope of this study , these results also involve the dimeric Duts in SaPI signalling . It is predicted that dimeric and trimeric Duts acquire a completely unrelated fold ( Penadés et al . , 2013 ) . However , since both dimeric and trimeric Duts perform the same enzymatic activity , we hypothesised that these proteins could have conserved domains responsible for the interaction with Stl . To test this hypothesis , and since the structure of the staphylococcal phage-encoded dimeric Duts remains unsolved , the structure of the dimeric ϕO11 Dut in complex with the nonhydrolyzable dUTP analog α , β-imido-dUTP ( dUPNPP ) and Mg2+ was determined at 2 . 1 Å resolution ( Supplementary file 1 ) . The crystal structure showed 2 molecules in the asymmetric unit organized as a homodimer ( Figure 2 ) . The structure shows that ϕO11 is an all-helix protein composed of only seven α-helices ( α1 , residues 7–23; α2 , 29–47; α3 61–82; α4 , 86–98; α5 , 104–108; α6 , 110–121 and α7 , 127–141 ) per protomer . Interestingly , the ϕO11 dimeric Dut has a ‘compact’ conformation compared to counterparts in other organisms ( which encompass ten or more α-helices ) ( Figure 2 and Figure 2—figure supplement 1 ) . Nevertheless , the ϕO11 protomer presents the characteristic structural core of dimeric Duts composed of four helices ( α1-α3 and α7 in ϕO11 ) that conforms the active centre where the nucleotide binds ( Figure 2 ) . In the ϕO11 dimer both active centres are oriented towards the same molecule face , forming a long channel that accommodates two molecules of dUPNPP . The rest of the protomer is placed on the opposite molecule face ( residues 83–138 ) , which corresponds to the divergent region in phage-encoded dimeric Duts but also adopts a helical fold ( helices α4-α6 in ϕO11 ) ( Figure 2 and Figure 2—figure supplement 1 ) . Since in the trimeric Duts the motifs IV , V and VI are essential for interaction with the Stl repressor ( Maiques et al . , 2016; Tormo-Más et al . , 2013 ) , we looked for the presence of structural elements with similar topology in the dimeric Dut . As could be anticipated by the difference in folding between the trimeric ( all-beta ) and dimeric ( all-alpha ) proteins , none of these motifs are present in the ϕO11 Dut ( Figures 2 and 3 ) . In the trimeric Duts , these three motifs place together surrounding the nucleotide in the active centre ( Figure 3 ) , thus we wondered whether the Stl recognition site was generated spatially by the disposition of specific residues provided by these three motifs rather than by the motifs themselves . To check this possibility we spatially compared the active sites of both types of Duts by superimposing the nucleotide-binding sites of the trimeric 80α and dimeric ϕO11 phagic Duts ( Figure 3 ) . As was previously observed in the comparison of the active centres from other dimeric and trimeric Duts ( Harkiolaki et al . , 2004 ) , the way of dUTP recognition and binding is completely different in both Dut types , not only in the orientation of the plane of the uracil moiety , which showed a relative rotation of more than 75° , but also in the disposition of the phosphates . In trimeric Duts , the α-phosphates acquire a gauche catalytic-competent geometry ( Kovári et al . , 2008 ) meanwhile a trans conformation is observed in the dimeric ϕO11 Dut ( Figure 3 ) . Furthermore , the β and γ phosphates differ in their relative disposition , chelating a single divalent metal in the trimerics , versus two in the dimerics ( Hemsworth et al . , 2013 ) ( Figures 2 and 3 ) . Therefore , the active centres in both types of enzymes show divergent architecture and , consequently , the spatial disposition of the residues surrounding the nucleotides , including those provided by motif IV , V and VI , is completely different . Taken together , these results strongly suggest that the SaPIbov1 Stl repressor has different interacting domains/ways to recognise the unrelated trimeric and dimeric Duts . To go further with these analyses , we generated a set of deletional mutants in the SaPIbov1 Stl repressor , with the idea that some of these mutants would specifically affect the interaction of the SaPIbov1 Stl repressor with one of the Dut types under study , but not with the other . Sequence analysis and in silico modelling indicates that SaPIbov1 Stl is mainly an α-helical protein composed of a N-terminal HTH DNA-binding domain ( residues 1–80 ) and C-terminal portion of unknown function that seems to be conformed of two domains connected by a region of low complexity ( residues 167–179 ) ( Figure 4—figure supplement 1 and Supplementary file 2; [Nyíri et al . , 2015] ) . Thus , we generated Stl deletional variants lacking the N-terminal DNA binding domain ( residues 1–86; StlΔHTH ) or the most C-terminal subdomain ( residues 176–267; StlΔCter ) ( Figure 4—figure supplement 1 ) . Unfortunately , these mutants couldn’t be analysed in vivo , since the generated Stl mutant repressors had lost the capacity to block the SaPI cycle . To solve that problem , we expressed the different Stl mutants in E . coli , and analysed in vitro their capacity to interact with the different Duts . Interestingly , deletion of the N-terminal DNA-binding domain abolished the interaction with the trimeric ϕ11 but not with the dimeric ϕO11 Dut . Conversely , the elimination of the C-terminal subdomain impairs the binding to the dimeric but not to the trimeric Dut ( Figure 4A ) . Moreover , it has been shown the interaction with the Stl repressor inhibits the dUTPase activity of both dimeric and trimeric Duts ( Hill and Dokland , 2016; Szabó et al . , 2014 ) . Here we have confirmed this inhibitory activity for the ϕ11 and ϕO11 Duts with the full-length Stl protein ( Figure 4B ) . Furthermore , and in agreement with the binding capacity shown by the Stl deletional variants , StlΔHTH inhibits the dUTPase activity of dimeric but not trimeric Duts , while StlΔCter has the opposite capacity ( Figure 4B ) . The fact that the SaPIbov1 Stl has particular regions for interacting with the trimeric and dimeric Duts supports the idea that the SaPIbov1 Stl repressor has evolved distinct ways to specifically interact with the dimeric or trimeric Duts . We next addressed the question of whether the previous phenomenon was exclusive to SaPIbov1 . To do that , we initially tried to identify the phage 80α inducer for SaPI2 , a SaPI frequently responsible for the clinically relevant menstrual toxic shock syndrome ( TSS; Subedi et al . , 2007 ) . Since SaPIs severely interfere with helper phage reproduction , a classical strategy used to identify non-essential SaPI inducers is to generate spontaneous phage mutants that are able to form plaques in the presence of the SaPIs . This strategy selects for phage mutants that have lost the ability to mobilise the islands because of mutations they carry in the SaPI inducer genes . These mutations usually generate non-functional proteins that have also lost their capacity to relieve Stl-mediated repression ( Frígols et al . , 2015; Tormo-Más et al . , 2010 ) . After many attempts , we obtained only a single spontaneous 80α phage mutant which was able to form plaques on S . aureus strain RN4220 containing SaPI2 , suggesting that the SaPI2 inducer is absolutely essential for the phage cycle even in laboratory conditions . In this mutant the 3’ region of the 80α ORF16 has been lost . Translation of this mutated gene generates a chimeric protein fused with the single strand binding protein ( Ssb; 80α ORF17; Figure 5—figure supplement 1 ) . Since in this mutant phage the ssb gene ( including its ribosomal binding site ) is unaffected and can be transcribed and translated independently of the chimeric structure , this result suggests that ORF16 is the SaPI2 inducer . The 80α ORF16 protein belongs to the Sak family of single strand annealing proteins ( SSAP , also called recombinases ) involved in homologous recombination ( Lopes et al . , 2010; Scaltriti et al . , 2011 ) . Although for many of these proteins their role in the phage cycle has not been established yet , we have recently demonstrated that this protein is essential for 80α phage replication ( Neamah et al . , 2017 ) . Note , however , that the chimeric Sak-Ssb protein is still functional for the phage , as demonstrated by the fact that the mutant phage encoding this protein still replicates and forms plaques in a sensitive recipient strain . Expression of the 80α sak ( ORF16 ) gene ( from the Pcad promoter in expression vector pCN51 ) in a SaPI2 positive strain demonstrated that Sak is sufficient to induce this SaPI . Thus , when overexpressed , the cloned sak ( but not the chimeric Sak-Ssb protein ) induced SaPI2 excision and replication ( Figure 5 ) . As the protein levels produced from these constructs are comparable ( Figure 5 ) , this result clearly shows that although expressed , the chimeric protein has lost its capacity to induce SaPI2 . Moreover , and to confirm that the mechanism involving Sak in SaPI2 induction matches with that previously reported for the other SaPIs , we demonstrated that 80α Sak induces expression of the SaPI2 Stl repressed genes . This was confirmed using plasmid pJP1977 , which carries a β-lactamase reporter gene fused to xis , downstream of str and the StlSaPI2-repressed str promoter , and also encodes StlSaPI2 ( see Figure 6A ) . The cloned sak gene was introduced on vector pCN51 and expression was tested in the presence or absence of an inducing concentration of CdCl2 . Induction of sak , but not the chimeric sak-ssb , strongly increased β-lactamase expression from the str promoter ( Figure 6A ) . Moreover , the predicted protein–protein interaction between Sak and the StlSaPI2 repressor was confirmed by co-expression and affinity purification of His6-StlSaPI2 and untagged Sak proteins . It was possible to co-purify a complex between His6-StlSaPI2 and Sak ( Figure 6B ) , whereas untagged Sak alone did not bind to the resin . The chimeric Sak-Ssb , which does not derepress SaPI2 ( Figure 6A ) , did not co-purify with His6-StlSaPI2 , confirming the specificity of the His6-StlSaPI2::Sak interaction . The identity of each of these bands was confirmed by amino acid sequencing and mass spectrometry . Next , and based on the fact that both dimeric and trimeric Duts induce SaPIbov1 , we explored the possibility that the SaPI2 Stl repressor could also target different phage proteins , significantly increasing the capacity of SaPI2 to be induced and transferred . Interestingly , phages ϕ80 and ϕ52A can also induce the SaPI2 cycle ( Ram et al . , 2014 ) , although none of them encodes a 80α Sak protein . To test the possibility that SaPI2 was targeting another protein , we tried to identify the SaPI2 inducer in phages ϕ80 and ϕ52A by generating spontaneous phage mutants that can grow in the presence of the island . After many attempts , we did not get any phage mutants capable of forming plaques in a SaPI2 positive strain , suggesting that the SaPI2 inducers are also essential for the biology of these phages , even in laboratory conditions . In view of this result , and bearing in mind that both the dimeric and trimeric Duts have the same biological ( enzymatic ) function for the phage , we hypothesised that the ϕ80 or ϕ52A SaPI2 inducers would be functionally related to the 80α Sak protein . Since the S . aureus phages display synteny , we speculated that the genes located in the same genome position as the 80α sak gene would be essential for the phage , would have a recombinase function , and would encode for the SaPI2 inducer . While phages ϕ80 and ϕ52A do not contain an orf homologue to 80α sak , all three phages encode identifiable ssb genes , which in the case of the 80α phage is located downstream of the sak gene ( Figure 5—figure supplement 2 ) . Thus , we analysed the possibility that the genes upstream of ssb were the SaPI2 inducers . Both ϕ80 and ϕ52A phages carried an identical gene , named ORF13 in phage ϕ80 and ORF16 in phage ϕ52A , which encodes a non-related protein to the 80α Sak ( Figure 5—figure supplement 3 ) . This protein belongs to a distinct family of SSAPs , Sak4 ( Lopes et al . , 2010 ) . While Sak4 and Sak are not homologous in sequence ( Figure 5—figure supplement 3 ) , we have recently demonstrated that they are both SSAPs ( recombinases ) performing a similar function in their cognate phages ( Neamah et al . , 2017 ) . The results above support the hypothesis proposing that unrelated proteins performing the same function for the phages could all act as inducers for a specific SaPI . Thus , expression of the ϕ80 and ϕ52A Sak4 proteins in a SaPI2 positive strain demonstrated that they are sufficient to induce the SaPI2 cycle ( Figure 5 ) . Moreover , expression of the sak4 genes strongly increased β-lactamase expression from the Stl-repressed str promoter ( Figure 6A ) . Since expression of the ϕ52A Sak4 protein in E . coli generated an insoluble protein which aggregates , we couldn’t co-purify a complex between His6-StlSaPI2 and untagged ϕ52A Sak4 . However , a two-hybrid assay confirmed the strong interaction between both the ϕ52A Sak4 recombinase and the SaPI2 Stl repressor and between the 80α Sak protein and the SaPI2 Stl pair ( used here as a control; Figure 6C ) , confirming that the phage Sak4 protein is a bona fide SaPI2 inducer . Since SaPI2 superimposes a high cost for the phage , it could be possible that staphylococcal phages would initially avoid this interference by encoding additional SSAPs , unrelated to Sak or Sak4 . In turn , and if the hypothesis we propose here is correct , it could also be possible that the SaPI repressor would evolve to target these new phage encoded recombinase proteins . In silico scrutiny looking at the genes located upstream of the ssb genes revealed that staphylococcal phages encode at least 4 distant SSAP families , including Erf , Redβ , and the aforementioned Sak and Sak4 ( Supplementary file 3 ) . All the staphylococcal phages encode one SSAP , in accordance with the fact that these proteins are essential for the phage ( Neamah et al . , 2017 ) . To test the possibility that these other unrelated recombinases also induced SaPI2 , we characterised in detail those present in phages ϕSLT ( ORF 17 ) and ϕN315 ( SA1794 ) , which belong to the Erf and Redβ families of SSAPs , respectively , and have completely different sequences ( Figure 5—figure supplement 4 ) . We selected phage ϕSLT because it is clinically relevant , encoding the Panton-Valentine leukocidin ( PVL ) toxin . Applying the same methodology and strategies previously used to characterise Sak and Sak4 , our results confirm that: ( i ) the expression of the ϕSLT Erf and ϕN315 Redβ proteins is sufficient to induce the SaPI2 cycle ( Figure 5 ) ; ( ii ) expression of these recombinases prevents Stl from binding to the SaPI2 stl-str divergent region ( Figure 6A ) and ( iii ) the two-hybrid assay confirmed the interaction between the SaPI2 Stl repressor and the Erf and Redβ recombinases ( Figure 6C ) . Finally , since the existence of different interacting domains in the Stl repressor explains why both the dimeric and trimeric proteins can induce SaPIbov1 , we wondered if a similar mechanism was employed by the SaPI2 island . Structure-based modelling of Sak , Sak4 , Erf and Redβ suggested they are unrelated , although Sak , Erf and Redβ can be connected through remote homology relationships ( Lopes et al . , 2010 ) . Thus , it has been proposed that Sak , Erf and Redβ belong to a large superfamily adopting a shortcut Rad52-like fold ( Lopes et al . , 2010 ) . However , structural models produced with I-Tasser ( Yang et al . , 2015 ) and Phyre2 ( Kelley et al . , 2015 ) servers for Sak ( phage 80α ) , Erf ( ϕSLT ) and Redβ ( ϕN315 ) only proposed the Rad52 fold for Sak , whereas for Erf and Redβ alternating foldings , non-related with Rad52 recombinases , were proposed with low confidence ( Figure 7 and Supplementary files 4A and B ) . By contrast , remote homologs to Sak4 are predicted to adopt a shortcut Rad51/RecA fold ( Lopes et al . , 2010 ) and models obtained from I-Tasser and Phyre2 servers proposed this fold for the ϕ52A Sak4 recombinase with good confidence ( Figure 7 and Supplementary files 4A and B ) . Taken together , these results suggest that the most likely scenario explaining why the SaPI2 Stl repressor can interact with different recombinases is the existence of different interacting domains in the repressor . Both we and others have previously demonstrated that the SaPIs can be inter- and intra-generically transferred ( Chen and Novick , 2009; Chen et al . , 2015; Maiques et al . , 2007 ) . Although this process occurs at astonishingly high frequencies in the lab , its impact in nature remains unsolved . The fact that the mechanisms involved in the life cycle of the phages are conserved among species raised an interesting possibility: by targeting conserved phage processes SaPI-like elements would be successfully spread and maintained in nature . To test this hypothesis , we searched for SaPIbov1 and SaPI2 Stl homologs in the database . Different SaPIbov1 Stl homologs were identified in PICI elements present in S . aureus , Staphylococcus hominis , Staphylococcus haemolyticus , Staphylococcus lugdunensis , Staphylococcus saprophyticus and Staphylococcus simulans . SaPI2 Stl homologs were also identified in many different Staphylococci , including Staphylococcus argenteus , Staphylococcus caprae , S . lugdunensis , Staphylococcus epidermidis , S . haemolyticus , S . simulans , Staphylococcus xylosus and Staphylococcus capitis , as well as in PICI from other Gram-positive bacteria , including Bacillus decisifrondis or Streptococcus pyogenes . Supplementary files 5A and B delineate characteristics of the different PICI elements and the identity among the different Stl repressors encoded by the Staphylococci PICIs , respectively . Of note is the fact that some islands , present in different species , encode identical Stl repressors , suggesting inter-species transfer . This was the case for SaPIbov1 ( S . aureus ) and SlCIVISLISI_25 ( S . lugdunensis ) , both encoding the SaPIbov1 repressor , and SaPI2 ( S . aureus ) , SarCISJTUF21285 ( S . argenteus ) , ScCIM23864:W1 ( S . caprae ) and SlCIFDAARGOS_141 ( S . lugdunensis ) , all encoding the SaPI2 Stl . To test if these Stl repressors interact with the SaPIbov1 or SaPI2 inducers , we used the aforementioned strategy to generate a set of plasmids in which the divergent str/str-xis region of the PICIs was fused to a β-lactamase reporter gene . These derivatives were generated for the PICIs encoding the most distantly related Stl repressors: ShoCI794_SEPI ( S . hominis ) and ShaCI51-48 ( S . haemolyticus ) , encoding a SaPIbov1 Stl homolog , and ShaCI137133 ( S . haemolyticus ) , SeCINIHLM095 ( S . epidermidis ) and SsCIUMC-CNS-990 ( S . simulans ) carrying a SaPI2 Stl homolog . Next , the capacity of the dimeric ϕO11 or trimeric ϕ11 Duts ( for the SaPIbov1-like Stl repressors ) , or the ability of the different SSAPs ( for the SaPI2-like Stl repressors ) to induce the PICI cycle was tested by introducing the pCN51 derivatives expressing the different SaPI inducers in the strains carrying the reporter plasmids . Remarkably , both the dimeric ϕO11 and trimeric ϕ11 Duts induced β-lactamase expression from the Stl-repressed str promoters present in the ShoCI794_SEPI and ShaCI51-48 islands ( Table 1 ) , suggesting that the Stl repressors encoded in all these islands have a common origin . Even more interesting were the results obtained with the SaPI2 Stl homologs ( Table 2 ) . All the islands were induced by at least one of the recombinases , although the distribution was not as uniform as with the Duts . Thus , the 80α coded Sak induced SaPI2 , SeCINIHLM095 and ShaCI137133 , but not ScCIUMC-CNS-990 ( Table 2 ) . The chimeric 80α Sak-Ssb protein induced none , supporting that the different PICI coded Stl repressors are structurally related ( Table 2 ) . Interestingly , the ϕ52A coded Sak4 recombinase induced SaPI2 and ScCIUMC-CNS-990 , while the ϕSLT Erf recombinase induced SaPI2 and SeCINIHLM095 . Finally , the ϕN315 coded Redβ recombinase uniquely induced SaPI2 , but not the other islands ( Table 2 ) . Taken together , including the previous results with the SaPIbov1 Stl mutants , these results strongly support the idea that although originally related , the different Stl repressors have evolved different domains to interact with the phage-coded inducers . It is striking how the SaPIs have evolved an elegant tactic to be highly transferred both intra- and inter-generically . However , and in the case of the inter-generic transfer of the elements , to be completely effective this strategy requires that the phages infecting the new SaPI-recipient species encode the conserved SaPI inducers . To test this , we analysed the presence of SaPIbov1 or SaPI2 inducing genes in a subset of staphylococcal phages infecting species other than S . aureus . As shown in Supplementary file 6 , we were able to identify homologs to the previously characterised SaPIbov1 or SaPI2 inducer in all the analysed phages , although with different degrees of identity among the members of the distinct families . Next , and to support the idea that once the inter-species transfer occurs the PICI can be maintained in the new recipient species , we tested whether the Dut encoded in the S . epidermidis phage IPLA6 , or the recombinase encoded in S . epidermidis phage PH15 , were capable of inducing the cycle of the different PICI elements encoding SaPIbov1 or SaPI2 Stl homologs . That was the case , and the behaviour of the S . epidermidis IPLA6 trimeric Dut was identical to that observed for the trimeric ϕ11 Dut ( Table 1 ) , while the functioning of the ϕPH15 Sak recombinase was indistinguishable from that observed with the homologue S . aureus 80α coded Sak ( Table 2 ) . In summary , our results confirm the idea that the PICIs have established a fascinating parasitic strategy that may allow their promiscuous transfer and widespread maintenance in nature . The fact that some islands present in different species encode identical proteins ( including not just the Stl repressors as demonstrated here but also other PICI proteins ) strongly supports the idea that some ancestral elements were horizontally transferred among species . In the different species these elements probably evolved independently , trying to adapt to the new cognate host . Our previous results , however , suggest that by pirating conserved phage mechanisms this inter-species transit probably occurs constantly in nature . To test that possibility , we scrutinised the genome databases looking for identical PICIs present in different staphylococcal species . We initiated this analysis by comparing the genomes of those PICIs encoding identical Stl proteins . Highlighting the versatility of these elements and the successful strategy they use to spread in nature , our analysis revealed that the ScCIM23864:W1 ( S . caprae ) and SlCIFDAARGOS_141 ( S . lugdunensis ) elements are identical ( just 3 mismatches over 13 , 847 nt; Supplementary file 7 ) .
The manner by which related SaPIs have acquired the ability to exploit conserved phage processes by targeting structurally unrelated proteins as antirepressors represents a remarkable evolutionary adaptation . Our results suggest that the most likely scenario explaining why the SaPI/PICI Stl repressors can interact with different phage coded inducers is the existence of different interacting domains in the SaPI Stl repressors . The presence of these different domains highlights the co-evolutionary and constant battle established between the helper phages , trying to avoid PICIs induction , and the parasitic PICIs , trying to interact with non-inducing phages ( Frígols et al . , 2015 ) . This mechanism could also be responsible , at least in part , for the widespread distribution of PICIs in nature . Note that we have recently demonstrated the existence of these elements in many Gram-positive cocci ( Martínez-Rubio et al . , 2017 ) . We hypothesised that at the beginning of the SaPI-phage war , a single phage protein may have been originally targeted; to escape from SaPI de-repression , because SaPIs interfere with phage maturation , substitution of the gene encoding this protein to one expressing a non-related , but functionally similar protein could have had a selective advantage for the phage . A second stage in SaPI evolution could have involved divergence of the SaPI repressor , enabling it to complex with structurally non-related phage proteins . The fact that the Stl repressors interact with structurally unrelated proteins performing the same function makes this strategy unique in nature and extremely effective . Note that in terms of increasing their transferability , a more simple strategy for the SaPIs could have been to select for Stl repressors that can interact with proteins performing different functions for the phage . However , since phages have mosaicism , encoding multiple versions of unrelated proteins performing the same function ( as also demonstrated here ) , this strategy would select for phages insensitive to the SaPIs that encode the correct combination of non-inducing proteins . By contrast , and since the processes targeted by the SaPIs are extremely well conserved in the staphylococcal phages , the fact that the SaPIs target different versions of proteins involved in the same biological processes limits the capacity of the phages to overcome SaPI parasitism , ensuring the transferability of these elements . Thus , our results show that SaPI-phage interactions represent a remarkable microcosm within the bacterial intracellular universe , highlighting SaPIs as one of the most fascinating and effective subcellular parasites . However , our results raise an interesting question . Why do some repressors interact just with one inducer , limiting their capacity to be transferred , while others seem to have a broader spectrum of inducers ? Our hypothesis is that although all the analysed phages encode putative SaPI inducers , these are different in sequence ( see Supplementary file 6 ) , so the repressors present in the different PICIs have evolved to increase their interaction with the specific inducers encoded in the cognate phages infecting these bacterial species . This also would explain the divergence in sequence observed in related Stl repressors . This hypothesis is currently under study . Lactococcus lactis encodes a plasmid with an abortive infection mechanism , AbiK ( Bouchard and Moineau , 2004 ) . As occurs with SaPI2 , the proteins targeted by the AbiK system are the different phage encoded SSAPs involved in homologous recombination ( Bouchard and Moineau , 2004 ) . Although the mechanism by which AbiK blocks phage reproduction remains unclear , it does not seem to involve the formation of a complex between the AbiK protein and the recombinases , as occurs with SaPI2 ( Bouchard and Moineau , 2004; Wang et al . , 2011 ) . Since the discovery of the SaPIs , it has gradually become apparent that prophages and PICIs have evolved in much more interesting ways than has generally been realised . PICIs are sophisticated , elegant and extremely effective parasites . They have incorporated an impressive arsenal of effective strategies to interfere with helper phages , ensuring their presence in nature ( Penadés and Christie , 2015 ) . We anticipate here that novel and unexpected mechanisms of PICI-mediated phage interference will soon be characterised , which will highlight the fascinating biology of these subcellular creatures and their cognate helper phages .
The bacterial strains used in this study are listed in Supplementary file 8A . S . aureus was grown in Tryptic soy broth ( TSB ) or on Tryptic soy agar plates . E . coli was grown in LB broth or on LB agar plates . Antibiotic selection was used where appropriate . Preparation and analysis of phage lysates was performed essentially as previously described ( Ubeda et al . , 2008 ) . General DNA manipulations were performed using standard procedures . Plasmid constructs used in this study ( Supplementary file 8B ) were generated by cloning PCR products obtained with oligonucleotide primers , listed in Supplementary file 8C . Detection probes for SaPI DNA in Southern blots were generated by PCR using primers SaPIbov1-112mE and SaPIbov1-113cB ( SaPIbov1 and SaPIbov5 ) or Tet-1m and Tet-2c ( SaPI2 ) , listed in Supplementary file 8C . Probe labelling and DNA hybridization were performed following the protocol provided with the PCR-DIG DNA-labelling and chemiluminescent detection kit ( Roche ) . Southern blot experiments were performed as previously described ( Tormo-Más et al . , 2010 ) . ϕO11 dut was cloned into pET28a vector ( Novagen ) using primers listed in Supplementary file 8C . Plasmids pETNKI-StlΔHTH and pETNKI-StlΔCter for expression of Stl deletional variants were produced using plasmid pETNKI-Stl as template ( Maiques et al . , 2016 ) . pETNKI-StlΔCter plasmid expressing Stl residues from 1 to 176 was generated by site direct mutagenesis introducing a stop codon in pETNKI-Stl after Lys176 using the Stl_M1-K176_Fw and Stl_M1-K176_Rv primers and Q5 Site-Directed Mutagenesis Kit ( NEB ) . pETNKI-StlΔHTH plasmid expressing Stl residues from 87 to 267 was generated by PCR-amplifying the encoding region with the primers Stl_T87-N267_Fw and Stl_T87-N267_Rv . The Ligation-Independent Cloning ( LIC ) system ( Savitsky et al . , 2010 ) was used to clone the PCR fragment into the pETNKI-his-SUMO3-LIC plasmid ( kindly supplied by Patrick Celie , NKI Protein facility ) previously digested with a KpnI ( Fermentas ) . All clones were sequenced at the IBV Core Sequencing facility or by Eurofins MWG Operon . Samples were taken at times 0’ and 3 hr following plasmid induction and pelleted . The samples were re-suspended in 50 μl lysis buffer ( 47 . 5 μl TES-Sucrose and 2 . 5 μl lysostaphin [12 . 5 μg ml-1] ) and incubated at 37°C for 1 hr . For the Southern blot analysis , 55 μl of SDS 2% proteinase K buffer ( 47 . 25 μl H2O , 5 . 25 μl SDS 20% , 2 . 5 μl proteinase K [20 mg ml-1] ) was added before incubation at 55°C for 30 min . Samples were vortexed for 1 hr with 11 μl of 10x loading dye . Cycles of incubation in dry ice and ethanol , then at 65°C were performed . Samples were run on 0 . 7% agarose gel at 25V overnight . DNA was transferred to a membrane and exposed using a DIG-labelled probe and anti-DIG antibody , before washing and visualisation . Preparation of S . aureus samples for western blot was performed by re-suspending pellets in 200 μl digestion/lysis buffer ( 50 mM Tris-HCl , 20 mM MgCl2 , 30% w/v raffinose ) plus 1 μl of lysostaphin ( 12 . 5 μg ml-1 ) , mixed briefly , and incubated at 37°C for 1 hr . 2X Laemmli sample buffer ( Bio-Rad , 2-mercaptoethanol added ) was added to the samples , which were heated at 95°C for 10 min , put on ice for 5 min and fast touch centrifuged . Samples were run on SDS-PAGE gels ( 15% Acrylamide , Bio-Rad 30% Acrylamide/Bis Solution ) before transferring to a PVDF transfer membrane ( Thermo Scientific , 0 . 2 μM ) using standard methods . Western blot assays were performed using anti-Flag antibody probes ( Monoclonal ANTI-FLAG M2-Peroxidase ( HRP ) antibody , Sigma-Aldrich ) as per the protocol supplied by the manufacturer . The two-hybrid assay for protein-protein interaction was done as described previously ( Battesti and Bouveret , 2012 ) using two compatible plasmids; pUT18c expressing T18 fusion with the individual recombinases , and pKNT25 expressing the T25 fusion with the StlSaPI2 . Both plasmids were co-transformed into E . coli BTH101 for the Bacterial Adenylate Cyclase Two Hybrid ( BACTH ) system and plated on LB +Ampicillin and Kanamycin + X gal as an indicator . After incubation at 30°C for 24 hr ( early reaction ) and 48 hr ( late reaction ) , the protein-protein interaction was detected by a colour change . Blue colonies represent an interaction between the two clones , while white/yellow colonies are negative for any interaction . For quantification of the BACTH analysis , overnight cultures were diluted 1/100 and grown to mid-log before induction with 5 mM IPTG . After 2 hr , 2 ml of culture was sampled and pelleted , before resuspension in the same volume of Z buffer ( 0 . 06M Na2HPO4 . 7H2O , 0 . 04M NaH2PO4 . H2O , 0 . 01M KCl , 0 . 001M MgSO4 , 0 . 05M β-mercaptoethanol , pH 7 . 0 ) . The OD600 was recorded and cells were permeabilized with chloroform and 0 . 1% SDS . The assay reaction was started using ONPG ( o-nitrophenyl-β-D-galactoside , 4 mg/ml ) , and vortexed and incubated at 30°C until yellow . The reaction was stopped using Na2CO3 and the reaction time recorded . Samples were spun down and the OD420 and OD550 were recorded . Miller Units were calculated as follows , where T is time of reaction ( minutes ) and V is the volume of culture used in the assay ( ml ) : Miller Units = 1000 x ( OD420 - 1 . 75 x OD550 ) / ( T xV x OD600 ) . For the β-Lactamase assays , cells were obtained at 0 . 2–0 . 3 OD540 and at 5 hr post-induction with/without 5 μM CdCl2 . β-Lactamase assays , using nitrocefin as substrate , were performed as described ( Tormo-Más et al . , 2010 ) , using a ELx808 microplate reader ( BioTek ) . An adjustment was made in reading time , with plates read every 20 s for 30 mins . β-Lactamase units/ml are defined as [ ( slope ) ( Vd ) ]/[ ( Em ) ( l ) ( s ) ] . Slope is the ∆absorbance/hour , V is the volume of the reaction , d is the dilution factor , Em is the millimolar extinction coefficient for the nitrocefin ( 20 , 500 M−1 cm−1 at 486 nm ) , l is the path length ( cm ) , and s is the sample amount . dUTPase activity was measured by Malachite Green assay as previously described ( Maiques et al . , 2016 ) . Briefly , Dut ( 30 nM ) alone or in presence of a 5X molar ratio ( monomer ) of Stl ( full length or truncated versions ) was incubated overnight at 4°C in Stl buffer ( 400 mM NaCl; 75 mM Hepes7 . 5; 5 mM MgCl2 ) . The experiment was carried out at 25°C and started by the addition of dUTP ( 10 μM final concentration ) . As indicated in the figure legends either a two-way ANOVA comparison with Sidak’s adjustment for multiple comparisons was conducted or a one-way ANOVA , as appropriate . All analysis was done using Graphpad Prism 6 software . Trimeric ϕ11 Dut and Stl were expressed and purified as previously described ( Maiques et al . , 2016 ) . StlΔCter was purified following an identical protocol as for the Stl full-length protein . StlΔHTH was produced from E . coli BL21 ( DE3 ) ( Novagen ) cultures harbouring the pETNKI-StlΔHTH plasmid . The culture was grown at 37°C in LB medium supplemented with 33 μg/ml kanamycin up to an OD600 of 0 . 5–0 . 6 , and then protein expression was induced with 0 . 1 mM isopropyl-β-D thiogalactopyranoside ( IPTG ) at 20°C for 16 hr . After induced cells were harvested by centrifugation at 4°C for 30 min at 3500 × g , the cell pellet was resuspended in buffer A ( 75 mM HEPES pH 7 . 5 , 400 mM NaCl and 5 mM MgCl2 ) supplemented with 1 mM PMSF and sonicated . A soluble fraction was obtained after centrifugation at 16 000 × g for 40 min , and it was loaded on a pre-equilibrated His Trap HP column ( GE Healthcare ) . After washing with 10 column volumes of buffer A , the protein was eluted by adding buffer A supplemented with 500 mM imidazole . The eluted protein was digested for His-SUMO3 tag removal using SENP2 protease at a molar ratio 1:50 ( protease:eluted protein ) for 16 hr at 4°C with slow shaking . After digestion , the sample was loaded one more time into the pre-equilibrated His Trap HP column to remove the His-SUMO3 tag and SENP2 protease from the Stl protein . Fractions were analysed by SDS-PAGE and those fractions with purest digested Stl protein were selected , concentrated and stored at −80°C . His-tagged dimeric ϕO11 Dut was overexpressed in E . coli BL21 ( DE3 ) ( Novagen ) harbouring the pJP1938 plasmid . The cells were grown to exponential phase at 37°C in LB medium supplemented with 33 μg/ml kanamycin , and then protein expression was induced by the addition of 1 mM IPTG for 3 hr . After induction , cells were harvested by centrifugation , re-suspended in buffer A supplemented with 1 mM phenylmethanesulfonyl fluoride ( PMSF ) and lysed by sonication . The lysate was clarified by centrifugation and the soluble fraction was loaded on a His Trap HP column pre-equilibrated with buffer A . The column was washed with the same buffer supplemented with 10 mM imidazole and proteins were eluted with buffer A supplemented with 500 mM imidazole . The eluted proteins were concentrated and loaded onto a Superdex S200 ( GE Healthcare ) equilibrated with buffer B ( 75 mM HEPES pH 7 . 5 , 250 mM NaCl and 5 mM MgCl2 ) for size exclusion chromatography . The fractions were analysed by SDS-PAGE and those fractions showing purest protein were selected , concentrated and stored at −80°C . Mass Spectrometry analyses were performed at the proteomics facility of SCSIE , University of Valencia . ϕO11 Dut protein in complex with dUPNPP protein was crystallized using the sitting drop method in the Crystallogenesis facility of IBV . ϕO11 ( at 10 mg/ml ) was incubated with 0 . 5 mM dUPNPP ( 2-Deoxyuridine-5-[ ( α , β ) -imido]triphosphate; Jena Biosciences ) and 5 mM MgCl2 during 8 hr at 4°C and sitting drops were set up at 21°C . The best crystals were obtained using 0 . 2 M magnesium chloride , 0 . 1 M Tris-HCl pH8 . 5 , 20% PEG 8000 as liquor mother . Crystals were frozen in liquid nitrogen respecting crystallization condition , increasing the cryobuffer to 35% PEG 8000 concentration for the diffraction process . Diffraction data was collected from single crystals at 100 K on ALBA ( Barcelona , Spain ) and DLS ( Didcot , UK ) synchrotrons and processed and reduced with Mosflm ( Powell et al . , 2013 ) and Aimless ( Evans and Murshudov , 2013 ) programs from the CCP4 suite ( Winn et al . , 2011 ) . The data-collection statistics for the best data sets used in structure determination are shown in Supplementary file 3 . Protein structure was solved by molecular replacement with Phaser ( McCoy et al . , 2007 ) and an edited PDB of the dimeric Dut from phage ϕDI as a model ( 5MYD ) . Based on sequence homology between ϕO11 and ϕDI Duts ( 70% identity ) , we excluded amino acids 82–140 , corresponding to the divergent regions present in the phage dimeric Duts , from the starting model . This decision was made in order to reduce the imposition of any initial structural conformation to this variable region . Iterative refinement , rebuilding and validation steps were done using programs Coot ( Emsley et al . , 2010 ) and Phenix ( Adams et al . , 2010 ) . The final model includes two Dut molecules ( amino acids sequence 4–160 and 3–158 ) forming one dimer in an asymmetric unit with one dUPNPP molecule and two Mg ions bound at each of the two active centres . The final structure has good geometry as indicated by the Ramachandran plots ( any residue in the disallowed region ) . A summary of structure refinement statistics is shown in Supplementary file 3 . Purified proteins were mixed at 40 μM 1:1 molar ratio in a buffer A ( final volume 18 μl ) and incubated at 4°C overnight . Samples were loaded into an 8% polyacrylamide gel and electrophoresis was performed at 4°C . Native gels were stained with coomassie brilliant blue . The 3D homology model of 80α , ϕSLT , 52A and ϕN315 SSAPs , and SaPIbov1 Stl were constructed using I-Tasser ( default mode ) ( Yang et al . , 2015 ) and Phyre2 ( intensive mode ) ( Kelley et al . , 2015 ) servers ( Supplementary files 2A , B and 4 ) . Intrinsic protein disorder was predicted by the meta-server Metadisorder ( Kozlowski and Bujnicki , 2012 ) . | Many harmful microbes can produce different molecules that make them more effective in causing and spreading diseases . These molecules can also be obtained from ‘mobile genetic elements’ that can be transferred between bacteria within a population . Pathogenicity islands are one such type of mobile genetic element and are very common among bacteria known as staphylococci . They spread toxin-encoding genes between bacteria , including one that can lead to a condition called toxic shock syndrome in humans . Pathogenicity islands are normally found within the DNA of the bacteria , where they are deactivated by specific repressor proteins . However , in the presence of another type of mobile genetic element – the bacteriophages – the repressor proteins start to interact with specific proteins encoded by the bacteriophages . This allows the pathogenicity islands to become active and spread to other bacteria . Previous research has shown that in the bacterium known as Staphylococcus aureus , different pathogenicity islands have different repressors . Scientists therefore assumed that the repressors are only able to interact with certain bacteriophage proteins . However , since pathogenicity islands are widespread in nature , it could be possible that they use other ways to hijack the bacteriophage machinery to ensure their transfer . To test this hypothesis , Bowring et al . studied two types of pathogenicity islands in S . aureus and revealed that their two different repressors did not interact with specific bacteriophage proteins as previously hypothesized . Instead , each repressor could interact with multiple bacteriophage proteins that had a variety of different structures , including proteins from completely different bacteriophages . Bowring et al . also discovered that each of the analyzed repressor proteins did not actually recognize any specific shared structural features on the bacteriophage proteins , but rather evolved to target proteins that play the same role in various bacteriophages . This suggests the repressors target a specific process rather than a single protein . This strategy allows them to be transferred within the same species , but also between different ones . A next step will be to better understand how a repressor can recognize structurally unrelated proteins , and establish what evolutionary forces are driving this phenomenon . A deeper knowledge of how pathogenicity islands spread between staphylococci is vital to understand how these bacteria can become resistant to treatments such as antibiotics . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2017 | Pirating conserved phage mechanisms promotes promiscuous staphylococcal pathogenicity island transfer |
Our ability to rationally optimize allosteric regulation is limited by incomplete knowledge of the mutations that tune allostery . Are these mutations few or abundant , structurally localized or distributed ? To examine this , we conducted saturation mutagenesis of a synthetic allosteric switch in which Dihydrofolate reductase ( DHFR ) is regulated by a blue-light sensitive LOV2 domain . Using a high-throughput assay wherein DHFR catalytic activity is coupled to E . coli growth , we assessed the impact of 1548 viable DHFR single mutations on allostery . Despite most mutations being deleterious to activity , fewer than 5% of mutations had a statistically significant influence on allostery . Most allostery disrupting mutations were proximal to the LOV2 insertion site . In contrast , allostery enhancing mutations were structurally distributed and enriched on the protein surface . Combining several allostery enhancing mutations yielded near-additive improvements to dynamic range . Our results indicate a path toward optimizing allosteric function through variation at surface sites .
In allosteric regulation , protein activity is modulated by an input effector signal spatially removed from the active site . Allostery is a desirable engineering target because it can yield sensitive , reversible , and rapid control of protein activity in response to diverse inputs ( Dagliyan et al . , 2019; Pincus et al . , 2017; Raman et al . , 2014 ) . One common approach for achieving allosteric regulation in both engineered and evolved systems is through domain insertion: the transposition , recombination , or otherwise fusion of an ‘input’ domain into an ‘output’ domain of interest ( Aroul-Selvam et al . , 2004; Dagliyan et al . , 2016; Ostermeier and Benkovic , 2000; Nadler et al . , 2016 ) . In natural proteins , domain insertions and rearrangements play a key role in generating regulatory diversity , with kinases serving as a prototypical example ( Fan et al . , 2018; Huse and Kuriyan , 2002; Peisajovich et al . , 2010; Shah et al . , 2018 ) . In engineered proteins , domain insertions have been used to generate fluorescent metabolite biosensors ( Nadler et al . , 2016 ) , sugar-regulated TEM-1 β-lactamase variants ( Guntas et al . , 2005 ) , and a myriad of light-controlled proteins including kinases , ion channels , guanosine triphosphatases , guanine exchange factors , and Cas9 variants ( Dagliyan et al . , 2016; Wang et al . , 2016; Karginov et al . , 2011; Toettcher et al . , 2013; Shaaya et al . , 2020; Coyote-Maestas et al . , 2019; Richter et al . , 2016 ) . In all cases , domain insertion provides a powerful means to confer new regulation in a modular fashion . However , naively created domain insertion chimeras sometimes exhibit relatively modest allosteric dynamic range , with small observed differences in activity between the constitutive and activated states ( Lee et al . , 2008 ) . These fusions then require further optimization by either evolution or empirical mutagenesis , but general principles to guide this process are largely absent . Which mutations tune or improve an allosteric system ? Because we lack comprehensive studies of allosteric mutational effects in either engineered or natural systems , it remains unclear whether such mutations are common or rare , and what magnitude of allosteric effect we might typically expect for single mutations . Additionally , it is not obvious if such mutations are structurally distributed or localized ( for example , to the insertion site ) . Answers to these questions would inform practical strategies for optimizing engineered systems and provide insight into the evolution of natural multi-domain regulation in proteins . To address these questions , we performed a deep mutational scan of a synthetic allosteric switch: a fusion between the E . coli metabolic enzyme Dihydrofolate Reductase ( DHFR ) and the blue-light sensing LOV2 domain from A . sativa ( Lee et al . , 2008; Reynolds et al . , 2011 ) . This modestly allosteric chimera shows a 30% increase in DHFR velocity in response to light . Focusing on mutations to the DHFR residues , we found that only a small fraction ( 4 . 4% ) of the mutations that retained DHFR activity had a statistically significant impact on allostery . Individual mutations exhibited generally modest effect sizes; the most allosteric single mutant characterized ( H124Q ) yielded a twofold increase in velocity in response to light relative to the starting construct . Structurally , allostery disrupting mutations tended to cluster near the LOV2 insertion site and were modestly enriched at both conserved and co-evolving amino acid positions . In contrast , allostery enhancing mutations were distributed across the protein , and strongly associated with the protein surface . We observed that combining a few of these mutations yielded near-additive enhancements to allosteric dynamic range . Collectively , our data elucidates practical strategies for optimizing engineered systems , and shows that weakly conserved , structurally distributed surface sites can contribute to allosteric tuning .
To begin our study of allostery tuning mutations , we selected a previously characterized synthetic allosteric fusion between DHFR and LOV2 generated in earlier work ( Lee et al . , 2008; Reynolds et al . , 2011 ) . In this fusion , the LOV2 domain of A . sativa is inserted between residues 120 and 121 of the E . coli DHFR βF-βG loop; we refer to this construct as DL121 ( Figure 1A , B ) . The choice of LOV2 insertion site was guided by Statistical Coupling Analysis ( SCA ) , an approach for analyzing coevolution between pairs of amino acids across a homologous protein family ( Rivoire et al . , 2016; Lockless and Ranganathan , 1999; Halabi et al . , 2009 ) . A central finding of SCA is that co-evolving groups of amino acids , termed sectors , often form physically contiguous networks in the tertiary structure that link allosteric sites to active sites ( Halabi et al . , 2009; Süel et al . , 2003; Pincus et al . , 2018 ) . To create the DL121 fusion , Lee et al . followed the guiding principle that sector connected surface sites in DHFR might serve as preferred sites ( or ‘hot spots’ ) for the introduction of allosteric regulation ( Lee et al . , 2008 ) . The resulting DL121 fusion covalently attaches the N- and C-termini of LOV2 into a sector connected surface on DHFR , and displays a twofold increase in DHFR hydride transfer rate ( khyd ) in response to blue light ( Lee et al . , 2008 ) . Under steady-state conditions , we measured a 28% increase in the turnover number ( kcat ) in response to light and a statistically insignificant change in the Michaelis constant ( Km ) ( Figure 1C ) . Thus , the DL121 fusion is modestly allosteric in vitro . As DHFR has no known natural allosteric regulation , the LOV2 insertion confers a new , evolutionarily unoptimized regulatory input . But can this relatively small allosteric effect generate measurable physiological differences that could provide the basis for evolutionary selection ? DHFR catalyzes the reduction of 7 , 8-dihydrofolate ( DHF ) to 5 , 6 , 7 , 8-tetrahydrofolate ( THF ) using NADPH as a co-factor . THF then serves as a one-carbon donor and acceptor in the synthesis of thymidine , purine nucleotides , serine , glycine , and methionine . Because of these critical metabolic functions , DHFR activity is strongly linked to growth rate , and under appropriate conditions , E . coli growth rate can be used as a proxy for DHFR activity ( Reynolds et al . , 2011; Thompson et al . , 2020 ) . Prior work found that the modest in vitro allosteric effect of DL121 conferred a selectable growth rate advantage in vivo: when an E . coli DHFR deletion strain ( ER2566 ΔfolAΔthyA ) was complemented with DL121 , the resulting strain grew 17% faster in the light than in the dark ( Reynolds et al . , 2011 ) . Thus , DL121 is a system where: ( 1 ) allosteric control is rapidly and reversibly applied , ( 2 ) the allosteric effects on activity can be readily quantified both in vitro and in vivo , and ( 3 ) there remains potential for large improvements in regulatory dynamic range through mutation . Our goal was to measure the effect of every single amino acid mutation in DHFR on the allosteric regulation of DL121 . To do this , we aimed to follow a strategy loosely akin to a double mutant cycle ( Figure 1D ) . The starting DL121 construct shows so-called V-type allostery , in which the effector ( light ) regulates the catalytic turnover number ( kcat ) ( Carlson and Fenton , 2016 ) . Thus , allostery can be quantified as the ratio of kcat between lit and dark states . More generally , allostery might be considered as a ratio of velocities ( v = kcat [S]/ ( Km + [S] ) ) between the lit and dark states , as the allosteric effector could regulate turnover , substrate affinity , or both . In either case , we defined the allosteric effect of mutation as the fold change in allosteric regulation upon mutation ( Figure 1D , blue box ) . We sought to infer this quantity for every mutation in a saturation mutagenesis library of DHFR by using growth rate as a proxy for catalytic activity . As in prior work , we measured the growth rate of many E . coli strains in parallel by using next generation sequencing ( NGS ) to monitor the frequency of individual DHFR mutants over time in a mixed culture ( Figure 2; Reynolds et al . , 2011; Thompson et al . , 2020 ) . Allele frequencies ( fa ) at each time point ( t ) were normalized as follows: fa=lnNaNWTt-lnNaNWTt=0 where Na and NWT are the number of mutant and wildtype ( WT ) counts at a given time point . By performing a linear fit of the log normalized allele frequencies vs . time we calculated a slope corresponding to relative growth rate: this value is the difference in growth rate for the mutant relative to a reference ( 'WT' ) construct . As individual mutations tend to exhibit modest effects on allosteric regulation , we optimized the linear regime and resolution of the growth rate assay in two ways ( Reynolds et al . , 2011 ) . First , we grew the E . coli populations in a turbidostat outfitted with blue LEDs to activate LOV2 ( Figure 2A ) . The turbidostat maintains each culture in exponential growth by dynamically sensing optical density and adjusting media dilution rate accordingly Toprak et al . , 2013; this ensures near-constant media conditions and eliminates the need for manual serial dilutions . Second , we selected media conditions – M9 minimal media with 0 . 4% glucose and 1 µg/ml thymidine supplementation – in which growth rate can resolve subtle differences in catalytic activity near the DL121 fusion . We evaluated the resolution of our assay using a ‘standard curve’ of 11 point mutations of known catalytic activity in non-chimeric DHFR ( Figure 2B ) . Under these conditions , we observed a log-linear relationship between relative growth rate and DHFR velocity over nearly four orders of magnitude; this relationship saturates ( plateaus ) for the most active mutants ( WT and M42F , Figure 2C ) . Importantly , the relative growth rate and velocity of DL121 were near the center of the linear regime of our assay . In using velocity to describe our data , we have incorporated two assumptions: ( 1 ) we presume minimal variation in protein abundance between mutants ( enzyme concentration is equal to one ) and ( 2 ) we fix the substrate concentration at 25 µM , which was previously reported as the endogenous concentration for WT E . coli ( Kwon et al . , 2008 ) . Individual mutations may cause variation in protein abundance , but because allostery concerns a relative change in activity , light-independent differences in abundance can be removed by appropriate normalization ( as discussed further below ) . As previously observed , the exponential divergence of mutants with different growth rates in a population makes it possible to detect even small biochemical effects ( Breslow et al . , 2008 ) . More specifically , we can discriminate a change of ±0 . 02 µM−1 s-1 in catalytic power ( kcat/Km ) under these conditions . This level of precision is on par with – and in some cases better than – literature-reported errors for in vitro steady state kinetics measurements of DHFR ( Reynolds et al . , 2011; Wagner et al . , 1992; Huang et al . , 1994 ) . Consequently , we can resolve small catalytic and allosteric effects of mutations on DL121 through this high-throughput growth-based assay . In order to map the coupling of individual DHFR positions to light , we constructed a deep mutational scanning library over all DHFR positions in the DL121 fusion ( Figure 3—figure supplements 1–2 ) . Then , we measured the growth rate effect of each mutation in triplicate under both lit and dark conditions using the above-described assay ( Figure 3A–C , Figure 3—figure supplements 3–4 , Figure 3—source data 1 ) . In this experiment , all growth rates were calculated relative to the unmutated DL121 fusion , which itself exhibits reduced activity ( and growth rate ) compared to WT DHFR . Mutations fell into four broad categories in terms of growth rate effects: neutral , uniformly deleterious ( Figure 3A ) , uniformly beneficial ( Figure 3B ) , or light dependent ( and thus allosteric , Figure 3C ) . We were unable to measure growth rate for 891 of the 3021 possible missense mutations ( 19 substitutions over 159 positions ) : 226 ( 7 . 5% ) were missing at the start of the experiment ( t = 0 ) for one or more replicates ( referred to as ‘no data’ ) , and an additional 665 ( 22% ) were depleted from the library before reaching the minimum of three time points required for growth rate estimation ( we refer to these as null mutants , see also Materials and methods , Figure 3—figure supplement 4 ) . We interpreted these 665 rapidly depleting null mutants as highly deleterious to growth rate and thus DHFR activity . The relative growth rates for the remaining 2130 mutations ( 70 . 5% ) were highly reproducible , with a correlation coefficient between replicate pairs above 0 . 9 ( Figure 3—figure supplement 3 ) . Before examining the allosteric effects of mutations , we first considered the effects of mutations on growth rate ( and thus DHFR activity ) in a single growth condition ( dark ) . Prior work has found that deleterious mutations are enriched at evolutionarily conserved positions and within the protein sector ( McLaughlin et al . , 2012 ) . The DHFR sector was defined by analyzing coevolution in a multiple sequence alignment of native DHFR domains , so we wished to examine if sector positions were indeed critical to function in the chimeric DL121 fusion . Good correspondence between the DHFR sector , evolutionary conservation , and deleterious mutations in DL121 would provide confidence that the core functional elements of native DHFR remain intact in the chimera . The vast majority of mutations were at least modestly deleterious to growth , with a median relative growth rate of −0 . 084 in the dark and −0 . 083 in the light ( Figure 3D ) . A cluster of beneficial mutations was observed just before the LOV2 insertion site at position 121 in both conditions , suggesting some potential to compensate for the inserted LOV2 ( Figure 3—figure supplement 4 ) . The overall distribution of fitness effects shows some differences relative to prior DMS studies of natural proteins including native E . coli DHFR ( Thompson et al . , 2020; Garst et al . , 2017 ) . First , the distribution of fitness effects for mutations in natural proteins is often centered around neutral , implying a certain degree of mutational robustness ( McLaughlin et al . , 2012; Stiffler et al . , 2015 ) . Secondly , DMS of native DHFR – under experimental conditions designed to resolve mutational effects near WT – revealed many beneficial ( activating ) mutations ( Thompson et al . , 2020 ) . There are two explanations for the relative paucity of beneficial and neutral mutations in the present dataset . First , the DL121 fusion is comparably less robust because the unoptimized LOV2 insertion introduces an initial compromise to DHFR function . Secondly , the conditions of our assay ( both expression and media ) differ from prior work ( Thompson et al . , 2020 ) and were selected to resolve mutational effects near DL121; consequently , mutations with native-like ( or better ) activity are in the saturating , non-linear regime of our assay . To identify the slowest growing – and presumably near , or entirely , inactivating – mutations , we applied an empirical growth rate cutoff of −0 . 13 to the lit and dark growth rates . This corresponds to the growth rate for DL121 D27N; D27N is an active site mutation that strongly reduces the activity of WT DHFR ( Figure 2B , C ) . The DL121 D27N mutant grows very slowly in the conditions of our assay and is inviable in the absence of thymidine supplementation ( Figure 3—figure supplement 5 ) . We found that mutations with growth rates at or below this cutoff ( including the null mutants ) were significantly enriched in both the sector ( p=7 . 9×10−8 , Figure 3E , Supplementary file 1b ) and at evolutionarily conserved positions ( p=8 . 7×10−20 , Figure 3—figure supplement 6 , Supplementary file 1c ) . When mapped to the WT DHFR structure , positions enriched for deleterious mutations surround the active site and co-factor binding pocket ( Figure 3F ) , structurally overlap with the sector ( Figure 3G ) , and include a number of positions known to play a critical role in WT DHFR catalysis ( e . g . W22 , D27 , M42 , and L54 ) ( Howell et al . , 1986; Fierke et al . , 1987 ) . These data are consistent with the view that sector positions continue to play a key role in conferring DHFR catalytic activity in the DL121 fusion . Following the thinking that ( near ) inactive DHFR variants are both inherently non-allosteric and associated with the least reproducible growth rate measurements ( Figure 3—figure supplement 3 ) , we removed the set of 1247 slow-growing ( growth rate <−0 . 13 ) and null mutations prior to the analysis of allostery . The retained 1548 mutations – representing 51% of the growth assay data – remain well-distributed between the DL121 surface , core , sector , and evolutionarily conserved positions ( Figure 3E ) . These present a high-confidence and representative subset of the data for evaluating mutational effects on DL121 allosteric regulation . To compute the allosteric effect of mutation , we considered the triplicate measurements of lit and dark relative growth rate for each mutant ( Figure 3A–C ) . Given the log-linear relationship between growth rate and DHFR velocity ( Figure 2C ) , subtracting growth rates approximates log-ratios of velocities . Thus , we estimated the allosteric effect of mutation by taking the difference in the average relative growth rates between lit and dark conditions: In the above equations , rgr is relative growth rate ( which is directly measured in our sequencing-based assay ) and gr refers to absolute growth rate . Accordingly , positive values indicate allostery enhancing mutations and negative values indicate allostery disrupting mutations ( Figures 1D and 4A ) . Of the 1548 mutations evaluated , the allosteric effect is normally distributed with a mean near zero ( µ = 0 . 0017 , Figure 4—figure supplement 1 ) . To assess the statistical significance of allosteric effects , we computed a p-value for each mutation by unequal variance t-test under the null hypothesis that the lit and dark replicate measurements have equal means . These p-values were compared to a multiple-hypothesis testing adjusted p-value of p=0 . 016 determined by Sequential Goodness of Fit ( SGoF , Figure 4B; Carvajal-Rodriguez and de Uña-Alvarez , 2011 ) . Under these criteria , only 69 mutations ( 4 . 5% of all viable mutants ) significantly influenced allostery: 56 mutations enhanced allostery while 13 disrupted allostery . We did not observe a strong association between the magnitude of growth rate effect and the allosteric effect size . Allostery-influencing mutations spanned a wide range of growth rates and exhibited comparatively modest effects on light regulation ( Figure 4C ) . To further examine the ability of the growth-based sequencing assay to quantitatively resolve mutation-associated changes in allosteric regulation , we selected 10 mutations spanning a range of allosteric and growth rate effects for in vitro characterization ( Figure 4B red dots , Figure 4—figure supplements 2–4 ) . As a control , we included the light insensitive variant DL121-C450S: the C450S mutation of LOV2 abrogates light-based signaling by blocking formation of a light-induced covalent bond between position 450 and the FMN chromophore ( Christie et al . , 2002 ) . We expressed and purified the selected DL121 mutants to near homogeneity; S148C and E154R did not yield sufficient quantities of active protein for in vitro studies . We find it noteworthy that E154R—one of the strongest allostery-enhancing mutations in vivo—was unstable in multiple purification strategies . For the remaining eight mutations we measured the kcat and Km of DHFR under lit and dark conditions ( Figure 4—figure supplement 2 ) . To confirm function of the fused LOV2 domain , we also measured relaxation of the FMN chromophore following light stimulation and collected absorbance spectra before and after the application of light ( Figure 4—figure supplements 3–4 ) . As expected , all the characterized DL121 mutations ( with the exception of DL121-C450S ) retained LOV2 domains with light-responsive absorbance spectra and chromophore relaxation constants similar to the unmutated DL121 construct . Evaluating the light dependence of DHFR activity , the change in Km value between lit and dark conditions was neither significant for any point mutation nor correlated to allosteric effect size ( R2 = 0 . 003 ) ( Supplementary file 1a , Figure 4—figure supplements 5–6 ) . The Km values for all characterized mutants ( 0 . 15–1 . 9 µM ) were similar to that of unmutated DL121 ( ~1 µM ) . Instead , we observed that light predominantly modulated catalytic turnover ( kcat ) . The ratio of kcat in the light relative to the dark ranged from 1 . 1 ( for the non-allosteric DL121-C450S construct ) to 2 . 0 ( for the most allosteric point mutation , H124Q ) ( Supplementary file 1a , Figure 4—figure supplements 5–6 ) . For reference , the starting DL121 construct has a lit:dark kcat ratio of 1 . 3 . So why might the characterized allosteric mutations predominantly effect kcat ? One plausible explanation is that the conditions of our in vivo experiments fall within a pseudo-zero-order kinetics regime ( [DHF]>>Km ) . In this scenario , light-associated changes in Km would have little impact on enzyme velocity ( and accordingly growth rate ) and go undetected in our assay . Consistent with this , the in vivo concentration of DHF for wildtype E . coli ( 25 µM ) is well above the Km for all the characterized DL121 mutations . Alternatively , it could be that the biophysical mechanism of the DL121 fusion somehow makes it more energetically feasible for light to modulate kcat than Km . In any case , the 1 . 3- to 2-fold changes in kcat translate to similar fold changes in enzyme velocity . A comparison of the in vitro allosteric effect on velocity to the in vivo growth rate effect yields a near-linear relationship with a correlation coefficient of 0 . 83 ( Figure 4D ) . Taken together , these data show that our growth-based assay is quantitatively reporting on changes in allostery , and that the allosteric mutations identified here modulate DHFR activity through changes in catalytic turnover number . Next , we examined the distribution of allostery-tuning mutations on the WT DHFR tertiary structure . The 13 allostery disrupting mutations localized to six DHFR positions concentrated near the LOV2 insertion site ( Figure 5A ) . More specifically , 90% of the allostery disrupting mutations occurred within 10 Å of the DHFR 121 cα atom ( Figure 5B ) . These mutations were modestly enriched in the protein sector ( Supplementary file 1d ) . Overall , the observed spatial distribution suggests these mutations may disrupt allostery by altering local structural contacts needed to ensure communication between DHFR and LOV2 . In contrast to this localized pattern , the 56 allostery enhancing mutations were observed at 25 positions distributed across the DHFR structure ( Figure 5C ) and enriched on the protein surface ( Figure 5D , Supplementary file 1e ) . These enhancing mutations were never found in the protein sector and were thus statistically significantly depleted from the protein sector ( Figure 5E , F ) . This relationship – wherein allostery disrupting mutations were modestly enriched and allostery enhancing mutations were strongly depleted from the sector – also holds when defining the set of allosteric mutations at a relaxed cutoff of p=0 . 05 ( Supplementary file 1d ) . Given the prior finding that sector connected surface sites were hotspots for introducing allostery in DHFR ( Reynolds et al . , 2011 ) , we also examined the association between allostery-influencing mutations and two other groups of DHFR positions: ( 1 ) surface sites that are either within or contacting the sector and ( 2 ) surface sites that are only contacting the sector ( but not within-sector ) . As for the analysis of sector positions only , we observed a statistically significant depletion of allostery enhancing mutations and enrichment of allostery disrupting mutations when considering the set of surface sites within or contacting the sector . This finding holds true over a range of significance thresholds for defining sector and allosteric mutations ( Supplementary file 1f ) . When considering the set of positions that contact ( but are not within ) the sector , we did not observe a statistically significant association at nearly all cutoffs ( Supplementary file 1g ) . Indeed , a number of allostery enhancing mutations do not contact the sector at all and occur in surface exposed loops ( e . g . from residues 84 to 89 , and from 116 to 119 ) . So , counter to our expectations , the optimization of allostery did not occur at sector connected sites or even proximal to the LOV2 insertion site . Instead , structurally distributed and weakly conserved surface sites provided a basis for tuning and enhancing allosteric regulation regardless of sector connectivity . Taken together , our data show that many distributed surface sites can make modest contributions to allosteric regulation . Can these mutants be combined to further improve allosteric dynamic range ? To test this , we created two mutant constructs by combining the most potent allostery enhancing mutations as characterized in vitro: the double mutant DL121-M16A , H124Q , and the triple mutant DL121-M16A , D87A , H124Q ( Figure 6A ) . For both constructs , we measured steady-state catalytic parameters ( Supplementary file 1a ) and verified LOV2 function through absorbance spectra and chromophore relaxation kinetics experiments ( Figure 6—figure supplement 1 ) . Interestingly , all three mutations exhibited near-log-additive improvements in allostery ( Figure 6B ) . The DL121-M16A , H124Q fusion exhibits a 2 . 74 fold increase in velocity upon light activation while the triple mutant shows a 3 . 87-fold increase in velocity . For both mutant combinations , the improvement in allostery is realized by reducing the dark state ( constitutive ) activity ( Figure 6—figure supplement 1 , Supplementary file 1a ) . The serial addition of allostery enhancing mutations also reduced the overall catalytic activity of DHFR , suggesting that further improvement could be obtained by combining these mutations with a non-allosteric but activity-enhancing mutation . Overall , these data suggest that a naïve sector connected fusion can be gradually evolved toward increased allosteric dynamic range through the stepwise accumulation of single mutations at structurally distributed surface sites ( Figure 6C ) .
We used deep mutational scanning to study the frequency and structural pattern of allostery tuning mutations in a synthetic allosteric system , with the goal of understanding how regulation between domains can be optimized . Overall , allostery-influencing mutations were rare – just under 5% of viable mutations had statistically distinguishable effects on the lit and dark states of the DL121 fusion . We found that mutations at conserved and co-evolving ( sector ) positions were often deleterious to DHFR function and infrequently influenced allosteric regulation . In a few cases , sector mutations served to disrupt allostery; nearly all allostery disrupting mutations were localized to the LOV2 insertion site on DHFR . Counter to our expectations , allostery enhancing mutations were distributed across the DHFR structure , depleted from the sector , and enriched on the protein surface . When considered individually , the allostery-enhancing mutations had modest effects ( up to twofold ) on regulation , but ( at least in some cases ) they can be combined to yield near-additive improvements in dynamic range . A triple mutant ( DL121-M16A , D87A , H124Q ) rationally designed using our point mutant data produces a 3 . 87-fold increase in velocity upon light stimulation , up from the 1 . 3-fold allosteric effect of our starting construct . These results should be considered in the context of our experiment: the DL121 fusion begins with sharply reduced DHFR activity , and our experiment intentionally used relatively stringent DHFR selection conditions to better resolve small differences in kinetic parameters . Thus , it is unsurprising that a large fraction of DHFR mutations in our library were deleterious , with an appreciable fraction near-inactive . This result echoes prior studies showing that the fraction of deleterious mutations ( and mutational robustness ) is strongly modulated by a variety of factors , including purifying selection strength and expression level ( Stiffler et al . , 2015; Jiang et al . , 2013; Lundin et al . , 2018 ) . Given the finding that stabilizing mutations can often improve protein evolvability ( Lundin et al . , 2018; Bloom et al . , 2006; Zheng et al . , 2020 ) , it would be interesting to examine how the distribution of mutational effects on both DL121 function and allostery would change in the background of a stability ( and/or activity ) enhancing mutation to DL121 . While we observed that the number of allosteric mutations is few and the effect sizes are generally small in our model system , a previous study of allostery tuning mutations in pyruvate kinase indicated that up to 30% of mutations can tune allostery , with the maximum observed effect size approaching 22-fold ( Tang and Fenton , 2017 ) . Nevertheless , our data serve to illuminate the pattern of mutational effects on a newly established ( and unoptimized ) domain fusion – the presumptive first step toward regulation in a number of both natural and synthetic systems . Interestingly , we observe a seeming disparity between the sites where we were able to introduce new allosteric regulation by domain fusion ( in our earlier work ) , and the sites where allosteric tuning takes place ( in this work ) . Previously , Reynolds et al . found that sector connected surface sites served as hotspots for the introduction of new light-based regulation in DHFR ( Reynolds et al . , 2011 ) . Indeed , allosteric regulation was never obtained when the LOV2 domain was inserted at a non-sector connected site . In contrast , in this work , we observed that allostery enhancing mutations were depleted both within the sector and at sector connected sites . For example , we observed a number of allostery enhancing mutations at positions 83–89 of the DHFR αD-βE loop , while LOV2 insertions in this region location did not initiate allostery as quantified either in vitro or in vivo ( Lee et al . , 2008; Reynolds et al . , 2011 ) . This suggests different structural requirements for establishing and tuning allostery in this system ( and possibly others ) : here allostery seems to be more easily introduced at evolutionarily conserved and co-evolving sites , but once established , can be optimized through less conserved sector-peripheral residues . Although our work focuses on a synthetic allosteric fusion , our results are broadly consistent with an emerging body of work characterizing allostery-influencing mutations in natural proteins . Together , these data point to a model in which mutations at evolutionarily conserved positions exert large ( and often disruptive ) effects on function while allostery is tuned at less conserved surface sites . For example , Leander et al . recently used deep mutational scanning to map the pattern of compensatory mutations that rescued allosteric function for non-allosteric tetracycline repressor ( TetR ) variants ( Leander et al . , 2020 ) . In that study a ‘disrupt-and-restore’ strategy was used: an already-allosteric system was inactivated and deep mutational scanning was then used to identify compensatory mutations . While there are significant differences between rescuing a deficient variant and the optimization of a novel allosteric construct , they likewise found that the mutations at highly conserved sites were often disruptive to stability and function , while allostery-rescuing mutations occurred at weakly conserved and structurally distributed sites ( Leander et al . , 2020 ) . Similarly , mutations at ‘rheostat’ sites – weakly conserved positions distal to the site of regulation – were found to modulate allosteric control in human liver pyruvate kinase and the lactose repressor protein ( lacI ) ( Campitelli et al . , 2020; Wu et al . , 2019 ) . Intriguingly , the association of allostery enhancing mutations with the protein surface hints at a possible role for solvent – and more specifically the protein hydration layer – in tuning regulation . The finding that the allostery initiated upon naïve fusion of the DHFR and LOV2 domains can be further enhanced by single mutations implies a path to improved allosteric dynamic range by stepwise mutagenesis and selection . Three of the most allostery enhancing mutations could be combined to yield a near-additive improvement in regulatory dynamic range . This has interesting implications for both evolved and engineered allosteric systems . In evolved systems , standing mutational variation is more likely at weakly conserved surface sites ( particularly under less stringent selection conditions ) , and this could provide a means for generating variation in allosteric regulation upon a domain fusion event . Moreover , while engineering studies sometimes use mutations near the domain insertion site to optimize regulation , our results suggest that diffuse surface site mutations could present an effective alternative . Whether by engineering or evolution , it seems that mutations at weakly conserved and structurally distributed residues can provide a path to the optimization of regulation .
DHFR DL121 sublibraries were transformed into ER2566 ∆folA ∆thyA E . coli by electroporation using a MicroPulser Electroporator ( Bio Rad ) and gene pulser cuvettes ( Bio Rad , cat#165–2089 ) . Cultures were grown overnight at 37°C in GM9 minimal media ( 93 . 0 mM Sodium ( Na+ ) , 22 . 1 mM Potassium ( K+ ) , 18 . 7 mM Ammonium ( NH4 ) , 1 . 0 mM Calcium ( Ca2+ ) , 0 . 1 mM Magnesium ( Mg2+ ) , 29 . 2 mM Chloride ( Cl- ) , 0 . 1 mM Sulfate ( SO42- ) , and 42 . 2 mM Phosphate ( PO43- ) , 0 . 4% glucose ) pH 6 . 50 , containing 50 µg/mL thymidine and 30 µg/mL chloramphenicol ( Sigma , cat#C0378-5G ) as well as folA mix which contains 38 µg/mL glycine ( Sigma , cat#50046 ) , 75 . 5 µg/mL L-methionine ( Sigma , cat#M9625 ) 1 µg/mL calcium pantothenate ( Sigma , cat#C8731 ) , and 20 µg/mL adenosine ( Sigma , cat#A9251 ) . Four hours before the start of the experiment , the overnight culture was diluted to an optical density of 0 . 1 at 600 nm in GM9 minimal media containing 50 µg/mL thymidine and 30 µg/mL chloramphenicol and incubated for four hours at 30°C . The cultures were centrifuged at 2000 RCF for 10 min and resuspended in the experimental conditions of GM9 minimal media containing 1 µg/mL thymidine and 30 µg/mL chloramphenicol . This was repeated two more times . The cultures were then back-diluted to an OD600 of 0 . 1 in 16 mL/vial of media . The turbidostat described in Toprak et al . , 2013 was used in continuous culture ( turbidostat ) mode with a clamp OD600 of 0 . 15 and a temperature of 30°C . Each vial had a stir bar . Vials designated as ‘lit’ had one 5V blue LED active . The optical density was continuously monitored throughout the experiment . 1 mL samples were taken at the beginning of selection ( 0 hr ) and at 4 , 8 , 12 , 16 , 20 , and 24 hr into selection and were centrifuged at 21 , 130 RCF for 5 min at room temperature with the pellet being stored at −20°C for sequencing sample preparation . Wild-type DHFR , 12 DHFR point mutants ( D27N , F31V , F31Y , F31Y-L54I , G121V , G121V-F31Y , G121V-M42F , L54I , L54I-G121V , M42F , and W22H ) , and three chimeric DHFR-LOV2 fusion constructs ( DL116 , DL121 , and DL121-C450S ) each in a pACYC-Duet vector with TYMS as described in Reynolds et al . , 2011 were transformed into ER2566 ∆folA ∆thyA E . coli by electroporation using a MicroPulser Electroporator ( Bio Rad ) and gene pulser cuvettes ( Bio Rad , cat#165–2089 ) ( Reynolds et al . , 2011 ) . Cultures were grown overnight at 37°C in GM9 minimal media ( 93 . 0 mM Sodium ( Na+ ) , 22 . 1 mM Potassium ( K+ ) , 18 . 7 mM Ammonium ( NH4 ) , 1 . 0 mM Calcium ( Ca2+ ) , 0 . 1 mM Magnesium ( Mg2+ ) , 29 . 2 mM Chloride ( Cl- ) , 0 . 1 mM Sulfate ( SO42- ) , and 42 . 2 mM Phosphate ( PO43- ) , 0 . 4% glucose ) pH 6 . 50 , containing 50 µg/mL thymidine and 30 µg/mL chloramphenicol ( Sigma , cat#C0378-5G ) as well as folA mix which contains 38 µg/mL glycine ( Sigma , cat#50046 ) , 75 . 5 µg/mL L-methionine ( Sigma , cat#M9625 ) 1 µg/mL calcium pantothenate ( Sigma , cat#C8731 ) , and 20 µg/mL adenosine ( Sigma , cat#A9251 ) . Four hours before the start of the experiment the overnight culture was diluted to an optical density of 0 . 1 at 600 nm in GM9 minimal media containing 50 µg/mL thymidine and 30 µg/mL chloramphenicol and incubated for four hours at 30°C . The cultures were centrifuged at 2000 RCF for 10 min and resuspended in the experimental conditions of GM9 minimal media containing 1 µg/mL thymidine and 30 µg/mL chloramphenicol . This was repeated two more times . The cultures were then back-diluted to an OD600 of 0 . 1 and pooled at equal ( 1/16th ) ratios and aliquoted into four ‘dark’ and four ‘lit’ vials with 16 ml culture . The turbidostat described in Toprak et al . , 2013 was used in continuous culture ( turbidostat ) mode with a clamp OD600 of 0 . 15 and a temperature of 30°C . Each vial had a stir bar . Vials designated as ‘lit’ had one 5V blue LED active . The optical density was continuously monitored throughout the experiment . One mL samples were taken at the beginning of selection ( 0 hr ) and at 4 , 8 , 12 , 16 , 20 , and 24 hr into selection and were centrifuged at 21 , 130 RCF for 5 min at room temperature with the pellet being stored at −20°C for sequencing sample preparation . Single point mutant DHFR-D27N , DL121 chimeric protein , and DL121 with a point mutant D27N each in a pACYC-Duet vector with TYMS as described in Reynolds et al . , 2011 were transformed into ER2566 ∆folA ∆thyA E . coli by electroporation using a MicroPulser Electroporator ( Bio Rad ) and gene pulser cuvettes ( Bio Rad , cat#165–2089 ) ( Reynolds et al . , 2011 ) . Cultures were grown overnight at 37°C in GM9 minimal media ( 93 . 0 mM Sodium ( Na+ ) , 22 . 1 mM Potassium ( K+ ) , 18 . 7 mM Ammonium ( NH4 ) , 1 . 0 mM Calcium ( Ca2+ ) , 0 . 1 mM Magnesium ( Mg2+ ) , 29 . 2 mM Chloride ( Cl- ) , 0 . 1 mM Sulfate ( SO42- ) , and 42 . 2 mM Phosphate ( PO43- ) , 0 . 4% glucose ) pH 6 . 50 , containing 50 µg/mL thymidine and 30 µg/mL chloramphenicol ( Sigma , cat#C0378-5G ) as well as folA mix which contains 38 µg/mL glycine ( Sigma , cat#50046 ) , 75 . 5 µg/mL L-methionine ( Sigma , cat#M9625 ) 1 µg/mL calcium pantothenate ( Sigma , cat#C8731 ) , and 20 µg/mL adenosine ( Sigma , cat#A9251 ) . Four hours before the start of the experiment , the overnight culture was diluted to an optical density of 0 . 1 at 600 nm in GM9 minimal media containing 50 µg/mL thymidine and 30 µg/mL chloramphenicol and incubated for four hours at 30°C . The cultures were centrifuged at 2000 RCF for 10 min and resuspended in the experimental conditions of GM9 minimal media containing either 0 , 1 , or 50 µg/mL thymidine and 30 µg/mL chloramphenicol . The cells were centrifuged and resuspended two more times . The cultures were then back-diluted to an OD600 of 0 . 005 into 96-well plates with six replicates each . Cell pellets were lysed by the addition of 10 µL sterile water , mixed by pipetting , and incubated at 98°C for 5 min . One µL of this was then combined with 5 µL Q5 buffer ( NEB , cat#M0491S ) , 0 . 5 µL 10 mM DNTP ( Thermo Scientific , cat#R0192 ) , 2 . 5 µL of 10 mM forward and reverse primers specific to the sublibrary and containing the TruSeq adapter sequence ( Appendix 1: SL1V2 , SL2V2 , SL3V2 , SL4V2 , DL121CLV3F , and DL_WTTS_R3 ) , 0 . 25 µL of Q5 enzyme ( NEB , cat#M0491S ) and 13 . 25 µL of sterile water . These samples were then heated at 98°C for 90 s and then cycled through 98°C for 10 s 63–65°C ( sublibrary 1: 66°C , sublibrary 2: 63°C , sublibrary 3: 64°C , and sublibrary 4: 65°C ) for 15 s and then 72°C for 15 s , repeating 20 times with a final 72°C heating for 120 s in a Veriti 96-well thermocycler ( Applied Biosystems ) . These samples were then amplified using TruSeq PCR reactions with a unique combination of i5/i7 indexing primers for each timepoint . 1 µL of this PCR reaction was then combined with 5 µL Q5 buffer ( NEB , cat#M0491S ) , 0 . 5 µL 10 mM DNTP ( Thermo Scientific , cat#R0192 ) , 2 . 5 µL of 10 mM forward and reverse primers , 0 . 25 µL of Q5 enzyme ( NEB , cat#M0491S ) and 13 . 25 µL of sterile water . These samples were then heated at 98°C for 30 s and then cycled through 98°C for 10 s 55°C for 10 s and then 72°C for 15 s , repeating 20 times with a final 72°C heating for 60 s in a Veriti 96 well thermocycler ( Applied Biosystems ) . Amplified DNA from i5/i7 PCR reaction was quantified using the picogreen assay ( Thermo Scientific , cat#P7589 ) on a Victor X3 multimode plate reader ( Perkin Elmer ) and the samples were mixed in an equimolar ratio . The DNA was then purified by gel extraction and a DNA Clean and Concentrator −5 kit ( Zymo Research , cat#D4014 ) . DNA quality was determined by 260 nm/230 nm and 260 nm/280 nm ratios on a DS-11 +spectrophotometer ( DeNovix ) and concentration was determined using the Qubit 3 ( Thermo Scientific ) . Pooled samples were sent to GeneWiz where they were analyzed by TapeStation ( Agilent Technologies ) and sequenced on a HiSeq 4000 sequencer ( Illumina ) with 2 × 150 bp dual index run with 30% PhiX spike-in yielding 1 . 13 billion reads . The control library was sequenced in-house using a MiSeq sequencer ( Illumina ) with 2 × 150 bp dual index 300 cycle MiSeq Nano Kit V2 ( Illumina cat#15036522 ) with 20% PhiX ( Illumina cat#FC-110–3001 ) spike-in yielding 903 , 488 reads . The E . coli DHFR LOV2 fusion was cloned as an NcoI/XhoI fragment into the expression vector pHIS8-3 ( Lee et al . , 2008; Reynolds et al . , 2011 ) . Point mutants were engineered into the DHFR gene using QuikChange II site-directed mutagenesis kits ( Agilent cat#200523 ) using primers specified in Appendix 1 . All DHFR/LOV2 fusions for purification were expressed under control of a T7 promoter , with an N-terminal 8X His-tag for nickel affinity purification . The existing thrombin cleavage site ( LVPRGS ) following the His-tag in pHIS8-3 was changed to a TEV cleavage site using restriction-free PCR to improve the specificity of tag removal ( Bond and Naus , 2012 ) . All constructs were verified by Sanger DNA sequencing . DHFR-LOV2 chimeric proteins were expressed in BL21 ( DE3 ) E . coli grown at 30°C in Terrific Broth ( 12 g/L Tryptone , 24 g/L yeast extract , 4 mL/L glycerol , 17 mM KH2PO4 , and 72 mM K2HPO4 ) . Protein expression was induced when the cells reached an absorbance at 600 nm of 0 . 7 with 0 . 25 mM IPTG , and cells were grown at 18°C overnight . Cell pellets were lysed by sonication in binding buffer ( 500 mM NaCl , 10 mM imidazole , 50 mM Tris-HCL , pH 8 . 0 ) added at a volume of 5 ml/g cell pellet . Next the lysate was clarified by centrifugation and the soluble fraction was incubated with equilibrated Ni-NTA resin ( Qiagen cat#4561 ) for 1 hr at 4°C . After washing with one column volume of wash buffer ( 300 mM NaCl , 20 mM imidazole , 50 mM Tris-HCL , pH 8 . 0 ) the DHFR-LOV2 protein was eluted with elution buffer ( 1M NaCl , 250 mM imidazole , 50 mM Tris-HCL , pH 8 . 0 ) at 4°C . Eluted protein was dialyzed into dialysis buffer ( 300 mM NaCl , 1% glycerol , 50 mM Tris-HCl , pH 8 . 0 ) at 4°C overnight in 10 , 000 MWCO Thermo protein Slide A Lyzer ( Fisher Scientific cat#PI87730 ) . Following dialysis , the protein was then purified by size exclusion chromatography ( HiLoad 16/600 Superdex 75 pg column , GE Life Sciences cat#28989333 ) . Purified protein was concentrated using Amicon Ulta 10 k M . W . cutoff concentrator ( Sigma cat#UFC801024 ) and flash frozen using liquid N2 prior to enzymatic assays . The protein was spun down at 21 , 130 RCF at 4°C for 10 min and the supernatant was moved to a new tube with any pellet being discarded . The concentration of the protein was quantitated by A280 using a DS-11 +spectrophotometer ( DeNovix ) with an extinction coefficient of 44920 mM−1 cm−1 . The parameters kcat and Km under Michaelis-Menten conditions were determined by measuring the initial velocity for the depletion of NADPH as measured in absorbance at 340 nm , with an extinction coefficient of 13 . 2 mM−1 cm−1 . This is done in a range of substrate concentrations with a minimum of 8 data points around 4 Km , 2 Km , 1 . 5 Km , Km , 0 . 8 Km , 0 . 5 Km , 0 . 25 Km and 0 . The initial velocities ( slope of the first 15 s ) were plotted vs . the concentration of Dihydrofolate and fit to a Michaelis Menten model using non-linear regression in GraphPad Prism 7 . The reactions are run in MTEN buffer ( 50 mM 2- ( N-morpholino ) ethanesulfonic acid , 25 mM tris base , 25 mM ethanolamine , 100 mM NaCl ) pH 7 . 00 , 5 mM Dithiothreitol , 90 µM NADPH ( Sigma-Aldrich cat#N7505 ) quantitated by A340 . Dihydrofolate ( Sigma-Aldrich cat#D7006 ) is suspended in MTEN buffer pH 7 . 00 with 0 . 35% β-mercaptoethanol and quantitated by A282 with an extinction coefficient of 28 mM−1 cm−1 . Depletion of NADPH is observed in 1 mL cuvettes with a path length of 1 cm in a Lambda 650 UV/VIS spectrometer ( Perkin Elmer ) with attached water Peltier system set to 17°C . Lit samples are illuminated for at least 2 min by full spectrum 125 watt 6400K compact fluorescent bulb ( Hydrofarm Inc cat#FLC125D ) . Dark samples were also exposed to the light in the same way as the lit samples but were in opaque tubs . Velocity , V=kcat[P][S]KM+[S] , was calculated using the concentration of DHF found in wild-type E . coli ( ~25 µM Kwon et al . , 2008 ) . The spectra of the LOV2 chromophore is determined with a Lambda 650 UV/VIS spectrometer ( Perkin Elmer ) at 350–550 nm using paired 100 μL Hellma ultra micro cuvettes ( Sigma cat#Z600350-1EA ) with a path length of 1 cm . Purified protein in was diluted ( when possible ) to 20 μM in MTEN buffer pH 7 . 00 with 0 . 35% β-mercaptoethanol The lit samples are illuminated for at least 2 min by full spectrum 125 watt 6400K compact fluorescent bulb ( hydrofarm Inc ) . Relaxation of the lit state chromophore is observed in the Lambda 650 UV/VIS spectrometer ( Perkin Elmer ) at 447 nm ( dark peak ) using paired 100 μL Hellma ultra micro cuvettes ( Sigma cat#Z600350-1EA ) with a path length of 1 cm . | Many proteins exhibit a property called ‘allostery’ . In allostery , an input signal at a specific site of a protein – such as a molecule binding , or the protein absorbing a photon of light – leads to a change in output at another site far away . For example , the protein might catalyze a chemical reaction faster or bind to another molecule more tightly in the presence of the input signal . This protein ‘remote control’ allows cells to sense and respond to changes in their environment . An ability to rapidly engineer new allosteric mechanisms into proteins is much sought after because this would provide an approach for building biosensors and other useful tools . One common approach to engineering new allosteric regulation is to combine a ‘sensor’ or input region from one protein with an ‘output’ region or domain from another . When researchers engineer allostery using this approach of combining input and output domains from different proteins , the difference in the output when the input is ‘on’ versus ‘off’ is often small , a situation called ‘modest allostery’ . McCormick et al . wanted to know how to optimize this domain combination approach to increase the difference in output between the ‘on’ and ‘off’ states . More specifically , McCormick et al . wanted to find out whether swapping out or mutating specific amino acids ( each of the individual building blocks that make up a protein ) enhances or disrupts allostery . They also wanted to know if there are many possible mutations that change the effectiveness of allostery , or if this property is controlled by just a few amino acids . Finally , McCormick et al . questioned where in a protein most of these allostery-tuning mutations were located . To answer these questions , McCormick et al . engineered a new allosteric protein by inserting a light-sensing domain ( input ) into a protein involved in metabolism ( a metabolic enzyme that produces a biomolecule called a tetrahydrofolate ) to yield a light-controlled enzyme . Next , they introduced mutations into both the ‘input’ and ‘output’ domains to see where they had a greater effect on allostery . After filtering out mutations that destroyed the function of the output domain , McCormick et al . found that only about 5% of mutations to the ‘output’ domain altered the allosteric response of their engineered enzyme . In fact , most mutations that disrupted allostery were found near the site where the ‘input’ domain was inserted , while mutations that enhanced allostery were sprinkled throughout the enzyme , often on its protein surface . This was surprising in light of the commonly-held assumption that mutations on protein surfaces have little impact on the activity of the ‘output’ domain . Overall , the effect of individual mutations on allostery was small , but McCormick et al . found that these mutations can sometimes be combined to yield larger effects . McCormick et al . ’s results suggest a new approach for optimizing engineered allosteric proteins: by introducing mutations on the protein surface . It also opens up new questions: mechanically , how do surface sites affect allostery ? In the future , it will be important to characterize how combinations of mutations can optimize allosteric regulation , and to determine what evolutionary trajectories to high performance allosteric ‘switches’ look like . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2021 | Structurally distributed surface sites tune allosteric regulation |
In some mammals and many social insects , highly cooperative societies are characterized by reproductive division of labor , in which breeders and nonbreeders become behaviorally and morphologically distinct . While differences in behavior and growth between breeders and nonbreeders have been extensively described , little is known of their molecular underpinnings . Here , we investigate the consequences of breeding for skeletal morphology and gene regulation in highly cooperative Damaraland mole-rats . By experimentally assigning breeding ‘queen’ status versus nonbreeder status to age-matched littermates , we confirm that queens experience vertebral growth that likely confers advantages to fecundity . However , they also upregulate bone resorption pathways and show reductions in femoral mass , which predicts increased vulnerability to fracture . Together , our results show that , as in eusocial insects , reproductive division of labor in mole-rats leads to gene regulatory rewiring and extensive morphological plasticity . However , in mole-rats , concentrated reproduction is also accompanied by costs to bone strength .
A hallmark of highly cooperative societies is reproductive division of labor . This phenomenon is best understood in eusocial insects , where environmental cues lead to reproductively and morphologically specialized castes , including one or few highly fecund ‘queens’ ( Wilson , 1971 ) . These changes help support the reproductive role of queens by differentiating them from nonbreeding colony members , who forage , care for young , and engage in colony defense ( Wilson , 1971; Keller and Genoud , 1997 ) . Queens are frequently much larger than their sterile colony mates ( e . g . , twice as large in honey bees and Pharaoh ants; Berndt and Eichler , 1987; Page and Peng , 2001 ) , reflecting dramatically altered growth and development programs that are explained by changes in gene regulation ( Smith et al . , 2008 ) . Social insects thus exemplify the tight evolutionary links between reproductive division of labor , cooperative behavior , and extreme morphological plasticity . Systems in which breeding is restricted to a single female supported by multiple nonbreeding helpers are also observed in vertebrates , including birds and mammals ( Koenig and Dickinson , 2016 ) . Here , breeding status is not determined during early development , but instead occurs in adulthood , and breeding is only achieved by those individuals who have the opportunity to transition into a reproductive role . In some species , new breeders undergo a period of accelerated growth , which may be important either for maintaining dominance or for supporting high fecundity ( Clutton-Brock et al . , 2006; Huchard et al . , 2016; O'Riain et al . , 2000; Russell et al . , 2004; Thorley et al . , 2018; Young and Bennett , 2010 ) . While substantial gene regulatory divergence with breeding status has been described for the brain and some peripheral organs ( Bens et al . , 2018; Mulugeta et al . , 2017; Sahm et al . , 2020 ) , we know little about the gene regulatory shifts responsible for breeder-associated patterns in growth . Because morphological change is often crucial for ramping up offspring production , these processes are key to understanding both the basis for , and limits of , status-driven differences in growth and development . Here , we investigate the morphological and molecular consequences of experimental transitions to breeding status in female Damaraland mole-rats ( Fukomys damarensis ) . Like naked mole-rats ( Heterocephalus glaber ) , Damaraland mole-rats are frequently classified as ‘eusocial’ ( Bennett and Faulkes , 2000; Jarvis , 1981; Jarvis and Bennett , 1993 ) , and female helpers who transition to queens experience accelerated vertebral growth associated with increases in fecundity ( O'Riain et al . , 2000; Thorley et al . , 2018 ) . However , it is not clear what triggers skeletal remodeling , where it is localized within the vertebral column , or whether it extends to other parts of the skeleton . Further , the gene regulatory changes that support skeletal remodeling in mole-rat queens are not known , nor are their consequences for skeletal growth potential and integrity . To address these questions , we experimentally assigned age-matched , female littermates to become queens or remain as nonbreeders and evaluated gene regulatory and morphological changes induced by the transition to queen status . Our results indicate that , as in eusocial insects , females that acquire breeding status experience substantial morphological remodeling , associated with pathway-specific changes in gene regulation . Notably , we found that queens not only experience lengthening of their lumbar vertebrae ( LV ) , but also show reductions in the growth potential and structural integrity of their long bones . These changes result from increased rates of bone resorption that may increase the risk of fracture , indicating that the presence of helpers does not annul the costs of reproduction to queens .
Adult female Damaraland mole-rats were randomly assigned to either transition to queen status ( n = 12 ) or remain as nonbreeders ( n = 18 ) for the duration of the experiment ( Figure 1A–C; Supplementary file 1 ) . Age at assignment ( mean age = 19 . 4 ± 4 . 4 s . d . months ) was consistent with the age at dispersal observed in wild Damaraland mole-rats ( 1–3 years , Thorley and Clutton-Brock , unpublished data ) . To resolve whether skeletal changes are a function of the queen transition per se versus release from reproductive suppression in the natal colony , nonbreeders were either kept in their natal colonies as helpers or placed into solitary housing in the absence of a breeding queen , recapitulating extended periods of dispersal in this species ( Jarvis and Bennett , 1993 ) ( n = 10 helpers and n = 8 solitaires ) . At the time of assignment , females assigned to the queen , helper , and solitaire treatments were statistically indistinguishable in body mass , age , and vertebral length ( as measured by LV5; unpaired t-tests between all pairwise combinations of treatments: p>0 . 05; Figure 1—figure supplement 1 ) . When possible , we assigned age-matched littermates to queen versus nonbreeding treatments ( 26 of 30 experimental animals were in sets of littermate sisters; Supplementary file 1 ) . Six nonexperimental animals ( one queen and five nonbreeders ) were also included in the sample , resulting in a total sample size of 13 breeders and 23 nonbreeders ( Supplementary file 1 ) . Females assigned to the queen treatment were each transferred to a new tunnel system containing only an unrelated adult male , simulating the natural process of dispersal and new colony formation in the wild ( Jarvis and Bennett , 1993 ) . This pairing procedure , which defines the queen treatment , typically leads to immediate sexual activity and rapid activation of the reproductive axis , including initiation of ovulation and the potential for conception ( Bennett et al . , 1996; Snyman et al . , 2006 ) . Queens gave birth to a mean of 6 . 92 ± 5 . 57 s . d . live offspring during the 12–22-month follow-up period , produced in a mean of 2 . 85 ± 1 . 75 s . d . litters ( range: 0–6; Supplementary file 1 ) . As expected , helpers and solitaires produced no offspring , and did not differ from each other in body mass or vertebral length after the 12–22-month follow-up period ( unpaired t-tests , all p>0 . 05; Figure 1—figure supplement 2 ) . Because helpers and solitaires were morphologically indistinguishable , and also exhibited no differences in gene expression in our subsequent genomic assays ( Supplementary file 2 ) , we grouped them together into a single ‘nonbreeder’ treatment for the remainder of our analyses . Compared to nonbreeders , queens showed rapid growth in the LV in the first 12 months post-pairing ( Figure 1D , E ) , especially in the vertebrae toward the caudal end of the vertebral column ( LV5 and LV6 ) . Based on longitudinal measurements , most of this differential growth was concentrated soon after the breeding status transition . Specifically , we observed a significant interaction between breeding status ( queen versus nonbreeder ) and post-pairing time point in the first four months of the experiment ( Figure 1D; β = 0 . 0794 , p=3 . 13×10−3; n = 49 x-rays from 28 animals ) , but not for measurements taken in later time point intervals ( 4 versus 8 months; 8 versus 12 months , 12 versus 16 months , 16 versus 22 months; all p>0 . 05 ) . Moreover , in the first four months , only queens that had already experienced pregnancy showed accelerated vertebral lengthening relative to nonbreeders ( unpaired t-test; LV5 of pregnant queens versus nonbreeders: t = −5 . 735 , df = 16 . 871 , p=2 . 50×10−5; LV5 of queens not yet pregnant versus nonbreeders: t = −0 . 789 , df = 13 . 007 , p=0 . 444; n = 14 nonbreeders , five pregnant queens , and two queens not yet pregnant ) . As a result of accelerated vertebral growth in queens post-transition , size differences persisted throughout the study . After 12 months , the absolute length of LV5 in queens was , on average , 4 . 8% longer than nonbreeders ( Figure 1E; LV5: unpaired t-test , t = 2 . 509 , df = 21 . 095 , p=0 . 020 ) , and the absolute length of the LV column in queens relative to nonbreeders was 3 . 5% longer , although the latter difference was not significant ( unpaired t-test , t = 1 . 945 , df = 22 . 49 , p=0 . 064 ) . Differences between queens and nonbreeders were even more apparent if LV measures were scaled by zygomatic arch ( head ) width , as in previous studies ( O'Riain et al . , 2000; Thorley et al . , 2018; Dengler-Crish and Catania , 2009 ) ( LV5: 9 . 3% longer , unpaired t-test , t = 4 . 12 , df = 15 . 135 , p=8 . 87×10−4; LV column length: 7 . 9% longer , unpaired t-test , t = 4 . 34 , df = 15 . 37 , p=5 . 58×10−4 ) . Thus , transitions to queen status induce reproductive investment , which in turn leads to organism-wide allometric changes that generate an elongated phenotype . The elongated phenotype appears to subsequently facilitate future fecundity . Queens with longer bodies ( which correlates with longer LV , Pearson’s r = 0 . 856 , p=5 . 99×10−59; Figure 1—figure supplement 3 ) had more pups per litter ( Figure 1F; β = 0 . 353 , p=1 . 35×10−3 , n = 328 litters from all breeding groups maintained in the same breeding facility; Supplementary file 3 ) . Controlling for litter size , longer queens also had larger pups: for every additional centimeter of maternal body length , pup body mass increased by 2 . 9% ( β = 0 . 29 , p=0 . 032 , n = 971 pups ) . Thus , the elongated queen phenotype is a strong candidate for adaptive plasticity that supports increased fertility in queen mole-rats . To identify the gene regulatory changes associated with skeletal plasticity , we cultured cells enriched for bone marrow-derived mesenchymal stromal cells ( bMSCs ) isolated from the LV ( pooled LV1–LV5 ) of queens and nonbreeders ( n = 5 queens , 11 nonbreeders ) . bMSC cultures include multipotent skeletal stem cells , the precursor of the osteoblast and chondrocyte lineages responsible for bone growth . In parallel , we cultured cells enriched for bMSCs from the pooled long bones ( humerus , ulna , radius , left femur , and left tibia ) of the same animals , which do not show increased elongation in queens ( femur at 12 months: unpaired t-test , t = −0 . 202 , df = 19 . 326 , p=0 . 842; tibia at 12 months: unpaired t-test , t = −0 . 860 , df = 16 . 759 , p=0 . 402 ) . To evaluate the potential role of sex steroid hormone signaling on bone growth , we treated cells from each bone sample for 24 hr with either 10 nM estradiol or vehicle control , resulting in 47 total samples . We then performed RNA-Seq on each sample to screen for genes that were systematically differentially expressed in the bone cells of queens versus nonbreeders . Of 10 , 817 detectably expressed genes , 171 genes showed a significant effect of breeding status at a false discovery rate ( FDR ) threshold of 10% in the long bones ( 329 at an FDR of 20%; Supplementary file 4 ) . Surprisingly , no genes showed a significant effect of breeding status in the LV at either FDR threshold . However , effect sizes were highly correlated between bone types overall ( R2 = 0 . 75 , p=4 . 60×10−53 ) , with more pronounced effects of breeding status in the long bone samples than in the LV ( paired t-test on breeding status effects in long bone versus vertebrae: t = 3 . 97 , df = 317 . 67 , p=8 . 73×10−5 ) . Importantly , breeding status-related differences were not readily attributable to differences in bone cell composition . Based on both canonical markers of bMSC lineage cells and deconvolution of the RNA-Seq data using data from 27 mesenchymal or hematopoietic lineage mouse cell types , the majority cell type in both queen and nonbreeder samples was most similar to cells from the bMSC lineage ( Dominici et al . , 2006; Hume et al . , 2010; Newman et al . , 2015; Figure 2—figure supplements 1 and 2 ) . Additionally , the top three principal components summarizing estimated cell-type proportions did not differ between queens and nonbreeders ( all FDR > 10% , Supplementary file 5 ) , and we identified no cases in which the effects of breeding status on gene expression were significantly mediated by the first principal component of cell composition ( p>0 . 05 for all 171 queen-associated genes at 10% FDR; Supplementary file 6 ) . The majority of breeding status-associated genes were upregulated in queens ( 151 of 171 genes , 88% ) . In support of their role in skeletal plasticity , upregulated genes were enriched for bone remodeling ( log2[OR]=4 . 07 , p=5 . 07×10−6 ) , a process that involves the balanced cycle between bone formation by osteoblasts and bone resorption by osteoclasts ( Redlich and Smolen , 2012; Figure 2 ) . Surprisingly , however , enriched pathways were specifically related to bone resorption , not formation ( Supplementary file 7 ) , including ‘positive regulation of bone resorption’ ( Figure 2A , C; log2[OR]=6 . 51 , p=1 . 55×10−6 ) and ‘superoxide anion generation , ’ which is involved in osteoclast activity and degradation of bone matrix ( Figure 2A , C; log2[OR]=5 . 29 , p=1 . 4×10−5 ) ( Darden et al . , 1996; Datta et al . , 1996; Key et al . , 1990; Key et al . , 1994 ) . Differentially expressed genes were also enriched for immune-related processes ( e . g . , ‘cytokine secretion , ’ ‘chemotaxis , ’ ‘leukocyte activation involved in immune response’; Supplementary file 7 ) . These observations suggest that transitions to queen status also involve changes in immunoregulatory signaling ( osteoclast cells are derived from monocytes ) . Omni-ATAC-seq profiling of open chromatin regions further supports a central role for bone resorption and osteoclast activity in the queen skeleton ( n = 8; Supplementary file 8 ) . Specifically , transcription factor binding motifs ( TFBMs ) located in accessible chromatin near queen upregulated genes were enriched for PU . 1 and MITF , two transcription factors that are essential for osteoclast differentiation ( Segeletz and Hoflack , 2016; Figure 2B , C; PU . 1 log2[OR]=1 . 041 , p=2 . 84×10−4; MITF log2[OR]=0 . 707 , p=7 . 36×10−3; see Supplementary file 8 for a complete list of enriched TFBMs ) . MITF was also among the 151 genes that were differentially expressed between queens and nonbreeders and upregulated in both queen long bones and LV . Surprisingly , given the role of sex steroid hormones in bone growth and elevated estradiol levels in queen versus helper Damaraland mole-rats ( Bennett , 1994 ) , we observed no significant effects of estradiol treatment on gene expression in either bone type ( all FDR > 10% ) . Queen upregulated genes were also not in closer proximity to androgen response elements ( AREs ) or estrogen response elements ( EREs ) than expected by chance ( ARE log2[OR]=0 . 207 , p=0 . 627; ERE log2[OR]=0 . 196 , p=0 . 652 ) . Consistent with this observation , transcription factor footprinting analysis showed no evidence of queen-associated differences in transcription factor activity of the androgen receptor ( AR ) , estrogen receptor 1 ( ESR1 ) , or estrogen receptor 2 ( ESR2 ) , in either the long bones or LV ( all paired t-tests: p>0 . 05; Figure 2—figure supplement 3 ) . Thus , our data point to the involvement of non-sex steroid-mediated signaling pathways in remodeling queen mole-rat bones , at least after 1 year post-transition . The gene expression data suggest that queen status-driven changes to the skeleton extend beyond the LV to the long bones . Further , they suggest that bone resorption—an important counterpoint to bone formation that is required for normal skeletal maintenance—also distinguishes breeding and nonbreeding females . To investigate this possibility , we performed high-resolution micro-computed tomography ( μCT ) scanning to generate 3D reconstructions of LV6 , LV7 , right femur , and right tibia of queens and female nonbreeders ( n = 140 bones from 36 animals; Figure 3A , Figure 3—figure supplement 1 ) . This approach substantially increases the level of resolution for investigating breeding status-linked differences in skeletal morphology , as previous studies relied on x-ray data alone ( O'Riain et al . , 2000; Thorley et al . , 2018 ) . We first asked whether breeding status could be predicted from morphological differences in the 3D reconstructions . We found that it could for the LV , but not for the femur: by applying the smooth Euler characteristic transform ( Crawford et al . , 2016 ) , we were able to predict queen versus nonbreeder status in LV6 ( 77 . 8% accuracy , p=0 . 01 , n = 36 ) , but not the femur ( 52 . 8% accuracy , p=0 . 53 , n = 36 ) . Including only highly fecund queens ( ≥6 total offspring ) improved predictive accuracy in the femur ( 70% accuracy , p=0 . 12 , n = 30 ) . Although these predictions did not reach statistical significance , they raised the possibility that morphological changes in femurs become enhanced with increasing reproductive effort . We next tested whether the transition to queen status affects the ability to continue bone lengthening . Lengthening requires the presence of a growth plate , a region of cartilage in the bone where longitudinal growth occurs through proliferation of cartilage cells ( chondrocytes ) ( Figure 3A , B , Figure 3—figure supplement 1 ) . Closure of the growth plate , which indicates that bone lengthening potential has terminated , typically occurs in mammals after reaching sexual maturation , when energy begins to be invested in reproduction instead of growth ( Kilborn et al . , 2002 ) . To test whether the transition to queen status alters bone lengthening potential , we performed Safranin O staining on sections of the right tibia and LV7 to visualize growth plates ( Figure 3B ) . In the proximal tibia but not LV7 , queens were less likely to have open growth plates ( Figure 3C , Figure 3—figure supplement 1 , and Supplementary file 9; tibia: two-sided binomial test , p=0 . 019; LV7: two-sided binomial test , p=0 . 422 ) . The increased probability of growth plate closure in the tibia of queens is linked to the number of offspring a female has produced: females with more offspring showed a higher expanse of closure across the growth plate ( β = 0 . 050 , p=4 . 51×10−3 , n = 12 , controlling for age ) . This pattern may be due in part to reduced chondrocyte proliferation , as females that produced more offspring had fewer chondrocyte columns in the remaining growth plate ( Figure 3D; β = −0 . 132 , p=0 . 020 , n = 12 , controlling for age ) . Thus , offspring production in queens is associated with loss of the ability to lengthen the long bones , but not the LV , consistent with the importance of abdominal lengthening for supporting larger litters . A major demand on reproductively active female mammals is a high requirement for calcium , particularly during lactation when mothers support rapid offspring bone growth . Maternal skeletons are remodeled to meet this demand , although in most mammals , these changes are not permanent ( reviewed in Kovacs , 2016 ) . Because of the particularly intense reproductive investment made by cooperatively breeding mole-rat queens , we therefore also evaluated the effect of queen status on trabecular and cortical bone volumes , which are thought to be important in satisfying short-term and long-term calcium demands , respectively . We found no effect of queen status on the amount of trabecular bone in the femur , tibia , LV6 , or LV7 ( all p>0 . 05 for bone volume/total volume ) . However , we found that cortical bone was significantly reduced at the femoral midshaft , but not in the LV , in queens compared to their nonbreeding sisters ( Figure 3E , Figure 3—figure supplement 2; femur: paired t-test of cortical area ( CA ) /total area , t = −4 . 067 , df = 8 , p=3 . 60×10−3; LV6: paired t-test of CA/total area , t = −0 . 741 , df = 6 , p=0 . 487 ) . Relative to nonbreeders , the femoral midshafts of queens showed significantly lower apparent density ( paired t-test , t = −3 . 734 , df = 8 , p=5 . 75×10−3 ) , but no difference in material density ( paired t-test , t = −0 . 074 , df = 8 , p=0 . 943 ) . Thus , reduced bone mass in queens is due to reduced bone volume , and not to reduced mass per unit volume ( e . g . , due to increased porosity ) . The reduction of apparent density in queens could be due to slowed bone growth , which typically occurs on the outer , periosteal bone surface , or to increased bone resorption , which typically affects the inner endosteal surface and increases the marrow cavity . Our analysis indicates that cortical bone loss in queens is due to the latter explanation: queens had a larger marrow cavity ( paired t-test , t = 5 . 355 , df = 8 , p=6 . 82×10−4; Figure 3F ) but showed no difference in periosteal area compared to their nonbreeding sisters ( paired t-test , t = 1 . 539 , df = 8 , p=0 . 162; Figure 3G ) . Because changes in cortical bone are thought to reflect accumulated demands over long time frames , we hypothesized that cortical thinning in queens is a consequence of repeated cycles of pregnancy and lactation over time , which can occur simultaneously in Damaraland mole-rat queens . In support of this idea , we found that , within queens , the relative amount of cortical bone is not predicted by the number of pups in a queen’s recent litter ( pups born within the past 30 days; β = −0 . 024 , n = 13 , p=0 . 287 ) , but instead by the total number of pups she produced in her lifetime . Specifically , queens who had more live births had reduced cortical bone thickness along the entire shaft of the femur ( Figure 4 and Supplementary file 10; across decile sections of the femur: all p<0 . 05 , controlling for mother’s litter as a random effect ) . Thus , cortical thinning does not commence with the transition to queen status per se ( i . e . , it is not a correlate of achieving breeder status ) , but instead appears to be a consequence of repeated investment in pregnancy and lactation . Notably , thinning is particularly marked in queens who had at least six offspring , which usually occurs by 14 months after a breeding status transition ( i . e . , by the third litter; Supplementary file 10 ) . Given that wild Damaraland mole-rat queens can maintain their status for many years ( Schmidt et al . , 2013 ) , our results suggest that long-lived queens may experience substantial morphological change ( although the long lives of wild Damaraland mole-rat queens [Schmidt et al . , 2013] suggest these changes must be manageable to some degree , or potentially even recoverable ) . In humans , accelerated bone resorption is a central cause of osteoporosis-related bone fragility ( Szulc et al . , 2006 ) . We therefore hypothesized that cortical thinning in queen mole-rat femurs would be linked to decreased bone strength . To test this hypothesis , we calculated three key indicators of femoral structural integrity: cortical area ( CA ) , the minimum second moment of inertia ( Imin , a correlate of minimum resistance to bending ) , and polar second moment of area ( J , a correlate of torsional rigidity ) . In nonbreeders , the three measures are positively correlated with body mass ( CA: R2 = 0 . 409 , n = 21 , p=1 . 08×10−3; Imin: R2 = 0 . 412 , n = 21 , p=1 . 03×10−3; J: R2 = 0 . 422 , n = 21 , p=8 . 62×10−4 ) . However , in queens , Imin and J are not significantly predicted by body mass ( Imin: R2 = 0 . 088 , n = 13 , p=0 . 17; J: R2 = 0 . 239 , n = 13 , p=0 . 0514 ) , but are instead a function of number of offspring produced ( Imin: R2 = 0 . 283 , p=0 . 0354; J: R2 = 0 . 271 , n = 13 , p=0 . 0393 ) . Queen CA is predicted by both offspring number and body mass , but offspring number explains almost twice the variance ( offspring number R2 = 0 . 634 , p=6 . 83×10−4; body mass R2 = 0 . 385 , p=0 . 014 ) . To evaluate the effects of reproductive activity on the risk of bone failure , we drew on data on the relationship between CA and bone mechanical failure in a large data set of mouse femurs ( Jepsen et al . , 2003 ) . In this data set , CA is the best predictor of maximum load ( the maximum force a bone can withstand prior to failure ) , and , crucially , the CA-max load relationship is highly linear ( Figure 5—figure supplement 1; R2 = 0 . 877 , p=6 . 64×10−38 ) . Scaling the mole-rat CA data to mouse suggests that transitions to queen significantly increase the risk of bone failure ( Figure 5; hazard ratio [95% confidence interval]=2 . 68 ( 1 . 18 , 6 . 08 ) , n = 34 , p=0 . 018 ) . Similar to growth potential and cortical thinning , this effect is driven by highly fertile queens , such that those who had at least six offspring showed the highest predicted risk of bone failure ( Figure 5; queens with ≥6 offspring relative to nonbreeders: HR = 3 . 74 ( 1 . 42 , 9 . 82 ) , n = 28 , p=0 . 007 ) . The risk of bone failure is thus predicted to increase by 21% for each additional pup ( HR = 1 . 21 ( 1 . 10 , 1 . 33 ) , n = 34 , p=1 . 22×10−4 ) .
Our results demonstrate that transitions to breeding status in Damaraland mole-rat queens lead to a cascade of skeletal changes linked to shifts in gene regulation . The vertebral lengthening observed in Damaraland mole-rat queens is concordant with previous reports of vertebral lengthening in both Damaraland mole-rats ( Thorley et al . , 2018 ) and naked mole-rats ( O'Riain et al . , 2000 ) . Like naked mole-rats , our analyses show that most growth is concentrated soon after the breeding status transition , especially in connection with the first post-transition pregnancies ( Dengler-Crish and Catania , 2009; Henry et al . , 2007 ) . However , our findings also suggest subtle differences: for instance , while the growth phenotype in naked mole-rats occurs at the cranial end of the LV ( Henry et al . , 2007 ) , it is concentrated at the caudal end of the vertebral column in Damaraland mole-rats . Given that Damaraland mole-rats and naked mole-rats independently evolved a similar , highly cooperative social structure ( Jarvis and Bennett , 1993; Faulkes and Bennett , 2016 ) , this difference suggests potential convergent evolution of the vertebral lengthening phenotype in queens , presumably in response to the selection pressure for increased fertility . In addition to previously described vertebral growth , we found that queen Damaraland mole-rats lose bone lengthening potential in the long bones and develop thinner femurs that are predicted to be more prone to mechanical failure . Moreover , gene expression levels in queens reflect a signature of bone resorption , rather than bone growth , at the time of sampling , which occurred 1–2 years post-transition . The molecular signature of bone resorption temporally aligns with changes in morphology , in which accelerated vertebral growth primarily occurs during a female’s first pregnancy , whereas cortical thinning in the long bones is a function of repeated cycles of offspring production . Thus , queens quickly progress from traits typically associated with pre-reproductive and pubertal growth in mammals ( e . g . , body elongation ) , to traits typically linked to aging ( e . g . , marrow cavity expansion and cortical thinning ) . The complex pattern of bone growth and bone resorption in queens likely involves multiple regulatory mechanisms . Because estrogen is known to impact bone growth and maintenance ( Cutler , 1997; Khalid and Krum , 2016 ) , and estrogen levels are higher in mole-rat queens relative to nonbreeding females ( Bennett , 1994 ) , we hypothesized that queens and nonbreeders would differ in their response to estradiol in bone marrow-derived cells . Surprisingly , we observed no gene expression response to estradiol treatment . By itself , this result could be a function of the specific concentration or duration of estradiol treatment we applied . However , we also observed no enrichment for estrogen receptor binding motifs near queen upregulated genes , and no evidence that estrogen or androgen receptor binding sites are differentially bound in cells from queens versus nonbreeders . Thus , our results suggest a role for other , as-yet unknown signaling pathways in the queen-specific signature of long bone cortical resorption ( although it does not exclude the possibility that estrogen signaling influences other phenotypes , such as bone elongation , growth plate closure , or collagen organization , which require further study; Juul , 2001; Cake et al . , 2005; Ham et al . , 2002 ) . Bone loss in Damaraland mole-rat queens may be an extreme of the typical mammalian pattern of bone remodeling , in which bone mineral density decreases during pregnancy and lactation , but recovers once offspring are weaned ( Kovacs , 2016 ) . Thinning in mole-rats may be sustained , however , because queens can begin gestating soon after lactating for the previous litter , leaving little to no time for recovery . One possible reason that this fast rate of breeding is achievable is that queens in colonies with more helpers work less and rest more ( Houslay et al . , 2020 ) , consistent with studies in other cooperative species that show that helpers alleviate breeding-associated efforts ( Bales et al . , 2000; Clutton-Brock and Manser , 2016; Crick , 1992; Paquet et al . , 2013; Russell , 2003; Scantlebury et al . , 2002 ) . Paradoxically , helpers might not only help offset costs of , but also contribute to , decreased bone mass in queens , given that large numbers of helpers are themselves produced via high queen fecundity , and reduced physical activity can also lead to decreases in bone mass ( Morseth et al . , 2011 ) . The extent to which helpers reduce the costs of breeding to queens may also differ between species depending on the relative numbers of helpers to breeders . For example , in eusocial insects , large colonies and the high ratio of helpers to queens reduce the costs of reproduction to queens to very low levels ( Wilson , 1971; Keller and Genoud , 1997 ) . Similarly , in naked mole-rats ( where colonies can include hundreds of animals compared to dozens in Damaraland mole-rat colonies; Jarvis , 1981; Jarvis and Bennett , 1993; Jarvis et al . , 1994 ) , a small sample of queens ( n = 6 ) suggests increased rather than decreased femoral cortical thickness relative to age-matched nonbreeders ( Pinto et al . , 2010 ) . Testing how the costs and benefits of reproduction are resolved across different levels of cooperativity , including the molecular mechanisms that mediate these differences , is an important next step towards understanding the evolution of cooperative breeding in mammals . Finally , despite frequent analogies between Damaraland mole-rats and eusocial insects ( Jarvis and Bennett , 1993; Jarvis et al . , 1994; Burda et al . , 2000 ) , our results suggest some key points of differences . Specifically , while abdominal lengthening allows queen mole-rats to increase fecundity per reproductive effort , loss of cortical bone in the femur is unlikely to directly benefit either fertility or survival . Instead , it reflects the cumulative burden of continuous cycles of pregnancy and lactation ( Kovacs , 2016 ) . Thus , unlike eusocial insect queens ( Rodrigues and Flatt , 2016; Rueppell et al . , 2016 ) , Damaraland mole-rat queens incur morphological costs to concentrated reproduction in addition to morphological changes that facilitate increased fitness . How these costs translate into fertility or survival outcomes in natural populations remains a fascinating , unanswered question .
Damaraland mole-rats ( F . damarensis ) were maintained in a captive colony at the Kuruman River Reserve in the Northern Cape Province of South Africa , within the species’ natural range . With the exception of the predictive Euler characteristic transform analysis ( which included two nonbreeding females born in the wild and subsequently maintained in captivity ) , only animals born in captivity , with known birthdates , ages , and litter composition , were used in this study . Animals were maintained in artificial tunnel systems built from PVC pipes with compartments for a nest-box and waste-box and transparent windows to allow behavioral observation ( Zöttl et al . , 2016 ) . Animals were fed ad libitum with sweet potatoes and cucumbers . Adult females ( >1 year ) from 16 natal colonies were randomly assigned to be either nonbreeders or queens , such that females assigned to queen status had age-matched littermates , where possible , who were assigned to the nonbreeding condition . To distinguish the effects of queen status from release from reproductive suppression , nonbreeders were either maintained in their natal colony as helpers or maintained alone , which models the social condition experienced by dispersing females . Females assigned to the breeder condition were transferred to a new tunnel system with an unrelated male from a separate social group . Nine new breeding females , six helpers , and eight solitaire females ( age-matched littermates where possible; Supplementary file 1 ) were set up in December 2015–July 2016 ( Thorley et al . , 2018 ) . With one exception ( animal G10F026 ) , animals maintained their breeding status for 14–22 months before sample collection . One queen and five helpers that were siblings , but not age-matched littermates , of experimental animals were also included in sample collection . To increase the final sample size , an additional four breeding colonies , matched against four age-matched littermate helpers , were formed in October 2017 and followed for 11–12 months ( Supplementary file 1 ) . One queen died before sample collection , and one nonexperimental helper was euthanized during the course of the study and included in sample collection . The final sample size included 13 queens , 15 helpers , and 8 solitaire females . For a subset of study subjects , full body X-rays were taken using the Gierth TR 90/20 battery-operated generator unit with portable Leonardo DR Mini plate ( OR Technology , Rostock , Germany ) every 2 months during the first 12 months of the experiment and at the time of sacrifice . From these X-rays , an experimenter blind to animal breeding status measured the length of each LV ( from vertebra 1 to 7 ) , the right femur , the right tibia , body length , and the width of the zygomatic arch using ImageJ ( Schneider et al . , 2012 ) . The caudal-most LV was labeled as LV7 . We tested for an effect of breeding status on LV5 using a linear mixed model in which post-pairing time point , breeding status , and the interaction of time point by breeding status were modeled as fixed effects and animal ID as a random effect . To test the effect of maternal body length on litter size and pup size , we used body length measurements obtained during routine colony monitoring of all queens maintained in the captive colony ( i . e . , not restricted to experimental animals ) . Following Thorley et al . , 2018 , we used body length measurements obtained nearest to , and no more than 90 days from , the date of parturition . The resulting data set included 328 litters ( 971 pups ) from 76 mothers , which represents a 76% increase over an earlier analysis of this relationship in Thorley et al . , 2018 . We fit two linear mixed effects models . In the first model , we modeled litter size as a function of maternal body length , controlling for whether the litter was the female’s first litter , and included maternal ID as a random effect . In the second model , we modeled pup mass as a function of maternal body length , controlling for litter size and whether the litter was the female’s first litter as fixed effects , and maternal ID and litter ID as random effects . Animals were deeply anesthetized with isoflurane and sacrificed with decapitation following USGS National Wildlife Health Center guidelines and under approval from the Animal Ethics Committee of the University of Pretoria . Immediately upon sacrifice , the LV and long bones were dissected , and attached muscle tissue was removed with forceps . LV6 and 7 and the right femur and tibia were collected into 50% ethanol for 24 hr , then transferred to 70% ethanol and stored at 4° C for μCT scans and histochemistry . To isolate bone cells for culture , LV1–5 were incubated in 2% Collagenase P ( Roche , Switzerland ) for 30 min at 30° C . Each bone was then cut in half and transferred to a 1 . 5 ml microcentrifuge tube containing a G-Tube microcentrifuge tube ( VWR , Radnor , PA , USA ) that had been punctured at the bottom with a 15 gauge needle . Tubes were spun at 3000 RCF for 5 s , allowing the marrow to collect into the 1 . 5 ml microcentrifuge tube . Cell pellets were resuspended in red blood cell lysis buffer , pooled , and incubated for 3 min at room temperature . 10 ml bMSC medium ( MEM-alpha [Sigma-Aldrich , St . Louis , MO , USA] + 15% fetal bovine serum [Hyclone , Logan , UT , USA] + 1% penicillin/streptomycin + 2 ng/ml recombinant human fibroblast growth factor-basic [Biocam , Centurion , Gauteng , South Africa] + 10 nM ROCK inhibitor Y-27632 [RI; Cayman Chemical , Ann Arbor , MI , USA] ) was added to stop lysis , and the tubes were spun for 5 min at 300 RCF . The cell pellet was resuspended in 1 ml bMSC medium and strained through a 70 µm cell strainer . Cells were plated at 1 . 6 × 105 cells per cm2 . The long bones ( excluding right femur and tibia ) were processed to enrich for bMSCs following the same procedure , but without incubation in Collagenase P . Cells were cultured at 37°C and 5% CO2 . 24 hr post plating , plates were carefully washed three times with 1× PBS and supplied with fresh medium to remove nonadherent cells . Once bMSC clusters were visible ( 2–9 days post plating ) , plates were fed bMSC medium without RI or fed bMSC medium without RI +10 nM estradiol ( E2 ) . 24 hr later , cells were collected into buffer RLT and stored at −80°C . Samples were shipped on dry ice to Duke University for RNA extraction using the Qiagen RNeasy Micro Kit . RNA-Seq libraries were generated using the NEBNext Single Cell/Low Input RNA Library Prep Kit for Illumina . RNA-Seq libraries were sequenced on an Illumina HiSeq 4000 ( 100 basepair single-end reads ) to a mean coverage of 16 . 1 ± 3 . 9 s . d . million reads . Reads were trimmed with cutadapt version 2 . 3 ( Martin , 2011 ) ( RRID:SCR_011841; parameters: -q 20 -e 0 . 2 --times 5 --overlap 2 -a AGATCGGAAGAGC -a ‘T’ --minimum-length=20 ) . Trimmed reads were then mapped to the Damaraland mole-rat v1 . 0 genome ( Fang et al . , 2014 ) ( DMR_v1 . 0 ) using two pass mapping with STAR ( RRID:SCR_004463 ) ( Dobin et al . , 2013 ) . Only uniquely mapped reads were retained . HTseq ( RRID:SCR_005514 ) ( Anders et al . , 2015 ) was used to quantify read counts mapping to genes ( using the v1 . 0 . 92 gtf file from Ensembl; we extended the genomic coordinates of the SERPINE1 gene by 2000 basepairs in both directions due to very high expression directly adjacent to the annotated coordinates ) . We transformed read counts to transcripts per million ( TPM ) ( Wagner et al . , 2012 ) and retained only genes with TPM ≥2 in at least 25% of samples . We performed voom normalization ( RRID:SCR_010943 ) ( Law et al . , 2014 ) on the raw counts using normalization factors produced by the trimmed mean of M-values ( TMM ) method ( Robinson and Oshlack , 2010 ) in DESeq ( RRID:SCR_000154 ) ( Anders and Huber , 2010 ) . We used the limma ( Smyth , 2005 ) function lmFit to regress out the proportion of uniquely mapped reads in genes ( which controls for efficiency of mRNA selection during RNA-Seq library preparation ) and animal natal colony ( which controls for littermate sets and date of sacrifice ) to obtain normalized , batch-corrected gene expression values for downstream analysis . We used the mixed effects model approach in emmreml ( Akdemir and Okeke , 2015 ) to estimate , for each gene , the effect of breeding status on gene expression within LV and within long bones using the following model:y1= μ+diβ1+biβ2+qiβ3∗I ( b=0 ) +qiβ4∗I ( b=1 ) +siβ5∗I ( b=0 ) +sβ6∗I ( b=1 ) +Zu+εi , u ~ MVN 0 , σu2Kε ~ MVN 0 , σe2where y is the vector of gene expression levels for n = 47 samples ( indexed by i ) ; μ is the intercept; d is the number of days in culture and β1 its effect size; b is bone type ( i . e . , long bones or LV ) and β2 its effect size; and q is a 0/1 variable representing breeder status and β3 and β4 its effect size in long bones and LV , respectively . I is an indicator variable for bone type ( 0 = long bone; 1 = LV ) . s is a 0/1 variable representing whether the cells were cultured with estradiol and β5 and β6 are its effect sizes in long bones and LV , respectively . Z is an incidence matrix that maps samples to animal ID to take into account repeated sampling from the same individual , and u is a random effect term that controls for relatedness . K is an m by m matrix of pairwise relatedness estimates ( derived from the genotype data , described below ) between all m animals . ε is the residual error , σu2 is the genetic variance component , and σe2 is the environmental variance component . We also ran an identical model but with an additional fixed effect of solitaire status in long bones and in LV , to test for a difference in gene expression between helpers and solitaires . To control for multiple testing , we calculated the FDR following Storey and Tibshirani , 2003 using an empirical null distribution derived from 100 permutations of each variable of interest . We used g:profiler ( RRID:SCR_006809 ) ( Raudvere et al . , 2019 ) to perform Gene Ontology enrichment analysis of the genes upregulated with queen status in LV and long bones ( 151 of 171 genes significantly associated with queen status at a 10% FDR ) . All genes in the original analysis set were used as the background gene set . We set both the minimum size of the functional category and the minimum size of the query/term intersection to 3 . Finally , we retained categories that passed a Bonferroni-corrected p-value of 0 . 05 . To control for relatedness when modeling the gene expression data , we performed single-nucleotide polymorphism ( SNP ) genotyping of the RNA-Seq data using the Genome Analysis Toolkit ( McKenna et al . , 2010 ) ( GATK; RRID:SCR_001876 ) . We used the SplitNCigarReads function on the trimmed , uniquely mapped reads and performed GATK indel realignment . Base recalibration was performed by using all SNPs with GQ ≥4 in an initial UnifiedGenotyper run on the full data set as a reference . Genotypes were called on the recalibrated bam files using HaplotypeCaller . Variants were filtered with the following GATK VariantFiltration parameters: QUAL < 100 . 0 , QD < 2 . 0 , MQ < 35 , FS > 30 , HaplotypeScore > 13 , MQRankSum < −12 . 5 , ReadPosRankSum < −8 . Variants were further filtered with vcftools ( RRID:SCR_001235 ) ( Danecek et al . , 2011 ) to only retain biallelic SNPs in Hardy–Weinberg equilibrium ( p>0 . 05 ) with minor allele frequency ≥ 0 . 1 , minimum mean depth of 5 , max missing count of 2 , and minimum GQ of 99 . Finally , SNPs were thinned to a distance of 10 kb basepairs , resulting in a final data set of 1965 stringently filtered biallelic SNPs . Missing values were imputed using beagle ( RRID:SCR_001789 ) ( Browning and Browning , 2007 ) , and the resulting vcf file was used to create a kinship matrix using vcftools ( Danecek et al . , 2011 ) . Values of the kinship matrix were confirmed to be higher in known siblings compared to non-siblings ( unpaired t-test , t = 27 . 939 , p=2 . 23×10−12; means = 0 . 513 and −0 . 097 ) . Two pairs of siblings were found to have different fathers ( G1F022 and G1F025; G4F020 and G4F019 ) . Although selection for adherent cells from bone marrow enriches for bMSCs , other cell types are also present ( Phinney et al . , 1999 ) . To assess whether cell-type heterogeneity accounts for queen-associated differential expression , we used CIBERSORT ( RRID:SCR_016955 ) to deconvolve the proportion of component cell types from the RNA-Seq data ( Newman et al . , 2015 ) . We trained CIBERSORT on a data set of quantile normalized gene expression values from mouse purified primary cell populations ( Hume et al . , 2010 ) . Specifically , we subset the training data to 27 purified cell populations of mesenchymal or hematopoietic origin ( Figure 2—figure supplement 2 ) and to genes that were included in our mole-rat gene expression data set . We then predicted the composition of the cells that contributed to the mole-rat quantile normalized gene expression data set , for each sample . To test whether cell-type heterogeneity was significantly explained by queen status , we also modeled cell-type proportion ( as summarized by the first principal component of CIBERSORT-estimated proportions for all 27 potential cell types; PC1 explains 50 . 9% of the overall variation ) following the same method used for gene-by-gene expression analysis but with PC1 included as an explanatory variable . We then performed mediation analysis on each of the 171 genes that showed a significant effect of breeding status at FDR < 10% . To do so , we first estimated the indirect effect of breeding status on gene expression through the mediating variable ( CIBERSORT PC1 ) . The indirect effect of breeding status through CIBERSORT PC1 was estimated by calculating the difference in the effect of breeding status between two models: one model that did not include the mediator ( i . e . , β3 and β4 from Equation 1 ) and the same model with the addition of the mediating variable . We performed 1000 iterations of bootstrap resampling to obtain the 95% confidence interval for the indirect effect and considered an indirect effect for a gene significant if the 95% interval did not overlap 0 . To investigate whether differentially expressed genes were associated with accessible binding motifs for specific transcription factors , we generated Omni-ATAC-seq data to profile regions of open chromatin ( Buenrostro et al . , 2013; Corces et al . , 2017 ) . We performed Omni-ATAC-seq on both LV bMSCs and long bone bMSCs from two female nonbreeding and two queen mole-rats ( n = 8 libraries total ) , following the published protocol ( Corces et al . , 2017 ) with the following modifications: 5000 cells were centrifuged at 500 RCF for 5 min at 4°C . The pellet was resuspended in 50 μl transposition mix ( 25 μl 2× TD buffer , 16 . 5 μl PBS , 6 . 75 μl water , 1 μl 10% NP40 , 1 μl 10% Tween-20 , 1 μl 1% digitonin , and 0 . 25 μl Tn5 transposase ) . The reaction was incubated at 37°C for 30 min without mixing , followed by a 1 . 5× Ampure bead cleanup . Omni-ATAC libraries were sequenced on a NovaSeq 6000 as 100 basepair paired-end reads to a mean coverage ( ± SD ) of 26 . 9 ( ± 4 . 4 ) million reads ( range: 16 . 8–38 . 3 ) . Reads were trimmed with Trim Galore ! ( RRID:SCR_011847 ) ( Krueger , 2015 ) to remove adapter sequence and low-quality basepairs ( Phred score <20; reads ≥25 bp ) . Read pairs were mapped to the DMR_v1 . 0 genome using bwa-mem ( RRID:SCR_010910 ) ( Li and Durbin , 2010 ) with default settings . Only uniquely mapped reads were retained . The alignment bam files for each treatment ( breeding or nonbreeding ) were merged , and open chromatin regions were identified using MACS2 v2 . 1 . 2 ( RRID:SCR_013291 ) ( Zhang et al . , 2008 ) with the following parameters: ‘-nomodel -keep-dup all -q 0 . 05’ . We combined open chromatin peaks with regions in the DMRv1 . 0 genome that match sequences of vertebrate transcription factor binding site motifs , using motifs defined in the HOMER database ( RRID:SCR_010881 ) ( Heinz et al . , 2010 ) . We used Fisher’s exact tests ( using a p-value threshold of 0 . 01 ) to test if TFBMs belonging to the same transcription factor were enriched in open chromatin regions within 2000 bp of the 5′ most transcription start site of queen upregulated genes . To compare genome-wide signatures of DNA-transcription factor binding for AR , ESR1 , and ESR2 , we characterized transcription factor footprints in queens and nonbreeders , in both the LV and long bones , using HINT-ATAC from the Regulatory Genomics Toolbox ( RGT ) with default parameters ( Li et al . , 2019 ) . We focused on the subset of peak regions called using MACS2 ( Zhang et al . , 2008 ) . We identified TF footprints by merging reads within each bone type-breeding status combination and calling footprints on the combined data . For each bone type , we then created a meta-footprint set by merging the respective footprint calls across queens and nonbreeders using the bedtools function merge ( Quinlan and Hall , 2010 ) . We identified transcription factor motifs in the DMR_v1 . 0 genome that fell within meta-footprints , based on the JASPAR CORE Vertebrates set of curated position frequency matrices ( Sandelin et al . , 2004 ) . Finally , we tested for differential footprints of AR , ESR1 , and ESR2 binding using the RGT differential function , using the activity score metric described in Li et al . , 2019 and default parameters . We performed μCT scans of LV6 , LV7 , right femur , and right tibia using a VivaCT 80 scanner ( Scanco Medical AG , Brüttisellen , Switzerland ) set at 55 kVp and 145 μA , with voxel size 10 . 4 μm . Trabecular bone was quantified using direct values ( i . e . , ‘No model’ ) from the 100 μCT slices below the proximal tibia growth plate , the 100 μCT slices above the distal femur growth plate , and the 100 μCT slices medial to the caudal growth plate of LV6 . To obtain midshaft cross-sections of the femur , tibia , and LV6 , we first reduced each bone mesh to 100 , 000 faces using Avizo Lite version 9 . 7 . 0 . Mesh files from the same bone type were auto-aligned using Auto3dgm ( Boyer et al . , 2015 ) in MATLAB ( RRID:SCR_001622 ) . Aligned mesh files were then back scaled to their original sizes in MATLAB , and the midshaft cross-section was generated using Rhinoceros version 6 . LV6 cross-sections required manual segmentation , which was performed in Adobe Illustrator CC version 23 . 0 . 2 . The MomentMacro plugin in ImageJ ( RRID:SCR_003070 ) was used to calculate bone area , minimum second moment of inertia , and polar second moment of area . To predict breeding status from bone shape , we reduced each bone mesh to 100 , 000 faces using Avizo Lite version 9 . 7 . 0 ( RRID:SCR_014431 ) . Mesh files from the same bone type were auto-aligned using Auto3dgm ( Boyer et al . , 2015 ) in MATLAB . The resulting aligned and scaled mesh files were used as input to perform the unsmooth Euler characteristic transform ( ECT ) for LV6 and to perform the smooth Euler characteristic transform ( SECT ) for femurs ( Crawford et al . , 2016 ) ( as performance was optimized for femurs by including smoothing ) . The ECT represents each shape as a two-parameter function of direction and height . We evaluate these functions at a discrete lattice of points , so that each point is represented as a vector . For each bone type , we calculated the Euler characteristic curve in 162 directions distributed approximately uniformly across the shape . For each of the 162 Euler characteristic curves , we ran 100 height parameter evaluations for LV6 and an average of 71 height parameter evaluations for the femur , so that each LV6 ( femur ) was represented with a vector of length 16 , 200 ( 11 , 423 ) . Using the ECT data , we performed leave-one-out predictions , running each bone type separately , using the linear kernel and c-classification with the support vector machine ( SVM ) implemented by the R package e1071 ( Dimitriadou et al . , 2008 ) . The SVM classifier was equipped with 1:100 , 000 weighting to balance for the different number of breeders and nonbreeders in the sample . The empirical p-values were estimated for each bone type by running 100 permutations of the queen/nonbreeder labels ( Golland et al . , 2005 ) . We note that the ECT does not provide information about specific regions of a shape , but rather is a topological summary statistic that captures the geometric complexity of a shape by quantifying the shape’s number of connected components , voids , and closed loops . This approach avoids the need for landmarks , which may be missing or vulnerable to observer error in some data sets ( Crawford et al . , 2016 ) . The ECT is amenable to regression models and is therefore useful for testing whether shape significantly predicts the value of an outcome variable ( here , breeding status ) . However , because it does not identify the regions of the shape that contribute most to predictive power , the regions of the LV and femur that most strongly differentiate queens from nonbreeders remain open to further study . For a subset of individuals ( Supplementary file 9 ) , the tibia and LV7 were plasticized , sectioned , and stained with Safranin O by the Washington University Musculoskeletal Research Center . The proportion of the tibia proximal growth plate that was fused , and the mean proportion of the LV7 cranial and caudal growth plates that were fused , were measured in ImageJ from Safranin O-stained sections . To quantify growth plate activity , we calculated the number of chondrocyte columns ( defined as linear stacks of at least three chondrocytes ) controlling for length of open growth plate . For each bone type ( tibia and LV7 ) , we ran two models: proportion of growth plate fusion or chondrocyte columns per mm growth plate as the dependent variable , and number of offspring born and age as the independent variables . We used Stradview ( Treece , 2019; Treece et al . , 2010 ) on dicom images from the μCT scans to measure and visualize , in an automated manner , cortical thickness across the surface of the femur . Bone surface was defined in Stradview by thresholding pixel intensity and contouring the bone at every 14 sections , with the following parameters: resolution = medium , smoothing = standard , strength = very low , contour accuracy = 6 . To measure cortical thickness , we used the auto threshold method in Stradview , with line width set to 5 , smooth set to 1 , and line length set to 3 mm . The smoothed thickness values of each femur were then registered ( i . e . , mapped ) to a single ‘canonical’ femur surface ( mole-rat GRF002 ) using wxRegSurf v18 ( http://mi . eng . cam . ac . uk/∼ahg/wxRegSurf/ ) . We sectioned the cortical thickness values into deciles according to location along the length of the femur . The top and bottom deciles were removed because cortical and trabecular bone towards the ends of the femur could not be easily differentiated by the automated method . Deciles were then recreated for the remaining length of the bone ( i . e . , the central 80% ) . From each bone decile , we estimated cortical thickness as the mean of all cortical thickness measures within that interval . For each decile across animals , we used a linear mixed model to model cortical thickness as a function of breeding status and number of offspring , with litter pair as a random effect . Previous research on mechanical properties of rodent femurs found that , among several morphological and compositional traits measured in eight morphologically varying mouse strains , CA at the midshaft was the best predictor of maximum load ( defined as the greatest force attained prior to bone failure , measured via four-point bending; published Pearson’s r = 0 . 95 ) ( Jepsen et al . , 2003 ) . We therefore used CA at the femoral midshaft to predict max load of Damaraland mole-rat femurs . To do so , we first fit a linear model of max load as a function of CA ( unadjusted for body weight ) using published mouse data ( R2 = 0 . 877 , n = 81 , p=6 . 64×10−38 ) ( Jepsen et al . , 2003 ) . We extrapolated from this linear fit to predict max load from CA at the midshaft of Damaraland mole-rat femurs . Predicted max loads were then used as input for Cox proportional hazards models , comparing either all queens to nonbreeders or queens with ≥6 offspring to nonbreeders . Models were fit using the R function coxph and were confirmed to meet the proportional hazards assumption using the cox . zph function in the R package survival ( Therneau , 2020 ) . Because max load was not directly measured in the Damaraland mole-rats , we used the Cox proportional hazards models to specifically evaluate the relative hazard of bone failure depending on queen status/number of offspring . We therefore report the results in Figure 5 based on relative force ( with the median predicted failure value for nonbreeders set to 1 ) instead of absolute force in Newtons . We note that the analysis makes an important assumption that the linear relationship between CA and maximum load observed in mice is shared with Damaraland mole-rats . Future mechanical loading tests of mole-rat bones , which were not possible in this study , will therefore be important for validating and refining the present modeling results . | Some social animals are highly cooperative creatures that live in tight-knit colonies . Bees and ants are perhaps the most well-known examples of social insects , while Damaraland mole-rats and naked mole-rats , two rodent species found in southern and eastern Africa , are among the most cooperative mammal species . In these colony-forming animals , only one or a few females reproduce and these fertile females are frequently referred to as “queens” . When an animal becomes a queen , her body shape can change dramatically to support the demands of high fertility and frequent reproduction . The molecular basis of such changes has been well-described in social insects . However , they are poorly understood in mammals . To address this knowledge gap , Johnston et al . studied how transitioning to queen status affects bone growth and structural integrity in Damaraland mole-rats , as well as body shape and size . The experiments compared non-breeding female mole-rats with other adult females recently paired with a male to become the sole breeder of a new colony . Johnston et al . also used bone-derived cells grown in the laboratory to assess underlying gene regulatory changes in new queen mole-rats . Johnston et al . showed that transitioning to the role of queen leads to a cascade of skeletal changes accompanied by shifts in the regulation of genetic pathways linked to bone growth . Queen mole-rats show accelerated growth in the spinal column of their lower back . These bones are called lumbar vertebrae and this likely allows them to have larger litters . However , queen mole-rats also lose bone growth potential in their leg bones and develop thinner thigh bones , which may increase the risk of bone fracture . Therefore , unlike highly social insects , mole-rats do not seem to have escaped the physical costs of intensive reproduction . This work adds to our understanding of the genes and physical traits that have evolved to support cooperative behaviour in social animals , including differences between insects and mammals . It also shows , with a striking example , how an animal’s genome responds to social cues to produce a diverse range of traits that reflect their designated social role . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"evolutionary",
"biology"
] | 2021 | Morphological and genomic shifts in mole-rat ‘queens’ increase fecundity but reduce skeletal integrity |
Microtubules are dynamic polymers that in cells can grow , shrink or pause , but the factors that promote pausing are poorly understood . Here , we show that the mammalian kinesin-4 KIF21B is a processive motor that can accumulate at microtubule plus ends and induce pausing . A few KIF21B molecules are sufficient to induce strong growth inhibition of a microtubule plus end in vitro . This property depends on non-motor microtubule-binding domains located in the stalk region and the C-terminal WD40 domain . The WD40-containing KIF21B tail displays preference for a GTP-type over a GDP-type microtubule lattice and contributes to the interaction of KIF21B with microtubule plus ends . KIF21B also contains a motor-inhibiting domain that does not fully block the interaction of the protein with microtubules , but rather enhances its pause-inducing activity by preventing KIF21B detachment from microtubule tips . Thus , KIF21B combines microtubule-binding and regulatory activities that together constitute an autonomous microtubule pausing factor .
The organization and function of microtubule ( MT ) networks critically depend on the dynamic instability of MTs – their ability to spontaneously switch between phases of growth and shrinkage ( Desai and Mitchison , 1997 ) . This MT behavior can be reconstituted in vitro using purified tubulin . In cells , numerous MT-associated proteins ( MAPs ) modulate the dynamic instability of MTs by controlling specific phases of MT dynamics . MAPs can accelerate MT polymerization , decorate and stabilize MTs , promote switching between growth and shortening ( catastrophes ) , or induce reverse transitions ( rescues ) . Many of these activities have been reconstituted in vitro in systems with purified components ( Akhmanova and Steinmetz , 2015; Gardner et al . , 2013 ) . Importantly , the plus ends of MTs growing from purified tubulin in vitro typically undergo sharp transitions between growth and shortening , while in cells MT plus ends often exist in a paused state . This difference is due to the presence of cellular factors that can dampen or even block MT dynamics , but the nature of these factors and the molecular mechanisms underlying their activity are still poorly understood . Proteins controlling MT dynamics can be broadly divided into molecular motors and MAPs that lack motor activity . The two types of MT-dependent motors , kinesins and dyneins , can both interact with MT ends to affect their dynamics ( Hu et al . , 2015; Laan et al . , 2012; Su et al . , 2012; Walczak et al . , 2013 ) . Amongst the kinesins , very different modes of regulation of MT polymerization have been reported . For example , the kinesin-13 family members have a centrally located motor domain , are immotile and use the energy of ATP hydrolysis to modify the structure of MT ends , induce catastrophes and enhance depolymerization ( Moores and Milligan , 2006; Walczak et al . , 2013 ) . Kinesin-8 family members have an N-terminal motor domain and can move processively to the plus ends where they induce MT disassembly or suppress MT dynamics ( Gardner et al . , 2008; Stumpff et al . , 2011; Su et al . , 2012 ) . Another family of MT-regulating kinesins is kinesin-4 . The best-studied family member , KIF4/Xklp1 , reduces the MT growth rate and suppresses catastrophes ( Bieling et al . , 2010; Bringmann et al . , 2004 ) . During mitosis , KIF4/Xklp1 binds to PRC1 , a potent anti-parallel MT bundler involved in the formation of the central spindle ( Kurasawa et al . , 2004; Zhu and Jiang , 2005 ) . The complex of KIF4/Xklp1 and PRC1 accumulates at MT ends and strongly inhibits MT elongation ( Bieling et al . , 2010; Subramanian et al . , 2013 ) . Another kinesin-4 family member , KIF7 , is immotile; it participates in organizing the tips of ciliary MTs by reducing the MT growth rate and promoting catastrophes ( He et al . , 2014 ) . Other members of the kinesin-4 family are the two large motors KIF21A and KIF21B . KIF21A has been studied quite extensively , because point mutations in this protein cause a dominant eye movement syndrome , Congenital Fibrosis of the Extraocular Muscles type 1 ( CFEOM1 ) ( Heidary et al . , 2008; Yamada et al . , 2003 ) . KIF21A is ubiquitously expressed , but the pathology in patients is associated with a specific defect in the development of the oculomotor nerve , likely due to a perturbation of axon guidance ( Cheng et al . , 2014; Heidary et al . , 2008 ) . In vitro , the KIF21A motor domain behaves similarly to that of Xklp1 – it reduces the MT growth rate and suppresses catastrophes ( van der Vaart et al . , 2013 ) . There are also indications that in addition to controlling MT dynamics , KIF21A plays a role in membrane transport ( Lee et al . , 2012 ) . All CFEOM1-associated mutations in KIF21A localize either to the motor domain or to a predicted short coiled-coil domain in the stalk region of the molecule , and each of them prevents the autoinhibitory interaction between these two elements ( Bianchi et al . , 2016; Cheng et al . , 2014; van der Vaart et al . , 2013 ) . The dominant character of the CFEOM1 syndrome is thus connected to the increased activity of the mutant KIF21A kinesin caused by the loss of autoinhibition ( Cheng et al . , 2014; van der Vaart et al . , 2013 ) . KIF21A and KIF21B are highly similar in sequence: they both contain an N-terminal motor domain followed by a stalk with several predicted coiled coils and a C-terminal WD40 domain ( Marszalek et al . , 1999 ) . KIF21B has been reported to be expressed in brain , eye and spleen and to be enriched in dendrites of neurons ( Marszalek et al . , 1999 ) . Polymorphisms in the KIF21B gene have been associated with multiple sclerosis and other inflammatory disorders ( Anderson et al . , 2009; Barrett et al . , 2008; Goris et al . , 2010; Yang et al . , 2015 ) . An increase in expression of KIF21B was connected to accelerated progression of neurodegenerative diseases ( Kreft et al . , 2014 ) , and microduplications of the locus bearing the KIF21B gene were linked to neurodevelopmental abnormalities ( Olson et al . , 2012 ) . Furthermore , it has been demonstrated that KIF21B binds to the ubiquitin E3 ligase TRIM3 , which can modulate the function of KIF21B ( Labonté et al . , 2013 ) . The motor was also implicated in the surface delivery of GABAA receptors in neurons , but the interaction is likely indirect ( Labonté et al . , 2014 ) . While this manuscript was in preparation , a paper describing a mouse knockout of KIF21B has been published ( Muhia et al . , 2016 ) . This work showed that mice lacking KIF21B are viable , but display defects in learning and memory , which are likely to be due to several dendritic phenotypes , such as reduced complexity of the dendritic arbor and diminished density of dendritic spines that correlate with defects in synaptic transmission . An even more recent paper showed that KIF21B contributes to activity-dependent regulation of some aspects of retrograde trafficking of brain-derived neurotrophic factor-TrkB complexes in cultured neurons ( Ghiretti et al . , 2016 ) . Both papers showed that KIF21B can affect MT plus-end dynamics , although the results were complex: while both studies reported an increase in MT growth processivity upon KIF21B loss , MT grew slower in Kif21b knockout neurons , but faster in neurons depleted of KIF21B by RNA interference ( Ghiretti et al . , 2016; Muhia et al . , 2016 ) . In vitro reconstitution work suggested that KIF21B increases MT growth rate and catastrophe frequency , although , surprisingly , the purified protein mostly associated with depolymerizing MT plus ends in these experiments ( Ghiretti et al . , 2016 ) . Here , we have used in vitro single molecule assays to systematically explore how the biochemical activity of KIF21B depends on its domain architecture . We found that KIF21B is a processive kinesin that walks to and accumulates at MT plus ends . The dimeric KIF21B motor domain was sufficient to reduce MT growth rate , while the full-length molecule could ‘hold on’ to the growing MT tip and induce its pausing . Strikingly , a few KIF21B molecules were sufficient to trigger and sustain a pause . In cases when KIF21B persisted at the MT tip but did not induce pausing , MT growth perturbation and catastrophes were observed . The potent effect of KIF21B on MT plus-end polymerization is due to the presence of two MT-binding regions in its tail , which help to prevent kinesin dissociation from the tip of the growing MT . We also found that the region responsible for autoinhibition in KIF21A ( Bianchi et al . , 2016 ) is conserved in KIF21B . However , instead of blocking the motor , this element reduced motor detachment from growing MT plus ends and thus contributed to MT pause induction . Taken together , our data show how the interplay between the motor domain and MT-binding and regulatory regions makes KIF21B a highly potent regulator of MT plus-end dynamics .
To get insight into the ability of KIF21B to regulate MT dynamics , we have expressed the full-length protein with a C-terminal GFP tag in COS-7 cells , which do not express endogenous KIF21B . Unlike its paralogue KIF21A , which is largely diffuse when expressed in similar conditions ( van der Vaart et al . , 2013 ) , KIF21B bound to MTs and accumulated at their ends at the cell periphery ( Figure 1A ) . Live cell imaging showed that KIF21B processively moves along MTs with an average speed of 0 . 63 ± 0 . 22 µm/s ( mean±SD ) ( Figure 1B ) ; this velocity is three times faster than that recently described for HaloTag-labeled KIF21B in neurons ( Ghiretti et al . , 2016 ) . In internal cell regions , where no clear accumulation of the motor at growing MT ends was observed , the expression of KIF21B led to a ~1 . 5 fold reduction in the MT growth rate measured with the MT plus-end marker EB3-TagRFP-T ( Stepanova et al . , 2003; van der Vaart et al . , 2013 ) ( Figure 1C ) . At the cell periphery , strong accumulation of KIF21B-GFP and stalling of MT growth were observed; however , the exact quantification of MT dynamics at the periphery of KIF21B-overexpressing cells was severely complicated by the frequent sliding of MT tips against each other . Interestingly , in cells with high expression levels of KIF21B , the MT network often strongly retracted , leaving significant portions of the cytoplasm largely devoid of MTs ( Figure 1D ) . The remaining MT network in such cells was still dense and appeared to be ‘corralled’ by KIF21B accumulations . Time lapse imaging showed that the retraction of the MT network in KIF21B-expressing cells was a gradual process that could be detected during 1–2 hr of observation ( Figure 1—figure supplement 1A ) . In addition , expression of KIF21B prevented full extension of MTs in experiments where the MT network recovered from treatment with the MT-depolymerizing drug nocodazole ( Figure 1—figure supplement 1B ) . We conclude that at high expression levels , KIF21B can accumulate at MT plus ends , block their polymerization and cause their very slow shortening ( Figure 1—figure supplement 1A ) . 10 . 7554/eLife . 24746 . 003Figure 1 . KIF21B inhibits MT growth in cells . ( A ) COS-7 cells were transiently transfected with KIF21B-FL-GFP and EB3-TagRFP-T and imaged using TIRF microscopy . Represented are a single-frame , maximum intensity projection of 500 frames for the GFP channel and an overlay of a single GFP frame in green and TagRFP-T in red . Kymographs illustrate the motility of KIF21B along the MT and its significant accumulation at a stationary but not a growing MT plus end . ( B ) Histogram of KIF21B-FL-GFP kinesin velocities in COS-7 cells is shown with black bars . Red line shows fitting with a normal distribution . n = 378 in 10 cells in two independent experiments . ( C ) Quantification of MT growth rate , measured by tracking EB3 labeled comets in cell interior . Three to ten MTs per cell were analyzed; n = 183 in 21 cells for GFP control , n = 214 in 12 cells for KIF21B-FL-GFP expressing cells , two independent experiments , p<0 . 0001 , Mann-Whitney U test ( indicated by an asterisk ) . ( D ) COS-7 cells were transiently transfected with KIF21B-FL-GFP , fixed the next day and stained for α-tubulin . Cell edges are indicated with yellow dashed lines in the overlay . ( E ) Histogram of KIF21B-MD-CC1-GFP velocities in COS-7 cells is shown with black bars . Red line shows fitting with a normal distribution . n = 431 in 14 cells in two independent experiments . ( F ) COS-7 cells transiently transfected with KIF21B-MD-CC1-GFP were fixed and stained for α-tubulin . ( G ) COS-7 cells were transiently transfected with KIF21B-MD-CC1-GFP and EB3-TagRFP-T and imaged using TIRF microscopy . Represented are a single-frame , maximum intensity projection of 500 frames for the GFP channel , an overlay of a single GFP frame in green and TagRFP-T in red and a kymograph along one of the EB3-labeled MTs showing the motility of the kinesin along the MT . ( H ) Kymographs showing EB3-TagRFP-T comet displacement in control COS-7 cells or cells expressing the MD-CC1 fragments of KIF21A or KIF21B . ( I ) Quantification of MT growth rate illustrated in H . n = 183 in 21 cells for GFP control , n = 136 in 15 cells for KIF21A-MD-CC1-GFP , n = 179 in 22 cells for KIF21B-MD-CC1-GFP , two independent experiments , p<0 . 0001 , Mann-Whitney U test ( indicated by asterisks ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 00310 . 7554/eLife . 24746 . 004Figure 1—source data 1 . An excel sheet with numerical data on the quantification of kinesin velocities and MT growth rate in COS-7 cells represented as plots in Figure 1B , C , E , I . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 00410 . 7554/eLife . 24746 . 005Figure 1—figure supplement 1 . Effects of KIF21B expression on MT organization and regrowth in cells ( A ) Time-lapse imaging of transiently transfected COS-7 cells expressing KIF21B-FL-GFP and TagRFP-tubulin . Yellow dashed lines in the overlay indicate the cell edge . ( B ) Nocodazole washout experiments of COS-7 cells expressing GFP or KIF21B-FL-GFP . Cells were transiently transfected with the indicated proteins and treated with 5 μM nocodazole for 2 hr . Subsequently , nocodazole was washed out and cells were fixed at the indicated time points . Antibodies against α-tubulin were used for cell staining . Yellow dashed lines in the overlay indicate the cell edge . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 005 To investigate whether blocking of MT growth could be caused by the motor domain of KIF21B alone , we used a C-terminally tagged truncated version encompassing the motor and a part of the dimeric coiled coil , but missing the tail region of the protein ( KIF21B-MD-CC1; see Figure 6A for the scheme of all constructs used in this study ) . KIF21B-MD-CC1 also bound to and walked along MTs with a velocity of 1 . 23 ± 0 . 27 µm/s ( mean±SD ) ( Figure 1E ) , but did not specifically accumulate at MT plus ends ( Figure 1F , G ) . The observed velocity was again ~3 times faster than that observed in neurons ( Ghiretti et al . , 2016 ) , which might be due to the fact that in neuronal cells kinesins are slowed down by specific MAPs . Expression of KIF21B-MD-CC1 reduced MT growth rate by ~1 . 6 fold ( Figure 1H , I ) , which is similar to what we previously observed with a comparable deletion mutant of KIF21A ( van der Vaart et al . , 2013 ) . To investigate whether the observed effect of KIF21B-MD-CC1 is direct , we next purified GFP alone and KIF21B-MD-CC1 , which was C-terminally tagged with GFP , from HEK293T cells ( Figure 2—figure supplement 1 ) . Using mass spectrometry , we confirmed that this purification method did not result in co-isolation of known MT regulators ( Supplementary file 1 ) . Analysis of fluorescence intensity of single KIF21B-MD-CC1-GFP molecules in comparison to monomeric GFP and dimeric EB3-GFP indicated that they were dimers , as expected ( Figure 2A , Supplementary file 2 ) . This conclusion was confirmed by two-step photobleaching profiles ( Figure 2B ) and was in agreement with the published data obtained in HeLa cell lysates with a similar construct ( Ghiretti et al . , 2016 ) . 10 . 7554/eLife . 24746 . 006Figure 2 . Dimeric motor domain of KIF21B slows down MT polymerization in vitro . ( A ) Histograms of fluorescence intensities at the initial moment of observation of single molecules of the indicated proteins immobilized on coverslips ( symbols ) and the corresponding fits with lognormal distributions ( lines ) . n = 3107 , 5802 and 4674 molecules and fluorophore density was 0 . 15 , 0 . 28 and 0 . 23 µm−2 for GFP , GFP-EB3 and KIF21B-MD-CC1-GFP proteins . ( B ) Representative photobleaching time traces of GFP , GFP-EB3 and KIF21B-MD-CC1-GFP individual molecules ( background subtracted ) . ( C ) Kymographs illustrating the dynamics of MTs grown in vitro in the presence of 20 nM mCherry-EB3 alone , with 10 nM purified GFP or with 2 and 10 nM KIF21B-MD-CC1-GFP . Zooms of the boxed areas are shown on the right . Kymographs were generated from movies acquired using a Photometrics Evolve 512 EMCCD camera ( Roper Scientific ) ( stream acquisition , exposure time 500 ms ) . ( D ) Histograms of fluorescence intensities of single GFP molecules immobilized on coverslips and KIF21B-MD-CC1-GFP moving on MTs in a separate chamber on the same coverslip ( symbols ) and the corresponding fits with lognormal distributions ( lines ) . n = 4815 and 1381 molecules; fluorophore density was 0 . 16 and 0 . 09 µm−2 for GFP and KIF21B-MD-CC1-GFP proteins ( for the latter , MT-containing regions were manually selected for analysis ) . Dashed lines show corresponding relative median values . ( E ) Histogram of KIF21B-MD-CC1-GFP velocities in vitro is shown with black bars . Red line shows fitting with a normal distribution . n = 675 in two independent experiments . ( F ) Histogram of KIF21B-MD-CC1-GFP run lengths in vitro is shown with black bars . Red line shows fitting with an exponential distribution . n = 675 in two independent experiments . ( G ) Upper panel - quantification of the MT growth rate illustrated in C . n = 71 for control , n = 65 for purified GFP , n = 71 , 67 and 54 for 2 , 5 and 10 nM KIF21B-MD-CC1-GFP , respectively . Lower panel shows quantification of the MT growth rate with 15 µM tubulin alone or with 10 nM purified GFP or with 2 , 5 and 10 nM KIF21B-MD-CC1-GFP as illustrated in Figure 2—figure supplement 2 . n = 67 for control , n = 57 for purified GFP , n = 71 , 66 and 80 for 2 , 5 and 10 nM KIF21B-MD-CC1-GFP , respectively , two independent experiments , p<0 . 0001 , Mann-Whitney U test ( indicated by asterisks ) . ( H ) Quantification of the MT growth rate with different concentrations of tubulin along with 20 nM EB3 in the absence and presence of 2 nM KIF21B-MD-CC1-GFP as illustrated in Figure 2—figure supplement 3 . ND; not determined , n = 71 for all conditions . two independent experiments , p<0 . 0001 , Mann-Whitney U test ( indicated by asterisks ) . ( I ) Kymographs illustrating the dynamics of MTs grown in vitro in the presence of 20 nM EB3 and 1 nM KIF21B-MD-CC1-GFP . Zoom of the boxed area is shown on the right . Kymographs were generated from a movie acquired using Photometrics Evolve 512 EMCCD camera ( Roper Scientific ) ( stream acquisition , exposure time 100 ms ) . ( J ) Distribution of EB3 fluorescence intensity fluctuations over time ( normalized to its maximum value during a course of a growth event ) at MT tip in the presence of 1 nM GFP or 1 nM KIF21B-MD-CC1-GFP ( solid line ) with Gaussian fit ( dotted line ) . n = 25 in both cases . Thick dotted lines show the peak of the Gaussian fitting . MT dynamics assay was performed in the presence of 15 µM tubulin , 20 nM EB3 and 1 nM GFP or 1 nM KIF21B-MD-CC1-GFP in the separate chambers of the same coverslip . ( I ) Plot of the mean and SD values of Gaussian fits shown in Figure 2J . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 00610 . 7554/eLife . 24746 . 007Figure 2—source data 1 . An excel sheet with numerical data on the quantification of KIF21B-MD-CC1-GFP dimer analysis , photobleaching-step analysis , velocities , run length , effects on MT growth rate and distribution of EB3 fluorescence intensity represented as plots in Figures 2A , B , D , E–H , J . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 00710 . 7554/eLife . 24746 . 008Figure 2—figure supplement 1 . Coomassie blue stained gels with purified GFP , KIF21B-FL-GFP and its deletion mutants . Protein purification was performed using TEV protease cleavage as described in the Materials and Methods section . Black arrows indicate isolated proteins; blue arrows indicate the TEV protease . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 00810 . 7554/eLife . 24746 . 009Figure 2—figure supplement 2 . Kymographs illustrating in vitro dynamics of MTs grown in the presence of 15 µM tubulin in the absence and presence of 10 nM purified GFP or 2 , 5 and 10 nM KIF21B-MD-CC1-GFP . Kymographs were generated from the movies of 600 frames ( stream acquisition , exposure time 500 ms ) using Photometrics Evolve 512 EMCCD camera ( Roper Scientific ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 00910 . 7554/eLife . 24746 . 010Figure 2—figure supplement 3 . Effects of the dimeric motor domain of KIF21B on MT polymerization in vitro ( A ) Kymographs illustrating the in vitro dynamics of MTs grown in the presence of different concentrations of tubulin along with 20 nM EB3 in the absence and presence of 2 nM KIF21B-MD-CC1-GFP . Kymographs were generated from the movies of 600 frames ( stream acquisition , exposure time 500 ms ) using Photometrics Evolve 512 EMCCD camera ( Roper Scientific ) . ( B ) Quantification of the velocity of MT minus end growth in the presence of 15 µM tubulin and 20 nM mCherry-EB3 without ( control ) or with 10 nM KIF21B-MD-CC1-GFP . n = 50 in both cases . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 01010 . 7554/eLife . 24746 . 011Figure 2—Figure Supplement 3—Source Data 1 . An excel sheet with numerical data on the quantification of the MT minus end growth rates represented as plot in Figure 2—figure supplement 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 011 We then examined the effect of KIF21B-MD-CC1 on dynamic MTs in vitro by using a Total Internal Reflection Fluorescence ( TIRF ) microscopy-based MT polymerization assay ( Bieling et al . , 2007; van der Vaart et al . , 2013 ) . In this assay , MTs are grown from GMPCPP-stabilized MT seeds attached to glass coverslips in the presence of fluorescently labeled or unlabeled tubulin and proteins of interest . We performed such assays in the presence of fluorescently labeled tubulin alone or with the addition of mCherry-EB3 ( Montenegro Gouveia et al . , 2010 ) , as this fluorescent protein greatly facilitates the detection of small changes in the position of the growing MT plus end , and our previous work showed that it did not alter the effect of KIF21A on the MT plus-end dynamics ( van der Vaart et al . , 2013 ) . Moreover , since EB proteins are highly abundant and ubiquitous MT plus-end binding proteins , EB-bound MT plus ends can be expected to represent ‘natural’ substrates for other MT regulators . KIF21B-MD-CC1 displayed short plus end-directed runs on MTs and could reach MT plus ends but detached from them upon arrival and thus did not accumulate at the MT tips ( Figure 2C ) . The intensity of individual KIF21B-MD-CC1 molecules moving on MTs was on average ~1 . 8 times higher than the intensity of individual monomeric GFP molecules immobilized in a separate chamber on the same coverslip ( Figure 2D , Supplementary file 2 ) . While a ratio of two might be expected for a dimer , we need to take into account that the motors are further away from the coverslip and that the evanescent field used for excitation decays exponentially . Given a penetration depth d of 80–200 nm , being 25 nm ( MT diameter ) away from the coverslip will yield a 12–27% reduction in intensity ( i . e . e-25/d ) ( Grigoriev and Akhmanova , 2010 ) . We further note that the intensity distribution of KIF21B-MD-CC1 molecules walking on MTs lacked the tail in the high-intensity range that was observed for molecules immobilized on glass ( compare Figure 2A and D ) , suggesting that larger KIF21B-MD-CC1 oligomers present in our preparations are unable to move on MTs . Single KIF21B-MD-CC1 molecules displayed an average velocity of 0 . 6 ± 0 . 3 µm/s ( mean and SD ) and an average run length of 0 . 34 ± 0 . 01 µm ( exponential fit to histogram and error of fit ) ( Figure 2E , F ) . KIF21B-MD-CC1 caused a concentration-dependent reduction of the MT plus-end growth rate both in the absence and in the presence of mCherry-EB3 , while GFP alone had no effect ( Figure 2C , G , Figure 2—figure supplement 2 ) . This effect was similar to that observed previously with the kinesin-4 Xklp1 ( Bieling et al . , 2010; Bringmann et al . , 2004 ) and with the dimeric motor domain of KIF21A ( van der Vaart et al . , 2013 ) . A decrease in MT growth rate was observed at tubulin concentrations ranging from 10 to 30 μM , while at 7 . 5 μM tubulin , 2 nM KIF21B-MD-CC1 was sufficient to almost completely prevent MT outgrowth ( Figure 2H , Figure 2—figure supplement 3A ) . In contrast , minus end growth was not affected in the presence of KIF21B-MD-CC1 ( Figure 2—figure supplement 3B ) , indicating that the effect of this kinesin on MT dynamics is plus end-specific . How can KIF21B-MD-CC1 inhibit MT growth without accumulating at MT tips ? It is possible that transient association of the dimeric motor with MT ends might be sufficient to briefly affect the structure of the tip and thus reduce its growth rate . If this were the case , even infrequent events of KIF21B-MD-CC1arrival to the MT tip could cause some perturbation of MT growth , and we reasoned such perturbations might be reflected in the brightness of the EB3 signal . To test this idea , we performed the assay in the presence of 1 nM KIF21B-MD-CC1 to create a situation when individual KIF21B-MD-CC1 would be occasionally hitting the MT tip , and used faster image acquisition conditions ( 100 ms/frame , Figure 2I ) , so that we could observe such events more clearly . Indeed , in these conditions , we observed irregularities of EB3 signal at the growing MT plus ends . To quantify this effect , we analyzed fluctuations of EB3 intensity in the presence of 1 nM GFP or 1 nM KIF21B-MD-CC1-GFP in separate chambers on the same coverslip . We excluded from our analysis MTs that were within ~40 s before catastrophe , since it is known that at this point the comet intensity is reduced ( Maurer et al . , 2012; Mohan et al . , 2013 ) . The distribution of EB3 intensities normalized to the maximum value was significantly broader ( with a lower mean and a higher standard deviation ) in the presence of KIF21B-MD-CC1-GFP than with GFP alone ( Figure 2J , K ) , indicating that the EB3 signal was indeed more irregular . We note that this analysis is not dependent on the absolute MT growth rate , which can affect the absolute EB3 signal , because the analyzed intensities were normalized to the maximum value . We conclude that the motor domain of KIF21B in a dimeric configuration is motile and can reduce MT plus-end polymerization rate , possibly by perturbing the structure of the growing MT tip . Next , we purified the full-length KIF21B-GFP from HEK293T cells ( Figure 2—figure supplement 1 ) and confirmed by single molecule analysis that it is a dimer ( Figure 3—figure supplement 1A , B , Supplementary file 2 ) . Mass spectrometry analysis of this purified protein revealed no known MT regulators ( Supplementary file 1 ) . Next , we assayed the activity of KIF21B-GFP on MTs in vitro ( Figure 3A , Figure 3—figure supplement 1C ) . Strikingly , the full-length protein showed a strong preference for GMPCPP-stabilized MT seeds , on which it landed and moved in the direction of the plus-end , while hardly any motor landing events were observed on the dynamic ( presumably GDP ) MT lattice ( Figure 3A , C ) . KIF21B-GFP motors accumulated at the tips of the seeds , and these accumulations could prevent MT outgrowth ( Figure 3A ) . Both the enrichment of KIF21B-GFP at the tip of seeds and the inhibition of MT outgrowth were more prominent for longer seeds ( Figure 3A , B ) . This indicates that GMPCPP-seeds act as ‘antennae’ that accumulate the kinesin motor at their ends in a length-dependent manner , similar to what has been previously described for the yeast kinesins Kip3 and Kip2 ( Hibbel et al . , 2015; Su et al . , 2012; Varga et al . , 2006 ) . Significant blocking of growth from MT seeds , especially the longer ones , was observed already with 3 nM KIF21B-GFP , while complete inhibition of MT outgrowth from all seeds was seen at higher KIF21B-GFP concentrations . At lower concentrations of KIF21B ( 0 . 5 nM ) growth of some seeds was still blocked , but some MTs were growing , and the effect of KIF21B on MT plus-end dynamics could be analyzed . 10 . 7554/eLife . 24746 . 012Figure 3 . KIF21B can induce MT pausing or catastrophe in vitro . ( A ) Kymographs showing the behavior of KIF21B in in vitro reconstitution assays on dynamic MTs grown from Rhodamine-tubulin-labeled seeds in the presence of 15 µM tubulin , 100 nM mCherry-EB3 ( red ) and 3 nM KIF21B-FL-GFP ( green ) . Kymographs were generated from movies acquired using a Photometrics Evolve 512 EMCCD camera ( Roper Scientific ) ( stream acquisition , exposure time 500 ms ) . Pausing and catastrophe events are indicated by arrows . ( B ) Quantification of MT seed length-dependent blocking of MT growth by 0 . 5 nM KIF21B- FL-GFP in the presence of 20 nM mCherry-EB3 . 188 MT seeds of different lengths were analyzed in four independent experiments . ( C ) Kymographs illustrating pausing events induced by KIF21B-FL-GFP ( 0 . 5 nM ) on dynamic MTs in vitro in the presence of 15 µM tubulin , 20 nM mCherry-EB3 , 3% Rhodamine-tubulin . MTs were grown from GMPCPP-stabilized seeds labeled with Rhodamine-tubulin . Kymographs were generated from the movies acquired using a CoolSNAP HQ2 CCD camera ( Roper Scientific ) with a 1 . 2-s interval between frames and an exposure time of 100 ms . The rightmost panels show tracked positions of the kinesins and MT tips together with the fluorescence intensity of the kinesins over time for the corresponding kymographs . ( D ) Kymographs illustrating various events induced by KIF21B-FL-GFP ( 0 . 5 nM ) on dynamic MTs in vitro in the presence of 15 µM tubulin , 20 nM mCherry-EB3 and 3% Rhodamine-tubulin . MTs were grown from GMPCPP-stabilized seeds labeled with Rhodamine-tubulin . Different events are indicated by arrows . Kymographs were generated from movies acquired as described for Figure 3C . ( E ) Quantification of different events observed after KIF21B-FL-GFP ( 0 . 5 nM ) reaches a growing MT plus end , as illustrated in C and D . n = 132 events , four independent experiments were analyzed . ( F ) Kymograph illustrating a long pause event induced by multiple KIF21B-FL-GFP molecules on dynamic MTs in vitro in the presence of 15 µM tubulin , 20 nM mCherry-EB3 and 3% Rhodamine-tubulin in solution . Kymographs are generated from a movie acquired as described for Figure 3C . ( G ) Kymographs illustrating the effects of KIF21B-FL-GFP ( 0 . 5 nM ) on dynamic MTs in vitro in the presence of 30 µM tubulin with 3% Rhodamine-tubulin and 20 nM mCherry-EB3 . MTs were grown from GMPCPP-stabilized seeds labeled with Rhodamine-tubulin . The arrows show the position of KIF21B at the site of MT pause and the asterisk indicates the growing MT tip beyond the position of KIF21B binding; note that the slope of the kymograph after KIF21B attachment is less steep than before , indicating that the growth rate is reduced . Kymographs are generated from movies acquired as described for Figure 3C . ( H , I ) Quantification of MT growth rate and catastrophe frequency in vitro in the presence of 15 or 30 µM tubulin with 20 nM mCherry-EB3 alone or together with 0 . 5 nM KIF21B-FL-GFP . MTs were grown in the presence of 3% Rhodamine-tubulin . For 15 µM tubulin , n = 71 for control and n = 100 for KIF21B-FL-GFP , three independent experiments . For 30 µM tubulin , n = 71 for control and n = 80 for KIF21B-FL-GFP , three independent experiments , p<0 . 0001 Mann-Whitney U test ( indicated by asterisks ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 01210 . 7554/eLife . 24746 . 013Figure 3—source data 1 . An excel sheet with numerical data on the quantification of KIF21B-FL seed blocking activity , pause induction , effects on MT growth rate and catastrophe frequency and outcomes of KIF21B-FL-GFP arrival at MT plus ends represented as plots in Figure 3B , C , E , H , I . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 01310 . 7554/eLife . 24746 . 014Figure 3—figure supplement 1 . Characterization of full-length KIF21B in vitro ( A ) Histograms of fluorescence intensities at the initial moment of observation of single molecules of the indicated proteins immobilized on coverslips ( symbols ) and the corresponding fits with lognormal distributions ( lines ) . n = 5063 , 8830 and 12230 molecules; fluorophore density was 0 . 23 , 0 . 4 and 0 . 45 µm−2 for GFP , GFP-EB3 and KIF21B-FL-GFP proteins . ( B ) Representative photobleaching time traces of GFP , EB3-GFP and KIF21B-FL-GFP individual molecules ( background subtracted ) . ( C ) Kymographs illustrating MT dynamics in vitro in the presence of 15 µM tubulin with 3% Rhodamine-tubulin and 0 . 5 nM KIF21B-FL-GFP . Kymographs were generated from the movies acquired using CoolSNAP HQ2 CCD camera ( Roper Scientific ) with a 1 . 2-s interval between frames and an exposure time of 100 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 01410 . 7554/eLife . 24746 . 015Figure 3—figure supplements 1—source data 1 . An excel sheet with numerical data on the quantification of the KIF21B-FL dimer and photobleaching step analysis represented as plots in Figure 3—figure supplement 1A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 01510 . 7554/eLife . 24746 . 016Figure 3—figure supplement 2 . KIF21B-FL-GFP induces pausing of a depolymerizing MT . The rightmost panel shows tracked positions of the kinesins and the MT tip together with the fluorescence intensities of the kinesins over time ( white boxed area in kymograph ) . See also Supplemental Video 1 . Kymograph was generated from a movies acquired using CoolSNAP HQ2 CCD camera ( Roper Scientific ) with a 1 . 2-s interval between frames and an exposure time of 100 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 01610 . 7554/eLife . 24746 . 017Figure 3—Figure Supplement 2—Source Data 1 . An excel sheet with numerical data on the quantification of tracked positions of the kinesins and the MT tip together with the fluorescence intensities of the kinesins over time represented as plot in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 017 Since KIF21B was present in the assay at low nanomolar concentrations , we could easily detect motility of individual molecules ( Figure 3A , C , D , Figure 3—figure supplement 1C ) . In cases where MT seed extension was observed , a large proportion of KIF21B-GFP motors was unable to transfer from the stabilized seed to the freshly grown part of the MT ( Figure 3A , C , D ) . However , the motors that did pass over to the freshly polymerized lattice exhibited an approximately two-fold faster motility on this lattice compared to the seed ( see below ) . These motors displayed a high degree of processivity ( runs with a length up to 8 . 5 µm were measured ) and typically reached the MT plus end ( Figure 3A , C , D ) . A number of distinct outcomes could be detected when KIF21B-GFP molecules reached MT plus ends . The most frequent one ( ~40% of all events ) was stalling of the kinesin at MT plus end accompanied by MT pausing or very slow growth , which could be distinguished by the loss of mCherry-EB3 signal from MT plus ends ( Figure 3C , E ) . We note that we cannot be sure that MTs did not undergo short ( a few hundred nanometers long ) growth and shrinkage episodes in these conditions , which we could not detect due to the resolution limit of fluorescence microscopy . Seventy-one percent of all observed pauses were induced by the arrival of what appeared to be a single kinesin dimer or a single small oligomer ( see below ) and had an average duration of 21 . 0 ± 5 . 9 s ( n = 36 ) . Tracking of the position of kinesin and MT tip together with kinesin’s fluorescence intensity over time further confirmed that during such pausing events no additional kinesins were recruited ( Figure 3C ) . We have also observed pauses where additional KIF21B-GFP molecules did arrive and stall at the plus end ( Figure 3F ) . Accumulation of multiple KIF21B-GFP motors resulted in prolonged inhibition of MT plus-end growth ( the longest pause detected was 231 s ) . We note that the pausing induced by KIF21B in this assay did not dependent on the presence of EB3 , because it was also observed in the presence of tubulin alone ( Figure 3—figure supplement 1C ) . Other possible outcomes of KIF21B-GFP arrival to the growing MT plus end , which all occurred at similar frequencies , were stalling of the kinesin on the MT without blocking MT elongation , catastrophe induction , which always led to kinesin dissociation from the MT tip , or immediate detachment of the kinesin from the MT plus end without perturbing MT growth ( Figure 3D , E ) . KIF21B-GFP molecules that reached the plus ends of shrinking MTs usually detached without affecting MT depolymerization ( Figure 3D , arrow in last kymograph ) . We did observe one example , where an event of KIF21B arrival to MT tip led to stalled MT depolymerization and a long pause with subsequent arrival of additional kinesins ( Figure 3—figure supplement 2 , Supplemental Video 1 ) . However , we did not observe any events of persistent KIF21B tracking of depolymerizing MT ends . 10 . 7554/eLife . 24746 . 018Video 1 . KIF21B induces pausing of a depolymerizing MT . The movie shows the arrival of KIF21B-FL-GFP at the end of a depolymerizing MT and a subsequent pausing event . The arrival of additional KIF21B-FL-GFP molecules results in a long pause . The experiment was performed in the presence of 15 µM tubulin , Rhodamine-tubulin ( 0 . 5 µM ) , mCherry-EB3 ( 20 nM ) and KIF21B-FL-GFP ( 0 . 5 nM ) . The movie consists of 200 frames acquired with a 1 . 2-s interval between frames and an exposure time of 100 ms . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 018 We also examined the behavior of the full-length KIF21B at a higher tubulin concentration , 30 μM , and found that in these conditions 5 nM KIF21B was needed to induce strong inhibition of seed elongation ( data not shown ) . At 0 . 5 nM KIF21B most MT seeds , including longer ones , could still grow . Since the seeds with a higher accumulation of KIF21B could still elongate , it was easier to observe multiple kinesins passing over to dynamic MTs ( Figure 3G ) . The overall effects of KIF21B on MT dynamics were similar at both tubulin concentrations: KIF21B could induce pausing and catastrophes ( Figure 3G ) . Measurement of MT dynamics showed that at both tubulin concentrations , KIF21B reduced MT growth rate and increased catastrophe frequency ( Figure 3H , I ) . How can relatively infrequent arrivals of KIF21B to growing MT tips significantly affect MT elongation rate ? We noticed that , even when the polymerization of a MT plus end was not fully suppressed by the incoming KIF21B molecule , in cases when the kinesin did not immediately detach from the MT , MT elongation was typically strongly perturbed ( Figure 4 , Figure 4—figure supplement 1A ) . We observed many events where a MT tip was undergoing short repeated growth and shortening excursions from the point of KIF21B stalling ( Figure 4A , B , Supplemental Video 2 ) . Such MT behavior indicates that KIF21B immobilized at a MT tip prevented both its normal elongation and also its depolymerization . At 30 μM tubulin , the KIF21B stalling events typically led to very irregular growth , which ended in catastrophe ( Figure 4—figure supplement 1B ) . Importantly , after the point where KIF21B was stalled , the path of the growing MT often became curved ( Figure 4 , Figure 4—figure supplement 1 , Supplemental Video 2 ) , while control MTs always grew straight in our assays . Taken together , these data suggest that KIF21B attached to the plus end might be blocking growth of a few protofilaments , leading to the extension of an incomplete and thus more flexible tube , which is more prone to undergo a catastrophe . 10 . 7554/eLife . 24746 . 019Figure 4 . KIF21B molecules persisting on a MT tip can perturb MT growth . ( A , B ) Kymographs illustrating perturbation of MT growth in vitro by 0 . 5 nM KIF21B-FL-GFP in the presence of 15 µM tubulin with 3% Rhodamine-tubulin and 20 nM mCherry-EB3 . Kymographs were generated from movies acquired as described for Figure 3C . Positions of the kinesins and MT tips together with the fluorescence intensity of the kinesins over time for the corresponding kymographs are also illustrated . Time lapse images on the right of the kymographs illustrate short excursions of MT plus tip ( A ) or curling of MT plus tip ( B ) after the binding of KIF21B-FL-GFP to the MT plus end . The position of the kinesin on the MT is indicated by arrows . Asterisks show the position of growing MT tips extending beyond the point of KIF21B attachment . See also Supplemental Video 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 01910 . 7554/eLife . 24746 . 020Figure 4—source data 1 . An excel sheet with numerical data on the quantification of tracking of kinesins and MT tips over time represented as plots in Figure 4A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 02010 . 7554/eLife . 24746 . 021Figure 4—figure supplement 1 . Perturbation of MT growth in vitro by full-length KIF21B ( A , B ) Kymographs illustrating perturbation of MT growth in vitro by 0 . 5 nM KIF21B-FL-GFP in the presence of 15 and 30 µM tubulin with 3% Rhodamine-tubulin and 20 nM mCherry-EB3 . Time lapse images on the right illustrate MT plus tip curling after the binding of KIF21B-FL-GFP to the MT plus end . The position of the kinesin on the MT is indicated by arrows . Asterisks show the position of growing MT tips extending beyond the point of KIF21B attachment . Boxed area is zoomed . See also Supplemental Video 2 . Kymographs were generated from movies acquired using CoolSNAP HQ2 CCD camera ( Roper Scientific ) with a 1 . 2-s interval between frames and an exposure time of 100 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 02110 . 7554/eLife . 24746 . 022Video 2 . KIF21B perturbs MT growth and induces MT bending . The combined movie shows the two different events illustrated in Figure 4—figure supplement 1A and Figure 4B . The movie shows bending of an MT growing beyond the point where KIF21B-FL-GFP was stalled ( top panel , upper MT ) and repeated short excursions from the point of KIF21B-FL-GFP stalling ( bottom panel ) . The experiment was performed in the presence of 15 µM tubulin , Rhodamine-tubulin ( 0 . 5 µM ) , mCherry-EB3 ( 20 nM ) and KIF21B-FL-GFP ( 0 . 5 nM ) . The movie consists of 128 frames acquired with a 1 . 2-s interval between frames and an exposure time of 100 ms . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 022 As indicated above , at 15 μM tubulin , we frequently observed MT pause induction by what appeared to be single KIF21B molecules or small oligomers . The ratio of the intensity of the motile KIF21B molecules responsible for the pausing events to the intensity of KIF21B dimers immobilized on the same coverslip was ~0 . 75–1 . 9 ( Figure 5A ) . This is consistent with one or two KIF21B molecules , if one takes into account the decay of the evanescent field . To further examine this , we compared KIF21B intensities with those of an N-terminal fragment of kinesin-1 , KIF5B residues 1–560 ( denoted KIF5B-560 ) , which is a well-studied motile dimeric kinesin ( Vale et al . , 1996 ) . The intensities of single GFP-tagged KIF5B-560 molecules moving on MTs in vitro were ~1 . 8 higher than the intensities of single GFP proteins immobilized in a separate chamber on the same coverslip ( Figure 5B , C , Supplementary file 2 ) , confirming that KIF5B-560 is a dimer in our preparation , similar to what we published previously ( Doodhi et al . , 2014 ) . We then compared the intensities of full length KIF21B molecules moving on seeds to the intensities of KIF5B-560 molecules moving on MTs in a separate chamber on the same coverslip , and found that they were also very similar ( Figure 5D , Supplementary file 2 ) . Importantly , the intensities of the few KIF21B molecules that transferred from seeds to the dynamic MT lattice were very similar to those of KIF21B molecules moving on seeds ( Figure 5E , Figure 5—figure supplement 1 , Supplementary file 2 ) . Together , these data indicate that motile KIF21B kinesins are mostly dimers , and that these kinesins do not need to form larger oligomers in order to ‘escape’ from the seed to the dynamic lattice . 10 . 7554/eLife . 24746 . 023Figure 5 . Single events of kinesin arrival to the MT plus end can induce MT pausing . ( A ) GFP intensity analysis of kinesins during MT pausing events . Values are normalized to the GFP intensity of proteins immobilized on the same coverslip in areas devoid of MTs . Data are from two independent experiments . ( B ) Kymographs showing the behavior of KIF5B-560-GFP in an in vitro reconstitution assay on dynamic MTs grown from Rhodamine-tubulin-labeled seeds in the presence of 15 µM tubulin , 20 nM mCherry-EB3 ( red ) and 5 nM KIF5B-560-GFP ( green ) . ( C ) Histograms of fluorescence intensities of single GFP molecules immobilized on coverslips ( initial moment of observation of single molecules ) or KIF5B-560-GFP moving on MTs in a separate chamber on the same coverslip ( symbols ) and the corresponding fits with lognormal distributions ( lines ) . n = 452 and 2040 molecules; fluorophore density was 0 . 05 and 0 . 09 µm−2 for GFP and KIF5B-560-GFP proteins ( for the latter , MT-containing regions were manually selected for analysis ) . Dashed lines show the corresponding relative median values . ( D ) Kymographs showing the behavior of 5 nM KIF5B-560-GFP ( moving on dynamic MTs , green ) and 0 . 5 nM KIF21B-FL-GFP ( moving on seeds , green ) in an in vitro reconstitution assay with MTs grown from Rhodamine-tubulin-labeled seeds in the presence of 15 µM tubulin and 20 nM mCherry-EB3 ( red ) . Histograms illustrate fluorescence intensities of KIF5B-560-GFP moving on MTs and KIF21B-FL-GFP moving on seeds in separate chambers on the same coverslip ( symbols ) and the corresponding fits with lognormal distributions ( lines ) . n = 5123 and 8728 molecules; fluorophore density was 0 . 15 and 0 . 18 µm−2 for KIF5B-560-GFP and KIF21B-FL-GFP proteins ( MT-containing regions were manually selected for analysis ) . Ratio of the corresponding median values is also indicated . ( E ) Kymographs illustrating KIF21B-FL-GFP ( 0 . 5 nM ) movement on seeds and dynamic MTs , histograms of the corresponding fluorescence intensities measured within the same sample ( symbols ) and their fits with lognormal distributions ( lines ) . Median values are also indicated . ( F–H ) Kymographs illustrating motility of KIF5B-560-GFP ( 5 nM ) on dynamic MTs and KIF21B-FL-GFP ( 0 . 5 nM ) moving from an MT seed to the dynamic MT lattice and inducing a MT pause in an in vitro reconstitution assay with MTs grown from Rhodamine-tubulin-labeled seeds in the presence of 15 µM tubulin and 20 nM mCherry-EB3 ( red ) . Histograms show fluorescence intensities of motile KIF5B-560-GFP molecules and KIF21B-FL-GFP inducing a MT pause in separate chambers on the same coverslip ( symbols ) and the corresponding fits with lognormal distributions ( lines ) . Median values are also indicated . In all the conditions , kymographs were generated from movies of 1500 frames ( stream acquisition , exposure time 100 ms ) using Photometrics Evolve 512 EMCCD camera ( Roper Scientific ) . Positions of seeds in each kymograph are indicated . ( I ) Fitted peak intensity time trace for the trajectory of a moving KIF21B-FL-GFP molecule from the event shown in Figure 5H . Dashed lines correspond to the scaled values of median fluorescence fitted peak intensity of KIF5B-560-GFP molecules moving on dynamic MT in a parallel chamber on the same coverslip . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 02310 . 7554/eLife . 24746 . 024Figure 5—source data 1 . An excel sheet with numerical data on the quantification of KIF21B-FL intensity during MT pausing events , KIF5B-560 dimer analysis and comparison of fluorescence intensities of KIF5B-560 with KIF21B-FL represented as plots in Figure 5A , C , D–I . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 02410 . 7554/eLife . 24746 . 025Figure 5—figure Supplement 2—Source data 1 . An excel sheet with numerical data on the quantification of photobleaching traces of KIF21B-FL-GFP represented as plots in Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 02510 . 7554/eLife . 24746 . 026Figure 5—figure supplement 1 . Kymographs illustrating KIF21B-FL-GFP ( 0 . 5 nM ) moving on seeds and dynamic MTs in in vitro reconstitution assays , histograms of the corresponding fluorescence intensities ( symbols ) and the corresponding fits with lognormal distributions ( lines ) . Median values are also indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 02610 . 7554/eLife . 24746 . 027Figure 5—figure supplements 1—Source data 1 . An excel sheet with numerical data on the quantification of KIF21B-FL fluorescence intensities represented as plots in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 02710 . 7554/eLife . 24746 . 028Figure 5—figure supplement 2 . Characteristic photobleaching traces of KIF21B-FL-GFP under two different imaging conditions . KIF21B-FL-GFP immobilized on coverslip was exposed to low laser power ( used for imaging shown in Figures 3–5 and 7 ) with 100 ms/stream acquisition or 100 ms exposure time , 1 frame per 1 . 2 s . Curves were fitted with one-phase exponential decay . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 028 We have also considered that we might be underestimating the actual size of KIF21B clusters that induce pausing because of their photobleaching . For our regular imaging experiments ( Figure 3C , D , G , Figure 4 ) , we acquired the data at 1 . 2 s/frame with an exposure time of 100 ms . For comparison , we collected data with the same laser power at a 12 times higher frame rate ( 100 ms/frame , stream acquisition ) , again in the presence of a ‘reference chamber’ with KIF5B-560 on the same coverslip . We observed several events of MT pause induction , in which the intensity of the KIF21B molecule inducing a pause was similar to that of single KIF5B-560 kinesins ( Figure 5F–H , Supplementary file 2 ) . For example , in the event shown in Figure 5H , the initial intensity of the analyzed KIF21B molecule when it starts its movement on the seed is in the range of the median of KIF5B-560 molecules ( Figure 5I ) . While this molecule moves on the freshly polymerized MT lattice , its intensity is reduced by half , which we attribute to the bleaching of one of the GFP molecules . After arriving to the MT tip and inducing a pause , the molecule bleaches or desorbs as the MT switches to catastrophe . These data show that the laser power used for illumination was gentle enough for 20 s of imaging of a single KIF21B molecule at 10 frames per second . Measurements of the average photobleaching time showed that it was ~16 s at 100 ms/stream acquisition and ~197 s when the images were acquired with a 100 ms exposure with the interval of 1 . 2 s ( Figure 5—figure supplement 2 ) . This means that in our regular imaging experiments , the average bleaching time is close to 200 s and is thus significantly longer than the duration of kinesin runs , which is typically tens of seconds . Photobleaching is thus unlikely to lead to a strong underestimate of the number of kinesins sufficient to trigger MT pausing . Our data presented so far indicate that a few KIF21B motors can prevent both growth and shortening by ‘holding on’ to a MT plus end . Since the dimeric motor domain of KIF21B by itself does not show such an activity , this result suggests that additional MT-binding sites that can associate with the MT plus ends must be present in the KIF21B tail . In line with this conclusion , we observed that when MTs were allowed to grow long in vitro in the presence of a low ( 0 . 5 nM ) KIF21B concentration , KIF21B motors could pull a MT along another MT ( Figure 6—figure supplement 1 ) . This observation suggests that KIF21B can bind to one MT and walk along another MT at the same time . We then set out to identify additional MT binding site ( s ) in the KIF21B tail by deletion mapping ( Figure 6A ) . Different KIF21B fragments were expressed in COS-7 cells , and their colocalization with MTs was observed by fluorescence microscopy ( Figure 6B , Figure 6—figure supplement 2A ) . We found that the tail of KIF21B alone strongly localized to MTs ( Figure 6A , B ) . Subsequent mapping showed that two separate parts of the KIF21B tail could bind to MTs: the centrally located predicted coiled-coil part with adjacent sequences ( CC2 ) , as well as the C-terminal WD40 domain together with the N-terminal linker region enriched in proline , serine and arginine residues ( termed L-WD40; Figure 6A , B , Figure 6—figure supplements 2A and 3 ) . Neither the WD40 domain alone , nor the linker alone showed robust MT binding , suggesting that the MT-binding affinity of this region depends on the combination of the two elements ( Figure 6A , B , Figure 6—figure supplement 2A ) . Together , these data indicate that KIF21B can interact with MTs through three non-overlapping regions , the motor domain , the stalk region and the WD40 domain , and that the full length KIF21B molecule is likely to be folded when attached to MTs . 10 . 7554/eLife . 24746 . 029Figure 6 . Mapping of the MT-binding domains in the tail of KIF21B . ( A ) Overview of deletion mutants used in this study . Colocalization of the GFP-tagged KIF21B deletion mutants with MTs in transiently transfected COS-7 cells is indicated . + , localization to MTs , - , diffuse distribution , -/+ , diffuse in most cells , with occasional MT localization observed in some cells . ( B ) COS-7 cells were fixed one day after transient transfection with the indicated constructs and stained for α-tubulin . ( C ) Electron micrographs of negatively stained taxol-stabilized MTs in complex with KIF21B-FL-GFP . ( D ) Live imaging of COS-7 cells transiently transfected with KIF21B-MD-CC-GFP or MD-CCΔrCC-GFP and EB3-TagRFP-T . Represented are a single-frame , maximum intensity projection of 500 frames for the GFP channel , an overlay of single GFP frame in green and TagRFP-T in red and a kymograph along one of the EB3-labeled MTs showing kinesin motility . ( E ) Streptavidin pull down assay with the extracts of HEK293T cell expressing BirA , KIF21B-MD-CC1-GFP-TEV-Bio and the indicated mCherry-labeled proteins . A and B stand for KIF21A and KIF21B; LZ , leucine zipper from GCN4 used for dimerization . The other abbreviations are explained in panel A . The results were analyzed by Western blotting with the antibodies against the GFP- and mCherry . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 02910 . 7554/eLife . 24746 . 030Figure 6—figure supplement 4—source data 1 . An excel sheet with numerical data on the quantification of far-UV CD spectra ( inset ) and thermal unfolding profile of recombinant KIF21B rCC1 represented as plots in Figure 6—figure supplement 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 03010 . 7554/eLife . 24746 . 031Figure 6—figure supplement 1 . In vitro reconstitution of MT growth in the presence of 20 nM mCherry-EB3 , 3% Rhodamine-tubulin and 0 . 5 nM KIF21B-FL-GFP . KIF21B-FL-GFP is attached to one MT and walks along another one , causing MT bending . Arrow indicates position of KIF21B-FL-GFP , yellow arrowheads trace the MT that bends . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 03110 . 7554/eLife . 24746 . 032Figure 6—figure supplement 2 . ( A ) COS-7 cells transiently transfected with the indicated KIF21B-GFP deletion constructs and stained for α-tubulin . ( B ) MT pelleting assay of taxol-stabilized MTs incubated with KIF21B-FL-GFP . Coomassie-stained SDS-PAGE of supernatant and pellet fractions of MTs alone or MTs incubated with KIF21B-FL-GFP are shown . ( C ) Electron micrograph of a negatively stained taxol-stabilized MT control specimens . ( D ) Gallery of electron micrographs of negatively stained taxol-stabilized MT specimens obtained in the presence of KIF21B-FL-GFP . ( E ) High-magnification views of taxol-stabilized MT ends decorated with KIF21B-FL-GFP . ( F ) Gallery of electron micrographs of negatively stained GMPCPP-MT specimens obtained in the presence of KIF21B-FL-GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 03210 . 7554/eLife . 24746 . 033Figure 6—figure supplement 3 . Alignment of human KIF21A and KIF21B sequences . Different protein domains and the autoinhibitory region in the coiled coil domain described for KIF21A ( van der Vaart et al . , 2013 ) are indicated , and CFEOM1-associated mutations found in KIF21A are boxed . The KIF21B sequence shown here corresponds to the longest KIF21B isoform ( Accession number O75037 ) ; a shorter isoform ( Accession number BAA32294 ) , which misses the amino acids 1269–1281 ( underlined ) , was used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 03310 . 7554/eLife . 24746 . 034Figure 6—figure supplement 4 . ( A ) Far-UV CD spectra ( inset ) and thermal unfolding profile of recombinant KIF21B rCC . CD measurements , performed in PBS at a protein concentration of 0 . 166 mg/ml . ( B ) Oligomerization state of recombinant KIF21B rCC determined by sedimentation velocity AUC at 20°C and at three different protein concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 034 To test whether KIF21B assumes a compact conformation when bound to MTs , we used MT pelleting assays and electron microscopy with the isolated full-length KIF21B together with taxol-stabilized MTs ( Figure 6—figure supplement 2B ) . As expected , after centrifugation , the full-length KIF21B was present in the pelleted fraction together with MTs . The pelleted fractions were further analyzed by negative stain electron microscopy . Electron micrographs of the full-length KIF21B bound to MTs indeed suggest that the motor has a highly folded , globular appearance , consistent with the presence of several MT interaction sites ( Figure 6C , Figure 6—figure supplement 2C–E ) . Similar results were obtained with GMPCPP-stabilized MTs ( Figure 6—figure supplement 2F ) . Interestingly , the deletion of the linker region located N-terminally of the WD40 strongly reduced the MT binding activity of the resulting KIF21B mutant , while the deletion of the C-terminal WD40 domain rendered the KIF21B protein completely diffuse ( KIF21B-MD-CC construct , Figure 6A , D ) . This was surprising , as the other two MT-binding sites , the motor domain and the CC2 , were still present in these deletion mutants . Since a shorter KIF21B fragment , KIF21B-MD-CC1 , bound to and moved along MTs , this result suggests that the CC2 region harbors not only a MT-binding , but also an autoregulatory activity . Previous work showed that the part of KIF21A that corresponds to the CC2 region of KIF21B contains an autoinhibitory element , which interacts with the motor domain , and that mutations in this region cause loss of autoinhibition and lead to CFEOM1 ( Cheng et al . , 2014; van der Vaart et al . , 2013 ) . Recently , we have characterized this interaction in detail and showed that it is mediated by a regulatory region that forms an intramolecular antiparallel coiled coil ( Bianchi et al . , 2016 ) . This sequence region , including the CFEOM1-associated residues , is well conserved in KIF21B ( rCC , amino acids 931–1010 ) ( Figure 6—figure supplement 3 ) , suggesting that it might have a similar autoinhibitory function . To test this possibility , we first assessed the secondary structure content and the stability of KIF21B rCC by performing circular dichroism ( CD ) spectroscopy experiments . The far-ultraviolet CD spectra recorded for the polypeptide chain fragments revealed a significant amount of α-helical structure with distinct minima centered around 208 and 222 nm ( Figure 6—figure supplement 4A , inset ) . The stability of KIF21B rCC was subsequently assessed by a thermal unfolding profile monitored by CD at 222 nm , which yielded a melting temperature of 43 . 6°C Figure 6—figure supplement 4A ) . To assess the oligomerization state of KIF21B rCC , we performed sedimentation velocity experiments ( Figure 6—figure supplement 4B ) , which revealed a molecular weight of 11 kDa for the polypeptide chain fragment ( calculated molecular weight of KIF21B rCC: 9 . 5 kDa ) . These biophysical results are consistent with KIF21B rCC forming an intramolecular antiparallel coiled coil in solution , very similar to the one we described for KIF21A ( Bianchi et al . , 2016 ) . In agreement with the expected autoinhibitory function of rCC , its deletion in the KIF21B-MD-CC restored MT binding activity and motility of this KIF21B fragment ( Figure 6A , D , Figure 6—figure supplement 2A ) . By itself , the rCC did not bind MTs in cells , and its deletion had no effect on the MT binding properties of the CC2 fragment ( Figure 6A , Figure 6—figure supplement 2A ) . Using immunoprecipitation assays , we detected an interaction between the CC2 and the MD-CC1 region of KIF21B ( Figure 6E ) . A weak binding was also observed with an rCC variant that was fused to the dimeric leucine zipper ( LZ ) of GCN4 , although not with the monomeric version of rCC ( Figure 6E ) . A stronger binding of the KIF21B motor domain was observed to the rCC region of KIF21A ( Figure 6E ) , suggesting that the autoinhibitory interaction within KIF21B is attenuated compared to KIF21A . In agreement with this view , overexpressed KIF21A is largely diffuse in cells and presumably only becomes active when bound to appropriate partners ( van der Vaart et al . , 2013 ) , while KIF21B shows constitutive MT association . If the rCC does not fully inhibit the full-length KIF21B motor , what is the function of this region ? To address this question , we have purified the KIF21B protein lacking the rCC ( KIF21B-FL-ΔrCC ) , and a shorter version of this protein , which also lacked the WD40 domain ( KIF21B-MD-CCΔrCC ) ( Figure 2—figure supplement 1 ) . Mass spectrometry analysis demonstrated that the contaminants present in these two KIF21B preparations were essentially the same as in the isolated full-length KIF21B ( Supplementary file 1 ) . Analysis of fluorescence intensity and photobleaching confirmed that both deletion mutants are dimers , similar to the full-length molecule ( Figure 7A , Supplementary file 2 ) . In vitro assays showed that unlike the full-length protein , both kinesins lacking the rCC could land not only on GMPCPP-seeds but also on newly polymerized MT lattices ( Figure 7B ) . Both full-length KIF21B and KIF21B-FL-ΔrCC exhibited slower motility on seeds compared to fresh GDP-MT lattices ( Figure 7C ) . In contrast , the KIF21B-MD-CCΔrCC protein showed no reduced velocity on the MT lattice ( Figure 7B , C ) , suggesting that the C-terminal WD40-containing region creates friction on GMPCPP-seeds . Consistent with this notion , we found that the L-WD40-GFP fragment displayed high preference for GMPCPP-seeds in vitro , both when present in cell extracts and in purified form , while the CC2 fragment showed no such preference ( Figure 7D , Figure 7—figure supplement 1A–C ) . The preference for the GMPCPP seeds was not due to their attachment to glass or inclusion of biotinylated tubulin , because L-WD40-GFP showed no preference for taxol-stabilized MT seeds prepared in the same way as the GMPCPP seeds ( Figure 7D , Figure 7—figure supplement 1B , C ) . However , we did not observe any accumulation of L-WD40-GFP at the growing MT plus ends , indicating that the preference for a specific nucleotide state is not sufficient to induce plus-end tracking of this protein fragment . 10 . 7554/eLife . 24746 . 035Figure 7 . The WD40 domain and the autoinhibitory coiled coil region contribute to the pause-promoting activity of KIF21B . ( A ) Histograms of fluorescence intensities at the initial moment of observation of single molecules of the indicated proteins immobilized on coverslips ( symbols ) and the corresponding fits with lognormal distributions ( lines ) . n = 3907 , 5002 , 6725 and 6943 molecules; fluorophore density was 0 . 19 , 0 . 24 , 0 . 30 and 0 . 33 µm−2 for GFP , GFP-EB3 , KIF21B-FL-ΔrCC-GFP and KIF21B-MD-CCΔrCC-GFP proteins . Insets show representative photobleaching traces of individual molecules ( background subtracted ) . ( B ) Kymographs illustrating the behavior of the indicated deletion mutants of KIF21B at 3 nM concentration on dynamic MTs in the presence of 100 nM mCherry-EB3 . GMPCPP-stabilized MT seeds were labeled with Rhodamine-tubulin ( lines below kymographs ) . Kymographs were generated from the movies acquired using Photometrics Evolve 512 EMCCD ( Roper Scientific ) camera ( stream acquisition with an exposure time of 500 ms ) . ( C ) Quantification of the velocity of KIF21B-FL-GFP and the deletion mutants on seeds and freshly polymerized MT lattices , shown in Figures 3A and 7B . Seed: n = 295 for KIF21B-FL-GFP , n = 195 for KIF21B-FL-ΔrCC-GFP , n = 434 for KIF21B-MD-CCΔrCC-GFP; lattice: n = 131 for KIF21B-FL , n = 133 for KIF21B-FL-ΔrCC-GFP , n = 434 for KIF21B-MD-CCΔrCC-GFP . Data are from two or three independent experiments . Values significantly different from each other are indicated by asterisks , p<0 . 0001 , Mann-Whitney U test . ( D ) Kymographs illustrating the interaction of purified GFP-L-WD40 ( 100 nM ) with dynamic MTs grown from Rhodamine-tubulin labeled GMPCPP- or taxol-stabilized seeds ( as indicated ) in the presence of 20 nM mCherry-EB3 . Kymographs were generated from the movies acquired in stream acquisition mode with an exposure time of 500 ms using Photometrics Evolve 512 EMCCD camera ( Roper Scientific ) . ( E ) Kymographs illustrating the behavior of KIF21B deletion mutants on dynamic MTs in the presence of 20 nM mCherry-EB3 and 3% Rhodamine-tubulin . Pauses and KIF21B detachment from a depolymerizing MT end are indicated by arrows and asterisks , respectively . Arrowheads indicate kinesin detachment from the growing MT tip . Kymographs were generated from the movies acquired using CoolSNAP HQ2 CCD camera ( Roper Scientific ) with a 1 . 2-s interval between frames and an exposure time of 100 ms . ( F ) Quantification of MT growth rate in vitro in the presence of 15 µM tubulin with 20 nM mCherry-EB3 alone ( n = 71 ) or together with 3 nM KIF21B-FL-ΔrCC-GFP ( n = 79 ) or KIF21B-MD-CCΔrCC-GFP ( n = 79 ) . MTs were grown in the presence of 3% Rhodamine-tubulin . two independent experiments . ( G ) Quantification of different events observed after KIF21B-FL or its mutants reach a growing MT plus end . Data shown in Figure 3E are included here for comparison . n = 501 for KIF21B-FL-ΔrCC-GFP , n = 647 for KIF21B-MD-CCΔrCC-GFP . Data are from at least two independent experiments . ( H ) Percentage of pausing events induced by a single event of kinesin arrival from all detected pauses . Total number of pausing events: n = 51 for KIF21B-FL-GFP , n = 46 for KIF21B-FL-ΔrCC-GFP , n = 22 for KIF21B-MD-CCΔrCC-GFP . Data are from at least two independent experiments . ( I ) Quantification of the duration of MT pausing induced by a single kinesin arrival event at the growing MT plus end . n = 36 for KIF21B-FL-GFP , n = 8 for KIF21B-FL-ΔrCC-GFP , n = 3 for KIF21B-MD-CCΔrCC-GFP . Data are from at least two independent experiments . **p<0 . 0001 , *p<0 . 0004 Mann-Whitney U test . ( J ) Model for the regulation of KIF21B motility and pause induction by the tail domain . In solution , KIF21B motor domains are inhibited by the regulatory region , while the WD40 domains are available for the interaction with MTs; WD40 domains show preference for the GMPCPP-stabilized seeds ( red ) . After binding to seeds , KIF21B becomes activated and can walk to the plus end; it is likely that both the WD40 and the CC2 region contribute to MT binding . The kinesin can transfer from the seed to the freshly polymerized MT lattice; the interaction of the CC2 but not of the WD40 with the lattice promotes motor processivity . At the tip , the conversion to the autoinhibited conformation and the WD40 domain can prevent KIF21B from stepping off the MT plus end . This allows the motor to prevent both elongation and shortening of a small number of protofilaments with which it interacts . The remaining protofilaments might undergo short excursions of growth and shrinkage ( upper panel ) ; alternatively , they might elongate for some time and such an incomplete MT will be prone to bending and catastrophe ( lower panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 03510 . 7554/eLife . 24746 . 036Figure 7—source data 1 . An excel sheet with numerical data on the quantification of KIF21B mutants dimer analysis , photobleaching step analysis , velocities on seeds and MT lattices , MT growth rate in vitro and outcomes of the arrival of KIF21B mutants at MT plus ends , represented as plots in Figure 7A , C , F–I . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 03610 . 7554/eLife . 24746 . 037Figure 7—figure supplement 1 . Characterization of KIF21B tail fragments in vitro ( A ) In vitro reconstitution of MT dynamics in the presence of 100 nM GFP-EB3 and the extracts of HEK293T cells expressing mCherry-CC2 or mCherry-L-WD40 . GMPCPP-stabilized MT seeds were labeled with HiLyte Fluor 488-tubulin ( lines below kymographs ) . ( B ) Quantification of the intensity of purified GFP-L-WD40 on GMPCPP- and taxol-stabilized seeds or dynamic MTs grown from GMPCCP or taxol-stabilized seeds . ( C ) Overview of the interactions of GFP-tagged KIF21B and its deletion mutants with MTs in vitro . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 03710 . 7554/eLife . 24746 . 038Figure 7—figure supplements 1—source data 1 . An excel sheet with numerical data on the quantification of the intensity of KIF21B-L-WD40 on seeds and dynamic MTs represented as plot in Figure 7—figure supplement 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 24746 . 038 We next examined the ability of the two KIF21B mutants to affect MT growth . Similar to the full-length molecule , both proteins showed a high degree of processivity , with run lengths of up to ~8–8 . 5 µm ( Figure 7E ) . The two mutants had no effect on MT depolymerization , as they detached from shrinking MTs ( asterisks in Figure 7E ) . Both mutants reduced MT growth rate , similar to the full-length KIF21B and the KIF21B-MD-CC1 mutant ( Figure 7F ) . Furthermore , upon arrival to the growing MT plus end , the two mutants could cause MT pausing and induce catastrophes , again similar to what was observed with the full-length molecule ( Figure 7E , G ) . However , both mutated motors were much less potent than full-length KIF21B , because in the majority of the cases ( ~60% ) , the mutated motors detached from the MT tip without pausing MT growth ( Figure 7G ) . The pauses induced by the two mutant kinesins were typically due to the presence of multiple independently arriving motors ( Figure 7E , G , H ) , and the duration of pauses induced by a single arrival event of either KIF21B-FL-ΔrCC or KIF21B-MD-CCΔrCC were significantly shorter than those triggered by the full-length KIF21B protein ( Figure 7I ) . Taken together , these results suggest that both the regulatory rCC region and the C-terminal WD40-containing domain can contribute to the ability of KIF21B to stay attached to the growing MT plus end and to induce pausing ( Figure 7J ) .
In this study , we showed that KIF21B is a highly processive MT plus-end directed motor , which can potently induce pausing of MT plus ends . This activity depends on several regions of this large motor protein . The N-terminally located motor domain is motile , thus ensuring protein accumulation at the MT plus ends , and similar to the motor domains of other kinesin-4 family members , it slows down MT polymerization ( Bieling et al . , 2010; Bringmann et al . , 2004; van der Vaart et al . , 2013 ) . The dimeric version of the motor is sufficient to reduce MT growth , even though it does not accumulate at the MT tips . It is possible that KIF21B motors arriving at the MT tip somehow affect the conformation of terminal tubulin dimers , and in this way transiently perturb the structure of the polymerizing MT end . Our analysis of EB3 intensity in the presence of a low concentration of KIF21B-MD-CC1 supports this idea . Importantly , in contrast to the full-length kinesin , the dimeric version of the motor domain alone does not show high processivity or the ability to stay attached to growing MT plus ends , thereby inducing their pausing . These properties are conferred by the stalk and the tail regions of the protein , which constitute two separate MT-binding sites . The presence of additional MT-binding domains is quite common in kinesins and has been established for kinesin-1 , kinesin-5 , kinesin-8 family members and CENP-E ( Gudimchuk et al . , 2013; Navone et al . , 1992; Stumpff et al . , 2011; Su et al . , 2011; van den Wildenberg et al . , 2008; Weaver et al . , 2011 ) . These domains are often basic polypeptide regions that can interact with the negatively charged surface of MTs . The distinguishing feature of KIF21B is the presence of a C-terminal WD40 domain involved in MT binding together with the adjacent positively charged linker region . The combined MT-binding activity of a folded domain augmented by a basic polypeptide region is reminiscent of that found in other MAPs such as EBs , CLIPs and the Ndc80 complex , in which globular calponin homology or CAP-Gly domains cooperate with positively charged linkers for MT binding ( Alushin et al . , 2012; Hoogenraad et al . , 2000; Komarova et al . , 2009 ) . An interesting feature of the KIF21B C-terminus is its ability to distinguish between different types of MT lattices , as it binds much better to GMPCPP than to taxol-stabilized GDP-MTs . This property has an impact on the full-length protein , as the WD40 domain promotes the binding of KIF21B to GMPCPP seeds and slows down its motility on the seeds . Since GMPCPP-MTs are believed to mimic certain features of the GTP-MT lattice ( Alushin et al . , 2014 ) , which is enriched at growing MT plus ends , it is tempting to speculate that in the context of the full-length protein this property helps to prevent kinesin detachment from the polymerizing plus ends . We should note , however , that the mechanistic basis of the preference of the L-WD40 fragment of KIF21B for the GMPCPP lattice is unclear , and it is not sufficient to confer MT plus-end tracking behavior . It is possible that at growing MT plus ends the number of binding sites for which the C-terminus of KIF21B would have preference or its affinity for these sites would be affected by protofilament curvature ( Brouhard and Rice , 2014 ) . Still , it could act in concert with other MT-binding domains of KIF21B to increase the bias for end-binding . Another feature that helps to prevent the detachment of KIF21B from the MT plus end once it arrives at the tip is the presence of the autoregulatory region . Autoinhibition mechanisms that prevent motility of the cargo-unbound motors are common in different kinesins , including kinesin-1 , kinesin-2 and the close KIF21B homologue KIF21A ( van der Vaart et al . , 2013; Verhey and Hammond , 2009 ) . However , in contrast to other autoinhibited kinesins , which require an activating partner , KIF21B proteins can still interact with MTs through the binding of the WD40 domain-containing tail with MTs . In our in vitro assays , the WD40 region induces a strong preference of the motor for GMPCPP-MTs , suggesting that the motor domains are autoinhibited , while the WD40 domains are available for MT binding ( Figure 7J ) . KIF21B with a deleted WD40 domain is fully autoinhibited , which indicates that the interaction between the motor and the rCC blocks the ability of both the motor and the MT-binding stalk ( CC2 region ) to interact with MTs . Once the motor is loaded on an MT , the autoinhibition is relieved and the motor starts to walk . It is possible that both the WD40 and the MT-binding CC2 regions contribute to the processivity of KIF21B ( Figure 7J ) . We observe frequent detachment of full-length motors at the border between the GMPCPP-seed and the GDP-MT lattice , suggesting that KIF21B might be switching back to an autoinhibited state . Such switching might be stimulated by the presence of MT lattice defects at the border between the seed and the freshly grown lattice . In contrast , motors lacking the rCC region land more easily on dynamic MT lattices and pass more frequently to such lattices from GMPCPP-seeds . The motors that do move along the GDP-MT lattice are highly processive , most likely due to the MT-binding CC2 domain in the stalk region , because the KIF21B-FL-ΔrCC and KIF21B-MD-CCΔrCC behave very similarly . In contrast , the KIF21B-MD-CC1 , which lacks the CC2 region , displays only short runs . The WD40 domain and the interaction between the motor domains and the autoregulatory rCC region become important once the kinesin reaches the growing MT plus end . KIF21B mutants lacking these regions often detach from the MT tip , while the full-length motor frequently persists and promotes pausing . It is possible that this behavior depends on the switching of the kinesin from the stepping mode to a conformation in which it attaches to the MT through its MT-binding tail domains ( Figure 7J ) . As discussed above , in these conditions , the WD40 domain-containing C-terminus might help to recognize some structural feature of the plus end that is related to the presence of the GTP-cap . It is striking that a few kinesins can induce a stable pause with an average duration of ~20 s , suggesting that different MT-binding domains within one molecule might interact with several protofilaments and prevent both their growth and depolymerization , thus inducing a pausing state . Noteworthy in this respect is our observation that when a MT continued growing after KIF21B was stalled , its growth was often strongly perturbed: we observed switching between growth and shortening episodes , as well as strong MT bending after the point of KIF21B attachment ( Figure 7J ) . These data are reminiscent of our work on the effect of binding of a protofilament-blocking drug , Eribulin ( Doodhi et al . , 2016 ) . We recently showed that attachment of a single Eribulin molecule , which , based on structural data can inhibit growth of only one MT protofilament , was sufficient to either cause a catastrophe or induce MT growth perturbation , suggesting elongation of an incomplete MT ( Doodhi et al . , 2016 ) . It is tempting to speculate that similar to Eribulin , a KIF21B molecule stalled at the MT tip would be sufficient to block a small number of protofilaments , and this would result in inefficient elongation of the remaining protofilaments . Importantly , in contrast to Eribulin , KIF21B can also stabilize the protofilaments which it blocks , and therefore it is able to prevent , at least for some time , MT depolymerization . The resulting event often appears as a pause at the level of fluorescence microscopy , although the protofilaments that are not occluded by KIF21B are still likely to be dynamic ( Figure 7J ) . The presence of multiple KIF21B molecules would result in blocking and stabilization of more protofilaments and thus more effective pausing , as we have observed . KIF21B-induced pausing events were typically followed by a catastrophe . This is in line with the slow retraction of the whole MT network observed in KIF21B-overexpressing cells ( Figure 1—figure supplement 1A ) , which can be explained by the gradual loss of tubulin subunits from KIF21B-stabilized MT plus ends . The effects induced by purified KIF21B in vitro are partly consistent with the recent analyses of MT plus-end dynamics in neuronal cells ( Ghiretti et al . , 2016; Muhia et al . , 2016 ) , in which a reduction of MT growth processivity was observed upon Kif21b knockout or depletion . This observation is in line with the idea that the presence of KIF21B induces either catastrophes or pausing , since both types of events would cause disappearance of an EB-positive comet . The effects on MT growth rate were opposite in the two studies: a decrease in MT growth rate was observed in knockout cells , while an increase was seen after RNA interference-mediated knockdown of KIF21B ( Ghiretti et al . , 2016; Muhia et al . , 2016 ) . These complexities might be due to some indirect effects on the tubulin pool or other MAPs , and are therefore difficult to compare to our in vitro analyses . Ghiretti et al . ( 2016 ) also carried out in vitro experiments with purified KIF21B and its fragments . Similar to our study , they have identified a MT-binding domain in the stalk of the kinesin , but since MT binding was only investigated by co-pelleting assays with stabilized MTs , the WD40-containing C-terminal MT-binding region was not detected in these experiments , consistent with our observation that this domain does not bind to taxol-stabilized MTs . We note that the results of the analyses of the effect of KIF21B on MT dynamics were very different from ours , as the full length KIF21B , although motile in cells , did not seem to display a motile behavior in vitro even at 300 nM concentration and mostly associated with depolymerizing MT ends . However , it could still increase the polymerization rate and promote catastrophes of growing MT plus ends ( Ghiretti et al . , 2016 ) . In contrast , in our experiments , 3–5 nM KIF21B was sufficient to block MT outgrowth from seeds at different tubulin concentrations . Furthermore , we did not observe KIF21B accumulation on depolymerizing MT ends , and the impact of KIF21B on MT growth ( pausing , growth perturbation or catastrophe induction ) was strongly associated with motile motors ‘catching up’ with growing MT plus ends . Finally , while in our experiments the KIF21B motor alone could slow down MT growth at 2–10 nM , Ghiretti et al . observed no effect of a similar construct even at 300 nM . We attribute these discrepancies to the differences in kinesin preparations and assay conditions ( e . g . different ionic strength of the assay buffer , which seemed to be significantly lower in the experiments by Ghiretti et al . than in our study ) . Our results suggest that in cells , KIF21B might use some additional factors for its loading onto MTs , and it is of course also possible that these or other factors would contribute to the association of KIF21B with MT plus ends . For example , the L-WD40 fragment of KIF21B fully decorates dynamic MTs in cells while it fails to do so in our in vitro assays , suggesting involvement of additional MAPs or post-translational modifications of tubulin . Further , an interesting implication of the observation that KIF21B is highly processive is that it is expected to display a stronger accumulation and thus a stronger pausing effect on the plus ends of longer MTs . Such a length-dependent effect would be similar to that described for other processive kinesins regulating MT plus-end dynamics ( Hibbel et al . , 2015; Su et al . , 2012; Varga et al . , 2006 ) . It is possible that MT length-dependent regulation of pausing or catastrophe might help to achieve more uniform MT lengths in long neurites of neuronal cells , where MT growth is not bounded by the cell margin . In addition , since KIF21B can bind to one MT and step on another one , it might also play a role in organizing MT arrays by sliding MTs against each other . KIF21B is thus an interesting player in the cell’s versatile toolbox responsible for MT-based transport and shaping of MT arrays . Changes in these arrays caused by alterations in KIF21B activity combined with its potential transport-related functions might explain the involvement of KIF21B in human diseases .
We used previously described COS-7 cells ( van Bergeijk et al . , 2015 ) and HEK293T cells ( Bouchet et al . , 2016 ) , which were cultured in DMEM/F10 ( 1/1 ratio , Lonza , Basel , Switzerland ) supplemented with 10% fetal calf serum and penicillin and streptomycin . The cell lines were routinely checked for mycoplasma contamination using LT07-518 Mycoalert assay ( Lonza ) . KIF21B expression constructs were made using human cDNA clone KIAA0449 ( Kazusa DNA Research Institute , Japan ) in pEGFP-N3 , pEGFP-C1 , mCherry-C1 or TagRFP-N3 vectors by PCR-based strategies . Additional TEV-protease recognition ( ENLYFQG ) and Biotinylation tag sequences ( MASGLNDIFEAQKIEWHEGGG ) were introduced in the EGFP vectors for protein purification purposes ( as described previously , ( van der Vaart et al . , 2013 ) ) . Biotin ligase BirA expression construct ( Driegen et al . , 2005 ) was a gift from D . Meijer ( University of Edinburgh , UK ) , EB3-TagRFP-T was described previously ( van der Vaart et al . , 2013 ) and TagRFP-α-tubulin was from Evrogen . Plasmids were transfected with polyethylenimine ( PEI ) or FuGene6 ( Roche , Basel , Switzerland ) . Rabbit-anti-GFP ( ab290 , Abcam , Cambridge , UK ) , mouse-anti-mCherry ( 632543 , Clontech , CA ) , rat-anti-α-tubulin ( YL1/2 ) ( MA1-80017 , Pierce Antibodies , MA ) were used on fixed cells and Western blotting . We used the following secondary antibodies: IRDye 800CW Goat anti-rabbit and anti-mouse ( Li-Cor Biosciences , Lincoln , NE ) , Alexa-488 and Alexa-568 conjugated goat antibodies against rat IgG ( Molecular Probes , Eugene , OR ) . For tubulin staining , COS-7 cells were fixed with –20°C methanol for 10 min and subsequently fixed with 4% paraformaldehyde ( PFA ) in phosphate-buffered saline ( PBS ) for 15 min at RT . Cell membranes were permeabilized with 0 . 1% Triton X-100 in PBS and washed with 0 . 1% Tween-20 in PBS . Blocking and labeling were done in 0 . 1% Tween-20 in PBS supplemented with 1% bovine albumin serum . Slides were rinsed with 70% ethanol in the last wash step , air-dried and mounded in Vectashield mounting medium ( Vector laboratories , Burlingame , CA ) . For the nocodazole wash-out , cells were treated with 5 µM nocodazole for 2 hr , subsequently washed four times and re-incubated in normal culture medium at 37°C for indicated time points . Cells were fixed and stained as described above . All cell biological experiments were performed at least twice . Bio-GFP-tagged bait constructs and mCherry-tagged prey constructs were cotransfected in HEK293T cells . A construct encoding BirA was co-transfected to induce biotinylation of the Bio-tag . Cell lysates were prepared in 20 mM Tris pH7 . 5 , 100 mM NaCl , 1% Triton-X100 , 1x cOmplete protease inhibitor cocktail tablet ( Roche ) and incubated with M-280 Streptavidin Dynabeads ( Invitrogen , CA ) for 1 hr . Samples were washed three times in 20 mM Tris pH 7 . 5 , 100 mM NaCl , 0 . 1% Triton-X100 and analyzed by SDS-PAGE and Western blotting . Constructs tagged with GFP-TEV-Bio were co-transfected with BirA in HEK293T cells as described for streptavidin pull down assays . Cell lysates were prepared in 50 mM Hepes pH 7 . 4 , 300 mM NaCl , 1 mM MgCl2 , 0 . 5% Triton-X100 , 1 mM DTT , 1x cOmplete protease inhibitor cocktail tablet ( Roche ) and incubated with M-280 Streptavidin Dynabeads ( Invitrogen ) for one hour . Samples were subsequently washed with 50 mM Hepes pH 7 . 4 , 300 mM NaCl , 1 mM MgCl2 , 0 . 5% Triton-X100 , 1 mM DTT three times and another three times with cleavage buffer ( 50 mM Hepes pH 7 . 4 , 150 mM NaCl , 1 mM MgCl2 , 0 . 05% Triton-X100 , 1 mM DTT , 1 mM EGTA ) , after which they were incubated in cleavage buffer supplemented with 40 ng/µl ( 770 nM ) TEV protease ( Sigma-Aldrich , St Louis , MO ) for 2 hr at 4°C . Supernatant was collected and stored at −80°C prior to use . Purity of the samples was analyzed via SDS-PAGE and Coomassie staining . Concentrations of stock solutions varied between ~50–250 nM for full-length KIF21B-GFP , KIF21B-FLΔrCC-GFP and KIF21B-MD-CCΔrCC-GFP , and 0 . 6–1 . 2 µM for MD-CC1-GFP and GFP-L-WD40 . Strep-tag-based KIF5B-560-GFP protein purification from HEK293T cells was done using Strep ( II ) -streptactin affinity purification method ( Sharma et al . , 2016 ) . cDNA encoding KIF21B rCC ( residues 930–1010 ) were PCR amplified from a human cDNA library ( Frey et al . , 2007 ) and cloned into the pET-based bacterial expression vector PSTCm1 ( Olieric et al . , 2010 ) . Subsequently , the protein was expressed in BL21 ( DE3 ) at 37°C grown in LB media supplemented with a mixture of 50 μg/ml kanamycin and 30 μg/ml chloramphenicol to an OD600 of 0 . 4–0 . 6 . Expression was induced with 0 . 5 mM isopropyl 1-thio-β- galactopyranoside ( IPTG; Sigma-Aldrich ) and grown overnight at 20°C . Cell pellets were resuspended in lysis buffer ( 50 mM HEPES , pH 8 , 500 mM NaCl , 10 mM Imidazole , 10% glycerol , 2 mM β-mercaptoethanol and 1 cOmplete EDTA-free protease inhibitor cocktail tablet ( Roche ) and lysed on ice by ultrasonication . Lysates were cleared by ultracentrifugation . Resulting supernatants were subsequently filtered ( 0 . 45 µm filter ) . The protein was affinity purified by IMAC on a 5 ml HisTrap FF Crude column ( GE Healthcare , Chicago , Illinois ) according to manufacturer’s instructions . The 6xHis tag was cleaved using 2 units of human thrombin ( Sigma-Aldrich ) per milligram of recombinant protein and cleavage was performed over night at 4°C by dialysis in thrombin cleavage buffer ( 20 mM Tris-HCl pH 7 . 4 , 150 mM NaCl , 2 . 5 mM CaCl2 and 2 mM DTT ) . The 6xHis tag was separated from the target protein by re-application to the IMAC column . The processed protein was concentrated and further purified by size exclusion chromatography on a HiLoad Superdex 75 16/60 size-exclusion column ( GE Healthcare ) equilibrated in 20 mM Tris-HCl , pH 7 . 5 , supplemented with 150 mM NaCl and 2 mM DTT . CD spectra were recorded at 5°C and at a protein concentration of 0 . 166 mg/ml in PBS supplemented with 0 . 5 mM TCEP using a Chirascan spectropolarimeter ( Applied Photophysics Ltd , Leatherhead , UK ) and a cuvette of 0 . 1 cm path length . Thermal unfolding profiles between 5°C and 90°C were recorded by increasing the temperature at a ramping rate of 1°C/min monitoring the CD signal at 222 nm . Midpoints of thermal unfolding were calculated using the Glob3 program ( Applied Photophysics ) . Sedimentation velocity experiments were performed at 20°C and 42 , 000 rpm in Tris-HCl , pH 7 . 5 , 150 mM NaCl , 2 mM DTT using a Beckman XLI analytical ultracentrifuge ( Beckman Coulter Inc . , CA ) . Sedimentation profiles were recorded by UV absorbance ( 280 nm ) and interference scanning optics . The partial specific volume of the samples as well as the density and viscosity of the buffer were calculated with SEDNTERP ( http://sednterp . unh . edu/ ) . Data were fitted with SEDFIT ( Schuck , 2000 ) using the continuous distribution model . Graphical representations were processed with GUSSI ( biophysics . swmed . edu/MBR/software . html ) . In vitro assays were performed as described previously ( van der Vaart et al . , 2013 ) . MT seeds were grown using 20 μM tubulin mix containing 18% biotin-tubulin and 12% Rhodamine- or HiLyte Fluor 488-tubulin ( Cytoskeleton , Inc . , Denver , CO ) and 1 mM GMPCPP by polymerization at 37°C for 30 min , pelleting by centrifugation in an Airfuge for 5 min and depolymerization on ice . After a subsequent round of polymerization and pelleting , seeds were stored in MRB80 buffer ( 80 mM K-PIPES , pH 6 . 8 , 4 mM MgCl2 , 1 mM EGTA ) with 10% glycerol . Flow chambers were made with microscopy slides and plasma-cleaned glass coverslips . Coating was done with 0 . 2 mg/ml PLL-PEG-biotin ( Surface Solutions , Dübendorf , Switzerland ) in MRB80 buffer and 0 . 8 mg/ml NeutrAvidin for 5 min each . The seeds were attached to the coverslips via biotin-NeutrAvidin links and blocked with 0 . 8 mg/ml κ-casein . Reaction mixtures consisting of MRB80 supplemented with different concentrations of tubulin ( indicated in figure legends ) , containing 3% Rhodamine-tubulin when indicated , 50 mM KCl , 0 . 1% methylcellulose , 0 . 5 mg/ml κ-casein , 1 mM GTP , oxygen scavenging system ( 20 mM glucose , 200 μg/ml catalase , 400 μg/ml glucose-oxidase , 4 mM DTT ) , 2 mM ATP , 20 or 100 nM mCherry-EB3 when indicated and the specified concentration of purified GFP or KIF21B-GFP proteins ( stored in 50 mM Hepes pH 7 . 4 , 150 mM NaCl , 1 mM MgCl2 , 0 . 05% Triton-X100 , 1 mM DTT , 1 mM EGTA , 770 nM TEV protease and diluted by at least 10 or more times in MRB80 buffer for in vitro assays ) or KIF5B-560 ( stored in 50 mM Hepes pH 7 . 4 , 150 mM NaCl , 1 mM MgCl2 , 0 . 05% Triton-X100 , 1 mM DTT , 1 mM EGTA , 2 . 5 mM d-Desthiobiotin and diluted 20 times in MRB80 buffer for in vitro assays ) were added to the flow chambers . Movies were collected using TIRF microscopy . For mCherry-CC2 and mCherry-L-WD40 , extracts of HEK293T cells expressing the proteins , prepared in MBR80 supplemented with 1x cOmplete protease inhibitor cocktail tablet ( Roche ) and 1% Triton-X100 , were used in the reaction mixture in a ratio of 1:4 . All samples were incubated at 30°C during imaging . The quantitative data reported for each experiment were collected in at least two or more independent assays . Fixed cells were imaged with a Nikon Eclipse 80i upright fluorescence microscope equipped with Plan Apo VC N . A . 1 . 40 oil 100x and 60x objectives , or Nikon Eclipse Ni-E upright fluorescence microscope equipped with Plan Apo Lambda 100x N . A . 1 . 45 oil and 60x N . A . 1 . 40 oil objectives microscopes , Chroma ET-BFP2 , - GFP or -mCherry filters and Photometrics CoolSNAP HQ2 CCD ( Roper Scientific , Trenton , NJ ) camera . The microscopes were controlled by Nikon NIS Br software . Live cell imaging and in vitro assays were performed on an inverted research microscope Nikon Eclipse Ti-E ( Nikon ) with the perfect focus system ( PFS ) ( Nikon ) , equipped with Nikon CFI Apo TIRF 100 × 1 . 49 N . A . oil objective ( Nikon , Tokyo , Japan ) , Photometrics Evolve 512 EMCCD ( Roper Scientific ) and Photometrics CoolSNAP HQ2 CCD ( Roper Scientific ) and controlled with MetaMorph 7 . 7 software ( Molecular Devices , CA ) . The microscope was equipped with TIRF-E motorized TIRF illuminator modified by Roper Scientific France/PICT-IBiSA , Institut Curie , and an ET-GFP filter set ( Chroma , Bellow Falls , VT ) for imaging of GFP-tagged proteins . For simultaneous imaging of green and red fluorescence we used triple-band TIRF polychroic ZT405/488/561rpc ( Chroma ) and triple-band laser emission filter ZET405/488/561m ( Chroma ) , mounted in the metal cube ( Chroma , 91032 ) together with Optosplit III beamsplitter ( Cairn Research Ltd , Faversham , UK ) equipped with double emission filter cube configured with ET525/50m , ET630/75m and T585LPXR ( Chroma ) . Long-term imaging was performed on an inverted research microscope Nikon Ti equipped with Plan Fluor 40x/1 . 30 Oil DIC and Plan Apochromat 20 × 0 . 75 Phase Contrast objectives , ET-GFP ( 49002 ) and ET-mCherry ( 49008 ) filters ( Chroma ) and controlled with MicroManager . Cells were kept at 37°C , and in vitro samples at 30°C in a Tokai Hit INUBG2E-ZILCS Stage Top Incubator . Images and movies were processed using ImageJ . All images were modified by adjustments of brightness and contrast; smooth and sharp masks were applied in some cases . Maximum intensity projections were made using z projection . MT growth rates and kinesin velocities were obtained from kymograph analysis , using ImageJ plugin KymoResliceWide v . 0 . 4 https://github . com/ekatrukha/KymoResliceWide ( Katrukha , 2015 ) ; copy archived at https://github . com/elifesciences-publications/KymoResliceWide ) . Results were plotted in Graphpad Prism 6 . Statistical analysis was performed using non-parametric Mann-Whitney U-test . To build single molecule fluorescence histograms ( Figure 2A , Figure 3—figure supplements 1A ) , purified GFP or GFP-fusion proteins were diluted in phosphate buffered saline ( PBS ) and added to the different imaging flow chambers of the same plasma cleaned coverslips . Chambers were subsequently washed with PBS , leaving a fraction of the GFP-tagged proteins immobilized on the coverslip . After sealing with vacuum grease to prevent evaporation , samples were imaged at room temperature using TIRF microscopy . Protein dilution was optimized to provide images of 0 . 1–0 . 4 fluorophores per µm2 for each condition . At least 20 images were acquired at different positions on the coverslip to avoid pre-bleaching . ImageJ plugin DoM_Utrecht v . 0 . 9 . 1 https://github . com/ekatrukha/DoM_Utrecht ( Katrukha et al . , 2016 ) ; copy archived at https://github . com/elifesciences-publications/DoM_Utrecht ) was used for detection and fitting of single molecule fluorescent spots as described previously ( Yau et al . , 2014 ) . . In short , individual spots were fitted with 2D Gaussian and the amplitude of the fitted Gaussian function was used as a measure of the fluorescence intensity value of an individual spot . The same parameter was used to build histograms in Figure 2A , Figure 3—figure supplements 1A and 7A . The histograms were fitted to lognormal distributions using GraphPad Prism 6 . To estimate the number of GFP molecules per kinesin ( Figure 5B–H ) , KIF5B-560 and KIF21B-FL moving along the lattice of MTs in the different imaging flow chambers of the same plasma cleaned coverslips were analyzed and compared . ImageJ plugin DoM_Utrecht v . 0 . 9 . 1 was used for detection and fitting of single molecule fluorescent spots . The histograms were fitted to lognormal distributions using GraphPad Prism 6 . For MT pausing events induced by kinesin ( Figure 5A ) , particles on the MT lattice and tip were detected using ComDet v . 0 . 3 . 5 https://github . com/ekatrukha/ComDet ( Katrukha , 2016 ) ; copy archived at https://github . com/elifesciences-publications/ComDet ) and DoM_Utrecht v . 0 . 9 . 1 ( Katrukha et al . , 2016 ) ImageJ plugins . Their single molecule intensity values were normalized to the average intensity of GFP-kinesins non-specifically attached to the coverslip in areas devoid of MTs . Kinesin intensity and position for Figure 3C , Figure 3—figure supplement 2 , Figure 4A , B were calculated using the same plugins . The position of MT tip was estimated from fitting of each x-profile of corresponding kymograph with error function with offset using custom written Matlab script . All mentioned ImageJ plugins have source code available and are licensed under open-source GNU GPL v3 license . To monitor the decay of fluorescence caused by photobleaching , single particles of KIF21B-FL-GFP were immobilized on the surface of a coverslip . Movies were recorded at 100 ms/stream acquisition for 100s or at 100 ms/1 frame per 1 . 2 s for 720s with low laser power ( same laser power used for imaging in Figures 3–5 and 7 ) to allow KIF21B-FL-GFP to be photobleached . GraphPad Prism 6 software was used for the data fitting . Taxol- or GMPCPP-stabilized MTs were polymerized to a final concentration of 10 μM as described ( Kevenaar et al . , 2016 ) . Afterwards , 50 µl of 0 . 19 µM of HEK293T purified full length KIF21B was incubated together with 0 . 45 μM stabilized MTs for 10 min at room temperature . As a control , the same amount of MTs was incubated separately . A taxol-glycerol cushion containing 55% 2x BRB80 buffer ( 80 mM K-PIPES , pH 6 . 8 , 1 mM EGTA , 1 mM MgCl2 ) , 44% glycerol and 6% 2 mM paclitaxel was added to the centrifugation tubes prior to sample addition . After centrifugation of the samples at 174 , 500 x g for 10 min , the supernatants were carefully removed and the pellets were resuspended in 50 μl BRB80 buffer . Twenty microliters of each supernatant and pellet were mixed with 5 μl 5x SDS loading dye and analyzed on Coomassie stained 7 . 5% SDS-PAGE . For negative staining electron microscopy , 5 μl aliquots of pellet samples prepared in the presence of either 1 µM taxol or 1 µM GMPCPP were transferred to freshly UV activated homemade carbon-coated copper grids . After 20 s of incubation , excess liquid was removed by side-blotting and the grids were washed twice with BRB80 buffer supplemented with either 1 µM taxol or 1 µM GMPCPP and once with double distilled water . Subsequently , the grid was stained three times with a freshly prepared uranyl acetate solution . Electron micrographs were taken at a nominal magnification of 40 k with a JEM2200FS ( JEOL , Peabody , MA ) electron microscope operated at 200 kV and equipped with a TVIPS F416 camera . Samples of purified proteins were ran on SDS-PAGE gel ( 150 ng FL , 30 ng FLΔrCC , 45 ng MD-CCΔrCC , 255 ng L-WD40 , 150 ng MD-CC1 ) . After in-gel digestion , samples were resuspended in 10% formic acid ( FA ) /5% DMSO and were analyzed with an Agilent 1290 Infinity ( Agilent Technologies , CA ) LC , operating in reverse-phase ( C18 ) mode , coupled to a TripleTOF 5600 ( AB Sciex , Nieuwerkerk aan de IJssel , Netherlands ) . MS spectra ( 350–1250 m/z ) were acquired in high-resolution mode ( R > 30 , 000 ) , whereas MS2 in high-sensitivity mode ( R > 15 000 ) . Raw files were processed using Proteome Discoverer 1 . 4 ( version 1 . 4 . 0 . 288 , Thermo Scientific , Bremen , Germany ) . The database search was performed using Mascot ( version 2 . 4 . 1 , Matrix Science , UK ) against a Swiss-Prot database ( taxonomy human ) . Carbamidomethylation of cysteines was set as a fixed modification and oxidation of methionine was set as a variable modification . Trypsin was specified as enzyme and up to two miss cleavages were allowed . Data filtering was performed using percolator , resulting in 1% false discovery rate ( FDR ) . Additional filters were; search engine rank 1 peptides and ion score >20 . | Microtubules are tiny tubes that cells use as rails to move various cell compartments and structures to different locations within the cell . They are made of building blocks called tubulin and form extensive networks across the cell . Depending on the cell’s needs , microtubule networks can be rapidly assembled and disassembled by adding or removing tubulin subunits at the ends of individual microtubules . While a lot is known about how cells regulate the growth and shrinkage of microtubules , much less is known about the factors that can pause these processes and thus stabilize a microtubule . Proteins belonging to the kinesin family are molecular motors that can walk along microtubules and control how microtubules grow and shrink . A kinesin known as KIF21B is found in several types of cells including neurons and immune cells and genetic alterations in this protein have been linked with several neurodegenerative diseases . KIF21B is made up of three regions: a motor domain , a stalk and a tail domain that binds to microtubules . Recent studies have suggested that this kinesin affects the ability of one end of microtubules ( known as the plus end ) to grow . Here , van Riel , Rai , Bianchi et al . used a biochemical approach to investigate the activity of KIF21B . The experiments show that KIF21B can walk to the plus end of microtubules and efficiently pause growth . Small numbers of KIF21B molecules are enough to inhibit microtubule growth and this activity depends on the motor domain and the tail domain of KIF21B working together . These experiments were performed a cell-free system and so the next challenge is to investigate how KIF21B works in living cells , including neurons and immune cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Kinesin-4 KIF21B is a potent microtubule pausing factor |
The functions performed by the concentric shells of multilayered dsRNA viruses require specific protein interactions that can be directly explored through their mechanical properties . We studied the stiffness , breaking force , critical strain and mechanical fatigue of individual Triple , Double and Single layered rotavirus ( RV ) particles . Our results , in combination with Finite Element simulations , demonstrate that the mechanics of the external layer provides the resistance needed to counteract the stringent conditions of extracellular media . Our experiments , in combination with electrostatic analyses , reveal a strong interaction between the two outer layers and how it is suppressed by the removal of calcium ions , a key step for transcription initiation . The intermediate layer presents weak hydrophobic interactions with the inner layer that allow the assembly and favor the conformational dynamics needed for transcription . Our work shows how the biophysical properties of the three shells are finely tuned to produce an infective RV virion .
The advent of single-molecule techniques have opened the door to understand how the mechanics of biomolecular assemblies is essential for their function ( Howard , 2001; Müller et al . , 2002 ) . In the case of viruses , the infectious particle must be robust enough to protect the viral genome outside the cell but also competent to undergo the required structural changes once the host cell is recognized , overcome its barriers and carry out the events necessary for a productive viral replication cycle ( Flint et al . , 2004 ) . Double-stranded RNA ( dsRNA ) viruses have a number of common challenges derived from the very nature of their genome . Specifically , since there are no host cell enzymes that can recognize dsRNA as template for transcription , the viral particle must incorporate a transcription machinery able to synthesize the required mRNAs to initiate the viral replication cycle . In addition , dsRNA is an inducer of the innate cell-based antiviral response , including interferon synthesis and apoptosis ( Mertens , 2004; Arnold et al . , 2013 ) . The virus must evade the host sentinels that trigger these mechanisms and control the host response ( Akira et al . , 2006; Frias et al . , 2012 ) . Most dsRNA viruses exhibit a common solution to these problems , which consists of the assembly of a stable protein cage in the host cytoplasm that isolates the viral dsRNA molecules to prevent the cellular antiviral response . This cage ( the viral core ) incorporates the necessary enzymes for transcription and replication of the dsRNA genome , which are accomplished without disassembly the particle . This core presents a common architecture that consists of an icosahedral T = 1 shell formed by 60 asymmetric dimers ( a 120-subunit capsid ) ( Jaing et al . , 2008 ) present in most of the dsRNA virus families ( King et al . , 2011 ) . Most of these viruses present a single protein shell and lack an extracellular cycle ( Ghabrial et al . , 2015 ) . However , Cystoviridae and Reoviridae families display concentric protein layers surrounding the core that are responsible of host cell recognition , entry , etc . This modularity facilitates the study of the relationship between the layer functions , their structure and physical properties . RV , a major causative agent of severe dehydrating diarrhea in children under five years ( GBD Diarrhoeal Diseases Collaborators , 2017 ) , is a multilayered virus of clinical relevance and one of the main study models for the Reoviridae family . The RV infectious particle is a 100 nm non-enveloped triple-layered particle ( TLP ) composed of three concentric protein shells enclosing the dsRNA genome and the viral RNA polymerase and capping enzyme ( Figure 1A ) ( Settembre et al . , 2011 ) . The inner layer is a T = 1 capsid formed by 60 asymmetric dimers of the VP2 protein ( 102 kDa ) that surrounds the eleven dsRNA genomic segments associated with the RNA-dependent RNA-polymerase VP1 ( 125 kDa ) and the RNA-capping enzyme VP3 ( 88 kDa ) at the pentameric positions ( Estrozi et al . , 2013; Periz et al . , 2013 ) . This thin single-layered particle ( SLP ) , an intermediate structure that is involved in the packing and replication of the viral genome , is surrounded by a thick T = 13 layer formed by 260 VP6 pear-shaped trimers ( 45 kDa ) ( Settembre et al . , 2011; McClain et al . , 2010 ) in the double-layered particle ( DLP ) . This particle , which does not disassemble during the infection , constitutes the transcriptional machinery that initiates the core steps of the viral replication cycle once delivered in the host cell cytoplasm ( Cohen et al . , 1979; Bass et al . , 1992; Lawton et al . , 1997 ) . The DLP is not infectious since it cannot recognize , bind to and penetrate the host target cell . These abilities are incorporated in the outer layer of the TLP formed by VP4 and VP7 . The VP7 glycoprotein is organized as 260 Ca2+-stabilized trimers that cap and embrace through its N-terminal arm each VP6 trimer of the DLP ( Settembre et al . , 2011; Chen et al . , 2009 ) . Sixty spikes are anchored on the VP6 layer depressions that surround the pentameric positions and are clamped by the VP7 layer . The viral spike is formed by three copies of VP4 that must be proteolytically processed to VP5* and VP8* by trypsin-like proteases from the intestinal lumen or from within cells to generate a fully-infectious virion ( Settembre et al . , 2011; Estes et al . , 1979; Estes et al . , 1981; Clark et al . , 1981 ) . Interestingly , the assembly of the VP6 T = 13 layer on the 60 VP2 dimers ( T = 1 ) that build the SLP is one of the best examples of symmetry mismatch , of which the consequences for virus particle stability are still not well understood . This mismatch is preserved in most reoviruses , and has been associated with the regulation of the polymerase activity ( McClain et al . , 2010 ) . In contrast with the plethora of information obtained during 30 years of structural studies on the particle components ( Trask et al . , 2012 ) , little is known about the mechanical properties of the RV particle layers , subviral particles and TLP , and their mutual influence in contributing to the virus stability along its cycle . Both the application ( Rief et al . , 1997; Perrino and Garcia , 2016 ) and measurement ( Hua et al . , 2002; Alsteens et al . , 2017 ) of forces on single molecules are key methodologies to decipher the function of biomolecular systems . Specifically , the study of viral capsids by Atomic Force Microscopy ( AFM ) enables the exploration of physicochemical properties , such as mechanics and electrostatics , in liquid milieu by using a sharp tip attached to a cantilever to probe individual particles ( Roos et al . , 2010 ) . Single indentation assay consists on deforming a virus particle with the AFM tip while recording the cantilever bending vs . the virus deformation to induce the virus breakage ( Roos , 2018 ) . The force-indentation curves ( FIC ) so obtained inform about the virus stiffness or spring constant ( elasticity ) , breaking force and brittleness . AFM also allows applying repetitive loading cycles to individual viruses at low force ( ~100 pN ) which results in mechanical fatigue experiments ( Moreno-Madrid et al . , 2017 ) . AFM directly probed the existence of pressure ( Kindt et al . , 2001; Smith et al . , 2001 ) in some phages ( Evilevitch et al . , 2011; Hernando-Pérez et al . , 2012 ) that is used to translocate their genome into the host ( González-Huici et al . , 2004 ) . In a similar way , it has been found that human adenovirus pressurizes during maturation , and that this pressure is related to the degree of condensation of the dsDNA of the viral minichromosome ( Ortega-Esteban et al . , 2015a; Ortega-Esteban et al . , 2015b ) . In addition , the influence of both homologous ( Mertens et al . , 2015; Zeng et al . , 2017a ) and heterologous ( Llauró et al . , 2016a; Snijder et al . , 2016 ) cargos have been explored in virus mechanics . The alteration of the capsid structure with maturation ( Roos et al . , 2012; Hernando-Pérez et al . , 2014a ) , mutations ( Castellanos et al . , 2012; van Rosmalen et al . , 2018 ) or cementing proteins ( Hernando-Pérez et al . , 2014b; Llauró et al . , 2016b ) also influences virus mechanics . However , these studies have been never applied to multilayered virus particles , which enable direct measurements of the inter-layer interactions magnitude . Here , we explore the mechanical properties of individual TLP , DLP and SLP particles by single indentation assay and probe their stability against mechanical fatigue . Our experiments , in combination with Finite Element ( FE ) analysis , the atomic structure of the layers and the calculation of the electrostatic properties of each particle , allow to probe and interpret the intra and interlayer interactions and relate them to their role during the virus replication cycle .
Previous studies have shown that RV TLP can be converted to DLP by disassembling the outer VP4-VP7 layer with chelating agents such as ethylenediaminetetraacetic acid ( EDTA ) ( Estes et al . , 1979 ) . Once purified , DLP can be converted to SLP by chaotropic agents such as CaCl2 ( Figure 1B ) ( Bican et al . , 1982 ) . TLP were purified from infected cells , and DLP and SLP were produced and purified combining the above described treatments with several ultracentrifugation steps to remove the proteins of the disassembled layers . Homogeneous populations of TLP , DLP and SLP were obtained , as indicated by SDS-PAGE and negative staining electron microscopy analysis ( Figure 1C–E ) . Spike polypeptides ( VP5*/VP8* ) and VP7 glycoprotein are totally removed in purified DLP ( Figure 1D ) while VP6 is absent in the isolated cores ( Figure 1E ) . After the adsorption of particles on substrate , we used AFM in jumping mode ( Ortega-Esteban et al . , 2012 ) for the topographical characterization of individual particles in liquid . Our high resolution images ( Figure 2 ) are compatible with the structures obtained from cryo-EM ( Settembre et al . , 2011; Zhang et al . , 2008 ) and x-ray ( McClain et al . , 2010 ) , where thousands and millions of particles are averaged , respectively . Spikes protruding from the TLP ( Figure 2A ) as well as the DLP pentameric and hexameric depressions ( Figure 2B ) are resolved . In contrast , SLP offers featureless structure ( Figure 2C ) . Although the distinctive topography of TLP allows their unambiguous identification , they exhibit a broad distribution of height values ( Figure 2D–E ) . This behavior is probably due to the number of spikes ( Figure 2—figure supplement 1 ) present at the interface between the particles and the substrate surface and to the mode of how they influence the particle adsorption . The height data of TLP ( Figure 2E , red ) suggest two populations centered at ~74 and ~62 nm , represented by filled and empty symbols , respectively . We propose that these data correspond to the presence ( red filled squares , Figure 2E ) or absence ( red empty squares , Figure 2E ) of spikes at the particle-surface interface . In the first case the presence of spikes would prevent partially the contact between the VP7 layer and the substrate ( Figure 2—figure supplement 2 ) , thus precluding virus adsorption and deformation . However , when the VP7 layer directly rests on the surface , TLP collapse to an average height value of 62 nm ( Figure 2 ) probably due to a strong VP7-surface interaction ( Zeng et al . , 2017a ) . In contrast , DLP and SLP present a narrower height distribution ( Figure 2E ) whose average values are compatible with the nominal values ( 70 nm for DLP and 55 nm for SLP ) , indicating a little deformation due to the adsorption on the surface of 6% and 2% for DLP and SLP respectively ( Llauró et al . , 2015; Zeng et al . , 2017b ) . Although icosahedral symmetry imposition renders an ideal RV particle with 60 trimeric spikes ( Figure 1A and Figure 1—figure supplement 1 ) , previous studies have shown that some positions are unoccupied in the purified TLP ( Chen and Ramig , 1992; Trask and Dormitzer , 2006; Rodríguez et al . , 2014 ) . To estimate the amount of spike protein in TLP , VP5* was quantified relative to protein VP6 ( occupancy ) . Densitometric analysis of Coomassie-stained gels ( Figure 1C ) produced an occupancy of 52% . Cryo-EM analysis and three-dimensional reconstruction ( 3DR ) of these TLP showed an equivalent occupancy ( ~50% ) when the relative density of the spikes in the 3DR is determined using the VP2-VP6-VP7 shell density as a reference ( Figure 1—figure supplement 1 ) . This occupancy correlates with the different number of spikes detected in the AFM images of single TLP ( Figure 2—figure supplement 1 ) . Since lateral spikes are easily removed by the AFM tip ( Video 1 ) , we analyzed the upper ~1/3 region of the virus surface , where the spikes point upwards and present a maximum resistance to AFM imaging . Although we cannot ignore that the AFM tip could remove some spikes , we minimized this effect by using the first image obtained for each particle . Our AFM topographies , which uniquely allow for the first time the direct imaging of the individual spikes , provide a more realistic view of the RV virion as a distribution ranging from fully decorated to almost naked particles . ( Figure 2 and Figure 2—figure supplement 1 ) . We can directly observe an average occupancy of 35% , compatible with electrophoretic and cryo-EM bulk analysis results . These data support the in vitro recoating assays , demonstrating that an occupancy as low as 10% is enough to generate particles with high specific infectivity ( Trask and Dormitzer , 2006 ) . In order to investigate the contribution of the different layers to the mechanical stability of the RV particle , systematic single indentation experiments of the different particles were performed ( Figure 3 ) resulting in broken structures . In order to understand the nature of each particle breakage it is interesting to compare their topographies before and after fracture ( Figure 3A–C ) , and to consider the average indentation curves for each type of structure ( Figure 3D , strong colors ) . While TLP breaks into large fragments ( Figure 3A , right ) , both DLP and SLP show circular deformations that can be attributed to the tip apex ( Figure 3B–C , right ) . The average of TLP nanoindentation curves ( Figure 3D , strong red ) shows a linear regime that corresponds to the virus elastic deformation up to ~2 . 0 nm at ~2 . 1 nN , where the elastic limit is reached . Afterwards the structure yields plastically without breaking until 2 . 5 nN at 4 . 7 nm , during 2 . 7 nm ( red double headed arrow , Figure 3D ) , where the downwards slope indicates fracture . The same reasoning applied to both DLP and SLP result in plastic deformations of ~13 nm and ~39 nm , respectively ( blue and green double headed arrows , Figure 3D ) . Virus topographies and indentation assays indicate that while TLP undergoes a brittle ( glass-like ) fracture , both DLP and SLP experience ductile ( rubber-like ) breakage . The analysis of single particle FIC charts ( Figure 3D ) provides some mechanical parameters . In particular , the linear fitting of the curves before reaching the elastic limit informs about the particle elastic constant or stiffness ( Figure 4A ) . Statistical analysis of the FIC linear part result in spring constants of kTLP = 0 . 76 ± 0 . 30 N/m , kDLP = 0 . 34 ± 0 . 20 N/m and kSLP = 0 . 22 ± 0 . 07 N/m . The elastic limit can be linked to the breaking force of the probed particle . The analysis of the breaking force provides values of 2 . 9 ± 0 . 5 nN , 0 . 9 ± 0 . 3 nN and 0 . 45 ± 0 . 10 nN for TLP , DLP and SLP , respectively ( Figure 4B ) . This monotonic decrease of both the spring constant and breaking force with the reduction of the number of layers indicates that virus mechanics captures the reinforcement nature of concentric shells: the more layers in the structure , the stronger it becomes . The calculation of the yield strain ε=Δhh0 ( Figure 4—figure supplement 1 ) , where Δh is the indentation corresponding to the force at the elastic limit and h0 the height of the intact particle , reveals that TLP , although with a high dispersion , can sustain larger elastic deformations than DLP and SLP . This simultaneous high rigidity and yield strain is exceptional since an increment in the spring constant and breaking force is usually associated with a lower yield strain as it happens , for example , with glass ( Schijve , 2009 ) . The analysis of how the TLP and DLP inform upon the mechanical properties of VP6 and VP7 layers has to be considered with care . The only layer for which an individualized analysis of the mechanical properties can be performed is the SLP . Although the genome and the replication/transcription machinery reside inside the VP2 shell , it is expected that they have no relevant effect on the particle's response to deformation . The relatively low packing fraction of RV ( ~20% ) compared to pressurized dsDNA viruses ( Purohit et al . , 2005 ) suggests a small pressure whose influence on the effective elastic constant will be smaller than our error bars . In any case , the presence of the core would only affect to the estimation of the Young’s Modulus of the VP2 layer , but not to the inferred properties of the VP6 and VP7 layers . For TLP and DLP , the isolation of the mechanical parameters for VP7 and VP6 layers is also complex because they include internal shells with their mutual interactions . Specifically , the mechanical response of DLP is due to the VP6 shell and the internal SLP , whereas in the TLP there is an additional contribution of the VP7 layer . Taking this into account , FE simulations ( Gibbons and Klug , 2008 ) were performed to extract the effective Young’s moduli for the different capsid layers from the measured spring constants kSLP , kDLP , kTLP ( see Materials and methods and Figure 4—figure supplement 1 ) . The nanoindentation of SLP was implemented first , yielding a value for the Young’s modulus of YVP2 = 0 . 53 ± 0 . 20 GPa . A second layer of 8 nm thickness , representing that of VP6 , was added on top of the VP2 layer , and a Young’s modulus YVP6 = 0 . 08 ± 0 . 07 GPa was needed to recover the spring constant of DLP , kDLP . Finally , a 3 . 5 nm thick third layer was placed on top of the DLP , requiring a Young’s modulus YVP7 = 1 . 0 ± 0 . 9 GPa to yield the same spring constant as the TLP , kTLP . We can compare the Young’s modulus between layers resulting in YVP7 / YVP6=12 and YVP2 / YVP6=6 . 5 . Figure 4C shows the map of the stress supported by the constituent layers of each subviral particle , demonstrating that the VP7 layer accumulates most of the stress in TLP . Thus , nanoindentation experiments and FE analysis , indicate the VP7 shell to be the stiffer layer of the RV structure , but also the most elastic , whereas the thick VP6 layer is remarkably soft and brittle . While the single indentation assay probes the global mechanical response of virus particles , fatigue experiments explore the local response of the virus building blocks ( capsomers ) . Mechanical fatigue experiments are performed by applying cyclic forces of ~100 pN , well below the breaking force ( ~1 nN ) , at every pixel of the virus ( Ortega-Esteban et al . , 2013 ) and the gradual disassembly of viral particles is typically induced ( Hernando-Pérez et al . , 2014a ) . Cyclic imaging of the TLP ( Figure 5A , left ) at forces between 100 pN to 200 pN per pixel shows that , while the VP4 spikes are removed from the particle surface in a few frames ( Figure 5A , middle and Video 1 ) , the VP7 layer remains mostly intact ( Figure 5A , right ) during 80 frames ( light red in Figure 5E , Figure 5—figure supplement 1 and Video 5 ) . These results illustrate that the spikes are easily removed by the AFM tip and are not strongly anchored . However , the VP7 layer displays a strong resistance against fatigue , in agreement with the high stiffness and breaking force demonstrated in single indentation assay experiments . A strong binding energy between capsomers would not only result in a high resistance of individual proteins against fatigue , but also will contribute to a high breaking force when all capsomers are probed in a single indentation assay experiment . We have found similar results before in lambda phage ( Hernando-Pérez et al . , 2014a ) . The current model proposes a calcium concentration drop in endosomal compartments during RV entry as the factor that triggers VP7 disassembly and membrane penetration ( Arias et al . , 2015 ) . In fact , calcium depletion by chelating agents ( as EDTA ) ( Estes et al . , 1979 ) is used to uncoat TLP to DLP by inducing VP7 trimer dissociation ( Figure 1 ) . To explore the structural consequences of this process in real time , we carried out fatigue assays on TLP while EDTA simultaneously flowed in the AFM liquid chamber , as described in Material sand methods , to induce the gradual depletion of Ca ions of the particles ( Figure 5B , Video 2 ) . In these conditions , fatigue induces the neat VP7 detachment from the VP6 subjacent layer ( indicated by a circle in Figure 5B#19 ) even before the spikes are removed . Indeed , the evolution of the topographic profiles ( dark red in Figure 5E ) show abrupt downwards steps very close to the VP7 thickness ( red arrow of Figure 5E ) indicating that TLP particle loses VP7 completely while keeping VP6 ( Video 2 ) . These results not only suggest that Ca ions mediate the interaction between VP7 and VP6 layers , but also that the absence of ions weakens the interaction between VP7 subunits . If fatigue continues , VP6 subunits are neatly removed from VP2 layer ( circle in Figure 5B#29 ) . Therefore , VP6 shell appears as a weak shell whose interaction with the beneath VP2 layer is not very strong , since it peels off rapidly to reveal the SLP . We found similar results on experiments performed on DLP . Again , fatigue induced a clean VP6 disassembly after less than 10 frames ( circle in Figure 5C#8 , and Video 3 ) . In this case the evolution of the topographic profiles ( blue in Figure 5E ) undergoes sharp reductions very close to the VP6 thickness , inducing the gradual uncovering of the innermost VP2 ( blue arrow in Figure 5E ) . These experiments not only illustrate a weak interaction between VP6 and VP2 layers , but also a very feeble VP6-VP6 binding force . Finally , the thin SLP VP2 is highly unstable under fatigue experiments collapsing well before reaching 10 frames ( green in Figure 5D , and Video 4 ) . We have seen that removing of Ca ions is key for inducing the transition from TLP to DLP in the fatigue experiments ( Figure 5A–B , and red charts in Figure 5E ) . To access to VP6 and VP2 layers in the presence of Ca ions , we combined single indentation with fatigue assays . Our aim is to produce local disruptions in the TLP shell by performing a controlled FIC and then monitor the progressive disassembly induced by fatigue experiments . Therefore , we intent to crack the three layers at once without tearing apart the particle like in Figure 3A , by adjusting the indentation up to 40 nm ( data not shown ) after imaging the TLP 24 times ( Figure 6A#24 ) . Right after the FIC ( Figure 6A#25 ) , the induced fracture reaches a maximum depth of ~23 nm that includes the thickness of the three layers ( Figure 6B ) . However , the shape of the crack shows that some VP2 layer has been exposed ( Figure 6B , dotted line ) and its distance to the VP7 layer external face is compatible with the thickness of VP7 and VP6 layers ( 3 , 5 and 8 nm , respectively ) . The subsequent fatigue cycles increased the VP2 uncovered area ( Figure 6A#56 , Video 6 ) without any signature of the TLP-DLP transition . These experiments indicate that VP6 hardly survives to VP7 removal , supporting a strong interaction between VP7 and VP6 layers in the presence of Ca and , once again , a weak binding between VP6 and VP2 .
The characterization of the biophysical properties of viral particles has proven to be a powerful approach to understand the connection between structure and function in different systems ( Moreno-Madrid et al . , 2017 ) . Our mechanical analysis of the multilayered RV particle offers new opportunities to explore the interplay between structure , function and mechanics . In particular , the atomic structure of the layers provided by X-ray crystallography and cryo-EM ( Settembre et al . , 2011; McClain et al . , 2010; Zhang et al . , 2008 ) , allows the discussion of our results at a molecular level . This architecture informs about the interactions among the viral proteins , including the analysis of contact surfaces and their electrostatic nature . Analysis of the electrostatic potential of the different RV particles ( Figure 7A–C ) shows that the core shell presents a mainly hydrophobic outer surface ( Figure 7A ) in agreement with its tendency to form aggregates ( Labbé et al . , 1991; Desselberger et al . , 2013 ) . While treatment of these cores with electrolytes or different pH do not solubilize them , incubation with some detergents like deoxycholate ( Desselberger et al . , 2013 ) or with trehalose ( Figure 1E ) disperse them and suggest that particle aggregation is produced by hydrophobic forces . In a RV infection SLP are localized in the viroplasm , where viral RNA packaging and replication occur and where extensive protein-RNA and protein-protein interactions prevent its aggregation ( Zeng et al . , 1998; Berois et al . , 2003; Vende et al . , 2003 ) . Over the hydrophobic outer surface of the VP2 T = 1 shell , VP6 pear-shaped trimers assemble into five non-equivalent positions ( Figure 7D–E , triangles ) to build a T = 13 architecture , in what constitutes an extreme example of symmetry mismatch . These mismatched interactions are mainly mediated by the hydrophobic VP6 inward-projecting loop 64–72 ( Figure 7—figure supplement 1 ) that contacts with the SLP outer surface , and are not only essential for assembly but also for transcription ( Charpilienne et al . , 2002 ) . Intertrimeric VP6 contacts are established through their pedestal domains and have local 2-fold contacts . Both the VP2-VP6 and the intertrimeric VP6-VP6 contacts are of modest extent . These weak protein-protein interactions , described in the structure , are in agreement with our experiments . In particular , VP6 trimers are quickly disassembled in fatigue experiments ( Figure 5C; blue in Figure 5E ) , proving poor lateral and perpendicular interactions between VP6 trimers and VP6-VP2 units . In contrast with the hydrophobic nature of the SLP outer surface , the calculation of the electrostatic potential surface of the DLP reveals a very negative outer surface ( Figure 7B ) ( Mathieu et al . , 2001 ) . The structures of the transcriptionally active particles of other dsRNA viruses present a similar negatively charged outer surface ( Figure 7—figure supplement 2 ) , which may reflect a common strategy to avoid the interaction with the newly synthesized negative charged transcripts . In addition to transforming the SLP particle in the transcription-active DLP ( Lawton et al . , 1997 ) , the VP6 that polymerizes on the surface of the SLP acts as an adaptor for the interaction with the outer RV shell ( Figure 7F–G ) . VP7 trimers are stabilized through the binding of calcium ions at each subunit interface ( Aoki et al . , 2009 ) . The bottom inner surface of the VP7 trimer has minimal contacts with the VP6 trimer apex of which the most intense is mediated by the VP7 N termini that embraces the underlying VP6 trimer ( Figure 7H ) . These arms also interact with adjacent VP7 trimers generating a cooperative lattice that reinforce the RV outer shell . Our fatigue experiments ( Figure 5 ) demonstrate weak interactions of VP6 , both intertrimeric and with the VP2 layer . These analyses also suggest a strong interaction of the VP7 trimers with the underlying VP6 and with the surrounding VP7 trimers in the presence of calcium ( Figure 6 ) . Many viral particles are stabilized by calcium ions bound to the interfaces between their capsomers which is allowed by the unique coordination chemistry of the Ca ion ( Zhou et al . , 2009; Carafoli and Krebs , 2016 ) . These ions are required to maintain the capsid structural integrity and/or regulate its proper assembly/disassembly ( Zhou et al . , 2009 ) . Examples include bacteriophages of the Leviviridae and Microviridae families ( McKenna et al . , 1996; Persson et al . , 2008 ) ; plant Tombusviruses ( and its associate satellite virus ) , Sobemoviruses , Bromoviruses or Virgaviruses ( Harrison et al . , 1978; Jones and Liljas , 1984; Speir et al . , 1995 ) and different animal viruses including members of the Polyoma , Noda , Picorna , Birna and Parvoviridae families ( Tsao et al . , 1991 ) . Actually , previous studies have directly probed the mechanical role of calcium ions in the shell stability of tomato bushy stunt virus nanoparticles ( Llauró et al . , 2015 ) . Surprisingly , the inward facing electrostatic potential surface of the VP7 layer ( Figure 7F ) is highly negative . We propose that calcium ions , beyond stabilizing the VP7 trimers , would be sandwiched between the VP7 inner and VP6 outer surfaces to allow their assembly . VP7 assembles into trimers that are stabilized through the binding of two calcium ions at each subunit interface ( Aoki et al . , 2009 ) . Thus , the depletion of calcium will promote the destabilization of the VP7 intertrimeric interactions and induce the rapid disassembly of this shell by the destabilization of the VP7/VP6 electrostatic interactions ( Figure 5B ) . Mechanical parameters , such as stiffness , breaking force and yield strain also inform of important differences between the three layers ( Figure 4 ) . Similar to that observed for the height distribution ( Figure 2E ) , the dispersion detected for TLP stiffness kTLP ( Figure 4A ) could be correlated with the unequal presence and distribution of spikes in each particle . The incorporation of the VP7 layer on the DLP produces a significant increase in stiffness ( Figure 4A ) and yield strain ( Figure 4—figure supplement 2 ) . Thus , while the Young’s modulus value of the VP7 shell is within the highest values as obtained for bacteriophages ( Roos et al . , 2012; Ivanovska et al . , 2004 ) , the VP6 layer presents the lowest value ever reported for a viral protein shell ( Marchetti et al . , 2016 ) . In fact , the FE simulations of the TLP show that the stiff VP7 layer accumulates most of the stress during the indentation ( Figure 4C ) , protecting the internal VP6 and VP2 layers by shielding the stress transmission to these layers . Taken together , nanoindentation and mechanical fatigue experiments demonstrate that the VP7 shell provides the resistance needed by the RV particle to bear with the severe conditions of extracellular media . RV is transmitted through the faecal-oral route and has to overcome the stringent physicochemical conditions of digestion at both the stomach and small intestine , where it infects mature enterocytes ( Estes and Greenberg , 2013; Ramig , 2004 ) . The viscosity of the chyle ( Jonas , 1976 ) is about 10 to 100 times larger than the host cytoplasm ( Luby-Phelps et al . , 1993 ) and presents higher molecular crowding ( Hernando-Pérez et al . , 2014a ) . Therefore , the VP7 shell has to be stable enough to overcome the constant barrage of molecular impacts in the small intestine . In fact , fatigue experiments provide a good approximation for these molecular impacts on RV particles ( Hernando-Pérez et al . , 2014a ) . Interestingly , the VP7 shell of TLP is able to withstand fatigue even at 200 pN ( Figure 5A ) indicating a strong intercapsomeric linkage . The labile nature of VP6 layer , showing both the lowest values of elasticity and Young’s modulus , is related with their structure ( weak contacts of the VP6 trimers with VP2 and between them ) and we propose that this feature is necessary for its function . It has been suggested that removal of VP7 causes the dilation of the particle pentameric channels allowing the flux of nucleotides , ions and transcripts ( Chen et al . , 2009; Aiyegbo et al . , 2013 ) . The removal of VP7 promotes the outward movement of the VP6 pentameric trimers . This conformational change is transmitted through the underlying VP2 decamer to the VP1 polymerase , enabling its activity . In other viruses , such as MVM ( Castellanos et al . , 2012 ) a similar conformational dynamics is favored by a low mechanical stability . In particular , the increase of local stiffness in MVM mutants blocks the conformational changes required for dsDNA translocation . Similarly , the high flexibility resulting from the low mechanical stability of the trimeric VP6 layer would favor its functional roles: this thick layer becomes the adaptor that allows the transformation of a highly hydrophobic SLP into a negative-charged DLP , overcoming the symmetry mismatch between the T = 1 and T = 13 layers , and generating a transcriptionally active particle . The VP6 shell constitutes the thickest and , according to our data , the softest layer of the RV particle , which allows for large deformations when TLP or DLP are adsorbed ( Figure 2 and Figure 2—figure supplement 2 ) . Finally , the SLP exists only in the viroplasm environment during RNA packaging and replication . The high electrodensity of the viroplasm is a signature of a large concentration of macromolecules that results in a higher molecular crowding than the cytoplasm . This fact would explain the higher Young’s modulus value of VP2 layer when compared with that of VP6 . This Young’s modulus combined with a presumably smaller adsorption energy with the substrate , result in non-deformed particle after adsorption , as it happens with other virus capsids ( Carrasco et al . , 2009 ) . In this mechanical study of a multi-layered virus we have shown how the biophysical properties and interactions of the three particle shells are finely tuned to produce an infective RV virion . While the high mechanical strength provided by the strong VP7-VP7 and VP7-VP6 interactions ( Figure 7H , black springs ) relates to protection tasks , the lower resistance of the VP6-VP6 and VP6-VP2 interactions ( Figure 7H , white springs ) guarantees the conformational dynamics required for transcription . Importantly , the interference with this finely tuned mechanical regulation offers new venues for development of antiviral strategies .
The simian rotavirus strain SA11-C4111 ( Rodríguez et al . , 2014 ) was used in this study . Viruses were grown using the monkey epithelial cell line MA104 ( ECACC 85102918 ) , cultured in MEM with 10% fetal calf serum , and used between passages 10 and 25 . The amplified viruses were used within three passages of the last plaque isolation step . For the production of TLP , 3 day post-confluent monolayers of MA104 cells were infected with a multiplicity of 0 . 5 PFU/cell . Activation of the viruses was performed for 30 min at 37°C with 100 BAEE U/ml of TPCK-treated trypsin ( TPCK Trypsin , Thermo Scientific Pierce ) . To remove serum , cell monolayers were washed twice with MEM prior to absorption ( 60’ , 37°C ) . After absorption , monolayers were washed with MEM and incubated in MEM containing 10 BAEE U/ml TPCK-trypsin . Cells and extracellular media were harvested when total cytopathic effect was observed . TLP were purified from these extracts as previously indicated ( Rodríguez et al . , 2014 ) . Purified TLP were diluted to 0 . 2 mg/ml of protein content in 1xTNC ( 10 mM Tris:HCl pH 7 . 5 , 140 mM NaCl , 10 mM CaCl2 ) containing 10% glycerol and 0 . 02% sodium azide , flash frozen in liquid nitrogen as small ( 5 µl ) aliquots , and stored at −80°C . The preparation of DLP from purified TLP by treatment with EDTA at 37°C and its isolation in CsCl gradients has been performed as described by Patton et al ( Patton et al . , 2000 ) . Purified DLP were diluted to 0 . 2 mg/ml of protein content in 1xTNE ( 10 mM Tris:HCl pH 7 . 5 , 140 mM NaCl , 1 mM EDTA ) containing 10% glycerol and 0 . 02% sodium azide flash frozen in liquid nitrogen as small ( 5 µl ) aliquots , and stored at −80°C . SLP were prepared form purified DLP by treatment with 1 . 25M CaCl2 in a solution containing 0 . 75M trehalose , 0 . 15M NaCl , 20 mM Borate buffer ( pH 8 . 45 ) and Complete-EDTA Free protease inhibitors ( Roche ) at the manufacturer recommended concentration . DLP , at a concentration of 100 µg/ml , where incubated for 2 hr at 37°C with gentle agitation . After the treatment , the concentration of trehalose in the mixture was reduced to 0 . 25M by dilution with two volumes of the buffer without trehalose , and incubated at room temperature ( 22°C ) during 90 min , with gentle agitation . SLP were concentrate by centrifugation ( 20 . 000 g , 60 min , 22°C ) and resuspended in a buffer containing 1 . 50M trehalose , 0 . 15M NaCl , 20 mM Tris:HCl ( pH 8 . 45 ) and Complete-EDTA Free protease inhibitors ( Roche ) at the manufacturer recommended concentration . Purified SLP were diluted to 0 . 2 mg/ml in 1xTNC containing 0 . 5M trehalose , flash frozen in liquid nitrogen as small ( 5 µl ) aliquots , and stored at −80°C . For transmission electron microscopy , purified particles were applied to glow-discharged carbon-coated grids and negatively stained with 2% aqueous uranyl acetate . Images were recorded on a Gatan 1 k CCD camera in a FEI Tecnai 12 microscope operated at 120 kV . For cryo-EM , samples were applied to Quantifoil R 2/2 holey grids , blotted , and plunged into liquid ethane using a Leica EM CPC cryo-fixation unit . Cryo-EM images were recorded in low-dose conditions ( ~10 e-/Å [Müller et al . , 2002] ) on a FEI Eagle 4 k CCD using a Tecnai G2 electron microscope operating at 200 kV and a detector magnification of 67 , 873X ( 2 . 16 Å/pixel sampling rate ) . Image processing operations were performed using Xmipp ( Marabini et al . , 1996 ) and Relion ( Scheres , 2012 ) and graphic representations were produced by UCSF Chimera ( Pettersen et al . , 2004 ) . Xmipp automatic picking routine was used to select 4238 particles and defocus was determined with CTFfind3 ( Mindell and Grigorieff , 2003 ) . Images were 2D classified using the corresponding Relion routine to select 4200 homogenous particles . To avoid any bias at the spike density , the published structure of the rotavirus VP7-recoated particle ( Chen et al . , 2009 ) , low-pass filtered to 30 Å , was used as initial model for Relion to obtain a 3DR using the corresponding Relion autorefinement routine . Resolution was assessed by gold standard Fourier Shell Correlation ( FSC ) between two independently processed half datasets . Applying a correlation limit of 0 . 5 ( 0 . 3 ) , the resolution is 14 . 2 ( 12 . 6 ) . The electrostatic potentials were calculated using DelPhi software ( Rocchia et al . , 2002 ) and surface-colored with UCFS Chimera . Measurements were carried out with an AFM ( Nanotec Electrónica S . L . , Madrid , Spain ) operating in Jumping Mode Plus ( Ortega-Esteban et al . , 2012 ) . This intermittent-contact imaging mode consists on performing low force-versus-Z-piezo-displacement ( FZ ) curves at every point of the imaging area , with nanometric lateral movements of the sample where it is far ( ~40 nm ) from the tip . All the experiments were carried out with rectangular silicon-nitride cantilevers ( RC800PSA , Olympus , Tokyo , Japan ) with nominal springs constants of 0 . 05 N/m , and were routinely calibrated using the Sader's method ( Sader et al . , 1999 ) . The obtained images were processed using the WSxM software ( Horcas et al . , 2007 ) . For adsorption of particles , one 5 µl aliquot of particles was thawed on ice and diluted to 50 μl with TNC ( for TLP and DLP ) or in TNC-Trehalose ( for SLP ) . They were incubated for 15 min on freshly cleaved highly oriented pyrolytic graphite ( HOPG; ZYA quality; NT-MDT , Tempe , AZ ) . The non-adsorbed particles were removed by performing several washes consisting in the addition of 50 μl of TNC and the extraction of 50 μl of the sample . The tip was also prewetted with a 20 μl drop of TNC before starting the image acquisition process . For single nanoindentation assays , individual particles were deformed with the AFM tip by performing single force curves at a constant speed ( 150 nm/s ) and with a high Z piezo displacement ( 150 nm ) to ensure that the tip always reached the substrate after the disruption of the particle . Images before and after the FZ were obtained to observe the structural damages suffered by each particle . The mechanical properties ( elastic constant , breaking force and critical strain ) were obtained from these FZ curves . For cyclic loading assays , the topographic image acquisition with the AFM tip was used to mechanically fatigue single particles ( TLP , DLP and SLP ) , causing their guided disassembly and allowing to image the dynamics of the process . The number of scanning points in the ‘x’ and ‘y’ coordinates ( 128 in each direction ) , and the size of the image ( ~300 nm ) were established to apply one loading cycle each ~2–3 nm . Real time experiments of TLP disassembly while removing Ca ions was carried out as follows . TLP were initially in the AFM liquid chamber with 70 µl of TNC buffer . This chamber was connected to two syringe pumps ( NE-1000 , New Era Pump Systems , Inc . ) . One of the syringes was used for pumping TNE buffer into the chamber while , simultaneously , the other syringe was withdrawing liquid . The pumping/withdrawing rate was 1 µl/min , and the fatigue experiments lasted ~80 min . Under these conditions TNC buffer was totally replaced by TNE at the end of the experiment , thus ensuring the chelation of all the Ca2 +ions initially present in the TLP . Finite elements simulations mimicking the AFM nanoindentation of the different rotavirus particles were performed using the program COMSOL Multiphysics 4 . 3 ( Comsol , Stockholm , Sweden ) . In the simulations , each layer was modeled as a homogenous spherical shell made of a material with Young’s modulus E and Poisson ratio ν = 0 . 3 ( a standard value for protein-like materials ) . This model shell was placed on a hard flat substrate and indented by a hard spherical object with radius Rin = 15 nm , mimicking the nominal radius of the AFM tip . The system was simulated using a 2D axisymmetric model that was meshed with over 1400–6000 triangular elements . The contacts between the shell and the tip as well as the supporting surface during indentation were implemented with a contact normal penalty factor . This parameter controls the hardness of the interface surface and it is used to prevent the penetration of the two boundaries coming into contact . The penalty factor used was Y/Δx , where Y is the Young’s modulus and Δx is the minimum element size of the mesh of the material which is indented . A parametric , non-linear solver was used to simulate the stepwise lowering of the tip onto the capsid . The spring constant was obtained in all cases from the slope of the force versus indentation curves at a small value of the indentation of 2 nm . For multilayer shells , two different cases were simulated: a model in which the shells are joint and coupled ( using the COMSOL option Union to finalize the geometry ) , and a second case in which the layers are independent and uncoupled ( using the option Assembly to finalize the geometry ) . In both cases , the results for the stress distribution , the force-indentation curves and the spring constant for small indentations were identical . The error bars in the values of Young’s modulus for the different layers were calculated in the FEM simulations in the following way . For each value of the experimental spring constant k±δk , we did FEM simulations to find which value of the Young’s modulus , Y , was giving a slope of k; which value , Ymin , was giving a slope k-δk; and which value , Ymax , was yielding k+δk . The best estimate and approximated uncertainty in the Young’s modulus were reported as Y± ( Ymax-Ymin ) /2 . The SLP was modeled as a spherical shell with an external radius R = 27 nm and thickness h = 3 . 5 nm ( see inset in Figure 4—figure supplement 1 ) . A Young’s modulus of Y1 = 0 . 53 ± 0 . 20 GPa was used in order to recover the same slope in the simulations as the one measured experimentally . The DLP was modeled as a double-layer spherical shell with an external radius R = 35 nm , made of an outer layer with Young’s modulus Y2 = 0 . 0815 ± 0 . 070 GPa and thickness h = 8 . 0 nm , and an inner layer with Young’s modulus Y1 = 0 . 53 ± 0 . 20 GPa and thickness h = 3 . 5 nm ( see inset in Figure 4—figure supplement 1 ) . Finally , the TLP was modeled as a triple-layer spherical shell , by adding a third layer with Young’s modulus Y3 = 1 . 0 ± 0 . 9 GPa and thickness h = 3 . 5 nm , mimicking the VP7 ( see inset in Figure 4—figure supplement 1 ) . | Viruses are small agents that enter and hijack cells to create more of themselves . Most of them are made of a protein shell that encases the viral genome and certain molecular tools . During the life cycle of a virus , this shell fulfils many roles , from protecting the genetic information to recognising the appropriate host cell . It must also disassemble at the right time for replication to take place . A number of viruses wrap themselves in several layers of protective casing , resulting in an onion-like structure . For example , the rotaviruses that sometimes cause severe diarrhoea in young children have three layers , each with specific properties . Rotavirus subparticles may exist with only one or two of these coats , which allows researchers to study each layer in detail . Here , Jiménez-Zaragoza et al . use a method called atomic force microscopy to look into the physical properties of the layers of the rotavirus . The technique uses an extremely sharp stylus attached to a tiny cantilever to deform the shells of a single virus . How the structure reacts can then be recorded using a powerful microscope . This helps to determine the stiffness of the layers , and how much force is required to break or weaken each of them . The experiments reveal that the mechanical properties of the layers are tailored to help the virus survive and infect cells . The outer coat is stiff and resistant to strain , which shields the virus during its travel through the digestive system . The middle layer is the thickest and the softest of the three . It allows the virus to cope with deformation , which is necessary for the expression of its genome . The outer and middle layers are strongly connected , in part through calcium ions that may be ‘sandwiched’ between the two . By contrast , the middle and inner layers are only loosely attached to each other . When the virus enters the cell , the calcium ions get dislodged , helping the external coating to easily disassemble . In turn , this creates structural changes in the middle layer , which activate molecules required for the genome to get expressed . Ultimately , disrupting the finely tuned properties of the layers could create new ways of fighting rotaviruses . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"physics",
"of",
"living",
"systems"
] | 2018 | Biophysical properties of single rotavirus particles account for the functions of protein shells in a multilayered virus |
Most humans are exposed to Tropheryma whipplei ( Tw ) . Whipple’s disease ( WD ) strikes only a small minority of individuals infected with Tw ( <0 . 01% ) , whereas asymptomatic chronic carriage is more common ( <25% ) . We studied a multiplex kindred , containing four WD patients and five healthy Tw chronic carriers . We hypothesized that WD displays autosomal dominant ( AD ) inheritance , with age-dependent incomplete penetrance . We identified a single very rare non-synonymous mutation in the four patients: the private R98W variant of IRF4 , a transcription factor involved in immunity . The five Tw carriers were younger , and also heterozygous for R98W . We found that R98W was loss-of-function , modified the transcriptome of heterozygous leukocytes following Tw stimulation , and was not dominant-negative . We also found that only six of the other 153 known non-synonymous IRF4 variants were loss-of-function . Finally , we found that IRF4 had evolved under purifying selection . AD IRF4 deficiency can underlie WD by haploinsufficiency , with age-dependent incomplete penetrance .
Whipple’s disease ( WD ) was first described as an intestinal inflammatory disease by George H . Whipple in 1907 ( Whipple , 1907 ) . Its infectious origin was suspected in 1961 ( Yardley and Hendrix , 1961 ) , and the causal microbe , Tropheryma whipplei ( Tw ) , a Gram-positive actinomycete , was detected by PCR in 1992 ( Relman et al . , 1992 ) , and cultured in 2000 ( Raoult et al . , 2000 ) . Tw is probably transmitted between humans via the oro-oral or feco-oral routes . WD is a chronic condition with a late onset ( mean age at onset: 55 years ) ( Braubach et al . , 2017 ) affecting multiple organs . The clinical manifestations of classical WD are arthralgia , diarrhea , abdominal pain , and weight loss ( Dobbins , 1987; Durand et al . , 1997; Fleming et al . , 1988; Mahnel et al . , 2005; Maizel et al . , 1970 ) . However , about 25% of WD patients display no gastrointestinal or osteoarticular symptoms , instead presenting with cardiac and/or neurological manifestations ( Durand et al . , 1997; Fenollar et al . , 2014 , Fenollar et al . , 2001; Gubler et al . , 1999; Schneider et al . , 2008 ) . WD is fatal if left untreated , and relapses occur in 2% to 33% of treated cases , even after prolonged appropriate antibiotic treatment ( Lagier et al . , 2011; Marumganti and Murphy , 2008 ) . WD is rare and has been estimated to affect about one in a million individuals ( Dobbins , 1981 , 1987; Fenollar et al . , 2007 ) . However , about two thousand cases have been reported in at least nine countries worldwide , mostly in North America and Western Europe ( Bakkali et al . , 2008; Desnues et al . , 2010; Fenollar et al . , 2008a; Lagier et al . , 2010; Puéchal , 2016; Schneider et al . , 2008 ) . Chronic asymptomatic carriage of Tw is common in the general population , and this bacterium has been detected in feces , saliva , and intestinal mucosae . The prevalence of Tw carriage in the feces has been estimated at 2% to 11% for the general population , but can reach 26% in sewer workers and 37% in relatives of patients and carriers ( Amsler et al . , 2003; Ehrbar et al . , 1999; Fenollar et al . , 2014 , Fenollar et al . , 2007; Maibach et al . , 2002; Rolain et al . , 2007; Schneider et al . , 2008; Street et al . , 1999 ) . Seroprevalence for specific antibodies against Tw in the general population varies from 50% in France to 70% in Senegal ( Fenollar et al . , 2009 , Fenollar et al . , 2014; Raoult et al . , 2000; Schneider et al . , 2008 ) . At least 75% of infected individuals clear Tw primary infections , but a minority ( <25% ) become asymptomatic carriers , a very small proportion of whom develop WD ( <0 . 01% ) ( Fenollar et al . , 2008b ) . Tw infection is therefore , necessary but not sufficient , for WD development , and it is unclear whether prolonged asymptomatic carriage necessarily precedes WD . The hypothesis that WD results from the emergence of a more pathogenic clonal strain of Tw was not supported by bacterial genotyping ( Li et al . , 2008 ) . WD mostly affects individuals of European origin , but does not seem to be favored by specific environments . WD is typically sporadic , but six multiplex kindreds have been reported , with cases often diagnosed years apart , suggesting a possible genetic component ( Durand et al . , 1997; Fenollar et al . , 2007; Ponz de Leon et al . , 2006 ) . WD patients are not prone to other severe infections ( Marth et al . , 2016 ) . Moreover , WD has never been reported in patients with conventional primary immunodeficiencies ( PIDs ) ( Picard et al . , 2018 ) . This situation is reminiscent of other sporadic severe infections , such as herpes simplex virus-1 encephalitis , severe influenza , recurrent rhinovirus infection , severe varicella zoster disease , trypanosomiasis , invasive staphylococcal disease , and viral infections of the brainstem , which are caused by single-gene inborn errors of immunity in some patients ( Andersen et al . , 2015; Casanova , 2015a , Casanova , 2015b; Ciancanelli et al . , 2015; Israel et al . , 2017; Lamborn et al . , 2017; Zhang et al . , 2015; Ogunjimi et al . , 2017; Vanhollebeke et al . , 2006; Zhang et al . , 2018 ) . We therefore hypothesized that WD might be due to monogenic inborn errors of immunity to Tw , with age-dependent incomplete penetrance .
We investigated four related patients diagnosed with WD ( P1 , P2 , P3 , and P4 ) with a mean age at diagnosis of 58 years . They belong to a large non-consanguineous French kindred ( Figure 1A ) . The proband ( P1 ) , a 69-year-old woman , presented with right knee arthritis in 2011 , after recurrent episodes of arthritis of the right knee since 1980 . Tw was detected in the synovial fluid by PCR and culture , but not in saliva , feces , or small intestine tissue . Treatment with doxycycline and hydroxychloroquine was effective . At last follow-up , in 2016 , P1 was well and Tw PCR on saliva and feces was negative . P2 , a second cousin of P1 , is a 76-year-old woman with classical WD diagnosed at 37 years of age in 1978 by periodic acid–Schiff ( PAS ) staining of a small intestine biopsy specimen . She was treated with sulfamethoxazole/trimethoprim . At last follow-up , in 2016 , Tw PCR on saliva and feces was positive . P3 , the father of P1 , is a 92-year-old man with classical WD diagnosed at 62 years of age , in 1987 , based on positive PAS staining of a small intestine biopsy specimen . Long-term sulfamethoxazole/trimethoprim treatment led to complete clinical and bacteriological remission . P4 , the brother of P2 , is a 70-year-old man who consulted in 2015 for arthralgia of the knees and right ulna-carpal joints . PCR and culture did not detect Tw in saliva and feces , but serological tests for Tw were positive . Treatment with methotrexate and steroids was initiated before antibiotics , the effect of which is currently being evaluated . All four patients are otherwise healthy . Saliva and/or feces samples from 18 other members of the family were tested for Tw ( Figure 1A; Figure 1—source data 1 ) . Five individuals are chronic carriers ( mean age: 55 years ) and 13 tested negative ( mean age: 38 years ) . Nine additional relatives could not be tested . The distribution of WD in this kindred was suggestive of an AD trait with incomplete clinical penetrance . We analyzed the familial segregation of WD by genome-wide linkage ( GWL ) , using information from both genome-wide single-nucleotide polymorphism ( SNP ) microarrays and whole-exome sequencing ( WES ) ( Belkadi et al . , 2016 ) . Multipoint linkage analysis was performed under an AD model , with a very rare disease-causing allele ( <10−5 ) and age-dependent incomplete penetrance . Twelve chromosomal regions linked to WD were identified on chromosomes 1 ( x3 ) , 2 , 3 , 6 , 7 , 8 , 10 , 11 , 12 and 17 , with a LOD score close ( >1 . 90 ) to the maximum expected value ( 1 . 95 ) ( Figure 1—figure supplement 1A ) . These regions covered 27 . 18 Mb and included 263 protein-coding genes . WES data analysis for these 263 genes identified 54 heterozygous non-synonymous coding variants common to all four WD patients ( Figure 1—source data 2 ) . Only one , a variant of the interferon regulatory factor ( IRF ) four gene encoding a transcription factor from the IRF family ( Ikushima et al . , 2013 ) , located in a 200 kb linked region on chromosome 6 ( Figure 1—figure supplement 1A and B ) , was very rare , and was even found to be private [not found in the gnomAD database , http://gnomad . broadinstitute . org , or in our in-house WES database ( HGID ) ] , whereas all other variants had a frequency >0 . 001 , which is inconsistent with the frequency of WD and our hypothesis of a very rare ( <10−5 ) deleterious heterozygous allele . The variant is a c . 292 C > T substitution in exon 3 of IRF4 , replacing the arginine residue in position 98 with a tryptophan residue ( R98W ) ( Figure 1A , B and C ) . IRF4 is a transcription factor with an important pleiotropic role in innate and adaptive immunity , at least in a few strains of inbred mice ( Shaffer et al . , 2009 ) . Mice heterozygous for a null Irf4 mutation have not been studied , but homozygous null mice have various T- and B-cell abnormalities and are susceptible to both Leishmania and lymphocytic choriomeningitis virus ( Klein et al . , 2006; Lohoff et al . , 2002; Mittrücker et al . , 1997; Suzuki et al . , 2004; Tamura et al . , 2005; Tominaga et al . , 2003 ) . We confirmed the IRF4 R98W mutation by Sanger sequencing genomic DNA from the blood of the four WD patients ( Figure 1C ) . Thirteen relatives of the WD patients were WT/WT at the IRF4 locus , and 10 of these relatives ( 77% ) tested negative for Tw carriage . Eight other relatives were heterozygous for the IRF4 R98W mutation , five of whom ( 62 . 5% ) were Tw carriers ( mean age: 55 years ) ( Figure 1A; Figure 1—source data 1 ) . Overall , 12 individuals from the kindred , including the four patients , the five chronic carriers of Tw , two non-carriers of Tw and one relative not tested for Tw , were heterozygous for IRF4 R98W ( Figure 1A; Figure 1—source data 1 ) . The familial segregation of the IRF4 R98W allele was therefore consistent with an AD pattern of WD inheritance with incomplete clinical penetrance . Chronic Tw carriage also followed an AD mode of inheritance . The R98 residue in the DNA-binding domain ( DBD ) of IRF4 is highly conserved in the 12 species for which IRF4 has been sequenced ( Figure 1B and D ) . It has been suggested that this residue is essential for IRF4 DNA-binding activity , because the R98A-C99A double mutant is loss-of-function ( LOF ) ( Brass et al . , 1999; Escalante et al . , 2002 ) . The R98W mutation is predicted to be damaging by multiple programs ( Kircher et al . , 2014 ) ; it has a combined annotation–dependent depletion score ( CADD ) score of R98W ( 26 . 5 ) , well above the mutation significance cutoff ( MSC ) of IRF4 ( 11 . 125 ) ( Figure 2 ) ( Itan et al . , 2016; Kircher et al . , 2014 ) . The R98W variant was not present in the gnomAD database or our in-house HGID database of more than 4 , 000 WES from patients with various infectious diseases . The mutant allele was not found in the sequences for the CEPH-HGDP panel of 1 , 052 controls from 52 ethnic groups , or in 100 French controls , confirming that this variant was very rare , probably private to this kindred . Therefore , the minor allele frequency ( MAF ) of this private allele is <4 × 10−6 . Moreover , the IRF4 gene has a gene damage index ( GDI ) of 2 . 85 , a neutrality index score of 0 . 15 ( Itan et al . , 2015 ) , and a purifying selection f parameter of 0 . 32 ( among the <10% of genes in the genome subject to the greatest constraint; Figure 2—figure supplement 1 ) , strongly suggesting that IRF4 has evolved under purifying selection ( i . e . strong evolutionary constraints ) ( Eilertson et al . , 2012 ) . Biologically disruptive heterozygous mutations of IRF4 are therefore likely to have clinical effects . We identified 156 other high-confidence heterozygous non-synonymous coding or splice variants of IRF4 ( Figure 2—source data 1 ) in public ( gnomAD: 153 variants , all with MAF <0 . 009 ) and HGID ( three variants ) databases: 147 were missense variants ( two of which were also found in the homozygous state: p . S149N and p . A370V ) , four were frameshift indels leading to premature stop codons ( p . W27YfsTer50 , p . W74GfsTer28 , p . Y152LfsTer60 , and p . S160RfsTer11 ) , three were in-frame indels ( p . E46del , p . G279_H280del , and p . S435del ) , one was a nonsense variant ( p . R82* ) , one was an essential splice variant ( c . 4032T > C ) , and two were missense variants found only in a non-canonical transcript predicted to undergo nonsense-mediated decay ( p . L406P and p . R407W ) . Up to 150 of the 156 variants are predicted to be benign , whereas only six were predicted to be potentially LOF according to the gnomAD database classification ( the four frameshift indels , the nonsense variant , and the essential splice variant ) . Comparison of the CADD score and MAF of these IRF4 variants showed R98W to have the second highest CADD score of the four variants with a MAF <4 × 10−6 ( Figure 2 ) . These findings suggest that the private heterozygous IRF4 variant of this kindred is biochemically deleterious , unlike most other rare ( MAF <0 . 009 ) non-synonymous variants in the general population , 150 of 156 of which were predicted to be benign ( Lek et al . , 2016 ) . We first characterized IRF4 R98W production and function in vitro , in an overexpression system . We assessed the effect of the IRF4 R98W mutation on IRF4 levels by transiently expressing WT or mutant R98W in HEK293T cells . IRF4 R98A-C99A , which is LOF for DNA binding ( Brass et al . , 1999 ) , was included as a negative control . In total cell extracts , mutant IRF4 proteins were more abundant than the WT protein and had the expected molecular weight ( MW ) of 51 kDa , as shown by western blotting ( Figure 3A ) . The R98 residue has been shown to be located in a nuclear localization signal , the complete disruption of which results in a loss of IRF4 retention in the nucleus ( Lau et al . , 2000 ) . We therefore analyzed the subcellular distribution of IRF4 WT and R98W proteins , in total , cytoplasmic , and nuclear extracts from transiently transfected HEK293T cells . The R98W mutant was more abundant than the WT protein in total cell and cytoplasmic extracts , but these proteins were similarly abundant in nuclear extracts ( Figure 3B ) . We performed luciferase reporter assays to assess the ability of the mutant IRF4 protein to induce transcription from interferon-stimulated response element ( ISRE ) motif-containing promoters . Unlike the WT protein , both R98W and R98A-C99A failed to activate the ( ISRE ) 3 promoter ( Figure 3C ) . We also assessed the ability of IRF4 to induce transcription from an ( AP-1 ) -IRF composite element ( AICE ) motif-containing promoter ( Li et al . , 2012 ) . Both R98W and R98A-C99A failed to activate the AICE promoter ( Figure 3D ) . Moreover , we observed no dominant-negative effect of the IRF4 R98W protein , with either the ISRE or AICE motif-containing promoter ( Figure 3—figure supplement 1 ) . We assessed the ability of R98W to bind DNA , in an electrophoretic mobility shift assay ( EMSA ) ( Figure 3E and F ) . Signal specificity was assessed by analyzing both supershift with an IRF4-specific antibody and by competition with an unlabeled competitor probe . The R98W mutation abolished IRF4 binding to the ISRE cis element ( Figure 3E ) , and binding of the IRF4-PU . 1 complex to interferon composite elements ( EICEs ) containing both IRF4 and PU . 1 recognition motifs ( Brass et al . , 1999 ) ( Figure 3F ) . The R98W allele of IRF4 is therefore LOF for both DNA binding and the induction of transcription . We then tested 153 of the other 156 IRF4 variants: 150 variants previously described in the gnomAD database and three variants found in the HGID database . The essential splice variant and the two variants present only in a non-canonical transcript were not tested . All the variants tested were normally expressed ( two with a higher MW ) , except for the five predicted LOF variants tested ( the epitope of the antibody being at the C-terminus of IRF4 ) ( Figure 3—figure supplements 2; 3 ) . Transfection with the five predicted LOF plasmids was efficient , as assessed by cDNA amplification and sequencing ( data not shown ) . When tested for ( ISRE ) 3 promoter activation , the five stop-gain or frameshift variants predicted to be LOF in the gnomAD database ( very rare variants , MAF <6 × 10−6 ) were found to be LOF . We also showed that among all the non-synonymous coding variants tested , only one rare very rare ( MAF 9 × 10−6 ) inframe deletion of one amino acid ( E46del ) reported in the gnomAD database , was hypomorphic , and another variant from our in-house WES database ( G279_H280 del , private to one family ) was LOF ( Figure 3—figure supplements 4; 5 ) . The cumulative frequency of these seven LOF ( n = 6 ) or hypomorphic ( n = 1 ) variants was <4 × 10−5 , fully consistent with the frequency of WD ( occurring only in adults chronically infected with Tw ) . Overall , our data show that the R98W IRF4 allele is LOF , like only six other very rare non-synonymous IRF4 coding variants of the 153 variants tested . Moreover , R98W is not dominant negative . We investigated the cellular phenotype of heterozygosity for the R98W allele in EBV-transformed B-cell lines ( EBV-B cells ) from patients . We performed reverse transcription-quantitative polymerase chain reaction ( RT-qPCR ) on EBV-B cells from P1 , P3 , two healthy heterozygous relatives ( IRF4 WT/R98W ) , four healthy IRF4-WT homozygous relatives , and seven healthy unrelated controls . We also investigated 25 unrelated WD patients with Tw carriage . We sequenced all IRF4 coding exons for these patients , who were found to be WT . They were also found to have an intact IRF4 cDNA structure and normal IRF4 protein levels in EBV-B cells ( data not shown ) . Cells from individuals heterozygous for the R98W mutation ( patients and healthy carriers ) had higher IRF4 mRNA levels than those from WT homozygous relatives , unrelated WD cohort patients and EBV-B cells from healthy unrelated controls ( Figure 4A ) . We compared the relative abundances of WT and R98W IRF4 mRNA in EBV-B cells from heterozygous carriers of the mutation , by performing TA-cloning experiments on P1 , P3 , one healthy heterozygous relative , one relative homozygous for WT IRF4 , and two previously tested healthy unrelated controls . In heterozygous carriers of the mutation ( patients and healthy relatives ) , the R98W mutation was present in 48 . 1–60% of the total IRF4 mRNA , whereas the rest was WT ( Figure 4B ) . We evaluated the levels and distribution of IRF4 protein by western blotting on EBV-B cells from P1 , P2 , P3 , one healthy heterozygous relative , three healthy homozygous WT relatives and five unrelated healthy individuals . As in transfected HEK293T cells , IRF4 protein levels were high both in total cell extract and even more so in cytoplasmic extracts of EBV-B cells from heterozygous carriers ( Figure 4—figure supplement 1 ) . By contrast , IRF4 protein levels in EBV-B cell nuclei were similar in heterozygous carriers and controls ( Figure 4—figure supplement 1 ) . As IRF4 is a transcription factor , we then analyzed the steady-state transcriptome of EBV-B cells from three healthy homozygous WT relatives and three WT/R98W heterozygotes ( P1 , P3 , VI . 6 ) . We identified 37 protein-coding genes as differentially expressed between subjects heterozygous for IRF4 and those homozygous WT for IRF4 ( 18 upregulated and 19 downregulated; data not shown ) . We identified no marked pathway enrichment based on these genes . EBV-B cells from individuals heterozygous for IRF4 had a detectable phenotype , in terms of IRF4 production and function , consistent with AD IRF4 deficiency underlying WD . We assessed IRF4 levels in peripheral blood mononuclear cells ( PBMCs ) from healthy controls ( Figure 5—figure supplement 1 ) . IRF4 were also expressed in CD4+ T cells , particularly after stimulation with activating anti-CD2/CD3/CD28 monoclonal antibody-coated ( mAb-coated ) beads ( data not shown ) . We therefore assessed the IRF4 protein expression profile in CD4+ T cells from four healthy unrelated controls , P1 and P3 , with and without ( non-stimulated , NS ) stimulation with activating anti-CD2/CD3/CD28 mAb-coated beads . The results were consistent with those for transfected HEK293T and EBV-B cells , as IRF4 levels were higher in activated CD4+ T cells from P1 and P3 than in controls , both for total cell extracts , and even more so for the cytoplasmic compartment ( Figure 5A and B ) . By contrast , IRF4 levels in the nucleus were similar and , possibly , even slightly lower in patients than in controls ( Figure 5C ) . We also investigated peripheral myeloid ( Figure 5—figure supplements 1–4 ) and lymphoid blood cell subsets in patients ( Figure 5—figure supplements 5–7; Figure 5—source data 1 ) , which display an apparently normal development compared to healthy controls’ cells . Then , we checked for transcriptomic differences associated with genotype and/or infection , by investigating the transcriptomes of PBMCs from six IRF4-heterozygous individuals ( three patients , P1-P3; and three healthy relatives , HET1-HET3 ) and six IRF4 WT-homozygous individuals ( four healthy relatives , WT1-WT4; and two healthy unrelated controls , C1-C2 ) with and without in vitro infection for 24 hr with Tw , or Mycobacterium bovis-Bacillus Calmette-Guerin ( BCG ) , which , like Tw , belongs to phylum Actinobacteria . We performed unsupervised hierarchical clustering of the differentially expressed ( DE ) transcripts ( infected versus uninfected ) to analyze the overall responsiveness of PBMCs from individual subjects to BCG and Tw infections in vitro . Heterozygous individuals clearly clustered separately from homozygous WT individuals ( Figure 6A ) , revealing a correlation between genotype and response to infection . Overall , we found that 402 transcripts from 193 unique genes were responsive to BCG infection ( Figure 6—source data 1 ) , and 119 transcripts from 29 unique genes were responsive to Tw infection ( Figure 6—source data 2 ) in homozygous WT subjects , according to the criteria described in the Materials and methods . Due to the small number of Tw-responsive transcripts linked to unique genes , we were unable to detect any pathway enrichment for this specific condition . However , we identified 24 canonical pathways as enriched after the exposure of PBMCs to BCG . We ranked these pathways according to the difference in mean z-score between homozygous WT and heterozygous subjects ( Figure 6B ) . The top 10 pathways included the interferon signaling network , the Th1 pathway network , the HMGB1 signaling network , the p38 MAPK signaling network , the NF-κB signaling network , the dendritic cell maturation network and the network responsible for producing nitric oxide and reactive oxygen species . These pathways were highly ranked mostly due to IFNG and STAT1 , which were strongly downregulated in IRF4 heterozygotes , particularly in P1 , P2 and P3 , relative to WT homozygotes . IRF4 is predicted to bind the promoter regions of 47% of the genes identified in the BCG study ( 91 of 193 genes ) , including those of IFNG and STAT1 . Subjects heterozygous for IRF4 also had lower levels of LTA expression , and lower levels of IL2RA expression were observed specifically in patients ( GSE102862 ) . These data suggest a general impairment of the T-cell response in subjects heterozygous for IRF4 upon BCG infection in vitro . Moreover , the lower levels of CD80 expression suggest a possible impairment of myeloid and/or antigen-presenting cell function upon BCG infection in patients , but not in healthy heterozygous or homozygous WT subjects ( GSE102862 ) . Peripheral leukocytes from IRF4-heterozygous individuals therefore had a phenotype in terms of IRF4 production and function .
WD was initially described as an inflammatory disease ( Whipple , 1907 ) but was subsequently shown to be infectious ( Raoult et al . , 2000; Relman et al . , 1992; Yardley and Hendrix , 1961 ) . We provide evidence that WD is also a genetic disorder . We show here that , in a large multiplex kindred , heterozygosity for the private , LOF R98W mutation of IRF4 underlies an AD form of WD with incomplete penetrance . The causal relationship between IRF4 genotype and WD was demonstrated as follows . First , the IRF4 R98W mutation is the only non-synonymous rare variant segregating with WD in this kindred . Second , the mutation was demonstrated experimentally to be LOF , unlike 146 of 153 other non-synonymous coding IRF4 variants in the general population . Only seven of the 153 non-synonymous coding variants identified ( including the five predicted to be LOF in the gnomAD database ) were found to be LOF ( n = 6 ) or hypomorphic ( n = 1 ) and they were all extremely rare ( cumulative MAF <4 × 10−5 ) . Moreover , IRF4 has evolved under purifying selection , suggesting that deleterious heterozygous variants of this gene entail fitness costs ( Barreiro and Quintana-Murci , 2010; Quintana-Murci and Clark , 2013; Rieux-Laucat and Casanova , 2014 ) . Third , EBV-B cells and activated CD4+ T cells heterozygous for IRF4 R98W have a distinctive phenotype , particularly for IRF4 expression in the cytoplasm . This mutation also has a strong functional impact on gene expression in IRF4 R98W-heterozygous PBMCs stimulated with BCG or Tw . These findings unequivocally show that heterozygosity for the R98W allele of IRF4 is the genetic etiology of WD in this kindred . Although we did not find any IRF4 mutations in a pilot cohort of 25 patients with sporadic WD , these and other patients may also develop WD due to other inborn errors of immunity , possibly related to IRF4 , as suggested by the apparent genetic heterogeneity and physiological homogeneity underlying severe infectious diseases ( Andersen et al . , 2015; Casanova , 2015a; Casanova , 2015b; Ciancanelli et al . , 2015; Israel et al . , 2017; Lamborn et al . , 2017; Ogunjimi et al . , 2017; Vanhollebeke et al . , 2006; Zhang et al . , 2018; Zhang et al . , 2015 ) . This observation therefore extends our model , in which life-threatening infectious diseases striking otherwise healthy individuals during primary infection can result from single-gene inborn errors of immunity . In this kindred with AD IRF4 deficiency , haploinsufficiency was identified as the key mechanism , although IRF4 protein levels in the cytoplasmic compartment were higher in patients with the mutation than in wild-type homozygotes . The protein was not more abundant in the nucleus , where IRF4 exerts its effects on transcription . Moreover , half of the IRF4 mRNA in EBV-B cells from heterozygous subjects is WT . The total amount of IRF4 mRNA was higher in the EBV-B cells of heterozygous subjects , but the total amount of IRF4 protein in the nuclear compartment of heterozygous EBV-B and activated CD4+ T cells was similar to that in WT homozygous cells . These data suggest that no more than half the IRF4 protein in the nucleus is WT in heterozygous cells . In addition , not only is IRF4 subject to purifying selection , but also the R98W mutation is itself LOF , with no detectable dominant-negative effect at cell level . Haploinsufficiency is an increasingly recognized mechanism underlying AD inborn errors of immunity ( Afzali et al . , 2017; Rieux-Laucat and Casanova , 2014 ) . It is commonly due to loss-of-expression alleles , contrasting with the negative dominance typically exerted by expressed proteins , but many mutations are known to cause haploinsufficiency without actually preventing protein production ( Afzali et al . , 2017; Pérez de Diego et al . , 2010; Rieux-Laucat and Casanova , 2014 ) . Haploinsufficiency in this kindred is not due to loss-of-expression of IRF4 . Instead , it results from a lack of activity of the R98W IRF4 proteins present in the nucleus . Incomplete penetrance is common in conditions resulting from haploinsufficiency . In this kindred , incomplete penetrance may result from a lack of Tw infection ( in heterozygous individuals IV . 5 and VI . 7 ) , or a lack of WD development in infected individuals ( in heterozygous individuals III . 6 , IV . 4 , V . 3 , V . 4 , VI . 6 ) . All five chronic carriers of Tw were heterozygous for the IRF4 R98W mutation , suggesting that AD IRF4 deficiency also favors the development of chronic Tw carriage . The five asymptomatic carriers were 24 to 82 years old , whereas the four patients were 69 to 92 years old . The impact of IRF4 R98W may therefore increase with age , initially facilitating chronic carriage in Tw-infected individuals , and subsequently predisposing chronic carriers to the development of WD . We cannot exclude the possibility that a modifier allele at another locus contributes to the development of WD in infected heterozygous individuals with IRF4 mutations . Future studies will attempt to define the cellular basis of WD in individuals with IRF4 mutations . The apparently normal development of all peripheral myeloid and lymphoid blood cell subsets studied in patients , and the selective predisposition of these individuals to WD suggest that the disease mechanism is subtle and specifically affects protective immunity to Tw and that it may act in the gastrointestinal ( GI ) tract . Interestingly , the data of several public databases indicate that IRF4 RNA is expressed in the stomach , colon , and small intestine ( https://www . gtexportal . org , http://biogps . org ) , and that the IRF4 protein is expressed in glandular cells from the stomach , duodenum , small intestine , and rectum ( https://www . proteinatlas . org ) . In addition , a recent analysis of human intestinal macrophage subsets ( Bujko et al . , 2018 ) showed IRF4 RNA to be expressed in human intestinal myeloid resident cells . Further studies of GI-tract-resident cells , including myeloid and lymphoid cells in particular , should make it possible to decipher the molecular and cellular mechanisms by which human IRF4 haploinsufficiency underlies WD upon infection by Tw .
All members of the multiplex kindred studied , the pedigree of which is shown in Figure 1A , live in France and are of French descent . Informed consent was obtained from all family members , and the study was approved by the national ethics committee . Patient 1 ( P1 , proband ) was born in 1948 and presented arthritis of the right knee in 2011 , after recurrent episodes of arthritis of this joint associated with effusion since 1980 . Tropheryma whipplei ( Tw ) was detected in synovial fluid by PCR and culture in 2011 , but was not detected by PCR in saliva , feces , and small-bowel biopsy specimens . Physical examination revealed a large effusion of the right knee , limiting mobility . The fluid aspirated from this joint contained 4 , 000 erythrocytes/mm3 and 8 , 800 leukocytes/mm3 , but no crystals or evidence of microbes . Synovial hypertrophy of the right knee and a narrowing of the right internal femoro-tibial joint were detected on MRI . X-ray showed an extension of the right femoro-tibial joint and erosion of the posterior part of the femoro-tibial joint . However , erythrocyte sedimentation rate ( ESR ) ( 3 mm/h ) and C-reactive protein ( CRP ) ( 1 . 8 mg/l ) determinations gave negative results . P1 received methotrexate ( 15 mg/week ) for 4 months , without remission . Antibiotic treatment with doxycycline ( 200 mg/day ) was then immediately initiated . The arthralgia resolved , but right knee effusion persisted . Hydroxychloroquine was therefore added to the treatment regimen . At last follow-up , in 2016 , the patient was well . P2 , a second cousin of P1 , was born in 1941 and was diagnosed with classical WD and digestive problems in 1978 , based on positive periodic acid–Schiff ( PAS ) staining of a small intestine biopsy specimen . She was treated with sulfamethoxazole/trimethoprim . At last follow-up , in 2016 , Tw PCR was positive for the saliva and feces . P3 , the father of P1 , was born in 1925 and was diagnosed with classical WD in 1987 on the basis of positive PAS staining of a small intestine biopsy specimen . Clinical manifestations included diarrhea , abdominal pain and weight loss . P3 displayed no extraintestinal manifestations . He was successfully treated with sulfamethoxazole/trimethoprim , with complete clinical and bacteriological remission . P4 , the brother of P2 , was born in 1947 and sought medical advice in 2015 for arthralgia affecting the knees and right ulna-carpal joints . The other joints were unaffected . A culture of the joint fluid was negative for bacteria , but Tw was not sought . Tw was not detected in the saliva and feces by PCR or culture , but serological tests for Tw were positive . The fluid aspirated from the right knee contained 4 , 800 erythrocytes/mm3 and 10 , 900 leukocytes/mm3 ( 91% neutrophils and 9% lymphocytes ) without crystals . Blood tests revealed an ESR of 30 mm/h and a CRP concentration of 50 mg/l , with no rheumatoid factor , anti-cyclic citrullinated peptide antibodies ( anti-CCP ) or anti-nuclear antibodies . An X-ray revealed a narrowing of the joint space in the knees and vertebral hyperostosis was visible . The joints of the hands were unaffected . The patient was treated with anti-inflammatory drugs , without success . Treatment with methotrexate and steroids was introduced , followed by antibiotics , the effect of which is currently being evaluated . Saliva and/or feces samples from 18 other members of the family were checked for the presence of Tw , by a PCR specifically targeting T . whipplei , as previously described ( Figure 1A , Figure 1—source data 1 ) ( Edouard et al . , 2012 ) . Five individuals were found to be chronic carriers ( mean age: 55 years ) and 13 were not ( mean age: 38 years ) . Testing was not possible for nine other relatives . The overall distribution of WD in this kindred was suggestive of an AD trait with incomplete penetrance . Genome-wide linkage analysis was performed by combining genome-wide array and whole-exome sequencing ( WES ) data ( Belkadi et al . , 2016 ) . In total , nine family members were genotyped with the Genome-Wide Human SNP Array 6 . 0 . Genotype calling was achieved with the Affymetrix Power Tools Software Package ( http://www . affymetrix . com/estore/partners_programs/programs/developer/tools/powertools . affx ) . SNPs were selected with population-based filters ( Purcell et al . , 2007 ) , resulting in the use of 905 , 420 SNPs for linkage analysis . WES was performed as described in the corresponding section , in four family members , P1 , P2 , P3 and P4 . In total , 64 , 348 WES variants were retained after application of the following filtering criteria: genotype quality ( GQ ) >40 , minor read ratio ( MRR ) >0 . 3 , individual depth ( DP ) >20 x , retaining only diallelic variants with an existing RS number and a call rate of 100% . Parametric multipoint linkage analysis was performed with the Merlin program ( Abecasis et al . , 2002 ) , using the combined set of 960 , 267 variants . We assumed an AD mode of inheritance , with a frequency of the deleterious allele of 10−5 and a penetrance varying with age ( 0 . 8 above the age of 65 years , and 0 . 02 below this threshold ) . Data for the family and for Europeans from the 1000 Genomes project were used to estimate allele frequencies and to define linkage clusters , with an r2 threshold of 0 . 4 . The method used for WES has been described elsewhere ( Bogunovic et al . , 2012; Byun et al . , 2010 ) . Briefly , genomic DNA extracted from the patients’ blood cells was sheared with a Covaris S2 Ultrasonicator ( Covaris ) . An adapter-ligated library was prepared with the Paired-End Sample Prep kit V1 ( Illumina ) . Exome capture was performed with the SureSelect Human All Exon kit ( 71 Mb version - Agilent Technologies ) . Paired-end sequencing was performed on an Illumina Genome Analyzer IIx ( Illumina ) , generating 72- or 100-base reads . We used a BWA-MEM aligner ( Li and Durbin , 2009 ) to align the sequences with the human genome reference sequence ( hg19 build ) . Downstream processing was carried out with the Genome analysis toolkit ( GATK ) ( McKenna et al . , 2010 ) SAMtools ( Li et al . , 2009 ) , and Picard Tools ( http://picard . sourceforge . net ) . Substitution calls were made with a GATK UnifiedGenotyper , whereas indel calls were made with a SomaticIndelDetectorV2 . All calls with a read coverage <2 x and a Phredscaled SNP quality <20 were filtered out . Single-nucleotide variants ( SNV ) were filtered on the basis of dbSNP135 ( http://www . ncbi . nlm . nih . gov/SNP/ ) and 1000 Genomes ( http:browser . 1000genomes . org/index . html ) data . All variants were annotated with ANNOVAR ( Wang et al . , 2010 ) . All IRF4 mutations identified by WES were confirmed by Sanger sequencing . PCR and serological tests for Tw were performed as previously described ( Fenollar et al . , 2009 ) . PBMCs were isolated by Ficoll-Hypaque density centrifugation ( GE Healthcare ) from cytopheresis or whole-blood samples obtained from healthy volunteers and patients , respectively . PBMCs and EBV-B cells ( purified B cells were immortalized with EBV in the laboratory ) were cultured in RPMI medium supplemented with 10% FBS , whereas HEK293T cells ( ATCC; CRL-3216 ) were cultured in DMEM medium supplemented with 10% FBS . Subsets were separated by MACS , using magnetic beads conjugated with the appropriate antibody ( Miltenyi Biotec ) according to the manufacturer’s protocol . All cells used in this study were tested for mycoplasma contamination and found to be negative . The full-length cDNA of IRF4 and PU . 1 was inserted into the pcDNA 3 . 1D/V5-His-TOPO vector with the directional TOPO expression kit ( Thermo Fisher Scientific ) . BATF anf JUN were obtained from Origene companie ( #RC207104 and #RC209804 , respectively ) . Constructs carrying mutant alleles were generated from this plasmid by mutagenesis with a site-directed mutagenesis kit ( QuikChangeII XL; Agilent Technologies ) , according to the manufacturer’s instructions . HEK293T cells were transiently transfected with the various constructs , using the Lipofectamine LTX kit ( Thermo Fisher Scientific ) in accordance with the manufacturer’s instructions . Total protein extracts were prepared by mixing cells with lysis buffer ( 50 mM Tris-HCl pH 7 . 4 , 150 mM NaCl , 0 . 5% Triton X-100 , and 2 mM EDTA ) supplemented with protease inhibitors ( Complete , Roche ) and phosphatase inhibitor cocktail ( PhoStop , Roche ) , 0 . 1 mM dithiothreitol DTT ( Life Technologies ) , 10 µg/ml pepstatin A ( Sigma , #P4265 ) , 10 µg/ml leupeptin ( Sigma , #L2884 ) , 10 µg/ml antipain dihydrochloride ( Sigma , #A6191 ) and incubating for 40 min on ice . A two-step extraction was performed to separate the cytoplasmic and nuclear content of the cells; cells were first mixed with a membrane lysis buffer ( 10 mM Hepes pH 7 . 9 , 10 mM KCl , 0 . 1 mM EDTA , 0 . 1 mM EGTA , 0 . 05 % NP40 , 25 mM NaF supplemented with 1 mM PMSF , 1 mM DTT , 10 µg/ml leupeptin , 10 µg/ml aprotinin ) and incubated for 30 min on ice . The lysate was centrifuged at 10 , 000 x g . The supernatant , corresponding to the cytoplasm-enriched fraction , was collected and the nuclear pellet was mixed with nuclear lysis buffer ( 20 mM Hepes pH 7 . 9 , 0 . 4 M NaCl , 1 , mM EDTA , 1 , mM EGTA , 25% glycerol supplemented with 1 mM PMSF , 1 mM DTT , 10 µg/ml leupeptin , 10 µg/ml aprotinin ) . Equal amounts of protein , according to a Bradford protein assay ( BioRad , Hercules , CA ) , were resolved by SDS-PAGE in a Criterion TGX 10% precast gel ( Biorad ) and transferred to a low-fluorescence PVDF membrane . Membranes were probed with unconjugated antibody: anti-IRF4 ( Santa Cruz , M-17 ) antibody was used at a dilution of 1:1000 and antibodies against GAPDH ( Santa Cruz , FL-335 ) , topoisomerase I ( Santa Cruz , C-21 ) , and/or lamin A/C ( Santa Cruz , H-110 ) were used as loading controls . The appropriate HRP-conjugated or infrared dye ( IRDye ) -conjugated secondary antibodies were incubated with the membrane for the detection of antibody binding by the ChemiDoc MP ( Biorad ) or Licor Odyssey CLx system ( Li-Cor , Lincoln , NE ) , respectively . Double-stranded unlabeled oligonucleotides ( cold probes ) were generated by annealing in TE buffer ( pH 7 . 9 ) supplemented with 33 . 3 mM NaCl and 0 . 67 mM MgCl2 . The annealing conditions were 100°C for 5 min , followed by cooling overnight at room temperature . After centrifugation at 3000 x g at 4°C for 30 min , the pellet was suspended in water . We labeled 0 . 1 µg of cold probe in Klenow buffer supplemented with 9 . 99 mM dNTP without ATP , 10 U Klenow fragment ( NEB ) and 50 µCi d-ATP-32P , at 37°C for 60 min . Labeled probes were purified on Illustra MicroSpin G-25 Columns ( GE Healthcare Life Sciences ) , according to the manufacturer’s protocol . We incubated 10 µg of nuclear protein lysate on ice for 30 min with a 32P-labeled ( a-dATP ) ISRE probe ( 5′ – gat cGG GAA AGG GAA ACC GAA ACT GAA-3′ ) designed on the basis of the ISG15 promoter or the λB probe ( 5′- gat cGC TCT TTA TTT TCC TTC ACT TTG GTT AC-3′ ) described by Brass et al . in 1999 ( Brass et al . , 1999 ) . For supershift assays , nuclear protein lysates were incubated for 30 min on ice with 2 µg of anti-IRF4 ( Santa Cruz , M-17 ) antibody or anti-goat Ig ( Santa Cruz ) antibody . Protein/oligonucleotide mixtures were then subjected to electrophoresis in 12 . 5% acrylamide/bis-acrylamide 37 . 5:1 gels in 0 . 5% TBE migration buffer for 80 min at 200 mA . Gels were dried on Whatman paper at 80°C for 30 min and placed in a phosphor-screen cassette for 5 days . Radioactivity levels were analyzed with the Fluorescent Image Analyzer FLA-3000 system ( Fujifilm ) . The ( ISRE ) 3 reporter plasmid ( pGL4 . 10[luc2] backbone , Promega #E6651 ) , which contains three repeats of the ISRE sequence separated by spacers , was kindly provided by Prof . Aviva Azriel ( Department of Biotechnology and Food Engineering , Technion-Israel Institute of Technology ) . The AICE reporter plasmid ( pGL4 . 10[luc2] backbone , Promega #E6651 ) contains part of the IL23R promoter ( −254 to −216 ) . HEK293T cells were transiently transfected with the ( ISRE ) 3 reporter plasmid ( 100 ng/well on a 96-well plate ) , the pRL-SV40 vector ( Promega # E2231 , 40 ng/well ) and a IRF4 WT or mutant pcDNA 3 . 1D/V5-His-TOPO plasmid ( Invitrogen #K4900-01 , 25 ng/well or the amount indicated , made up to 100 ng with empty plasmid ) , with the Lipofectamine LTX kit ( Thermo Fisher Scientific ) , according to the manufacturer’s instructions . For the AICE assay , we used the same protocol , but with the addition of BATF and JUN expression plasmids ( 25 ng/well each ) . Cells were used 24 hr after transfection for the ISRE assay and 48 hr after transfection for the AICE assay , with the Dual-Luciferase 1000 assay system kit ( Promega #E1980 ) , according to the manufacturer’s protocol . Signal intensity was determined with a Victor X4 plate reader ( Perkin Elmer ) . Experiments were performed in triplicate and reporter activity is expressed as fold-induction relative to cells transfected with the empty vector . Negative dominance was assessed by performing the same protocol with the following modifications: ( ISRE ) 3 reporter plasmid ( 100 ng/well for a 96-well plate ) , pRL-SV40 vector ( 40 ng/well ) , +IRF4+ WT and mutant plasmids were used to cotransfect cells , with a constant amount of WT plasmid ( 25 ng/well ) but various amounts of mutant plasmid ( 25 ng/well alone made up to 100 ng with empty plasmid or with the indicated amount , made up to 100 ng with empty plasmid ) amounting to a total of 240 ng/well . For the AICE assay , the same protocol was used , but with the addition of BATF and JUN expression plasmids ( 25 ng/well each ) . For the microarray analysis of PBMCs , cells from six IRF4-heterozygous individuals ( three patients , P1-P3; and three asymptomatic relatives , HET1-HET3 ) and six IRF4 WT-homozygous individuals ( four healthy relatives , WT1-WT4; and two healthy unrelated controls , C1-C2 ) were dispensed into a 96-well plate at a density of 200 , 000 cells/well and were infected in vitro with live Tw at a multiplicity of infection ( MOI ) of 1 , or with live BCG ( M . bovis-BCG , Pasteur substrain ) at a MOI of 20 , or were left uninfected ( mock ) . Two wells per condition were combined 24 hr post-infection for total RNA isolation with the ZR RNA Microprep kit ( Zymo Research ) . For the microarray on EBV-B cells , we used 400 , 000 cells from three IRF4-heterozygous mutation carriers and three WT individuals from the kindred for total RNA isolation with the ZR RNA Microprep kit ( Zymo Research ) . Microarray experiments on both PBMCs and EBV-B cells were performed with the Affymetrix Human Gene 2 . 0 ST Array . Raw expression data were normalized by the robust multi-array average expression ( RMA ) method implemented in the affy R package ( Gautier et al . , 2004; Irizarry et al . , 2003 ) . Normalized expression data were processed as previously described ( Alsina et al . , 2014 ) and briefly summarized here . First , fold-changes ( FC ) in expression between mock-infected and BCG-infected or Tw-infected conditions were calculated for each individual separately . For each set of conditions , transcripts were further filtered based on a minimal 1 . 5 FC in expression ( up- or downregulation ) . In a final stage , transcripts satisfying the previous filters in at least four of the six homozygous WT individuals for each in vitro infection condition were retained for downstream analysis . We counted the differentially expressed ( DE ) transcripts affected by stimulation in samples from all subjects for each stimulus , and determined the mean counts for these DE transcripts in all homozygotes . The mean values obtained were then used to normalize the counts of DE transcripts , yielding an overall transcriptional responsiveness for each individual separately , and for each stimulus . This overall responsiveness of subjects to either Tw or BCG is shown as a heatmap , and individual subjects were grouped by unsupervised hierarchical clustering . Responsive transcripts were further analyzed with Ingenuity Pathway Analysis ( IPA ) Software , Version 28820210 ( QIAGEN ) ( Alsina et al . , 2014 ) for functional interpretation . In brief , the FC values for each individual and treatment were used as input data for the identification of canonical pathway enrichment ( z-score cut-off set at 0 . 1 ) . The activation z-score values calculated for the identified pathways were exported from IPA and used to calculate mean values and differences between WT homozygotes and heterozygotes , and for graphical representation , with Microsoft Excel and GraphPad Prism Version 7 . 0 , respectively . The direction of the difference was not considered further . Negative mean difference values were converted into positive values before the ranking of the canonical pathways according to the difference between the genotypes . The microarray data used in this study have been deposited in the NCBI Gene Expression Omnibus ( GEO ) database , under accession number GSE102862 . Total RNA was prepared from the EBV-B cells of individuals heterozygous for IRF4 mutations ( two patients and two asymptomatic relatives ) , and healthy homozygous WT relatives ( n = 4 ) . We also included samples from unrelated individuals ( seven healthy controls and 25 patients with Tw carriage; all IRF4 coding exons for each individual were sequenced and shown to be WT ) . Moreover , the 25 WD patients included in this experiment were found to have an intact IRF4 cDNA structure and normal levels of IRF4 protein production in EBV-B cells . RNA was prepared from 500 , 000 cells with the ZR RNA Microprep kit ( Zymo Research ) , according to the manufacturer’s instructions . A mixture of random octamers and oligo dT-16 was used , with the MuLV reverse transcriptase ( High-Capacity RNA-to-cDNA kit , Thermo Fisher Scientific ) , to generate cDNA . Quantitative real-time PCR was performed with the TaqMan Universal PCR Master Mix ( Thermo Fisher Scientific ) , the IRF4-specific primer ( Hs01056533_m1 , Thermo Fisher Scientific ) and the endogenous human β-glucuronidase ( GUSB ) as a control ( 4326320E , Thermo Fisher Scientific ) . Data were analyzed by the ΔΔCt method , with normalization against GUSB . The full-length cDNA generated from the EBV-B cells of IRF4-heterozygous and WT homozygous individuals was used for the PCR amplification of exon 3 of IRF4 . The products obtained were cloned with the TOPO TA cloning kit ( pCR2 . 1-TOPO TA vector , Thermo Fisher Scientific ) , according to the manufacturer’s instructions . They were then used to transform chemically competent bacteria , and 100 clones per individual were Sanger-sequenced with M13 primers ( forward and reverse ) . Isolated CD4+ T cells from patients ( P1 and P3 ) and from healthy unrelated controls ( C1-C4 ) were either left unstimulated ( NS ) or stimulated with activating anti-CD2/CD3/CD28 mAb-coated beads ( Miltenyi Biotec ) for 24 hr in RPMI medium supplemented with 10% FBS ( 24-well plate ) . After isolation , monocytes from P1 or two healthy unrelated controls were plated ( 24-well plate; 600 , 000 cells/well ) in RPMI medium supplemented with 40% human serum ( M1-like ) or 10% FBS ( M2-like ) . Differentiation cytokines ( R and D Systems ) were immediately added: 0 . 5 ng/ml rhGM-CSF ( M1-like ) or 20 ng/ml rhM-CSF ( M2-like ) . Every 3 days , we replaced 30% of the medium with fresh complete medium supplemented with the appropriate cytokines . After 14 days of differentiation , cells were left unstimulated ( NS ) or were activated by incubation with 2 . 5 ng/ml IFN-γ ( M1-like ) or 50 ng/ml rh-IL-4 ( M2-like ) for 48 hr . Differentiated/activated MDMs were detached by treatment with trypsin ( 1 . 6 µg/ml ) in PBS . Cells were treated with Fc receptor blocking agent ( Miltenyi Biotec ) and Aqua Dead Cell Stain kit ( Thermo Fisher Scientific ) for 1 hr . They were then washed and stained for 1 hr at room temperature with appropriate antibodies ( see Figure 5—figure supplement 4 ) and appropriate isotype controls ( BD biosciences ) . Samples were analyzed on a Beckman Coulter Gallios flow cytometer . PBMCs from healthy unrelated controls and patients ( P1 , P2 and P3 ) were stained with antibodies against CD20 , CD10 , and CD27 , and IgM , IgD , IgG , or IgA . The percentages of transitional ( CD20+ CD10+ CD27- ) , naïve ( CD20+ CD10- CD27- ) and memory ( CD20+ CD10- CD27+ ) B cells were determined by flow cytometry . We then assessed the IgM/IgD or IgG or IgA expression of the memory B cells , to determine the extent of Ig isotype switching in the memory compartment . CD4+ T cells were isolated as previously described ( Ma et al . , 2012 ) . Briefly , cells were labeled with anti-CD4 , anti-CD45RA , and anti-CCR7 antibodies , and naive ( defined as CD45RA+ CCR7+ CD4+ ) T cells or effector/memory T cells ( defined as CD45RA- CCR7± CD4+ ) were isolated ( >98% purity ) with a FACS Aria cell sorter ( BD Biosciences ) . Purified naive or effector/memory CD4+ T cells were cultured with T-cell activation and expansion beads ( anti-CD2/CD3/CD28; Miltenyi Biotec ) for 5 days . Culture supernatants were then used to assess the secretion of IL-2 , IL-4 , IL-5 , IL-6 , IL-9 , IL-10 , IL-13 , IL-17A , IL-17F , IFNγ , and TNFα in a cytometric bead array ( BD biosciences ) , and the secretion of IL-22 , by ELISA ( Peprotech ) . Naïve CD4+ T cells ( CD45RA+ CCR7+ ) were isolated ( >98% purity ) from healthy unrelated controls or patients , with a FACS Aria sorter ( BD Biosciences ) . They were cultured under polarizing conditions , as previously described ( Ma et al . , 2016 ) . Briefly , cells were cultured with T-cell activation and expansion beads ( anti-CD2/CD3/CD28; Miltenyi Biotec ) alone or under Th1 ( IL-12 [20 ng/ml; R and D Systems] ) or Th17 ( TGFβ , IL-1β [20 ng/ml; Peprotech] , IL-6 [50 ng/ml; PeproTech] , IL-21 [50 ng/ml; PeproTech] , IL-23 [20 ng/ml; eBioscience] , anti-IL-4 [5 μg/ml] , and anti-IFN-γ [5 μg/ml; eBioscience] ) polarizing conditions . After 5 days , culture supernatants were used to assess the secretion of the cytokines indicated , by ELISA ( IL-22 ) , or with a cytometric bead array ( all other cytokines ) . | In 1907 , George Hoyt Whipple – an American pathologist working at Johns Hopkins University in Baltimore – described a new inflammatory disease that affects the intestine . Patients with this condition , now known as Whipple’s disease , experience diarrhea , weight loss , and abdominal and joint pain . The disease is rare; it affects about one in a million people , mostly those over the age of 50 who are of European descent . Later it was discovered that bacteria called Tropheryma whipplei cause Whipple’s disease and that antibiotics can cure it . These bacteria are widespread and yet only a small minority of individuals infected with T . whipplei goes on to develop Whipple’s disease . In some populations , over 50% of individuals have been infected with the bacteria at some point in their lives , but most will get rid of the infection . This raised the question: when exposed to the same microbe , why do some individuals develop a severe disease , while others remain unharmed ? From the 1950s onwards , scientists identified a few families with multiple members who have developed Whipple’s disease . These observations suggested that human genes may play a role in determining whether a person infected with T . whipplei becomes ill . Rare genetic mutations that affect the immune system have also been linked to the development of life-threatening cases of influenza or tuberculosis . Now , Guérin et al . report that , in one French family , an extremely rare mutation in the gene that codes for a protein called IRF4 may contribute to the development of Whipple’s disease . This family had four members with Whipple’s disease , all of whom had one copy of the gene with this rare mutation and one normal copy of the gene . The IRF4 protein acts like a switch that turns on and off some genes involved in the body’s response to infection . In patients with this mutation , the IRF4 protein does not work as it should . Guérin et al . suggest that Whipple’s disease may be caused by specific genetic mutations affecting the immune system in subjects infected by T . whipplei . More studies are needed to see if other genetic mutations also contribute to other cases of Whipple’s disease . Such studies may help physicians to better understand why some people become sick after T . whipplei infections while others do not . They may also help physicians to diagnose the disease , and even lead to better treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2018 | IRF4 haploinsufficiency in a family with Whipple’s disease |
During gamete formation , crossover recombination must occur on replicated DNA to ensure proper chromosome segregation in the first meiotic division . We identified a Mec1/ATR- and Dbf4-dependent replication checkpoint in budding yeast that prevents the earliest stage of recombination , the programmed induction of DNA double-strand breaks ( DSBs ) , when pre-meiotic DNA replication was delayed . The checkpoint acts through three complementary mechanisms: inhibition of Mer2 phosphorylation by Dbf4-dependent Cdc7 kinase , preclusion of chromosomal loading of Rec114 and Mre11 , and lowered abundance of the Spo11 nuclease . Without this checkpoint , cells formed DSBs on partially replicated chromosomes . Importantly , such DSBs frequently failed to be repaired and impeded further DNA synthesis , leading to a rapid loss in cell viability . We conclude that a checkpoint-dependent constraint of DSB formation to duplicated DNA is critical not only for meiotic chromosome assortment , but also to protect genome integrity during gametogenesis .
During meiosis , a single round of DNA replication is followed by two nuclear divisions to produce haploid gametes from diploid progenitor cells . In most organisms , the faithful segregation of homologous chromosome pairs in meiosis I relies on physical connections between homologs produced by meiotic recombination . In budding yeast , DNA exchanges between homologs are the result of the repair of ∼160 DNA double-strand breaks ( DSBs ) , which occur immediately following pre-meiotic S phase ( meiS ) and are distributed across all 16 chromosome pairs . DSB formation requires the concerted activity of 10 proteins , including the catalytic subunit Spo11 and its partner Ski8 , the Mre11/Rad50/Xrs2 ( MRX ) complex , and the meiosis-specific proteins Mer2 , Rec114 , Mei4 , Rec102 and Rec104 ( reviewed in ( Keeney and Neale , 2006 ) ) . Additionally , DSB formation requires the phosphorylation of Mer2 by two cell cycle kinases , cyclin dependent-kinase ( CDK ) and the Dbf4-dependent Cdc7 kinase ( DDK ) ( Henderson et al . , 2006; Sasanuma et al . , 2008; Wan et al . , 2008 ) . The complexity of this process reflects the fact that meiotic genome fragmentation needs to be carefully controlled to limit genome instability . Under normal circumstances , meiotic DSBs are introduced after DNA replication . This temporal separation is necessary because the crossover-mediated linkages between homologs require sister-chromatid cohesion distal to the crossover site . Thus , only crossovers formed after DNA replication serve to hold homologous chromosomes together in metaphase I . Moreover , replication forks are unable to cross a DSB ( Doksani et al . , 2009 ) , so the presence of >100 DSBs in the genome would severely interfere with the completion of DNA replication . The mechanisms that coordinate pre-meiotic DNA replication and DSB formation are not well understood . DNA replication and DSB formation are coordinated at the local level , because delayed replication of a single chromosome arm delays DSB formation on that arm ( Borde et al . , 2000 ) . However , DSB formation does not require DNA replication; pre-meiotic replication initiation mutants introduce full levels of DSBs on chromosomes that are not replicated in both budding and fission yeasts ( Murakami and Nurse , 2001; Hochwagen et al . , 2005; Blitzblau et al . , 2012 ) . In addition , the initiation of meiotic recombination is prevented globally when DNA replication is delayed by nucleotide depletion ( Schild and Byers , 1978; Tonami et al . , 2005; Ogino and Masai , 2006 ) . In fission yeast , a replication checkpoint blocks DSB formation in this situation ( Tonami et al . , 2005; Ogino and Masai , 2006 ) . Although a related checkpoint was found to delay the meiotic divisions upon nucleotide depletion in budding yeast ( Stuart and Wittenberg , 1998 ) , a subsequent study came to the conclusion that it did not regulate DSB formation ( Borde et al . , 2000 ) . Thus , it remains unclear how budding yeast prevent DSB formation on unreplicated DNA . The replication checkpoint couples DNA replication and cell cycle progression by sensing and coordinating the response to delayed replication forks ( reviewed in ( Labib and De Piccoli , 2011 ) . In vegetatively growing yeast cells , stalled replication forks activate a conserved kinase cascade including Mec1/ATR and Rad53/CHK2 . Mec1 and Rad53 inhibit cell cycle progression by preventing chromosome segregation and mitotic entry , respectively ( Clarke et al . , 2001; Clarke et al . , 2003 ) . Additionally , their activation stabilizes replication forks , preventing catastrophic fork collapse . Finally , activated Rad53 also phosphorylates and directly interacts with the Dbf4 subunit of DDK ( Weinreich and Stillman , 1999; Duncker et al . , 2002; Chen et al . , 2013 ) , which delays further initiation of DNA replication . The replication checkpoint has not been characterized in budding yeast meiosis . It presumably functions , because cells can respond to and recover from replication inhibition ( Schild and Byers , 1978; Blitzblau et al . , 2012 ) and meiotic functions of both Mec1 and Rad53 have been described ( reviewed in ( MacQueen and Hochwagen , 2011 ) ) . The fact that the replication checkpoint inhibits DDK activity in mitotic cells suggests that the checkpoint could be easily adapted to prevent meiotic DSBs by preventing the DDK-dependent phosphorylation of Mer2 . To characterize the replication checkpoint during meiotic cell division , we investigated the effects of inhibiting pre-meiotic DNA replication . We found that the replication checkpoint is active in budding yeast meiosis and inhibits DSB formation . The checkpoint uses parallel mechanisms to regulate the abundance , DNA loading and DDK-dependent phosphorylation of DSB factors . Cells that formed DSBs on partially replicated chromosomes were unable to complete either DSB repair or genome duplication , revealing that the separation of DNA replication and meiotic DSB formation is critical for maintaining genome integrity and viability .
To determine whether DSB formation is coordinated with bulk DNA replication during meiS in budding yeast , we exposed cells to increasing doses of the replication inhibitor hydroxyurea ( HU ) and measured the kinetics of DNA replication and DSB formation ( using a rad50S mutation to prevent DSB repair ) . Because we previously observed significantly delayed meiotic entry when cells were treated with high concentrations of HU ( Blitzblau et al . , 2012 ) , all of our analyses were carried out with 20 mM or lower amounts of HU . FACS analysis of total DNA content revealed that DNA replication occurred between 1–3 hr for wild-type cells in the absence of HU , was significantly delayed in 5 mM HU , and arrested in early S phase in 20 mM HU ( Figure 1A ) . DSB formation was comparably affected when measured by Southern blot analysis of a prominent DSB hotspot on chromosome 3 ( Figure 1B ) . Quantification of FACS profiles and Southern analysis revealed that DSBs appeared just after bulk DNA replication was completed ( 4C DNA content appeared ) in 0 or 5 mM HU samples , and were fully suppressed when replication was blocked by 20 mM HU ( Figure 1C ) . Consistent with the idea that slowing DNA replication activates the replication checkpoint , we detected HU-dependent Rad53 autophosphorylation by Western blotting ( Figure 1D ) , which has been shown to be a direct effect of checkpoint activation in pre-mitotic cells ( Pellicioli et al . , 1999 ) . In addition , we found that Mec1 and Rad53 were essential to maintain viability in HU-treated pre-meiotic cells ( Figure 1—figure supplement 1A ) , indicating that activation of the pre-meiotic replication checkpoint is critical to maintain replication forks in the presence of replication inhibition , as in pre-mitotic cells . 10 . 7554/eLife . 00844 . 003Figure 1 . Ongoing DNA replication delays meiotic DSB formation . rad50S ( H156 , A–D ) or cdc6-mn rad50S ( H155 , E–H ) cells were induced to enter meiosis in 0 , 5 or 20 mM HU and analyzed at the indicated time points . ( A and E ) FACS analysis of total DNA content . ( B and F ) Southern blot analysis of DSB formation at the yCR048W DSB hotspot . Arrows indicate the major DSB bands quantified in ( C and G ) . ( C and G ) Quantification of 4C DNA content from FACS is shown in the upper panel . The measurement of DSBs from Southern blot is plotted in the lower panel . ( D and H ) Western blot analysis of Rad53 protein mobility is shown as a measurement of phosphorylation and activation . Slower migrating bands correspond to phosphorylated Rad53 . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 00310 . 7554/eLife . 00844 . 004Figure 1—figure supplement 1 . Replication and checkpoint requirements during meiS . ( A ) The viability of sml1Δ cells ( H3554 ) , mec1-1 sml1Δ ( H2560 ) and rad53Δ sml1Δ ( H2591 ) cells is plotted with respect to time in SPO . The number of viable colonies was normalized to the number at 0 hr for each culture . ( B ) Mcm2-7 genome-wide location ( ChIP-chip ) analysis is presented for wild-type ( H2544 , Blitzblau et al . , 2012 ) and cdc6-mn ( H154 ) cells as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 004 We tested whether the DSB delay in HU-treated cells was due to inhibited DNA replication , as the mitotic replication checkpoint is characteristically activated by ssDNA at the replication fork . DNA replication is strongly decreased in cdc6-mn cells ( Hochwagen et al . , 2005; Blitzblau et al . , 2012 ) . This decrease was due to impaired replicative helicase loading ( Figure 1—figure supplement 1B ) , and little DNA replication was observed in 0 , 5 or 20 mM HU ( Figure 1E ) . The cdc6-mn cells formed DSBs with wild-type kinetics in all concentrations of HU despite the absence of any completed DNA replication ( Figure 1F , G ) , consistent with the idea that depleting the number of active replication forks abrogates the replication checkpoint signal . Importantly , we were unable to detect phosphorylation of Rad53 at 2–3 hr when DSBs formed ( Figure 1H ) , indicating that the replication checkpoint is not efficiently activated in these cells . When cdc6-mn cells were exposed to concentrations of HU greater than 20 mM , DSB formation was either reduced or abolished without activation of Rad53 ( data not shown ) , consistent with our previous report that high levels of HU can inhibit meiotic cell cycle entry ( Blitzblau et al . , 2012 ) . This could explain why the checkpoint was previously not observed ( Borde et al . , 2000 ) . Together , these data confirm that the canonical Mec1- and Rad53-dependent replication checkpoint responds to delayed DNA replication in budding yeast meiosis , and that DSB formation is delayed while DNA replication is ongoing . Given that Rad53 inhibits DDK in mitotically dividing cells and that DDK activates the meiotic DSB factor Mer2 , we explored whether inhibition of pre-meiotic DNA replication delayed phosphorylation of Mer2 . As shown in Figure 2A , when pre-meiotic cells were treated with HU , Dbf4 accumulated mainly in a hyperphosphorylated state . The amount of hyperphosphorylated Dbf4 was reduced in both mec1Δ and rad53Δ cells treated with HU ( top panel ) , indicating that the massive accumulation of this form of the protein is checkpoint-dependent in meiotic cells as it is in mitotic cells ( Weinreich and Stillman , 1999 ) . For this analysis we used a polyclonal antibody to Dbf4 , having noted that the C-terminally myc-tagged protein was unstable and present at much lower levels and with degradation products , compared to the untagged protein ( data not shown ) . Consistent with the idea that the hyperphosphorylated form of Dbf4 is inactive in the cell , Mer2 accumulated in a hypophosphorylated form in HU-treated cells ( Figure 2A , bottom panel ) only when Mec1 and Rad53 were present . Because Mer2 is sequentially phosphorylated by both CDK and DDK prior to meiotic DSB formation ( Sasanuma et al . , 2008; Wan et al . , 2008 ) , we confirmed that Mer2 was phosphorylated by CDK in HU-treated cells ( Figure 2—figure supplement 1A , B ) , indicating that HU treatment specifically inhibited DDK-dependent phosphorylation of Mer2 . These results indicate that the replication checkpoint kinases Mec1 and Rad53 restrain DSB formation by limiting the activity of DDK , a role well established in the mitotic replication checkpoint . 10 . 7554/eLife . 00844 . 005Figure 2 . The pre-meiotic replication checkpoint inhibits DDK kinase activity . ( A ) Western blot analysis of Dbf4 ( top panel ) and Mer2-5myc ( bottom panel ) in sml1Δ ( H5157 ) , mec1Δ sml1Δ ( H5220 ) and rad53Δ sml1Δ ( H5127 ) cells . The sml1Δ mutation was used to maintain viability of mec1Δ and rad53Δ mutants . For Mer2-5myc blotting , only 20% ( wild-type ) or 50% ( mec1Δ and rad53Δ ) of total protein was loaded for HU-treated samples as high accumulation of the Mer2 protein obscured the analysis of mobility shifts . ( B ) Schematic of wild-type and mutant Dbf4 proteins analyzed in this study . N , M , and C refer to the N-terminal , middle and C-terminal conserved domains . ( C ) Southern blot analysis of DSB formation at the yCR048W DSB hotspot in rad50S cells for DBF4/DBF4 ( H6097 ) , dbf4-NLS-ΔN221/dbf4-NLS-ΔN221 ( H6146 ) and dbf4-NLS-ΔN221/DBF4 ( H7335 ) cells . ( D ) FACS analysis of total DNA content of the strains in ( C ) . See also Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 00510 . 7554/eLife . 00844 . 006Figure 2—figure supplement 1 . Regulation of DDK prevents DSBs in HU-treated cells . ( A ) Western blot analysis of Mer2-5myc in an ndt80Δ spo11-Y135F-HA strain ( H5079 ) probed with anti-myc ( top panel ) for total protein and anti-Mer2 phospho-S30 antibody ( bottom panel ) in the absence or presence of HU . ( B ) Specificity of the anti-Mer2 phospho-S30 antibody was confirmed by comparing the signals on proteins from wild-type ( H4695 ) and clb5Δ clb6Δ ( H5076 ) cells after 5 hr in SPO medium by Western blotting using the same conditions as ( A ) . These cells are NDT80 ( i . e . , not blocked in prophase ) , so Mer2 is largely degraded in wild-type cells without HU treatment . ( C ) Southern blot analysis of DSBs at the yCR048W hotspot in dbf4-Δ71-221 rad50S cells ( H6296 ) . ( D ) Southern blot analysis of DSBs at the yCR048W hotspot in NLS-DBF4 rad50S cells ( H7309 ) . ( E ) Southern blot analysis of DSBs at the yCR048W hotspot in rad50S ( H4226 ) and dbf4-m25 rad50S ( H5603 ) cells . ( F ) Western blot analysis of Dbf4 ( H6097 ) and dbf4-ΔN221 ( H6146 ) proteins in the absence or presence of HU . ( G ) Southern blot analysis of DSBs at the yCR048W hotspot in dbf4-T163A rad50S ( H4883 ) cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 006 To test whether the activation of DDK is sufficient to allow for DSB formation in pre-meiotic cells , we sought an unregulated version of Dbf4 that could allow for DSB formation in HU-treated cells . We tested several previously characterized alleles of DBF4 in our Southern blot assay for DSBs ( Figure 2B , C , Figure 2—figure supplement 1 ) . Cells expressing dbf4-NLSΔN221 , an N-terminal truncation protein containing an SV40 nuclear localization signal ( NLS ) to support viability , formed high levels of DSBs in the presence of HU ( Figure 2C ) without a substantial increase in DNA replication ( Figure 2D ) . The replication checkpoint bypass was dominant as DSBs also occurred in cells heterozygous for this dbf4 allele ( Figure 2C ) . This result indicates that deregulating DDK activity is sufficient to allow the initiation of meiotic recombination in the presence of ongoing DNA replication . The dbf4-NLSΔN221 protein lacks the conserved N domain that has been shown to interact with Rad53 and other proteins ( Gabrielse et al . , 2006; Chen et al . , 2013 ) . However , cells harboring a smaller truncation of DBF4 that maintains the native NLS but removes the Rad53 interaction domain , dbf4-Δ71-221 , bypassed the replication checkpoint poorly ( Figure 2—figure supplement 1C ) , indicating that disrupting the Rad53 interaction alone is not sufficient to deregulate Dbf4 activity . Similarly , addition of the SV40 NLS to the wild-type Dbf4 protein did not allow for DSBs in HU-treated cells ( Figure 2—figure supplement 1D ) , indicating the strong NLS was not solely responsible for the replication checkpoint bypass of the dbf4-NLSΔN221 allele . Thus , multiple functions of the Dbf4 N-terminus are likely required for the regulation of meiotic DSBs in response replication stress . We conclude that DDK regulation is critical in the pre-meiotic replication checkpoint . Given that Dbf4 is regulated by Rad53-dependent phosphorylation , we explored the role of such phosphorylation in preventing meiotic DSBs in response to HU-treatment . A dbf4 allele lacking 25 potential phosphorylation sites , dbf4-m25 , allows the initiation of DNA replication from late origins in HU-treated pre-mitotic cells ( Lopez-Mosqueda et al . , 2010 ) . However , this mutation did not permit DSB formation in pre-meiotic cells treated with HU ( Figure 2—figure supplement 1E ) . This result indicates that simply preventing the Rad53-dependent phosphorylation of Dbf4 is insufficient to accumulate enough DDK activity to form meiotic DSBs in HU-treated cells . Furthermore , the dbf4-NLSΔN221 allele produced a truncated protein that shifted mobility consistent with phosphorylation in the presence of HU ( Figure 2—figure supplement 1F ) , despite the fact that 21 of the 25 mutations in the dbf4-m25 allele are within the deleted region . These results suggest either the dbf4-NLSΔN221 mutant protein is unable to respond to Rad53-dependent phosphorylation , or that separate phosphorylation sites are required to prevent meiotic DSBs . We ruled out a function for the sole Mec1 consensus site on Dbf4 , because mutation of threonine 163 to alanine did not allow DSB formation in HU ( Figure 2—figure supplement 1G ) . Together , these data reveal that deregulating DDK is sufficient to allow DSBs in HU-treated cells . However , simply preventing the phosphorylation of Dbf4 or its interaction with Rad53 does not produce enough DDK activity to allow for DSB formation in HU-treated cells . Given that Dbf4 activity is critical for DSB formation , we tested whether removing the checkpoint kinases Mec1 and Rad53 that control Dbf4 activity was sufficient to allow DSB formation in HU-treated cells . We found that disruption of Mec1 , either by deletion or using the mec1-1 allele ( Figure 3A , Figure 3—figure supplement 1 and Figure 4—figure supplement 1 ) , or removal of the Mec1-interacting protein Ddc2 ( Figure 3—figure supplement 1A ) was sufficient to allow DSB formation in HU-treated cells . DSB formation in mec1Δ cells occurred along entire chromosomes ( Figure 3—figure supplement 1B ) without significantly increased DNA replication ( Figure 3B , Figure 3—figure supplement 1C ) , indicating the replication checkpoint was bypassed . The levels of DSB formation in mec1Δ cells treated with HU were lower than those observed in the absence of HU ( Figure 3C ) , which we believe is due to relocalization of DSB factors in checkpoint mutants ( discussed in more detail later ) . In contrast , rad53Δ cells failed to form DSBs in the presence of HU ( Figure 3 , Figure 3—figure supplement 1C ) . These results were confirmed in dmc1Δ repair-deficient strains , which arrest in meiotic prophase with resected DNA ends ( Figure 3—figure supplement 1D ) , indicating they were not specific to the rad50S allele used in our initial assays . We eliminated the possibility that the pre-meiotic replication checkpoint utilizes alternate signaling kinases by simultaneously removing Mek1 , Chk1 and Rad53 , which did not allow DSB formation in HU-treated cells ( Figure 3—figure supplement 1E ) . Similarly , neither removal of Rad53 kinase activity nor the deletion of the Rad53 activating proteins Rad9 and Mrc1 allowed DSBs to form in HU-treated cells ( Figure 3—figure supplement 1F , G ) . Indeed , removal of Mec1 , but not Rad53 , Rad53 kinase activity or Rad53-dependent phosphorylation sites on Dbf4 , allowed DSB formation across entire chromosomes in both rad50S and dmc1Δ repair-deficient strains ( Figure 3—figure supplement 1B , D ) . We noted that the lack of checkpoint bypass in the absence of Rad53 is in contrast to the bypass observed in Dbf4 truncation mutants ( Figure 2 ) . These findings argue against a strictly linear pathway , in which Mec1 acts through Rad53 to inhibit DDK , but rather suggest that there are Rad53-independent functions of the replication checkpoint in meiosis . 10 . 7554/eLife . 00844 . 007Figure 3 . Removal of MEC1 , but not RAD53 , allows DSB formation in HU-treated cells . ( A ) Southern blot analysis of DSB formation at the yCR048W DSB hotspot in sml1Δ rad50S ( H4898 ) , mec1Δ sml1Δ rad50S ( H4935 ) and rad53Δ sml1Δ rad50S ( H4969 ) and mec1Δ rad53Δ sml1Δ rad50S cells ( H4932 ) in the absence or presence of HU . ( B ) FACS analysis of total DNA content of the strains in ( A ) . ( C ) Quantification of the relative DSB levels from the Southern blot in ( A ) in the absence ( solid lines ) or presence ( dashed lines ) of HU . DSBs levels were measured as in Figure 1 and normalized to the maximum measurement in the sml1Δ rad50S ( H4898 ) ‘wild-type’ strain ( shown as grey lines for comparison ) . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 00710 . 7554/eLife . 00844 . 008Figure 3—figure supplement 1 . Removing Mec1 , but not Rad53 , allows DSBs in HU-treated cells . ( A ) Southern blot analysis of DSBs at the yCR048W hotspot in sml1Δ rad50S ( H4898 ) , mec1Δ sml1Δ rad50S ( H4935 ) and ddc2Δ sml1Δ rad50S ( H6002 ) strains . ( B ) CHEF gel electrophoresis and Southern blot analysis of DSB formation on Chromosome 8 in sml1Δ rad50S ( H4898 ) , rad53Δ sml1Δ rad50S ( H4969 ) , and mec1Δ sml1Δ rad50S cells ( H4935 ) . The probe used for Southern blotting in SGD coordinates was: Chromosome VIII: 23 , 771-25 , 410 . ( C ) FACS analysis of total DNA content in sml1Δ rad50S ( H4898 ) and mec1-1 sml1Δ rad50S ( H4557 ) strains . ( D ) CHEF gel electrophoresis and Southern blot analysis of DSB formation on Chromosome 8 in sml1Δ dmc1Δ ( H4618 ) , mec1-1 sml1Δ dmc1Δ ( H4557 ) rad53Δ sml1Δ dmc1Δ ( H6813 ) , dmc1Δ cells ( H118 ) , rad53-kd dmc1Δ ( H6815 ) and dbf4-m25 dmc1Δ ( H6814 ) cells . The probe used for Southern blotting in SGD coordinates was: Chromosome VIII: 23 , 771-25 , 410 . ( E ) Southern blot analysis of DSBs at the yCR048W hotspot in sml1Δ rad50S ( H4898 ) and rad53Δ chk1Δ mek1Δ sml1Δ rad50S ( H5241 ) cells . ( F ) Southern blot analysis of DSBs at the yCR048W hotspot in rad50S ( H4226 ) and rad53-kd rad50S ( H5884 ) cells . ( G ) Southern blot analysis of DSBs at the yCR048W hotspot in sml1Δ rad50S ( H4898 ) and mrc1Δ rad9Δ sml1Δ rad50S ( H5776 ) cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 008 In the course of carrying out these experiments , we noticed that rad53Δ and rad53-kd cells exhibited lower DSB levels in untreated cells , specifically in the rad50S background ( Figures 3 and 4A , Figure 3—figure supplement 1C , D ) , suggesting that Rad53 may play a second role in promoting meiotic DSBs in this context . A DSB-promoting role for Rad53 is also supported by our observation that deletion of RAD53 reduced the DSB levels of mec1Δ cells ( Figure 3 ) . In this case , DSBs were formed in HU-treated mec1Δ rad53Δ cells , but the break levels were substantially lower than in mec1Δ cells alone ( Figures 3A , C and 4A ) , indicating Mec1 and Rad53 regulate DSBs using separate pathways . The rad50S mutation leads to the formation of blunt-ended DSBs that are detected by a Tel1 and Rad53-dependent checkpoint ( Xu et al . , 1997; Usui et al . , 2001 ) . We found that removal of Tel1 also lowered DSB levels in the absence of HU at the yCR048W hotspot ( Figure 4A , Figure 4—figure supplement 1 ) and on whole chromosomes analyzed by pulsed-field gel analysis ( Figure 4B ) , consistent with a recent report ( Argunhan et al . , 2013 ) . Similar to the rad53Δ mutation , deletion of TEL1 also reduced DSBs in a mec1–1 background , while still allowing for replication checkpoint bypass in the presence of HU ( Figure 4—figure supplement 1 ) . Intriguingly , DSBs levels were also lowered in dbf4-NLSΔN221 cells ( Figure 4A ) , but not dbf4-m25 mutants ( Figure 2—figure supplement 1D ) , suggesting that Rad53 might promote DDK activity in a phosphorylation-independent manner . Thus , Tel1 and Rad53 are required to achieve maximal DSB levels in rad50S cells , whereas Mec1 inhibits DSBs in response to delayed replication using both Rad53-dependent and -independent mechanisms . These data extend recent findings that multiple checkpoint pathways modulate DSB formation in budding yeast ( Zhang et al . , 2011; Argunhan et al . , 2013; Carballo et al . , 2013; Gray et al . , 2013 ) . 10 . 7554/eLife . 00844 . 009Figure 4 . A Tel1-dependent feedback mechanism increases DSB levels in rad50S cells . ( A ) The maximum levels of DSBs in untreated rad50S strains containing the indicated mutations were measured from the Southern blots shown in Figure 2 , Figure 3 , Figure 2—figure supplement 1 , Figure 3—figure supplement 1 , and Figure 4—figure supplement 1 . The amount of broken DNA was calculated as in Figure 1 and all values were normalized to the wild-type strain from the same experiment . ( B ) CHEF gel electrophoresis and Southern blotting was conducted to assess DSB levels on whole chromosomes in sml1Δ rad50S ( H4898 ) , rad53Δ sml1Δ rad50S ( H4969 ) , and tel1Δ sml1Δ rad50S ( H4849 ) cells . Chromosomes 8 ( left panel ) and 16 ( right panel ) were resolved on separate gels , blotted and visualized with the following probes in SGD coordinates: Chromosome VIII: 23 , 771-25 , 410 and Chromosome XVI: 20 , 281-21 , 012 . See also Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 00910 . 7554/eLife . 00844 . 010Figure 4—figure supplement 1 . TEL1 is required for wild-type DSB levels in rad50S cells . Southern blot analysis of DSBs at the yCR048W hotspot in sml1Δ rad50S ( H4850 ) , mec1-1 sml1Δ rad50S ( H4851 ) , tel1Δ sml1Δ rad50S ( H4849 ) and mec1-1 tel1Δ sml1Δ rad50S ( H4853 ) cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 010 To address the mechanism of action of the pre-meiotic replication checkpoint , we analyzed DSB factor abundance and modifications in the absence or presence of HU , as DSB factor accumulation is limited by the checkpoint in Schizosaccharomyces pombe ( Ogino and Masai , 2006 ) . To ensure that the modifications we observed were the result of the replication checkpoint and not of DSBs themselves ( which also activate Mec1 ) , the analysis was carried out in spo11-Y135F ndt80Δ mutants that arrest in meiotic prophase without DSBs . The protein levels of most DSB factors were unchanged in the absence or presence of HU ( Figure 5—figure supplement 1A ) , although we found that Mer2 accumulated to high levels in HU-treated cells ( Figure 2A and legend , Figure 2—figure supplement 1A and Figure 5—figure supplement 1A ) . The exceptional protein in this analysis was Spo11 , whose levels were approximately 10-fold lower in HU-treated cells than untreated cells ( Figure 5A ) . This result was confirmed in wild-type cells ( Figure 5—figure supplement 1B ) , indicating it was not the result of the spo11-Y135F allele . Spo11 levels were partially restored in mec1Δ and cdc6-mn mutants , but not rad53Δ cells ( Figure 5A ) , indicating that downregulation of Spo11 protein is a result of Rad53-independent replication checkpoint activity . Northern blot analysis indicated that checkpoint-dependent downregulation of Spo11 occurs at the level of the SPO11 transcript ( Figure 5B , C , Figure 5—figure supplement 1C ) . SPO11 RNA accumulation halted 1 hr after meiotic induction ( the time of replication onset ) , ultimately leading to a reduction in steady-state RNA levels of at least 5-fold ( Figure 5C , Figure 5—figure supplement 1C ) . We note that SPO11 RNA and protein were less abundant in rad53Δ cells than wild type , which may contribute the decreased DSBs levels in this strain background . We conclude that SPO11 expression is under checkpoint control . 10 . 7554/eLife . 00844 . 011Figure 5 . The replication checkpoint regulates DSB factor levels and DNA loading . ( A–C ) Analysis of spo11-Y135F-HA in sml1Δ ndt80Δ ( H5233 ) , cdc6-mn sml1Δ ndt80Δ ( H7447 ) , mec1-1 sml1Δ ndt80Δ ( H5227 ) , rad53Δ sml1Δ ndt80Δ ( H5230 ) strains in the presence or absence of HU . ( A ) Western blot analysis of spo11-Y135F-HA levels . A twofold dilution series of the 5h ( –HU ) time point was used to estimate changes in protein levels in the presence of HU . The cross-reacting band marked with red asterisks serves as a loading control . ( B ) Northern blot analysis of spo11-Y135F-HA RNA . ( C ) Quantification of the Northern blots in ( B ) . Northern blots were reprobed for UBC6 ( Teste et al . , 2009 ) for normalization . RNA levels of the spo11-Y135F-HA sml1Δ ndt80Δ ‘wild-type’ strain are shown as grey lines in all panels for comparison . ( D ) Binding profiles for Mer2-13myc ( H4585 ) from genome-wide location analysis ( ChIP-chip ) along Chromosome 8 in the absence ( blue lines ) or presence ( red lines ) of HU . ( E ) As in ( D ) , binding profiles from ChIP-chip analysis for Rec114-13myc ( H4890 ) and Mre11-13myc ( H5547 ) in wild-type cells in the absence ( blue lines , Vader et al . , 2011 ) or presence ( red lines ) of HU . ( F ) ChIP-chip binding profiles of and Rec114-13myc in mec1Δ sml1Δ ( H7305 ) and rad53Δ sml1Δ ( H7302 ) cells and for Mre11-13myc in mec1Δ sml1Δ ( H7323 ) and rad53Δ sml1Δ ( H7320 ) cells in the absence ( blue lines ) or presence ( red lines ) of HU . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 01110 . 7554/eLife . 00844 . 012Figure 5—figure supplement 1 . The meiotic replication checkpoint regulates DSB factor abundance , phosphorylation and DNA binding . ( A ) Western blot analysis in the absence or presence of HU in ndt80Δ spo11-Y135F strains for: Mre11-13myc ( H5085 ) , Rad50-6HA ( H7312 ) , Xrs2-13myc ( H5098 ) , Rec114-13myc ( H5092 ) , Mei4-13myc ( H5095 ) , Ski8-13myc ( H5154 ) , Rec102-13myc ( H5249 ) , spo11-Y135F-HA ( H5082 ) , Mer2-5myc ( H5079 ) , Rec104-13myc ( H5088 ) , and Sae2-13myc ( H5082 ) . Rec8-3HA ( H4695 ) cells are not ndt80Δ spo11Y135F . Red asterisks indicate the presence of bands that are absent in the HU-treated cells . ( B ) Western blot analysis in the absence or presence of HU for Spo11-18myc ( H2087 ) in otherwise wild-type cells , therefore not ndt80Δ spo11Y135F . A twofold dilution series of the 5h ( –HU ) time point was used to estimate changes in protein levels in the presence of HU . Fpr3 serves as loading control . ( C ) Time course of SPO11 transcripts accumulation in Spo11-18myc cells ( H2087 ) in the presence or absence of HU as determined by Northern blotting . SPO11 signals were normalized to FPR4 transcript levels . Shown are the mean and standard deviations of four independent Northern assays . ( D ) ChIP-chip binding profiles for Mre11-13myc in mec1Δ sml1Δ ( H7323 ) cells in the absence ( blue ) or presence ( red ) of HU are shown for Chromosome 8 . The sites of pre-meiotic replication origins are indicated by vertical dashed grey lines . ( E ) The distribution of values of either all points or replication origins ( as indicated ) were plotted for the Mre11-13myc binding profiles in mec1Δ sml1Δ ( H7323 ) cells in the absence ( blue ) or presence ( red ) of HU . The p values ( Student's t test ) of the differences between the distributions are shown above the plot . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 012 Our Western blot analysis also revealed that several DSB factors exhibited altered electrophoretic mobility upon HU treatment , indicative of a possible altered phosphorylation state . These changes are unlikely to be a consequence of direct phosphorylation by Mec1 , as no protein showed the expected reduced mobility in HU-treated cells ( Figure 5—figure supplement 1A ) . However , we noted that several protein bands showed faster mobility upon HU treatment , including Mer2 , Rec104 ( Figure 5B ) , Sae2/CtIP and Rec8 ( Figure 5—figure supplement 1A ) . As Mer2 and Rec8 are both reported to be DDK targets , it is possible that the increased mobility of these proteins in the presence of HU is the result of the decreased DDK activity we observed . Sae2 shows checkpoint- and cell-cycle dependent phosphorylation ( Huertas et al . , 2008; Manfrini et al . , 2010 ) and Rec104 has previously been described as a meiotic phospho-protein ( Kee et al . , 2004 ) . These data raise the intriguing possibility that multiple DSB and repair factors are targets of DDK-dependent phosphorylation , including Rec104 and Sae2 . To test whether the meiotic replication checkpoint affected the localization of DSB factors , we analyzed their DNA binding by genome-wide location analysis in the absence and presence of HU . Mer2-5myc was similarly detected on the same core meiotic chromosome binding sites occupied by axial proteins ( Panizza et al . , 2011 ) in the absence or presence of HU ( Figure 5D ) , in spite of the fact that it is not fully phosphorylated in HU-treated cells ( Figure 2A , Figure 5—figure supplement 1A ) . In contrast , we were unable to detect chromosomal loading of Rec114-13myc and Mre11-13myc in HU-treated cells , although both proteins associated with the DNA robustly in the absence of HU ( Figure 5E ) . These data suggest that the loading of specific DSB factors is prevented by the replication checkpoint . To understand how the replication checkpoint prevents Rec114 and Mre11 DNA loading , we monitored their chromosomal association in mec1Δ and rad53Δ cells . We found that deletion of MEC1 restored Rec114-13myc loading in the presence of HU whereas deletion of RAD53 did not ( Figure 5F , left panels ) , indicating that inhibition of Rec114 chromosome loading is Mec1-specific . Given that Rec114 forms a complex with Mer2 ( Arora et al . , 2004 ) , it is possible that Rec114 DNA loading depends on Mer2 phosphorylation . However , we do not believe this to be case , as Mer2 was equally phosphorylated in both mec1Δ and rad53Δ cells ( Figure 2A ) , yet Rec114 loading was specifically regulated by Mec1 . Furthermore , Rec114-13myc could not be detected on chromosomes in HU-treated cells containing the mer2-DDD allele that mimics the DDK-dependent phosphorylations ( Wan et al . , 2008 ) ( data not shown ) , consistent with the idea that Mer2 phosphorylation is not sufficient to recruit Rec114 to the DNA when Mec1 is activated . Similar to our results for Rec114 , we found that chromosome loading of Mre11 was detectible in HU-treated mec1Δ , but not rad53Δ cells ( Figure 5F , right panels ) . However , the pattern of Mre11 DNA-binding was altered dramatically in checkpoint mutant cells ( mec1Δ and rad53Δ ) treated with HU; Mre11 association with core meiotic chromosomal binding sites was substantially lower than in untreated cells , and the protein instead accumulated close to every pre-meiotic replication origin ( Blitzblau et al . , 2012 ) ( Figure 5—figure supplement 1D , dashed grey lines ) . Indeed , Mre11 was significantly enriched specifically at replication origins in mec1Δ cells treated with HU ( Figure 5—figure supplement 1E , p values from Student’s t test ) . We believe this to be the result of Mre11 recruitment to the aberrant DNA damage structures that form at replication forks when the replication checkpoint is impaired ( Feng et al . , 2006 ) . This extensive relocalization of Mre11 may explain why DSB levels in mec1Δ cells are lower in the presence of HU than without treatment ( Figure 3 ) . The specific inhibition of Rec114 and Mre11 loading by Mec1 reveals an additional Rad53-independent replication checkpoint pathway . The presence of redundant mechanisms to inhibit meiotic DSB factors in response to DNA replication suggests that there are negative consequences of DSBs on replicating chromosomes . We were unable to test this hypothesis in the mec1Δ strain , as these cells die due to replication fork problems upon HU treatment ( Figure 1—figure supplement 1A ) . Therefore , we employed the cdc6-mn strain , which experiences high levels of meiotic DSBs on chromosomes that have undergone very little replication ( schematic Figure 6A ) . Although Cdc6 protein levels are strongly depleted in cdc6-mn mutants ( data not shown ) , low levels of DNA replication were observed in this strain ( Figure 6B , shoulder to the right of 2C DNA ) . To rule out the possibility that this was break-induced replication ( BIR ) , we depleted DSBs in the cdc6-mn strain using a hypomorphic SPO11 allele . Remarkably , rather than observing a reduction in DNA replication , as would be expected if the shoulder in the FACS profile was due to BIR , we observed a substantial increase in DNA replication ( Figure 6B ) . This suggests that DSBs in the cdc6-mn strain inhibit DNA replication , possibly by directly blocking replication forks and/or activating a DNA damage checkpoint . 10 . 7554/eLife . 00844 . 013Figure 6 . DSBs on replicating chromosomes are lethal . ( A ) Schematic of a DSB that occurs ahead of a DNA replication fork . ( B ) FACS analysis of total DNA content in cdc6-mn ( H2655 ) and cdc6mn spo11-HA3-His6/spo11-Y135F ( H3598 ) cells as they progress through meiosis . ( C ) Southern blot analysis of DSB formation and repair at the HIS4-LEU2 hotspot in wild-type ( H2636 ) and cdc6-mn ( H2655 ) cells . Cells were treated with psoralen and DNA was crosslinked with UV light to preserve recombination intermediates . The relative positions of parental bands , repair intermediates and recombinants are marked . The red asterisk marks the position of an alternative recombination product . ( D ) Schematic of the experiment shown in ( E and F ) . Cells from ndt80Δ ( H385 ) , cdc6-mn ndt80Δ ( H386 ) and cdc6-mn spo11Δ ndt80Δ ( H3682 ) pre-sporulation cultures were split into YPD or SPO to induce mitosis or meiosis , respectively . After 5 hr , half of the SPO culture was returned to growth in YPD . ( E ) FACS analysis of total DNA content was performed for the experiment described in ( D ) . ( F ) Viability of cells was measured for the strains described in ( D ) at the indicated time points during meiotic induction . The number of colonies at each time point was normalized to the 1-hr time point for each culture . ( G ) Viability of wild-type ( H7099 ) , dbf4-NLS-ΔN221/DBF4 ( 7401 ) , ndt80Δ ( H7494 ) , dbf4-NLS-ΔN221/DBF4 ndt80Δ ( H7493 ) , ndt80Δ spo11-Y135F-HA ( H7468 ) and dbf4-NLS-ΔN221/DBF4 ndt80Δ spo11-Y135F-HA ( H7469 ) strains induced to enter meiosis in the presence or absence of HU and transferred onto YPD medium at the indicated time points . Each point is the average of two or three independent experiments . Viabilities were normalized to the 1-hr time point for each culture . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 01310 . 7554/eLife . 00844 . 014Figure 6—figure supplement 1 . Characterization of DNA replication and SC formation in cdc6-mn cells . ( A ) Relative copy number of mitochondrial DNA to single-copy chromosomal DNA in wild-type ( H2775 ) and cdc6-mn ( H2776 ) cells in the presence or absence of HU calculated by Southern blot analysis of HaeIII-digested genomic DNA probed for COX2 for the mitochondrial DNA and CEN15 for single-copy DNA . ( B ) Cells containing a single GFP-marked chromosome using TetR-GFP and a TetO array integrated on one homolog at TELV , CENV or LYS2 , respectively , for wild-type ( H3758 , H3755 , H3805 , blue bars ) , cdc6-mn ( H3756 , H3753 , H3803 , orange bars ) and cdc6-mn spo11Δ ( H3757 , H3754 , H3804 , green bars ) were analyzed after 24-hour incubation in SPO . The number of GFP dots in each tetranucleate cell was counted as a measure of the ability of the cell to completely replicate and segregate the given chromosome . The average number of tetranucleates produced by each strain is indicated next to the key . ( C ) Comparative genome hybridization of total genomic DNA from wild-type ( H2636 , blue dots ) and cdc6-mn ( H2655 , orange dots ) cells vs a G1 DNA control ( H1785 ) after 8 hrs in SPO . ( D ) Indirect immunofluorescence of Rad51 to mark DSBs and Zip1 to mark synaptonemal complex ( SC ) formation on spread nuclei from wild-type ( H2636 ) and cdc6-mn ( H2655 ) cells . Representative nuclear spreads at the 5-hr time point are shown . ( E ) Quantification of the number of cells showing full SC formation at the indicated times after inoculation into SPO for the strains shown in ( D ) . The defect in SC formation is relatively mild , which is probably why it was not observed in a previous study ( Brar et al . , 2009 ) . 200 cells were counted for each condition . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 014 We investigated the source of the DNA replication we observed in cdc6-mn cells . The DNA content increase was not due to mitochondrial DNA replication ( Figure 6—figure supplement 1A ) , but rather because whole chromosomes were sporadically replicated and segregated in the cdc6-mn strain , as revealed by the occasional duplication and segregation of a single chromosome marked with a TetO array/TetR-GFP ( Figure 6—figure supplement 1B ) . Although we were unable to detect the loading of Mcm2-7 helicase in cdc6-mn cells by genome-wide location analysis ( Figure 1—figure supplement 1B ) , we believe that the SCC1 promoter driving CDC6 expression in the cdc6-mn strain allows for leaky expression that permits an undetectable amount of Cdc6 to act stochastically at replication origins throughout the genome . We suggest that the infrequency of these events precludes detection by population-based assays . Consistent with this idea , when we measured DNA replication in the cdc6-mn cells after 8 hr in sporulation medium compared to G1 unreplicated DNA , no specific chromosomes or regions of the genome were preferentially replicated except the ribosomal DNA ( rDNA ) ( Figure 6—figure supplement 1C ) . Because the rDNA contains ∼100 tandem 9 . 1 kb repeats , each with its own potential origin of replication , the chance that one of these repeats would load Mcm2-7 and initiate DNA replication is extremely high , compared to the 11–46 origins present in the single-copy regions of each chromosome . Therefore , we conclude that cdc6-mn cells undergo sporadic chromosomal replication initiation , although clearly the levels of replication are too low to promote a checkpoint response in the critical period , during which DSBs must be prevented ( Figure 1H ) . To understand the consequences of forming DSBs on replicating chromosomes , we analyzed DSB repair in cdc6-mn cells at the well-characterized HIS4-LEU2 locus on chromosome 3 ( Hunter and Kleckner , 2001 ) . Engineered restriction sites at this locus permit the measurement of DSB repair by Southern blot analysis . In wild-type cells , DSB formation is followed by the accumulation of double-Holliday junction repair intermediates . These intermediates are subsequently resolved into crossover products that show an altered molecular weight from the parental fragments ( Figure 6C ) . DSBs formed at wild-type levels at HIS4-LEU2 in cdc6-mn cells , and a subset of DSBs were converted into crossovers , indicating that cdc6-mn cells are able to undergo homologous recombination . However , we also observed many DSBs that accumulated and migrated faster in the gel at 8–10 hr , indicating that repair of these breaks is defective and they become hyperresected ( Figure 6C ) . Similarly , cdc6-mn cells exhibited a small defect in synaptonemal complex formation between homologous chromosomes ( Figure 6—figure supplement 1D , E ) , likely a result of the repair defects we observed . Thus , a subset of DSBs persist in cdc6-mn mutants , indicating that cells that form DSBs on partially replicated chromosomes are unable to complete DSB repair . Given that cdc6-mn cells have problems completing both DNA replication and DSB repair , we asked whether these defects affected later meiotic events or cell viability . Spindle and DAPI analysis revealed that despite ongoing DNA replication and unrepaired DSBs , cdc6-mn cells entered into the meiotic divisions with little delay from the wild-type cells ( Hochwagen et al . , 2005 ) , but exhibited strongly reduced levels of tetranucleate formation ( Figure 6—figure supplement 1C ) and no viable spores were produced ( data not shown ) . This result suggests that DSBs formed during meiS are catastrophic for meiotic cells , which are unable to complete DNA replication or restrain the nuclear divisions . We employed the return-to-growth protocol to determine the contribution of precocious DSB formation to cell lethality . DSB repair in meiotic cells is constrained to promote homologous recombination . Some DSBs , for example in dmc1Δ cells , cannot be repaired during meiosis , but can be repaired if cells are returned to mitotic growth in rich medium ( Shinohara et al . , 1997; Zenvirth et al . , 1997 ) . We analyzed ndt80Δ , cdc6-mn ndt80Δ and cdc6-mn spo11Δ ndt80Δ strains that were blocked in meiotic prophase to prevent meiotic chromosome segregation ( experimental outline in Figure 6D ) , and measured DNA replication ( Figure 6E ) and viability ( Figure 6F ) . Cells returned to rich medium prior to meiotic entry ( 0 hr ) , should express CDC6 normally from the SCC1 promoter , and , accordingly , we observed no defect in DNA replication or cell division ( Figure 6E , middle three panels ) . When cells remained in sporulation medium , little DNA replication occurred in cdc6-mn ndt80Δ cells , but much more was observed in cdc6-mn spo11Δ ndt80Δ cells by 30 hr ( Figure 6E , left three panels ) , showing that DSBs substantially impeded the completion of DNA replication in the cdc6-mn background . When cells were returned to growth in rich medium after 5 hr in sporulation medium , wild-type cells had already completed DNA replication ( Figure 6E , first right panel ) , and returned to cycling with 100% viability ( Figure 6F ) . In contrast , cdc6-mn cells were able to complete very little DNA replication upon return to growth ( Figure 6E , second to last panel ) and lost viability quickly after exposure to sporulation medium ( Figure 6F ) . Removal of Spo11 allowed more DNA replication upon return to growth ( Figure 6E , compare last two panels ) and the spo11Δ strain exhibited significantly increased viability over the SPO11 strain . We confirmed that replication-checkpoint bypass similarly results in HU- and Spo11-dependent lethality in the dbf4-NLSΔN221 strain background in unblocked meiosis ( top panel ) , as well as in prophase-arrested cells ( ndt80Δ ) ( Figure 6G ) . Together , these results reveal that preventing DSB formation on replicating chromosomes is essential to inhibit a lethal meiotic chromosome fragmentation event that significantly impedes both the completion of DNA replication , and the repair of meiotic DSBs .
In this study we have defined the core components of a pre-meiotic replication checkpoint in budding yeast . Meiotic cells detect replication stress using the canonical replication checkpoint machinery , which then regulates meiosis-specific processes . As during mitotic S phase , detection of HU-induced replication inhibition relies on Mec1 and Ddc2 . Mec1 activates the Rad53 effector kinase , which , in turn , inhibits DDK activity . The essential roles of Mec1 and Rad53 in maintaining replication fork stability appear to be equally important in meiotic cells , as removing either protein is lethal when cells are exposed to HU in meiS . Similar to the mitotic replication checkpoint , redundant and separable Mec1- and Rad53-dependent mechanisms maintain genome stability and prevent precocious cell-cycle progression in the presence of replication inhibition during meiosis ( Labib and De Piccoli , 2011 ) . However , the targets of the checkpoint differ . We found that the pre-meiotic replication checkpoint prevents accumulation , DNA-loading and phosphorylation of DSB factors ( Figure 7 , model ) . First , Mec1 downregulates SPO11 transcript levels . Although Spo11 production is not completely prevented , we believe that downregulation of Spo11 has functional consequences because genetically decreasing Spo11 activity was able to significantly rescue DNA replication in the cdc6-mn strain ( Figure 6B ) . Second , Mec1 activity strongly reduces the DNA loading of Rec114 and Mre11 . Rec114 is phosphorylated in a Mec1-dependent manner in response to meiotic DSBs ( Carballo et al . , 2013 ) . However , such phosphorylation was found to increase Rec114 chromosomal association and we found no evidence that Rec114 or Mre11 are direct targets of Mec1 during meiS . The lack of DSB factor loading could be due to impaired axis formation , in spite of the fact that Rec8 , Hop1 and Red1 load robustly onto the DNA in HU-treated cells ( Blitzblau et al . , 2012 ) . Alternatively , the DSB factors may be directly regulated by the replication checkpoint . Third , Rad53-dependent inhibition of DDK prevents Mer2 phosphorylation , which is expected to block meiotic DSBs ( Sasanuma et al . , 2008; Wan et al . , 2008 ) . The redundancy resulting from the separable inhibition of all of the major DSB complexes likely increases the speed and robustness of the checkpoint response , which is important given the severe repair defects and lethality associated with precocious DSB formation on unreplicated DNA . 10 . 7554/eLife . 00844 . 015Figure 7 . Model for the pre-meiotic replication checkpoint in budding yeast . A schematic for the assembly of the four budding yeast DSB factor complexes ( as defined below ) in the absence ( left ) and presence ( right ) of replication inhibition is shown . In the absence of inhibition , all factors load onto the DNA and Mer2 and Rec104 are fully phosphorylated , allowing Spo11 to introduce DSBs . In the presence of HU , the levels of SPO11 transcripts are reduced , the DNA loading of Mre11 and Rec114 is prevented and the phosphorylation of Mer2 and Rec104 is abrogated . Illustration by Tom DiCesare ( Whitehead Institute ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00844 . 015 The role of Dbf4 in the replication checkpoint remains elusive . Studies in pre-mitotic cells indicate that DDK activity is regulated by Rad53-dependent phosphorylation and interaction ( Zegerman and Diffley , 2010; Chen et al . , 2013 ) , but our results separate the phosphorylation status and activity of the protein from the ability to form DSBs in HU-treated cells . HU treatment resulted in Mec1- and Rad53-dependent Dbf4 phosphorylation and a block of DDK target phosphorylation . However , neither removal of Rad53 ( which prevented Dbf4 hyperphosphorylation and allowed for Mer2 phosphorylation ) nor the dbf4-m25 non-phosphorylatable protein allowed DSBs in the presence of HU , indicating that simply allowing DDK activation does not bypass the pre-meiotic replication checkpoint . Furthermore , the dbf4-Δ71-221 protein that cannot interact with Rad53 ( Chen et al . , 2013 ) also did not allow for checkpoint bypass . The only form of Dbf4 that allowed for efficient DSB formation in HU-treated cells was dbf4-NLSΔN221 , which appears to be an unregulated version of the protein that is dominant to the wild-type . The fact that the dbf4-NLSΔN221 strain is not sensitive to HU in mitotic cells ( Gabrielse et al . , 2006 ) suggests that this allele does not bypass checkpoint activation , but rather acts downstream or in parallel to Mec1 and Rad53 . It is possible that dbf4-NLSΔN221 is hyperactive compared to the wild-type protein and overcomes the checkpoint , or that it accumulates DDK activity early and ‘licenses’ DSBs before the replication checkpoint is activated . Regardless of the mechanism , our results clearly indicate that controlling DDK activity is a critical step in regulating DSB formation . It is intriguing that there is so little Rad53-dependent checkpoint activation in cdc6-mn cells , in spite of the observation that they can initiate significant DNA replication , as evidenced in spo11 mutant cells . One possibility is that DSB formation in cdc6-mn cells triggers the activation of the recombination checkpoint , in which Mec1 specifically activates the Mek1 kinase and not Rad53 ( Usui et al . , 2001; Cartagena-Lirola et al . , 2008 ) to direct DSB repair activity to the homologous chromosomes ( Carballo et al . , 2008; Niu et al . , 2009 ) . Therefore , Rad53 activity may be specifically suppressed during prophase by the DSBs in cdc6-mn cells . Rad53 can be activated later in meiosis ( Cartagena-Lirola et al . , 2008 ) , and consistently , we noted that Rad53 became activated at later time points in cdc6-mn cells treated with HU ( Figure 1H ) , which proceed into the meiotic divisions with replication forks and DSBs . An alternative and non-exclusive model is that no replication checkpoint is activated in cdc6-mn cells either because Cdc6 itself is a checkpoint factor , as has been suggested for S . pombe ( Hermand and Nurse , 2007 ) , or the number of replication forks is too low to promote checkpoint activation . In either case , the observed restriction of Rad53 activity could allow cells to treat DNA lesions differently depending on the type of damage and phase of the cell cycle . The pre-meiotic replication checkpoint delays DSB formation and cell cycle progression during impaired DNA replication , but it is not the only mechanism to preserve cell cycle order . Measuring the kinetics of DNA replication and DSB formation in both unchallenged and HU-treated cells indicated that DSB formation occurs with a fixed delay with respect to the appearance of 4C ( replicated ) DNA , consistent with previous observations ( Padmore et al . , 1991 ) . Therefore , it appears that under normal circumstances , DSBs occur only after bulk DNA replication . Furthermore , DSB formation seems to be timed independently from DNA replication , as it did not occur significantly earlier in cdc6-mn replication-depleted or checkpoint mutants cells , suggesting the order and timing of the two processes are under independent cell cycle controls . A second mechanism to directly couple DNA replication and DSB formation has been described; delaying DNA replication on the left arm of chromosome 3 similarly delays local DSB formation ( Borde et al . , 2000 ) . However , we do not believe this coordination is the same as the checkpoint described here , as delaying DNA replication with low concentrations of HU was sufficient to block DSB formation until 4C DNA appeared , that is when meiS was completed . The local delay of DSB formation with late forks is likely necessary , because we have shown here that the replication checkpoint is insensitive to very low levels of DNA replication , such as in the cdc6-mn mutant . Therefore , any genomic regions with late forks that do not complete replication inside of the normal S phase would be subject to this second coupling mechanism to ensure DSBs do not occur prior to replication completion . The severe phenotype exhibited by cells that make breaks on replicating chromosomes could explain the existence of multiple coordinating mechanisms for meiS and DSB formation . A pre-meiotic replication checkpoint that prevents DSBs in response to replication inhibition is conserved in the distantly related fission yeast S . pombe , where Rad3/ATR/Mec1 and Cds1/CHK2/Rad53 are similarly activated in response to HU treatment . Some mechanistic differences exist , which could be due to the lack of identity of the DSB factors themselves . First , unlike budding yeast , fission yeast lacking the Rad53 homolog Cds1 form DSBs in the presence of HU ( Tonami et al . , 2005; Ogino and Masai , 2006 ) . Second , although specific DSB factors are transcriptionally downregulated in S . pombe ( Ogino and Masai , 2006; Miyoshi et al . , 2012 ) , their identity is not conserved in budding yeast and the S . pombe Spo11-homolog Rec12 was not affected . Other mechanistic parallels remain to be tested , in particular , the role of DDK in the pre-meiotic replication checkpoint , which we identified as a central player in coupling DNA replication with DSB formation in budding yeast . The S . pombe Cdc7 homolog Hsk1 is essential for DSB formation ( Ogino et al . , 2006 ) , and is regulated by the pre-mitotic replication checkpoint ( Snaith et al . , 2000 ) . The close temporal succession of pre-meiotic DNA replication and DSB formation is a universal feature of meiotic recombination , and our data demonstrate that the separation of these two processes is vital for maintaining genomic integrity . Given that both the DDK cell cycle kinase and the replication checkpoint are highly conserved in all eukaryotes , it seems likely that similar coupling mechanisms exist to protect the gametes in other species , including humans .
Strains used in this study are isogenic to SK1 and are listed in Supplementary file 1 . Gene disruptions and tagging were carried out using a PCR-based protocol ( Longtine et al . , 1998 ) . Synchronous meiosis was induced as previously described ( Blitzblau et al . , 2012 ) . For HU experiments , cells were inoculated into sporulation medium ( SPO ) containing 20 mM HU at 30°C , except in Figure 1 when 5 mM HU was used as indicated . The proportion of viable cells in the culture was measured at each indicated time point by removing and plating ∼500 cells on YPD plates and measuring the number of colonies that grew after 3 days at 30°C . The number of colonies present was normalized to the number observed at the first time point ( 0 or 1 hr after introduction into SPO ) . FACS analysis for total DNA content was performed as in ( Blitzblau et al . , 2012 ) . Clamped-homogeneous electric field ( CHEF ) gel electrophoresis and Southern blotting for small chromosomes ( including chromosome 8 ) were performed as described ( Blitzblau et al . , 2007 ) . Large chromosome CHEF analysis was carried out similarly , using a 1% gel for 15 hr with 60 s pulses followed by 9 hr at 90 s . For resolution of recombination intermediates , cells were killed with 0 . 1% sodium azide . They were resuspended in 0 . 1 mg/ml psoralen in TE ( 50 mM Tris pH 7 . 5 , 50 mM EDTA ) and crosslinked with 365-nm UV light for 12 min on a UV lightbox ( 5 mW/cm2 ) in a polystyrene culture dish . DNA was isolated via standard Southern protocol . All time course experiments for Southern analysis were repeated at least twice with similar results . Whole cell protein extracts were prepared by TCA precipitation as in Blitzblau et al . ( 2012 ) . An equal number of cells were loaded for each sample and equivalent loading was confirmed by Ponceau S staining . SDS-polyacrylamide gel electrophoresis and blotting were performed as described in Falk et al . ( 2010 ) . The following antibodies were used for detections , all diluted in PBS-T + 3% milk ( TBS-T was used for Mer2 phospho-S30 ) and incubated overnight; anti-Rad53 yC-19 ( Santa Cruz Biotechnology Inc , Santa Cruz , CA ) used at 1:5 , 000 dilution and Rad53 separated on an 8% gel; rabbit polyclonal anti-Dbf4 HM5765 ( Steven P Bell , Francis , et al . , 2009 ) was used at 1:1 , 000 dilution and Dbf4 separated on a 7 . 5% gel; anti-myc 9E10 ( Covance , Princeton , NJ ) was used at 1:1 , 000 for the following proteins: Mer2 , Rec104 , and Sae2 separated on 14% gels , Rec114 , Mei4 , and Rec102 separated on 12% gels , Spo11 , Mre11 , Xrs2 , and Ski8 separated on 10% gels; anti-HA 12CA5 ( Roche , Basel , Switzerland ) was used at a 1:1 , 000 dilution for Rad50 separated on a 7 . 5% gel and Rec8 separated on a 10% gel; anti-HA 3F10 ( Roche ) was used at 1:1 , 000 for spo11-Y135F-HA separated on a 12 . 5% gel; anti-Mer2 phospho-S30 ( Abcam , Cambridge , MA ) was used at a 1:1 , 000 dilution . The appropriate species secondary antibodies were diluted 1:5 , 000 in PBS-T + 3% milk ( except TBS-T was used for Mer2 phospho-S30 ) and incubated for 2 hr at room temperature . Northern blot analysis was performed as in Hochwagen et al . ( 2005 ) with minor modifications . 6 ml of cells were harvested at the indicated time points , washed once with 1 × TE ( 10 mM Tris pH 7 . 5 , 1 mM EDTA ) and frozen at −80°C . Cell pellets were ruptured by vigorous shaking for 30 min at 4°C in equal volumes glass beads , phenol-chloroform-isoamylalcohol ( 25:24:1 ) and cold RNA buffer 1 ( 300 mM NaCl , 10 mM Tris pH 7 . 5 , 1 mM EDTA , 0 . 2% SDS ) . RNA was precipitated in ethanol and resuspended at 65°C in RNA buffer 2 ( 10 mM Tris pH 7 . 5 , 1 mM EDTA , 0 . 2% SDS ) . 30 μg of each RNA sample were denatured at 65°C in denaturing solution ( 50% formamide , 6 . 5% formaldehyde , 40 mM MOPS pH 7 . 0 , 10 mM sodium acetate , 0 . 1 mM EDTA ) and separated by electrophoresis in a 1 . 1% agarose gel ( containing 6% formaldehyde ) in MOPS buffer ( 40 mM MOPS pH 7 . 0 , 10 mM sodium acetate , 0 . 1 mM EDTA ) . RNA was blotted onto a Zeta probe GT membrane ( BioRad ) in 10 × SSC ( 1 . 5 M NaCl , 150 mM sodium citrate , pH 7 . 0 ) and UV-crosslinked using a Stratalinker . Probe hybridization was performed as for Southern blots . 25 ml of cells were harvested after 3 hr in SPO . Chromatin immunoprecipitation ( ChIP ) for genome-wide location analysis was performed as described ( Blitzblau et al . , 2012 ) . One tenth of the lysate was removed as an input sample . Samples were immunoprecipitated for 16 hr at 4°C with anti-HA 3F10 ( Rec8-3HA , Roche , used 2 µg per immunoprecipitation ) , ChIP grade anti-myc 9E11 ( Abcam , 2 µl used per immunoprecipitation ) or UM185 ( Rabbit polyclonal anti-Mcm2-7 , Stephen P Bell , 2 µl serum used per immunoprecipitation ) . Total genomic DNA extraction for CGH analysis was performed as described in Blitzblau et al . ( 2012 ) . For ChIP experiments , one half of the immunoprecipitated DNA and one tenth of the input DNA were labeled . Samples were labeled and hybridized as in Blitzblau et al . ( 2012 ) . For each co-hybridization , Cy3 and Cy5 levels were calculated using Agilent Feature Extractor CGH software . Background normalization , log2 ratios for each experiment and scale normalizations across each set of biological replicates were calculated using the sma package ( Yang et al . , 2001 ) in R , a computer language and environment for statistical computing ( v2 . 1 . 0 , http://www . r-project . org ) . The value of the feature closest to each potential pre-meiotic replication origin ( Blitzblau et al . , 2012 ) was used to estimate Mre11 binding close to replication origins . The raw data and log ratios analyzed in this study are available from the NCBI Gene Expression Omnibus ( http://www . ncbi . nlm . nih . gov/geo/ ) , accession number GSE46841 . Meiotic nuclear spreads were performed as described ( Falk et al . , 2010 ) . In brief , the nuclei of spheroplasted cells were spread on a glass slide in the presence of paraformaldehyde fixative and 1% lipsol . After drying , the slides were blocked in blocking buffer ( 0 . 2% gelatin , 0 . 5% BSA in PBS ) and stained with anti-Rad51 y-180 ( Santa Cruz ) used at 1:200 dilution and anti-Zip1 yN-16 ( Santa Cruz ) used at 1:100 dilution . | Most cells in an organism contain two sets of chromosomes , one inherited from the mother and the other from the father . However , sexual reproduction relies on the production of gametes—eggs and sperm—which contain only one set of chromosomes . These are produced through a specialized form of cell division called meiosis . Meiosis begins with a cell replicating its entire genome . Maternal and paternal versions of each chromosome then pair up and swap sections of their DNA through a process known as homologous recombination . This gives rise to chromosomes with new combinations of maternal and paternal genes . Finally , the cell undergoes two successive rounds of division—the first to produce a cell with two nuclei containing two sets of chromosomes each , and the second to produce four gametes , each containing a single set of chromosomes . Homologous recombination requires the formation of double-strand breaks in the DNA , but it is essential that these do not form before DNA replication is complete . Now , Blitzblau and Hochwagen have used yeast , which is easy to maintain in the lab and to manipulate genetically , to reveal the molecular components of a checkpoint that controls this process . Blitzblau and Hochwagen first used an inhibitor called hydroxyurea to block DNA replication in yeast cells , and confirmed that this treatment also suppressed the formation of double-strand breaks . By selectively inhibiting the activity of individual proteins , it was shown that break formation was controlled by a checkpoint that relies on two conserved proteins , the checkpoint kinase Mec1 ( homologous to the human tumour suppressor ATR ) and the cell-division kinase DDK . Moreover , when double-strand breaks were allowed to form on partially replicated chromosomes , DNA replication stalled and meiosis could not proceed normally , with lethal results for the yeast . These results explain how DNA replication and recombination are coordinated during meiosis in yeast . Moreover , because the genes that control meiosis are highly conserved from yeast to humans , they have implications for research into human fertility . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cell",
"biology"
] | 2013 | ATR/Mec1 prevents lethal meiotic recombination initiation on partially replicated chromosomes in budding yeast |
Embryonic growth occurs predominately by an increase in cell number; little is known about growth mechanisms later in development when fibrous tissues account for the bulk of adult vertebrate mass . We present a model for fibrous tissue growth based on 3D-electron microscopy of mouse tendon . We show that the number of collagen fibrils increases during embryonic development and then remains constant during postnatal growth . Embryonic growth was explained predominately by increases in fibril number and length . Postnatal growth arose predominately from increases in fibril length and diameter . A helical crimp structure was established in embryogenesis , and persisted postnatally . The data support a model where the shape and size of tendon is determined by the number and position of embryonic fibroblasts . The collagen fibrils that these cells synthesise provide a template for postnatal growth by structure-based matrix expansion . The model has important implications for growth of other fibrous tissues and fibrosis .
Tendons transmit tensile forces from muscles to bone and are amongst the heaviest loaded tissues in vertebrates; the forces have been estimated at ∼ 16 kN/Kg of body mass ( Harrison et al . , 2010 ) . The ability of tendons to transmit such large forces is directly attributable to an extracellular matrix ( ECM ) comprising collagen fibrils aligned parallel to the tissue long axis . Such forces occur during extended periods of activity with potentially devastating consequences on tissue integrity and cell survival . However , despite the high forces , tendon cells ( tenocytes ) remain viable and are able to sustain the integrity of the tissue . Our motivation for this study was to understand how tendons increase in size from an anlage of condensed mesenchyme to a mature tissue that is predominately ECM and which can withstand large forces without cell death . The discovery of transcription factors expressed by tendon progenitors and the identification of signalling pathways in tendon has contributed greatly to our understanding of the earliest stages of tendon specification . However , the absence of a robust method of imaging cell-matrix organisation has precluded a detailed study of how the cell-rich anlage found in embryogenesis grows into a mechanically strong connective tissue in the adult . Advances in serial block face-scanning electron microscopy ( SBF-SEM ) provide a method to address this problem ( Starborg et al . , 2013 ) . SBF-SEM is a practical addition to serial section reconstruction and can typically provide data at ∼10 nm resolution in the sectioning plane for ∼100 , 000 µm3 volume of tissue , which is sufficient to quantify cell number , cell shape , and cell–cell interactions , as well as estimate the number and organisation of individual collagen fibrils . When used in combination with a developmental series , SBF-SEM provides new information on the changes in cell organization and matrix assembly during tissue growth . The discovery of Scleraxis ( Scx ) , amongst all other transcription factors , transformed studies of tendon development ( Brent et al . , 2003 ) . The generation of Scx-GFP mice showed the precise location of tendons in the proximal–distal axis of the limb ( Schweitzer et al . , 2001 ) . Scx specifies important interactions between somatic muscle and cartilage cell lineages leading to tendon development ( Brent et al . , 2003; Sugimoto et al . , 2013a ) . However , Scx is also expressed in non-tendon tissues including ligaments , intervertebral discs , joints , and cartilage around the chondro-tendinous/ligamentous junction ( Sugimoto et al . , 2013b ) . Tenomodulin , a type II transmembrane glycoprotein , is expressed in tendons and ligaments and is a regulator of tenocyte proliferation and is involved in collagen fibril maturation ( Docheva et al . , 2005 ) . Further studies have shown that Mohawk ( Liu et al . , 2010 ) , EGR1 , and EGR2 are transcription factors that are also involved in tendon formation ( Lejard et al . , 2011 ) . Furthermore , the ability of EGR1 to promote tendon differentiation is partially mediated by TGFβ2 ( Guerquin et al . , 2013 ) . Two main signalling pathways , TGFβ and FGF , have been identified in vertebrate tendon development ( Maeda et al . , 2011 , and for review see Tozer and Duprez , 2005; Schweitzer et al . , 2010 ) . Collagen type I occurs in virtually all fibrous tissues along with type III , V ( Birk and Mayne , 1997 ) , XII , XIV ( Ansorge et al . , 2009 ) , fibronectin and small proteoglycans such as decorin ( Berenson et al . , 1996; Zhang et al . , 2006 ) . Therefore , the expression of these transcription factors and structural proteins , and the presence of particular signalling pathways , does not , in itself , provide information on the development of tendon structure and function . The defining event that signifies the onset of functional tendon development is the appearance of collagen fibrils in the ECM . The distribution of collagen fibril diameters distinguishes two stages in tendon development . In mouse , stage 1 begins at E12 . 5 when narrow ( ∼35 nm ) diameter collagen fibrils are formed within actin-dependent fibripositors at the cell surface ( Canty et al . , 2004 , 2006; Kalson et al . , 2013 ) . The length and width of the tissue doubles in a few days , and is accompanied by two-orders of magnitude increases in elastic modulus and ultimate tensile strength ( McBride et al . , 1985 , 1988 ) . Serial section transmission electron microscopy ( ss-TEM ) of chick embryonic tendon showed that the tendon matrix contains bundles of collagen fibrils that undergo gradual rotation over several microns ( Birk and Trelstad , 1986; Birk et al . , 1989 ) . Moreover , the collagen fibril bundles are stabilised by cell–cell connections containing cadherin-11 ( Richardson et al . , 2007 ) . During stage 1 there is a modest increase in fibril diameter ( to ∼40 nm ) , which has been replicated in vitro by slow stretching of a 3-dimensional ( 3D ) tendon-like construct containing embryonic tenocytes ( Kalson et al . , 2011 ) . At birth ( in the mouse ) , the unimodal distribution of narrow fibrils is quickly ( within a few days ) replaced by a bimodal distribution of fibril diameters with a range of 35–400 nm ( Goh et al . , 2012 ) . The transition from unimodal to bimodal distributions specifies the onset of stage 2 . In addition to increased fibril diameter and changes to cell morphology , the collagen fibrils in tendon develop regular undulations , known as crimp ( Diamant et al . , 1972 ) . Crimp becomes evident early in embryonic development ( Shah et al . , 1982 ) and provides the biomechanically important ‘toe-region’ to the force–extension curve ( Shah et al . , 1982 ) . Crimp structure has been investigated by plane polarised light microscopy and by SEM ( Raspanti et al . , 2005; Franchi et al . , 2007 ) , but the structure of crimp is still debated; numerous studies describe a planar 2D undulating crimp , but mathematical modelling of crimped collagen fibrils in tendon suggests crimp to be helical in structure ( Grytz and Meschke , 2009 ) . Again , technical limitations of techniques used to study crimp have prevented a definitive description; for example , SEM requires chemical treatment of tissue to remove cellular material , precluding analysis of the composite 3D architecture . Although ss-TEM provided insights into the short-range hierarchical structure of tendon , technical challenges of generating undistorted serial sections and achieving an absolute alignment have been major hurdles in studying hierarchical organisation beyond the level of a few cells . We here describe SBF-SEM studies of tendon development and show 3D reconstructions at three key stages of tendon development: stage 1 ( embryonic ) , the stage 1–2 transition ( newborn ) , and stage 2 ( 6 week ) . We interpret our findings as suggesting that the hierarchical structure of tendon , with collagen fibrils organised into fibril bundles with a biomechanically important spiral crimp structure , and the number of collagen fibrils in bundles , is established during embryogenesis . Subsequent postnatal tendon growth is achieved by increase in collagen fibril diameter and fibril length , likely by interface-limited molecular accretion , during which the spatial relationship between cells , fibril bundles and the crimp spiral , is maintained .
To learn more about tendon development we examined embryonic day 15 . 5 ( E15 . 5 ) , newborn and 6 week postnatal mouse-tail tendon . First we investigated the collagen fibrils in the ECM , using TEM of ultra-thin sections cut perpendicularly to the tendon long axis , in which the collagen fibrils are transected transversely ( Starborg et al . , 2013 ) . The results showed that at E15 . 5 the vast majority of collagen fibrils had near-circular outlines and were therefore perpendicular ( to a close approximation ) to the tendon long axis . The fibrils were organised in small groups , which occurred in channels defined by the plasma membranes of surrounding cells . These groups of fibrils are termed ‘fibril bundles’ ( Figure 1A ) . At this stage of development the collagen fibrils are being synthesised in fibripositors at the cell-matrix interface ( Kalson et al . , 2013 ) . 10 . 7554/eLife . 05958 . 003Figure 1 . 2D analysis of mouse-tail tendon through development . ( A ) Transmission electron microscope ( TEM ) images of transverse sections at embryonic day 15 . 5 ( E15 . 5 ) . A closed arrowhead marks a fibripositor , seen during embryonic development but absent in postnatal tissue . Fibril bundles ( marked * ) are small . An open arrowhead marks a cell nucleus . ( B ) Newborn ( P0 ) . A cell–cell junction is circled between two cytoplasmic processes enclosing a fibril bundle . Bundles are larger and contain more fibrils than at E15 . 5 . ( C ) 6 weeks postnatal . Cells have elongated processes that divide the matrix into bundles and connect with other cells . Bundles are larger than in newborn tissue . Open arrowhead marks cell nuclei . Stars ( * ) mark fibril bundles . Scale bar = 1 µm . The areas marked with a square in the left image in A , B and C are magnified in the right image . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 003 At birth the number of fibril profiles in fibril bundles increases >fourfold , and the fibril bundles are well defined by cellular plasma membrane extensions ( Figure 1B ) . It is important to note that fibril profiles are fibrils in transverse section . It was not possible to count entire fibrils because some are too long to be contained within the SBF-SEM volumes we studied . TEM images showed electron dense cell–cell junctions between adjacent cell membrane extensions ( Waggett et al . , 2006; Richardson et al . , 2007 ) . In 6 weeks postnatal tendon , fibril bundles have grown in lateral size and are separated by striking elongated cell processes ( Figure 1C ) . Analysis of TEM images demonstrated that fibril diameter increased during development , with mean fibril diameter increasing from 35 . 4 nm ± 0 . 2 nm at E15 . 5 , to 46 . 5 ± 1 . 1 nm at birth and 160 . 1 ± 3 . 4 nm at 6 weeks postnatal ( Table 1 , Figure 2A , B , C ) . A bimodal distribution of diameters was found in 6 week postnatal tendon , as previously described ( Parry et al . , 1978 ) . Notably , the increase in fibril diameter occurred uniformly throughout the ECM , and was not confined to fibrils close to the cell . This simple , key observation is critical to understanding how collagen fibrils grow in diameter independent of distance from the source of newly synthesized collagen ( i . e . , the cell ) . 10 . 7554/eLife . 05958 . 004Table 1 . Summary of dataDOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 004StageMean fibril diameter ( nm ) Mean fibril area ( nm2 ) Mean number of fibrils per bundleMean fibril length ( µm ) FAF ( % ) Crimp length ( µm ) Crimp helix radius ( µm ) Tail length ( mm ) E15 . 535 . 4 ± 0 . 21006 . 5 ± 13 . 8156 ± 11125 ± 181 . 4 ± 0 . 3––6 . 1 ± 0 . 30Newborn46 . 5 ± 1 . 12137 . 9 ± 370 . 9659 ± 23578 ± 1717 . 6 ± 1 . 314 . 0 ± 0 . 31 . 6 ± 0 . 110 . 9 ± 0 . 96W160 . 1 ± 3 . 424 , 606 . 8 ± 933 . 0684 ± 221250* ± 30576 . 4 ± 1 . 399 . 3 ± 2 . 52 . 3 ± 0 . 167 . 6 ± 0 . 3*For fibrils of diameter <150 nm . Raw data are provided in Supplementary file 1 . 10 . 7554/eLife . 05958 . 005Figure 2 . Fibril diameter and fibril bundling during development . ( A ) Fibril diameter plot for embryonic day 15 . 5 ( A ) , newborn ( B ) and 6 week old tendon ( C ) . In embryonic development there is a gradual increase in fibril diameter , which has a unimodal size distribution . Postnatal tendon has larger fibrils with a bimodal distribution . ( D ) There is a major ( ∼fourfold ) increase in fibril profiles in transverse section per bundle from E15 . 5 to newborn tissue . This fibril number remains constant postnatal in mature tissue . ( E ) Fibril bundle number per cell nucleus is constant postnatal . ( F ) The number of fibrils per µm2 decreases as a function of increasing fibril diameter . ( G ) The area occupied by fibrils ( as a proportion of total tendon area ) increases significantly at each time point . ( H ) Mean fibril area increases significantly at each time point ( derived from fibril diameter data assuming circularity ) . A logarithmic scale has been used on the y-axis . ** Indicates significant difference ( p = <0 . 05 ) , † indicates p = >0 . 05 . Source data in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 00510 . 7554/eLife . 05958 . 006Figure 2—source data 1 . Fibril diameter and bundle data . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 006 Analysis of the bundles showed a pronounced increase in the number of fibril profiles per bundle during E15 . 5 to newborn , which is the period that we know de novo fibril assembly is occurring ( Table 1 ) ( Trelstad and Hayashi , 1979; Kalson et al . , 2013 ) . After birth the number of fibrils stabilised; approximately 700 fibrils were found in each fibril bundle at birth and also at 6 weeks postnatal ( Figure 2D , p = >0 . 1 , Data in Figure 2—source data 1 ) . At the same time as the number of fibril profiles per bundle had stabilised the number of fibril bundles per nucleus was also constant ( 7 . 4 ± 0 . 3 bundles per cell vs 7 . 3 ± 0 . 3 in newborn vs 6 week , p = >0 . 1 , Figure 2E ) . Fibripositors ( sites of fibril formation during stage 1 ) were absent from 6 week tail tendon . The number of fibril profiles per μm2 fell by a factor of ∼10 during development as the diameter of the fibrils increased ( Figure 2F ) and fibril area fraction ( FAF ) increased during development from 1 . 4 ± 0 . 3% at E15 . 5 , to 17 . 6 ± 1 . 3% at newborn , to 76 . 4 ± 1 . 3% at 6 weeks ( Figure 2G ) . Corresponding with increase in fibril diameter , the mean fibril area increased from 1006 . 5 ± 13 . 8 nm2 at E15 . 5 , to 2137 . 9 ± 370 . 9 nm2 at newborn to 24 , 606 . 8 ± 933 . 0 nm2 at 6 weeks ( Figure 2H ) . TEM provides detailed ultra-structural information in two dimensions but does not easily permit study of 3D architecture , which is critical to the biomechanical function of collagenous connective tissues . We therefore used SBF-SEM to extend our studies of tendon development ( Starborg et al . , 2013 ) . We studied mouse-tail tendon at the same developmental time points as Figure 1 . Representative reconstructions of SBF-SEM datasets are presented in Figure 3 . A transverse view ( perpendicular to the long axis of the tendon ) is shown in Figure 3A . The cells at E15 . 5 are rounded and close together . The cell bodies move apart as fibril bundles increase in size , eventually becoming star shaped ( in transverse section ) at 6 weeks postnatal . Longitudinal views ( B and C ) show the close relationship of cells , which are stacked on top of each other , thereby maintaining the precise longitudinal alignment of the fibril bundles . This end-to-end relationship of cells is maintained throughout development and is particularly striking in 6 week-old tendon . Reconstructing the fibril bundles in three dimensions ( in red in Figure 3C ) showed that in newborn and 6 week tendon the fibril bundles had a regular wavy configuration ( crimp ) . The fibril bundles reconstructed in red ( Figure 3C ) are shown in the corresponding SBF-SEM images ( Figure 3D ) . 10 . 7554/eLife . 05958 . 007Figure 3 . The 3D morphology of tendon changes from embryonic to postnatal tissue . 3D reconstructions from SBF-SEM images of E15 . 5 , newborn and 6 week mouse-tail tendon . Individual cells are reconstructed in a different colour . ( A ) A transverse view of reconstructed tendon cells . E15 . 5 is comprised predominantly of rounded cells , with relatively little collagenous ECM . During development the ECM grows rapidly; by 6 weeks postnatal there are large spaces between cells containing bundles of collagen fibrils . One bundle ( red ) is marked with a white arrowhead . ( B ) Longitudinal views demonstrate the development of ( 1 ) well-defined channels formed by cells stacked end-to-end , and ( 2 ) wavy orientation of tenocytes corresponding to crimped bundles of collagen fibrils , particularly notable in newborn tendon . ( C ) The same reconstructions as in B with a bundle of collagen fibrils highlighted ( red , marked with a white arrow ) . Longitudinally continuous bundles of fibrils are seen at E15 . 5 closely associated with tendon cell membranes . Crimp is clearly seen in newborn tendon and persists in 6 week tendon with a longer wavelength . ( D ) SBF-SEM images corresponding to C . The fibril bundle in C is circled red . Cell membranes are also outlined . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 007 The close relationship during embryogenesis between the cell membrane and fibril bundles led us to investigate 3D cell shape using SBF-SEM ( Figure 4 ) . SBF-SEM images at each of the time points studied are shown in the top row ( Figure 4A—E15 . 5 , Figure 4B—newborn and Figure 4C—6 week ) . One cell membrane is highlighted to demonstrate the membrane protrusions that form channels surrounding the fibril bundles . Longitudinal views demonstrate that the fibril bundle conforms to the shape of the cells; the cell and the fibril bundle remain in close apposition along the length of the cell . Contact between adjacent cell processes ( which form the boundaries of these channels ) is established during embryogenesis and is maintained during postnatal development of cell–cell contacts . The cell processes lengthen to accommodate the larger diameter collagen fibrils , resulting in a stellate ( star shaped ) appearance of the cell in transverse view in postnatal tendon ( Figure 4C ) . Cell body length decreased during development from 60 ± 2 μm at E15 . 5 , to 55 ± 3 μm at birth , to 27 ± 2 μm at 6 weeks ( Data in Figure 4—source data 1 ) . Calculation of cell dimensions demonstrated that cell volume remains relatively constant during this period of matrix growth between newborn and 6 weeks ( Figure 4D ) , but the cell surface area increases greatly due to the growth of these long processes ( Figure 4E ) . When isolated from tendon and plated on plastic the cell shape changed radically: in 2D cell culture the cells become flat and have the characteristic appearance of a fibroblast in culture ( Figure 4—figure supplement 1 ) . There was no difference in the surface area of the cells when cultured on plastic comparing E15 . 5 cells and 6 week cells ( Data in Figure 4—source data 1 ) . 10 . 7554/eLife . 05958 . 008Figure 4 . Changes in cell morphology during development . ( A ) E15 . 5 mouse-tail tendon . A cell membrane is highlighted in blue ( top panel , single SBF-SEM transverse image ) and reconstructed ( middle and bottom panels ) . A fibril bundle , circled in dark blue ( top ) , is reconstructed , and lies closely associated with the cell along the entire cell length . ( B ) Newborn mouse-tail tendon . A cell membrane is highlighted in yellow ( top panel , single SBF-SEM transverse image ) . A fibril bundle is highlighted in red and reconstructed in red , together with the corresponding cell ( in yellow , middle and bottom panels ) . The fibril bundle is again closely associated with the cell . Cell protrusions surrounding fibrils are better defined than at E15 . 5 . ( C ) 6 week old mouse-tail tendon . Two adjacent cell membranes are highlighted blue and pink ( top panel , single SBF-SEM transverse section image ) . The same cells are reconstructed below . A fibril bundle is marked with * . The two cells have elongated processes that reach out and contact each other ( highlighted with a rectangle , top reconstruction ) , enveloping a bundle of collagenous ECM ( * ) . ( D ) There is a small increase in cell volume between E15 . 5 and newborn time points , but no further increase in cell volume , which remains relatively constant during postnatal development . ( E ) Cell surface area increases markedly ( 1 . 5 fold and 1 . 8 fold for E15 . 5 to newborn and newborn to 6 week old , respectively ) as cells develop large cell membrane processes to enlarge the channels containing the fibril bundles . ** Indicates significant difference ( p = <0 . 05 ) , † indicates p = >0 . 05 . * Marks a fibril bundle . Source data in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 00810 . 7554/eLife . 05958 . 009Figure 4—source data 1 . Cell morphology data . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 00910 . 7554/eLife . 05958 . 010Figure 4—figure supplement 1 . Cell shape in 2D culture . Cells isolated from E15 . 5 ( A ) , and 6 week ( B ) mouse-tail tendon by trypsin digest plated on plastic have similar morphology . Cell surface area is not significantly different in 2D cell culture ( C ) . † indicates p = >0 . 05 . This is in contrast to the marked difference in cell surface area of these cells in vivo ( Figure 4E ) . Scale bar = 10 µm . Source data in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 010 To obtain a better understanding of the cell changes that underpin tendon growth we used SBF-SEM to analyse the change in cell number during embryonic and postnatal development ( Figure 5 , data in Figure 5—source data 1 ) . We identified cell nuclei in 3D reconstructions from SBF-SEM data ( Figure 5A , B ) . The number of cells per unit volume of tendon tissue fell significantly from 1 . 26 ± 0 . 02 to 0 . 91 ± 0 . 02 cells per 1000 µm3 between E15 . 5 and birth , to 0 . 14 ± 0 . 01 at 6 weeks ( p = <0 . 05 , Figure 5C ) . Measurement of cells per unit area of transverse tendon area showed a significant fall from 30 . 1 ± 0 . 6 cells per 1000 µm2 ( E . 15 . 5 ) to 20 . 5 ± 0 . 8 ( newborn ) to 1 . 9 ± 0 . 1 ( 6 week , p = <0 . 05 , Figure 5D ) . 10 . 7554/eLife . 05958 . 011Figure 5 . Cell number during development . ( A ) 3D reconstructions of longitudinal views of tendon tissue . Individual cell membranes are reconstructed in different colours . Inset images show a lower magnification view . Note the longitudinal-axis cell overlap in newborn tendon , which is not seen at 6 weeks . ( B ) Cell nuclei from reconstructions in ( A ) are marked with red spheres . ( C ) Calculation of cell number per unit volume of tissue . There is a significant decrease in cells per unit volume from E15 . 5 to newborn ( 1 . 26 ± 0 . 02 to 0 . 91 ± 0 . 02 cells per 1000 µm3 ) and from newborn to 6 weeks ( an almost sevenfold change , 0 . 91 ± 0 . 02 to 0 . 14 ± 0 . 01 cells per 1000 µm3 ) . ( D ) Calculation of cells per 1000 µm2 of transverse tissue area . The number of nuclei in transverse section was determined . There was a significant decrease in cells per 1000 µm2 from E15 . 5 to newborn and from newborn to 6 weeks . ( E ) The longitudinal distance between nuclei of cells in stacks decreases from newborn to 6 weeks ( E15 . 5 cells are not stacked , precluding this analysis ) . ( F ) The transverse distance between cell nuclei in adjacent cell stacks increases significantly between newborn and 6 weeks . * Indicates p = <0 . 05 . Source data in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 01110 . 7554/eLife . 05958 . 012Figure 5—source data 1 . Cell number data . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 012 Tenocytes in newborn and 6 week tendon are arranged in stacks , between which sit bundles of collagen fibrils . The observation that cell body length decreases during development led us to measure the distance between cell nuclei in stacks . At birth cell nuclei were spaced 39 . 9 ± 1 . 8 µm apart ( on the longitudinal axis ) and at 6-weeks they were 27 . 5 ± 1 . 1 µm apart ( p = <0 . 05 , Figure 5E ) . The discrepancy between cell length and distance between nuclei in newborn tissue was due to overlap of cells in stacks ( Figure 5A ) . These data suggest that there is a relative increase in the number of cells in longitudinal axis in cell channels during tendon growth comparing newborn with 6 week tendon . We also analysed the change in distance between adjacent cells in different cell stacks in transverse sections . The horizontal distance between cell nuclei increased from 6 . 2 ± 0 . 2 to 31 . 5 ± 0 . 9 µm from birth to 6 weeks ( p = <0 . 05 , Figure 5F ) . The maintenance of similar numbers of fibril bundles per cell ( in transverse section ) during postnatal development and the formation of channels that defined fibril bundles by cell processes led us to investigate cell–cell contacts in more detail . At E15 . 5 no distinct cell membrane channels were formed via membrane protrusions contacting adjacent cells; instead groups of narrow-diameter fibrils were seen between adjacent cell membranes or enclosed in cell membrane invaginations ( as shown in Figures 1A , 6A and Figure 6—figure supplement 1A , data in Figure 6—source data 1 ) . Newborn tenocytes formed protrusions that contacted adjacent cells; these formed channels surrounding fibril bundles ( Figure 6B and Figure 6—figure supplement 1B ) . Modeling of these cell–cell contact regions in 3D revealed close apposition of membranes along the longitudinal axis of the tissue , thereby defining the fibril channels . At 6 weeks a similar morphology of cell–cell contact was found , although the cell processes were much longer ( Figure 6C and Figure 6—figure supplement 1C ) . Furthermore , immunofluorescence staining against connexins 32 and 43 revealed positive immunolocalisation of cell–cell junctions in mature ( 6 week ) tendon ( Figure 6D , E ) . 10 . 7554/eLife . 05958 . 013Figure 6 . Cell–cell contacts during development . ( A ) SBF-SEM images of E15 . 5 ( left and middle panel ) , 3D cell reconstruction ( right panel ) . Two cells are outlined ( blue and green tracing their cell membranes ) . At E15 . 5 cells are packed closely , without distinct membrane protrusions forming well-defined channels for collagen fibrils . Narrow diameter fibrils are found between adjacent cell membranes and in cell membrane invaginations . 3D reconstruction of E15 . 5 cell membranes shows the cells to be in close apposition along their entire length . ( B ) SBF-SEM images of newborn tendon cells . Two cell membranes are traced ( brown and purple ) . Cell-protrusions are seen , forming channels for ECM fibrils . ( C ) SBF-SEM images of 6 week tendon . Two cell membranes are traced ( pink and green , upper row , three images ) , and reconstructed ( lower row , three images ) . Distinct cell–cell contacts are seen between adjacent cells ( circled in red in SBF-SEM images and in the 3D reconstruction ) . ( D , E ) Positive immunofluorescence staining for connexin 32 and connexin 43 in 6 week tendon tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 01310 . 7554/eLife . 05958 . 014Figure 6—source data 1 . Cell-contact data . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 01410 . 7554/eLife . 05958 . 015Figure 6—figure supplement 1 . Quantification of cell–cell contacts . ( A ) Reconstructions from SBF-SEM data of E15 . 5 tendon cells . Cells are closely packed , with multiple neighbouring cells in close apposition . ( B ) Reconstructions from SBF-SEM data of newborn tendon cells . One cell ( red ) is reconstructed with its four surrounding neighbouring cells , which are all contacted via channel-forming protrusions . ( C ) Reconstructions from SBF-SEM data of 6 week tendon cells . One cell ( green ) is reconstructed with its four neighbouring cells , which are contacted channel forming-protrusions . ( D ) Quantification of cell–cell contacts in 3D . The number of other cells contacted per cell was similar in newborn and at 6 weeks . ( E ) Newborn and 6 week tendon cells both formed a similar number of cell–cell contacts that formed fibril channels . † Indicates p = >0 . 05 . Source data in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 015 Quantitation of cell–cell contacts showed that in newborn tendon each tendon cell forms distinct channel-defining protrusions that contact , on average , 4 . 6 ± 0 . 1 adjacent cells . At 6 weeks each tendon cell contacts 4 . 5 ± 0 . 1 adjacent cells ( p = >0 . 05 , Figure 6—figure supplement 1D ) . The number of protrusions per cell that contacted adjacent cells was also similar in newborn and 6 week tendon ( 8 . 2 ± 0 . 2 vs 8 . 3 ± 0 . 2 , Figure 6—figure supplement 1E ) . The observation that the fibril bundles had a regular undulating configuration ( in Figure 3 ) led us to investigate further the 3D shape of the fibril bundles using SBF-SEM ( Figure 7 ) . Single transverse images taken with SBF-SEM are shown in Figure 7A . Fibril bundles are highlighted . These fibril bundles were reconstructed and shown in longitudinal views in Figure 7B . At E15 . 5 an undulation or wavy pattern is seen , but is neither regular nor well defined . By birth the fibril bundles have developed regular undulations , of defined and consistent wavelength . Importantly the undulations are in axial-register ( see Figure 7B , middle panel ) . This wavy pattern is seen in 6 week tendon ( Figure 7B , right panel ) . Oblique views of the reconstructed fibril bundles are shown in Figure 7C to demonstrate the spiral nature of the crimp structure , which is not clearly defined at E15 . 5 , but is marked in newborn and 6 week old tendon . These data show the crimp to derive from a helical spring sub-structure , and is not simply a planar ( 2D ) undulation . The wavelength of the crimp increases during development from 14 ± 0 . 3 μm at birth to 99 ± 2 . 5 μm at 6 weeks ( Table 1 , Figure 7—figure supplement 1 , Data in Figure 7—source data 1 ) . The crimp helix radius ( the distance from the central axis ) was 1 . 6 ± 0 . 1 μm at birth and 2 . 3 ± 0 . 1 at 6 weeks . The irregular pattern of the crimp at E15 . 5 precluded accurate measurement of crimp wavelength and translation distance . 10 . 7554/eLife . 05958 . 016Figure 7 . The 3D crimp structure of tendon . ( A ) Single SBF-SEM images of transverse sections at E15 . 5 , newborn and 6 week old mouse-tail tendon . Fibril bundles are ringed in colour . At 6 weeks the centre of the bundle is marked with a red circle . Centre of bundle points were used to track the bundles in B and C ( 6 weeks , right hand panels ) . ( B ) Longitudinal views of bundle reconstructions . Colours correspond to bundles circled in A . The crimp is apparent at E15 . 5 , but it is not in register between different bundles . By birth the crimp is in register , all bundles have the same wavelength , in a helix throughout the tendon . This morphology is also present in 6 week tendon . ( C ) Oblique views of bundle reconstructions demonstrate that the crimp structure is a regular helix . A bird's-eye view of a fibril bundle reconstruction is shown for 6 weeks , demonstrating a spiral structure . Reconstructed bundles are tracked at E15 . 5 and newborn using the outline of the bundle . At 6 weeks the large size of the bundle made appreciation of the crimp structure difficult . Therefore the centre of the bundle was used as the marker point for reconstruction ( marked in A , right hand panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 01610 . 7554/eLife . 05958 . 017Figure 7—source data 1 . Crimp structure data . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 01710 . 7554/eLife . 05958 . 018Figure 7—figure supplement 1 . Crimp wavelength . Crimp wavelength increases significantly from 14 µm at newborn to almost 100 µm at 6 weeks . ** Indicates significant difference ( p = <0 . 05 ) . Source data in Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 01810 . 7554/eLife . 05958 . 019Figure 7—figure supplement 2 . Fibril length calculations . ( A ) E15 . 5 fibrils tracked through a z-depth of 10 µm ( 100 SBF-SEM sections cut at 100 nm thickness ) . Fibrils traversing the complete z distance are shown in purple , fibrils with tips identified are shown in yellow . Estimated fibril length 125 ± 18 µm . Scale bar = 1 µm . ( B ) Newborn fibrils tracked over 20 µm ( 200 SBF-SEM sections ) . Fibrils traversing the complete z distance ( green ) and fibrils with a tip identified ( blue ) are reconstructed . Estimated fibril length 578 ± 17 µm . Scale bar = 2 µm . Source data in Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 019 To give an indication of the macroscopic degree of tissue growth we measured tail length ( Table 1 , Data in Figure 2—source data 1 ) . Tail length increased from 6 . 1 ± 0 . 3 mm ( E15 . 5 ) to 10 . 9 ± 0 . 9 mm ( newborn ) to 67 . 6 ± 0 . 3 mm ( 6 week ) ; overall a factor of 10 increase . The increase in length between newborn and 6 weeks closely matches the change in crimp wavelength ( ∼factor of 6–7 fold increase ) . In 119 out of 120 specimens we have so far examined we have consistently found the helical coil to turn in a left-handed spiral . In one specimen from 6 week mouse we observed a right-handed helix . This continues the alternating handedness from polypeptide chains ( left-handed helix ) , to collagen molecules ( right-handed ) , to microfibrils ( left-handed ) and collagen fibrils ( right-handed , as described in [Holmes et al . , 2001] ) . These data show that the fibrils in tendon bundles then form a left-handed helical macrostructure . We have observed this left-handed spiral structure of fibril bundles in all the anatomically distinct tendons we have examined using SBF-SEM , including mouse Achilles tendon and chick metatarsal tendons , with different biomechanical properties compared with mouse-tail tendon , an axial tendon ( Benjamin et al . , 2008 ) . Using SBF-SEM we were also able to make estimates of mean fibril length by tracking fibrils and counting fibril tips ( Starborg et al . , 2013 ) . This demonstrated a significant increase in fibril length through development , from 125 ± 18 µm at E15 . 5 , to 578 ± 17 µm in newborn tendon ( Figure 7—figure supplement 2 , p = <0 . 05 , Data in Figure 7—source data 1 ) . At E15 . 5 160 fibril tips were identified ( within 10 , 000 µm fibril lengths tracked ) . In newborn mouse tendon , 69 tips were found ( over ∼20 , 000 µm fibril lengths tracked ) . At 6 weeks fibril tips were rare; 16 tips were identified within 10 , 000 µm fibril lengths tracked , all occurring in fibrils less than 150 nm in diameter ( the smaller population of postnatal fibrils ) . No tips were found in fibrils >150 nm in diameter . This gave an estimate of mean-length of 1250 ± 305 µm for fibrils less than 150 nm diameter . We are unable to estimate the length of larger-diameter fibrils in 6 week tendon . These data , together with the information in Figure 2 and Table 1 , are interpreted in a model of fibril growth in Figure 8 . 10 . 7554/eLife . 05958 . 020Figure 8 . Model for development of tendon ECM . Left side panel . Schematic diagram to represent a model of fibril nucleation and growth during tendon development . Fibrillar arrays representing bundles are shown in longitudinal and transverse section ( labelled LS and TS , respectively ) at the 3 time points ( E15 . 5 , newborn and 6 week ) . The embryonic growth stage involves fibril nucleation to increase fibril numbers along with fibril growth to increase both fibril diameter and length . The numbers of fibrils in the transverse section of the bundle is increased not only by fibril nucleation but also by axial growth of the fibrils into neighbouring regions of the bundle . The postnatal growth in stage 2 involves only fibril growth , with a major increase in fibril diameter and length , with no increase in fibril number . To maintain the observed invariant number of fibrils in the bundle transverse section , the increase in fibril length must match the local increase in tissue length . Right side panel . Schematic to represent cell and fibril growth during tendon development . Early embryonic tendon tissue ( E10 ) is composed of rounded cells with little ECM . Collagen fibrils begin to appear in small bundles in poorly defined channels formed by cell–cell adhesions by E15 . 5 . These channels become more developed with expansion of the ECM and are well defined in newborn tendon tissue . Postnatal expansion of the ECM results in large fibril bundles in these channels that were established during embryonic development . DOI: http://dx . doi . org/10 . 7554/eLife . 05958 . 020
The events during tendon development when ECM overtakes cells as the major component of the tissue are poorly understood , largely because of the lack of a suitable technique with which to examine the micro- and macro-scale architecture . At a micro-scale , it was unknown how collagen fibrils increase in diameter from ∼35 nm to ∼400 nm ( stage 1 to 2 transition ) , and curiously , why diameters are independent of the distance from the cell . At a macro-scale it was unknown how tenocytes become aligned in tramlines parallel to the tissue long axis in adult tendon , and how crimp develops . In this study we have shown that structure-based matrix expansion in the presence of stable cell-to-cell junctions is a novel mechanism for driving tissue growth . The appearance of collagen fibrils and thus the start of tendon development proper , coincides with the formation of extracellular channels whose borders are delineated by cellular extensions ( Birk and Trelstad , 1985 ) . As shown in Figure 1 , the wrapping of plasma membrane extensions around bundles of collagen fibrils in newborn and 6 week tendon produces a highly convoluted cross-sectional shape of the cell . Contacts between projections of neighbouring cells are stabilised by cell junctions ( McNeilly et al . , 1996; Ralphs et al . , 2002; Richardson et al . , 2007 ) . Thus , neighbouring cells define vertical channels into which newly formed collagen fibrils are assembled , thereby generating bundles of collagen fibrils that are corralled into lying parallel to the tendon long axis . Although there is an increase in the number of cells longitudinally ( in cell stacks ) between newborn and 6 weeks , the fall in cells per unit volume of tissue due to expansion of the ECM suggest that the major driver of tissue growth is an increase in the ECM in channels between cell stacks . The extended longitudinal cell–cell contacts seen in both newborn and in 6 week tendon tissue , and the maintenance of similar numbers of cell–cell contacts and cell protrusions at birth and 6 weeks , suggests that the cell–cell contacts formed in embryogenesis form the basic pattern for the mature tendon tissue . Positive immunolocalisation of important gap junction components , connexin 32 and 43 , at 6 weeks supports this conclusion . The increase in cell number along the longitudinal axis of the tendon in postnatal growth is compatible with this interpretation . New cells formed by mitosis during tissue growth can be added in cell stacks whilst at the same time maintaining the longitudinal channels containing fibril bundles , which are growing laterally due to a dramatic increase in fibril diameter . We were able to quantitate the number of fibril profiles , diameter and average length of collagen fibrils during embryonic to postnatal development . As shown in Figure 2 , fibril diameters increased during E15 . 5 to newborn but maintained a unimodal diameter distribution . During embryonic development the number of fibrils per bundle increases ( from E15 . 5 to birth ) . There was a significant increase in fibril diameter in postnatal tendon , and fibrils developed a bimodal diameter distribution . Also during postnatal growth ( between birth and 6 weeks ) the number of bundles per cell remained constant and the number of fibril profiles per bundle remained constant . The fact that the number of bundles per cell remained constant was a good indication that the cell–cell junctions between neighbouring cells are stable . SBF-SEM allowed estimation of fibril length , which showed a significant increase in fibril length from E15 . 5 to newborn . Taken together , these data showed that growth during the period E15 . 5 to newborn was the result of new fibril formation together with an increase in fibril diameter and length . Recent studies have shown that the collagen fibrils in the extracellular channels , at this early stage of tendon development , are formed in cell-surface fibripositors ( Canty et al . , 2004; Kalson et al . , 2013 ) , and therefore , the cell strictly controls the number of fibrils deposited to the bundles . SBF-SEM analysis of 6 week postnatal tendon provided explanations for features that have long been a puzzle . As shown in Figure 2 , the number of bundles per cell and the number of collagen fibril profiles per bundle remain stable between birth and 6 weeks postnatal . During the same period , fibril diameters increase markedly and the distribution becomes bimodal . These data showed that , in mouse , cells set the number of bundles and number of collagen fibrils per bundle at the time of birth . Local growth in tendon size between newborn and 6 weeks postnatal is driven primarily by increase in collagen fibril diameter and length . Growth models have been proposed for the increase in fibril diameter based on inter-fibrillar fusion and accretion of newly-synthesised collagen molecules ( Birk et al . , 1995; Kadler et al . , 2000; Trotter et al . , 2000 ) . Inter-fibrillar fusion can potentially involve tip-to-tip , tip-to-shaft , and shaft-to-shaft fusion . Tip-to-tip fusion results in fibril lengthening . Quantitative mass mapping by scanning transmission electron microscopy and analysis of fibril staining patterns by TEM showed that tip-to-tip fusion occurs during early embryonic tendon morphogenesis and relies on unipolar collagen fibrils , in a process that is regulated by collagen-proteoglycan interactions ( Graham et al . , 2000 ) . Tip-to-shaft fusion also occurs , which generates branched networks ( Kadler et al . , 2000 ) . A model for linear and lateral growth of fibrils during tendon development has been proposed in which decorin , and perhaps related small proteoglycans , regulates surface interactions of participating fibrils ( Birk et al . , 1995 ) . Thus , fibril fusion is regulated by the directionality of collagen molecules in the fibril ( Kadler et al . , 1990 ) , the availability of tips from unipolar fibrils ( Graham et al . , 2000 ) , and by the presence of proteoglycans at the fibril surface ( Birk et al . , 1995; Graham et al . , 2000 ) . An important and often overlooked consequence of fibril fusion is the inevitable reduction in fibril numbers as fibrils fuse together , and the decrease in the number of fibrils per cell . It is also to be noted that fibril fusion has no effect on fibril volume fraction as no new fibril volume is created by any type of fusion; we found very few fibril fusion ( tip-shaft ) events when we tracked fibrils with SBF-SEM . The data presented here show that collagen fibril numbers ( in profile ) increased during stage 1 of development and remained constant during stage 2 , despite increases in fibril diameter ( in stage 1 and stage 2 ) . Although we cannot rule out tip-to-tip fusion ( which would not be expected to affect fibril diameter ) our diameter and fibril number data do not support shaft-to-shaft fusion as a major contributor to increases in fibril diameter during tendon development . We also have not observed irregularly shaped fibrils , which would be expected to occur if lateral fusion was occurring . An alternative mechanism of fibril growth is the surface-nucleation-and-propagation ( SNP ) model in which collagen molecules accrete onto the growing tips of fibrils ( Holmes et al . , 1992 , 1998 ) and to the molecular reversal region in bipolar fibrils ( Trotter et al . , 1998 , 2000 ) . The SNP model predicts that collagen fibrils have specific binding sites for collagen molecules , leading to growth in diameter and length . Thus , fibril growth is determined by local structure at the fibril surface and is an example of interface-controlled growth as opposed to diffusion-limited accretion onto the fibril surface . The occurrence of an abrupt limitation in fibril diameter at the growing tips of vertebrate collagen fibrils also suggests a structure-based mechanism for fibril growth ( Holmes et al . , 1998 ) . Evidence from collagen fibril diameter measurements in vivo also supports a structure-based assembly mechanism . If collagen molecules could assemble onto non-specific binding sites on the fibril in an uncoordinated manner , the axial mass profiles would be expected to show pronounced random fluctuations . In the case of diffusion-limited growth , collagen fibrils closer to the cell might be expected to have larger diameters because of the ready availability of collagen molecules emerging from the cell . However , as fibril diameters are independent of distance from the cell , it seems more likely that fibril growth ( especially in postnatal tendon ) is strongly influenced by the local structure of the fibril surface . Thus , the establishment of fibril number during stage 1 of development and subsequent growth in fibril diameter during stage 2 by the SNP mechanism ( in the absence of overt shaft-to-shaft fibril fusion ) would explain the observed increase in fibril diameter in stage 2 and the lateral growth of the tissue . The ability of SBF-SEM to examine relatively large volumes of tissue at good resolution enabled us to assess cell shape with precision . As shown in Figure 4 , cells at E15 . 5 and newborn had convoluted profiles , were slender cylinders with overlapping ends , and were ∼50–60 µm in length . In contrast , cells in 6 weeks postnatal had prominent lateral cellular extensions that reached deep into the ECM , were short ( ∼25 µm ) and stacked in rows on their blunt ends . Importantly , cells at 6 weeks postnatal tendon remained in contact with each other via their cellular extensions regardless of the distance between the cell bodies where the nucleus was located . The number of collagen bundles at the cell periphery ( and therefore the number of cell–cell contacts ) was stable between newborn and 6 weeks postnatal . The data show that the main cell bodies move laterally apart because of the expansion of the ECM but maintain their longitudinal relationship , thereby generating the tramline organisation seen in mature tendon . An unexpected result was visualisation of a spiral 3D organisation of collagen bundles to form the tendon crimp . In relaxed tendons , the bundles of collagen fibrils are buckled into a wavy pattern , called crimp , which is readily visible using plane polarized light ( Kapacee et al . , 2008 ) . Unbuckling of the crimp during longitudinal loading functions as a natural shock-absorber on initial loading as well as being important in elastic recoil ( reviewed by [Benjamin et al . , 2008] ) . Crimp first appears during embryonic tendon development ( Shah et al . , 1982 ) but the mechanism of crimp formation remains unknown , largely because of the impracticalities of studying its formation in vivo . In our studies we were careful to preserve the tension in the tendon , avoiding major changes in crimp structure ( Shah et al . , 1977 ) , by fixing whole tails without disturbing the axial attachment to the skeleton . Crimp 3D structure has been debated over a prolonged period . We found no kinks as previously described ( Raspanti et al . , 2005 ) but instead observed a left-handed spiral running the length of our 3D reconstructions . A spring/spiral nature of the tendon crimp has been predicted in a mathematical model ( Grytz and Meschke , 2009 ) . It is important to note that fibril bundles are continuous over the distances we have measured here , forming parallel spirals in axial register to provide uniform biomechanical properties when loaded , and there is a close relationship between the cell and the fibril bundle . The close association of cells with bundles suggests that the cell shape strongly influences the configuration of the crimp structure . This is consistent with studies suggesting that cell force generated by the actin-myosin system can create a crimp structure ( Herchenhan et al . , 2012 ) . As illustrated in Figure 7 , we showed that the crimp wavelength lengthens from ∼14 µm at E15 . 5 to ∼99 µm at 6 weeks postnatal . The length increase of the mouse-tail from newborn to 6 weeks by a factor of ∼6 . 2× closely matches the increase in crimp wavelength of ∼7 . 1× . This suggests a simple crimp elongation process in synchrony with bundle and tissue elongation . In conclusion , we have shown that tendon development occurs in a two-stage process . In stage 1 the organisation of fibroblasts and the cellular channels for fibril deposition and growth are established , and cell-mediated control of collagen fibril deposition occurs . This sets the developmental template for structure-based matrix expansion of the existing matrix in stage 2 , which drives the increase in the size of the matrix that underpins tissue growth . Important features include the cell-regulated number of cell processes that define the boundaries of collagen fibril bundles and help to maintain the tramline organization of cells in the mature tissue . These mechanisms could be a paradigm for the growth of other matrix-rich tissues that are abundant in collagen fibrils .
C57/Black6 mice were sacrificed using a UK Home Office approved S1 Schedule method . Male mice were sacrificed at embryonic day 15 . 5 , birth ( newborn ) and at 6 weeks postnatal for SBF-SEM and TEM . Whole tendons were fixed in situ in the tail , preserving the native state of the tail tendon , which at E15 . 5 and at birth was small enough to allow penetration of the fixative . The skin from 6 week mouse-tail was removed , before fixation of the entire underlying structures to allow penetration of the fixative . Tendon from the tail axial mid-point was examined with TEM and SBF-SEM to allow comparison between different ages . Samples were prepared as described ( Starborg et al . , 2013 ) . In brief , mouse-tails were fixed in situ by using 2% ( wt/vol ) glutaraldehyde ( Agar Scientific , UK ) in 0 . 1 M cacodylate buffer ( pH 7 . 2 ) , en-bloc stained in 1% ( wt/vol ) osmium tetroxide , 1 . 5% ( wt/vol ) potassium ferrocyanide in 0 . 1 M cacodylate buffer , followed by 1% ( wt/vol ) tannic acid in 0 . 1 M cacodylate buffer ( pH 7 . 2 ) . After washing with 0 . 1 M cacodylate buffer ( pH 7 . 2 ) , more osmium was added by staining in 1% ( wt/vol ) osmium tetroxide . The final staining step involved soaking in 1% ( wt/vol ) uranyl acetate . Samples were dehydrated in ethanol and infiltrated in Araldite resin ( CY212; Agar Scientific ) . A minimum of three samples per time point were imaged using SBF-SEM . In this experiment and other investigations in our laboratory we have imaged >100 mouse and chick tendons using SBF-SEM ( 12 were used for SBF-SEM in this study ) ( Kalson et al . , 2013 ) . Resin-embedded samples were sectioned using a Gatan 3View microtome within an FEI Quanta 250 scanning electron microscope as described ( Starborg et al . , 2013 ) . For E15 . 5 and newborn mouse-tail samples , a 41 μm × 41 μm field of view was chosen and imaged by using a 4096 × 4096 scan , which gave an approximate pixel size of 10 nm . The section thickness was set to 100 nm in the Z ( cutting ) direction . For 6 week-old mouse-tail samples the field of view was increased to accommodate the larger cells ( 200 μm × 200 μm ) . Typically , Z volumes datasets comprised 1000 images ( 100 μm z depth ) . The IMOD suite of image analysis software was used to build image stacks , reduce imaging noise , and generate 3D reconstructions ( Kremer et al . , 1996 ) . Cell volume , cell surface area , cell length and crimp wavelength were calculated using functions in the IMOD image analysis suite . 30 measurements ( 10 from each sample ) of 3D cell length , 3D cell volume and 3D cell surface area were made for each time point . Cell length measurement was made along the tendon long axis as the distance between the tip and the tail of the cell . Measurement of crimp wavelength and helix radius was performed from reconstructions in IMOD . 50 measurements of crimp wavelength and helix radius were made at each time point ( 20 measurements from sample 1 , 15 from samples 2 and 3 ) . Cell number measurement was made on three separate SBF-SEM samples for each time point ( E15 . 5 , newborn , 6 weeks ) . The volume of the tendon tissue in each SBF-SEM dataset was calculated and all the cells in the volume were reconstructed using IMOD . Each cell nucleus contained within the reconstruction was identified and counted . Cells per 1000 μm3 of tissue were calculated to allow comparison between samples . The longitudinal distance between nuclei in cell stacks in newborn and 6 week tendon was measured in three separate SBF-SEM datasets . 20 measurements were made on each sample as the distance between the mid-point ( in the longitudinal plane ) of each cell nucleus . Cells per μm2 of transverse tissue area and the transverse distance between cell nuclei in adjacent cell stacks in the transverse tissue plane were calculated from three separate SBF-SEM datasets at each time point ( E15 . 5 , newborn and 6 week for cells per μm2 , newborn and 6 weeks for transverse distance between nuclei ) . For each of these analyses 20 measurements per sample were made , each at 2 . 5 μm intervals through the tissue volume ( along the longitudinal tendon axis ) . For cells per μm2 all cells in a single transverse image were counted , and the area of the tendon calculated . For distance between cell nuclei in adjacent cell stacks the distance between cell nuclei in the transverse tissue plane in cell stacks at newborn and 6 weeks was measured in three separate samples in newborn and 6 week tendon . Measurements were made between the edges of adjacent nuclei . Cell–cell contacts were investigated in newborn and 6 week tendon . Three separate SBF-SEM datasets were used to generate cell reconstructions . For each dataset 10 individual cells were reconstructed , together with each cell's neighbouring cells . The number of cell protrusions that formed fibril channels was counted for each cell ( 30 cells in total ) . The number of different cells contacted by channel-forming cell membrane protrusions were counted for 10 cells per sample ( 30 cells in total ) . Estimation of average fibril length was made as previously described ( Starborg et al . , 2013 ) . For E15 . 5 , 1000 fibrils in four different fibril bundles ( two bundles from separate SBF-SEM datasets ) were tracked for 10 μm ( 10 , 000 μm total length tracked ) . For newborn , 1000 fibrils from two separate SBF-SEM datasets ( in four different bundles ) were tracked over 20 μm ( 20 , 000 μm total length tracked ) . For 6 weeks , 1000 fibrils from two separate SBF-SEM datasets ( four different bundles ) were tracked over 10 μm ( 10 , 000 μm total length tracked ) . Total length of fibrils tracked and the number of fibril tips were recorded and the equationLm=2×Σ ( total length of fibrils in a bundle ) number of fibril tips in a bundle . was used to estimate average fibril length . For TEM analysis a minimum of three samples per time point ( E15 . 5 , newborn and 6 week ) were prepared as previously described ( Starborg et al . , 2013; Canty et al . , 2004 ) . Sections were examined using an FEI Tecnai 12 Twin transmission electron microscope ( TEM ) . Images were captured using a 2 k × 2 k cooled CCD camera ( F214A , Tietz Video and Image Processing Systems , Gauting , Germany ) . For each sample a minimum of three different sections were reviewed . An unbiased sampling of each section was performed . Three different magnifications were used: 2100× for FAF and fibril bundle number per nucleus ( giving a large-area survey ) , and 11 , 000× for fibril diameter , fibril area and fibrils per μm2 . The sampling procedure generated 20 images of each section at 2100× , 40 views per section at 6800× and 60 views per section at 11 , 000× . All measurements were made using ImageJ software ( NIH freeware , http://rsb . info . nih . gov/nih-image ) . For fibril diameter measurement 500 fibrils were measured in total from three separate samples per time point ( 200 from sample 1 , 150 each from sample 2 and 3 ) . Fibril area was calculated from fibril diameter assuming circularity ( 1/4 × π × diameter2 ) . Magnification calibration was performed for each magnification using a diffraction-grating replica grid ( 2160 lines/mm , Agar Scientific , Stansted , UK ) . 20 measurements of FAF were made for each time point ( 7 from sample 1 , 7 from sample 2 and 6 from sample 3 [Starborg et al . , 2013] ) . 30 measurements were made for fibrils per bundle , bundles per nucleus and fibrils per μm2 at each time point ( 10 measurements from each sample ) on TEM images of transverse sections of tendon . This gave a measurement of fibril bundles per nucleus in transverse section . Fibril bundles are continuous along the length of the tendon , so there are significantly fewer bundles than tendon cells when the tendon is considered as a whole composite unit containing cells and ECM . Measurements of tail length were made on four mice at E15 . 5 , two newborn , and two 6 week mice ( all male ) using photographs of whole mice . Measurements were made using Image J . C57/Black6 mice were sacrificed using a UK Home Office approved S1 Schedule method and tail tendons immediately dissected and the cells released by placing the tendon in trypsin ( 37 , 000 U ) and bacterial collagenase ( 522 U ) in DMEM at 37°C for 2 hr . Cells were passed through a 70 μm cell strainer ( BD Biosciences , UK ) , collected by centrifugation ( 240×g for 5 min ) and washed 3 times in PBS . Cells were re-suspended in DMEM4 with 100 U/ml penicillin , 100 μg/ml streptomycin , 2 mM L-glutamine and 10% FCS . Cells were not passaged before examination by light microscopy . Three separate tendon cell isolations were performed for each time point . Cells on coverslips were rinsed 3 times with PBS containing 0 . 9 mM Ca2+ and 0 . 49 mM Mg2+ ( Sigma D8662 ) and fixed with 1% paraformaldehyde in 0 . 1 M HEPES ( pH 7 . 4 ) for 15 min at room temperature . After being permeabilised cells were blocked with 1% BSA in PBS at room temperature for 30 min . FITC labelled phalloidin ( Sigma ) was added and incubated for 1 hr in the dark . Cells were washed , then left to air dry before mounting with vector shield containing DAPI and left to set at 4°C . Samples were examined with a Leica light microscope . Cell area was measured using ImageJ . 10 cells were measured from each isolate ( n = 30 per time point ) . Cryosections of mouse-tail tendon ( 10 µm ) were fixed in 100% acetone at 20°C for 10 min and blocked at 4°C overnight with 5% normal goat serum in PBST ( PBS supplemented with 0 . 1% Triton X-100 ) . Sections were incubated with primary antibody ( 1:250 ) diluted in 1% bovine serum albumin in PBS for 1 hr , washed 3 times for 5 min each with PBST , and incubated with goat anti-rabbit–Cy3 ( 1:1000 ) for 1 hr . Tissue was washed 3 times for 5 min each with PBST and mounted with Vectashield mounting medium containing DAPI ( 4 , 6-diamidino- 2-phenylindole ) . Cryosections of mouse-tail tendons ( 10 µm ) were fixed in 2% PFA and blocked for 1 hr at 4°C with 3% BSA in PBST ( PBS supplemented with 0 . 1% Triton X-100 ) . Sections were incubated with primary antibody ( 1:500 ) diluted in blocking buffer , overnight at 4°C then washed 3 times for 5 min each with PBST , and incubated with goat anti-rabbit–Cy3 ( 1:1000 ) for 1 hr . Tissue was washed 3 times for 5 min each with PBST and mounted with Vectashield mounting medium containing DAPI . Three separate tendon samples ( three slides per sample ) were stained for connexin 32 and 43 . Images were collected on an Olympus BX51 upright microscope using a 20×/0 . 50 Plan Fln objective and captured using a Coolsnap ES camera using MetaVue Software ( Molecular Devices ) . Images were then processed and analysed using ImageJ . Data are presented as mean ± SEM . For all statistical tests type I error was set to 0 . 05 and p values less than 0 . 05 considered to be significant . Three groups were compared for all tests , so the one-way ANOVA was used with a Tukey's post-test . Tests were performed using SPSS version 20 . A summary of raw data is presented in Supplementary file 1 . | Young animals are able to grow in a way that allows them to maintain roughly the same shape until they reach their adult size . The growth of embryos is driven by increases in cell size and number , but it is less clear how the body grows after birth . By this point , many of the cells in the body are part of tendons and other fibrous tissues , where they are surrounded by a mesh of fibres made of collagen and other proteins . These fibres provide strength to the tissue , but may also restrict its ability to grow . Tendons connect muscles to bones . They contain fibres of collagen that run along their length , which enables them to cope with very strong pulling forces . Kalson et al . used electron microscopy to generate highly detailed three-dimensional models of mouse tendons at three stages: in the embryo , at birth and six weeks later . The experiments identified two stages in tendon development . During the first stage , the number of cells and fibres across the tendon is determined in the embryo . The fibres also slightly expand in diameter and form regular waves called crimps that are important for the structural strength of the tendon . The second stage happens after birth , during which the number of cells and fibres remains constant , but the tendons continue to grow because the fibres increase in diameter and length . The cells also move to form towers of cells running along the tendon . From these observations , Kalson et al . propose that the numbers and locations of the cells and collagen fibres that determine the shape and size of tendons are established in the embryo . The collagen fibres create a framework for the continued growth of the tendon after birth . Future challenges are to understand how the number and the arrangement of cells in the tendon is determined before the collagen fibres are made , and how these cells control the number of collagen fibres that form . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2015 | A structure-based extracellular matrix expansion mechanism of fibrous tissue growth |
Wingless ( Wg ) /Wnt signaling is conserved in all metazoan animals and plays critical roles in development . The Wg/Wnt morphogen reception is essential for signal activation , whose activity is mediated through the receptor complex and a scaffold protein Dishevelled ( Dsh ) . We report here that the exon junction complex ( EJC ) activity is indispensable for Wg signaling by maintaining an appropriate level of Dsh protein for Wg ligand reception in Drosophila . Transcriptome analyses in Drosophila wing imaginal discs indicate that the EJC controls the splicing of the cell polarity gene discs large 1 ( dlg1 ) , whose coding protein directly interacts with Dsh . Genetic and biochemical experiments demonstrate that Dlg1 protein acts independently from its role in cell polarity to protect Dsh protein from lysosomal degradation . More importantly , human orthologous Dlg protein is sufficient to promote Dvl protein stabilization and Wnt signaling activity , thus revealing a conserved regulatory mechanism of Wg/Wnt signaling by Dlg and EJC .
Canonical Wingless ( Wg ) /Wnt signaling plays an evolutionarily conserved role in dictating cell proliferation , pattern formation , stem cell maintenance and adult tissue homeostasis . Given the importance of Wg/Wnt signaling in many cellular processes , it is not surprising that dysregulation of Wg/Wnt signaling in humans results in developmental defects as well as cancer ( MacDonald et al . , 2009; Clevers and Nusse , 2012 ) . In Drosophila , Wg ligand binds to the seven-pass transmembrane receptors Frizzled/Frizzled2 ( Fz/Fz2 ) and a co-receptor Arrow ( Arr ) , a homolog of vertebrate LRP5/6 . Formation of this trimeric complex activates a scaffold protein Dishevelled ( Dsh ) on the plasma membrane , leading to disruption of the Axin-mediated degradation complex and hence stabilization of Armadillo ( Arm ) , a homolog of vertebrate β-catenin . Accumulated Arm then translocates to the nucleus to activate target gene transcription ( MacDonald and He , 2012 ) . Although core components of the Wg/Wnt signaling cascade have been identified , gaps in the understanding of this critical signaling pathway still remain to be filled . To unveil novel regulators of Wg signaling , we conducted a genome-wide RNAi screen in the developing Drosophila wing , from which a RNA binding exon junction complex ( EJC ) emerged as a positive regulator of Wg signaling . The EJC is known to act in several aspects of posttranscriptional regulation , including mRNA localization , translation and degradation ( Tange et al . , 2004; Le Hir et al . , 2016 ) . After transcription , the pre-mRNA associated subunit eIF4AIII is loaded to nascent transcripts about 20–24 bases upstream of each exon junction , resulting in binding of Mago nashi ( Mago ) /Magoh and Tsunagi ( Tsu ) /Y14 proteins to form the pre-EJC core complex . The pre-EJC then recruits other proteins including Barentsz ( Btz ) to facilitate its diverse function ( Shibuya et al . , 2004 ) . In vertebrates , the EJC is known to ensure translation efficiency ( Nott et al . , 2004 ) as well as to activate nonsense-mediated mRNA decay ( NMD ) ( Gatfield et al . , 2003; Chang et al . , 2007 ) . In Drosophila , however , the EJC does not contribute to NMD ( Gatfield et al . , 2003 ) . It is instead required for the oskar mRNA localization to the posterior pole of the oocyte ( Newmark and Boswell , 1994; Hachet and Ephrussi , 2001; Mohr et al . , 2001; van Eeden et al . , 2001; Palacios et al . , 2004 ) . Very recently , the pre-EJC has been shown to play an important role in alternative splicing of mRNA in Drosophila . Reduced EJC expression results in two forms of aberrant splicing . One is the exon skipping , which occurs in MAPK and transcripts that contain long introns or are located at heterochromatin ( Ashton-Beaucage et al . , 2010; Roignant and Treisman , 2010 ) . The other is the intron retention on piwi transcripts ( Hayashi et al . , 2014; Malone et al . , 2014 ) . Furthermore , transcriptome analyses in cultured cells indicates the role of EJC in alternative splicing is also conserved in vertebrates ( Wang et al . , 2014 ) . In this study , we have utilized the developing Drosophila wing as an in vivo model system to investigate new mode of regulation of Wg signaling . We find that the pre-EJC positively regulates Wg signaling through its effect on facilitating Wg morphogen reception . Further studies reveal that the basolateral cell polarity gene discs large 1 ( dlg1 ) is an in vivo target of the pre-EJC in Wg signaling . We show that Dlg1 acts independently from its role on cell polarity to stabilize Dsh protein , thus allowing Wg protein internalization required for signaling activation . Furthermore , we demonstrate that human Dlg2 exhibits a similar protective role on Dvl proteins to enhance Wnt signaling in cultured human cells . Taken together , our study unveils a conserved regulatory mechanism of the EJC and Dlg in Wg/Wnt signaling .
The majority of the Wg/Wnt signaling components have been identified through classical forward genetic screens in Drosophila ( Swarup and Verheyen , 2012; Jenny and Basler , 2014 ) . However , these screens failed to uncover a regulatory role of RNA processing in Wg signaling , probably due to the fact that most components of RNA machineries exhibit pleiotropic effects in early development . In an in vivo RNAi screen , we found that knocking down three core components of the pre-EJC , tsu , mago and eIF4AIII , respectively , resulted in loss of marginal tissues and sensory bristles in adult Drosophila wing blade ( Figure 1—figure supplement 1E–G ) , which resembles stereotypical phenotypes associated with reduced Wg signaling . Furthermore , loss-of-function tsuΔ18 or mago93D mutants ( Roignant and Treisman , 2010 ) displayed similar defects in wing development ( Figure 1A , B ) . To confirm that Wg signaling was indeed altered in pre-EJC mutants , we examined in wing imaginal discs the expression of two Wg signaling targets , senseless ( sens ) and Distal-less ( Dll ) ( Seto and Bellen , 2006 ) , which respond to graded Wg morphogen . We observed obvious loss of Sens and Dll protein production in tsu or mago somatic clones ( Figure 1C–F; Figure 1—figure supplement 2A–D ) . However , Sens expression was not altered in somatic clones of btz ( Figure 1—figure supplement 3A , B ) , which is a cytoplasmic component of the EJC ( Palacios et al . , 2004 ) , suggesting that the role of EJC in Wg signaling is independent of its cytoplasmic function . To directly monitor transcriptional activity of Wg signaling in pre-EJC defective wing discs , two lacZ enhancer traps inserted in the genomic loci of Wg targets , frizzled3 ( fz3 ) ( Sivasankaran et al . , 2000 ) and Dll , were used . As expected , the expression of fz3-lacZ and Dll-lacZ was decreased when tsu activity was reduced ( Figure 1G , H , Figure 1—figure supplement 1H–J ) . Taken together , the above data indicate that the pre-EJC activity is required for Wg signaling activation in the developing fly wing . 10 . 7554/eLife . 17200 . 003Figure 1 . The pre-EJC positively regulates Wg signaling . ( A , B ) A typical loss of Wgsignaling wing margin phenotype was observed when tsuΔ18 or mago93D mutant somatic clones were generated in adult wings . Arrows indicate serrated wing margin . ( C–H ) The production of Wg signaling targets Sens ( C , D ) , Dll ( E , F ) and fz3-lacZ ( G , H ) was reduced in tsuΔ18 clones ( marked by the absence of GFP and hereafter in subsequent figures ) . The positions of clones are indicated by arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 00310 . 7554/eLife . 17200 . 004Figure 1—figure supplement 1 . Knocking down individual components of the pre-EJC reduces Wg signaling in the developing wing . ( A ) The vgBE-Gal4 driver used in the screen confers expression of RNAi transgenes along the dorsal-ventral ( D-V ) boundary in the third instar wing imaginal disc as marked by the expression of UAS-gfp transgene ( bottom panel ) . DAPI labeling marks the nuclei . ( B ) Expression of gfp alone by vgBE-Gal4 did not produce any defect in the adult wing . ( C , D ) Shown are differential activation of Wg signaling target genes along the D-V boundary in wildtype wing discs . Moderate Wg signaling induces the expression of Dll-lacZ in a broad region across the wing pouch ( C ) . High level of Wg activity results in activation of Sens immediately adjacent to the D-V boundary ( D ) . ( E–G ) Reduced expression of either component of the pre-EJC , tsu ( E ) , mago ( F ) or eIF4AIII ( G ) by RNAi resulted in a typical loss of wing margin phenotype . ( H–J ) The activity of Dll-lacZ was reduced when tsu RNAi was expressed by hh-Gal4 in the posterior compartment of the wing imaginal disc ( I ) . Similarly , the expression of a high threshold Wg signaling target Sens was abolished ( J ) . Arrows mark the posterior part of the wing disc . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 00410 . 7554/eLife . 17200 . 005Figure 1—figure supplement 2 . The pre-EJC component Mago positively regulates Wg signaling . ( A–D ) The expression of Wg signaling targets Sens ( A , B ) and Dll ( C , D ) was reduced in loss of function mago93D somatic clones ( marked by the absence of GFP ) . ( E–F ) Wg protein stained with the conventional method was accumulated in mago93D clones . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 00510 . 7554/eLife . 17200 . 006Figure 1—figure supplement 3 . The EJC cytoplasmic component Btz does not regulate Wg signaling . The expression of Sens ( A , B ) and the accumulation of Wg ( C , D ) were not altered in loss-of-function btz2 somatic clones ( marked by the absence of GFP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 006 To understand how the pre-EJC regulates Wg signaling , we examined the expression of Wg signaling components in wing discs with altered EJC activity . We found that the amount of Wg protein present in tsu or mago , but not btz , mutant clones was significantly increased ( Figure 2A–C; Figure 1—figure supplement 2E , F; Figure 1—figure supplement 3C , D ) . This finding was surprising given the observation that Wg signaling was downregulated in mutant cells . Clue to the understanding of this apparent contradiction came from the observation that Wg signaling was ectopically activated in wildtype cells immediately next to mutant clones ( insets of Figures 1D and 2C ) . This result suggests that increased Wg protein present in mutant clones is sufficient to activate signaling in neighboring wildtype cells whilst mutant cells are incapable of receiving Wg input . Therefore , the Wg signaling defects observed in mutant cells could be caused by blockage of Wg reception at the plasma membrane . Consistent with this hypothesis , extracellular Wg protein ( Strigini and Cohen , 2000 ) accumulated significantly in both apical and basolateral extracellular spaces ( Figure 2D–F ) . Furthermore , we showed that the increased extracellular Wg was not due to heightened wg gene transcription nor Wg protein secretion in producing cells because neither the expression of wg-lacZ ( a wg transcription reporter ) nor the amount of NRT-Wg ( a membrane-tethered form of Wg ) was altered ( Figure 2G–K; also see Figure 3—figure supplement 3B for wg transcription analyses ) . 10 . 7554/eLife . 17200 . 007Figure 2 . The pre-EJC is required for Wg morphogen reception . ( A–C ) Wg protein stained with the conventional method ( B ) was accumulated in tsuΔ18 clones where the expression of Sens was reduced ( arrows ) . The regions marked by arrowheads are shown in insets . Note that Sens was activated in cells outside the mutant clone ( C ) . ( D–F ) Extracellular Wg was accumulated at the basal ( E ) and apical extracellular spaces of the tsuΔ18 clones ( F ) . ( G–I ) The activity of wg-lacZ ( H ) did not change when tsu RNAi was expressed by ptc-Gal4 ( marked by GFP and arrows ) . Note that Dll expression was reduced in tsu RNAi-expressing cells ( I ) . ( J , K ) The expression of plasma membrane bound NRT-HA-Wg did not change when tsu RNAi was expressed by hh-Gal4 . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 007 Previous studies have shown that inhibiting Wg protein endocytosis by reducing either the activity of Wg receptor complex Fz2/Arr or the scaffold protein Dsh results in accumulation of Wg protein on the plasma membrane ( Han et al . , 2005 ) . In tsu and mago mutant clones , we found that the expression of Dsh ( Figure 3A–D ) , but not Fz2 or Arr ( Figure 3—figure supplement 1 ) , was obviously reduced . This result was further verified in cultured Drosophila Schneider 2 ( S2 ) cells in which tsu or mago was knocked down by RNAi ( Figure 3I ) . To unveil the functional importance of the EJC regulation on Dsh , overexpressed dsh was able to rescue the wing margin defects caused by tsu knockdown ( Figure 3H; cf . Figure 3E ) , whereas fz2 or arr had little effect ( Figure 3F , G ) . The above experiments suggest that the pre-EJC acts primarily through Dsh in wing discs to regulate Wg signaling reception . 10 . 7554/eLife . 17200 . 008Figure 3 . The pre-EJC regulates Wg signaling through Dsh . ( A–D ) The amount of Dsh protein was reduced in tsuΔ18 and mago69B clones in the wing disc . ( E–H ) Overexpressing dsh ( H ) , but not fz2 ( F ) or arr ( G ) , rescued the loss of Wg signaling wing margin phenotype caused by tsu knockdown ( E ) . Arrows indicate sensory bristles along the wing margin . ( I ) The production of Myc-tagged Dsh was reduced when tsu or mago dsRNA was expressed in S2 cells . Yeast Gal80 dsRNA served as a negative control for RNAi treatment . MAPK/Rl , a known pre-EJC target served as a positive control for defective pre-EJC . β-Tubulin was used as a loading control for all experiments . ( J ) Real time RT-PCR revealed that the abundance of dlg1 , RhoGEF2 and l ( 2 ) gl , but not dsh mRNA , was reduced when tsu dsRNA was expressed in wing discs . mapk/rl served as a positive control . α-Tubulin 84B was used to normalize the amount of cDNA template . The experiments were performed in triplicates , and data were represented as the mean+S . D . ( *p<0 . 05; **p<0 . 01; Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 00810 . 7554/eLife . 17200 . 009Figure 3—figure supplement 1 . The pre-EJC does not obviously regulate the expression of Fz2 and Arr . Shown are expression patterns of Fz2 ( A ) and Arr ( D ) , two co-receptors for Wg morphogen reception , in wildtype wing discs . The expression of Fz2 ( B , C ) was slightly upregulated in some tsuΔ18 clones , while the expression of Arr ( E , F ) remained largely unchanged . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 00910 . 7554/eLife . 17200 . 010Figure 3—figure supplement 2 . The mRNA abundance of cDNA-derived dsh is not altered when the pre-EJC activity is knocked down in S2 cells . The mRNA expression of cDNA-derived dsh was not altered when tsu ( A ) or mago ( B ) dsRNA was expressed in S2 cells . Yeast Gal80 dsRNA served as a negative control for RNAi treatment . mapk/rl served as a positive control . α-Tubulin 84B served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 01010 . 7554/eLife . 17200 . 011Figure 3—figure supplement 3 . The mRNA abundance of dlg1 but not those encoding other Dsh-interacting proteins is reduced when the pre-EJC is knocked down in wing discs . The mRNA expression of dlg1 ( B ) but not dsh ( A ) nor genes that encode for other Dsh-interacting proteins ( C ) was obviously reduced when tsu RNAi was expressed by T80-Gal4 in wing discs . mapk/rl served as a positive control . α-Tubulin 84B served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 011 The pre-EJC is known to function at the level of target mRNA splicing . However , we were not able to correlate reduced Dsh protein production observed in pre-EJC defective cells with decreased dsh mRNA expression ( Figure 3J; Figure 3—figure supplement 2; Figure 3—figure supplement 3A ) . This inconsistency could be due to the fact that the dsh locus does not contain any intron , and in principle , may not be subjected to the EJC regulation . Thus , we reasoned that the pre-EJC must regulate Wg signaling through a Dsh-interacting protein whose mRNA expression may be controlled by the pre-EJC . To uncover bona fide EJC targets whose encoded protein products interact with Dsh to control Wg signaling reception , we utilized whole transcriptome RNA-seq to compare mRNAs extracted from wildtype ( i . e . overexpressing lacZ ) and pre-EJC-defective ( i . e . overexpressing tsu RNAi ) wing discs , respectively . We found that the expression of 1447 mRNAs was altered by more than 25% when the pre-EJC activity was downregulated ( Supplementary file 2 ) . This list likely includes both direct as well as indirect pre-EJC targets as 629 genes were identified whose expression was increased . Among those genes whose mRNA levels were reduced we found that the expression of sens and Dll mRNA , two Wg signaling transcription targets , was decreased by more than 50% in wing discs expressing tsu RNAi . Consistent with the RT-PCR result ( Figure 3—figure supplement 3 ) , wg and dsh mRNA did not show an obvious change . To effectively narrow down the pre-EJC targets acting in Wg signaling , we compared our candidate list with annotated information extracted from the FlyBase describing validated protein interactions with Dsh ( Carmena et al . , 2006; Chung et al . , 2009; Kaplan and Tolwinski , 2010; Varelas et al . , 2010; Johnston et al . , 2013; Schertel et al . , 2013; Strutt et al . , 2013; Warrington et al . , 2013; Garcia et al . , 2014 ) . Among 818 mRNAs whose expression was downregulated , we identified three genes , dlg1 , lethal ( 2 ) giant larvae [l ( 2 ) gl] and diablo ( dbo ) , whose protein products are known to directly interact with Dsh . qPCR and RT-PCR analyses were then performed to confirm bona fide Dsh interactors that potentially mediate the pre-EJC activity on Dsh ( Figure 3J; Figure 3—figure supplement 3B ) . In both cases , dlg1 was the candidate consistently exhibiting reduced expression in the pre-EJC-defective wing disc cells when compared with that in wildtype cells . These results fit well with RNA-seq analyses that the amount of dlg1 mRNA was reduced by about 30% when the pre-EJC was dysfunctional . We suspected that reduced dlg1 expression may be a consequence of altered RNA splicing . Apart from previously reported exon skipping caused by dysfunctional EJC ( Ashton-Beaucage et al . , 2010; Roignant and Treisman , 2010 ) , we uncovered two additional aberrant splicing events ( Figure 4—figure supplement 1 ) . The first event was exon inclusion , which retained exons that were normally efficiently spliced in specific isoforms ( Figure 4—figure supplement 1E ) . This event likely increases the usage of splice sites . Indeed the RNA-seq analyses revealed a slightly higher usage of annotated splice sites ( i . e . 1 . 4% higher usage for 5’ splice sites and 1 . 5% for 3’ splice sites ) in pre-EJC depleted wing disc cells compared with those in wildtype cells ( Figure 4—figure supplement 2A ) . The second event utilized previously unidentified splice sites to generate new introns and exons ( Figure 4—figure supplement 1F–H , 2B , C; Supplementary file 3 ) . Together , our bioinformatic analyses suggest that the pre-EJC plays a critical role in alternative splicing by preserving correct usage of splice sites to generate functional mRNA products . The impact of the pre-EJC regulation on dlg1 splicing was further verified at the protein level as the amount of Dlg1 protein was reduced in tsu mutant wing disc clones ( Figure 4A , B ) as well as in S2 cells treated with dsRNA against individual components of the pre-EJC ( Figure 4E ) . In contrast , the level of Dlg1 protein derived from a cDNA expression construct did not change ( Figure 4F ) . Furthermore , we demonstrated that reduced dlg1 expression led to accumulation of extracellular Wg ( Figure 4C , D ) , attenuation of Wg signaling in wing discs ( Figure 4—figure supplement 3A–D ) , and consequently loss of wing margin and sensory bristles in the adult wing blade ( Figure 4—figure supplement 3E ) . Significantly , overexpressed dlg1 was sufficient to rescue wing margin defects caused by dysfunctional pre-EJC ( Figure 4—figure supplement 3H ) . It is known that additional splicing factors are required for the pre-EJC activity ( Reichert et al . , 2002 ) . Inhibiting one of such factors , Rnps1 , in the wing disc resulted in the same defect in Dlg1 and Dsh protein production as well as Wg signaling activation ( Figure 4—figure supplement 4 ) . Together , these experiments indicate that the pre-EJC mediated splicing activity positively regulates dlg1 to control Wg signaling in Drosophila . 10 . 7554/eLife . 17200 . 012Figure 4 . The pre-EJC regulates Dsh-interacting protein Dlg1 . ( A , B ) The expression of endogenous Dlg1 protein was reduced in tsuΔ18 clones ( indicated by an arrow ) . ( C , D ) Extracellular Wg was accumulated in dlg114 clones . Shown in the inset is an enlarged clone with accumulated extracellular Wg . ( E , F ) The amount of endogenous ( E ) but not a cDNA-derived Dlg1 protein ( F ) was reduced when tsu or mago dsRNA was expressed in S2 cells . MAPK/Rl served as a positive control for dysfunctional pre-EJC activity . ( G , H ) Endogenous Dlg1 interacted with Myc-tagged Dsh in S2 cells ( G ) and GFP-tagged Dsh in wing discs expressing the dsh-gfp under the control of the dsh promoter ( H ) . A Myc-tagged irrelevant protein X or GFP alone served as negative controls , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 01210 . 7554/eLife . 17200 . 013Figure 4—figure supplement 1 . The pre-EJC regulates the precise splicing of dlg1 . ( A ) Aberrant dlg1 transcripts ( indicated by asterisks ) were detected by RT-PCR in RNA samples prepared from fly larvae expressing tsu RNAi driven by da-Gal4 . ( B ) Shown is the genomic structure of dlg1 that contains 23 coding exons ( blue boxes ) . The positions of primer pairs used to amplify RB , RH and RL splicing isoforms are indicated by red arrows . ( C ) Shown are RB , RH , RL splicing isoforms that are amplified by the above indicated primer pair . ( D–H ) Aberrant isoforms indicated by asterisks in panel A . were excised and purified from the agarose gel , cloned into a pGEM-T vector and subsequently subject to sequencing . After comparing the mRNA sequences deduced from sequencing results with wildype RB , RH and RL isoforms , three classes of aberrant splicing events were detected . The first class represents a previously reported exon skipping event , which was observed in all splicing forms ( D–G ) . The second class represents exon inclusion , which contains the exons that are normally efficiently spliced in wildtype isoforms ( green boxes ) . The third class includes exons that are generated by previously unidentified splice sites ( SS; red boxes ) . These three classes are not mutually exclusive as classes 1 and 2 were found in aberrant isoforms shown in panel E , classes 1 and 3 in panel F , and classes 1–3 in panel G . Δ indicates missing exons . 1’ indicates a 5’ SS in exon 1 . 16’’ indicates a 3’SS sites in exon 16 . 7’’’ and 23’’’ indicate 5’ and 3’ SS present in the exon 7 and 23 , respectively . Dashed lines indicate novel ligation of exons . Details on usage of previously unidentified splice sites are shown ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 01310 . 7554/eLife . 17200 . 014Figure 4—figure supplement 2 . The pre-EJC regulates the alternative splicing . ( A ) The previously identified 5’ splice sites ( SS ) and 3’ SS extracted from reference ( BDGP v5 . 78 ) were used 206 , 299 and 207 , 082 times in the tsu sample , respectively , while those used in the lacZ sample were 204 , 053 and 204 , 713 times . Thus , compared with the lacZ sample , the tsu sample has a 1 . 4% ( 2246/204 , 053 ) higher usage for previously identified 5’SS and a 1 . 5% ( 2269/204 , 713 ) for 3’SS , respectively . p-value for 5' splice site usage is 0 . 079 , and 0 . 057 for 3’ splice sites ( Student’s t-test ) . ( B ) Previously unidentified splice sites were detected in RNA-seq when the pre-EJC was knocked down . In total , 2 , 207 of 5’ SS and 2 , 081 of 3’ SS were identified , among which 394 of 5’ SS and 395 of 3’ SS were located in annotated exons in reference ( BDGP v5 . 78 ) , respectively . ( C ) Distribution plot highlights the correlation between maximum intron length and genes whose splicing were subject to differential regulation by the pre-EJC . Shown are distribution plots of genes whose splicing pattern was neither affected ( black line ) nor changed ( red line ) by reduced pre-EJC activity . The genes with altered splicing have an overall larger maximum intron length ( average length>1000 nt ) . This result is consistent with previous reports ( Ashton-Beaucage et al . , 2010; Roignant and Treisman , 2010 ) . In addition , a new class of genes was identified in which previously unidentified splice sites were utilized for generating novel transcripts ( green line ) . Interestingly , no correlation with overall larger maximum intron length was detected in this new class of genes ( green lines ) , implying an unknown mechanism for the pre-EJC to recognize splice sites . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 01410 . 7554/eLife . 17200 . 015Figure 4—figure supplement 3 . Dlg1 is a positive regulator of Wg signaling . ( A–D ) The expression of Sens was reduced in dlg114 mutant clones ( A , B , marked by the absence of GFP ) or when dlg1 RNAi was expressed by the hh-Gal4 in the posterior compartment of the wing disc ( C , D , marked by GFP ) . The expression pattern of Sens in a wildtype wing disc is shown in Figure 1—figure supplement 1D . ( E–H ) Overexpressing dlg1 ( F , H ) or dsh ( G ) largely rescued the loss of Wg signaling phenotype along the wing margin caused by knockdown of dlg1 ( E ) or pre-EJC ( H , cf . Figure 3E ) . Enlarged images of the wing margin from panels E–H are shown as E’–H’ . Arrows indicate sensory bristles along the wing margin . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 01510 . 7554/eLife . 17200 . 016Figure 4—figure supplement 4 . Splicing factor Rnps1 regulates Wg signaling . ( A , D , G ) Shown are the expression patterns of Dll ( A ) , Dsh ( D ) and Dlg1 ( G ) in wildtype wing discs . ( B , C , E , F , H , I ) The production of Dll ( C ) , Dsh ( F ) and Dlg1 ( I ) protein was reduced when Rnps1 RNAi was expressed by hh-Gal4 in the posterior compartment of the wing disc ( marked by GFP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 01610 . 7554/eLife . 17200 . 017Figure 4—figure supplement 5 . The interaction between Dlg1 and Dsh in S2 cells . ( A , B ) The interaction between Dlg1-PD ( A ) or Dlg1-PB ( B ) and Dsh was detected in S2 cells transiently transfected with HA-dlg1 and dsh-Myc by co-immunoprecipitation ( co-IP ) . Irrelevant proteins tagged with either HA or Myc served as negative controls for co-IPs , respectively . β-Tubulin served as a loading control . ( C ) Overexpressing dlg1-RB increased the abundance of Dsh protein in S2 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 01710 . 7554/eLife . 17200 . 018Figure 4—figure supplement 6 . The pre-EJC components Mago and Tsu regulate cell polarity in the wing disc . ( A , B ) Shown are the expression patterns of cell polarity proteins aPKC ( A ) and DE-Cad ( B ) in wildtype wing discs . ( C–J ) The expression patterns of aPKC ( C , D , G , H ) and DE-Cad ( E , F , I , J ) were altered in loss-of-function mago93D ( C–F ) or tsuΔ18 ( G–J ) somatic clones ( marked by the absence of GFP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 018 Dlg1 is important for the maintenance of cell polarity , acting together with L ( 2 ) gl to form a basolateral complex to counteract with the apical complex in epithelium ( St Johnston and Ahringer , 2010 ) . Therefore , it is not surprising that reduced pre-EJC activity led to altered cell polarity in wing disc epithelium ( Figure 4—figure supplement 6 ) . To investigate if the maintenance of cell polarity is required for Wg signaling , we examined the effects of L ( 2 ) gl as well as Bazooka ( Baz ) and Cdc42 ( Figure 5—figure supplement 1 ) , two apical complex members ( Harris and Peifer , 2004; Warner and Longmore , 2009 ) , in the Wg signal transduction . Surprisingly , loss of these polarity regulators in wing disc clones did not obviously affect Wg signaling ( Figure 5—figure supplement 2 ) , implying that the regulation of Wg signaling by Dlg1 may be independent of its function on cell polarity . Since cell polarity does not contribute significantly to Wg signaling , we reasoned that Dlg1 may directly interact with Dsh to regulate its activity . It has been reported that Drosophila Dsh binds in vitro to the K-K-x-x-x-Ψ motif within the I3-insert of the Hook domain in Dlg1 ( Garcia et al . , 2014 ) . We confirmed this interaction in S2 cells as well as in wing discs by immunoprecipitation ( Figure 4G , H; Figure 4—figure supplement 6A ) . The relevance of such interaction was further illustrated in rescue experiments in which overexpressed dsh was able to largely rescue the wing margin defects associated with reduced dlg1 activity ( Figure 4—figure supplement 3G ) . Next , we investigated how Dlg1 modulates Dsh to facilitate Wg signaling . Knocking down dlg1 by RNAi in S2 cells led to reduced production of Dsh protein , whilst overexpressing dlg1 had an opposite effect ( Figure 5A , B; Figure 4—figure supplement 5C ) . A similar result was observed in vivo when dlg1 expression was manipulated in wing discs ( Figure 5C–F ) . The effect of Dlg1 on Dsh production may be a direct consequence of altered protein stability . Dsh protein ectopically produced in S2 cells exhibited a short half-life of around two hours ( Figure 5—figure supplement 3 ) . The turnover of Dsh protein could be controlled through the ubiquitin-proteasome or lysosome mediated degradation . We found that the degradation of Dsh protein occurred mainly in lysosome when nascent protein synthesis was blocked by cycloheximide ( CHX ) ( Figure 5G ) . This is consistent with the observation that Dsh colocalized with early endosome and late endosome/lysosome markers in wing discs as well as in S2 cells ( Figure 5H–M; Figure 5—figure supplement 4 ) . Furthermore , heightened dlg1 expression counteracted with the CHX effect on Dsh degradation ( Figure 5N ) . In contrast , Dsh degradation potentiated by tsu or dlg1 RNAi could be prevented when S2 cells were treated with a lysosomal inhibitor chloroquine ( CQ ) ( Figure 5O , P ) . The above data indicate that the interaction between Dlg1 and Dsh protects Dsh protein from degradation in the lysosome . This conclusion was supported further by the observation that increased dlg1 expression correlated with reduced degree of ubiquitination modification on Dsh , a form of posttranslational modifications required for protein degradation ( Figure 5Q ) . 10 . 7554/eLife . 17200 . 019Figure 5 . Dlg1 regulates Dsh protein turnover . ( A ) Knocking down dlg1 by RNAi in S2 cells reduced the amount of Dsh protein . ( B ) Overexpressing dlg1 increased the abundance of Dsh protein in S2 cells . Note that Dlg1-PD form was used in all experiments unless mentioned otherwise . ( C–F ) The amount of Dsh was altered respectively when dlg1 RNAi ( D ) or dlg1-gfp ( F ) was expressed in wing discs by ap-Gal4 . ( G ) CHX treatment-induced Dsh protein degradation was inhibited by lysosome inhibitor chloroquine ( CQ ) but not by proteasome inhibitor MG132 ( MG ) . Cyclin B is known to be degraded in the proteasome , which served as a positive control for MG treatment ( Zhang et al . , 2014 ) . DMSO served as a negative control . ( H–M ) Dsh protein was detected in endocytic compartments in wing discs expressing Ubpy RNAi to prevent lysosome function . Over 10% of GFP-tagged Dsh ( dsh-gfp under the control of the dsh promoter ) colocalized with early endosome protein Rbsn-5 [H–J; n ( field of view ) = 8] and late endosome/lysosome protein marker LAMP1 in wing discs [K–M; n ( field of view ) = 10] . ( N ) CHX treatment-induced Dsh degradation was inhibited when dlg1 was overexpressed in S2 cells . ( O , P ) Dsh degradation resulted from tsu RNAi ( O ) or dlg1 RNAi ( P ) treatment was inhibited by lysosome inhibitor CQ . ( Q ) Overproduced Dlg1 reduced the level of ubiquitination of Dsh . Asterisk indicated non-specific signal of Myc antibody reactivity . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 01910 . 7554/eLife . 17200 . 020Figure 5—figure supplement 1 . Reduced activity of cell polarity determinants l ( 2 ) gl , baz or cdc42 result in polarity defects in wing disc cells . The localization of adherens junction , marked by DE-cadherin , was obviously changed in l ( 2 ) gl27S3 ( A–B ) , baz4 ( C–D ) or cdc424 ( E–F ) loss-of-function somatic clones ( marked by the absence of GFP ) in wing discs . Reconstituted optical cross sections along the Z-axis ( indicated by the yellow arrowheads ) are shown in A’–F’ . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 02010 . 7554/eLife . 17200 . 021Figure 5—figure supplement 2 . Reduced activity of cell polarity determinants l ( 2 ) gl , baz or cdc42 does not result in an obvious Wg signaling defect . The expression pattern of Sens ( B , F , J ) or Dsh ( D , H , L ) was not obviously changed in l ( 2 ) gl27S3 ( A–D ) , baz4 ( E–H ) or cdc424 ( I–L ) loss-of-function somatic clones ( marked by the absence of GFP ) induced in wing discs . Shown in insets ( B , F , J ) are enlarged images of regions marked by arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 02110 . 7554/eLife . 17200 . 022Figure 5—figure supplement 3 . The turnover of fly Dsh protein was measured in S2 cells treated with CHX followed by a four-hour chase . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 02210 . 7554/eLife . 17200 . 023Figure 5—figure supplement 4 . Dsh colocalizes with endosome and lysosome markers in S2 cells . S2 cells expressing dsh-Myc were treated with CQ for 8 hr to disrupt lysosome function . Over 60% of Dsh-Myc protein colocalized with early endosome protein Rbsn-5 [A–C , n ( number of cells ) = 10] and late endosome/lysosome marker LAMP1 [D–F , n ( number of cells ) = 10] . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 023 Consistent with a role of Dlg1 on Dsh stability in Drosophila wing development , regulated protein degradation serves as a control mechanism on Dvl protein homeostasis in cultured vertebrate cells ( Gao and Chen , 2010 ) . It is thus likely that Dlg1 orthologs could act in a similar manner to control Dvl protein stability in vertebrates . Human cells encode five orthologous Dlg proteins , Dlg1-5 . However , only Dlg2 contains an intact K-K-x-x-x-Ψ motif required for Dsh interaction as elucidated in Drosophila ( Figure 6A ) . Indeed , overproduced Dlg2 , but not Dlg3 , was sufficient to stabilize Dvl proteins in HEK293T cells ( Figure 6B , E , Figure 6—figure supplement 1A , B , Figure 6—figure supplement 2 ) . To further illustrate the importance of the K-K-x-x-x-Ψ motif of Dlg2 in stabilizing Dvl proteins , we generated a mutant form of Dlg2 ( hDlg2KKAA ) where two key lysine residues of the Hook domain were mutated to alanines . As predicted , this Dlg2 mutant failed to stabilize Dvl due to its inability to interact with Dvl ( Figure 6C , D ) . Dlg1 and Dlg2 share extensive homology within the Hook domain except that an amino acid is missing in Dlg1’s K-K-x-x-x-Ψ motif ( Figure 6A ) . We replaced this motif in Dlg2 with amino acids found in Dlg1 ( Figure 6—figure supplement 1C ) . The resulting Dlg2SFI-NL mutant showed reduced interaction with Dvl and no effect on Dvl protein stabilization ( Figure 6—figure supplement 1D , E ) . To further demonstrate functional consequence of Dlg2 regulation on Dvl protein stability in Wnt signaling , we performed TOPFlash Luciferase Wnt signaling reporter assay in HEK293T cells . Overexpressed wildtype but not the KKAA mutant form of dlg2 enhanced further the Dvl-induced TOPFlash reporter activity ( Figure 6F ) . The above experiments together indicate that the Dlg activity on Dvl protein homeostasis is conserved from flies to vertebrates . 10 . 7554/eLife . 17200 . 024Figure 6 . The regulation of Dvl proteins by Dlg is conserved in vertebrates . ( A ) Sequence alignment of the I3-insert of Hook domains presented in Dlg proteins reveals sequence conservation between fly Dlg1 and human Dlg orthologs . The K-K-x-x-x-Ψ motif is highlighted in a red box . Amino acids essential for Dsh interaction are shown in blue . ( B , C ) Overproduced wildtype ( B ) , but not the KKAA mutant form of hDlg2 ( C ) , increased the amount of Dvl proteins in HEK293T cells . ( D ) Dvl protein interacted with wildtype , but not the KKAA mutant form of hDlg2 . ( E ) CHX treatment-induced Dvl protein degradation was prevented by overproduced hDlg2 . ( F ) Overexpressing wildtype , but not the KKAA mutant form of hDlg2 , further increased Dvl2-induced TOPFlash Wnt signaling reporter activity . Ectopic Wnt3a served as a positive control . Experiments were repeated thrice , and data were represented as the mean+S . D . after normalized to Renilla activity ( **p<0 . 01; *p<0 . 05; Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 02410 . 7554/eLife . 17200 . 025Figure 6—figure supplement 1 . An intact K-K-x-x-x-Ψ motif is required for human Dlg proteins to stabilize Dvl2 in HEK293T cells . ( A ) Overexpressing hdlg3 in HEK293T cells did not increase the amount of hDvl2 proteins . ( B ) No interaction between hDlg3 and hDvl2 was detected by co-IP in HEK293T cells . ( C ) Shown is the alignment of the K-K-x-x-x-Ψ motives in human Dlg 1 and 2 . Two hDlg2 mutant forms , KKAA and SFI-NL , are also compared . ( D ) Overexpressing the SFI-NL mutant form of hdlg2 ( B ) did not result in increased hDvl2 production . ( E ) The SFI-NL mutant form of hDlg2 displayed reduced interaction with hDvl2 in HEK293T cells when compared with that of wildtype hDlg2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 02510 . 7554/eLife . 17200 . 026Figure 6—figure supplement 2 . Protein turnover of human Dvl2 protein was measured in HEK293T cells treated with CHX followed by a four-hour chase . DOI: http://dx . doi . org/10 . 7554/eLife . 17200 . 026
The EJC as well as other RNA binding protein complexes are thought to function in a pleiotropic manner . However , our data presented here together with several recent studies argue that RNA regulatory machineries can act specifically on developmental signaling for pattern formation and organogenesis . It has been increasingly recognized that the production , transport or the location of mRNA are subject to precise regulation in Wg/Wnt signaling . For example , apical localization of wg RNA is essential for signal activation in epithelial cells ( Simmonds et al . , 2001; Wilkie and Davis , 2001 ) , whilst the RNA binding protein RBM47 regulates Wnt signaling in zebrafish head development ( Guan et al . , 2013 ) as well as in cancer ( Vanharanta et al . , 2014; Venugopal et al . , 2015 ) . The specific role of RNA machineries on cell signaling is not limited to Wg/Wnt signaling . It has been reported that RNA-binding protein Quaking specifically binds to the 3’UTR of transcription factor gli2a mRNA to modulate Hedgehog signaling in zebrafish muscle development ( Lobbardi et al . , 2011 ) . RNA binding protein RBM5/6 and 10 could differentially control alternative splicing of a negative Notch regulator gene NUMB , thus antagonistically regulating the Notch signaling activity for cancer cell proliferation ( Bechara et al . , 2013 ) . Therefore , generally believed pleotropic RNA regulatory machineries emerge as important regulatory means to specifically control cell signaling and related developmental processes . The most studied function of the EJC in development is to localize oskar mRNA to the posterior pole of the oocyte for oocyte polarity establishment and germ cell formation in Drosophila ( Newmark and Boswell , 1994; Hachet and Ephrussi , 2001; Mohr et al . , 2001; van Eeden et al . , 2001; Palacios et al . , 2004 ) . Further study suggests that the proper oskar RNA localization relies on its mRNA splicing ( Hachet and Ephrussi , 2004 ) . In light of our study of the EJC activity on dlg1 mRNA as well as the roles of EJC on mapk and piwi splicing ( Ashton-Beaucage et al . , 2010; Roignant and Treisman , 2010; Hayashi et al . , 2014; Malone et al . , 2014 ) , we suspect that EJC might regulate oskar mRNA splicing to mediate its mRNA localization . Our RNA-seq analyses identified several hundreds of candidate mRNAs whose expression may be directly or indirectly subjected to EJC regulation . Apart from defects in Wg and MAPK signaling , however , we did not observe altered wing patterning associated with other developmental signaling systems in EJC defective flies , arguing that EJC may primarily regulate Wg and MAPK signaling in patterning the developing wing . Wg/Wnt signaling plays a fundamental role in development and tissue homeostasis in both flies and vertebrates . Its activation and maintenance rely on appropriate activity of the ternary receptor complex including Fz family proteins . In Drosophila , polarized localization of Fz and Fz2 proteins is essential for activation of non-canonical and canonical Wg signaling , respectively ( Boutros et al . , 2000; Wu et al . , 2004 ) . Dsh , which acts as a hub mediating both canonical and non-canonical Wg signaling , however , is found at both the apical cell boundary and in the basal side of the cytoplasm ( Kaplan and Tolwinski , 2010 ) . Thus , the polarized activity of Dsh must require distinct regulatory mechanisms at different sub-membrane compartments ( Mlodzik , 2016 ) . Our results provide the in vivo evidence suggesting that the basolateral polarity determinant Dlg1 may play a dominant role to control the Dsh abundance/activity in canonical Wg signaling . Altered Dvl production or activity has been linked with several forms of cancer ( Kafka et al . , 2014 ) . The stability of Dvl proteins can be controlled through regulated protein degradation both in vertebrates ( Gao and Chen , 2010 ) and in Drosophila as reported in this study . In HEK293T cells , Dapper1 induces whilst Myc-interacting zinc-finger protein 1 ( MIZ1 ) antagonizes autophagic degradation of Dvl2 in lysosome ( Zhang et al . , 2006; Huang et al . , 2015; Ma et al . , 2015 ) . It is also reported that a tumor suppressor CYLD deubiquitinase inhibits the ubiquitination of Dvl ( Tauriello et al . , 2010 ) . As Dlg1 prevents Dsh from degradation in Drosophila , it is important to investigate if Dlg1 participates in a posttranslational regulatory network of Dvl to integrate endocytosis and autophagy . Furthermore , upregulation of dvl2 and dlg2 expression has been found in various forms of cancer as shown in the COSMIC database ( Forbes et al . , 2015 ) . The study of the interaction between Dlg1 and Dsh may aid the development of novel approaches to prevent or treat relevant diseases . Dlg1 acts together with L ( 2 ) gl to form a basolateral complex in polarized epithelium . Dsh is known to interact with L ( 2 ) gl . On one hand , Dsh activity is required for correct localization of L ( 2 ) gl to establish apical-basal polarity in Xenopus ectoderm and Drosophila follicular epithelium ( Dollar et al . , 2005 ) . On the other hand , L ( 2 ) gl can regulate Dsh to maintain planar organization of the embryonic epidermis in Drosophila ( Kaplan and Tolwinski , 2010 ) . Despite the complex interaction between L ( 2 ) gl and Dsh , not much is known about mutual regulation between Dlg1 and Dsh . A recent report suggests that Dsh binds to Dlg1 to activate Guk Holder-dependent spindle positioning in Drosophila ( Garcia et al . , 2014 ) . Our results unveil another side of the relationship in which Dlg1 controls the turnover of Dsh to ensure developmental signal propagation . Apart from its apical localization at the cell boundary , Dsh is also found in the basal side of the cytoplasm ( Kaplan and Tolwinski , 2010 ) . It is likely that the interactions among Dsh , Dlg1 and L ( 2 ) gl may be dependent on their localization , and Dsh may serve as a bridge to connect cell signaling and polarity . Developmental signaling and cell polarity intertwine to control a diverse array of cellular events . It is well known that Wg/Wnt signaling controls cell polarity in distinct manner . Non-canonical signaling acts through cytoskeletal regulators to establish planar cell polarity ( Yang and Mlodzik , 2015 ) . Canonical signaling may also directly affect apical-basal cell polarity ( Karner et al . , 2006 ) . On the other hand , disruption of epithelial cell polarity has a profound impact on protein endocytosis and recycling ( Barbieri et al . , 2016 ) , both of which are essential regulatory steps for signal activation and maintenance ( Shivas et al . , 2010 ) . Our results add another layer of complexity by which polarity determinants could contribute to cell signaling independent of their conventional roles in polarity establishment and maintenance . Interestingly , this mode of regulation is also observed for other signaling processes . Loss of Dlg5 impairs Sonic hedgehog-induced Gli2 accumulation at the ciliary tip in mouse fibroblast cells that may not rely on cell polarity regulation ( Chong et al . , 2015 ) . Similarly , L ( 2 ) gl regulates Notch signaling via endocytosis , independent of its role in cell polarity ( Parsons et al . , 2014 ) . We believe that other cell polarity determinants may similarly participate in polarity-independent processes , however , the exact mechanism of how they cooperate to modulate developmental signaling awaits further investigation .
The following fly stocks were obtained from the Bloomington Drosophila Stock Center: ap-Gal4 , da-Gal4 , dpp-Gal4 , hh-Gal4 , T80-Gal4 , ts-Gal80 ( McGuire et al . , 2003 ) , vgBE-Gal4 , baz4 FRT9-2 ( #23229 ) , cdc424 FRT19A ( #9106 ) , dlg114 FRT101 ( #36283 ) , l ( 2 ) gl27S3 FRT40A ( #41561 ) , Dll-lacZ ( #10981 ) , wg-lacZ ( #1672 ) , UAS-dsh ( #9524 ) , UAS-eIF4AIII RNAi ( #32444 ) , UAS-mago RNAi ( #28931 ) , UAS-Rnps1 RNAi ( #36580 ) , and UAS-tsu RNAi ( #28955 ) . UAS-eIF4AIII RNAi ( #108580 ) , UAS-mago RNAi ( #28132 ) , UAS-tsu RNAi ( #107385 ) and UAS-Ubpy RNAi ( #107623 ) were obtained from the Vienna Drosophila RNAi Center ( VDRC ) . UAS-arr was a gift of Xinhua Lin , btz2 ( Palacios and St Johnston , 2002 ) was a gift of Daniel St Johnston , UAS-dlg1-gfp ( Zhang et al . , 2007 ) was a gift of Bingwei Lu , dsh-gfp ( Axelrod et al . , 1998 ) was a gift of Jeffrey Axelrod , esg-flp ( esg-Gal4 , UAS-flp; Chen et al . , 2005 ) and NRT-HA-wg ( Alexandre et al . , 2014 ) were gifts of Jean-Paul Vincent , UAS-fz2 ( Chen and Struhl , 1999 ) was a gift of Gary Struhl , fz3-lacZ ( Sato et al . , 1999 ) was a gift of Roel Nusse , FRT42D mago93D and FRT42D tsuΔ18 ( Roignant and Treisman , 2010 ) and FRT42D M ( 2 ) 58F ubi-gfp were gifts of Jessica Treisman . All fly crosses were maintained at 25°C unless noted otherwise . Detailed crossing schemes for each figure are shown in Supplementary file 1 . Loss-of-function somatic clones were induced in the wing disc by Flp/FRT-mediated homologous recombination . Second instar larvae from parental crosses were heat-shocked at 37°C for half an hour . Phenotypes of the adult wing and wing discs are all fully penetrant ( n>20 ) . For conventional immunofluorescence staining , wing discs dissected from third instar larvae were fixed in 4% paraformaldehyde , blocked in 0 . 2% BSA and incubated overnight at 4°C with the following primary antibodies: rabbit anti aPKC ( 1:500; sc-216; Santa Cruz Biotechnology , Dallas , TX ) , rabbit anti-Arr ( 1:15000; a gift of Stephen DiNardo; Rives et al . , 2006 ) , mouse anti β-galactosidase ( 1:50; 40-1a; Developmental Studies Hybridoma Bank , DSHB , Iowa City , IA ) , rabbit anti-β-galactosidase ( 1:4000; Cappel , Durham , NC ) , mouse anti-Dlg1 ( 1:50; 4F3; DSHB ) , mouse anti-Dll ( 1:400; a gift of Ian Duncan; Duncan et al . , 1998 ) , rat anti-DE-Cad ( 1:500; DCAD2; DSHB ) , rat anti-Dsh ( 1:500; a gift from Tadashi Uemura; Shimada et al . , 2001 ) , mouse anti-Fz2 ( 1:20; 12A7; DSHB ) , mouse anti-HA ( 1:100; 6E2 , Cell Signaling Technology , CST , Danvers , MA ) , rabbit anti-LAMP1 ( 1:500; ab30687; abcam , Cambridge , MA; Bouché et al . , 2016 ) , rabbit anti-Rbsn-5 ( 1:2000; a gift of Akira Nakamura; Tanaka and Nakamura , 2008 ) , guinea pig anti-Sens ( 1:1000; a gift of Hugo Bellen; Nolo et al . , 2000 ) and mouse anti-Wg ( 1:200; 4D4; DSHB ) . The wing discs were incubated with Alexa fluor-conjugated secondary antibodies ( 1:400; Invitrogen , Carlsbad , CA ) for one hour at room temperature before mounting . Fluorescence images were acquired with a Zeiss Axio Imager Z1 microscope equipped with an ApoTome or a Leica SP8 confocal microscope . The figures were assembled in Adobe Photoshop CS5 . Minor image adjustments ( brightness and/or contrast ) were performed in AxioVision 4 . 8 . 1 or Photoshop . Extracellular Wg staining was performed based on a previously described protocol with minor modification ( Strigini and Cohen , 2000 ) . Briefly , third instar larvae discs were dissected in ice-cold Schneider’s Drosophila medium ( Invitrogen ) supplemented with 10% FBS , 100 U/ml of penicillin and 100 mg/ml of streptomycin ( full medium ) and then incubated on ice with mouse anti-Wg antibody ( 1:10; 4D4 ) diluted in the full medium for one hour . Larval discs were rinsed and then fixed for 20 minutes in ice-cold PBS containing 4% paraformaldehyde before proceeding for immunofluorescence staining . To detect Dsh protein that undergoes endocytic degradation in wing discs in which dsh-gfp expression was controlled by the dsh promoter , Ubpy RNAi was used to disrupt ESCRT-0 complex that is required for delivery of internalized cargos for lysosomal degradation ( Zhang et al . , 2014 ) . Similarly , S2 cells overexpressing dsh-Myc was treated with chloroquine ( 10 mg/ml , Sigma , St . Louis , MO ) to disrupt lysosome function . Antibodies against Rbsn-5 and LAMP1 were used to label early endosomes and late endosomes/lysosomes , respectively . Myc-tagged fly dsh was generated by cloning the dsh-Myc fragment amplified from the UAS-dsh ( Bloomington #9524 ) transgenic strain into the NotI/XbaI site of the pUAST vector ( Brand and Perrimon , 1993 ) or a pCaSpeR-hs vector derived from pCaSpeR . dsh-Flag construct was generated by cutting full-length fly dsh cDNA from the pUAST vector and then subcloned into the EcoRI site of a pUAST-3×Flag vector . The HA-tagged fly dlg1 was generated by fusing an HA tag at the N-terminus of the full-length dlg1-RB or dlg1-RD cDNA and then cloned into the pUAST vector . HA-tagged human dvl1 , dvl2 and dvl3 were generated by fusing an HA tag at the C-terminus of the full-length human dvl1 , dvl2 and dvl3 cDNA , respectively , and the resulting HA-fusions were then cloned into a pcDNA3 . 1 vector . Human dlg with a Flag tag at the C-terminus was generated by cloning the full-length dlg2 or dlg3 cDNA into a pCMV3×Flag vector . This vector was also used to generate the dlg2KKAA or dlg2SFI-NL mutant expression vector by site directed mutagenesis ( Stratagene , La Jolla , CA ) . pAct-Myc-Ub plasmid was provided by Shunsuke Ishii ( Dai et al . , 2003 ) . Primers used in molecular cloning are listed in Supplementary file 1 . Drosophila Schneider S2 cells were cultured at 25°C in Schneider’s Drosophila full medium . HEK293T cells were grown in DMEM medium ( Invitrogen ) supplemented with 10% FBS , 100 µ/ml of penicillin and 100 mg/ml of streptomycin at 37°C . DNA transfection was carried out using a standard calcium phosphate protocol . In some experiments , S2 cells or HEK293T cells were treated for up to four hours with cycloheximide ( CHX; 50 µg/ml; Sigma ) before harvest to inhibit nascent protein synthesis . MG132 ( 50 µM; Sigma ) was used to inhibit the proteasome activity , while chloroquine ( 10 mg/ml ) was used to inhibit lysosome function ( Zhang et al . , 2014 ) . S2 cells transfected with hs-dsh and indicated vectors were heat shock for half an hour at 37°C after transfection for 48 hours . Then the cells were recovered at 25°C for one hour followed by drug treatment . dsRNA was generated with the MEGAscript high yield transcription kit ( Ambion , Austin , TX ) according to the manufacturer’s instruction . DNA template targeting tsu ( encoding amino acids 45–165 ) , mago ( encoding amino acids 31–140 ) , dlg1 ( encoding amino acids 800–917 ) and full length yeast Gal80 was generated by PCR and used for dsRNA synthesis . dsRNA targeting yeast Gal80 coding sequence was used as a negative control ( Su et al . , 2011 ) . For RNAi knockdown in S2 cells , dsRNA transfection was carried out using a standard calcium phosphate protocol . Primers used to generate dsRNAs are listed in Supplementary file 1 . HEK293T cells grown in 24 well plates were transfected with the TOPFlash luciferase reporter ( a gift of Yeguang Chen; Zhang et al . , 2006 ) and indicated vectors for two days before harvest . The pRL-TK Renilla reporter was co-transfected to normalize transfection efficiency . Luciferase activity was measured following the Dual-Glo luciferase assay protocol ( Promega , Madison , WI ) . S2 , HEK293T cells and 3rdinstar larval wing imaginal discs were lysed in NP-40 buffer ( 1% NP-40 , 150 mM NaCl and 50 mM Tris-Cl , pH 8 ) supplemented with protease inhibitor cocktail ( Roche , Germany ) . The concentration of protein cell lysate was quantified using a BCA protein assay kit ( Thermo , Waltham , MA ) . Immunoblotting was carried out using standard protocols . The following antibodies were used for immunoblotting: mouse anti-β-Tubulin ( 1:10000; E7 , DSHB ) , mouse anti-Cyclin B ( 1:50; F2F4; DSHB ) , mouse anti-Dlg1 ( 1:2000; 4F3; DSHB ) , rabbit anti-Flag tag ( 1:2000; D6W5B; CST ) , rabbit anti-HA tag ( 1:1000; Y-11; Santa Cruz ) , mouse anti-HA tag ( 1:2000; 6E2; CST ) , rabbit anti-MAPK ( 1:1000 , 137F5 , CST ) and mouse anti-Myc tag ( 1:2000; 9B11; CST ) . Immunoprecipitation was performed using agarose anti-HA ( Vector Labs , Burlingame , CA ) , agarose anti-Myc ( Vector Labs ) , agarose anti-GFP ( Vector Labs ) or Flag M2 affinity gel ( Sigma ) according to manufacturers’ instructions . Immunoblots presented in all figures are representatives of at least three independent experiments . Ubiquitination assays were carried out with hot lysis-extracted protein lysates based on the protocol described previously ( Row et al . , 2006; Zhang et al . , 2014 ) . Briefly , S2 cells transfected with dsh-Flag , HA-dlg and 6×Myc-Ub were hot-lysed in denaturing buffer ( 1% SDS , 50 mM Tris , pH 7 . 5 , 0 . 5 mM EDTA ) by boiling for five minutes . Lysates were then diluted 10-fold with NP-40 lysis buffer and subject to immunoprecipitation with anti-Flag M2 affinity gel ( Sigma ) . Total RNA of pooled third instar larvae or imaginal wing discs was extracted using TRIzol reagent ( Invitrogen ) . Residual genomic DNA was removed by RNase-free DNase ( New England Biolabs , NEB , Ipswich , MA ) . First strand cDNA was synthesized using oligo-dT primers and SuperScript III reverse transcriptase ( Invitrogen ) . Quantitative real-time PCR was performed using SYBR Green PCR master mix ( Applied Biosystems , Waltham , MA ) on a 7500 real time PCR system ( Applied Biosystems ) . Using α-Tubulin 84B as an internal control , relative fold changes of transcripts were calculated using comparative CT ( 2–ΔΔCT ) method . Three independent samples were prepared and run in triplicates . RT-PCR was performed to compare with the quantitative real-time PCR results . Primers used in quantitative real-time PCR and RT-PCR are listed in Supplementary file 1 . Total RNA from pooled third instar larval wing imaginal discs ( 1000 pairs per sample preparation ) expressing UAS-lacZ or UAS-tsu RNAi ( VDRC#107385 ) driven by T80-Gal4 was extracted in duplicates using TRIzol . Poly ( A ) + mRNAs were enriched using Dynabeads oligo ( dT ) beads ( NEB ) . RT reactions and purification of cDNA templates were performed following the RNA-seq sample preparation protocol from Illumina . Each cDNA sample was sequenced on an IlluminaHiseq 2500 system . Whole transcriptome reads were aligned using the TopHat ( v2 . 0 . 13 ) ( Trapnell et al . , 2012 ) with the Ensembl Drosophila_melanogaster . BDGP5 . 78 . gtf as a reference ( http://www . ensembl . org/index . html ) . In total , 22 , 563 , 753 and 24 , 660 , 199 of 125 bp reads pairs for duplicated tsu samples and 22 , 610 , 210 and 28 , 651 , 178 of 125 bp reads pairs for lacZ samples were sequenced , respectively . The transcription analysis was performed using Cufflinks ( v2 . 2 . 1 ) ( Trapnell et al . , 2012 ) . DEXseq was used to plot the transcripts to each gene ( Anders et al . , 2012 ) . RPKM method ( reads per kilobase of transcript per million mapped sequence reads; Mortazavi et al . , 2008 ) was used for normalizing gene counts . We calculated the ratio of RPKM between the tsu and lacZ samples . The R density plot package ( R Core Team , 2015 , R: A language and environment for statistical computing . R Foundation for Statistical Computing , Vienna , Austria . https://www . R-project . org// ) was used to generate the distribution plot shown in Figure 4—figure supplement 2C . The Seq data was deposited to NCBI website: http://www . ensembl . org/index . html http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? token=ujyhqauybbujhun&acc=GSE81220 . | Animal development involves different signaling pathways that coordinate complex behaviors of the cells , such as changes in cell number or cell shape . One such pathway involves a protein called Wingless/Wnt , which controls cell fate and growth and is also involved in tumor formation in humans . In recent decades , scientists have made a lot of progress in understanding how this signaling pathway operates . However , it is not well understood how the Wingless/Wnt signaling pathway interacts with other regulatory networks during development . Now , Liu , Li et al . unveil a new regulatory network that controls the Wingless/Wnt pathway in fruit flies and in mammalian cells grown in the laboratory . The experiments show that an RNA binding protein family named the Exon Junction Complex positively regulates a protein called Dishevelled , which serves as a hub in the Wingless/Wnt pathway . The Exon Junction Complex keeps the amount of Dishevelled protein in check via an interaction with another protein referred to as Discs large . Further experiments indicated that Discs large binds to and protects Dishevelled from being degraded inside the cell . Liu et al . 's findings highlight a new control mechanism for the Wingless/Wnt signaling pathway . In the future , the findings may also aid the development of new approaches to prevent or treat birth defects and cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2016 | The exon junction complex regulates the splicing of cell polarity gene dlg1 to control Wingless signaling in development |
Decisions are influenced by recent experience , but the neural basis for this phenomenon is not well understood . Here , we address this question in the context of action selection . We focused on activity in the pedunculopontine tegmental nucleus ( PPTg ) , a mesencephalic region that provides input to several nuclei in the action selection network , in well-trained mice selecting actions based on sensory cues and recent trial history . We found that , at the time of action selection , the activity of many PPTg neurons reflected the action on the previous trial and its outcome , and the strength of this activity predicted the upcoming choice . Further , inactivating the PPTg predictably decreased the influence of recent experience on action selection . These findings suggest that PPTg input to downstream motor regions , where it can be integrated with other relevant information , provides a simple mechanism for incorporating recent experience into the computations underlying action selection .
Selecting actions in a dynamic environment should take into account both sensory input and internally-generated estimates of action value . Integrating these sources of information is well-described by a Bayesian framework in which estimates of action value are continually updated by incoming sensory information in order to select the most valuable action ( Körding and Wolpert , 2006; Gold and Shadlen , 2007; Kim and Basso , 2010 ) . This updating of action values , based on experiencing the outcomes associated with past actions , is thought to be mediated primarily by striatal circuits ( Lau and Glimcher , 2007 , 2008; Histed et al . , 2009; Tai et al . , 2012; Kim et al . , 2013 ) . These action values can be maintained in striatal activity and ultimately used to bias activity in downstream motor centers such that the most valuable actions are more likely to be selected ( Hikosaka et al . , 2006 , 2014 ) . This system is capable of flexibly encoding action value estimates in arbitrarily complex and dynamic contexts over a range of time-scales . However , in environments in which only the recent past is relevant to action value , which is often the case in the real world , a simpler complementary mechanism would be to maintain short-term representations of the most recent actions and their outcomes that directly modulate the action selection process . We studied this possibility by examining behavior and neural activity in well-trained mice performing a task requiring them to select an action based on the dominant component of an odor mixture ( Uchida and Mainen , 2003 ) . We have previously shown that the superior colliculus ( SC ) plays a critical role in selecting the action – a leftward or rightward orienting movement – required by this task ( Felsen and Mainen , 2008 , 2012; Stubblefield et al . , 2013 ) , consistent with its role in selecting orienting movements in other species ( Glimcher and Sparks , 1992; Horwitz and Newsome , 2001; Bergeron et al . , 2003; Krauzlis et al . , 2004; Song et al . , 2011; Wolf et al . , 2015 ) . In this study , we asked whether the pedunculopontine tegmental nucleus ( PPTg ) , a mesencephalic sensorimotor hub that provides direct input to the SC ( Graybiel , 1978; Beninato and Spencer , 1986; Stubblefield et al . , 2015 ) , encodes information about recent actions and their outcomes by recording from individual neurons in behaving mice . While numerous regions provide input to the SC ( Sparks and Hartwich-Young , 1989 ) , many of which may modulate its processing underlying action selection ( Wolf et al . , 2015 ) , the PPTg holds particular interest because it is engaged by sensorimotor tasks across species ( Matsumura et al . , 1997; Dormont et al . , 1998; Kobayashi and Isa , 2002; Kobayashi et al . , 2002; Okada and Kobayashi , 2009; Norton et al . , 2011; Thompson and Felsen , 2013; Lau et al . , 2015 ) . We found that actions in this task are influenced by actions and outcomes in the recent past . Further , at the time of action selection and even throughout much of the trial , most PPTg neurons represented the choice ( left or right movement ) , outcome ( rewarded or non-rewarded ) , or both , on the previous trial . Furthermore , we found that these representations influenced action selection: the probability of particular upcoming choices was predictably related to the firing rates of neurons selective for choice on the previous trial , and pharmacological inactivation of the PPTg causally affected behavior , in part by decreasing the influence of recent choices on upcoming choices . Our results suggest a novel mechanism , subserved by the PPTg , for efficiently modulating action selection based on the recent history of actions and their outcomes .
Our overall goal was to address the role played by the PPTg within the interconnected network of brain regions responsible for selecting actions ( Gold and Shadlen , 2007; Wolf et al . , 2015 ) . Given its connectivity with the basal ganglia and input to the SC , we hypothesized that PPTg activity may represent recent information that could be relevant to the selection of upcoming choices . To test this hypothesis , we examined first whether behavior in our task was influenced by the choices ( left or right ) and outcomes ( rewarded or non-rewarded ) on previous trials , and next whether PPTg activity similarly reflected these previous choices and outcomes . Towards this end , we trained seven mice on a well-established odor-guided spatial choice task ( Uchida and Mainen , 2003; Thompson and Felsen , 2013 ) . In each trial of the task , the mouse samples a binary odor mixture at a central port , waits for a go signal , and reports which odor is dominant by moving to the left or right port for a water reward ( Figure 1A; see Materials and methods ) . We have previously used this task to show that the activity of a population of PPTg neurons tends to be higher preceding contralateral than ipsilateral orienting movements ( Thompson and Felsen , 2013 ) . Although optimal performance on this task requires that choices be based only on the current stimulus , previous studies – in mice , primates and humans – have shown that behavior in such tasks is , nevertheless , often influenced by previous trials ( Lau and Glimcher , 2005; Gold et al . , 2008; Busse et al . , 2011; Akaishi et al . , 2014 ) . We therefore examined how , in our task , upcoming choices depend on the choices and outcomes of previous trials by first examining psychometric functions conditional on previous trial history . As illustrated in the example data from one mouse in Figure 1B , C , the mouse tended to choose the left port more often when it had chosen the left port on the previous trial and had been rewarded ( Figure 1B , compare solid black line to gray line ) , and tended to choose the right port more often when it had chosen the right port on the previous trial whether or not it had been rewarded ( Figure 1C , compare solid and dashed black lines to gray line ) . To quantify the influence of previous trials on choices , separately for each mouse ( 12 sessions per mouse; 451 ± 115 ( mean ± STD ) trials per session ) , we employed a logistic regression model with terms for previous choice ( β Choice1 ) and outcome ( β Outcome: L1 and β Outcome: R1 ) ( Extended model , 1 trial back; see Materials and methods ) . The sign of β Choice1 reflects the tendency to stay with the same ( positive ) or switch to the opposite ( negative ) choice as on the previous trial , while the sign of β Outcome: L1 and β Outcome: R1 reflects the tendency to choose left ( positive ) or right ( negative ) when the previous choice was rewarded at the side indicated in the subscript . Consistent with the example data shown in Figure 1B , C , and with some previous work ( Lau and Glimcher , 2005 ) , we found that , while there was some variability across mice , they tended to choose the same side as they had on the previous trial ( reflected in positive values of β Choice1 ) [although in different tasks , a tendency to choose the opposite side has been observed ( Lau and Glimcher , 2005; Kim et al . , 2007; Sul et al . , 2011 ) ] , and that this tendency was enhanced when the mouse was rewarded on the previous trial ( reflected in positive values of β Outcome: L1 and negative values of β Outcome: R1; Figure 1D ) . In order to determine whether including these terms improved our behavioral model , we compared the accuracy of the Extended model , 1 trial back to the accuracy of the Simple model at predicting choices ( Materials and methods ) . We found that including the choice and outcome on the previous trial improved the predictive accuracy of the model for five of seven mice ( Figure 1E , p<0 . 0001 , one-tailed paired Student’s t test separately by mouse ) . However , including the choice and outcome from two trials back did not further improve the predictive accuracy of the model in any mice ( Figure 1F , p>0 . 16 , one-tailed paired Student’s t test ) . 10 . 7554/eLife . 16572 . 003Figure 1 . Behavior is influenced by previous trials . ( A ) Timing of behavioral events for two consecutive trials ( ‘previous’ and ‘current’ ) of the spatial choice task . Gray box shows pre-stimulus epoch , the primary focus of our neurophysiological analyses . ( B and C ) Behavioral performance conditional on whether the choice on the previous trial was left ( B ) or right ( C ) , and rewarded or unrewarded , for 1 example mouse ( 12 sessions ) . Gray circles and lines include all trials and are identical in B and C , shown for comparison with the conditional data ( black ) . Lines show best-fit logistic functions using the Simple model . ( D ) Influence of previous choice and outcome on choice behavior for 7 mice ( M1-M7 ) , estimated with the Extended model , 1 trial back . Error bars , 95% confidence intervals . ( E and F ) Improvement in predicting choices realized by including in the behavioral model choices and outcomes 1 trial back ( E ) and 2 trials back ( F ) for all seven mice ( corresponding to M1-M7 in D ) . Each symbol represents 1 session . For each session , simulated choices on a test set of trials ( 50% of trials in the session ) were separately generated by three models ( Simple model; Extended model , 1 trial back; and Extended model , 2 trial back ) in which all regression coefficients were estimated from the remaining trials ( in all sessions ) for that mouse . The accuracy of each model’s prediction of choices in the test set was calculated as the percentage of trials in which the predicted choice matched the actual choice . This process was repeated , with a new test set , 50 times/session . Each symbol shows the average , across repeats for each session , of the improvement in accuracy of the Extended model , 1 trial back over the Simple model ( E ) , and of the Extended model , 2 trials back over the Extended model , 1 trial back ( F ) . *p<0 . 0001 , one-tailed paired Student’s t test across sessions per mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 16572 . 003 In principle , the influence of the previous choice on behavior ( Figure 1E ) could be mediated in part by body position . For example , if the orientation at which the mouse enters the odor port depends on its previous choice , its subsequent choice behavior may be biased . If this were the case , we reasoned that would likely , although not necessarily , observe that movement duration ( from odor port exit to reward port entry ) would depend on the previous choice . However , we found that this was not the case ( durations for rightward movements preceded by right choices and preceded by left choices , analyzed separately for each mouse , did not differ in 7/7 mice; durations for leftward movements preceded by right choices and preceded by left choices did not differ in 6/7 mice; p>0 . 05 , two-tailed unpaired Students t-tests separately by mouse ) . Therefore , although it is not possible to entirely rule out a role for the position of the body or any of its parts ( e . g . , whiskers ) in contributing to our observed effect , as we did not measure all dimensions of body position , these results reduce the likelihood that choices depend on patterns of body position arising from the previous choice . Together , these results indicate that choices in our task are influenced by what happened on the previous trial – primarily by the choice rather than the outcome – and little influenced by the choice or the outcome from two trials in the past . Having established the relevance of the previous trial for behavior , we next examined PPTg activity recorded in the same seven mice while they performed the task ( see Materials and methods ) . We first asked whether PPTg activity reflects the choice or outcome of the current trial . Figure 2A shows rasters and peristimulus time histograms ( PSTHs ) for a representative PPTg neuron aligned to the time of odor valve open and grouped by the choice on the current trial , and Figure 2B shows the same data grouped by outcome on the current trial . Although the firing rate increased around the time of odor valve open , it does not appear to depend on either the choice ( Figure 2A ) or outcome ( Figure 2B ) on the current trial . To quantify dependence on choice across the population of 506 PPTg neurons recorded , we used an ROC-based preference analysis that uses the firing rates in a particular epoch to assign values ranging from −1 ( ipsilateral preference ) to 1 ( contralateral preference ) , where the magnitude reflects preference strength ( see Materials and methods ) . We were initially interested in preference during the pre-stimulus epoch – from odor port entry to 100 ms after odor valve open ( Figure 1A , gray shading; Figure 2A , gray bar ) – when the mouse is presumably preparing to update its estimates of the value of each option ( i . e . , to move left or right ) with the information provided by the stimulus . We found that few neurons exhibited a preference for choice during this epoch ( Figure 2 , p<0 . 05 , Monte Carlo permutation test ) . We then used a similar preference analysis to quantify dependence on outcome during this same epoch , where negative values indicate preference for no-reward and positive values indicate preference for reward ( see Materials and methods ) . We found that few neurons exhibited outcome preference ( Figure 2 , p<0 . 05 ) . It is not surprising that few neurons exhibit choice or outcome preference during this epoch , because the mouse presumably does not commit to a choice and cannot predict the outcome of the trial , before receiving the stimulus . 10 . 7554/eLife . 16572 . 004Figure 2 . PPTg activity during the pre-stimulus epoch is influenced by choices and outcomes on previous trials . ( A and B ) Rasters ( top ) and PSTHs ( bottom ) for one example neuron grouped by choice ( A ) and outcome ( B ) on the current trial . Fifty pseudorandomly selected ipsilateral and contralateral trials ( A ) and reward and no-reward trials ( B ) are shown in the rasters; all trials are included in the PSTHs . Activity is aligned to the time of odor valve open ( blue lines ) and sorted within each group by the time since odor port entry ( orange ticks ) . PSTHs show average firing rate across all trials in each group , smoothed with a Gaussian filter ( σ = 15 ms ) . Shading , ± s . e . m . Gray bar shows mean pre-stimulus epoch . ( C and D ) Preference during the pre-stimulus epoch for choice ( C ) and outcome ( D ) on the current trial across the population of neurons . Arrowheads indicate preferences for example neuron shown in A and B . ( E , F , G and H ) As in A , B , C and D with respect to 1 trial back instead of current trial . Data from same example neuron are shown . ( I , J , K and L ) As in A , B , C and D with respect to two trials back instead of current trial . Data from same example neuron are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16572 . 004 We next examined whether PPTg activity depended on the choice and outcome on the previous trial , which have been shown to influence behavior ( Figure 1 ) . Figure 2E , F shows data from the same neuron as in Figure 2A , B but with trials grouped by the choice ( Figure 2E ) or outcome ( Figure 2F ) on the previous , as opposed to the current , trial . In contrast to the lack of dependence on the choice or outcome on the current trial ( Figure 2A , B ) , there appears to be a noticeable difference in firing rate around the time the odor valve opens depending on the choice or outcome on the previous trial ( Figure 2E , F ) . We quantified this dependence across the population using the same preference analysis previously described , but with trials grouped according to the choice ( Figure 2G ) or outcome ( Figure 2H ) on the previous trial . Across the population , a sizeable fraction of neurons exhibited a preference for the choice ( Figure 2G; 26% preferred ipsilateral , 49% preferred contralateral , p<0 . 05; more preferred contralateral than ipsilateral , p=0 . 001 , χ2 test ) and outcome ( Figure 2H; 46% were selective , p<0 . 05 ) on the previous trial . We found a modest correlation between preference for the choice on the previous ( Figure 2G ) and current ( Figure 2C ) trials ( r = 0 . 25 , p=1 . 42 × 10−8 ) . In principle , the dependence of PPTg activity on previous choice could actually reflect a dependence on the direction of movement from the chosen reward port back to the odor port . While either representation would be useful for conveying information about previous choice , to determine whether PPTg activity reflects choice per se or movement direction , we examined how activity during movement to and from the reward port ( which are in opposite directions ) depended on which reward port was selected . We found that , rather than representing movement direction , two-thirds of neurons maintained their preference for which reward port was chosen ( p=9 . 12 × 10−6 , χ2 test ) , suggesting that the preference of PPTg neurons for previous choice ( Figure 2G ) cannot be accounted for by a dependence on movement direction . We then calculated choice and outcome preference with respect to two trials back ( Figure 2I–L ) and found that , as was the case for the current trial , few neurons exhibited a preference for choice ( Figure 2K ) or outcome ( Figure 2L ) , consistent with the relatively little influence on behavior of the choice and outcome from two trials in the past ( Figure 1F ) . Thus , a larger proportion of neurons exhibited a preference for choice or outcome one trial back than on either the current trial or two trials back ( choice , p=2 . 0 × 10–48; outcome , p=3 . 0 × 10–5 , Kruskal-Wallis ANOVA; post-hoc comparisons: choice , one trial back vs . current trial , p=5 . 6 × 10–30; one trial back vs . two trials back , p=1 . 6 × 10–30; outcome , one trial back vs . current trial , p=4 . 0 × 10–4; one trial back vs . two trials back , p=0 . 002 ) , and there were no differences between the fractions of neurons exhibiting a preference for choice or outcome on the current trial and two trials back ( choice , two trials back vs . current trial , p=0 . 92; outcome , two trials back vs . current trial , p = 0 . 99 ) . These results suggest that PPTg activity during the pre-stimulus epoch is primarily modulated by the choice and outcome on the previous trial ( although our analyses again cannot entirely rule out the possibility that body position also modulates neural activity ) . Previous work using the same task has shown that PPTg activity in other epochs reflects the choice and outcome on the current trial ( Thompson and Felsen , 2013 ) , consistent with other studies of PPTg function ( Matsumura et al . , 1997; Dormont et al . , 1998; Kobayashi et al . , 2002; Norton et al . , 2011; Maclaren et al . , 2013 ) . Together , these results raise the question of how neural activity evolves over the course of the trial from representing previous trial information during the pre-stimulus epoch ( Figure 2G , H ) to current trial information in later epochs . To quantify these temporal dynamics , we performed a linear regression analysis investigating how PPTg activity is influenced , over time , by four factors: previous choice ( β Choice1 ) , current choice ( β Choice0 ) , previous outcome ( β Outcome1 ) and current outcome ( β Outcome0 ) [Materials and methods; we did not include terms for two trials back because this trial had little influence on PPTg activity ( Figure 2K , L ) ] . Figure 3A shows these βs as a function of time , for the same example PPTg neuron shown in Figure 2 . Consistent with the PSTHs shown in Figure 2 , around the time the mouse enters the odor port ( which precedes odor valve open by about 100 ms ) , the activity of this neuron began to be influenced by previous choice and previous outcome , but not by current choice and current outcome ( Figure 3 , left ) . The influence of the previous trial declines by the time the mouse exits the odor port ( Figure 3A , middle ) , and the influence of the current choice increases upon reward port entry ( Figure 3A , right ) , consistent with the fact that many neurons exhibit preference for the current choice during and following entrance to the reward port ( Thompson and Felsen , 2013 ) . 10 . 7554/eLife . 16572 . 005Figure 3 . Dynamics of the influence of the choice and outcome on the previous and current trial on PPTg activity throughout the trial . ( A ) For the example neuron shown in Figure 3 , 95% confidence intervals are shown for each regression coefficient calculated in 100 ms bins ( shifted by 10 ms ) , aligned to three trial events . ( B ) Average regression coefficients across all neurons that exhibited a significant preference for choice on the previous trial . Ribbons reflect mean ± SEM β Choice1 ( black ) β Choice0 ( red ) coefficient values as a function of time . ( C ) Fraction of neurons with activity in each 100 ms bin significantly influenced by choice on the previous or current trial , aligned as in A . ( D ) As in B , with respect to outcome instead of choice . ( E ) As in C , with respect to outcome instead of choice . DOI: http://dx . doi . org/10 . 7554/eLife . 16572 . 005 We performed this regression analysis on each neuron that exhibited a significant preference for choice on the previous trial ( corresponding to the black bars in Figure 2G; 378/506 total neurons ) . Figure 3B and C show , over the course of the trial , the mean β Choice1 and β Choice0 , as well as the fraction of neurons with activity influenced by the choice on the previous ( solid line ) and current ( dashed line ) trial . Initially – even preceding odor delivery – the activity of many neurons reflects the choice on the previous trial , while the activity of virtually no neurons reflects the choice on the current trial ( Figure 3C , left ) . During stimulus presentation many neurons begin to reflect the choice on the current trial , and by the time the movement is initiated more neurons reflect the choice on the current than the previous trial ( Figure 3C , middle ) . Interestingly , the sign of β Choice1 of many neurons briefly inverted immediately preceding movement initiation ( Figure 3B ) , consistent with the dip in the fraction of significant neurons at this time ( Figure 3C ) . The choice on the current trial retains its influence for the remainder of the trial while the influence of the choice on the previous trial declines ( Figure 3C , right ) , although , interestingly , a sizeable fraction of neurons continues to reflect the choice on the previous trial until well-after the mouse has made its choice on the current trial . We also examined , for the same population shown in Figure 3B , C , how the fraction of neurons with activity influenced by the outcome on the previous trial ( solid line ) and current trial ( dashed line ) changes over the course of the trial ( Figure 3D , E ) . Similar to the evolving representation of choice , the activity of more neurons is more strongly influenced by previous trial outcome early in the trial ( Figure 3D , E , left ) , and by current trial outcome later in the trial ( Figure 3D , E , right ) . In contrast to the long-lasting representation of previous choice ( Figure 3B , C ) , virtually none of these neurons represent previous outcome by the time the mouse initiates its movement to the reward port ( Figure 3D , E , middle; however , note that the population of neurons analyzed was selected based on preference for previous choice , and not previous outcome , during the pre-stimulus epoch ) . These analyses show how PPTg activity evolves from representing information about the previous trial to representing information about the current trial , which may reflect the transfer of previous-trial information from the PPTg to a downstream region where it can be integrated with sensory evidence in order to form a motor plan . In order to address whether the PPTg activity representing previous trial information is used to guide behavior , we next examined whether this activity influences the choice on the current trial . For simplicity we focused on PPTg activity representing previous choice , and not outcome , because we found that the former representation was more robust ( Figure 2G , H ) . If this activity indeed contributes to behavior – e . g . , by providing information about the value of each option – we would predict a trial-by-trial correlation between the activity of individual neurons that exhibit a preference for previous choice and the likelihood of the mouse choosing the corresponding reward port , particularly on trials in which the sensory cue is ambiguous . Specifically , on trials in which a given neuron exhibits higher activity during the pre-stimulus epoch , we would expect the mouse to be more likely to choose the reward port corresponding to the preferred choice ( ipsilateral or contralateral ) of that neuron . Therefore , for each neuron exhibiting a significant preference for choice on the previous trial ( corresponding to the black bars in Figure 2G; 378/506 total neurons ) , we examined how choice on ambiguous trials ( % Left odor = 40 , 50 , or 60; Figure 1B ) depended on firing rate during the pre-stimulus epoch ( Figure 4A ) . Specifically , for each trial of the session we calculated the firing rate ( normalized to its maximum across trials , separately for each previous choice condition ) and classified it by whether the mouse chose the preferred or antipreferred choice of the neuron , quantified as 1 and 0 , respectively . We then calculated the slope of the best-fit line through these points , each of which correspond to a trial . For a given neuron , a positive slope indicates that the mouse is more likely to choose the reward port corresponding to the preferred choice of the neuron on trials in which its activity is high during the pre-stimulus epoch , as we had predicted . Across this population of neurons , many more exhibited a significantly positive slope than negative slope ( p<0 . 05 , Monte Carlo permutation test; positive slopes: 136/378; negative slopes: 30/378; positive > negative , p=0 . 001 , χ2 test; Figure 4B ) , indicating a similar relationship between firing rate and current choice as shown in Figure 4A . However , since firing rate and current choice both depend on previous choice ( Figuers 1 and 2 ) , it is possible that this relationship is indirect . To determine whether this was the case , we examined the distribution of slopes separately for each mouse and found that each exhibited more neurons with positive than negative slopes ( Figure 4C ) , even though the behavior of some mice did not depend on previous choice ( Figure 1E ) . These results suggest that the representation of the choice on the previous trial ( Figure 2 ) can directly influence the choice made by the mouse on the current trial . 10 . 7554/eLife . 16572 . 006Figure 4 . PPTg representation of previous choice affects behavior . ( A ) Probability that the mouse chose the reward port corresponding to the previous-trial choice preference of an example neuron , plotted as a function of the firing rate of that neuron during the pre-stimulus epoch . Firing rates were normalized to the maximum for the neuron , separately within each previous choice condition . Only current trials with an ambiguous sensory cue are shown . Dashed line , best linear fit to raw data ( antipreferred choice , y = 0; preferred choice , y = 1 ) . Circles represent mean ± s . e . m . fraction preferred choices for binned normalized firing rates and are shown for display only ( not used to fit the line ) . ( B ) Slope of best-fit line calculated as in A for all neurons exhibiting a significant preference for choice on the previous trial during the pre-stimulus epoch . Slopes were determined to be significantly different from 0 ( p<0 . 05 ) with a Monte Carlo permutation test with 1000 repeats . Significance of slopes was identical when the data were fit to a logistic function rather than a line as shown in A . Arrowhead indicates slope for example neuron shown in A . ( C ) Fraction of significantly positive ( white ) and negative ( black ) slopes , and non-significant ( gray ) slopes ( calculated as in B ) , separately for each mouse , numbered as in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16572 . 006 To test the causality of this relationship , we next asked whether inactivating the PPTg , with muscimol ( a GABAA agonist ) would change the degree to which the previous choice influences behavior . We first examined the direct effect on behavior of unilateral PPTg inactivation ( see Materials and methods ) . We found that choices were biased ipsilateral to the inactivated PPTg , as compared to the preceding and following control sessions in which saline was infused to the same PPTg ( Figure 5A ) . We quantified this effect across all 30 sets of sessions [saline ( pre ) , muscimol , and saline ( post ) ] from three mice by estimating the influence of muscimol on choice ( represented by βMuscimol; see Materials and methods ) . Positive values of βMuscimol correspond to an ipsilateral influence and negative values correspond to a contralateral influence . We found that unilaterally inactivating the PPTg resulted in a modest ipsilateral influence on choices ( Figure 5B; p<0 . 05 , two-tailed unpaired Student’s t test , pooled across 3 animals and 30 sets of sessions ( 424 ± 105 trials; mean ± STD ) ; βMuscimol > 0 in 14/30 individual sets of sessions , βMuscimol < 0 in 6/30 individual sets of sessions , p<0 . 05 , two-tailed one-sample Student’s t test; black bars ) , indicating a causal relationship between PPTg activity and contralateral movements in this task . Consistent with the modest but significant effect of unilateral inactivation on choice bias , we found that movement latency and duration – both to and from the reward port – were longer in muscimol than in saline sessions , even when contralateral and ipsilateral choices were pooled ( Materials and methods; p=4 . 55 × 10−20 , 5 . 24 × 10−32 , and 0 . 0026 , respectively , two-tailed unpaired Student’s t tests ) . 10 . 7554/eLife . 16572 . 007Figure 5 . Inactivating the PPTg causally affects behavior . ( A ) Behavioral performance during example sessions after which either muscimol ( black circles ) or 0 . 9% saline ( gray circles ) was infused into the left PPTg . Saline was infused during the session before ( gray circles , solid gray line ) and after ( gray triangles , dashed gray line ) the muscimol session . One session was performed per day . Lines show best-fit logistic functions using the Simple model . Error bars , ± s . d . Behavioral data shown here and in all subsequent figures were collected from well-trained mice ( Materials and methods ) . ( B ) βMuscimol calculated for all 3 mice and 30 sets of sessions . Positive values represent leftward ( ipsilateral ) influence of muscimol . Arrowhead indicates βMuscimol for example sessions shown in A . ( C ) Behavioral influence of the previous choice calculated separately for muscimol and saline sessions ( same sessions shown in B ) . *p=0 . 021 , one-tailed Mann-Whitney U test . DOI: http://dx . doi . org/10 . 7554/eLife . 16572 . 007 Finally , we wondered whether the effect on behavior of inactivating PPTg could be due to a decreased dependence of current choice on previous choice . We tested this idea by estimating β Choice1 separately for the muscimol and saline sessions and found that β Choice1 was indeed lower when PPTg activity was inhibited ( Figure 5C , p=0 . 021 , one-tailed Mann-Whitney U test ) , perhaps accounting for the modest effect on choices shown in Figure 5B [with respect to the other terms in the model , we found that βOdor: L and βOdor: R did not differ between muscimol and saline sessions ( p=0 . 44 and p=0 . 28 , respectively , two-tailed Mann-Whitney U tests ) , but that β0 was higher in muscimol than saline sessions ( p=0 . 040 , two-tailed Mann-Whitney U test ) , indicating an odor-independent leftward bias consistent with Figure 5B] . These results support our hypothesis that the PPTg activity representing previous experience ( Figure 2 ) influences action selection .
This study examined the neural basis by which action selection is influenced by recent experience . We first found that behavior in a stimulus-cued spatial-choice task can be accounted for by a Bayesian framework ( among others ) in which choices are influenced by both recent trial history and the sensory stimulus ( Figure 1 ) , consistent with previous studies ( Gold et al . , 2008; Busse et al . , 2011; Akaishi et al . , 2014 ) . We then found that activity of a subpopulation of PPTg neurons reflects choices and outcomes in the recent past ( Figures 2 and 3 ) and correlated with upcoming choice ( Figure 4 ) , suggesting a role in mediating the influence of recent experience on behavior . Furthermore , inactivating PPTg decreased the influence of recent experience on behavior ( Figure 5C ) . Together , our results demonstrate that the PPTg encodes representations of recent experience that can contribute to action selection . Below , we consider our findings in light of the evolving view of the role of the PPTg within the interconnected network of brain regions involved in integrating priors with sensory evidence in order to select and control motor output ( Mena-Segovia et al . , 2004; Gold and Shadlen , 2007 ) . The motor-related role of the PPTg has traditionally been considered in the context of central pattern generation ( Garcia-Rill , 1991 ) . More recent studies in behaving animals have found that PPTg activity encodes specific actions and reflects their outcomes ( Matsumura et al . , 1997; Dormont et al . , 1998; Okada and Kobayashi , 2009; Norton et al . , 2011; Maclaren et al . , 2013; Thompson and Felsen , 2013 ) and have suggested that the PPTg may be involved in attention and other cognitive processes ( Steckler et al . , 1994; Winn , 2008 ) . The results described here are consistent with these findings and suggest that PPTg output may play a role in providing information about recent actions and their outcomes that can be used to guide the selection of actions mediated by downstream motor circuits . However , downstream circuits do not necessarily utilize this information , given that it is represented in PPTg activity even in mice that do not exhibit a behavioral dependence on previous choice ( Figure 1D ) . To where might PPTg representations of recent experience be conveyed in order to modulate action selection ? While the PPTg projects to several motor-related regions including other brainstem nuclei and the striatum ( Beninato and Spencer , 1987; Inglis and Winn , 1995 ) , and the current study cannot rule out the involvement of these regions , an attractive candidate is the SC . Cholinergic PPTg neurons project to and excite neurons in the SC responsible for motor output ( Beninato and Spencer , 1986; Sooksawate et al . , 2008; Stubblefield et al . , 2015 ) , and nicotinic signaling in the SC has been shown to modulate orienting behavior ( Weldon et al . , 1983; Aizawa et al . , 1999; Watanabe et al . , 2005 ) . The SC is thought to integrate a wide range of inputs in order to select orienting actions ( Kobayashi and Isa , 2002; Krauzlis et al . , 2004; Grossberg et al . , 2015; Wolf et al . , 2015 ) and is thus well-positioned to combine – possibly additively – incoming sensory information with prior representations of action value ( provided before , during , or even after the onset of sensory input; Figure 3B , C ) in order to select the most valuable action ( Trappenberg et al . , 2001; Dorris et al . , 2007; Kim and Basso , 2010 ) . Our results suggest that the source of these prior representations may be the PPTg , which could complement other SC inputs – such as inhibition from the basal ganglia ( Hikosaka and Wurtz , 1983; Chevalier et al . , 1985 ) – in biasing action selection by modulating SC processing according to recently-experienced action-value associations ( Kobayashi and Isa , 2002; Hikosaka et al . , 2006; Wolf et al . , 2015 ) . While we have found it useful to think of the input from the PPTg to the SC as representing priors , in a Bayesian sense , accepting this framework – which may be imperfect given that choices in our task should not depend on previous trials – is not necessary for interpreting our results . Given the importance of trial history for action selection , it is not surprising that representations of trial history similar to that shown here have been observed in several brain areas , including the striatum ( Lau and Glimcher , 2007 , 2008; Histed et al . , 2009; Kim et al . , 2013 ) , prefrontal cortex ( Histed et al . , 2009 ) and premotor cortex ( Marcos et al . , 2013 ) . Activity in these regions typically reflects information from several previous trials , while PPTg activity was primarily associated with the immediately preceding trial ( Figure 2 ) . While this difference may be due to different task demands ( recall that tracking past trials conferred no behavioral benefit in our task ) , the representations in the PPTg that we observed suggest a complementary mechanism to the more computationally expensive processing in these other regions for integrating internal representations of experience with sensory evidence: Specifically , that PPTg directly influences the circuits underlying action selection based on recent experience , which may be well-suited to the relevant dynamics of some real-world situations . Future studies can expand upon our findings by recording from specific types of neurons ( e . g . , cholinergic ) in the PPTg ( Lima et al . , 2009; Cohen et al . , 2012; Roseberry et al . , 2016 ) , as well as by examining how associations between stimuli and reward location are initially learned , in order to further elucidate the function of the PPTg in selecting actions .
All experiments were performed according to protocols approved by the University of Colorado School of Medicine Institutional Animal Care and Use Committee . Mice were bred in the animal facilities of the University of Colorado Anschutz Medical Campus or purchased ( Jackson Labs ) . We used 10 male adult mice , aged 124–186 days at the start of experiments . Pharmacology experiments were performed in three C57BL/6J mice , and electrophysiological experiments were performed in three C57BL/6J and four ChAT-Cre mice ( Jackson Labs , strain B6; 129S6-Chattm2 ( cre ) Lowl ) ( we observed no differences across backgrounds and data are combined here ) . Mice were housed singly in a vivarium with a 12-hr light/dark cycle with lights on at 5:00 am . Food ( Teklad Global Rodent Diet No . 2918; Harlan ) was available ad libitum . Access to water was restricted to the behavioral session ( see below ) unless less than ~1 ml was received , in which case free water was provided for 2–5 min following the session ( Thompson and Felsen , 2013 ) . Mice were trained on an odor-guided spatial choice task ( Uchida and Mainen , 2003 ) as described in detail in Thompson and Felsen ( 2013 ) . Briefly , in each trial of the task , the mouse waited for a central port to be illuminated , entered the port , waited 144 ± 64 ms ( mean ± s . d . ) for the odor valve to open , sampled a binary odor mixture , waited 434 ± 68 ms ( mean ± s . d . ) for a go signal ( simultaneous port light off and tone presentation; 5kHz , ~85dB ) , exited the odor port , and moved toward the left or right reward port [Figure 1A; these delays were selected in order to temporally segregate behavioral events for determining how they were correlated with neural activity , minimize training time , and for consistency with our previous work ( Thompson and Felsen , 2013 ) ] . In this previous work , we used a photo-ionization-detector to estimate the latency between odor valve open and the odor first arriving at the port to be 75–100 ms ( Thompson and Felsen , 2013 ) . Exiting the odor port prior to the go signal resulted in the unavailability of reward on that trial . All training and experimental behavioral sessions were conducted during the light cycle . Odors were comprised of binary mixtures of ( + ) -carvone and ( − ) -carvone ( Acros ) , commonly perceived as caraway and spearmint , respectively . In all sessions – including training on the task , as well as during neural recording and manipulation – mixtures in which ( + ) -carvone > ( − ) -carvone indicated reward availability at the left port , and ( − ) -carvone > ( + ) -carvone indicated reward availability at the right port . When ( + ) -carvone = ( − ) -carvone , the probability of reward at the left and right ports , independently , was 0 . 5 . The full set of ( + ) -carvone/ ( − ) -carvone ratios used was 95/5 , 80/20 , 60/40 , 50/50 , 40/60 , 20/80 , 5/95 . Mixtures were diluted in mineral oil and carrier air and delivered to the odor port at 800 ml/min . The mixture presented in each trial was selected pseudo-randomly . Reward ( 5 µL water ) was delivered by transiently opening a calibrated water valve 41 ± 32 ms ( mean ± s . d . ) after reward port entry . Mice completed training in 6–8 weeks and were then implanted with a drug-delivery cannula or a neural recording drive , as described below . Since our neural recording and manipulation experiments were performed in mice that were well-trained on the task , we do not attempt to examine how neural activity underlies task acquisition here ( e . g . , via reinforcement learning ) . Two types of implants were used in these experiments: ( 1 ) a steel infusion cannula ( Plastics One , Minneaspolis , MN ) for muscimol delivery and ( 2 ) a Versa Drive 4 microdrive ( containing four independently-adjustable tetrodes; Neuralynx ) for tetrode recordings . All implants were targeted to the PPTg using the same general stereotactic procedure; implant-specific details are described below . Mice were placed in a ventilated chamber and briefly exposed to a volatile anesthesia ( isoflurane , 2%; Priamal Healthcare Limited ) . Immediately following the onset of deep anesthesia ( verified by toe-pinch ) , the mouse was placed in a stereotaxic device with a nose cone that continuously delivered 1%–1 . 5% isoflurane to maintain anesthesia . When the mouse was fully unresponsive to foot pinch and appeared to maintain a consistent breathing rate , the fur on the surface of the scalp was removed , and topical antiseptic ( Betadine; Purdue Products ) was applied along with ophthalmic ointment on the eyes . Before exposing the skull , a bolus of topical anesthetic ( 250 µL 2% lidocaine; Aspen Veterinary Resources ) was injected under the surface of the scalp . With the skull exposed by central incision and scalp retraction , we adjusted head angle to align the elevation of bregma and lambda and drilled a 1 . 5 mm diameter cranial window centered on coordinates for the left PPTg ( 4 . 5 mm posterior from bregma , 1 . 1 mm lateral from the midline [Paxinos and Watson , 2006] ) . Following craniotomy and durotomy , we implanted one of the following: Prior to each session , an injection cannula was prepared with either muscimol ( test sessions ) or saline ( control sessions ) and inserted into the chronically implanted guide sans anesthesia . An infusion pump ( Harvard Apparatus ) was used to administer 150 nL of solution at 0 . 075 μL/min . Muscimol dosages ranged from 22 to 44 pmol . Mice recovered for at least 10 min before beginning the behavioral session . Recordings were collected using four tetrodes . Each tetrode consisted of four polyimide-coated nichrome wires ( 12 . 5 µm diameter; Sandvik ) gold plated to 0 . 2–0 . 4 MΩ impedance . Electrical signals were amplified and recorded using the Digital Lynx S multichannel acquisition system in conjunction with Cheetah data acquisition software ( Neuralynx ) . Tetrode depths , estimated by calculating the rotation of the screw affixed to the shuttle holding the tetrode ( one rotation = ~250 µm ) , were adjusted ~75 µm between recording sessions to sample independent populations of neurons across sessions . Offline spike sorting and cluster quality analysis was performed using MCLUST software ( MClust-4 . 0 , A . D . Redish et al . ) in MATLAB . Briefly , single units were isolated by manual clustering based on features of the sampled waveforms ( amplitude , energy , and the first principal component normalized by energy ) . Clusters with L-ratio <0 . 75 and isolation distance >12 were deemed single units ( Schmitzer-Torbert et al . , 2005 ) , which resulted in excluding 30% of clusters . Although units were clustered blind to inter-spike interval ( ISI ) , clusters with ISIs <1 ms were excluded . Final tetrode location was confirmed histologically using electrolytic lesions made after the last recording session and tetrode tracks ( Thompson and Felsen , 2013 ) . On day-one , post-lesion , mice were overdosed with an intraperitoneal injection of sodium pentobarbital ( 100 mg/kg; Sigma Life Science ) and transcardially perfused with saline and ice-cold 4% paraformaldehyde ( PFA ) in 0 . 1 M phosphate buffer ( PB ) . After perfusion , brains were submerged in 4% PFA in 0 . 1 M PB for 3 . 5 hr for post-fixation and then cryoprotected overnight by immersion in 20% sucrose in 0 . 1 M PB . The brain was embedded in optimal cutting temperature compound ( ThermoFisher Scientific ) and frozen rapidly on dry ice . Serial coronal sections ( 50 µm ) were cut on a cryostat . Alexa 555 fluorescent Nissl ( 1:500 , NeuroTrace; Invitrogen , catalog #N-21480 ) was used to identify cytoarchitectural features of the PPTg and verify tetrode tracks and lesions . We quantified choice behavior with a logistic function of the form p= 11+e−η , where p is the choice made on a given trial ( right choice , p=0; left choice , p=1; trials in which no reward port was entered within 1 . 5 s of odor port exit were excluded ) and η is the linear predictor , which was adapted to specific analyses as described below . For all analyses , η consisted of at least η0=β0+ βOdor: LxOdor: L+ βOdor: RxOdor: R , where β0 represents overall choice bias , xOdor: L and xOdor: R represent the strength of the odors associated with the left and right reward port , respectively , and βOdor: L and βOdor: R represent the influence of the odors on choice . xOdor: L and xOdor: R are calculated as ( fraction of odor - 0 . 5 ) / 0 . 5 and range from 0 to 1; we used separate terms for left and right odors to allow for the possibility that they asymmetrically influenced choice . In Figure 1E , for the Simple model , η=η0 . To assess the effect on choice behavior of inhibiting PPTg activity with muscimol ( Figure 5A , B ) , we combined each muscimol session with its pre- and post-saline session and set η=η0+βMuscimol xMuscimol , where xMuscimol=0 for saline trials , xMuscimol=1 for muscimol trials , and βMuscimol represents the influence of muscimol on choice [positive values represent leftward ( ipsilateral ) influence] ( Salzman et al . , 1992 ) . In addition , we examined the effect of inhibiting PPTg activity on movement latency and duration by examining the duration between the go signal and odor port exit , between odor port exit and reward port entry , and between reward port exit and odor port entry ( to initiate the next trial ) . To assess the effect on choice behavior of the choices and outcomes on previous trials , we set η= η0+ ∑n=1N ( β Choicenx Choicen+ β Outcome: Lnx Outcome: Ln+ β Outcome: Rnx Outcome: Rn ) , wherex Choicen={−1 for right choice0 for no choicentrialsback , 1 for left choicex Outcome: Ln={−1 for unrewarded left choice0 for nonleft choicentrialsback , 1 for rewarded left choicex Outcome: Rn={−1 for unrewarded right choice0 for nonright choicentrialsback , 1 for rewarded right choice β Choicen represents the influence of the choice n trials back ( ‘no choice’ includes any excluded trials and those in which odor port entry to initiate the next trial did not occur within 1 . 5 s of water port exit ) , β Outcome: Ln represents the influence of the outcome at the left n trials back , and β Outcome: Rn represents the influence of the outcome at the right n trials back . This set of coefficients provided the most intuitive interpretation of our data; our results did not depend on whether we instead included separate coefficients for previous left and right choice or a single coefficient for previous outcome . In Figure 2D–F , for the Extended model , 1 trial back , n = 1; for the Extended model , 2 trials back , n = 2 . To assess how inhibiting PPTg activity modulated the influence of the choice on the previous trial on behavior , we used a reduced form of the Extended model , 1 trial back by setting η= η0+ βChoice1x Choice1 , and calculated β Choice1 separately for inhibited ( muscimol ) and control ( saline ) sessions ( Figure 5C ) . To quantify the selectivity of single neurons for choice and outcome , we used an ROC-based algorithm ( Green and Swets , 1966 ) that calculates the ability of an ideal observer to classify whether a given spike rate was recorded in one of two conditions ( e . g . , on trials in which the left or right reward port was selected ) . ‘Preference’ was calculated as 2 ( ROCarea – 0 . 5 ) , a measure ranging from −1 to 1 , where −1 signifies the strongest possible preference for one alternative , 1 signifies the strongest possible preference for the other alternative , and 0 signifies no preference ( Thompson and Felsen , 2013 ) . For choice preference , −1 = left and 1 = right; for outcome preference , −1 = no reward and 1 = reward . Since choice is correlated with movement direction , this measure of choice preference alone formally cannot disambiguate whether activity reflects the chosen side itself or the direction of movement either towards or from the reward port . However , either movement is a perfect proxy for the choice – e . g . , leftward movement towards the reward port and rightward movement from the reward port both indicate a left choice – and we report in the Results section an analysis derived from choice preference that can disambiguate whether PPTg activity reflects choice or movement direction . For clarity , we refer to neural activity with respect to the chosen reward port and the direction of movement towards that port ( i . e . , left and leftward in the above example ) . Statistical significance of preference was determined with a Monte Carlo permutation test: we recalculated preference after randomly reassigning each firing rate to one of the two groups , repeated this procedure 1000 times to obtain a distribution , and calculated the fraction of randomly generated preferences exceeding the actual preference . We examined preference during the pre-stimulus epoch ( from odor port entry to 100 ms after odor valve open , before stimulus information can reach the PPTg; Figure 1A ) . For all preference analyses , we tested for significance at α = 0 . 05 . Neurons with fewer than four trials for each condition or with a firing rate below 2 spikes/s for both conditions were excluded from analysis ( Thompson and Felsen , 2013 ) . To assess the influence on PPTg activity of the choices and outcomes on the previous and current trials ( Figure 3 ) , we fit the electrophysiological data with a multi-variable linear regression model of the formFR ( t ) = β0 ( t ) + β Choice1 ( t ) x Choice1+ β Outcome1 ( t ) x Outcome1+ β Choice0 ( t ) x Choice0+ β Outcome0 ( t ) x Outcome0 , where FR ( t ) is the mean firing rate in a given time bin t ( 100 ms duration , sampled every 10 ms ) , x Choice1={−1 for an ipsilateral ( left ) choice0 for no choiceontheprevioustrial , 1 for a contralateral ( right ) choicex Outcome1={−1 for a nonrewarded choice0 for no choiceontheprevioustrial , 1 for a rewarded choicex Choice0={−1 for an ipsilateral choice0 for no choiceonthecurrenttrial , 1 for a contralateral choicex Outcome0={−1 for a nonrewarded choice0 for no choiceonthecurrenttrial , 1 for a rewarded choice β0 ( t ) represents the mean firing rate across trials in time bin t , and β Choice1 ( t ) , β Outcome1 ( t ) , β Choice0 ( t ) , and β Outcome0 ( t ) represent the influence on firing rate of the previous choice , previous outcome , current choice and current outcome , respectively , in time bin t ( Rorie et al . , 2010 ) . Positive values for β Choice1 and β Choice0 indicate that firing rate is increased by contralateral choices and negative values indicate that firing rate is increased by ipsilateral choices . Positive values for β Outcome1 and β Outcome0 indicate that firing rate is increased by rewarded choices and negative values indicate that firing rate is increased by unrewarded choices . | The decisions we make are influenced by recent experience , yet it is not known how this experience is represented in the brain . For decisions about when , where and how to move , researchers have hypothesized that recent experience might influence activity in a region of the brainstem – the central trunk of the brain – that is known to be involved in movement . When deciding when , where and how to move , several areas of the brain are involved in selecting the optimal action . Recent studies suggest that groups of neurons known as locomotor brainstem nuclei may also contribute to making decisions about movements . Thompson et al . investigated whether a brainstem locomotor area called the pedunculopontine tegmental ( PPTg ) nucleus in mice might contribute to decision making rather than just conveying the selected response . The mice were trained to recognize particular odors and move to either the left or right to collect a food reward . While the mice were selecting an action , the activity of neurons in the PPTg nucleus reflected the action they had chosen on a previous experience and the outcome of that choice ( i . e . whether they received a reward ) . These representations of past experiences influenced the upcoming decision the mice were about to take . The findings of Thompson et al . suggest that the PPTg nucleus might play a critical role in the process of selecting the optimal action . Future work will examine what kinds of information about the environment or recent experience have the biggest effect on the activity of this region . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Mesencephalic representations of recent experience influence decision making |
Heart regeneration is limited in adult mammals but occurs naturally in adult zebrafish through the activation of cardiomyocyte division . Several components of the cardiac injury microenvironment have been identified , yet no factor on its own is known to stimulate overt myocardial hyperplasia in a mature , uninjured animal . In this study , we find evidence that Neuregulin1 ( Nrg1 ) , previously shown to have mitogenic effects on mammalian cardiomyocytes , is sharply induced in perivascular cells after injury to the adult zebrafish heart . Inhibition of Erbb2 , an Nrg1 co-receptor , disrupts cardiomyocyte proliferation in response to injury , whereas myocardial Nrg1 overexpression enhances this proliferation . In uninjured zebrafish , the reactivation of Nrg1 expression induces cardiomyocyte dedifferentiation , overt muscle hyperplasia , epicardial activation , increased vascularization , and causes cardiomegaly through persistent addition of wall myocardium . Our findings identify Nrg1 as a potent , induced mitogen for the endogenous adult heart regeneration program .
Myocardial infarction ( MI ) is a common injury that causes permanent loss of hundreds of millions of cardiac muscle cells , increasing susceptibility to heart failure and sudden death . Major goals of regenerative medicine are methodologies to enhance cardiomyocyte recovery after MI and to restore cardiac function to patients with heart failure . Although there has been much investment in candidate cardiac stem cell populations over the past decade ( Laflamme and Murry , 2011; Behfar et al . , 2014 ) , many promising alternative strategies for heart regeneration have emerged , including activation of cardiomyocyte division , reprogramming of non-muscle cells into cardiomyocyte-like cells , and delivery of stem cell-derived cardiomyocytes ( Bersell et al . , 2009; Qian et al . , 2012; Shiba et al . , 2012; Song et al . , 2012; Chong et al . , 2014 ) . Heart regeneration occurs naturally after extreme tissue damage in non-mammalian vertebrates like zebrafish ( Poss et al . , 2002 ) . New myocardium is created through division of spared cardiomyocytes , and lineage-tracing experiments have not yielded evidence for a stem cell contribution ( Jopling et al . , 2010; Kikuchi et al . , 2010 ) . Zebrafish regenerate after injuries that deplete 60% or more of their cardiomyocytes , suggesting broad potential of most or all cardiomyocytes to participate in regeneration in these animals ( Wang et al . , 2011 ) . By contrast , cardiomyocyte division is robust through early postnatal life in mice and can enable regeneration , but by the adult stage is profoundly reduced ( Porrello et al . , 2011 ) . Factors that on their own can stimulate spontaneous creation of patterned , vascularized adult cardiac muscle would hold great potential for addressing human cardiovascular disease . The adult heart is famously resistant to forced hyperplasia , and tumors of myocardial origin are exceedingly rare . Adult mammalian cardiomyocyte proliferation has been reported to be stimulated to a minor extent after ischemic injury ( Senyo et al . , 2013 ) . Genetic manipulations in cardiomyocytes of cell cycle genes like cyclin D , Rb , and/or p130 can increase cardiomyocyte cell cycle entry or division but do not cause significant muscularization ( Pasumarthi et al . , 2005; Sdek et al . , 2011 ) . Similarly , manipulations of Hippo signaling or miRNA function can increase cardiomyocyte proliferation after injury in mice ( Eulalio et al . , 2012; Heallen et al . , 2013; Xin et al . , 2013 ) , but no compelling evidence for an impact on adult cardiogenesis has been reported . In adult zebrafish , several manipulations have boosted cardiomyocyte proliferation after trauma ( Jopling et al . , 2012; Yin et al . , 2012; Choi et al . , 2013 ) , yet no genetic or pharmacologic method has stimulated cardiomyocyte proliferation or obvious cardiogenic growth in the absence of injury . The extracellular factor Neuregulin1 ( Nrg1 ) has multiple roles in cardiovascular biology and has been implicated as a cardiomyocyte mitogen . In developing zebrafish or mouse embryos , Nrg1 signaling is critical for cardiac myofiber trabeculation . Mice mutant in Nrg1 , or in its receptors ErbB2 or ErbB4 , die at mid-gestation , or later with cardiac-restricted mutations , from thin ventricular walls and aberrant trabeculation ( Gassmann et al . , 1995; Lee et al . , 1995; Meyer and Birchmeier , 1995; Liu et al . , 2010 ) . Moreover , different forms of Nrg1 peptides have been delivered to adult animals , with various effects on the cardiovascular system . Nrg1 delivery to embryonic mouse cardiac explants or embryonic rat cardiomyocytes increased their proliferation in culture ( Zhao et al . , 1998 ) , and Nrg1 treatment of cultured adult mouse cardiomyocytes or systemic injection into adult mice induced cardiomyocyte proliferation ( Bersell et al . , 2009 ) . Although the proliferative increases in this latter study were relatively small , the authors also found evidence that Nrg1 injection improves cardiac repair after MI . Nrg1 has reported effects on cell survival , metabolism , angiogenesis , and myofiber structure in addition to cardiomyocyte proliferation ( Parodi and Kuhn , 2014 ) , influences that are possibly reflected by functional improvement after Nrg1 infusion in congestive heart failure patients ( Jabbour et al . , 2011 ) . It is unclear from these recombinant protein delivery experiments whether and how Nrg1 is part of an endogenous regeneration program . Moreover , corroborating evidence generated by other research groups for Nrg1 as a pro-regenerative cardiomyocyte mitogen is lacking . Indeed , a recent study argued that Nrg1 delivery to adult mice has no effects on cardiomyocyte proliferation ( Reuter et al . , 2014 ) . To address this , we examined the role Nrg1 might play in the strong , endogenous regenerative response of the adult zebrafish heart . We find that zebrafish nrg1 is upregulated in perivascular cells following cardiac injury , and that blockade of Nrg1 signaling inhibits injury-induced cardiomyocyte proliferation . Most strikingly , transgenic reactivation of Nrg1 expression in the absence of cardiac injury stimulated many hallmarks of cardiac regeneration and markedly enhanced ventricular size . These findings implicate Nrg1 as a key mitogenic node between injury and the endogenous heart regeneration program .
Using quantitative PCR , we found that nrg1 levels increase after genetic ablation of ∼50% of cardiomyocytes in the adult zebrafish ventricle . nrg1 levels rise above baseline at 3 days post-injury and peak at ∼11-fold above uninjured levels by 7 days , an injury timepoint at which cardiomyocyte proliferation also peaks ( Wang et al . , 2011 ) . nrg1 levels lower to ∼fourfold above uninjured levels by 14 days post injury . To visualize nrg1 expression , we used RNAScope , a modified in situ hybridization technique with improved sensitivity over standard methodology ( Wang et al . , 2012 ) . While nrg1 was rarely detectable in uninjured hearts , nrg1 expression was induced 7 days after various injury methods . After resection of the ventricular apex , we observed nrg1 staining in small regions surrounding cardiac damage . We saw larger stretches of expression after genetic cardiomyocyte ablation , distributed throughout the ventricular wall with occasional expression in the trabecular compartment ( Figure 1B–H ) . nrg1 signals in the ventricular wall were commonly in perivascular regions ( Figure 1D–F ) . 10 . 7554/eLife . 05871 . 003Figure 1 . Induction of Nrg1 after cardiac injury . ( A ) Time course of nrg1 induction in cardiac ventricles following severe genetic ablation of cardiomyocytes . nrg1 mRNA levels were assessed by qPCR at 3 , 7 , and 14 days after ablation injury in tamoxifen-treated cmlc2:CreER; β-act2:RSDTA animals relative to control cmlc2:CreER animals ( closed circles ) . cmlc2:CreER; β-act2:RSDTA ( open circle ) vehicle-treated animals serve as an additional control . Data are presented as mean ± SE . *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 , Student's t-test , two-tailed . ( B–F ) Section images of in situ hybridization experiments assessed for nrg1 expression in uninjured ventricles ( B ) or at 7 days after induced cardiomyocyte ablation ( C and D ) . Dashed lines delineate the ventricular wall from the trabecular compartment . Higher magnification of boxes in ( E ) and ( F ) reveal nrg1 signals surrounding vessels ( bv ) . Arrowheads indicate examples of RNAscope signals . Scale bar represents 100 µm ( A and C ) . ( G and H ) Section images of RNAScope in situ hybridization analysis for nrg1 expression at 7 days after ventricular resection surgery . Image in ( H ) is a higher magnification of box in ( G ) . Arrowheads indicate examples of RNAscope signals . Scale bar represents 100 µm . ( I–K ) Confocal slice images , with accompanying orthogonal views , of nrg1 expression colocalized with tcf21:nucEGFP ( I ) , fli1:EGFP ( J ) , or cardiac muscle ( Troponin , K ) . Arrows point to RNAscope signal , and red arrows indicate area for orthogonal views . Scale bar represents 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 003 To define the cells inducing nrg1 , we combined RNAscope with transgenic reporter lines marking known cardiac cell types . In these experiments , the majority of nrg1 signals could be localized to cells also positive for tcf21:EGFP fluorescence ( Figure 1I ) . During heart development , tcf21+ cells contribute to the epicardial layer as well as vascular support cells ( Kikuchi et al . , 2011a; Acharya et al . , 2012 ) . After cardiac injury , epicardial cells and their progeny proliferate and incorporate into the injury site ( Kikuchi et al . , 2011a ) , where they have been assigned numerous pro-regenerative roles ( Gemberling et al . , 2013 ) . In other contexts , Nrg1 is known to be expressed in Schwann cells and endothelial cells ( Meyer et al . , 1997; Cote et al . , 2005; Stassart et al . , 2013 ) . Yet , we found that nrg1 signals rarely overlapped during heart regeneration with cells positive for fli1a:EGFP or cmlc2:EGFP , which mark vascular endothelial cells and cardiomyocytes , respectively ( Figure 1J , K ) . Thus , our results indicate that the dominant source of nrg1 in the postnatal ventricular wall is the tcf21+ epicardial derived , perivascular cell compartment . To test whether modulation of Nrg1 signaling can alter cardiomyocyte proliferation during regeneration , we employed loss- and gain-of-function approaches . Messages for Nrg1 receptors Erbb2 and Erbb4b were detectable by PCR methods in uninjured adult zebrafish ventricles ( Figure 2G ) . Previous studies reported that the administration of AG1478 , a small molecule inhibitor of Erbb receptors , mimics the effect of erbb2 mutations on cardiac trabeculation in zebrafish ( Liu et al . , 2010 ) . To examine Erbb activity requirements during regeneration , we treated adult zebrafish with 10 µM AG1478 from 6 to 7 dpa . We then quantified cardiomyocyte proliferation indices using nuclear markers of cardiomyocytes ( Mef2 ) and cell cycle stage ( PCNA ) , visual methodology that is required for accurate quantification of heart regeneration ( Wills et al . , 2008; Wang et al . , 2011; Yin et al . , 2012; Fang et al . , 2013 ) ( Figure 2A , B ) . This regimen decreased cardiomyocyte proliferation by ∼54% , indicating that Nrg1 signaling is essential for heart regeneration ( n = 20 , 22; Figure 2E ) . 10 . 7554/eLife . 05871 . 004Figure 2 . Nrg1 signaling modulates cardiomyocyte proliferation during regeneration . ( A and B ) Section images of injured ventricular apices of animals treated from 6 to 7 dpa with DMSO ( A ) or 10 µM AG1478 ( B ) and stained for Mef2+PCNA+ cells ( arrowheads ) . Wounds are indicated by dotted lines . Scale bar represents 100 µm . ( C and D ) Section images of 7 dpa ventricular apices of control β-act2:BSNrg1 ( C ) or cmlc2:CreER; β-act2:BSNrg1 ( D ) animals treated with tamoxifen at 3 days before injury , stained for Mef2+PCNA+ cells ( arrowheads ) . Scale bar represents 100 µm . ( E ) Quantification of cardiomyocyte proliferation at 7 dpa . DMSO-treated wild-type clutchmates ( n = 22 ) were used as controls for 10 µM AG1478 treatment ( n = 20 ) , and tamoxifen-treated β-act2:BSNrg1 clutchmates ( n = 15 ) were controls for cmlc2:CreER; β-act2:BSNrg1 ( n = 18 ) animals . Data are represented as mean ± SEM . *p < 0 . 05 , Mann–Whitney Ranked Sum Test . ( F ) Cartoon schematic of β-act2:BSNrg1 transgene . ( G ) RT-PCR results for erbb2 , erbb4a , and erbb4b , indicating the presence of erbb2 and erbb4b messages in the uninjured adult ventricle . cmlc2 is shown as a control . ( H ) Section image of RNAscope in situ hybridization analysis for nrg1 expression at 14 days after tamoxifen-released expression in uninjured cmlc2:CreER; β-act2:BSNrg1 ventricles . ( I ) Section image of RNAscope in situ hybridization analysis for EGFP expression in uninjured cmlc2:actinin3-EGFP ventricles , used as a control to detect transgenic signals . Scale bar represents 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 004 To increase Nrg1 levels , we created a transgenic line to inducibly express nrg1 in cardiomyocytes when combined with a cardiomyocyte-restricted , taxmoxifen-inducible transgene , ( Tg ( β-actin2:loxP-mTagBFP-STOP-loxP-Nrg1 ) pd107; referred to hereafter as β-act2:BSNrg1 ) ( Figure 2F; Kikuchi et al . , 2010 ) . We treated cmlc2:CreER; β-act2:BSNrg1 animals with tamoxifen to induce nrg1 expression ( Figure 2H , I ) . 3 days later , we resected ventricles , and found that elevated nrg1 expression led to an ∼84% increase in the cardiomyocyte proliferation index near the injury site at 7 dpa ( n = 15 , 18; Figure 2C–E ) . This finding reveals a mitogenic influence of Nrg1 signaling on heart regeneration and suggests that nrg1 levels are a limiting factor for cardiomyocyte proliferation after cardiac injury . To test the effects of activating nrg1 expression in the absence of injury , we treated 4- to 6-month-old cmlc2:CreER; β-act2:BSNrg1 animals with tamoxifen and collected ventricles 7–30 days after treatment ( dpt ) . Within 7 days of nrg1 reactivation , there was a marked increase in cardiomyocyte proliferation , in particular within the ventricular wall ( Figure 3A , B ) . Whereas the cardiomyocyte proliferation index in the trabecular compartment increased modestly from ∼0 . 2% to ∼1% after 7 days of nrg1 overexpression ( Figure 3C ) , the proliferation index of cortical muscle increased sharply from ∼1 . 6% to ∼28% ( n = 8 , 9; Figure 3D , E ) . Continued nrg1 overexpression over an additional 7 days had similar effects , maintaining trabecular proliferation indices at ∼1 . 6% and cortical muscle proliferation at ∼32% ( n = 10 , 10; Figure 3C–E ) . This increased cardiomyocyte proliferation manifested through marked changes in cardiac anatomy . The ventricular wall thickened by ∼76% at 7 dpt compared to controls , by ∼265% at 14 dpt , and by ∼459% after 30 days of overexpression ( n = 8–11; Figure 4A–E ) . To test the extent to which cellular hypertrophy contributes to this wall thickening , we assessed the size and numbers of cardiomyocytes , which are predominantly mononuclear in adult zebrafish ( Wills et al . , 2008 ) . We found large increases in the number of cardiomyocyte nuclei within the ventricular wall after 14 days of nrg1 expression , and no evidence for increased cardiomyocyte size ( Figure 5A–C ) . Thus , most or all of the effects of nrg1 reactivation are hyperplastic . Such an extreme hyperplastic response to a single factor , which we term Nrg1-induced cardiac hyperplasia ( iCH ) for brevity , was unexpected from the published body of literature . 10 . 7554/eLife . 05871 . 005Figure 3 . Nrg1 reactivation without injury induces proliferation of ventricular wall cardiomyocytes . ( A and B ) Section images from uninjured cmlc2:CreER; β-act2:BSNrg1 and control ventricles at 7 days post-tamoxifen treatment ( dpt ) , stained for Mef2+PCNA+ cells . Insets show high-zoom views of the boxed regions , and arrowheads indicate Mef2+PCNA+ nuclei . Dashed lines delineate cortical ( wall ) from trabecular muscle . Scale bars represent 100 µm . ( C ) Quantification of cardiomyocyte proliferation in cmlc2:CreER; β-act2:BSNrg1 and controls in the trabecular muscle compartment at 7 ( n = 8 , 9 ) and 14 dpt ( n = 10 , 10 ) . Data are represented as mean ± SEM . *p < 0 . 05 , Mann–Whitney Ranked Sum Test . ( D ) Quantification of cardiomyocyte proliferation in cortical muscle at 7 ( n = 8 , 9 ) and 14 dpt ( n = 10 , 10 ) , from groups in ( A and B ) . Data are represented as mean ± SEM . *p < 0 . 05 , Mann–Whitney Ranked Sum Test . ( E ) Quantification of total cardiomyocyte proliferation at 7 ( n = 8 , 9 ) and 14 dpt ( n = 10 , 10 ) , from groups in ( C and D ) . Data are represented as mean ± SEM . *p < 0 . 05 , Mann–Whitney Ranked Sum Test . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 00510 . 7554/eLife . 05871 . 006Figure 4 . Nrg1-induced cardiomyocyte proliferation expands the ventricular wall . ( A ) Quantification of cortical muscle thickness at 7 ( n = 8 , 9 ) , 14 ( n = 10 , 11 ) , and 30 dpt ( n = 11 , 11 ) . Data are represented as mean ± SEM . *p < 0 . 05 , Student's t-test , two-tailed . ( B–E ) Section images of cmlc2:CreER; β-act2:BSNrg1 ( Nrg1 on ) and control ventricles from 7 to 30 dpt , using animals also transgenic for cmlc2:actinin3-EGFP to indicate sarcomere organization . Brackets indicate cortical muscle , and dashed lines delineate cortical from trabecular muscle . Scale bar represents 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 00610 . 7554/eLife . 05871 . 007Figure 5 . Nrg1 induces a hyperplastic , not hypertrophic , response . ( A ) Quantification of total ventricular wall cardiomyocytes in cmlc2:CreER; β-act2:BSNrg1 animals and controls at 7 ( n = 8 , 9 ) and 14 dpt ( n = 10 , 10 ) . Data are represented as mean ± SEM . *p < 0 . 05 , Student's t-test , two-tailed . ( B ) Quantification of cardiomyocyte area in cmlc2:CreER; β-act2:BSNrg1 animals and controls at 14 dpt . Data are represented as mean ± SD , with all data points represented . *p < 0 . 05 , Student's t-test , two-tailed . ( C ) Confocal images of dissociated cardiomyocytes from cmlc2:CreER; β-act2:BSNrg1 ( right ) and controls ( left ) at 14 dpt ( n = 124 , 172 ) . Only cardiomyocytes with visible sarcomeres and nuclei were measured . Examples of quantified cells are marked with arrowheads . Scale bar represents 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 007 To determine long-term effects of iCH , we examined hearts 6 months after tamoxifen treatment . While animals displayed no outward differences from control clutchmates at this stage , analysis of dissected tissues revealed obvious cardiomegaly ( Figure 6A ) . On average , experimental animals at 6 months post-tamoxifen treatment had ventricles with a ∼2 . 1-fold greater ventricular section area than control animals ( n = 9 , 10; Figure 6B ) . Tissue sections indicated elevated cardiomyocyte proliferation even at this late stage , although in some cases regions of the massively thickened ventricular wall showed mild fibrin and collagen deposition ( Figure 6C–G ) . No fibrosis was detectable in hearts of animals after just 30 days of iCH ( n = 7; Figure 6H , I ) . 10 . 7554/eLife . 05871 . 008Figure 6 . Nrg1-induced hyperplasia causes cardiomegaly . ( A ) Whole-mount images of cmlc2:CreER; β-act2:BSNrg1 and control ventricles at 6 months post-tamoxifen treatment . ( B ) Quantification of the cross-sectional surface area of cmlc2:CreER; β-act2:BSNrg1 ( n = 9 ) and control ventricles ( n = 10 ) 6 months post-treatment , revealing cardiomegaly effects of nrg1 overexpression . Data are represented as mean ± SEM . *p < 0 . 05 , Student's t-test , two-tailed . ( C and D ) Section images of ventricular walls of 6 mpt control β-act2:BSNrg1 ( C ) or cmlc2:CreER; β-act2:BSNrg1 animals ( D ) stained for Mef2+PCNA+ cells ( arrowheads ) . Scale bar represents 100 µm . ( E ) Section images of control β-act2:BSNrg1 ventricles stained with Acid-Fuchsin Orange G ( AFOG ) , revealing minimal collagen ( blue ) , or fibrin deposition ( red ) . Scale bar represents 100 µm . ( F and G ) Section images of cmlc2:CreER; β-act2:BSNrg1 ventricles stained with AFOG , revealing collagen ( blue ) and fibrin deposition ( red ) in the inner portions of the thickened ventricular wall . Image in ( G ) is a high-zoom view of box in ( F ) and also indicates two examples of large coronary vessels ( bv ) . ( H and I ) Acid-Fuchsin Orange ( AFOG ) staining reveals minimal fibrosis in cmlc2:CreER; β-act2:BSNrg1 ventricle at 30 dpt despite the thickened ventricular wall ( n = 7 , 7 ) . Scale bar represents 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 008 To further understand the consequences of iCH , we monitored physiologic parameters over a period of 9 months . First , we measured cardiac function using echocardiography . The spatial resolution of standard B-mode echocardiography has limitations in zebrafish; yet we were able to use Doppler echocardiography to detect changes in atrioventricular inflow after 3 months of iCH ( Figure 7A , B ) . These findings suggested that ventricular filling is altered as thickness of the ventricular walls increases ( Figure 7C ) . Importantly , we were unable to detect cardiac dysfunction at 3 months of iCH , and instead our results suggested that stroke volume is enhanced with iCH ( Figure 7C , D ) . By 3 months of iCH , ventricular wall hyperplasia was obvious by echocardiography ( Figure 7; Videos 1 , 2 ) . Accurate echocardiography at 8 or 9 months of iCH was challenged by the extreme cardiac dysmorphology in these animals . 10 . 7554/eLife . 05871 . 009Figure 7 . Effects of Nrg1 reactivation on cardiac function . ( A ) Doppler measures of ventricular filling obtained at the AV valve in cmlc2:CreER; β-act2:BSNrg1 and β-act2:BSNrg1 animals ( n = 9 and 7 ) . Data are represented as mean ± SEM . *p < 0 . 05 , Student's t-test , two-tailed . ( B ) . Representative PW Doppler at the AV valve in cmlc2:CreER; β-act2:BSNrg1 and β-act2:BSNrg1 animals . Arrows indicate E waves . ( C ) Doppler measures of cardiac output and stroke volume using the velocity time integral ( VTI ) obtained at the outflow tract ( OFT ) in cmlc2:CreER; β-act2:BSNrg1 and β-act2:BSNrg1 animals ( n = 9 , 7 ) . Data are represented as mean ± SEM . *p < 0 . 05 , Student's t-test , two-tailed . ( D ) Representative PW Doppler at the OFT in cmlc2:CreER; β-act2:BSNrg1 and β-act2:BSNrg1 animals . ( E ) Quantification of graded swimming performance of animals at varying times of nrg1 overexpression plotted as box and whisker plots . Two-way ANOVA was performed looking at the effect of Nrg1 overexpression ( p < 0 . 05 ) , age ( p = 0 . 38 ) , and the interaction of Nrg1 overexpression and age ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 00910 . 7554/eLife . 05871 . 010Video 1 . B-mode video of echocardiography performed on adult control β-act2:BSNrg1 animals 3 months post-tamoxifen treatment , indicating a thin ventricular wall . Relates to Figure 7A–D . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 01010 . 7554/eLife . 05871 . 011Video 2 . B-mode video of echocardiography performed on adult cmlc2:CreER; β-act2:BSNrg1 animals 3 months post-tamoxifen treatment , showing increased wall thickness . Relates to Figure 7A–D . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 011 We next examined the ability of transgenic animals to swim against an increasing water current , an assay of cardiac function that is sensitive to massive cardiac damage and heart failure ( Wang et al . , 2011 ) . Animals at 3 months of iCH showed normal endurance swimming , consistent with echocardiographic results . At later time points ( 8 or 9 months iCH ) , animals displayed a sharp reduction in swimming endurance , suggesting that continued addition of myocardium eventually turns deleterious ( Figure 7E ) . Thus , forced nrg1 reactivation in the absence of cardiac injury induces robust myocardial hyperplasia in adult zebrafish , adding many new layers of myocardium over weeks of continued cardiomyocyte proliferation . To identify mechanisms of iCH , we examined several hallmarks of injury-induced regeneration in zebrafish . First , to reveal the spatiotemporal growth patterns of cardiogenesis , we coupled iCH with multicolor clonal analysis . The cortical muscle in the ventricular wall typically forms from a small number of large cardiomyocyte clones . These clones expand laterally on the ventricular surface and largely retain discernable boundaries through adulthood ( Gupta and Poss , 2012 ) . By contrast , regeneration of resected muscle occurs through roughly uniform proliferation by many cardiomyocytes near the injury site , generating a mixed conglomeration of small clones in the restored ventricular wall ( Gupta et al . , 2013 ) . We induced nrg1 expression simultaneously with permanent multicolor labeling in 5 weeks post-fertilization ( wpf ) juvenile animals , around the time of the initial emergence of cortical muscle . iCH caused ectopic wall thickening by 10 wpf , with obvious clone mixing and clone growth in the Z-plane away from the lumen ( 14/14 iCH , 0/11 controls; Figure 8A–F ) . These findings indicate that Nrg1 does not activate a gradual layering process but builds muscle radially with proliferation dynamics that are more reminiscent of injury-induced regeneration . 10 . 7554/eLife . 05871 . 012Figure 8 . Nrg1 reactivation is sufficient to induce the heart regeneration program . ( A–F ) Section images of ventricles from control cmlc2:CreER; priZM ( A–C ) and cmlc2:CreER; β-act2:BSNrg1; priZM ( D–F ) animals treated with tamoxifen at 5 weeks post-fertilization ( wpf ) and assessed at 10 wpf . Cortical myocyte clones show clear boundaries between clones in control ventricles ( B and C; n = 11 ) . During nrg1 overexpression , cortical muscle thickens appreciably via mixing and radial growth of distinct clones ( E and F; n = 14 ) . Dashed lines delineate cortical from trabecular muscle . Scale bar represents 100 µm . ( G and H ) Section images of ventricles from cmlc2:CreER; β-act2:BSNrg1 ( Nrg1 on ) and control animals at 7 days post-treatment , using animals also transgenic for gata4:EGFP . EGFP induction is clear in the cortical layer during nrg1 overexpression . Scale bar represents 100 µm . ( I and J ) Section images of ventricles from cmlc2:CreER; β-act2:BSNrg1 and control animals at 7 days post-treatment , visualized for tgfβ3 expression by in situ hybridization . Scale bar represents 100 µm . ( K , L , Q , R ) Section images of ventricles from cmlc2:CreER; β-act2:BSNrg1 and control animals at 7 days post-treatment , visualized for raldh2 ( K and L ) or fn1 expression ( Q and R ) in epicardial cells by in situ hybridization . Scale bar represents 100 µm . ( M and N ) Section images of the ventricular wall from cmlc2:CreER; β-act2:BSNrg1 and control animals at 14 days post-treatment , using animals transgenic for cmlc2:actinin3-EGFP . EGFP marks sarcomeric z-bands . Control animals ( M ) show organized sarcomeres in ventricular wall . nrg1 overexpression ( O ) leads to reduced EGFP fluorescence and disorganization of sarcomeres . Arrowheads point to areas of reduced EGFP intensity and sarcomere organization . Boxes in ( M and O ) are represented as high-zoom in ( N and P ) . Scale bars represents 50 µm . ( S and T ) Section images of ventricles from cmlc2:CreER; β-act2:BSNrg1 and control animals at 14 days post-treatment , visualized for epicardial cells using a tcf21:nucEGFP transgene . nrg1 overexpression grossly increases epicardial cell presence . Scale bar represents 100 µm . ( U and V ) Section images of ventricles from cmlc2:CreER; β-act2:BSNrg1 and control animals at 30 days post-treatment , visualized for endothelial cells using a fli1:EGFP transgene . Increased endothelial cells and vasculature are evident in the thickened ventricular wall . Scale bar represents 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05871 . 012 We next examined molecular signatures of iCH . After partial ventricular resection , cardiomyocytes near the injury activate regulatory sequences of the embryonic cardiogenic transcription factor gata4 , before dividing to create new muscle ( Kikuchi et al . , 2010 ) . Moreover , Gata4 activity is essential for regeneration of this muscle ( Gupta et al . , 2013 ) . Using a transgenic reporter strain , we found that 7 days of iCH in mature adults induced gata4:EGFP expression in the outermost layer of cortical muscle ( 9/10 iCH , 1/11 controls; Figure 8G , H ) . Additionally , by coupling iCH with a transgenic reporter visualizing cardiomyocyte Actinin3 localization , we found that cortical muscle displayed reduced cmlc2 expression and poorly organized sarcomeres ( Figure 8M–P ) . These observations together suggest a reduction in the contractile program that is indicative of dedifferentiation . iCH also activated myocardial expression of tgfβ3 , which was implicated previously in regeneration ( 4/5 iCH , 0/5 controls; Figure 5I , J ) ( Chablais and Jazwinska , 2012; Choi et al . , 2013 ) . We examined other cardiac cells types for their response to iCH . Retinoic acid synthesis in endocardial and epicardial cells is induced by myocardial injury , where RA signaling is essential , but not sufficient , for cardiomyocyte proliferation ( Kikuchi et al . , 2011b ) . Fibronectin ( Fn ) synthesis is also activated in the epicardium by injury and is an essential extracellular matrix ( ECM ) component of regeneration ( Wang et al . , 2013 ) . Both raldh2 and fn1 were induced in the epicardium by 7 days of iCH ( 7 of 9 iCH , 0 of 10 controls for each marker ) consistent with the presence of a regenerative program ( Figure 5K , L , Q , R ) . Along with epicardial marker induction , 7–14 days of iCH induced expansion of the epicardial layer that is reminiscent of that observed during regeneration ( Lepilina et al . , 2006; Wang et al . , 2011 ) ( Figure 8S , T ) . Finally , we assessed myocardial vascularization , which occurs concomitantly with injury-induced regeneration ( Lepilina et al . , 2006 ) . The adult zebrafish ventricle typically has a thin muscular wall penetrated with vessels , but 30 days of iCH stimulated formation of major vascular network throughout the expanded ventricular wall ( Figure 8U , V ) . After 6 months of iCH , many large coronary vessels were evident in the ventricle ( Figure 6G ) . Together , this analysis indicates that expression of a single molecule , Nrg1 , is sufficient to induce and maintain critical aspects of the heart regeneration program that encompass several cell types . Here , we identify the extracellular factor Nrg1 as a potent activator of the heart regeneration program in zebrafish . Nrg1 is induced by cardiac injury and pharmacological blockade of its receptor decreases injury-induced cardiomyocyte proliferation—each indicating involvement in the endogenous process . Moreover , whereas several experimental manipulations have been reported to increase cardiomyocyte proliferation in an injured adult heart , the Nrg1 protocol we describe here is potently cardiogenic in the absence of trauma . Nrg1 was reported to have mitogenic effects on cultured adult cardiomyocytes ( Bersell et al . , 2009 ) , and its most remarkable property in our experiments was the ability to induce and maintain adult cardiomyocyte proliferation in the absence of injury . Thus , the primary cellular target of Nrg1 is likely to be cardiomyocytes . Nrg1 effects additionally involve organizing a tissue microenvironment that includes expression of additional mitogenic factors , ECM regulation , and vascular perfusion . This recruitment of various non-myocyte cell types is likely to be an endogenous role of Nrg1 , given that it is also known to stimulate vascularization after ischemic injury in the hindlimb and is capable of inducing expression of ECM components in fibroblasts ( Hedhli et al . , 2012; Kim et al . , 2012 ) . The next generation of genome editing-inspired tools for zebrafish researchers should enable tests of cell-restricted , inducible genetic deletion of key Nrg1 components in multiple cell types . These experiments promise to dissect and define Nrg1 signaling requirements for regenerative responses within the complex cardiac milieu . Factors that on their own can stimulate the creation of full myocardial units have unique potential to avoid complications from immunosuppression and arrhythmia compared to cell-based approaches and can have straightforward pharmacologic entrypoints . Here , we have described effects of endogenous myocardial Nrg1 delivery , whereas previous studies have injected purified Nrg1 protein ( Parodi and Kuhn , 2014 ) . These distinct experimental delivery methods could result in markedly different Nrg1 doses , target cells , and target receptor responses . It will be critical to define the mechanisms by which Nrg1 is induced by injury in zebrafish and restricted in the absence of trauma , as well as downstream Nrg1 targets in the regeneration program . Such investigation of the endogenous regulation of Nrg1 after cardiac injury can help guide methodology to optimize its delivery and impact on heart regeneration in mammals .
nrg1 cDNA was amplified with the following primers and ( Forward 5′-ACCGGTGCACCATGGCTGAGGTGAAAGCAGG-3′ , Reverse 5′-GcGGCCGCTCACACAGCTATAGGATCCT-3′ ) and then subcloned into the AgeI/NotI site of the β-act:loxP-TagBFP-STOP-lox-P vector ( Gupta et al . , 2013 ) . This construct was co-injected into one-cell-stage wild-type embryos with I-SceI . One founder was isolated and propagated . The full name of this transgenic line is Tg ( βactin2:loxP-mTagBFP-STOP-loxP-Neuregulin1 ) pd107 . Wild-type or transgenic zebrafish of the hybrid EK/AB strain of the indicated ages were used for all experiments . All transgenic strains were analyzed as hemizygotes . Published transgenic strains or other alleles used in this study were gata4:EGFP ( Tg ( gata4:EGFP ) ae1 ) ( Heicklen-Klein and Evans , 2004 ) ; cmlc2:CreER ( Tg ( cmlc2:CreER ) pd10 ) ( Kikuchi et al . , 2010 ) ( used with priZm , β-act2:BSNrg1 , and bactin2:loxp-mCherry-STOP-loxp-DTA ) ; tcf21:nucEGFP ( Tg ( tcf21:nucEGFP ) pd41 ) ; bactin2:loxp-mCherry-STOP-loxp-DTA ( Tg ( bactin2:loxP-mCherry-STOP-loxP-DTA176 ) pd36 ) ; and cmlc2:actinin3-EGFP ( Tg ( myl7:actnb3-EGFP ) ) ( Wang et al . , 2011 ) ; priZm ( Tg ( β-act2:Brainbow1 . 0L ) pd49 ) ( Gupta and Poss , 2012 ) ; and fli1:EGFP ( Tg ( fli1:EGFP ) y1 ) ( Lawson and Weinstein , 2002 ) . Ventricular resection surgeries were performed as described previously ( Poss et al . , 2002 ) . To induce expression of nrg1 , adult cmlc2:CreER; β-act2:BSNrg1 animals were bathed in 5 µM tamoxifen ( Sigma-Aldrich , St . Louis , MO ) for 18–24 hr , made from a 2 mM stock solution dissolved at room temperature in 100% propylene glycol . Animals were treated at a density of 3 per 125 ml of water and then returned to recirculating water . Juvenile cmlc2:CreER; β-act2:BSNrg1; priZm animals were incubated in 2 µM Tamoxifen from the same stock solution . Animals were treated at a density of 8 per 100 ml of water and then returned to recirculating water . For genetic cardiomyocyte ablation , adult cmlc2:CreER; βactin2:loxp-mCherry-STOP-loxp-DTA animals were placed in 0 . 3 µM Tamoxifen for 16 hr . Animals were treated at a density of 3–4 per 125 ml of water and then returned to recirculating water . Adult animals were incubated in 10 µM AG1478 ( Selleck Chemical , Houston , TX ) diluted from a 10 mM stock in DMSO for a 24-hr period from 6 dpa to 7 dpa . TriReagent was used to isolate RNA from whole ventricles , with 4–6 chambers pooled for each sample . From partially resected ventricles , only the apical halves were collected . A total of 0 . 5–1 µg of total RNA was used in each cDNA synthesis reaction . cDNA was synthesized using the Roche Transcriptor first strand synthesis kit . Quantitative PCR was performed using the Roche Light Cycler 480 , Roche UPL probes , and LightCycler 480 Probes Master . Intron spanning primer sets were designed using the Roche UPL design center . All experiments were performed using biological and technical triplicates . Primers were tested for efficiency and all primer sets were found to have efficiencies between 1 . 95 and 2 . 05 . Primer sets used were ef1alpha ( Forward 5′-CCTCTTTCTGTTACCTGGCAAA-3′ , Reverse 5′-CTTTTCCTTTCCCATGATTGA-3′ , used with probe #73 ) and nrg1 ( Forward 5′-CACAAATGAGTTCACATCACCA-3′ , Reverse 5′-TCTGCTTTGCCATTACTCCA-3′ , used with probe #76 ) ; nrg1 levels were normalized to ef1alpha levels for all experiments . RT-PCR was performed to assess erbb receptor levels . We used the following primers for each receptor: erbb2 ( Forward 5′-GATGGCAACATGGTTTTCCT-3′ , Reverse 5′-TGGGTTCTCCACACTGTTCC-3′ ) , erbb4a ( Forward 5′-ATGTCAGGATCAGGGGATGA-3′ , Reverse 5′-TTCCGATGGTTTACGAAAGG-3′ ) , and erbb4b ( Forward 5′-TTATTGCGGCAGGGGTTATTGGAGG-3′ , Reverse 5′-CAACAGGAATCTTCACAGTCTCACCCTCA-3′ ) . In situ hybridization was performed on 10-µm sections of paraformaldehyde-fixed hearts as described ( Poss et al . , 2002 ) . In situ hybridization was performed manually or with the aid of an InSituPro robot ( Intavis ) . In situ hybridization data were imaged as described ( Kikuchi et al . , 2011b ) . RNAscope ( Advanced Cell Diagnostics , Hayward , CA ) was performed on hearts fixed with paraformaldehyde for 1 hr at room temperature , equilibrated in 30% sucrose overnight , embedded in OCT , and sectioned to 10 µm . Tissue was washed with PBS for 5 min to remove OCT , followed by boiling in Pretreat 2 for 20 min . After Pretreat 2 , slides were briefly washed with water and incubated for 10 min at 40°C with Pretreat 4 . Following Pretreat 4 , the manufacturer's protocol for RNAscope 2 . 0 HD detection Kit—Red was followed . Immunostaining was performed following nrg1 detection as described in Kikuchi et al . ( 2011b ) , with primary antibodies incubated overnight at 4°C . Advanced Cell Diagnostics designed nrg1 and EGFP probes used in this study . RNAscope was performed on 6–8 animals for resection and ablation studies . Primary and secondary antibody staining was performed as described ( Kikuchi et al . , 2011b ) . Acid Fuchsin-Orange G staining was performed on 10-µm sections as described ( Poss et al . , 2002 ) . AFOG was performed on 6–10 animals per time point analyzed . Mef2/PCNA staining on sections from 7 dpa ventricles was performed and imaged as described ( Kikuchi et al . , 2011b ) . A Zeiss 700 confocal microscope was used to image Mef2/PCNA-stained sections from whole uninjured ventricles , using the tilescan function to acquire entire longitudinal sections from each ventricle . Images were taken of the three largest sections from each ventricle . Mef2+ and Mef2+/PCNA+ cells were counted manually . Three sections from each heart were averaged to compute a proliferative index for each animal . Cross-sectional areas of ventricles at 6-month post-tamoxifen or -vehicle treatment were calculated ( ImageJ ) using images of the three largest sections from each heart stained for TroponinT . priZm samples were imaged and processed as described ( Gupta and Poss , 2012 ) . A Zeiss 700 confocal microscope was used to image RNAscope for colocalization using a DIC filter to capture the nrg1 signal . Z-stacks were taken for orthogonal views to show co-localization . Primary antibodies used in this study: anti-PCNA ( mouse; sigma ) at 1:250 , anti-Mef2 ( rabbit; Santa Cruz Biotechnology , Dallas , TX ) at 1:75 , anti-troponinT ( mouse; Thermo Scientific , Waltham , MA ) at 1:100 , anti-GFP ( rabbit; Life Technologies , Carlsbad , CA ) at 1:100 , and anti-α-actinin ( mouse; Sigma-Aldrich ) at 1:400 . Secondary antibodies used in this study: Alexa Fluor 594 goat anti-mouse IgG ( H + L ) for anti-Mef2 , anti-PCNA , and anti-α-actinin; and Alexa Fluor 488 goat anti-rabbit IgG ( H + L ) for anti-Mef2 , anti-PCNA , and anti-GFP . Secondary antibodies ( Life Technologies ) were all used at 1:200 . Previously described probes for raldh2 , fn1 , and tgfb3 were used for in situ hybridization ( Wang et al . , 2011; Choi et al . , 2013 ) . Raw data spreadsheets from experiments in this study are included in Supplementary file 1 . Swim tunnel analysis was performed as described ( Wang et al . , 2011 ) , with two exceptions: ( 1 ) fish were exercised in groups of 10–15 in a larger swim tunnel ( Respirometer 5L [120 V/60 Hz]; Loligo cat #SW10060 ) ; and ( 2 ) there was no time limit for swimming . Swimming speed was increased every two minutes after a 20 min acclimation period . Water velocities were measured up to 102 cm/s and values above were extrapolated using the other measures . Exhausted animals were removed from the chamber without disturbing the remaining fish , while others continued to swim . After all fish swam to exhaustion , they were allowed to recover and then placed back in recirculating water . To determine significance , 2-way ANOVA was performed looking at the effect of Nrg1 overexpression , time , and the interaction of Nrg1 overexpression and time . The number of total animals analyzed per time point: pre-recombination ( 24 ) , 1 month ( 24 ) , 2 months ( 24 ) , 3 months ( 23 ) , 5 months ( 21 ) , 6 months ( 19 ) , 8 months ( 17 ) , and 9 months ( 17 ) . Hearts were removed and washed in PBS wt/heparin and dissociated using a previously described protocol with minor changes ( Sander et al . , 2013 ) . Three ventricles were placed in each tube and dissociated for 1 . 5 hr , leaving the majority of the ventricles intact while enriching dissociated cells for wall cardiomyocytes . After dissociation , cells were plated onto a glass slide using a cytospin and spun for 3 min at 400×g . Then , cells were fixed for 10 min with 4% PFA and washed three times with PBS wt/0 . 1% Tween-20 for 5 min . Slides were stained for α-actinin overnight at 4°C and imaged using a Zeiss LSM 700 confocal microscope . Cells with the following criteria were measured using ImageJ software: ( 1 ) flattened appearance with visible sarcomeric staining; ( 2 ) clear dissociation from other cells; ( 3 ) single nucleus . Echocardiography was performed on conscious zebrafish using a Vevo 2100 high-resolution imaging system with an MS-550S transducer ( VisualSonics ) . Fish were sedated in phenoxyethanol and immobilized on a sponge immersed in fish water . Imaging was performed in the short axis ( perpendicular to the fish ) and long axis ( parallel to the fish ) using B-mode and PW imaging . Doppler parameters were measured using the Vevo2100 Cardiac Package by taking the average measurement of three consecutive contractions . Velocity time integral ( VTI ) and VTI*HR were used as measures of cardiac performance and output . Cardiac output could not be formally calculated , as outflow tract ( OFT ) diameter could not be directly measured by echocardiography . However , OFT diameter was not significantly different between experimental groups when measured in tissue sections ( data not shown ) . Ventricular filling was measured by PW Doppler across the atrioventricular valve during diastole . | Heart attacks—which are a major cause of death in humans—occur when a blocked blood vessel stops blood from flowing to the heart . This causes many heart muscle cells to die , which can result in permanent damage that makes survivors more susceptible to heart failure in the future . A major goal of regenerative medicine is to develop therapies that can improve the recovery of heart muscle cells after a heart attack and restore normal heart activity to patients with heart failure . Unlike the human heart , the heart of an adult zebrafish is able to regenerate even after extensive damage . After an injury , the remaining heart muscle cells divide to replace the lost heart muscle , but it is not clear how this works . A protein called Neuregulin1 ( or Nrg1 for short ) can stimulate heart muscle cells to divide . Gemberling et al . investigated the role of this protein in the regeneration of the heart in adult zebrafish . The experiments show that when the heart is injured , the gene encoding the Nrg1 protein is switched on in cells of the outer layer of the heart wall . When Nrg1 is deliberately activated in uninjured adult zebrafish hearts , it causes the muscle cells to divide , leading to many new layers of heart muscle forming over the course of several weeks . Along with promoting cell division , Nrg1 also makes the heart muscle cells return to an immature state more like stem cells . Gemberling et al . found that Nrg1 also supports regeneration of the heart by changing the environment surrounding the muscle cells . For example , it stimulates the growth of new blood vessels and recruits non-muscle cells to the injury site . Gemberling et al . 's findings demonstrate that Nrg1 is sufficient to induce the growth of heart muscle growth in an adult animal , even in the absence of injury . To develop its therapeutic potential , future work will also need to identify how the gene that encodes Nrg1 is switched on by injury and identify the other molecules that interact with Nrg1 . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine"
] | 2015 | Nrg1 is an injury-induced cardiomyocyte mitogen for the endogenous heart regeneration program in zebrafish |
Choanoflagellates , the closest living relatives of animals , can provide unique insights into the changes in gene content that preceded the origin of animals . However , only two choanoflagellate genomes are currently available , providing poor coverage of their diversity . We sequenced transcriptomes of 19 additional choanoflagellate species to produce a comprehensive reconstruction of the gains and losses that shaped the ancestral animal gene repertoire . We identified ~1944 gene families that originated on the animal stem lineage , of which only 39 are conserved across all animals in our study . In addition , ~372 gene families previously thought to be animal-specific , including Notch , Delta , and homologs of the animal Toll-like receptor genes , instead evolved prior to the animal-choanoflagellate divergence . Our findings contribute to an increasingly detailed portrait of the gene families that defined the biology of the Urmetazoan and that may underpin core features of extant animals .
The biology of the first animal , the ‘Urmetazoan , ’ has fascinated and confounded biologists for more than a century ( Dujardin , 1841; James-Clark , 1867; Haeckel , 1869; Haeckel , 1873; Haeckel , 1874; Kent , 1880; Leadbeater and McCready , 2000 ) . What features defined the biology of the Urmetazoan , and which of those features represent animal innovations ? Despite the fact that the first animals originated over 600 million years ago ( Douzery et al . , 2004; Hedges et al . , 2004; Peterson et al . , 2004; Narbonne , 2005; Knoll , 2011 ) , features of their genomes can be reconstructed through phylogenetically-informed comparisons among extant animals , their closest living relatives , the choanoflagellates , and other closely related lineages ( King , 2004; Rokas , 2008; Richter and King , 2013; Grau-Bové et al . , 2017; Sebé-Pedrós et al . , 2017 ) . Although close to 1000 genomes of animals have been sequenced ( NCBI Resource Coordinators , 2017 ) , only two choanoflagellate genomes have been previously published ( King et al . , 2008; Fairclough et al . , 2013 ) . These two choanoflagellates are the strictly unicellular Monosiga brevicollis and the emerging model choanoflagellate Salpingoeca rosetta , which differentiates into a number of sexual and asexual cell types , ranging from single cells to multicellular rosette colonies ( Fairclough et al . , 2010; Dayel et al . , 2011; Levin and King , 2013 ) . The M . brevicollis and S . rosetta genomes revealed that many genes critical for animal biology , including p53 , Myc , cadherins , C-type lectins , and diverse tyrosine kinases , evolved before the divergence of animals and choanoflagellates ( King et al . , 2008; Fairclough et al . , 2013 ) , whereas many other genes essential for animal biology , including components of the Wnt , Notch/Delta , Hedgehog , TGF-β , and innate immune pathways ( e . g . , Toll-like receptors ) have not been detected in choanoflagellates , and therefore have been considered textbook examples of animal innovations . Nonetheless , M . brevicollis and S . rosetta are relatively closely related ( Carr et al . , 2017 ) , leaving the bulk of choanoflagellate diversity unexplored . Moreover , both species have demonstrably experienced gene loss , as some genes conserved among animals and non-choanoflagellates are apparently missing from M . brevicollis and S . rosetta . Examples include RNAi pathway components , which are present across eukaryotes ( Shabalina and Koonin , 2008 ) , the cell adhesion protein β-integrin , and T-box and Runx transcription factor families , which have been detected in the filasterean Capsaspora owczarzaki ( Sebé-Pedrós and Ruiz-Trillo , 2010; Sebé-Pedrós et al . , 2010; Sebé-Pedrós et al . , 2011; Sebé-Pedrós et al . , 2013a; Ferrer-Bonet and Ruiz-Trillo , 2017 ) . Gene loss can lead to false negatives during ancestral genome reconstruction , and the phenomenon in choanoflagellates parallels that of animals , where two species selected for early genome projects , Drosophila melanogaster and Caenorhabditis elegans , were later found to have lost numerous genes ( e . g . , Hedgehog and NF-κΒ in C . elegans and fibrillar collagens in both C . elegans and D . melanogaster ) that are critical for animal development and otherwise conserved across animal diversity ( C . elegans Sequencing Consortium , 1998; Aspöck et al . , 1999; Gilmore , 1999; Rubin et al . , 2000 ) . To counteract the impact of gene loss in M . brevicollis and S . rosetta , and gain a more complete picture of the Urmetazoan gene catalog , we analyzed the protein coding genes of 19 previously unsequenced species of choanoflagellates representing each major known lineage ( Carr et al . , 2017 ) . By comparing their gene catalogs with those of diverse animals and other phylogenetically relevant lineages , we have greatly expanded and refined our understanding of the genomic heritage of animals . This more comprehensive data set revealed that ~372 gene families that were previously thought to be animal-specific actually evolved prior to the divergence of choanoflagellates and animals , including gene families required for animal development ( e . g . , Notch/Delta ) and immunity ( e . g . , Toll-like receptors ) . We find that an additional ~1944 gene families evolved along the animal stem lineage , many of which likely underpin unique aspects of animal biology . Although most of these animal-specific genes were subsequently lost from one or more species , 39 core animal-specific genes are conserved in all animals within our data set , likely because of their importance to core features of animal biology .
To reconstruct the genomic landscape of animal evolution , we first cataloged the protein coding potential of nineteen diverse choanoflagellate species by sequencing and assembling their transcriptomes ( Figure 1 , Figure 1—figure supplement 1 , Supplementary file 1 ) . Because most of these species were previously little-studied in the laboratory , two important stages of this project were the establishment of growth conditions optimized for each species and the development of improved protocols for isolating choanoflagellate mRNA for cDNA library construction and sequencing ( Supplementary file 1 , Materials and methods ) . After performing de novo transcriptome assembly and filtering for cross-contamination , we predicted a catalog of between 18 , 816–61 , 053 unique protein-coding sequences per species . [These counts likely overestimate the true numbers of underlying protein-coding genes , as they may include multiple alternative splice variants for any given gene and redundant contigs resulting from intra-species polymorphisms or sequencing artifacts ( Grabherr et al . , 2011; Haas et al . , 2013 ) ] . Using multiple independent metrics , we found that the new choanoflagellate transcriptomes approximate the completeness of choanoflagellate genomes for the purposes of cataloging protein-coding genes . For example , by comparing the S . rosetta genome with its transcriptome , we found that 93% of S . rosetta genes predicted from the genome were represented in its transcriptome with coverage over at least 90% of their length ( Figure 1—figure supplement 2a ) . Furthermore , compared with the genomes of M . brevicollis and S . rosetta , which contain 83 and 89% , respectively , of a benchmark set of conserved eukaryotic genes [BUSCO; ( Simão et al . , 2015 ) ] , each of the new choanoflagellate transcriptomes contains between 88–96% ( Supplementary file 2 ) . We also investigated the phylogenetic diversity of the choanoflagellate species we sequenced , finding it comparable to that of animals: the average phylogenetic distance between pairs of choanoflagellates was slightly larger than the phylogenetic distance between the mouse Mus musculus and the sponge Amphimedon queenslandica ( Figure 2—figure supplement 1 ) . Next , we used subsets of our data to reconstruct the phylogeny of choanoflagellates . We found that the positions of two species lying on long terminal branches ( Salpingoeca dolichothecata and Codosiga hollandica ) were poorly supported or recovered at inconsistent locations ( Materials and methods ) . Therefore , to avoid basing our comparative genomics efforts on a potentially incorrect phylogeny , and because the focus of our study was on reconstructing large-scale patterns of gene family evolution between animals and choanoflagellates , we designed our analyses to be independent of species relationships within either group . [For display purposes only , we relied on a consensus of previously published phylogenies ( Philippe et al . , 2009; Burki et al . , 2016; Carr et al . , 2017 ) ] . We next compared the choanoflagellate gene catalogs with those of diverse animals and phylogenetically relevant outgroups ( Supplementary file 3 ) to identify orthologous gene families and determine the ancestry of genes present in animals ( see Materials and methods for rationale underlying inferences of gene family orthology ) . In summary , two features that distinguish our analyses from prior reconstructions of ancestral animal gene content are ( 1 ) the additional breadth and depth provided by 19 phylogenetically diverse and newly-sequenced choanoflagellates and ( 2 ) a probabilistic and phylogenetically-informed approach designed to avoid the artificial inflation of ancestral gene content resulting from methods that rely on binary decisions for gene family presence or absence in each species while remaining independent of currently unresolved or contentious species relationships ( Figure 2—figure supplement 2 , Materials and methods ) . By grouping gene families by their phylogenetic distribution on a heat map , we were able to visualize and infer their evolutionary history , as well as their presence or absence in each species analyzed ( Figure 2 , Figure 2—figure supplement 3 , Figure 2—figure supplement 4 , Supplementary file 4; Supplementary file 5 ) . Several notable observations emerged from this visualization . First , the origins of animals , choanoflagellates , and choanozoans [the monophyletic group composed of animals and choanoflagellates ( Brunet and King , 2017 ) ] were each accompanied by the evolution of distinct sets of gene families ( i . e . , synapomorphies ) , some of which likely underpin their unique biological features . Second , the numbers of gene families gained on the animal and choanoflagellate stem lineages are roughly equivalent ( ~1944 and ~2 , 463 , respectively ) , indicating that the specific functions of novel gene families , not their quantity , were critical to the very different phenotypes each clade went on to have . Finally , although different sets of gene families can reliably be inferred to have been present in the last common ancestor of each group , gene family loss was rampant during animal and choanoflagellate diversification . [After these analyses were complete , several additional genomes from early-branching holozoans and animals became available . Incorporating them post hoc into the heat map did not substantially change any of the above observations ( Figure 2—figure supplement 5; Materials and methods ) ] . While the phenomenon of gene loss has been well documented in the evolution of animals and other eukaryotes ( Wolf and Koonin , 2013; Albalat and Cañestro , 2016; O'Malley et al . , 2016 ) , it has been unclear which extant animals retained the most gene families from the Urmetazoan . Using the Urchoanozoan and Urmetazoan gene family catalogs reconstructed in this study , we ranked extant species based on their conservation of ancestral gene families ( Figure 3 , Figure 3—figure supplement 1 ) . Compared with other animals in our study , the cephalochordate Branchiostoma floridae retains the most gene families that evolved along the animal stem lineage and also the most gene families with pre-choanozoan ancestry [extending prior observations that B . floridae preserved a comparatively large portion of the gene content of the last common ancestor of chordates ( Louis et al . , 2012 ) ] . Among the non-bilaterian animal lineages , the cnidarian Nematostella vectensis most completely retains the Urmetazoan genetic toolkit [consistent with previous findings of conservation between N . vectensis and bilaterians ( Putnam et al . , 2007; Sullivan and Finnerty , 2007 ) ] , followed by the sponge Oscarella pearsei . Importantly , B . floridae , N . vectensis , and O . pearsei each retain different subsets of the Urmetazoan gene catalog , as only two thirds ( 67% ) of the genes retained in any one of these species are found in all three species . In contrast , the more rapidly evolving ecdysozoans C . elegans , Pristionchus pacificus and Tetranychus urticae , as well as the ctenophore Mnemiopsis leidyi , retain the fewest ancestral gene families , suggesting widespread gene family loss in these lineages , although the draft nature of some of their genome assemblies and high rates of sequence evolution may artificially inflate counts of missing genes . Of the 21 choanoflagellates in our analysis , S . dolichothecata [which , despite the shared genus name , is not closely related to S . rosetta ( Carr et al . , 2017 ) ] retains the most choanozoan-specific gene families , and therefore may be relatively more informative for comparative genomic studies of animal origins than other choanoflagellate species ( Figure 3 , Figure 3—figure supplement 1 ) . Notably , the two choanoflagellate species with previously-sequenced genomes , M . brevicollis and S . rosetta , are among the choanoflagellates that have retained the fewest ancestral gene families . Thus , they are less representative of Urchoanoflagellate gene content than are most choanoflagellate species we sequenced . Indeed , several key gene families that were previously thought to be absent from choanoflagellates ( due to their absence in M . brevicollis and S . rosetta ) are conserved in S . dolichothecata and other choanoflagellates: the ancient ribonucleases Argonaute and Dicer , which are required for RNAi across eukaryotes ( Jinek and Doudna , 2009 ) , and holozoan gene families previously found in C . owczarzaki that are important for the regulation of animal development , including the transcription factors Churchill and Runx ( Sebé-Pedrós et al . , 2011 ) and a diagnostic domain for integrin β ( Sebé-Pedrós et al . , 2010 ) ( Figure 2—figure supplement 6 , Figure 3—figure supplement 2 , Supplementary file 6 , Materials and methods ) . These findings of lineage-specific gene family loss in certain animals , M . brevicollis and S . rosetta echo the more general observation that the criteria used to select species for genome sequencing also frequently select for those with streamlined genomes [e . g . , ( Gu et al . , 2005 ) ] . Gene families that originated on the stem lineage leading to animals are more likely to function in pathways or processes that distinguish animals from other eukaryotic groups . We identified ~1944 such animal-specific gene families ( Supplementary file 5 ) , including well-known developmental receptors , signaling proteins and transcription factors such as TGF-β , Hedgehog , Pax and Sox [consistent with previous reports ( Srivastava et al . , 2010; Riesgo et al . , 2014 ) ] . Notably , we detected many animal-specific gene families with no known function; 203 gene families ( 12% of total ) lack any Pfam domain , and a further 50 ( 3% ) are annotated only with Pfam domains of unknown function . The biochemical activities of these uncharacterized animal-specific gene families remain to be discovered , along with their roles in animal development and evolution . We next sought to characterize the extent to which the ~1944 gene families that originated on the animal stem lineage were subsequently retained in the 21 animal genomes we analyzed ( Table 1 ) . We found only 39 gene families that are universally conserved in all 21 animal genomes in our study; we refer to these as core animal-specific gene families . This count of core animal-specific gene families is likely to be an underestimate due to methodological tradeoffs in the genome-scale analysis that we used to identify gene families ( see Materials and methods ) . By reducing the stringency of the requirement for conservation , we identified a total of 153 gene families that were missing in no more than two animals from our data set ( i . e . , approximately 10%; Figure 3—figure supplement 3 ) , leaving ~1791 gene families that , despite being specific to animals , were lost in three or more extant lineages . In addition , recent studies in organisms not included in our genomic data set – myxozoans , a parasitic lineage of cnidarians , and glass sponges , which develop into syncytial larvae and adults – indicate that even among the 39 core animal-specific genes , some appear to be dispensable in animals with dramatically derived body plans ( Chang et al . , 2015; Schenkelaars et al . , 2017 ) . Focusing on the 39 core animal-specific gene families , we asked whether they might participate in pathways known to be critical for animal biology . Indeed , this set of genes includes seven from the Wnt pathway ( including Frizzled , Dishevelled , TCF/LEF and β-catenin ) , five involved in cell-cell adhesion ( including integrin α , laminin , and vinculin ) , and other well-known animal gene families such as JNK , caspases , and metabotropic glutamate receptors . The 39 core animal gene families also include several that are less well characterized or whose specific contributions to animal origins and animal biology are not immediately obvious , such as two subunits of the transcription-regulating Mediator complex ( Malik and Roeder , 2010 ) and the ubiquitination-associated BTB/POZ domain-containing protein 9 ( DeAndrade et al . , 2012 ) . For comparison , choanoflagellates have a similarly small set of 75 gene families ( out of ~2463 choanoflagellate-specific gene families ) that are conserved in all 21 choanoflagellate species that we sampled; 27% of these gene families encode Pfam domains related to kinase signaling ( including protein kinases , phosphatases and adapters; Supplementary file 7 ) . While novel features of animal biology might have evolved with the emergence of new gene families , the loss of ancient genes also influenced animal origins . We detected ~1645 gene families that evolved prior to the choanoflagellate-animal divergence that were retained in choanoflagellates and lost entirely from animals . These include gene families in pathways necessary for the biosynthesis of the amino acids leucine , isoleucine , valine , methionine , histidine , lysine and threonine ( Figure 2—figure supplement 7 , Supplementary file 8 ) . The shikimic acid pathway , which is required for the synthesis of the aromatic amino acids tryptophan and phenylalanine , and other aromatic compounds , was also lost along the animal stem lineage [although subsequently regained in cnidarians through horizontal gene transfer from bacteria ( Fitzgerald and Szmant , 1997; Starcevic et al . , 2008 ) ] . We thus demonstrate that components of the biosynthesis pathways for nine amino acids that are essential in animals ( Payne and Loomis , 2006; Guedes et al . , 2011 ) were lost on the animal stem lineage , and not prior to the divergence of choanoflagellates and animals . The SLN1 two-component osmosensing system , which has been shown in fungi to regulate acclimation to changes in environmental salinity ( Posas et al . , 1996 ) , is also conserved in choanoflagellates but absent in animals . [Although these amino acid synthesis and osmosensing pathway components were retained in choanoflagellates , several other gene families involved in diverse biosynthetic pathways were instead lost on the choanoflagellate stem lineage ( Supplementary file 9 ) ] . Together , the ensemble of animal gene family losses reflects the substantial changes in metabolism and ecology that likely occurred during early animal evolution . In addition to the set of gene families that evolved on the animal stem lineage , those that originated on the holozoan and choanozoan stem lineages also contributed to the genomic heritage of animals . Our increased sampling of choanoflagellate diversity allowed us to ask whether gene families previously thought to have been animal innovations , due to their absence from M . brevicollis , S . rosetta and other outgroups , may in fact have evolved before the divergence of animals and choanoflagellates . We found that ~372 gene families previously thought to be restricted to animals have homologs in one or more of the 19 newly sequenced choanoflagellates ( Supplementary file 5; see Supplementary file 10 for a list of pathways with components gained or lost on the Choanozoan stem lineage ) . Within this set of genes are the Notch receptor and its ligand Delta/Serrate/Jagged ( hereafter Delta ) , which interact to regulate proliferation , differentiation and apoptosis during animal developmental patterning ( Artavanis-Tsakonas et al . , 1999 ) . Intact homologs of these important signaling proteins have never previously been detected in non-animals , although some of their constituent protein domains were previously found in M . brevicollis , S . rosetta and C . owczarzaki ( King et al . , 2008; Suga et al . , 2013 ) . In our expanded choanoflagellate data set , we detected a clear Notch homolog in Mylnosiga fluctuans with the prototypical EGF ( epidermal growth factor ) , Notch , transmembrane and Ank ( ankyrin ) domains in the canonical order , while five other choanoflagellates contain a subset of the typical protein domains of Notch proteins ( Figure 2—figure supplement 6 , Figure 2—figure supplement 8 ) . Similarly , the choanoflagellate species S . dolichothecata expresses a protein containing both of the diagnostic domains of animal Delta ( MNNL and DSL , both of which were previously thought to be animal-specific ) and four other choanoflagellate species express one of the two diagnostic domains of Delta , but not both . The distributions of Notch and Delta in choanoflagellates suggest that they were present in the Urchoanozoan and subsequently lost from most ( but not all ) choanoflagellates , although it is formally possible that they evolved convergently through shuffling of the same protein domains in animals and in choanoflagellates . A similar portrait emerges for the cadherins Flamingo and Protocadherin ( Chae et al . , 1999; Usui et al . , 1999; Frank and Kemler , 2002 ) that were previously thought to be animal-specific , but are found in a subset of choanoflagellates within our data set ( Figure 2—figure supplement 6 , Materials and methods ) . We also found evidence that numerous gene families and pathways that originated in animals arose through shuffling of more ancient protein domains and genes that were already present in the Urchoanozoan ( Ekman et al . , 2007; King et al . , 2008; Grau-Bové et al . , 2017 ) . For example , the new choanoflagellate gene catalogs confirm that several signature animal signaling pathways , such as Hedgehog , Wnt , JNK , JAK-STAT , and Hippo , are composed of a mixture of gene families that were present in the Urchoanozoan and others that evolved later on the animal stem lineage or within animals ( Snell et al . , 2006; Adamska et al . , 2007; Hausmann et al . , 2009; Richards and Degnan , 2009; Srivastava et al . , 2010; Sebé-Pedrós et al . , 2012; Babonis and Martindale , 2017 ) ( Supplementary file 8 ) . For another animal signaling pathway , TGF-β , the critical ligand , receptor and transcription factor gene families are composed of animal-specific domain architectures ( Heldin et al . , 1997; Munger et al . , 1997 ) , although all three contain constituent protein domains that evolved on the choanozoan stem lineage ( Figure 2—figure supplement 9 ) . One surprise from our analyses was the existence in choanoflagellates of genes required for innate immunity in animals . Although innate immunity is a feature of both animal and plant biology , the receptors and pathways used by these evolutionarily distant organisms are thought to have evolved independently ( Ausubel , 2005 ) . The animal immune response is initiated , in part , when potential pathogens stimulate ( either directly or indirectly ) the Toll-like receptors ( TLRs ) , which have previously been detected only in animals ( Leulier and Lemaitre , 2008 ) . Importantly , although TLRs are found in nearly all bilaterians and cnidarians , they are absent from placozoans , ctenophores , and sponges [proteins with similar , but incomplete domain architectures have been detected in sponges ( Miller et al . , 2007; Riesgo et al . , 2014 ) ] and were therefore thought to have evolved after the origin of animals . We found that 14 of 21 sequenced choanoflagellates encode clear homologs of animal TLRs ( Figure 4a , c ) , implying that TLRs first evolved on the Urchoanozoan stem lineage ( Materials and methods ) . Animal and choanoflagellate TLRs are composed of an N-terminal signal peptide , multiple leucine rich repeat ( LRR ) domains , a transmembrane domain , and an intracellular Toll/interleukin-1 receptor/resistance ( TIR ) domain . In the canonical TLR signaling pathway , the interaction of the intracellular TIR domain of Toll-like receptors with TIR domains on adapter proteins ( e . g . , MyD88 ) initiates one of a number of potential downstream signaling cascades that ultimately lead to activation of the NF-κB transcription factor ( Janeway and Medzhitov , 2002 ) . To investigate whether the Urchoanozoan TLR might have activated a downstream signaling pathway that resembled the canonical TLR pathway in animals , we searched for TLR adapters , downstream kinases and NF-κB in choanoflagellates ( Figure 4c , Supplementary file 6 ) . While many choanoflagellates encode NF-κB , we found no evidence for two critical Death domain-containing proteins involved in the TLR-dependent activation of NF-κB: the adapter protein MyD88 ( Wiens et al . , 2005; Gauthier et al . , 2010 ) and the downstream kinase IRAK ( Song et al . , 2012 ) . However , we did detect two new classes of choanoflagellate-specific proteins that pair kinase domains directly with LRR and/or TIR domains , potentially bypassing the need to recruit kinases into multi-protein signaling complexes ( Figure 4b ) : TLR-like proteins with an intracellular kinase domain positioned between the transmembrane domain and TIR domain ( which we provisionally term ‘Kinase TLRs’ ) and proteins encoding TIR and kinase domains , but lacking a transmembrane domain ( which we provisionally term ‘Kinase TIRs’ ) . In addition , we detected homologs of the TIR-containing adapter SARM1 , a multi-functional protein that can trigger both NF-κB-dependent and NF-κB-independent responses ( Couillault et al . , 2004; Sethman and Hawiger , 2013; Liu et al . , 2014 ) . Choanoflagellate SARM1 homologs contain a conserved glutamic acid residue that is necessary for SARM1 NADase activity in animals ( Essuman et al . , 2017 ) ( Figure 4—figure supplement 1 ) . Finally , although we did not detect most animal cytosolic innate immune sensors in choanoflagellates , including the LRR-containing NLR family , ALRs , MAVS , MDA5 and RIG-I , we did find evidence for both cGAS and STING in diverse choanoflagellates [as previously reported in M . brevicollis ( Wu et al . , 2014 ) ] . Thus , critical components of the animal innate immune pathway , including both extracellular and intracellular pattern sensing receptors , predate animal origins .
Our increased sampling of choanoflagellates reveals how the Urmetazoan genome evolved as a mosaic of old , new , rearranged , and repurposed protein domains , genes and pathways . We have identified ~8418 gene families that were present in the Urmetazoan [consistent with recent findings ( Simakov and Kawashima , 2017 ) ] , about 75% of which were also present in the Urchoanozoan and the remainder of which evolved on the animal stem lineage ( Supplementary file 5 ) . The patchwork ancestry of the Urmetazoan genome is illustrated by the fact that many gene families responsible for animal development , immunity and multicellular organization evolved through shuffling of protein domains that first originated in the choanozoan stem lineage together with ancient or animal-specific domains ( e . g . the TGF-β ligand and receptor; Figure 2—figure supplement 6 , Figure 2—figure supplement 9 ) . In addition , other gene families found in the Urchoanozoan were subsequently combined into new pathways in the animal stem lineage along with newly evolved genes ( e . g . , the TLR and Hedgehog pathways; Figure 4 , Supplementary file 8 ) . Moreover , the history of the Urmetazoan genome is not simply one of innovation and co-option , as ~1 , 645 Urchoanozoan genes were lost on the animal stem lineage , including genes for the synthesis of nine essential amino acids [Figure 2—figure supplement 7 , Supplementary file 8; ( Payne and Loomis , 2006; Guedes et al . , 2011; Erives and Fassler , 2015 ) ] . A study based on similar methodology that incorporated two of the 21 choanoflagellate species analyzed here ( M . brevicollis and S . rosetta ) was recently published and found a similar pattern of gene innovation on the animal stem lineage , while also identifying many of the same core animal-specific genes ( Paps and Holland , 2018 ) . The origin of multicellularity in animals provided novel niches for bacteria to exploit , requiring the first animals to evolve new mechanisms for mediating interactions with pathogenic and commensal bacteria . In addition , the progenitors of animals interacted with bacteria – both as prey and pathogens – and the roots of animal innate immunity clearly predate animal origins . We have found that choanoflagellates express TLRs , transmembrane receptors that trigger the animal innate immune response , as well as its canonical downstream signaling target , NF-κB , suggesting that both existed in the Urchoanozoan ( Figure 4a , c ) . Like modern choanoflagellates and sponges , the Urchoanozoan likely preyed upon bacteria ( McFall-Ngai et al . , 2013; Richter and King , 2013 ) , and bacterial cues can induce life history transitions in choanoflagellates ( Alegado et al . , 2012; Woznica et al . , 2016; 2017 ) , although the mechanisms by which choanoflagellates capture bacteria and sense bacterial cues are unknown . We hypothesize that the core TLR/NF-κB pathway functioned in prey sensing , immunity , or more complex processes in the Urchoanozoan that subsequently formed the basis of a self-defense system in animals . Because critical pathway members linking TLR and NF-κB appear to be animal innovations ( e . g . , MyD88 ) , the animal signaling pathway may have evolved to diversify downstream signaling processes to tailor responses in a multicellular context . This pathway diversification may have included the evolution of roles in development , as TLRs have been implicated in both NF-κB-dependent and NF-κB-independent developmental signaling ( in addition to their functions in immunity ) in bilaterians and in the cnidarian N . vectensis ( Leulier and Lemaitre , 2008; Brennan et al . , 2017 ) . The uncharacterized choanoflagellate-specific Kinase TLRs and Kinase TIRs ( Figure 4b , c ) may function as part of a streamlined signaling pathway that mediates responses to extracellular cues , including bacteria , although further research will be required to test this hypothesis . Our study provides a detailed view of the changes in gene content that laid the foundation for animal origins . Innovations in gene and protein regulation in the Urmetazoan also likely contributed to animal evolution , as features of animal phosphoproteome remodeling , gene co-regulation and alternative splicing ( but not animal promoter types and enhancers ) have been found in C . owczarzaki ( Sebé-Pedrós et al . , 2013b; 2016a; 2016b ) and in the holozoan Creolimax fragrantissima ( de Mendoza et al . , 2015 ) . Multicellularity has also evolved independently in a number of other eukaryotic lineages , including slime molds , brown algae , fungi , and chlorophytes . In each transition to multicellularity , the underlying genomic changes are distinct from those that occurred on the animal stem lineage . For example , in the social amoeba Dictyostelium discoideum , many novel gene families are involved in extracellular sensing ( Glöckner et al . , 2016 ) , similar to the marked increase in signal transduction gene families found in the multicellular brown alga Ectocarpus siliculosus ( Cock et al . , 2010 ) . Gene innovations in multicellular ascomycete fungi are enriched for functions related to endomembrane organelles ( Nguyen et al . , 2017 ) and the gene complement of the multicellular green alga Volvox carteri is largely distinguished from its unicellular relative Chlamydomonas reinhardtii by expansions or contractions within gene families , rather than the evolution of new families ( Prochnik et al . , 2010 ) . Through our analyses of genomes and transcriptomes representing the full breadth of choanoflagellate and animal diversity , we have provided a genome-scale overview of the gene families whose invention or co-option distinguished the Urmetazoan from all other organisms and therefore may have provided a basis for the evolution of unique mechanisms regulating development , homeostasis and immunity in animals .
We acquired 18 of 20 cultures used for transcriptome sequencing from external sources ( Supplementary file 1 ) . We isolated the remaining two cultures , Acanthoeca spectabilis ( Virginia ) ATCC PRA-387 and Codosiga hollandica ATCC PRA-388 . ( A . spectabilis ( Virginia ) is a different isolate from A . spectabilis ATCC PRA-103 , which was originally collected in Australia . ) We collected the water sample from which A . spectabilis ( Virginia ) was isolated on December 19 , 2007 near Hog Island , Virginia ( GPS coordinates: 37 . 472502 , –75 . 816018 ) and we isolated C . hollandica from a sample collected on June 25 , 2008 from Madeira , Portugal ( GPS coordinates: 32 . 762222 , –17 . 125833 ) . C . hollandica was formally described in Carr et al . ( 2017 ) . We isolated choanoflagellates with a micromanipulator and a manual microinjector ( PatchMan NP 2 and CellTram vario 5176 ( Eppendorf , Hamburg , Germany ) for A . spectabilis , and XenoWorks Micromanipulator and Analog Microinjector ( Sutter Instrument , Novato , California , United States ) for C . hollandica ) . We pulled glass needles used for isolation from 1 mm diameter borosilicate glass ( GB100-10 , Science Products GmbH , Hofheim , Germany ) using a Flaming/Brown needle puller ( P-87 , Sutter Instrument ) with the following program: heat = 820 , pull = 50 , velocity = 140 , time = 44 . We polished and sterilized needles by passing them briefly through a low flame . We used a separate needle for each attempted isolation , transferring single cells into separate culture flasks containing appropriate growth medium ( see Supplementary file 1 ) . In order to reduce the possibility of contamination during the isolation procedure , we generated sterile air flow across the microscope and micromanipulator apparatus using a HEPA-type air purifier ( HAP412BN , Holmes , Boca Raton , Florida , United States ) . One culture we obtained from ATCC , Salpingoeca infusionum , was contaminated by an unidentified biflagellated unicellular eukaryote . To remove the contaminant from the culture , we counted then diluted cells into separate wells of two 24-well plates . After 7 days of growth , we found 4 of 48 wells containing only S . infusionum and bacteria , one well containing only the contaminant and bacteria , and the remaining 43 wells containing only bacteria . We selected one of the four wells containing only S . infusionum for use in transcriptome sequencing . Choanoflagellates are co-isolated with diverse bacterial prey , which serve as a food source . In order to limit bacterial diversity to the species that led to optimal choanoflagellate growth , we treated each culture with a panel of ten different antibiotics ( Supplementary file 11 ) . We obtained all antibiotics from Thermo Fisher Scientific ( Waltham , Massachusetts , United States ) with the exception of erythromycin , gentamicin , ofloxacin , and polymyxin B ( Sigma-Aldrich , St . Louis , Missouri , United States ) . We sterilized antibiotic solutions before use by filtration through a 0 . 22 µm syringe filter ( Thermo Fisher Scientific ) in a laminar flow hood . We initially treated each culture with all ten antibiotics . We selected initial treatments that decreased bacterial density and diversity , and then repeatedly diluted the cultures into fresh medium with the same antibiotic until no further change in bacterial density or diversity was observed . We then re-treated each of these cultures with an additional test of all ten antibiotics , as their modified bacterial communities often responded differently from their initial communities . We repeated successive rounds of treatment until no further reduction in bacterial density or diversity was observed ( Supplementary file 1 ) . We tested a range of concentrations of different temperatures and growth media ( Supplementary file 12 ) in order to maximize choanoflagellate cell density , with three types of nutrient sources: quinoa grains , proteose peptone/yeast extract , and cereal grass . We used filtered water ( Milli-Q , Millipore , Burlington , Massachusetts ) when preparing all solutions . For marine species , we used 32 . 9 grams per liter of artificial seawater ( Tropic Marin , Montague , Massachusetts , United States ) . We used proteose peptone ( Sigma-Aldrich Chemical ) at a final concentration of 0 . 002% ( w/v ) and yeast extract ( Becton Dickinson Biosciences , San Jose , California , United States ) at a final concentration of 0 . 0004% ( w/v ) . To prepare cereal grass media ( also known as chlorophyll alfalfa , Basic Science Supplies , St . Augustine , Florida , United States ) , we added it to autoclaved water and allowed it to steep until cool . Once cool , we removed large particles of cereal grass by repeated vacuum filtration through a ceramic Buchner funnel with a double layer of Grade one cellulose filter paper ( Whatman , GE Healthcare Life Sciences , Marlborough , Massachusetts , United States ) . We autoclaved organic quinoa grains and added them to the medium after filtration , with roughly two grains added per 25 cm2 of culture vessel surface area . We measured final nutrient content of each type of medium by Flow Injection Analysis at the University of California , Santa Barbara Marine Science Institute ( Supplementary file 3 ) . We tested buffered medium for two freshwater species that experienced lowered pH during growth , C . hollandica ( pH 5 . 5 ) and Salpingoeca punica ( pH 5 ) , using 50 mM HEPES ( Thermo Fisher Scientific ) adjusted to a pH of 7 . We sterilized all media with a 0 . 22 µm vacuum filter ( Steritop , Millipore ) in a laminar flow hood prior to use . We selected final culture conditions that maximized choanoflagellate density and variety of cell types present , as we hypothesized that different cell types , each potentially expressing different subsets of genes , would yield the greatest diversity of transcripts for sequencing . We defined five generic choanoflagellate cell types: ‘attached’: attached to the bottom of the culture vessel or to a piece of floating debris , either directly , within a theca , within a lorica , or on a stalk; ‘slow swimmer’: a typical swimming cell; ‘fast swimmer’: a cell with reduced collar length swimming at higher speed; ‘free-swimming colonial’: in a colony swimming in the water column; ‘attached colonial’: in a colony attached to a stalk; ‘passively suspended’: suspended in the water column , either naked or within a lorica . See ( Dayel et al . , 2011; Carr et al . , 2017 ) for further information on choanoflagellate cell types and ( Leadbeater , 2015; Richter and Nitsche , 2016 ) for descriptions of extracellular structures ( thecae , loricae , etc . ) . We routinely grew choanoflagellates in 25 cm2 angled neck cell culture flasks with 0 . 2 µm vented caps ( Corning Life Sciences , Corning , New York , United States ) containing 25 ml of medium . We performed all cell culture work in a laminar flow hood . To reduce the possibility of cross-contamination among samples in the hood , we dispensed media into growth vessels prior to the introduction of cultures , we only worked with a single culture at a time , and we cleaned thoroughly with 70% ethanol before and after introducing cultures . To grow and collect large amounts of cells for RNA preparation , we used different growth vessels , volumes , growth durations and centrifugation times as appropriate to each culture ( Supplementary file 1 ) . Vessels included long neck Pyrex glass culture flasks ( Corning ) , 150 mm plastic tissue culture dishes ( Becton Dickinson ) , and 75 cm2 angled neck cell culture flasks with 0 . 2 µm vented caps ( Corning ) . We harvested cultures depending on the cell types present ( Supplementary file 1 ) . For cultures with five percent or fewer attached cells , we collected liquid by pipetting without scraping ( to reduce the number of bacteria collected ) . For cultures containing between 5 and 90 percent attached cells , we harvested single plates by pipetting after scraping cells from the bottom of the plate . For cultures with 90 percent or greater attached cells , we combined multiple plates as follows: we removed and discarded the liquid from the first plate by pipetting , added 50 ml of either artificial sea water or filtered water , as appropriate , scraped cells from the plate , removed the liquid from the second plate , transferred the liquid from the first to the second plate , and repeated the procedure on subsequent plates . For cultures containing quinoa grains or large bacterial debris , we filtered with a 40 µm cell strainer ( Fisher ) after collection . We pelleted cells in 50 ml conical tubes at 3220 x g in a refrigerated centrifuge at 4°C , removed the first 47 . 5 ml of supernatant by pipetting , and the last 2 . 5 ml by aspiration . When we harvested more than 50 ml for a culture , we spun in separate tubes , removed all but 2 . 5 ml of supernatant , resuspended , combined into a single 50 ml conical tube , and repeated the centrifugation as above . We flash froze pellets in liquid nitrogen and stored them at −80°C . To reduce the possibility of cross-contamination , we harvested and centrifuged each culture separately , we used disposable plastic pipette tubes , conical tubes , and cell scrapers , and we cleaned all other material ( bench tops , pipettes , etc . ) with ELIMINase ( Decon Laboratories , King of Prussia , Pennsylvania , United States ) between cultures . We isolated total RNA from cell pellets with the RNAqueous kit ( Ambion , Thermo Fisher Scientific ) . We modified the manufacturer’s protocol to double the amount of lysis buffer , in order to increase RNA yield and decrease degradation . We performed both optional steps after adding lysis buffer: we spun for 3 min at top speed at 1°C and passed the supernatant through a 25 gauge syringe needle several times . We used the minimum suggested volumes in the two elution steps ( 40 and 10 µl ) . We measured RNA concentration using a NanoDrop ND-1000 spectrophotometer ( Thermo Fisher Scientific ) . For all species except C . hollandica , we immediately proceeded to digest genomic DNA using the TURBO DNA-free kit ( Ambion , Thermo Fisher Scientific ) , following the manufacturer’s protocol with a 30 min incubation . After digestion , we removed DNase with DNase Inactivation Reagent for all species except S . punica , whose RNA extract was incompatible with the reagent . We instead removed DNase from S . punica by extracting with two volumes of pH eight phenol:chloroform:isoamyl alcohol , removing residual phenol with two volumes of chloroform:isoamyl alcohol , and precipitating with 0 . 3 M sodium acetate pH 5 . 2 ( all three from Sigma-Aldrich ) , 25 µg of GlycoBlue ( Thermo Fisher Scientific ) as a carrier and two volumes of 100% ethanol . We washed the pellet in 70% ethanol and resuspended in 50 µl of nuclease-free water . For Didymoeca costata , after DNase removal with the Inactivation Reagent , the RNA still appeared to be slightly contaminated with protein , so we performed a pH eight phenol:chloroform extraction to remove it . For C . hollandica , we observed significant total RNA degradation in the presence of DNase buffer . Instead , to remove genomic DNA we performed three successive rounds of extraction with pH 4 . 5 phenol:chloroform:isoamyl alcohol ( Sigma-Aldrich ) , followed by the chloroform:isoamyl and precipitation steps described above . To reduce the possibility of cross-contamination among samples , we performed RNA isolation and DNase digestion for a single culture at a time , we used disposable materials when possible , and we cleaned all other materials ( bench tops , centrifuges , dry baths , pipettes , etc . ) thoroughly with ELIMINase before use . We evaluated total RNA on Bioanalyzer 2100 RNA Pico chips ( Agilent Technologies , Santa Clara , California , United States ) with four criteria to be considered high-quality: ( 1 ) four distinct ribosomal RNA peaks ( 16S and 23S for bacteria , 18S and 28S for choanoflagellates ) , ( 2 ) low signal in all other regions , as a non-ribosomal signal is evidence of degradation , ( 3 ) at least a 1:1 ratio of 28S ribosomal area to 18S ribosomal area , since 28S ribosomal RNA is likely to degrade more easily than is 18S ribosomal RNA , and ( 4 ) an RNA Integrity Number ( RIN ) of 7 or greater ( Schroeder et al . , 2006 ) . ( We note that the Bioanalyzer software could not calculate RIN for several cultures . ) If we were not able to obtain high-quality total RNA after the first attempt for any culture , we repeated cell growth , centrifugation and total RNA isolation up to a maximum of 5 times , and selected the best available total RNA sample to use for transcriptome sequencing . We produced a rough estimate of the amount of choanoflagellate total RNA present in each sample by calculating the ratio of choanoflagellate to bacterial ribosomal RNA peaks ( 18S vs . 16S and 28S vs . 23S ) and multiplying the resulting fraction by the total amount of RNA present in the sample ( Supplementary file 1 ) . The standard library preparation protocol for Illumina mRNA sequencing used poly-dT beads to separate polyadenylated mRNA from other types of non-polyadenylated RNA such as rRNA and tRNA . For choanoflagellates , the bead selection step also served to separate choanoflagellate mRNA from bacterial RNA ( which is not polyadenylated ) . Because the amount of bacterial RNA isolated from a culture often exceeded the amount of choanoflagellate RNA by one to several orders of magnitude , we reasoned that the standard bead selection might not be sufficient . We tested this hypothesis on S . rosetta Px1 , a culture grown with a single species of bacterium , Algoriphagus machipongonensis . Because both species have sequenced genomes ( Alegado et al . , 2011; Fairclough et al . , 2013 ) , we could identify the origin of sequenced RNA using a straightforward read mapping procedure . We cultivated S . rosetta Px1 cells as described previously ( Fairclough et al . , 2010 ) . We scraped and harvested 50 ml of culture after three days of growth in a 150 ml tissue culture dish . We performed centrifugation ( with a 10 min spin ) , RNA isolation , DNase digestion and total RNA quality assessment as described above . We compared the standard Illumina TruSeq v2 mRNA preparation protocol , which performs two rounds of polyA selection with a single set of poly-dT-coated beads , against a modified protocol that repeats the polyA selection steps , for a total of four rounds of polyA selection with two sets of beads . For all other aspects of library preparation , we followed the manufacturer’s protocol . We quantified libraries by qPCR ( Kapa Biosystems , Sigma-Aldrich ) and sequenced them on a HiSeq 2000 machine ( Illumina , San Diego , California , United States ) at the Vincent J . Coates Genomics Sequencing Laboratory at the California Institute for Quantitative Biosciences ( Berkeley , California , United States ) . We generated 16 , 970 , 914 single-end 50 bp reads for the library prepared with two rounds of polyA selection , and 17 , 182 , 953 for the four-round library . We mapped reads to the S . rosetta and A . machipongonensis genomes using BWA version 0 . 6 . 1 ( Li and Durbin , 2009 ) and SAMtools version 0 . 1 . 18 ( Li et al . , 2009 ) with default parameter values . We counted reads mapping to S . rosetta ribosomal loci on supercontig 1 . 8 ( 5S: positions 1900295–1900408 , 18S: 1914756–1916850 and 28S: 1917502–1923701 ) . The number of reads mapping to the non-ribosomal portion of the S . rosetta genome did not differ substantially between the two data sets: 12 , 737 , 031 reads mapped for the two-round data , and 12 , 585 , 647 for the four-round data . Similarly , 10 , 509 , 262 reads from the two-round data mapped to S . rosetta transcripts and 10 , 181 , 522 for the four-round data . We also asked whether additional rounds of polyA selection would cause increased RNA breakage due to pipetting or heating during the selection process [e . g . , ( Kingston , 2001 ) ] , leading to lower coverage of the 5’ ends of transcripts ( because the poly-dT sequence binds to the 3’ end of RNA molecules ) . We estimated the loss of 5’ transcript ends due to shear to affect less than roughly 5% of transcripts ( Figure 1—figure supplement 2a ) . The four-round method removed roughly an order of magnitude more non-polyadenylated RNA than the two-round method ( Figure 1—figure supplement 2b ) . We observed that the four-round data set had a slightly lower overall read quality . To address this , we tested whether a difference in read quality between the two data sets could account for the difference in read mapping by randomly resampling the two-round data set to contain the same number of either Phred-like quality 20 or Phred-like quality 30 bases as the four-round data set , but neither resampling affected the results . We also tested whether transcript assembly quality would suffer in the four-round data set by assembling both data sets de novo with Trinity release 2012-03-17 ( Grabherr et al . , 2011 ) with default parameter values and mapping the resulting contigs to the S . rosetta genome using BLAT version 34 ( Kent , 2002 ) with default parameter values , but we observed no substantial difference between the two data sets . Given the superior ability of four rounds of polyA selection to remove contaminating bacterial RNA with little to no loss of transcript coverage , we adopted this methodology for subsequent transcriptome sequencing . The raw sequence reads for this experiment are available at the NCBI Short Read Archive with the BioProject identifier PRJNA420352 . We began the Illumina TruSeq v2 mRNA library preparation protocol with approximately 2 µg of total RNA per culture , if available ( Supplementary file 1 ) . We performed four rounds of polyA selection ( instead of the standard two ) and introduced two further modifications to the standard protocol: first , we repeated the Agencourt AMPure XP ( Beckman Coulter , Indianapolis , Indiana , United States ) bead clean-up step to enhance adapter removal , and second , we used 1 . 5 µl less volume in all bead elution steps , in order to reduce loss during the protocol . We prepared libraries from 5 RNA samples at a time , and the libraries were later multiplexed into groups of 6 or seven per sequencing lane . To allow us to detect evidence of potential cross-contamination during either process , we ensured that the groupings for sample preparation and sequencing were distinct ( Supplementary file 1 ) . We estimated library concentration using the Qubit dsDNA HS Assay ( Thermo Fisher Scientific ) and determined quality and fragment size distribution with a Bioanalyzer 2100 High Sensitivity DNA chip ( Agilent ) . We quantified libraries by qPCR ( Kapa Biosystems , Sigma-Aldrich ) and sequenced them on an Illumina HiSeq 2000 at the Vincent J . Coates Genomics Sequencing Laboratory at the California Institute for Quantitative Biosciences ( Berkeley , California , United States ) . One group of libraries was sequenced twice ( consisting of A . spectabilis , Diaphanoeca grandis , Helgoeca nana , S . helianthica , S . infusionum and Salpingoeca urceolata ) due to a drop-off in quality after base 50 on the forward read of sequencing pairs; quality scores up to base 50 on the forward read and on reverse reads were not affected . The second , repeat sequencing run did not experience this issue . We incorporated both sequencing runs for affected libraries into subsequent analyses ( including Quality trimming and Error correction , see below ) . We produced between 23 million and 61 million paired-end 100 bp sequencing reads per library ( Supplementary file 1 ) . Raw sequence reads are available at the NCBI Short Read Archive with the BioProject identifier PRJNA419411 ( accession numbers for each species are listed in Supplementary file 1 ) . We trimmed sequence reads using Trimmomatic version 0 . 30 ( Lohse et al . , 2012 ) with two separate filters: ( 1 ) removal of TruSeq adapter sequence and ( 2 ) trimming very low quality bases from the ends of each read . To implement these filters , we ran Trimmomatic in three phases . In the first phase , we clipped palindromic adapters using the directive ILLUMINACLIP:2:40:15 and discarded resulting reads shorter than 25 bases with MINLEN:25 . This resulted in two data sets: one containing reads whose mate pair remained in the set , and the other composed of reads whose pair was removed due to adapter contamination . In the second phase , we clipped simple adapters from the remaining paired data set using the directive ILLUMINACLIP:2:40:15 , imposed a minimum Phred-like quality cutoff of 5 on the first ten and last ten bases using LEADING:5 and TRAILING:5 , subjected the read to a minimum sliding window quality using SLIDINGWINDOW:8:5 and discarded resulting reads shorter than 25 bases with MINLEN:25 . The third phase operated on the remaining unpaired reads from the first phase , and implemented the same directives as the second phase . We used a permissive minimum quality of 5 in order to remove very low quality bases , as these might interfere with read error correction in the subsequent processing step . We discarded reads less than 25 in length because they were shorter than the k-mer size of the Trinity assembler ( see De novo transcriptome assembly below ) . In all adapter clipping operations , we used sequences appropriate to the index used for multiplexed sequencing . The number of sequence reads and total bases remaining after trimming for each library are given in Supplementary file 1 . We performed error correction on trimmed reads using Reptile v1 . 1 ( Yang et al . , 2010 ) following the authors’ instructions , with the modifications described below . We began by using the ‘fastq-converter . pl’ script to convert from FASTQ and to discard reads with more than one ambiguous character ‘N’ in any window of 13 bases . For reads with one ‘N’ , we chose the character ‘a’ as the substitute for ‘N’ , as all of the characters in our input reads were in upper case ( A , C , G , or T ) ; thus , we could later recognize ‘N’ bases converted in this step . Next , we tuned parameters using the ‘seq-analy’ utility following the authors’ instructions , in four steps: ( 1 ) Running ‘seq-analy’ with default settings . ( 2 ) Adjusting the input settings to ‘seq-analy’ using the results from step 1 . For all species , we set MaxBadQPerKmer to eight and KmerLen to 25 ( to match the k-mer length used in Trinity ) . ( 3 ) Re-running ‘seq-analy’ using the adjusted input settings . ( 4 ) Creating the input settings to ‘Reptile’ based on the output of step 3 . We set KmerLen to 13 and step to 12 for all species . The values of QThreshold , T_expGoodCnt , T_card and Qlb differed by species ( Supplementary file 1 ) . All other parameters were left at their defaults to run Reptile . We noticed that the locations within reads of errors identified by Reptile fell into two general classes: sporadic errors not located adjacent to any other error , and clustered errors , in which several adjacent bases within the same k-mer window were identified as errors . In some extreme cases , every single base within a sequence read was identified as a target for error correction; we observed this phenomenon in the set of read corrections for every species . We reasoned that this was an unintended consequence of the iteration-to-exhaustion approach ( step 2d ) of the Reptile algorithm . Therefore , we designed a method to correct sporadic errors , but not clustered errors . For each species , we began by grouping each read according to the total number of errors identified . Within each group , we built a distribution of the number of adjacent errors within the same k-mer window . For sporadic errors , this number should be close to 0 , but for clustered errors , the number could be up to the k-mer size minus one . There was a clear pattern within each of these distributions , with some errors identified with no neighbors ( sporadic errors ) , a smaller number identified with one neighbor , and an increasing number beginning at two or more neighbors ( clustered errors ) . We used these empirical distributions to set the maximum allowable amount of neighboring errors within a k-mer window as the count just prior to the beginning of the secondary increase within each distribution . For example , for D . grandis , in the case of the group of reads containing four total identified errors , there were 316 , 103 errors with no neighbors within the same k-mer , 197 , 411 with one neighbor , 156 , 043 with two neighbors , and 353 , 639 with three neighbors ( that is , all four errors were within the same k-mer window ) . Thus , for the group of reads containing four total identified errors in D . grandis , we only corrected errors with up to two neighboring errors in the same k-mer window . After running Reptile error correction of sequence reads and quality files subject to these cutoffs , we performed a final step of restoring ambiguous bases converted by ‘fastq-converter . pl’ ( from ‘N’ to ‘a’ ) that were not subsequently corrected by Reptile back to their original value of ‘N’ . We performed de novo transcriptome assembly on trimmed , corrected sequence reads and quality files with Trinity release 2013-02-25 ( Grabherr et al . , 2011 ) with ‘--min_contig_length’ set to 150 and all other parameters at their default values . We chose a minimum contig length of 150 ( rather than the default of 200 ) so as not to exclude assembly fragments that might encode predicted proteins with lengths between 50 and 66 amino acids , because some domains in the Pfam database are as short as 50 amino acids . Because none of the species we sequenced had an available genome assembly , we did not know whether transcripts might be encoded in overlapping positions within the genome . To test this possibility , we repeated each Trinity assembly with the addition of the ‘--jaccard-clip’ option and compared the estimated number of fusion transcripts predicted by Transdecoder release 2012-08-15 ( Haas et al . , 2013 ) . We found essentially no difference in the number of predicted fusion transcripts between the original and ‘--jaccard-clip’ assemblies , and so we continued with the original assemblies . Assembly statistics are reported in Supplementary file 1 . Cross-contamination within a multiplexed Illumina sequencing lane is estimated to cause incorrect assignment of roughly 0 . 5% of index pairs ( Kircher et al . , 2012 ) . We designed a procedure to eliminate transcriptome assembly contigs resulting from incorrect index assignments . We ran blastn version 2 . 2 . 26 ( Altschul et al . , 1997 ) with a maximum E value of 1 × 10−10 to query contigs from each species against all other species . Because of the evolutionary distances among the choanoflagellates we sequenced ( Figure 2—figure supplement 1 ) , most contigs from one species had no matches in any other species . Within the contigs that did have cross-species matches ( Figure 1—figure supplement 3a ) , we observed a large number that were identical or nearly-identical , which were likely cross-contaminants , and another set of matches distributed around roughly 80% identity , likely representing highly conserved genes . The two cases were separated at roughly 96% identity . After exploring the distribution of match lengths in a similar manner ( Figure 1—figure supplement 3b ) , we considered matches at 96% or greater identity of at least 90 bases in length to be cross-contaminated . Next , we identified the sources of cross-contaminated contigs by comparing the number of reads mapping from both species for each match . We first masked contigs with Tandem Repeats Finder version 4 . 04 ( Benson , 1999 ) , with the following parameter values: match = 2 , mismatch = 7 , indel penalty = 7 , match probability = 80 , mismatch probability = 10 , min score = 30 , max period = 24 . We next mapped reads to masked contigs using the Burroughs-Wheeler Aligner , BWA , version 0 . 7 . 5a ( Li and Durbin , 2009 ) and SAMtools version 0 . 1 . 18 ( Li et al . , 2009 ) . We ran BWA ‘aln’ with the ‘-n 200’ option to allow up to 200 equally best hits to be reported , and all other parameter values left at their defaults . Based on the distribution of read mapping ratios between the pair of species matching for each cross-contaminated contig ( Figure 1—figure supplement 3c ) , we retained only contigs for the species in a pair with 10 times or more reads mapping , and discarded all other contigs , with one exception: if a contig had at least 10 , 000 reads mapping , we did not discard it , regardless of read mapping ratio . We observed that many contigs encoding conserved genes ( for example , α-tubulin and elongation factor 1α ) also tended to be the most highly expressed , and thus the read mapping ratio was often close to one for these contigs . We identified as cross-contaminated and removed between 1 . 7 and 8 . 8% of contigs for each species ( Supplementary file 1 ) . We note that our procedure would also be expected to discard sequences from any bacterial species that were present in two different choanoflagellate cultures . For a more detailed examination of the cross-contamination removal process , see ( Richter , 2013 ) . We predicted proteins from decontaminated contigs with Transdecoder release 2012-08-15 ( Haas et al . , 2013 ) with a minimum protein sequence length of 50 . We noticed that many of the proteins originating from different contigs within a species were completely identical along their entire length . Furthermore , we also observed many contigs whose predicted proteins were a strict subset of the predicted proteins from another contig . For example , contig one might encode predicted proteins A and B , and contig two might encode two predicted proteins exactly matching A and B , and a third predicted protein C . We removed both types of redundancy ( exact matches and subsets ) from the set of predicted proteins , and we also removed the contigs from which they were predicted ( Supplementary file 1 ) . To estimate expression levels , we mapped sequence reads to decontaminated , non-redundant , Tandem Repeats-masked contigs using the Burroughs-Wheeler Aligner , BWA , version 0 . 7 . 5a ( Li and Durbin , 2009 ) . We ran BWA ‘mem’ with the ‘-a’ option to report all equally best hits , with all other parameter values left at their defaults . We converted BWA output to BAM format using SAMtools version 0 . 1 . 18 ( Li et al . , 2009 ) and ran eXpress version 1 . 4 . 0 ( Roberts and Pachter , 2013 ) with default parameter values in order to produce estimated expression levels , in fragments per kilobase per million reads ( FPKM ) . The distribution of FPKM values ( Figure 1—figure supplement 3d ) had a peak near 1 , with steep decreases in the number of contigs at lower values . Therefore , we chose an extremely conservative noise threshold two orders of magnitude below the peak , at FPKM 0 . 01 , and discarded contigs ( and their associated predicted proteins ) below this value ( Supplementary file 1 ) . The final sets of contigs are available as Dataset 1 , and the proteins as Dataset 2 ( Richter et al . , 2018 ) . FPKM values for contigs are given in Dataset 3 ( Richter et al . , 2018 ) . We also submitted the final sets of contigs and proteins to the NCBI Transcriptome Shotgun Assembly ( TSA ) sequence database . The contigs ( and their associated proteins ) available in the TSA differ from our final sets in three ways: ( 1 ) The TSA database does not accept contigs with lengths less than 200 , whereas our minimum was 150; ( 2 ) The submission system identified and rejected a small number of contigs as bacterial contaminants; ( 3 ) The submission system identified and required us to trim a small number of contigs to remove Illumina adapter sequences that were missed by our screen with Trimmomatic . Differences between the assemblies we analyzed and those submitted to NCBI TSA are summarized in Supplementary file 1 , and a complete list of affected contigs can be found in Dataset 9 ( Richter et al . , 2018 ) . To determine whether the conserved gene content of the transcriptomes we produced was similar to the two sequenced choanoflagellate genomes , we searched our data for conserved eukaryotic proteins with BUSCO version 3 . 0 . 2 ( Simão et al . , 2015 ) . We used default parameter values and the 303 BUSCOs from the ‘eukaryota_odb9’ set , and performed searches with HMMER version 3 . 1b2 ( Eddy , 2011 ) . We note that each final transcriptome assembly contained between 18 , 816 and 61 , 053 proteins per species ( Supplementary file 1 ) , markedly more than the 9196 and 11 , 629 genes predicted , respectively , from the assembled genomes of M . brevicollis ( King et al . , 2008 ) and S . rosetta ( Fairclough et al . , 2013 ) . The relatively higher protein counts predicted from choanoflagellate transcriptomes likely represent an overestimate resulting from the inherent complexities in assembling unique contig sequences from short read mRNA sequencing data in the absence of a reference genome ( Grabherr et al . , 2011 ) , including the fact that sequence reads from different splice variants may have assembled into separate contigs while being encoded by the same gene . Because our goal was to reconstruct the full diversity of genes in the Urchoanozoan and Urmetazoan , the tendency of transcriptomes to yield overestimates of gene numbers was not a significant concern . To confirm the identity of the choanoflagellate cultures we sequenced , we compared our transcriptome data to seven protein coding genes with choanoflagellate sequences previously available in GenBank: actin , alpha tubulin , beta tubulin , EF-1A , EFL , HSP70 and HSP90 . To search the transcriptomes for each gene , we downloaded all previously available choanoflagellate sequences from GenBank , aligned them using FSA version 1 . 15 . 7 ( Bradley et al . , 2009 ) with default parameter values , and trimmed the alignments using Gblocks ( Talavera and Castresana , 2007 ) with allowed gap positions set to ‘half’ and all other parameter values set to their most permissive . We next built HMMs for each trimmed alignment using hmmbuild from the HMMer package version 3 . 0 ( Eddy , 2011 ) and searched the contigs from each transcriptome ( and their reverse complements ) with hmmsearch , both with default parameter values . In each case , the top hit we retrieved for each transcriptome ( the contig with the lowest E value ) matched the corresponding sequence for the species in GenBank . We tested for the possibility of animal contamination within choanoflagellate transcriptomes using the same set of seven HMMs to search the nucleotide coding sequences from all 59 species in our data set . For each target species , we retained the top three hits by E value . For each gene , we aligned the resulting sequences using using MAFFT version 7 . 130b ( Katoh and Standley , 2013 ) with the parameters ‘--maxiterate 1000 --localpair’ and trimmed alignments using trimAl version 1 . 2rev59 ( Capella-Gutiérrez et al . , 2009 ) with the parameter ‘-gt 0 . 8’ . We built phylogenetic trees with RAxML version 8 . 2 . 0 ( Stamatakis , 2014 ) with the options ‘-m GTRCAT’ to define the model of DNA substitution and ‘-f a -N 100’ option for a rapid bootstrap analysis with 100 bootstraps . We collapsed branches with bootstrap support below 50 using Archaeopteryx version 0 . 9813 ( Han and Zmasek , 2009 ) . We observed no instances with ≥50% bootstrap support in which a choanoflagellate sequence was nested within a clade of animal sequences ( Figure 2—source data 1 ) , indicating that none of the choanoflagellate transcriptome sequences we retrieved appeared to be of animal origin . To identify gene families , we chose a set of representative animals and outgroup species ( Supplementary file 3 ) . We used the 19 choanoflagellates we sequenced and the complete predicted protein sets from the M . brevicollis ( King et al . , 2008 ) and S . rosetta genomes ( Fairclough et al . , 2013 ) . As an internal control , we had sequenced two independent isolates of Stephanoeca diplocostata , whose predicted proteins we compared using CD-HIT version 4 . 5 . 4 ( Li and Godzik , 2006 ) with default parameter values . We found that the two S . diplocostata isolates contained essentially equivalent predicted protein sets , so we used only the French isolate for constructing orthologous groups . We compared the 21 choanoflagellate species to 21 representative animals with genome-scale sequence data available , with an emphasis on early-branching non-bilaterians: sponges , ctenophores , T . adhaerens and cnidarians . We included 17 outgroups with sequenced genomes in our analysis: C . owczarzaki , a holozoan representative of the closest relatives of animals and choanoflagellates , five fungi chosen to represent fungal diversity , two amoebozoa , and nine species representing all other major eukaryotic lineages . The predicted proteins of the chytrid Homolaphlyctis polyrhiza were released in two partially redundant sets , which we combined using CD-HIT version 4 . 5 . 4 with default parameter values . We applied OrthoMCL version 2 . 0 . 3 ( Li et al . , 2003 ) to construct gene families ( orthologous groups of proteins ) using recommended parameter values . As recommended in the OrthoMCL documentation , we performed an all versus all sequence homology search using blastp version 2 . 2 . 26 ( Altschul et al . , 1997 ) with an expectation value of 1 × 10−5 , and we determined orthologous groups with MCL-edge version 12–068 ( Enright et al . , 2002; van Dongen and Abreu-Goodger , 2012 ) using an inflation parameter of 1 . 5 . Our goal in constructing gene families was to identify groups composed of orthologous proteins . Although numerous approaches are currently available , no existing algorithm can yet identify all orthologous genes while perfectly separating orthologs from spurious BLAST hits ( Quest for Orthologs consortium et al . , 2016 ) . We chose OrthoMCL due to its widespread use in previous studies , including analyses of animal gene family evolution based on the genomes of S . rosetta and M . brevicollis ( Fairclough et al . , 2013 ) , and due to its relative balance of sensitivity and specificity in comparison to other algorithms ( Altenhoff and Dessimoz , 2009 ) . Spurious BLAST hits within gene families pose a particular problem for ancestral gene content reconstruction . Approaches based on binary presence/absence calls , originally developed for morphological characters where the probability of homoplasy is considered to be much lower ( Farris , 1977 ) , may be vulnerable to yielding hyperinflated estimates of the gene content of the Urchoanozoan , since any gene family containing at least one protein from animals and one from choanoflagellates could be inferred to have been present in the Urchoanozoan . Indeed , we observed a number of gene families containing many animals but only one or two choanoflagellates , and vice versa ( for example , there were 130 gene families containing one choanoflagellate and ≥15 animal species; Figure 2—figure supplement 2a ) . If these gene families represented true orthologous groups and were therefore present in the ancestral Choanozoan , they would require several independent loss events within one of the two groups; we reasoned that it was more parsimonious that some proportion of these genes evolved in only one of the two groups and that the isolated BLAST hits from the other group represented false positives . To address this problem , we produced membership probabilities for each protein within each gene family , based on its average BLAST E value to all other proteins within the gene family ( the absence of a BLAST hit between a pair of proteins was treated as the maximum possible ( i . e . , least probable ) E value ) . A true ortholog should have low E value hits to nearly all other members of its gene family , resulting in a low average E value . In contrast , a false ortholog should have higher E value hits to only one or a few members of the gene family , resulting in a high average E value . If there were multiple proteins from the same species in a gene family , we chose the protein with the lowest average BLAST E value . To produce probabilities from average E values , we rescaled them using the empirical cumulative density function ( van der Vaart , 1998 , p . 265 ) of all E values from the initial all versus all homology search . We ordered E values from highest to lowest to produce a probability between 0 and 1 ( Figure 2—figure supplement 2b ) . Within the resulting probabilities , a protein with low E value BLAST hits to all other members of its gene family had a probability close to 1 , and a protein with high E value BLAST hits to only a few members of its gene family had a probability close to 0 ( Figure 2—figure supplement 2c–d ) . We found that most proteins fell close to one of these extremes , and that the probabilities clearly distinguished hits between proteins within the same gene family versus hits between proteins in two different gene families ( Figure 2—figure supplement 2e ) . In addition , proteins within gene families identified above containing many choanoflagellates and many animals had probabilities closer to 1 , whereas some proteins within gene families with few choanoflagellates and many animals were assigned low probabilities , which are likely to be false orthologs , and some were assigned high probabilities likely representing true orthologs ( Figure 2—figure supplement 2f ) . We conclude that this approach was able to identify and remove a substantial proportion of false orthologs , resulting in a set of gene families highly enriched for ( although still not entirely composed of ) truly orthologous families . We tested the effectiveness of our approach to screen out false positive orthologs by building a phylogenetic tree using gene family 9066 as an example . This gene family contains numerous animal sequences and a single choanoflagellate sequence , which was assigned a probability of 0 . 03 for gene family membership due to its high average blastp E value to other members of the family ( Figure 2—figure supplement 2c ) . We first aligned the protein sequences in gene family 9066 using MAFFT version 7 . 130b ( Katoh and Standley , 2013 ) with the parameters ‘--maxiterate 1000 --localpair’ and trimmed alignments using trimAl version 1 . 2rev59 ( Capella-Gutiérrez et al . , 2009 ) with the parameter ‘-gt 0 . 5’ . We next built an HMM for the trimmed alignment using hmmbuild from the HMMer package version 3 . 0 ( Eddy , 2011 ) and searched the full protein complement of each species using hmmsearch , both with default parameter values . For each species , we retrieved the top hit among sequences not present in gene family 9066 . We aligned the sequences found by hmmsearch together with those from gene family 9066 using MAFFT and trimAl as above , modifying the trimAl missing sequences threshold to be ‘-gt 0 . 25’ , in order to include parts of the alignment only present in sequences from gene family 9066 ( which represented slightly more than one quarter of the sequences included in the alignment ) . We built phylogenetic trees with RAxML version 8 . 2 . 0 ( Stamatakis , 2014 ) with the options ‘-m PROTGAMMALGF’ and ‘-f a -N 100’ . We visualized the resulting tree and collapsed branches below 50% bootstrap support using the Interactive Tree of Life ( iTOL ) web site ( Letunic and Bork , 2016 ) . The choanoflagellate protein in gene family 9066 is more closely related to proteins outside the gene family than to proteins inside the family ( Figure 2—figure supplement 10 ) , and thus was correctly identified as a false positive by our probability assignment method . Protein sequences for gene families are available as Dataset 4 , and presence probabilities by species are listed as part of Dataset 5 ( Richter et al . , 2018 ) . For each gene family , we calculated the sum of membership probabilities for species from each of the five major groups in this study ( choanoflagellates , animals , C . owczarzaki , fungi , and other eukaryotes ) . Based on the resulting distribution ( Figure 2—figure supplement 3a ) , for a gene family to be considered present in a major group , we required this sum to be ≥10% of the number of species in the major group ( stated alternatively , we required the average presence probability in the group to be ≥10% ) . Notably , the 10% threshold was independent of the tree topology within each major group , thereby minimizing the impact of currently unresolved within-group species relationships [e . g . , among early-branching animals ( King and Rokas , 2017 ) ] . For choanoflagellates and animals , each of which have 21 species in our data set , this equates , for example , to a gene family represented at high probability ( ≥70% , corresponding to an average BLAST E value of 1 × 10−20; Figure 2—figure supplement 2a ) in three or more species , or at 35% probability ( E value of 1 × 10−10 ) in six or more species . Therefore , the only truly orthologous gene families likely to be excluded by this threshold are those with weak homology ( i . e . , average BLAST E values > 10−10 ) to only a few species in a major group , a rare case which would also require numerous independent losses of the gene family within the group . Next , we developed a set of parsimony-based rules ( Supplementary file 4 ) to determine the origin of each gene family . If a gene family was present within two separate major groups , we considered it to have been present in their last common ancestor . For gene families containing proteins from species in only one major group , we considered it to have been present in the last common ancestor of that major group only if it was present ( i . e . , satisfying the 10% probability threshold ) within two or more separate sub-groups ( Figure 2—figure supplement 3b ) . Within animals , we defined three sub-groups: sponges ( O . pearsei , Ephydatia muelleri , A . queenslandica ) , ctenophores ( M . leidyi ) , and later-branching animals ( T . adhaerens , cnidarians , bilaterians ) . An animal-specific gene family present in at least two of these three sub-groups would be considered to have been present in the Urmetazoan ( and thus to have evolved on the animal stem lineage ) ; this inference would not change if either sponges or ctenophores were the earliest-branching animals . Similarly , a choanoflagellate-specific gene family was required to be present within both craspedids and loricates [two well-defined groups of species Leadbeater , 2015; Carr et al . , 2017 ) ] to be considered present in the Urchoanozoan . Inferred group presences for each gene family are available in Dataset 5 ( Richter et al . , 2018 ) . In the text , we precede estimated gene family counts with a tilde . We selected a conservative probability threshold of 10% in order to minimize the number of gene families that might erroneously be assigned as specific to a given group ( e . g . , animals or choanozoans ) . This choice may have resulted in some gene families that are truly specific to a group instead being incorrectly assigned as shared with another group . As a consequence , counts of gene families originating in different groups should represent relatively conservative estimates . To produce a heat map for visual display of gene families , we ordered them by their patterns of presence within the five major species groups . Within a given pattern ( for example , absent in outgroups and fungi but present in C . owczarzaki , animals and choanoflagellates ) we clustered gene families by Pearson correlation using Cluster 3 . 0 ( de Hoon et al . , 2004 ) , with all other parameter values set to their defaults . We visualized heat maps using Java TreeView version 1 . 1 . 6r4 ( Saldanha , 2004 ) and color palettes from ColorBrewer ( Harrower and Brewer , 2003 ) . For display purposes , we restricted Figure 2 to gene families inferred to have been present in at least one of six nodes representing last common ancestors of interest: Ureukaryote , Uropisthokont , Urholozoan , Urchoanozoan , Urchoanoflagellate , and Urmetazoan . The full heat map for all gene families with representatives from at least two species is shown in Figure 2—figure supplement 4 . Gene families present , gained and loss at each ancestral node are listed in Dataset 6 ( Richter et al . , 2018 ) . We tested three classes of existing ancestral reconstruction methods using presence/absence data as input: Dollo parsimony , maximum likelihood , and Bayesian . The three different methods produced substantially different estimates of ancestral gene family content ( Supplementary file 13 ) . We performed Dollo parsimony analysis on presence/absence data for all gene families using PHYLIP version 3 . 695 ( Felsenstein , 2013 ) with default parameter values . For comparison to previous studies , we also gathered Dollo parsimony-based gene content estimates from Fairclough et al . , 2013 , which included two choanoflagellate species ( M . brevicollis and S . rosetta ) . We ran maximum likelihood analysis on presence/absence data for all gene families using Mesquite version 3 . 40 ( Maddison and Maddison , 2018 ) . When running Mesquite , we supplied the phylogenetic tree shown in Figures 2 and 3 as input ( see Materials and methods , ‘Species tree and phylogenetic diversity’ ) and analyzed gene family content evolution with the AsymmMk ( asymmetrical Markov k-state two parameter ) model . We performed Bayesian analysis with MrBayes version 3 . 2 . 6 ( Ronquist and Huelsenbeck , 2003 ) . We supplied presence/absence data only for gene families present in more than one species ( i . e . , we eliminated singletons ) , and specified the options ‘noabsencesites’ and ‘nosingletonpresence’ to correct for unobservable site patterns following ( Pisani et al . , 2015 ) , with Naegleria gruberi as the outgroup , gamma-distributed rate variation among sites , and other parameters left at their defaults . We ran separate analyses for each ancestral node of interest , each time including the species descended from that node as a topology constraint . We ran each analysis with a sampling frequency of 1000 generations until the average standard deviation of split frequencies between runs reached 0 . 01 ( between 5 million and 10 million generations , depending on the ancestral node of interest ) . For the two analyses that produced probabilities of gene family presence at internal nodes ( maximum likelihood and Bayesian ) , we calculated total counts as follows: for gene family presence , we summed the probabilities across all gene families . For gene family gains , we summed the difference in probability between each node of interest and its parent ( e . g . , the Urmetazoan and its parent , the Urchoanozoan ) only for the gene families with a higher probability of presence in the node of interest compared with its parent . For gene family losses , we performed an analogous sum only for the gene families with lower probabilities in the node of interest in comparison to its parent . Because new genome sequences are continuously becoming available , we tested whether the ancestral gene content reconstruction we performed would be influenced by the addition of new genome-scale data from two key sets of species: early-branching animals and early-branching holozoans . We chose species with high-quality publicly available protein catalogs that would maximize the phylogenetic diversity added to our data set . For early-branching animals , we selected two sponges , the demosponge Tethya wilhelma version v01_augustus_prots ( Francis et al . , 2017 ) and the calcareous sponge Sycon ciliatum version SCIL_T-PEP_130802 ( Fortunato et al . , 2014 ) , and one ctenophore , Pleurobrachia bachei version 02_P-bachei_Filtered_Gene_Models ( Moroz et al . , 2014 ) . For early-branching holozoans , we selected the teretosporeans C . fragrantissima , Ichthyophonus hoferi , Chromosphaera perkinsii and Corallochytrium limacisporum ( Grau-Bové et al . , 2017 ) . Because we planned to implement a best reciprocal BLAST approach , which might be confounded by paralogs present within any of the species , we began by removing redundancy separately for each species using CD-HIT version 4 . 6 ( Li and Godzik , 2006 ) with default parameter values . We then performed best reciprocal blastp from the protein catalog of each additional species versus all 59 species included in our original analysis ( Supplementary file 3 ) , with the same maximum E value ( 1 × 10−5 ) . We retained only top reciprocal blastp hits . We next calculated gene family presence probabilities for each additional species , using a slightly modified procedure . Because we retained only the top hit for each additional species protein to each original species , an additional species protein could only match to a single representative per species within each gene family , although the family might contain multiple representatives per species . Thus , instead of calculating average E value from each additional species protein to all members of each gene family ( as we did for the original set ) , we calculated the average of best hits to each species within the family . Next , because each protein from an additional species might hit multiple different gene families , we chose the single gene family match with the lowest average E value . Finally , for each gene family and each additional species , we selected the lowest average E value of any protein in the species to the gene family and translated those to probabilities using the same empirical cumulative density function as for the original analysis ( Figure 2—figure supplement 2b ) . To visualize whether the additional species might substantially impact our inferences of ancestral gene content , we inserted their presence probabilities into the original heat map of Figure 2 without reordering gene families ( Figure 2—figure supplement 5 ) . The additional sponges display similar patterns of gene family presence probability to the original sponge species in the analysis , as does the additional ctenophore in comparison to the original ctenophore . With these species added , the number of gene families present in the Urmetazoan in our original data set , ~8 , 418 , would increase by 37 , sixteen of which would be previously choanoflagellate-specific gene families newly classified as shared with animals . Within the early-branching holozoans , C . perkinsii appears to show slight evidence for the presence of animal-specific gene families , but generally with low probabilities , roughly comparable to the choanoflagellate S . dolichothecata . Thus , we estimate that a very small number of animal-specific gene families would instead be classified as originating in Holozoa with the addition of early-branching holozoans ( 13 of ~1 , 944 ) . In addition , since we only included one representative early-branching holozoan ( C . owczarzaki ) in our original analysis , a subset of gene families originally classified as originating in Choanozoa would instead be assigned to Holozoa with the additional species ( 30 ) ; in general , however , we did not strongly emphasize the distinction between choanozoan-origin and holozoan-origin gene families in our results . Furthermore , as described below ( Protein domains and evidence for gene presence based on domain architecture ) , the additional species would have a negligable effect on our inferences of gene family presence based on protein domain architecture . We attempted to produce a robustly supported phylogenetic tree reflecting the relationships among M . brevicollis , S . rosetta and the choanoflagellate species we sequenced . As input , we selected different subsets of gene families based on different criteria ( for example , one set of criteria selecting the 49 gene families missing at most five species , with remaining species having exactly one copy , or another set of criteria selecting the 24 gene families present in all species , with no species having more than two copies and at most 10 species having two copies ) . Within each subset , we separately aligned each gene family using MAFFT version 7 . 130b ( Katoh and Standley , 2013 ) with the parameters ‘--maxiterate 1000 --localpair’ and trimmed alignments using trimAl version 1 . 2rev59 ( Capella-Gutiérrez et al . , 2009 ) with the parameter ‘-gt 0 . 5’ . We then concatenated the trimmed alignments from each gene family and built trees with maximum likelihood ( RAxML version 8 . 2 . 0 ( Stamatakis , 2014 ) with the options ‘-m PROTCATGTR’ and ‘-f a -N 100’ ) and Bayesian ( PhyloBayes-MPI version 1 . 5a ( Lartillot et al . , 2013 ) with the option ‘-dc’ ) methods . We found that two species resting on long terminal branches , S . dolichothecata and C . hollandica , had low support values and inconsistent phylogenetic positions depending on the gene family subset and the analysis method used . We hypothesized that this might result from the fact that both lack sister species in our transcriptome data set . To test this , we instead built phylogenetic trees with the addition of nucleotide sequences from sister species for C . hollandica ( Codosiga botrytis and Sphaeroeca volvox ) and S . dolichothecata ( Salpingoeca tuba ) . For this analysis , we were limited by the small number of genes that had previously been sequenced for these species ( a maximum of 5 ) . We found that the phylogenetic positions of both C . hollandica and S . dolichothecata stabilized when their sister species were added [and that the choanoflagellate topology was consistent with that of ( Carr et al . , 2017 ) ] . We concluded that a robust phylogenomic tree of choanoflagellates , with stable , supported positions for C . hollandica and S . dolichothecata may require further genome-scale sequencing from sister groups of both species . Therefore , since the focus of this study was not phylogenetic tree construction , and because the topic has been addressed elsewhere , we selected species trees from prior publications . Furthermore , because our major findings were based on comparisons among groups whose phylogenetic relationships are well established ( animals , choanoflagellates , C . owczarzaki , fungi and other eukaryotes ) , differences in tree topology within any of these groups should be of minor importance . For choanoflagellates , we used the species tree from ( Carr et al . , 2017 ) , for animals ( Philippe et al . , 2009 ) and for other eukaryotes ( Burki et al . , 2016 ) . Because the branching order of early animals is still under active debate , we depicted the relationships among early-branching animals as a polytomy . We visualized the resulting tree with Archaeopteryx version 0 . 9813 ( Han and Zmasek , 2009 ) . To measure phylogenetic diversity , we selected a set of 49 gene families for which each species had exactly one protein representative and no more than five species were missing ( or roughly 10% of the 59 species in our data set ) . We aligned the protein sequences in each gene family separately using MAFFT version 7 . 130b ( Katoh and Standley , 2013 ) with the parameters ‘--maxiterate 1000 --localpair’ . We performed two rounds of alignment trimming with trimAl version 1 . 2rev59 ( Capella-Gutiérrez et al . , 2009 ) : in the first round , we used the parameter ‘-automated1’ , and we supplied the output of the first round to the second round , with the parameter ‘-gt 0 . 5’ . We constructed phylogenetic trees from the two-round trimmed alignments separately for each gene family with RAxML version 8 . 2 . 0 ( Stamatakis , 2014 ) , with the options ‘-m PROTGAMMALGF’ and ‘-f a -N 100’ . We measured cophenetic distances separating pairs of species on the resulting trees ( in units of substitutions per site along branches ) using the ape package version 4 . 1 ( Paradis et al . , 2004 ) and plotted pairwise distances averaged across all 49 gene families using the ggplot2 package version 2 . 2 . 1 ( Wickham , 2009 ) , both with R version 3 . 4 . 1 ( R Core Team , 2017 ) in the RStudio development platform version 1 . 0 . 143 ( RStudio Team , 2016 ) . To identify animal-specific or choanoflagellate-specific gene families that were also present in all species within either group , we required every species in the group to pass the 10% minimum probability criterion . These core gene families are subject to several potential technical artifacts . First , an incomplete genome or transcriptome assembly could result in a species appearing to lack a gene family . Second , gene families containing repeated or repetitive protein domains ( e . g . EGF or Ankyrin ) might cause inappropriate inferences of sequence homology in our BLAST-based approach . Third , a true orthologous gene family present in all animals could be erroneously partitioned by OrthoMCL into two or more gene families , neither of which would then be considered present in all animals . Fourth , a gene family which duplicated on the lineage leading to the last common ancestor of a group could produce two gene families , among which paralogs are incorrectly partitioned , resulting in one or more species appearing to lack one of the two families . Thus , the lists of core gene families should not be considered exhaustive , especially for serially duplicated or repeat-containing gene families . We annotated core animal-specific gene families ( Table 1 ) by selecting consensus features and functions in UniProt ( The UniProt Consortium , 2017 ) , FlyBase ( Gramates et al . , 2017 ) and WormBase ( Lee et al . , 2018 ) . Notably , BTB/POZ domain-containing protein 9 , whose function is relatively poorly characterized in comparison to other core animal-specific gene families , contains the BTB Pfam domain , which was identified as part of an expanded repertoire of ubiquitin-related domain architectures in animals ( Grau-Bové et al . , 2015 ) . To determine retention of ancestral gene families within extant species , we began with the sets inferred to have evolved on the stem lineages leading to the Ureukaryote , Uropisthokont , Urholozoan , Urchoanozoan , Urchoanoflagellate , and Urmetazoan ( Supplementary file 4 , Dataset 6; Richter et al . , 2018 ) . For each set of gene families partitioned by ancestral origin , we summed the membership probabilities for each species . Because we applied a 10% membership probability threshold to consider a gene family to be present within group of species ( see above , Inference of gene family origins and heat map ) , a species in one group might have a small residual sum of membership probabilities for gene families assigned to another group . As an example , some choanoflagellates may display a small amount of retention of animal-specific gene families , which represents the sum of non-zero membership probabilities that did not reach the 10% threshold . To test whether B . floridae , N . vectensis , and O . pearsei retained the same gene families , we applied the 10% presence probability threshold within each species . We found that there were 7 , 863 Urmetazoan gene families retained in at least one of the three species and 5 , 282 retained in all three ( 67% ) . Some animal species with draft genomes , for example P . pacificus , were among those that retained the fewest ancestral gene families . However , the lack of gene predictions resulting from an incomplete genome assembly is likely not as strong as the signal produced by gene loss . In the example case of P . pacificus , its sister species C . elegans , which has a finished genome , retains the second fewest gene families . We produced phylogenetic tree-based visualizations using the Interactive Tree of Life ( iTOL ) web site ( Letunic and Bork , 2016 ) . We produced bar charts using the ggplot2 package version 2 . 2 . 1 ( Wickham , 2009 ) with R version 3 . 4 . 1 ( R Core Team , 2017 ) in the RStudio development platform version 1 . 0 . 143 ( RStudio Team , 2016 ) . We annotated gene family function using the PANTHER Classification System ( Thomas et al . , 2003 ) . We used the PANTHER HMM library version 7 . 2 ( Mi et al . , 2010 ) and the PANTHER HMM Scoring tool version 1 . 03 ( Thomas et al . , 2006 ) with default parameter values and the recommended expectation value cutoff of 10−23 . We used PANTHER-provided files to associate PANTHER HMM hits with Gene Ontology ( GO ) terms ( Ashburner et al . , 2000 ) . Annotations for both sets of terms , for all input proteins ( including those not placed into gene families ) are available in Dataset 7 ( Richter et al . , 2018 ) . Annotations by gene family are listed in Dataset 8 ( Richter et al . , 2018 ) . We compared pathways present in the Urholozoan , Urchoanozoan , Urchoanoflagellate and Urmetazoan using MAPLE version 2 . 3 . 0 ( Takami et al . , 2016 ) . As input , we provided protein sequences from gene families present at each ancestral node , but only for species descending from the ancestral node ( for example , for the Urholozoan , we only included sequences from holozoans ) . We supplied the following parameters to MAPLE: NCBI BLAST; single-directional best hit ( bi-directional best hit would not have been appropriate , since our input database consisted of gene families containing closely-related proteins from multiple species ) ; KEGG version 20161101; and organism list ‘ea’ ( all eukaryotes in KEGG ) . We compared completeness based on the WC ( whole community ) module completion ratio . We listed modules which differed by 25% or greater in completeness between the Urchoanozoan and Urmetazoan in Supplementary file 8 , between the Urholozoan and Urchoanozoan in Supplementary file 10 , and between the Urchoanozoan and Urchoanoflagellate in Supplementary file 9 . To focus on amino acid biosynthesis pathways ( Figure 2—figure supplement 7 ) , we exported KEGG results from MAPLE for the Urchoanozoan and Urmetazoan gene sets and compared them using the KEGG Mapper Reconstruct Pathway tool , pathway ID 01230 ( Kanehisa et al . , 2012 ) . We predicted protein domains with the Pfam version 27 . 0 database and pfam_scan . pl revision 2010-06-08 ( Punta et al . , 2012 ) , which used hmmscan from the HMMER 3 . 0 package ( Eddy , 2011 ) . We performed all Pfam searches with default parameter values . We predicted signal peptides and transmembrane domains using Phobius version 1 . 01 ( Käll et al . , 2004 ) with default parameter values . Domains for all input proteins ( including those not placed into gene families ) are listed in Dataset 7 ( Richter et al . , 2018 ) . Domains by gene family are shown in Dataset 8 ( Richter et al . , 2018 ) . To determine which animal-specific gene families lacked a Pfam domain of known function , we calculated the proportion of proteins in each animal gene family that were annotated with a given Pfam domain ( Figure 2—figure supplement 11 ) . We accepted Pfam domains present in at least 10% of proteins in a gene family . Domains of unknown function in the Pfam database had names beginning with ‘DUF’ . To ensure that these domain names were not assigned a function in a more recent version of the Pfam database published after our initial annotation , we checked against Pfam version 31 . 0 and considered all domains whose names no longer began with ‘DUF’ as having an assigned function . We established a set of criteria defining ‘strong’ , ‘moderate’ and ‘weak’ evidence for the presence of genes of interest , based on domain content and OrthoMCL gene families; all criteria can be found in Supplementary file 6 . In all cases , ‘strong’ evidence for conservation consisted of the canonical domain architecture ( a diagnostic domain or set of domains in a specific order ) . Because Pfam domains are constructed from a set of aligned protein sequences selected from sequenced species , they are often strongly enriched for animals ( and , in particular , for animal models such as D . melanogaster and C . elegans ) and therefore may be biased against detecting instances of protein domains with more remote homology . To address this concern , for some gene families , we considered the presence of a protein in the same gene family as another protein with ‘strong’ evidence to constitute potential ‘moderate’ or ‘weak’ evidence . In addition , we considered previous reports to be ‘strong’ evidence in two cases: the presence of a canonical TLR in N . vectensis ( Miller et al . , 2007; Sullivan et al . , 2007 ) , and the presence of NF-κB in C . owczarzaki ( Sebé-Pedrós et al . , 2011 ) . Genome or transcriptome data for nine additional early-branching holozoans became available since our initial analyses of protein domain architecture ( Torruella et al . , 2015; Grau-Bové et al . , 2017 ) . To test whether the genes and protein domains of interest that we inferred to have originated in the Urchoanozoan did not in fact originate in the Urholozoan , we interrogated the domain content of the nine new species . In all cases , there were no proteins in any of the nine species with domain architectures that would qualify as ‘strong’ evidence to change our inference of ancestral origin ( i . e . , where there was not already ‘strong’ evidence in C . owczarzaki ) . To identify Notch and Delta , we relied solely on conserved protein domain architecture rather than on BLAST-based or phylogenetic evidence . In addition to their diagnostic protein domains , both families contain repeated , common protein domains . These can result in false inferences of homology for BLAST , and in difficulties in producing robust sequence alignments for phylogenetic analysis . Previous studies on Notch and Delta encountered difficulties in producing resolved phylogenetic trees , which were more pronounced for early-branching animals ( Gazave et al . , 2009 ) and potentially due to the presence of EGF repeats ( Rasmussen et al . , 2007 ) . To test these effects in our data , we built phylogenetic trees for both Notch and Delta . For Notch , we selected all proteins encoding a Notch domain , and for Delta , all proteins encoding a DSL ( Delta Serrate ligand ) domain . We aligned Notch and Delta separately using MAFFT version 7 . 130b ( Katoh and Standley , 2013 ) with the parameters ‘–maxiterate 1000 –localpair’ for high accuracy alignments , trimmed them using trimAl version 1 . 2rev59 ( Capella-Gutiérrez et al . , 2009 ) with the option ‘-gt 0 . 4’ to discard positions represented in fewer than 40% of proteins in the alignment , and built phylogenetic trees using RaxML version 8 . 2 . 0 ( Stamatakis , 2014 ) , with the options ‘-m PROTGAMMALGF’ to define the model of protein substitution and ‘-f a –N 100’ for a rapid analysis with 100 bootstraps . We visualized trees and removed branches with less than 50% bootstrap support using the Interactive Tree of Life ( iTOL ) web site ( Letunic and Bork , 2016 ) , yielding trees that were largely unresolved for both genes ( Figure 2—figure supplement 12 ) , consistent with previous results and reinforcing the use of protein domain architecture to identify Notch and Delta . For Notch , we defined ‘strong’ evidence for conserved domain architecture to be one or more EGF repeats , a Notch domain , a transmembrane domain and an Ank domain , in that order . This domain architecture was unique to animals and choanoflagellates in our data set . Many , but not all , Notch genes in animals also contain NOD and NODP protein domains . However , we did not use these as evidence , because numerous animal Notch genes do not encode these domains , including some bilaterians ( Gazave et al . , 2009 ) ( for example , the primate Macaca mulatta ( UniProt ID F7HCR6 ) , the tapeworm Hymenolepis microstoma ( A0A068XGW6 ) , and the nematode Ascaris suum ( U1NAR2 ) ) . In our data , the NOD and NODP domains were not present in any sponge or choanoflagellate species , although two of the three sponges ( E . muelleri and A . queenslandica ) encoded proteins with EGF , Notch , transmembrane and Ank domains , and a previous study found that A . queenslandica Notch and Delta were both expressed in in patterns consistent with other animals ( Richards et al . , 2008 ) . We hypothesize that the NOD and NODP domains were not detected in either choanoflagellates or sponges because the NOD and NODP Pfam models were constructed from sequence alignments that included proteins mostly from bilaterians . For Delta , we defined ‘strong’ evidence as the presence of both an MNNL ( N terminus of Notch ligand ) and a DSL domain . Both of these domains were considered to be animal-specific prior to this study [although we note that the DSL domain is found in two recently sequenced holozoans , the ichthyosporeans C . fragrantissima and Pirum gemmata ( Grau-Bové et al . , 2017 ) ] . The choanoflagellate S . dolichothecata expressed a protein encoding both the MNNL and DSL domains ( m . 249548 ) , but lacking the transmembrane domain typically found in Delta . However , the contig encoding this protein was only partially assembled , and the presence of other partially assembled contigs in S . dolichothecata encoding combinations of DSL , transmembrane and EGF domains increases our confidence that S . dolichothecata encodes a bona fide Delta . ( Predicted proteins encoding MNNL and DSL , but not a transmembrane domain , were also present in some animals in our data set; for example , Capitella teleta and Strongylocentrotus purpuratus . ) The canonical TGF-β signaling ligand ( e . g . , Activins or BMPs ) is characterized by the TGFb_propeptide ( LAP ) and TGF_beta domains ( Munger et al . , 1997 ) , the type I TGF-β receptor by the combination of a TGF_beta_GS and a kinase domain , and the transcriptional activator SMAD by the MH1 and MH2 domains ( Heldin et al . , 1997 ) . Although none of these combinations was present in choanoflagellates , we detected the individual TGFb_propeptide , TGF_beta_GS , MH1 and MH2 domains encoded in separate proteins ( Figure 2—figure supplement 9 ) . None of these four domains is found in eukaryotes outside Choanozoa ( kinases are ancient domains , and the TGF_beta domain is found only in animals ) . For our analysis of Toll-like receptors , we considered all proteins containing two of the three following Pfam domains: LRR , Pkinase and TIR , as these are found in immune receptors and signaling proteins in both animals and plants . We found that the domain architecture of LRR , transmembrane domain and TIR , in that order , was unique to choanozoans , consistent with the independent evolution in plants of immune-related proteins that contain the same protein domains but with distinct domain architectures , different pattern recognition , and different downstream signaling pathways ( Ausubel , 2005; Leulier and Lemaitre , 2008 ) . The domain architectures LRR , transmembrane , Pkinase , TIR ( ‘Kinase TLRs’ ) and Pkinase , TIR lacking a transmembrane domain ( ‘Kinase TIRs’ ) were both unique to choanoflagellates . We did not distinguish among members of the LRR and TIR families ( whose names in Pfam take the form of LRR_X , where X is a number , or TIR/TIR_2 ) , because we found that different animal TLRs could contain domain combinations of each type . For example , M . musculus TLR11 ( UniProt Q6R5P0 , NCBI GI 45429992 ) contains LRR_8 and TIR_2 domains , whereas M . musculus TLR6 ( UniProt Q9EPW9 , NCBI GI 157057101 ) contains LRR_4 and TIR domains . We also did not attempt to differentiate among single vs . multiple cysteine cluster LRR domains ( Imler and Zheng , 2004 ) , because we found that TLRs containing a multiple cysteine cluster LRRNT Pfam domain as the final LRR prior to the transmembrane region were restricted to D . melanogaster within our data set . We tested whether phylogenetic evidence could be used to confirm or contradict our findings based on protein domain architecture , although we anticipated that the repeated nature of LRR domains and the relatively short length of TIR domains might interfere with the resolution of phylogeny reconstructions . We aligned all proteins in our data set containing two of the three of LRR , Pkinase and TIR with MAFFT version 7 . 130b ( Katoh and Standley , 2013 ) with default parameters ( we were unable to build an alignment using MAFFT high accuracy parameters due to the large number of proteins involved ) . We trimmed the alignment using trimAl version 1 . 2rev59 ( Capella-Gutiérrez et al . , 2009 ) with the option ‘-gt 0 . 4’ and built phylogenetic trees using RAxML version 8 . 2 . 0 ( Stamatakis , 2014 ) , with the options ‘-m PROTGAMMALGF’ and ‘-f a -N 100’ . We visualized trees and removed branches with less than 50% bootstrap support using the Interactive Tree of Life ( iTOL ) web site ( Letunic and Bork , 2016 ) . The resulting tree was largely unresolved , as anticipated ( Figure 4—figure supplement 2 ) , although there were no cases in which a choanozoan-type TLR ( LRR , transmembrane , TIR ) was found in the same supported clade as a plant protein , consistent with the hypothesis that they evolved independently . Although we did not find strong evidence for canonical animal TLR domain architecture in sponges , M . leidyi and T . adhaerens , we did detect proteins in sponges and cnidarians matching the domain content of animal interleukin receptors , which can also signal via MyD88 ( O'Neill and O’Neill , 2008 ) : an extracellular immunoglobulin ( Ig ) domain , a transmembrane domain , and an intracellular TIR domain [as found previously for the sponge A . queenslandica ( Gauthier et al . , 2010 ) ] . This architecture was not present in choanoflagellates nor any other non-animal in our data set . We note that , in our analyses , the proteins previously identified as TLR2 homologs in A . queenslandica , O . carmela and other sponges ( Riesgo et al . , 2014 ) appear to have domain architectures matching interleukin receptors or other proteins , but not TLRs . For genes in the canonical M . musculus TLR/NF-κB signaling pathway , we collected information on domain architectures in the Pfam 31 . 0 database ( which did not include the choanoflagellates sequenced in this study ) and on the species present in the same gene family as the M . musculus protein ( Supplementary file 14 ) . With the exception of adapters and kinases containing the Death domain , there were no architectures that were both specific to animals and present across animal diversity , two features required of diagnostic architectures for a gene family . Similarly , we did not identify any gene families restricted to holozoans ( as would be expected for genes participating in the TLR/NF-κB pathway ) and whose choanoflagellate members we could unequivocally identify as orthologous to the M . musculus proteins in the family . Thus , we restricted our current analyses to TLRs , adapters containing TIR or Death domains , and NF-κB . Future analyses more focused on detecting individual signaling components [e . g . , ( Gilmore and Wolenski , 2012 ) ] are likely to further elucidate their evolutionary histories . We built an alignment for all proteins in the gene family containing SARM1 ( gene family 6840 ) , with the addition of the human sequence ( UniProt Q6SZW1 ) , since previous work has focused on the human SARM1 protein . We aligned proteins with MAFFT 7 . 130b ( Katoh and Standley , 2013 ) with the parameters ‘--maxiterate 1000 --localpair’ for high accuracy alignments . We visualized the alignment with JalView 2 ( Waterhouse et al . , 2009 ) . In addition to SARM1 and Kinase TIRs , choanoflagellates also encoded a diversity of protein domains associated with intracellular TIR domain-containing proteins ( lacking a transmembrane domain ) . Domains found associated with TIR in multiple choanoflagellate species included ankyrin repeat ( Ank ) and armadillo ( Arm ) protein-protein interaction domains and Src homology 2 ( SH2 ) kinase signal transduction domains . In addition to TLRs , we also searched the choanoflagellates in our data set for cytosolic immune sensing genes . Several classes of nucleotide-binding domain and leucine-rich repeat ( NLR ) proteins have previously been detected in the A . queenslandica genome , suggesting that NLRs were present in the Urmetazoan ( Yuen et al . , 2014 ) . The architecture of NLRs consists of a protein interaction domain ( either Death or CARD ) , a nucleotide binding domain ( NACHT ) and multiple LRRs . Although we detected proteins encoding both NACHT and LRR domains in multiple choanoflagellates , neither the Death or CARD domain was present in any choanoflagellate we studied . The CARD domain in animal NLRs either directly or indirectly mediates activation of caspases ( von Moltke et al . , 2013 ) , which are among the 39 core animal-specific gene families that we detected , suggesting that both NLRs and associated caspase downstream signaling activity may be animal innovations . Plants possess an analogous intracellular sensing pathway , NBS-LRR , which , like the TLR pathway , is thought to have evolved independently ( Urbach and Ausubel , 2017 ) . Diagnostic domains were absent in choanoflagellates for four additional cytosolic sensors: three containing CARD domains [mitochondrial antiviral signaling protein ( MAVS ) , melanoma differentiation-associated protein 5 ( MDA5 ) and retinoic acid inducible gene 1 ( RIG-I ) ] and one containing HIN and PYRIN domains [absent in melanoma 2 ( AIM2 ) ] . However , we did find evidence for two other sensors previously reported in M . brevicollis ( Wu et al . , 2014 ) : STING is present in both Salpingoeca macrocollata and M . brevicollis , based on the presence of the diagnostic Pfam domain TMEM173 , and cGAS , identified by the Pfam domain Mab-21 , is present in S . macrocollata , M . brevicollis and M . fluctuans . To define ‘strong’ evidence for the presence of Argonaute in our data set , we used the conserved Pfam domain architecture DUF1785 , PAZ and Piwi ( Supplementary file 6 ) . To detect ‘strong’ evidence for Dicer , we searched for the architecture Dicer_dimer , PAZ and Ribonuclease_3 ( we considered the Dicer_dimer domain alone to be ‘moderate’ evidence ) . We did not detect Piwi in choanoflagellates , although it is present in most other eukaryotic lineages . Piwi is thought to repress transposable elements ( Aravin et al . , 2001; Aravin et al . , 2007 ) , and M . brevicollis , the only choanoflagellate species that has been investigated for transposable elements , appears to have very few ( Carr et al . , 2008 ) . Argonaute and Dicer were each lost five times independently in choanoflagellates ( Figure 3—figure supplement 2 ) , although these could both be overestimates if the genes were not expressed in our culture conditions . A similar pattern of repeated parallel loss has also been previously observed in kinetoplastids , a group of single-celled eukaryotes ( Matveyev et al . , 2017 ) . Curiously , in contrast to the case for kinetoplastids , where Argonaute and Dicer were generally lost together , we detect five choanoflagellate species in which Argonaute is present and Dicer is absent or vice versa . These absences could reflect the presence of non-canonical RNAi genes or Dicer-independent RNAi pathways in choanoflagellates , as previously reported in Saccharomyces cerevisiae and in M . musculus , respectively ( Drinnenberg et al . , 2009; Cheloufi et al . , 2010 ) . Diverse cadherins have previously been identified in choanoflagellates ( Abedin and King , 2008; Nichols et al . , 2012 ) , but two classes of cadherins involved in animal planar cell polarity and development were thought to be animal-specific: Flamingo and Protocadherins . The first described Flamingo cadherin ( also known as starry night ) is involved in the Frizzled-mediated establishment of planar polarity in the D . melanogaster wing ( Chae et al . , 1999; Usui et al . , 1999 ) . In animals , Flamingo cadherins are distinguished by the combination of three diagnostic Pfam domains , the presence of which constituted ‘strong’ evidence for Flamingo in our data set ( Supplementary file 6 ) : Cadherin , GPS and 7tm_2 , a seven-pass transmembrane domain . The transcriptome of C . hollandica is predicted to encode a protein with all three diagnostic domains ( Figure 2—figure supplement 6 ) . In addition to these three domains , many animal Flamingo cadherins also contain HRM , EGF , and Laminin domains . Nine further choanoflagellate species express proteins with GPS , 7tm_2 , and one of these additional domains ( ‘moderate’ evidence ) , and all choanoflagellate species express proteins with GPS and 7tm_2 domains ( ‘weak’ evidence ) . Protocadherins are a large and diverse family of genes involved in animal development and cell adhesion ( Frank and Kemler , 2002 ) , and most , but not all , family members possess a diagnostic transmembrane Protocadherin domain paired with one or more extracellular Cadherin domains . Five species of choanoflagellates expressed proteins with a Protocadherin domain ( ‘strong’ evidence for conservation , Figure 2—figure supplement 6 ) , although none of these also contained a Cadherin domain . The Choanoeca perplexa Protocadherin domain-containing protein also contains a transmembrane domain and an intracellular SH2 phosphorylated tyrosine binding domain , a combination not found in any other organism . Tyrosine kinase signaling networks greatly expanded in the Urchoanozoan ( Manning et al . , 2008; Pincus et al . , 2008 ) , and the Protocadherin domain in C . perplexa may have been co-opted for tyrosine kinase signaling function . Integrins are thought to have been present in the Urchoanozoan , since homologs of both components of the animal integrin plasma membrane α/β heterodimer ( Hynes , 2002 ) have previously been identified in C . owczarzaki ( Sebé-Pedrós et al . , 2010 ) . We found ‘strong’ evidence for integrin β ( in the form of the diagnostic Integrin_beta Pfam domain ) in only a single species of choanoflagellate , D . costata ( Figure 2—figure supplement 6 ) . In contrast , we detected the integrin β binding protein ICAP-1 ( Chang et al . , 1997 ) , which has been proposed to act as a competitive inhibitor ( Bouvard et al . , 2003 ) , in all choanoflagellates except M . brevicollis and S . rosetta . | All animals , from sea sponges and reef-building corals to elephants and humans , share a single common ancestor that lived over half a billion years ago . This single-celled predecessor evolved the ability to develop into a creature made up of many cells with specialized jobs . Reconstructing the steps in this evolutionary process has been difficult because the earliest animals were soft-bodied and microscopic and did not leave behind fossils that scientists can study . Though their bodies have since disintegrated , many of the instructions for building the first animals live on in genes that were passed on to life forms that still exist . Scientists are trying to retrace those genes back to the first animal by comparing the genomes of living animals with their closest relatives , the choanoflagellates . Choanoflagellates are single-celled , colony-forming organisms that live in waters around the world . Comparisons with choanoflagellates may help scientists identify which genes were necessary to help animals evolve and diversify into so many different species . So far , 1 , 000 animal and two choanoflagellate genomes have been sequenced . But the gene repertoires of most species of choanoflagellates have yet to be analyzed . Now , Richter et al . have cataloged the genes of 19 more species of choanoflagellates . This added information allowed them to recreate the likely gene set of the first animal and to identify genetic changes that occurred during animal evolution . The analyses showed that modern animals lost about a quarter of the genes present in their last common ancestor with choanoflagellates and gained an equal number of new genes . Richter et al . identified several dozen core animal genes that were gained and subsequently preserved throughout animal evolution . Many of these are necessary so that an embryo can develop properly , but the precise roles of some core genes remain a mystery . Most other genes that emerged in the first animals have been lost in at least one living animal . The study of Richter et al . also showed that some very important genes in animals , including genes essential for early development and genes that help the immune system detect pathogens , predate animals . These key genes trace back to animals’ last common ancestor with choanoflagellates and may have evolved new roles in animals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2018 | Gene family innovation, conservation and loss on the animal stem lineage |
The microtubule-associated protein , tau , is the major subunit of neurofibrillary tangles associated with neurodegenerative conditions , such as Alzheimer's disease . In the cell , however , tau aggregation can be prevented by a class of proteins known as molecular chaperones . While numerous chaperones are known to interact with tau , though , little is known regarding the mechanisms by which these prevent tau aggregation . Here , we describe the effects of ATP-independent Hsp40 chaperones , DNAJA2 and DNAJB1 , on tau amyloid-fiber formation and compare these to the small heat shock protein HSPB1 . We find that the chaperones play complementary roles , with each preventing tau aggregation differently and interacting with distinct sets of tau species . Whereas HSPB1 only binds tau monomers , DNAJB1 and DNAJA2 recognize aggregation-prone conformers and even mature fibers . In addition , we find that both Hsp40s bind tau seeds and fibers via their C-terminal domain II ( CTDII ) , with DNAJA2 being further capable of recognizing tau monomers by a second , distinct site in CTDI . These results lay out the mechanisms by which the diverse members of the Hsp40 family counteract the formation and propagation of toxic tau aggregates and highlight the fact that chaperones from different families/classes play distinct , yet complementary roles in preventing pathological protein aggregation .
Tau is an intrinsically disordered protein ( IDP ) that is highly expressed in neurons and plays essential roles in microtubule self-assembly and stability ( Mandelkow and Mandelkow , 2012 ) , axonal transport ( Gustke et al . , 1994 ) , and neurite outgrowth ( Biernat and Mandelkow , 1999 ) . Tau binds to microtubules via its central microtubule-binding repeat ( MTBR ) domain ( Figure 1a ) , an interaction that is modulated by post-translational modifications ( PTMs ) . Aberrant PTMs such as hyperphosphorylation and acetylation ( Hanger et al . , 2009; Morris et al . , 2015; Cook et al . , 2014 ) were suggested to decrease the affinity of tau to microtubules , thus subsequently reducing microtubule stability . In addition , when tau dissociates from microtubules , it can form oligomers with the potential to disrupt cellular membranes , thereby impairing synaptic and mitochondrial functions , before ultimately forming amyloid fibers ( Shafiei et al . , 2017 ) . These abnormal forms of tau are thought to play a key role in the pathogenesis of various human tauopathies , including Alzheimer’s disease ( AD ) , frontotemporal dementias , and progressive supranuclear palsy ( Ballatore et al . , 2007 ) . In such cases , tau forms large intracellular aggregates , termed neurofibrillary tangles , whose abundance and localization in the brain correlates with cognitive decline ( Ballatore et al . , 2007; Brunello et al . , 2020 ) . It is still unclear , however , if the fibrils themselves are the neurotoxic species or whether prefibrillar soluble aggregates and oligomers of tau promote neuronal death by spreading tau pathogenicity from cell to cell in a prion-like manner . Chaperone machineries , and in particular members of the Hsp70 and the ATP-independent , small heat shock protein ( sHSP ) families , engage with tau during these pathogenic events , counteracting its aggregation into amyloids and targeting the misfolded species for degradation ( Dou et al . , 2003; Voss et al . , 2012; Petrucelli et al . , 2004; Mok et al . , 2018; Caballero et al . , 2021 ) . Moreover , in various cellular models , increased chaperone levels were shown to play an important role in tau cellular homeostasis by promoting tau solubility and microtubule binding while reducing the levels of pathological tau species ( Dou et al . , 2003; Perez et al . , 1991; Renkawek et al . , 1994; Dabir et al . , 2004; Sahara et al . , 2007; Jinwal et al . , 2013; Shimura et al . , 2004 ) . The chaperones do so by specifically recognizing the six-residue aggregation-prone regions located at the start of the second and third repeats in the MTBR domain ( Mok et al . , 2018; Freilich et al . , 2018; Baughman et al . , 2018; Weickert et al . , 2020; Figure 1A ) . These two motifs , 275VQIINK280 and 306VQIVYK311 ( also called PHF6* and PHF6 , respectively ) , are susceptible to the formation of β-sheet structures that have been found to be a prerequisite for tau aggregation ( von Bergen et al . , 2000; Mukrasch et al . , 2005 ) . Despite both Hsp70 and HSPB1 chaperones recognizing the same regions , their tau aggregation-prevention mechanisms have been found to be markedly different . Hsp70 interaction with tau suppresses the formation of aggregation-prone tau nuclei and sequesters tau oligomers and fibrils ( Baughman et al . , 2018; Patterson et al . , 2011 ) , thereby neutralizing their ability to damage membranes and seed further tau aggregation ( Kundel et al . , 2018a ) . In contrast , HSPB1 was shown to delay tau fiber formation by weakly interacting with early species in the aggregation reaction ( Baughman et al . , 2018 ) . Thus , it would appear that different cellular chaperones can bind to distinct tau species and affect tau homeostasis in different ways . Excitingly , though , HSPB1 is not the only ATP-independent chaperone reported to interfere with the tau aggregation pathway – members of the Hsp40 family ( also known as J-domain proteins [JDPs] ) have also recently been reported to affect tau aggregation in the cell ( Mok et al . , 2018; Brehme et al . , 2014; Fontaine et al . , 2015; Hou et al . , 2020 ) . JDPs are a diverse group of proteins that function as co-chaperones of the Hsp70 machinery , and are responsible both for selecting and delivering clients to the chaperones , and stimulating Hsp70 ATPase activity , thereby activating the chaperone cycle . These multidomain proteins are all structurally characterized by the conserved signature J-domain , essential for stimulation of Hsp70 ATPase activity ( Kityk et al . , 2018 ) . In addition , canonical class A and B JDPs also comprise a regulatory glycine‐rich ( GF ) region adjacent to the N-terminal J-domain ( Faust et al . , 2020; Karamanos et al . , 2019 ) , two structurally similar C-terminal β‐barrel domains ( CTDI and CTDII ) containing the substrate binding region , and a dimerization domain ( Kampinga and Craig , 2010; Rosenzweig et al . , 2019 ) . Class A JDPs further contain a zinc-finger-like region ( ZFLR ) protruding from CTDI . Recently , several studies have indicated that JDPs can also function as bona fide chaperones , utilizing holdase activity to prevent the aggregation of their client proteins ( Ayala Mariscal and Kirstein , 2021 ) . DNAJA2 , a member of this JDP family , was recently identified as a potent suppressor of tau aggregation , capable of effectively preventing the seeding of tau and formation of amyloids in cells ( Mok et al . , 2018; Abisambra et al . , 2012 ) , with DNAJA2 levels being selectively increased in AD patient neuronal cells ( Mok et al . , 2018 ) . Additionally , it was recently shown that , along with the Hsp70 system , DNAJB1 , a class B JDP , can break apart tau amyloid fibers extracted from AD brain tissues ( Nachman et al . , 2020 ) . Little is known , however , regarding how these co-chaperones interact with tau or the mechanism by which they modify tau disease-related amyloid states . We therefore used NMR spectroscopy , in combination with kinetic aggregation assays , to elucidate the effect of the DNAJA2 and DNAJB1 chaperones on tau aggregation . We found that the aggregation-prevention mechanisms of DNAJB1 and DNAJA2 are strikingly different from that of HSPB1 holdase chaperone . Moreover , we found that the two Hsp40 family members also diverge in their interactions with tau , whereas DNAJA2 interacts with all species along the tau aggregation pathway , including inert tau monomers , DNAJB1 only interacts with aggregation-prone tau conformers , such as seeding competent species or mature fibers .
We first investigated the effect of DNAJA2 and DNAJB1 on the fibril formation of full-length tau ( 2N4R ) and compared it to that of HSPB1 , a well-characterized suppressor of tau aggregation . Tau aggregation was monitored using Thioflavin-T ( ThT ) fluorescence ( Biancalana and Koide , 2010 ) . The 3D GXG variant of HSPB1 ( 22 ) was used to mimic the fully activated dimeric form of the chaperone . As expected , the addition of HSPB1 significantly inhibited tau fiber formation , in agreement with previous reports ( Mok et al . , 2018; Freilich et al . , 2018; Baughman et al . , 2018 ) . Interestingly , addition of DNAJA2 and DNAJB1 chaperones also completely inhibited the formation of tau fibrils for over 16 hr ( Figure 1B ) , with no observable ThT signal being detected over this length of time . Similar results were obtained with the shorter tau construct tau4R ( residues 244–372 ) ( Figure 1C ) , which forms the core of tau filaments and nucleates tau aggregation ( Barghorn et al . , 2004 ) . Despite the faster aggregation kinetics of tau4R relative to the full-length tau ( t½ ≈ 110 min vs . t½ ≈ 340 min ) , the presence of even a sub-stoichiometric concentration of either of the three molecular chaperones ( DNAJB1 , HSPB1 , or DNAJA2 ) fully suppressed aggregation for the duration of the experiment ( Figure 1C ) . As all three chaperones are able to suppress tau aggregation , we next aimed to understand the mechanisms by which they do so . In order to unravel the mechanism of aggregation prevention , we first identified the binding sites for the three chaperones on tau . To this end , we recorded 1H-15N HSQC spectra of either 15N-tau or 15N-tau4R in the absence and presence of each chaperone . Upon addition of HSPB1 , we observed significant peak broadening of tau residues 275–280 and 306–311 , which correspond to the PHF6* and PHF6 motifs ( Figure 1D and Figure 1—figure supplement 1 ) , in agreement with previous reports ( Mok et al . , 2018; Freilich et al . , 2018; Baughman et al . , 2018 ) . Notably , the PHF6 motifs , which are the most hydrophobic regions within tau , are also the preferential binding sites for the major ATP-dependent chaperone families , such as Hsp70 and Hsp90 ( Mok et al . , 2018 ) . A similar preference for the PHF6 and PHF6* motifs was also found for DNAJA2 , with residues 275–284 and 306–320 showing significant peak broadening ( Figure 1D and Figure 1—figure supplement 1 ) . Surprisingly , however , DNAJB1 showed no significant binding to tau , despite efficiently suppressing amyloid formation ( Figure 1D and Figure 1—figure supplement 1 ) . As DNAJB1 efficiently prevented tau aggregation in both full-length and tau4R experiments , yet showed no detectable binding to the monomers , we hypothesized that it functions , instead , through association with preformed tau fibers . In fact , a similar behavior was recently reported for DNAJB1 in the case of α-synuclein , where the chaperone displayed a remarkable preference ( >300-fold ) toward the amyloid state of α-synuclein over the monomer ( Wentink et al . , 2020 ) . We therefore checked whether DNAJB1 interacts with preformed tau4R fibrils using co-sedimentation experiments . Indeed , a large portion of DNAJB1 was detected in the insoluble fraction together with tau4R ( Figure 2A ) , indicating a strong interaction between tau fibrils and the chaperone . Co-sedimentation experiments with DNAJA2 and HSPB1 showed that also DNAJA2 co-precipitated with tau fibers , whereas HSPB1 was mainly found in the soluble fraction and only marginally interacted with tau amyloids ( Figure 2A ) . Similar results were also obtained with fluorescence anisotropy , resulting in an affinity ( dissociation constant ) for tau4R fibrils of 1 . 7 ± 0 . 2 µM ( mean ± s . e . m . ) for DNAJB1 and of 7 . 6 ± 0 . 6 µM for DNAJA2 ( Figure 2B ) . In contrast , the affinity of HSPB1 for tau fibers was outside our experimental concentration range . The selective interaction of DNAJB1 and DNAJA2 chaperones with tau fibers was also observed by negative stain electron microscopy ( EM ) . Here , tau alone formed characteristic long twisted filaments with a periodicity of 50–100 nm ( Figure 2D ) , consistent with previous observations ( Fitzpatrick et al . , 2017 ) . Upon addition of HSPB1 chaperone to the fibers , no changes to fiber length or morphology were detected , in agreement with our co-sedimentation assays that showed no binding of HSPB1 chaperone to the preformed fibers . HSPB1 itself , however , generated large protein assemblies that can be seen next to the fibers in the EM images ( marked by * ) . Addition of DNAJB1 or DNAJA2 , on the other hand , caused visible change in the appearance of the fibers , generating straighter , less twisted filaments decorated by periodically bound chaperones ( Figure 2D ) . The overall length of the fibers , however , did not change substantially , and also smaller tau fragments were not observed in our EM images , indicating that DNAJB1 and DNAJA2 do not enhance fiber breakage or fragmentation upon binding . In summary , HSPB1 only binds tau4R monomers , DNAJB1 interacts with tau4R fibrils , whereas DNAJA2 binds both monomers and fibers ( Figure 2B , C ) . We were next interested to see how the different tau binding modes of the three chaperones affect their respective aggregation-prevention mechanisms . We therefore performed a series of aggregation kinetics experiments , varying the concentrations of the chaperones while maintaining a constant concentration of tau4R . In the absence of chaperones , the aggregation of tau4R has previously been reported to occur through the following microscopic steps: primary nucleation , fibril growth through the addition and rearrangement of monomers ( saturating elongation ) , and fiber fragmentation ( Kundel et al . , 2018b; Yao et al . , 2020; Figure 3—figure supplement 1 ) . To confirm that this is indeed the aggregation mechanism for tau4R under our experimental conditions , we recorded aggregation kinetics data at various concentrations of monomeric tau4R ( 2 . 5–40 µM; Figure 3—figure supplement 1 ) . The half-saturation times ( half-times ) were then extracted and analyzed as a function of the concentration of tau4R monomer in accordance with a protocol by Meisl et al . , 2016 , allowing us to calculate the scaling exponent . This was found to be −0 . 34 ± 0 . 04 , which was consistent with a dominant primary nucleation pathway and a contribution stemming from the presence of fibril fragmentation ( Kundel et al . , 2018b; Yao et al . , 2020; Shammas et al . , 2015; see Materials and methods for more detail ) . In addition , a positive curvature of these double-logarithmic plots indicated the presence of saturation effects in the dominant mechanism ( Meisl et al . , 2016; Figure 3—figure supplement 1A ) . Tau4R aggregation kinetics data in the absence ( Figure 3—figure supplement 1B ) and presence of aggregation seeds ( Figure 3—figure supplement 1C ) were next globally fit , assuming a nucleus size of two tau4R monomers , to extract the kinetic rates for nucleation , elongation , and fragmentation , as well as the saturation constant ( Figure 3—figure supplement 1D ) . All chaperones caused a concentration-dependent retardation of tau aggregation at sub-stoichiometric concentrations ( Figure 3 ) . To understand the effect of the chaperones at the microscopic level , we fit the data to the kinetic model of tau aggregation using the kinetic parameters derived in the absence of chaperones . In order to determine which step in the aggregation mechanism is most likely affected by each of the chaperones , we only allowed an individual kinetic rate to vary in each of the analyses ( see Materials and methods for detail ) . As the addition of any of the chaperones to preformed tau fibrils did not change the overall fibril length ( Figure 2D ) , we could further assume that the chaperones have no effect on tau fragmentation rates ( km ) . Upon fitting HSPB1 aggregation-prevention data , only a poor agreement was achieved when allowing the perturbation of primary nucleation rates ( kn ) ( Figure 3A ) . Given the fact that HSPB1 binds tau monomers , the relatively moderate effect on tau primary nucleation was somewhat surprising as monomer binding should inhibit both nucleation and elongation . In contrast , fitting the fiber elongation rate ( kp ) provided a significantly better description of the kinetic data ( Figure 3A ) , indicating an order of magnitude reduction of this rate ( Figure 3D ) . Moreover , even when allowing the variation of both the nucleation and elongation rates ( kn and kp ) in the fit , only the elongation rates were reduced in the presence of HSPB1 chaperone , while the nucleation rates remained unchanged ( Figure 3—figure supplement 2A , Figure 3D ) . The effect of DNAJA2 on tau aggregation could neither be described by the reduction of nucleation rates nor elongation rates alone ( Figure 3B ) , which is not entirely surprising , given that the chaperone can bind to both monomeric tau and fibrils ( Figure 1D , Figure 2B , C ) . In general , DNAJA2 has at least two distinct pathways to impact aggregation , namely by ( i ) reducing the amount of monomeric tau accessible for nucleation and elongation and ( ii ) lowering the potency of fibrils to grow . In order to determine which rates are affected by the chaperone , we recorded an additional set of kinetic measurements in the presence of DNAJA2 chaperone , this time with the aggregation being initiated by tau fiber seeds . In the presence of seeds , tau primary nucleation events are negligible , thus allowing us to estimate the effect of the DNAJA2 chaperone only on rate constants of elongation and fragmentation ( Figure 3—figure supplement 2C ) . Using a global fit of the unseeded and seeded aggregation kinetics , we were then able to determine the effect of DNAJA2 on the nucleation and elongation rates ( Figure 3—figure supplement 2C and Figure 3D ) . Combined , these results show a substantial perturbation of both tau nucleation and elongation rates by the DNAJA2 chaperone ( Figure 3D ) , indicating that this chaperone affects the aggregation process in more intricate ways than HSPB1 . Surprisingly , the effect of DNAJB1 on tau aggregation was best described by its ability to reduce the rates of primary nucleation ( Figure 3C , left ) and , only to a lesser extent , fiber elongation ( Figure 3C , right ) . This ability of the DNAJB1 chaperone to effectively inhibit the rate of tau4R primary nucleation was unexpected as primary nucleation involves interaction between tau monomers , yet no interaction between DNAJB1 and this species of tau was observed in our NMR experiments . The effect of DNAJB1 on tau4R nucleation , despite its inability to interact with monomeric tau4R , can , however , be reconciled by the recent finding that soluble monomeric tau4R exists in two conformational ensembles – an ensemble that does not spontaneously aggregate ( ‘inert’ tau monomer ) and a seed-competent monomer that triggers the spontaneous aggregation of tau ( Mirbaha et al . , 2018; Chen et al . , 2019 ) . It is therefore possible that DNAJB1 only identifies and interacts with the ‘aggregation-prone’ tau species and not the inert monomers . Such seed-competent species have been proposed to be readily populated in tauopathy-associated tau mutants ( Chen et al . , 2019 ) , and can be generated in vitro by addition of polyanions such as heparin ( Eschmann et al . , 2017 ) . Unfortunately , the addition of heparin causes the rapid formation of tau fibers , thus preventing a detailed structural investigation of the seed-competent ensemble using NMR spectroscopy . Yet , whereas tau4R is inaccessible under conditions at which it rapidly aggregates , the rate of tau fibrillization can be tuned by altering the concentration of heparin present in the aggregation reaction . Low concentrations of heparin enhance the rate of fiber formation , whereas higher heparin concentrations potentially inhibit it ( Ramachandran and Udgaonkar , 2011 ) . We therefore performed aggregation kinetics at different heparin concentrations ( 0 . 1–40 µM ) to identify the conditions at which tau aggregation is sufficiently slow to permit NMR experiments . We found that heparin increased the rate of fiber formation up to a sub-stoichiometric concentration of 1 µM ( 1:0 . 1 tau:heparin ) ( Figure 4—figure supplement 1A ) . Above this threshold , further addition of heparin in fact slowed tau aggregation in a dose-dependent manner . At a one- to fourfold excess of heparin ( 10–40 µM ) , we found that tau amyloid formation was arrested for over 2 hr ( Figure 4—figure supplement 1A ) . An equimolar concentration of heparin to tau was therefore used to generate a soluble , aggregation-prone tau species ( Mirbaha et al . , 2018 ) that does not aggregate and is therefore amenable for NMR experiments . We then monitored chaperone binding to this tau species by recording 1H-15N HSQC spectra for 15N-tau-heparin complex alone , and upon addition of DNAJB1 , DNAJA2 , and HSPB1 chaperones to the mixture ( Figure 4 ) . To map the binding sites for the chaperones on this heparin-bound , aggregation-prone tau species , we first had to assign its spectrum . HNCA , CBCA ( CO ) NH , HN ( CA ) CO , and HNCO 3D NMR experiments were recorded , and assignments were obtained for 88% of the non-proline residues . Heparin binding resulted in chemical shift perturbations ( CSPs ) to the NMR spectrum of tau4R , mainly in the R1 , R2 , and PHF6* regions ( Figure 4A and Figure 4—figure supplement 1B ) . No changes were observed to the overall dispersion of the spectrum , indicating that heparin binding does not induce global folding of tau , in agreement with previous findings ( Mukrasch et al . , 2005 ) . With these assignments in hand , we were able to identify that , despite having negligible affinity toward the tau monomer , the DNAJB1 chaperone indeed interacts strongly with the aggregation-prone tau-heparin mixture ( Figure 4B ) , as we previously predicted . We further mapped this binding to tau residues 275–280 and 305–314 of the PHF6 and PHF6* repeats – the same regions to which both DNAJA2 and HSPB1 bind in the inert , monomeric tau . We next tested whether DNAJA2 and HSPB1 interact with the heparin-bound tau . DNAJA2 showed strong binding to the R2 and R3 PHF6* repeats of this aggregation-prone form of tau4R , similarly to its interaction with the free monomer ( Figure 4B ) . In contrast , HSPB1 only bound this species weakly , despite previously displaying a strong interaction with free tau4R . The lack of interaction between HSPB1 and aggregation-prone tau species explains our previous observation that HSPB1 does not affect the rate of fiber nucleation ( Figure 3A ) , as such an inhibition would require a direct interaction with the aggregate nucleus or aggregation-prone tau . Thus , while interacting with the same regions of tau , the three chaperones each display specific preferences for the different tau monomer conformers: HSPB1 interacts solely with tau monomers; DNAJB1 exclusively with the aggregation-prone species; and DNAJA2 with both . It was unclear , however , how the chaperones discriminate between the inert form of tau and the aggregation-prone tau-heparin complex . Secondary-structure propensity analysis ( Marsh et al . , 2006 ) of free and heparin-bound tau demonstrated that the different chaperone binding profiles cannot be explained by heparin-induced formation of extended β-strand structures in the PHF6 repeats ( Eschmann et al . , 2017 ) as these regions displayed no increase in secondary-structure propensity upon addition of heparin ( Figure 4—figure supplement 1C ) . Likewise , no notable changes were observed between the Cα and Cβ secondary chemical shifts patterns of heparin-bound and free tau samples ( Figure 4—figure supplement 1D ) . We then set out to determine whether the differences between monomeric tau4R and the heparin-tau complex are caused by electrostatic interactions . Heparin is a polyanion with a net charge of −3 per disaccharide at pH 7 . 0 ( or estimated charge density of −6 . 7 e-/kD; Lin et al . , 2020 ) . In contrast , tau4R has a positive net charge of +9 . 5 , suggesting that the heparin-tau4R complex is significantly stabilized by electrostatic attractions . As such , the ability of the chaperones to distinguish between the monomeric and aggregation-prone tau species may be related to simple differences in the net charge of the complex , with heparin binding , for example , reducing electrostatic repulsions that may exist between DNAJB1 and monomeric tau4R . Screening any such electrostatic interactions with higher ionic strengths ( 0 . 3 M ) , however , neither facilitated binding between DNAJB1 and tau4R ( Figure 4—figure supplement 2 ) nor altered tau4R interaction with HSPB1 or DNAJA2 chaperones ( compare Figure 1D and Figure 4—figure supplement 2 ) . Similarly , the interaction cannot be explained by attractive electrostatic interactions caused by the negative net-charge of the complex as we did not detect specific binding between heparin and any of the chaperones . Hence , electrostatic interactions cannot explain the different binding profiles of the chaperones to the two tau species . A clue to how heparin alters the conformational ensemble of tau4R came from a recent structural characterization of patient-derived , seeding-competent tau monomer . This species of tau was reported to have an expanded ensemble with a more exposed PHF6 motif compared to the inert monomeric protein ( Mirbaha et al . , 2018; Chen et al . , 2019 ) . This increased expansion of the PHF6 aggregation motifs was , in turn , suggested to drive the self-assembly and subsequent aggregation of tau ( Kaufman et al . , 2017 ) . Interestingly , binding of heparin to tau was also shown to expand the local conformation of the repeat regions ( R2 and R3 ) , thereby making the amyloidogenic PHF6 sequences more accessible ( Eschmann et al . , 2017 ) . Indeed , in agreement with previous reports ( Mukrasch et al . , 2007; Mukrasch et al . , 2009 ) , measurements of one-bond N-H RDCs in tau , which had been partially oriented in either a Pf1 bacteriophage or polyethylene glycol/hexanol alignment medium , showed relatively large H-N RDC values ( 10–20 Hz ) in the PHF6 repeat regions , which can arise from a locally compacted conformation ( Figure 4—figure supplement 3A , B ) . Similar analysis of the tau-heparin complex showed only very small RDCs in the PHF6 regions , despite the larger degree of alignment of this complex , indicating a potential expansion of these regions ( Figure 4—figure supplement 3C ) . This expansion , and thus the increased accessibility of the PHF6 motif in the heparin-bound state , may therefore be what enables the interaction of DNAJB1 with tau . To further test whether the expansion of the PHF6 repeats in the seeding-competent tau species is indeed the discriminating factor for chaperone binding , we monitored the interaction of the three chaperones with P301L/S missense mutations , which are known to cause dominantly inherited tauopathy ( Rizzu et al . , 1999 ) . These tau variants had been shown to contain a higher population of the expanded PHF6 conformation ( Chen et al . , 2019 ) , as also seen in our RDC measurements ( Figure 4—figure supplement 3D , E ) , and could therefore potentially mimic the aggregation-prone tau without requiring the addition of heparin . Similarly to what we observed for wild-type tau4R , we found that indeed all three chaperones efficiently suppressed the aggregation of these tau mutants ( Figure 4—figure supplement 4B ) , although with reduced efficacy in the case of HSPB1 and higher activity of DNAJB1 ( Figure 4—figure supplement 4C ) . The interaction of the P301L and P301S familial mutations of tau4R with HSPB1 , DNAJB1 , or DNAJA2 chaperones was then assayed using NMR spectroscopy . In the presence of HSPB1 and DNAJA2 , the resulting binding profiles of the two tau variants were very similar to those of wild-type tau , with residues 274–280 and 305–318 , corresponding to PHF6* and PHF6 repeats , displaying severe peak broadening ( Figure 4C and Figure 4—figure supplement 4A ) . This similarity to wild-type tau was not unexpected , though , as only a small portion of the P301L/S tau conformational ensemble adopts the extended conformation at any given moment ( Chen et al . , 2019; Kawasaki and Tate , 2020 ) . In contrast , DNAJB1 , while showing no binding to wild-type tau , did cause noticeable reductions in peak intensities in both PHF6 motifs of the P301S and P301L tauopathy mutants ( Figure 4C and Figure 4—figure supplement 4A ) . Moreover , the degree of observed intensity reduction ( 31 and 26% ) was found to be consistent with the relatively low population of the expanded conformation in tau P301L/S mutants ( Kawasaki and Tate , 2020; Figure 4—figure supplement 3D and E ) . Thus , DNAJB1 binding to monomeric tau indeed appears to depend on the exposure of the PHF6 region , thereby allowing it to distinguish between inert tau and the aggregation-prone species that eventually leads to amyloid formation . Our results show that all three chaperones efficiently inhibit tau aggregation via interactions with the hydrophobic aggregation-prone PHF6 repeats . Each of the chaperones interacts with a specific set of tau species , thus slowing down different microscopic processes in the aggregation reaction ( Figure 5D ) . However , it remained unclear whether the chaperones also affect the size and/or morphology of tau fibers . Since ThT fluorescence only reports on total fibril mass , with no differentiation to fibril length or number , we turned to EM in order to image the fibers . When adding DNAJA2 at the start of the aggregation reaction , a clear dose-dependent decrease in the density and length of the tau fibrils was observed ( Figure 5A ) . Similarly to DNAJA2 , DNAJB1-containing reactions yielded shorter fibrils that further shortened when repeating the experiment with increasing chaperone concentrations ( Figure 5B ) . This result is not unexpected given that these two chaperones also bind to tau4R fibers ( Figure 2 ) , which may contribute to the arrest of tau amyloid elongation . Interestingly , DNAJA2 was very efficient in reducing both the size and amount of formed tau fibers , and even at the sub-stoichiometric concentration of 0 . 1:1 DNAJA2:tau , a significant reduction in fiber length was observed . At a ratio of 0 . 25:1 DNAJA2:tau , the majority of fibers were between 0 . 5 and 2 . 0 µm long , and at 0 . 5:1 DNAJA2:tau the fibers were too short to be detected ( Figure 5A ) . These results were also in agreement with our kinetic measurements at 0 . 5:1 DNAJA2:tau ratio , which showed an 88% reduction in fibril mass ( Figure 5—figure supplement 1A ) . Thus , the ability of DNAJA2 chaperone to interact with all tau species and to prevent both elongation and nucleation rates makes it a potent suppressor of tau aggregation . DNAJB1 , which does not interact with tau monomers , was less effective than DNAJA2 in reducing fiber size and mass ( 67% reduction; Figure 5—figure supplement 1B ) , and at 0 . 5:1 DNAJB1:tau the fibers were still visible . These were , however , significantly shortened , with sizes ranging from 0 . 2 to 1 . 0 µm , and at a stoichiometric ratio of 1:1 no fibers were observed ( Figure 5B ) . HSPB1 chaperone , on the other hand , affected fiber formation differently when added at the start of the tau aggregation reaction . While a clear dose-dependent decrease in the amount of the tau fibrils was observed ( Figure 5C and Figure 5—figure supplement 1C ) , unlike DNAJB1 and DNAJA2 , HSPB1 was unable to completely suppress the formation of fibers at sub-stoichiometric concentrations . Even in the presence of twofold excess of HSPB1 chaperone , a few fibers per micrograph were still observed , which , interestingly , were somewhat shorter and straighter in appearance compared to the long twisted fibrils of tau alone ( Figure 5C ) . Although the reduced elongation rate in the presence of HSPB1 decelerates the aggregation process , HSPB1 only has a limited effect on the final fiber mass , as also evident in our kinetic measurements , showing only 45% reduction in fibril mass even at equimolar HSPB1:tau concentrations . Thus , the weak interaction of HSPB1 with tau monomers along with its limited ability to only slow down the fibril elongation rate is not sufficient to efficiently prevent tau incorporation into the amyloid fibers . Thus , DNAJB1 and DNAJA2 chaperones , which bind to both aggregation-prone tau species and fibers , are significantly more efficient in preventing the formation of mature fibers than HSPB1 , which can only bind to tau monomers . Overall it appears that the DNAJB1 and DNAJA2 chaperones , despite being very similar in structure , display distinct differences in their interaction with tau . While DNAJA2 binds to all tau species – monomers , aggregation-prone tau , and fibers – DNAJB1 does not bind at all to inert tau monomers , but interacts strongly with the aggregation-prone species as well as mature fibers . This disparity between the chaperones could be explained by utilization of different structural domains to recognize the various tau species; however , no structural information is currently available for either DNAJA2 or DNAJB1 in complex with tau4R . We therefore utilized NMR to map the binding sites on the chaperones for the various tau species . Both DNAJA2 and DNAJB1 are homodimeric proteins comprising an eponymous N-terminal JD that is essential for Hsp70 activation , two putative substrate binding domains , CTDI and CTDII , and a C-terminal dimerization domain . In addition , DNAJA2 , as all class A JDPs , has a ZFLR insertion in CTDI ( Kampinga and Craig , 2010; Ayala Mariscal and Kirstein , 2021; Jiang et al . , 2019 ) , while DNAJB1 has an autoinhibitory GF region connecting the JD to CTDI and blocking premature interaction of Hsp70 with the JD ( Faust et al . , 2020 ) . Due to the large size of the DNAJA2 dimer ( 90 kDa ) , which hampers NMR experiments , we used a monomeric version of Ydj1 , a DNAJA2 homologue from Saccharomyces cerevisiae , that contains only the substrate binding and ZFLR domains ( Li et al . , 2003 ) . This construct ( 27 kDa ) of 2H , 15N-labeled DNAJA2 ( Ydj1 ) 111-351 was far more amenable to NMR and , as previously reported , gave a high-quality 1H-15N HSQC-TROSY spectrum ( Jiang et al . , 2019; Figure 6—figure supplement 1A ) . Addition of twofold excess of tau4R caused CSPs in the first DNAJA2 substrate-binding domain ( Li et al . , 2003 ) , located in CTDI ( Figure 6A , C and Figure 6—figure supplement 1A ) . Specifically , tau4R ( most likely via the hydrophobic PHF6 and PHF6* motifs ) binds to a hydrophobic pocket located between β-strands 1 and 2 ( Figure 6C , colored purple ) . The binding was in fast exchange , and no reduction in peak intensities to DNAJA2 ( Ydj1 ) 111-351 residues was observed ( Figure 6—figure supplement 1A ) , in agreement with the relatively low affinity of DNAJA2 for monomeric tau ( 43 µM , Figure 2C ) . DNAJB1 , despite its CTDI having 56% identity to the CTDI tau-binding region of DNAJA2 , showed no interaction with the monomeric tau ( Figure 6E and Figure 6—figure supplement 1E , F ) , as previously seen from the tau side ( Figure 1C ) . We then repeated the binding experiment using aggregation-prone tau species , generated by the addition of heparin ( Eschmann et al . , 2017 ) . Interestingly , this tau species , unlike the inert tau monomer , caused significant peak broadening to residues in the second substrate binding region of DNAJA2 , located in CTDII ( Figure 6B and Figure 6—figure supplement 1B ) , whereas only small changes were detected in the CTDI region ( Figure 6—figure supplement 1B ) . Hence , the DNAJA2 substrate-binding groove in CTDI interacts predominantly with the inert , monomeric tau , while the aggregation-prone tau species preferentially binds to CTDII ( Figure 6C , D ) . In order to verify that our binding results were not affected by the deletion of the dimerization domain in the DNAJA2 ( Ydj1 ) 111-351 construct , we then repeated the binding experiments using a construct comprising only CTDII and the Ydj1 dimerization domains , termed DNAJA2 ( Ydj1 ) 256-409 . This DNAJA2 variant , as expected , showed no binding to monomeric tau in our NMR experiments , confirming that tau indeed binds to the CTDI region , which was lacking in this construct ( Figure 6—figure supplement 1C ) . Addition of heparin-bound tau , however , caused significant CSPs to the spectrum of DNAJA2 ( Ydj1 ) 256-409 ( Figure 6—figure supplement 1D ) , again confirming the binding of this aggregation-prone tau species to CTDII . Having this second , distinct tau-binding domain thus explains both the high affinity of DNAJA2 toward the aggregation-prone tau and the ability of the chaperone to interact with the two tau species . We then tested the binding of DNAJB1154-341 ( a dimeric protein lacking the N-terminal JD ) to this aggregation-prone tau species . The interaction with tau-heparin , however , caused severe peak broadening , preventing us from obtaining site-specific information . The high molecular mass of the complex formed between multiple aggregation-prone tau monomers and the DNAJB1-dimer is presumably responsible for this effect . To overcome this problem , we utilized a 2H 13CH3-ILVM sample of full-length DNAJB1 that gave a high-quality 1H-13C HMQC spectrum even upon complex formation with the aggregation-prone tau ( Figure 6—figure supplement 1G ) . Selective peak broadening was detected in methyl residues located in CTDII ( Figure 6F , G and Figure 6—figure supplement 1G ) , indicating that , similarly to DNAJA2 chaperone , the aggregation-prone tau binds to CTDII in DNAJB1 . This CTDII site in DNAJB1 was also recently identified as a binding site for another amyloid-forming protein , α-synuclein ( Faust et al . , 2020 ) . Hence , DNAJB1 and DNAJA2 recognize seed-competent tau4R species predominantly via CTDII , yet the structural differences between the CTDI domains of the chaperones likely cause the diverging specificities for the various tau4R species .
Hsp70 chaperones are known to be key factors in tau quality control and turnover ( Miyata et al . , 2011 ) ; however , the contributions of their Hsp40 co-chaperones remain poorly understood . In this study , we describe the effects of Hsp40 chaperone family members , DNAJA2 and DNAJB1 , on tau amyloid fiber formation in comparison to the well-characterized tau aggregation suppressor HSPB1 . DNAJA2 chaperone was previously identified as a potent suppressor of tau aggregation ( Mok et al . , 2018 ) . Our results demonstrate that DNAJA2 can interact simultaneously with multiple tau species , which can explain its high effectiveness in aggregation prevention . DNAJA2 does not only bind tau monomers via its CTDI , thus preventing fiber growth by monomer addition , but it also binds aggregation-prone tau species via its CTDII , which effectively reduces the speed of nucleation . Moreover , DNAJA2 even associates with mature tau fibers , which provides an additional pathway for inhibiting the incorporation of new monomers into the growing fibers ( Figure 7 ) . Given this multiplicity of interfering pathways , DNAJA2 has the potential to serve as an early protective cellular factor that limits tau aggregation , which explains the correlation between distorted cellular DNAJA2 levels and pathology-linked nucleation sites ( Mok et al . , 2018 ) . Remarkably , in comparison to DNAJA2 , DNAJB1 displayed significantly different interactions with tau despite the high structural similarity between both chaperones . In fact , unlike DNAJA2 , DNAJB1 does not interact with inert tau monomers , although its CTDI is highly homologous to that of the class A chaperone . DNAJB1 affinity , however , is increased by orders of magnitude for the aggregation-prone tau species . There , DNAJB1 also functions as a bona fide chaperone , effectively hindering tau aggregation by preventing the formation of the seeding nuclei , as well as by stably binding to the amyloid fibers themselves , slowing down their growth ( Figure 7 ) . Interestingly , both interactions are mediated by CTDII , thus leaving the CTDI site free to potentially recruit Hsp70 chaperones and initiate the disaggregation of fibers when these are formed . This mode of action of DNAJB1 differs significantly from that of the well-characterized tau aggregation-suppressor HSPB1 , as well as from the recently described activity of the DNAJC7 chaperone ( Hou et al . , 2020 ) . These chaperones preferentially bind to the PHF6 repeats in inert tau monomers , thus protecting these aggregation-prone regions and preventing them from being incorporated into the growing fibers . By solely interacting with inert tau monomers , though , HSPB1 can only effectively inhibit the rate of fibril elongation , and thus has a limited effect of preventing tau amyloid formation ( Figure 7 ) . In contrast , the actions of DNAJA2 and DNAJB1 result in a significant decrease not only in fibril mass , but also length . This function of the Hsp40 chaperones , compared to HSPB1 , could be attributed to their ability to interact with the forming tau fibers , which efficiently blocks the incorporation of additional tau monomers . In addition , the Hsp40 chaperones bind aggregation-prone tau species , thereby blocking their elongation into the mature tau fibers . The relatively minor differences in the aggregation-prevention abilities of DNAJB1 and DNAJA2 , along with the relatively poor performance of HSPB1 in preventing fibril growth , suggest that , while interaction with soluble tau monomers helps slow aggregation , this does not significantly contribute to the ability of chaperones to prevent amyloid growth once it has begun . It is therefore possible to envision a scenario in which HSPB1 delays amyloid formation during the early stages of tau aggregation via interaction with the monomers , and Hsp40 chaperones later interact with seeds and more mature species to further hinder the fibril formation process . One open question is how these chaperones affect the aggregation of tau mutants linked to tauopathies . It has been hypothesized that some variants , such as P301L , may be capable of ‘avoiding’ the chaperone system , thus possibly contributing to the disease pathology . Such behavior was indeed recently observed with the DNAJC7 chaperone , which was found to have a significantly reduced affinity to the tau P301L mutation ( Hou et al . , 2020 ) . Furthermore , in our aggregation-prevention assays , a reduction was observed in the activity of HSPB1 when incubated with the P301L variant compared to wild-type tau ( Mok et al . , 2018; Figure 4—figure supplement 4B ) . For these two chaperones , their reduced efficacy is likely due to the equilibrium of the P301L mutation shifting toward an aggregation-prone seeding conformation of tau ( Chen et al . , 2019 ) , for which both DNAJC7 ( 30 ) and HSPB1 ( Figure 4B ) display reduced affinities . In contrast , DNAJA2 and DNAJB1 remained effective in suppressing the aggregation of the P301L variant of tau4R ( with the anti-aggregation activity of DNAJB1 being even higher for this tauopathy mutant , Figure 4—figure supplement 4C ) , indicating that these Hsp40 chaperones could be effective in suppressing a wide range of tauopathies . A second , crucial question that has yet to be answered is whether the reduction of fibril growth by the Hsp40 chaperones , which results in generation of smaller tau fibrils , is indeed beneficial for cellular homeostasis . A recent study showed that the disaggregation of tau by the DNAJB1/Hsp70/Hsp110 chaperones generates low-molecular-weight tau species , which were seeding-competent in cell culture models ( Nachman et al . , 2020 ) . Hence , chaperone-mediated tau disaggregation may not be beneficial per se , but may instead be involved in the prion-like propagation of tau pathology . The smaller tau species generated during the DNAJA2 and DNAJB1 aggregation-prevention processes could then have similar prion-like propagation properties , acting as seeds that can sequester more tau protein into amyloid aggregates . In such a case , chaperone-mediated aggregation prevention would , in fact , accelerate the progression of disease , ultimately proving detrimental to cell health . HSPB1 chaperones , on the other hand , do not generate smaller tau species during their aggregation prevention and could therefore be more beneficial in slowing the progression of disease , despite their lower chaperoning activity . However , another important aspect to consider is that aggregation prevention in the cell does not occur in isolation and can also be coupled to protein degradation via the proteasome or autophagy . These pathways could potentially be more potent in degrading smaller tau fibrils and aggregation-prone monomers than the mature fibrils , thereby rendering the activity of DNAJA2 and DNAJB1 beneficial overall . Hence , further studies will be required to understand the full role of Hsp40-mediated tau aggregation prevention in the cell , as well as to evaluate the therapeutic potential of the Hsp40 chaperone machineries in combating tauopathies .
Tau4R ( residues 244–372 , C322A ) wt and mutants , HSPB1 ( S15D , S78D , S82D , I181G , V183G ) , DNAJA2 , and DNAJB1 ( residues 154–341 ) were expressed in Escherichia coli BL21 ( DE3 ) cells from pET-29b ( + ) vector with a N-terminal His6 tag followed by a tobacco etch virus ( TEV ) protease cleavage site . Tau ( C322S ) , DNAJB1 , and Ydj1 ( yeast orthologue of DNAJA ) constructs were expressed from the pET-SUMO vector with an N-terminal His6 purification tag and a Ulp1 cleavage site ( DNAJB1 plasmid was a gift from B . Bukau , University of Heidelberg ) . Cells were grown in Luria Bertani broth ( LB ) to OD600 ≈ 0 . 8 at 37°C , and expression was induced by addition of 1 mM isopropyl-β-D-thiogalactoside ( IPTG ) . Cells expressing HSPB1 , DNAJA2 , and DNAJB1 chaperone variants were allowed to proceed overnight at 25°C and cells expressing tau constructs at 18°C . Isotopically labeled tau and tau4R proteins for NMR were grown in M9 H2O media supplemented with 15NH4Cl ( and 13C-glucose ) as the sole nitrogen ( and carbon ) source . Protein expression was induced with 1 mM IPTG at 18°C overnight . Labeled DNAJB1154-341 , Ydj1111-351 , and Ydj1256-409 were grown at 37°C in M9 D2O media supplemented with [2H , 12C]-glucose and 15NH4Cl as the sole source of carbon and nitrogen . In the case of DNAJB1 , 2-ketobutyric acid-13C4 , 3 , 3-d2 sodium salt ( 60 mg/L ) , 2-ketoisovaleric acid-13C4 , d3 sodium salt ( 80 mg/L ) , and 13C-L-methionine ( 100 mg/L ) ( Cambridge Isotope Laboratories ) were added 1 hr prior to induction with 1 mM IPTG , following the procedure of Tugarinov et al . , 2006 to produce U-2H , 15N , 13CH3-ILVM labeled protein . Proteins were expressed at 25°C overnight . Proteins were purified on a Ni-NTA HiTrap HP column ( GE Life Sciences ) . The purification tag was cleaved by the appropriate protease ( see Construct preparation ) , and the cleaved protein was further separated from the uncleaved protein , the tag , and the protease on a Ni-NTA HiTrap HP column . HSPB1 , DNAJA2 , and DNAJB1 chaperone variants were concentrated on an Amicon Ultra-15 10K molecular weight cutoff ( MWCO ) filter ( Millipore ) and further purified on a HiLoad 16/600 Superdex 200 pg gel filtration column ( GE Healthcare ) , equilibrated with 25 mM HEPES pH 7 . 0 , 150 mM KCl , and 2 mM DTT . Tau constructs were concentrated on an Amicon Ultra-15 3 . 5K MWCO filter ( Millipore ) and further purified on a HiLoad 16/600 Superdex 75 pg gel filtration column ( GE Healthcare ) equilibrated with 25 mM HEPES pH 7 . 0 , 300 mM KCl , and 2 mM DTT . Purity of proteins was confirmed by SDS-PAGE . Aggregation kinetics were measured in Synergy H1 microplate reader ( BioTek ) in black , flat-bottom , 96-well plates ( Nunc ) . Tau or tau4R variants ( 10 μM ) were pre-incubated in the presence or absence of indicated chaperones for 10 min at 37°C . All proteins in the assay were buffer exchanged into the assay buffer ( 50 mM HEPES pH 7 . 4 , 50 mM KCl , and 2 mM DTT ) . ThT ( Sigma ) at a final concentration of 10 μM was added , and the aggregation was induced by the addition of 2 . 5 μM freshly prepared heparin salt solution ( Sigma ) . Aggregation reactions were run at 37°C with continuous shaking ( 567 rpm ) and monitored by ThT fluorescence ( excitation = 440 nm , emission = 485 nm , bandwidth ) , using an area scan mode with a 3 × 3 matrix for each well . Black , flat-bottom , 96-well plates ( Nunc ) sealed with optical adhesive film ( Applied Biosystems ) were used . For data processing , baseline curves at same conditions but without heparin were subtracted from the data . Samples were run in triplicate , and the experiments were repeated at least four times with similar results . Tau seeds were prepared from mature tau fibers generated under similar conditions to these in the aggregation-prevention assays , except that ThT was omitted . The fibers were then sonicated using a probe sonicator ( Vibra-Cell , SONICS ) with an amplitude of 40% , for 30 s on and 10 s off , for a total of 7 min . The sonicated fibers were immediately added to monomeric tau , ThT , and DTT in a 96-well plate in the ratios described above and ThT fluorescence was measured as a function of time . The hydrodynamic radius of tau4R seeds was measured by DLS on a DynaPro DLS Plate Reader III ( Wyatt Technology ) . Tau seeds ( 10 µM ) were loaded on a 96-well black , clear-bottom plates ( Nunc ) , and subjected to a 5 min 3000 × g centrifugation to remove air bubbles from the wells . Measurements were carried out 20 times per well before averaging , with 5 s acquisitions at 25°C . Resulting autocorrelation functions were fitted with the equation ( 1 ) g2τ=1+βe-2Dq2τwhere β is the coherence factor , D is the translational diffusion coefficient , and q is the scattering wave vector given by ( 2 ) q=4πnλ0sin ( θ2 ) where n is the solvent refractive index ( n = 1 . 334 was used ) , λ0 is the wavelength used by the instrument , and θ is the scattering angle . The Stokes radius ( Rs ) was calculated from the translational diffusion coefficient , D , using the Stokes–Einstein equation ( 3 ) D=kBT6πηRswhere kB is Boltzmann coefficient ( 1 . 38⋅10−23kgm2s2K ) , T is the temperature ( 298 K ) , and η is the dynamic viscosity of our buffer . All aggregation kinetics were fitted with a saturation-elongation-fragmentation model ( Meisl et al . , 2016 ) using a critical nucleus size of nc=2 . The differential equation system for this model is ( 4 ) P˙ ( t ) =knm ( t ) nc+kmM ( t ) M˙ ( t ) =2kpKEm ( t ) KE+m ( t ) P ( t ) Here , Pt is the number concentration of fibrils , Mt is the mass concentration of a fibril , mt is the monomer concentration , kn is the nucleation rate , km is the fragmentation rate , kp is the elongation rate , and KE is the equilibrium constant for monomer addition to an existing fibril . The effect of heparin was not included explicitly in this model and is implicitly contained in the kinetic rates . For fitting , the ThT fluorescence signal Sit of the ith time trace was converted to the mass concentration of the fibrils Mit according to ( 5 ) Mit=mmaxSitSmax∞ Here , mmax is the highest tau4R concentration used in the experiments and Smax∞ is the ThT signal of the long-term plateau for the time trace with the highest tau4R concentration . The resulting mass concentration Mit was then fitted by numerically solving the differential equation system Equation 4 for Mt , using the initial conditions M0=0 , P0=0 , and m0=mi , where mi is the initial concentration of tau4R monomers for the ith time trace . Prior to fitting , the time traces Mit were smoothed by binning data points to reduce noise and to speed up fitting . The bin size was 2–5 data points . Fitting was performed using the 'differential evolution' method ( Storn and Price , 1997 ) in Mathematica 11 . 2 ( Wolfram ) . Importantly , kp cannot be independently obtained from unseeded data . We therefore arbitrarily set kp=1 for the global fit of the unseeded data at all monomer concentrations , thus obtaining k'n=knkp and k'm=kmkp . In a second step , we determined kn , km , and kp in a global fit of a data set including aggregation seeds using ( 6 ) P˙ ( t ) =km′kpM ( t ) M˙ ( t ) =2kpKEm ( t ) KE+m ( t ) P ( t ) with the initial conditions M0=M0 , P0=M0/L , and m0=m0=10μM . Here , M0 is the mass concentration of the seeds and L is the length of the seeds . We determined L by measuring the Stokes radius Rs=55nm of the seeds using DLS ( see above ) . We then modeled the seeds as an ellipsoid with a long axis a ( seed length ) and a short axis b ( fibril thickness ) . The friction coefficient for an ellipsoid is given by ξe=6πηa/ln2a/b , which must be identical to that of a sphere with the Stokes radius Rs given by ξs=6πηRs . From the equality , the fibril length a can be determined given that the fibril thickness b is known . Based on existing cryo-EM structures of tau-fibrils ( 6QJH , 6QJM , 6QJP ) , we estimated b~10nm , which results in a~200nm . Given the spacing of tau monomers in a fibril of approximately 2 nm , we estimated a seed length of L=100 tau monomers . Error estimates of the kinetic rates were obtained by fitting two independent data sets . Fitting of the data in the presence of chaperones was performed for each kinetic trace individually by fixing the kinetic rates to those determined in the absence of chaperone and only allowing one rate to vary at a time . An exception was the data set for DNAJA2 in which we performed a simultaneous global fit of seeded and unseeded experimental traces ( Equations 4–6 ) to identify the simultaneous effect of DNAJA2 on kn , kp , and km . For all fits in which only a single parameter was scanned , we used 'simulated annealing' to optimize the parameters . We would like to note that the amplitude of the aggregation kinetics was not a free-fitting parameter but was determined by the total concentration of tau4R monomers ( 10 μM ) in the experiment . Hence , fitting of the aggregation kinetics in the presence of chaperones assumes that the presence of chaperone does not alter the ThT concentration accessible in solution to stain the fibrils . All NMR experiments were carried out at 25°C on 14 . 1 T ( 600 MHz ) , 18 . 8T ( 800 MHz ) , or 23 . 5 T ( 1000 MHz ) Bruker spectrometers equipped with triple resonance single ( z ) or triple ( x , y , z ) gradient cryoprobes . The experiments were processed with NMRPipe ( Delaglio et al . , 1995 ) and analyzed with NMRFAM-SPARKY ( Goddard and Kneller , 2000 ) and CCPN ( Vranken et al . , 2005 ) . Assignments for tau4R were transferred from the BMRB ( entry 19253 ) and corroborated by HNCACB , CBCA ( CO ) NH , HN ( CA ) CO , and HNCO experiments on a 4 mM sample of [U- 15N , 13C]-labeled tau4R in 50 mM HEPES pH 7 . 4 , 50 mM KCl , 1 mM DTT , 0 . 03% NaN3 , and 10% D2O . The assignment experiments were recorded on an 800 MHz magnet , resulting in the unambiguous assignment of 90% of non-proline residues . Tau-heparin complex assignments were obtained by recording 3D HNCA , CBCA ( CO ) NH , and HN ( CA ) CO on a 2 . 5 mM [U-13C , 15N]-labeled tau4R sample supplemented with 2 . 5 mM heparin . The experiments were recorded on an 800 MHz magnet and 88% of non-proline residues were assigned . Secondary structure propensities for tau4R and tau4R-heparin complex were calculated from backbone C′ , Cα , and Cβ , 1H , 15N chemical shifts following a procedure described in Marsh et al . , 2006 . Proline and cysteine residues were omitted from this calculation . Backbone amide 1DNH RDCs were measured using a 300 μM sample of tau4R , tau P301L , or tau P301S diluted in 50 mM HEPES pH 7 . 4 buffer with 100 mM KCl . The one-bond N-H RDCs were determined by using inphase-antiphase ( IPAP ) -HSQC experiments ( Ottiger et al . , 1998 ) , and 1DNH values were calculated as the difference between splittings measured in the isotropic phase and in a sample in which tau4R , tau4R P301S , or tau4R P301L had been aligned in 5 mg/mL Pf1 bacteriophage ( Asla ) or in 4 . 5% ( vov/vol ) C12E5/n-hexanol alignment medium ( Sigma ) . The experiments were recorded on a 1000 MHz Bruker spectrometer and RDCs ranged from +20 to −10 Hz . The heparin-bound tau sample ( 1:1 tau:heparin molar ratio ) was only aligned in 16 mg/mL bacteriophage pf1 and the measured RDCs ranged from +9 to −7 Hz . Tau interaction with chaperones was assayed for 200 µM samples of [U-15N]-labeled tau , tau4R , or tau4R P301S and P301L mutants . Tau variants were measured alone or upon addition of heparin ( 200 µM ) and/or chaperones ( 100 or 400 µM; as indicated in spectrum ) in 50 mM HEPES pH 7 . 0 , 50 mM KCl , 1 mM DTT , 0 . 03% NaN3 , and 10% D2O . 1H-15N HSQC-TROSY spectra were acquired for each sample , and peak intensities were determined by quantifying peak volumes . Regions of tau4R with signal loss greater than one standard deviation from the average intensity ratio were determined to be the regions of binding . High salt binding experiments were performed with samples of 15N-tau4R ( 200 µM ) and 400 µM [U-1H]-labeled chaperones in 50 mM HEPES pH 7 . 0 , 300 mM KCl , 1 mM DTT , 0 . 03% NaN3 , and 10% D2O . Binding was determined by calculating intensity ratios as described above . The binding of DNAJB1 and DNAJA2 chaperones to tau4R was measured by acquiring 1H-15N HSQC-TROSY spectra for 200 μM [U- 2H , 15N]-labeled DNAJB1154-341 or Ydj1111-351 alone or with twofold excess of deuterated tau4R . The reactions were measured in 50 mM HEPES pH 7 . 0 , 50 mM KCl , 1 mM DTT , 0 . 03% NaN3 , and 10% D2O . Backbone DNAJB1154-341 and Ydj1111-351 assignments were available through the BMRB ( entries 27998 and 28000 , respectively ) . The interaction of full-length DNAJB1 to tau4R was determined by acquiring 1H-13C HMQC methyl-TROSY spectra ( Tugarinov et al . , 2003 ) for 100 μM [2H , 13CH3]-ILVM-labeled DNAJB1 alone or with 200 μM 2H-tau4R ( or tau-heparin complex ) in 50 mM HEPES pH 7 . 4 , 100 mM KCl , 2 mM DTT and 0 . 03% NaN3 in 100% D2O . ILVM assignments for full-length DNAJB1 were taken from previous work in our lab ( Faust et al . , 2020 ) . Binding regions were determined by intensity ratio as described above . The interaction of tau4R with heparin was monitored by 2D 1H–15N HSQC experiments . Heparin ( 40–400 μM ) was titrated into 200 μM of 15N-labeled tau4R in 50 mM HEPES pH 7 . 0 , 50 mM KCl , 1 mM DTT , 0 . 03% NaN3 , and 10% D2O and chemical shifts were recorded . CSPs were calculated from the relation ( 7 ) Δδ=ΔδH2+ ( ΔδN5 ) 2where ΔδH is the amide proton chemical shift difference , and ΔδN is the 15N backbone chemical shift difference . CSPs greater than one standard deviation from the mean were considered significant . Tau fibrils or tau fibrils-chaperone mixtures ( 10 µl ) were deposited on glow-discharged carbon-coated copper EM grids ( Electron Microscopy Sciences ) , washed with three consecutive drops of 1% w/v Uranyl-formate , and air-dried . Imaging was performed on an FEI T12 Spirit transmission electron microscope at 120 kV and a magnification of 9300–30 , 000 times , equipped with a Gatan OneView CMOS 4K × 4K CCD camera . Preformed tau4R fibers ( 10 μM ) were incubated with HSPB1 , DNAJB1 , and DNAJA2 chaperones ( 10 μM ) for 20 min at 37°C in 50 mM HEPES pH 7 . 4 and 50 mM KCl . Tau fibers were separated from the unbound chaperones by centrifugation at 16 , 900 g for 30 min . The pellets were washed , resuspended in 50 μL of buffer with 20% SDS , and sonicated for 10 min . Samples were incubated for 5 min at 95°C and run on a 4–20% gradient SDS-PAGE gel ( GenScript ) . Steady-state equilibrium binding of DNAJA2 , DNAJB1 chaperones to preformed tau fibers was measured by fluorescence polarization using 100 nM of fluorescently tagged chaperones ( DNAJB1 G194C-AF488 or DNAJA2-AF488 ) . Steady-state equilibrium binding of DNAJB1 , DNAJA2 , and HSPB1 to monomeric tau was measured by fluorescence anisotropy using 100 nM of fluorescently tagged tau ( tau C291S , C322S , L243C-AF488 ) . Samples were allowed to equilibrate for 10 min at 37°C , and measurements were performed on a Tecan SPARK 10M plate reader in black , flat-bottomed 384 square well plates . The excitation filter was centered on 485 nm with a bandwidth of 20 nm , and the emission filter was centered on 535 nm with a bandwidth of 25 nm . The gain and Z position were optimized from a well in the center of the binding curve , followed by calibration of the G factor . 60 flashes were performed per well . Data were fit to a one-site binding model using OriginPro version 2018 . | Several neurological conditions , such as Alzheimer’s and Parkinson’s disease , are characterized by the build-up of protein clumps known as aggregates . In the case of Alzheimer’s disease , a key protein , called tau , aggregates to form fibers that are harmful to neuronal cells in the brain . One of the ways our cells can prevent this from occurring is through the action of proteins known as molecular chaperones , which can bind to tau proteins and prevent them from sticking together . Tau can take on many forms . For example , a single tau protein on its own , known as a monomer , is unstructured . In patients with Alzheimer’s , these monomers join together into small clusters , known as seeds , that rapidly aggregate and accumulate into rigid , structured fibers . One chaperone , HSPB1 , is known to bind to tau monomers and prevent them from being incorporated into fibers . Recently , another group of chaperones , called J-domain proteins , was also found to interact with tau . However , it was unclear how these chaperones prevent aggregation and whether they bind to tau in a similar manner as HSPB1 . To help answer this question , Irwin , Faust et al . studied the effect of two J-domain proteins , as well as the chaperone HSBP1 , on tau aggregation . This revealed that , unlike HSBP1 , the two J-domain proteins can bind to multiple forms of tau , including when it has already aggregated in to seeds and fibers . This suggests that these chaperones can stop the accumulation of fibers at several different stages of the aggregation process . Further experiments examining which sections of the J-domain proteins bind to tau , showed that both attach to fibers via the same region . However , the two J-domain proteins are not identical in their interaction with tau . While one of them uses a distinct region to bind to tau monomers , the other does not bind to single tau proteins at all . These results demonstrate how different cellular chaperones can complement one another in order to inhibit harmful protein aggregation . Further studies will be needed to understand the full role of J-domain proteins in preventing tau from accumulating into fibers , as well as their potential as drug targets for developing new treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2021 | Hsp40s play complementary roles in the prevention of tau amyloid formation |
Mutations in human Atrophin1 , a transcriptional corepressor , cause dentatorubral-pallidoluysian atrophy , a neurodegenerative disease . Drosophila Atrophin ( Atro ) mutants display many phenotypes , including neurodegeneration , segmentation , patterning and planar polarity defects . Despite Atro’s critical role in development and disease , relatively little is known about Atro’s binding partners and downstream targets . We present the first genomic analysis of Atro using ChIP-seq against endogenous Atro . ChIP-seq identified 1300 potential direct targets of Atro including engrailed , and components of the Dpp and Notch signaling pathways . We show that Atro regulates Dpp and Notch signaling in larval imaginal discs , at least partially via regulation of thickveins and fringe . In addition , bioinformatics analyses , sequential ChIP and coimmunoprecipitation experiments reveal that Atro interacts with the Drosophila GAGA Factor , Trithorax-like ( Trl ) , and they bind to the same loci simultaneously . Phenotypic analyses of Trl and Atro clones suggest that Atro is required to modulate the transcription activation by Trl in larval imaginal discs . Taken together , these data indicate that Atro is a major Trl cofactor that functions to moderate developmental gene transcription .
Atrophin family transcription factors are conserved transcriptional corepressors essential for development . Humans have two Atrophin genes; Atrophin 1 ( ATN1 ) and Atrophin 2 . A polyglutamine expansion in ATN1 is responsible for dentatorubral-pallidoluysian atrophy ( DRPLA ) a progressive disorder of ataxia , myoclonus , epilepsy , intellectual deterioration and dementia ( Koide et al . , 1994; Nagafuchi et al . , 1994 ) . Drosophila has only one Atrophin ( Atro ) gene , and expression of polyglutamine expansion Atro also lead to neurodegeneration ( Napoletano et al . , 2011 ) . Loss of Atro results in defects in planar polarity , segmentation , and eye , wing and leg developmental defects ( Erkner et al . , 2002; Zhang et al . , 2002; Fanto et al . , 2003; Charroux et al . , 2006; Napoletano et al . , 2011; Saburi et al . , 2012; Sharma and McNeill , 2013 ) . Atro is required for proper expression of en in the embryo and represses the gap genes Krüppel and knirps ( Erkner et al . , 2002; Zhang et al . , 2002; Haecker et al . , 2007 ) . Atro also affects epidermal growth factor receptor ( EGFR ) , Decapentaplegic ( Dpp ) and Hedgehog signaling ( Erkner et al . , 2002; Charroux et al . , 2006; Zhang et al . , 2013 ) . Despite regulating many pathways and processes , only three direct Atro targets have been defined ( knirps , fat and dpp ) ( Wang et al . , 2006; Napoletano et al . , 2011; Zhang et al . , 2013 ) . These targets cannot adequately explain the diverse phenotypes exhibited by loss of Atro . Atro physically interacts with histone deacetylase 1 ( HDAC1 ) and the histone methyltransferase , G9a , through its SANT and ELM2 domains , respectively ( Figure 1A ) ( Wang et al . , 2006 , 2008 ) . These interactions are thought to contribute to Atro’s repressor activity . Atro also interacts with other cofactors such as the co-repressor , Scribbler ( also called Brakeless ) ( Wang et al . , 2006; Haecker et al . , 2007 ) . Atro does not have a DNA-binding sequence and has been shown to interact with DNA via interactions with nuclear receptors ( Wang et al . , 2006 ) . We show here that a major means by which Atro regulates transcription is with the Drosophila GAGA factor , Trithorax-like ( Trl , also called GAF ) . Trl is an essential transcription factor that directly binds to GA repeats ( Biggin and Tjian , 1988; Soeller et al . , 1988 ) . Trl regulates many developmentally important genes including the segment polarity gene , engrailed ( en ) ( Farkas et al . , 1994; Bejarano and Busturia , 2004 ) . Trl can interact with other transcription factors ( e . g . Yorkie , Tramtrack ) to regulate transcriptional targets in patterning and growth control ( Pagans et al . , 2002; Oh et al . , 2013 ) . 10 . 7554/eLife . 23084 . 003Figure 1 . Atrophin ChIP-seq results . ( A ) A schematic of Atro protein is shown . Atro’s N-terminal side has ELM2 and SANT domains to interact with HDAC1 and G9a , respectively . Atro’s C-terminal domain interacts with Tailless and Fat . ( B ) Fraction of the common Atro regions overlapping with various genomic features ( genome release 5 . 57 ) ( Attrill et al . , 2016 ) . ( C ) The fraction of common Atro regions overlapping genes ( including 500 bp upstream of transcription start site ) with high , medium or low expression in S2 cells ( data from [Cherbas et al . , 2011] ) . ( D ) Venn diagram showing overlap of the common regions from three biological replicates ( S1-3-5 ) with two additional , independent biological replicates ( A1 and A2 ) using a different Atro antibody ( 4H6 ) . Only the peaks at the major chromosome arms ( heterochromatin are excluded ) are compared , and thus , only 1275 peaks are used in S1-3-5 for this analysis instead of 1377 peaks . ( E ) Example overlap of Atro ChIP-seq peaks at the engrailed ( en ) locus . Atro #1–3 are the triplicates Atro ChIP-seq data; IgG #1–3 are the corresponding IgG ChIP-seq controls for the Atro ChIP-seq data . Atro ( 4H6 ) #1–2 are the independent Atro ChIP replicates ( shown in D ) . ( F ) Top 10 GO-term enrichment hits of the Atro ChIP-seq data . GO-term enrichments were done with PANTHER overrepresentation test with default parameters and Bonferroni correction ( Mi et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 00310 . 7554/eLife . 23084 . 004Figure 1—source data 1 . PANTHER GO-Term enrichment results ( Mi et al . , 2016 ) for Figure 1F . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 00410 . 7554/eLife . 23084 . 005Figure 1—source data 2 . Overlap of ChIP data sets with the three classes defined in PCA analysis of Atro peaks ( modENCODE ( downloaded from http://intermine . modencode . org/ ) and CBP [Philip et al . , 2015] ) for Figure 1—figure supplement 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 00510 . 7554/eLife . 23084 . 006Figure 1—figure supplement 1 . Principal component analysis of Atro peaks . ( A ) Score plot showing the first two components from principal component analysis using enrichment of all factors mapped by modENCODE ( S2 cells ) within Atro-binding regions . The Atro-binding regions are colored according to the three classes defined by hierarchical clustering . R2X values represent the fraction of the total variation explained by each component . ( B ) The mean scaled enrichment values for each factor withing the three classes . All factors with enrichment values below 0 . 4 in all classes were excluded . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 006 Trl was initially discovered to bind to the promoter of Ultrabithorax ( Ubx ) to activate transcription of Ubx in vitro ( Biggin and Tjian , 1988 ) . Trl is also required in transcriptional repression . This is supported by three pieces of evidence: First , Trl binds to Polycomb response elements , DNA sequences where Polycomb group proteins bind to repress transcription of target genes ( Horard et al . , 2000; Busturia et al . , 2001 ) . Second , Trl can physically associate with Polycomb Repressive Complex I ( Poux et al . , 2001 ) . Third , Trl mutations enhance Polycomb group mutations , indicating that Trl is required for Polycomb repression ( Mahmoudi et al . , 2003 ) . In addition to Trl’s ability to regulate transcription , Trl binding leads to open chromatin ( Tsukiyama et al . , 1994 ) and Trl maintains open chromatin ( Fuda et al . , 2015 ) , at least in part by interacting with chromatin remodeling complexes such as NURF and FACT ( Xiao et al . , 2001; Shimojima et al . , 2003 ) . Trl is also required for transcriptional pausing ( Lee et al . , 2008; Fuda et al . , 2015 ) . A working model is that Trl first binds to DNA to maintain open chromatin and later associates with other proteins to activate or repress transcription ( reviewed in [Granok et al . , 1995; Lehmann , 2004] ) . Here , we identify the genome-wide targets of Atro by chromatin-immunoprecipitation followed by sequencing ( ChIP-seq ) . Our ChIP data show that Atro binds to regulatory regions of en , and in the putative regulatory regions of multiple Dpp and Notch signaling components . Our genetic and phenotypic analyses of Atro show that Atro regulates Dpp and Notch signaling , via transcriptional regulation of thickveins and likely fringe , respectively . We find that Atro negatively modulates En expression , while Trl promotes En expression . Bioinformatic analyses reveal that Atro and Trl ChIP-seq data strongly overlap and sequential ChIP and coimmunoprecipitation experiments confirm Atro and Trl bind to the same loci simultaneously and associate with one another . These data indicate that Atro modulates developmental gene expression via Trl binding . Our results suggest that Trl uses Atro as a cofactor to modulate transcription activation .
ChIP-seq against Atro was carried out in Drosophila Schneider 2 ( S2 ) cells in three biological replicates ( three sets of cells were grown and ChIP’ed on different days ) . ChIP peaks were called with MACS2 ( Zhang et al . , 2008 ) for each biological replicate using IgG ChIP-seq as the background model . The resulting three lists ( containing 1757 , 3064 and 3375 peaks ) were intersected and peaks found in all three lists and with summits within 100 bp of each other were selected . The resulting list of 1377 peaks were associated with 1300 unique genes by proximity and treated as potential Atro targets ( Supplementary file 1 ) . We further mapped the 1377 Atro peaks relative to gene features and gene expression , which showed that the majority of peaks are located in actively transcribed genes , often close to the promoters or in introns ( Figure 1B , C ) . In addition , we quantified the enrichment of all chromatin factors mapped by modENCODE ( 44 proteins and 23 histone modifications ) at the Atro peaks and performed principal component analyses of our Atro peaks from major chromosome arms ( with heterochromatin excluded , thus 1275 peaks were used ) . After hierarchical clustering of the significant principal components , we defined three classes ( Figure 1—figure supplement 1A ) . We noticed that Class 2 ( blue , Figure 1—figure supplement 1A and B ) peaks are strongly enriched with Polycomb factors , indicating a potential connection between Atro and Polycomb factors . To validate these peaks , we also performed independent ChIP-seq in two biological replicates with an antibody raised against a different part of Atro . Peaks were called with MACS2 using input as background model and intersected with the peaks identified above . We compared the peaks from major chromosome arms from each of the ChIP-seqs . A large fraction of the Atro-binding regions identified with the first antibody overlapped the ones found with this second antibody ( Figure 1D , E ) . GO term enrichment analysis ( PANTHER , [Mi et al . , 2016] ) of our ChIP-seq data revealed that Atro targets are enriched in chromatin organization , development and morphogenesis ( Figure 1F ) . Interestingly , a potential direct target gene of Atro is engrailed ( en , Figure 1E ) , a critical and conserved regulator of development . There are two strong Atro peaks in the en promoter ( Figure 1E , within 2 . 4 kb upstream of the transcription start site [Kassis et al . , 1992] ) . During embryogenesis , Atro is proposed to repress en expression with Even-skipped ( Zhang et al . , 2002 ) . The regulatory region of en is complex ( Cheng et al . , 2014 ) , and it is not known if Atro regulates en expression in larval stages . Therefore , we generated null mutant clones of Atro ( Atro35 ) in larval imaginal discs and examined the expression of En by immunofluorescence . We found Atro35 clones have increased En levels in antennal , wing and leg imaginal discs ( Figure 2A , B , and data not shown ) . Interestingly , Atro35 clones only show increased En levels in clones located in the posterior compartment , where En is endogenously expressed ( Figure 2A–C ) . Notably , Atro35 clones in the anterior compartment cannot induce ectopic expression of En ( Figure 2C ) . These data suggest that Atro negatively modulates En expression levels to maintain moderate expression in imaginal discs . 10 . 7554/eLife . 23084 . 007Figure 2 . Atro is required for modulating En levels in larval imaginal discs . ( A ) Antennal disc clones of Atro35 have increased En levels ( arrows ) . ( B ) Atro35 wing disc clones in the posterior compartment have increased En levels ( yellow arrow ) . But Atro35 clones cannot induce ectopic En expression in the anterior compartment ( red arrow ) . ( C ) Atro35 clones cannot induce ectopic En expression ( dotted lines mark clonal borders , wing disc is shown here ) . ( A ) has posterior to the right; ( B–C ) have posterior to the left , dorsal up . All scale bars are 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 007 Atro mutant clones exhibit phenotypes similar to Dpp signaling defects such as leg patterning defects and expanded wing veins ( Erkner et al . , 2002; Zhang et al . , 2002 ) , suggesting Atro regulates Dpp signaling . Interestingly , our ChIP-seq data show that several Atro peaks occur in putative regulatory regions of several Dpp signaling pathway components ( e . g . thickveins , Daughters against dpp and schnurri , Figure 3A and Supplementary file 1 ) , suggesting Atro regulates Dpp signaling by directly regulating the expression of Dpp signaling pathway components . Thickveins ( Tkv ) is a critical Dpp receptor , and Atro35 clones show upregulation of the tkv-LacZ reporter in the wing ( Wehn and Campbell , 2006 ) . To confirm that Atro represses tkv expression , tkv transcript levels were assessed in Atro knocked down wing discs using in situ hybridization . Indeed , tkv expression is increased if Atro is knocked down ( Figure 3B ) , confirming Atro represses tkv expression in the wing . 10 . 7554/eLife . 23084 . 008Figure 3 . Atro regulates Dpp signaling . ( A ) An Atro ChIP peak is found inside the tkv locus . This peak is directly upstream of the transcription start site of tkv isoform D . ( B ) In situ hybridization showing enGal4 control wing disc with normal tkv expression . ( B’ ) shows in situ hybridization of Atro RNAi driven in the posterior half of the wing disc by enGal4 . tkv expression is increased in the posterior half . ( C ) Atro35 wing disc clones that cross the endogenous pMad regions have increased pMad levels along the interior border of the clone that is closest to the middle of the disc ( where the Dpp source is located ) . This indicates that Atro35 clones have increased Dpp signaling . ( D ) Atro35 wing disc clones cannot cause ectopic Mad phosphorylation ( yellow arrow heads ) . ( E ) Atro35 wing disc clones cannot induce ectopic Dpp . All clones are marked by the absence of GFP; all figures have posterior to the left , dorsal up . Scale bars in A and C are 100 μm; in B and D are 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 00810 . 7554/eLife . 23084 . 009Figure 3—figure supplement 1 . Model of how loss of Atro and Trl affect Dpp signaling in wing discs . ( A ) Dpp signaling in the wild-type situation is shown ( Tanimoto et al . , 2000 ) . ( B ) In Atro35 clones ( dark rectangle ) , Tkv protein levels are increased . The increase of receptors causes pMad to increase near the interior clonal border closest to the Dpp source . Atro35 clones near the anterior/posterior edges of the wing also increase Tkv levels but does not affect pMad levels . ( C ) TrlR85 clones ( dark rectangle ) cause a loss of Tkv protein levels and thus pMad levels also decrease within the clones . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 009 Atro binds within tkv’s gene locus and represses tkv transcription; however , it is not clear if this repression is important for the regulation of Dpp signaling . Mothers against dpp ( Mad ) is a downstream component of Dpp signaling , that is phosphorylated when the cell receives the Dpp signal ( Newfeld et al . , 1997 ) . Phosphorylated Mad ( pMad ) is found in a broad stripe around the Dpp source and is used as a read-out for Dpp signaling . We stained Atro35 clones with anti-pMad antibodies to assess Dpp signaling . Atro35 clones that are found in the endogenous pMad regions have increased pMad levels along the interior border of the clones that is closest to the centre of the discs , where the endogenous Dpp source is located ( Figure 3C ) . This pattern of increased pMad expression is expected if Tkv levels are increased in the clone , since increased Tkv levels in Atro35 clones would lead to an increase in Mad phosphorylation , as these cells will receive more Dpp signal compared to the adjacent non-clone cells ( see Figure 3—figure supplement 1B ) . Atro35 clones also have increased Optomotor blind ( Omb , a downstream target of Dpp signaling ) expression ( data not shown ) , indicating increased Dpp signaling in the clones ( Grimm and Pflugfelder , 1996 ) . However , Atro35 clones far from the Dpp source do not induce ectopic Mad phosphorylation , suggesting Dpp levels are not increased in Atro35 clones ( Figure 3D , arrow heads ) . Consistent with this , Atro35 clones in the anterior or posterior compartments do not cause ectopic Dpp expression ( Figure 3E ) . Our data indicate that Atro directly regulates tkv expression and thereby Dpp signaling . Atro ChIP data also revealed that Atro binds the putative regulatory regions of several Notch signaling components ( mastermind ( mam ) , Delta ( Dl ) , neuralized ( neur ) and fringe ( fng ) ( Figure 4A , Supplementary file 1 ) , suggesting that Atro may regulate Notch signaling . Previous studies had not indicated that Atro loss of function affected Notch pathway activity . Therefore , to explore the biological relevance of the link with Notch , we first tested for genetic interactions between Atro and Notch ( N ) . N is required for the development of the wing margin , and heterozygous N264-39 mutant ( a null allele of N ) flies have wing notches in the adult wing ( Figure 4B’ ) . We find that transheterozygous N264-39/+; Atro35 /+ mutant flies exhibit strikingly more severe wing notching than heterozygous N264-39 mutant alone ( Figure 4B” ) . No notching is observed in Atro35 /+ mutant flies . Similar results were obtained with another independent Atro allele , Atroj5A3 ( a P-element insertion allele , data not shown ) . Thus , Atro genetically interacts with N , suggesting Atro may play a role in N signaling . 10 . 7554/eLife . 23084 . 010Figure 4 . Loss of Atro leads to loss of Notch signaling phenotypes . ( A ) Atro ChIP peaks are found in the fng and neur loci , indicating that Atro may regulate these genes during development . ( B ) Wild-type wings have no notches at the wing margin . N264-39 heterozygous wings have wing notches ( B’ ) . Wing notch phenotype is enhanced in transheterozygous N264-39 , Atro35 wings ( B’’ ) . ( C ) Atro35 clones cause the Notch signaling reporter ( GFP ) to express in a diffused pattern ( arrow ) . Dotted line marks clonal borders . ( D ) Large Atro35 clones cause an autonomous loss of wing margin marker Cut as well as ectopic Cut expression on the posterior side of the clone ( arrow ) . Clonal borders are marked with dotted lines . ( E ) Atro35 clones have extra cells expressing the early R8 marker , Sens . Extra Sens—positive cells are found in clusters of two to three cells but one cell within each cluster has more Sens staining than the rest ( arrowheads ) . Clones are marked by absence of ß-Gal ( red ) in C and clones are marked by the absence of GFP in D and E . B have posterior to the bottom and C to E have posterior to the left . C and D have dorsal side up . Scale bars in C and D are 50 μm; in E is 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 01010 . 7554/eLife . 23084 . 011Figure 4—figure supplement 1 . Loss of Atro causes loss of wing margin markers via Notch signaling . ( A ) Atro35 clones have an autonomous loss of Wg expression . ( B ) Atro35 clones do not affect Cut expression in the antennal disc . ( C ) In situ hybridization of enGal4 control wing disc with normal fng expression . ( C’ ) In situ hybridization of a wing disc with Atro RNAi driven in the posterior half by enGal4 . fng expression is increased in the posterior half . A and C have the posterior side to the left , dorsal side up . B has posterior to the right . Scale bars are 50 μm in A and B; 100 μm in C . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 01110 . 7554/eLife . 23084 . 012Figure 4—figure supplement 2 . Atro35 clones have normal Boss expression but altered Emc levels . ( A ) Atro35 clones have extra Sens-positive cells ( blue ) but normal amount of Boss staining ( red ) . ( B ) Atro35 clones have reduced Emc levels ( arrow ) anterior to the morphogenetic furrow ( arrowhead ) but normal levels posterior to the furrow . All clones are marked by absence of GFP . All pictures have posterior on the left . Scale bar in A is 25 μm; in B is 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 01210 . 7554/eLife . 23084 . 013Figure 4—figure supplement 3 . Loss of Atro have eye phenotypes similar to loss of Notch signaling . ( A ) Atro35 clones have a loss of R7 cells . R7 cells are marked by Pros and Runt co-expression ( purple ) . Yellow outline marks the clone boundary . ( B ) Atro35 clones have a loss of Cut expression but some Cut positive cells are still found inside the clone . ( C ) Atro35 clones have a loss of Tramtrack isoform 69 ( Ttk69 ) expression . Clonal boundary is marked by yellow outline . Clones are marked by the absence of GFP . All pictures have posterior to the left . Scale bars are 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 013 Since Atro and N genetically interact , we decided to test if loss of Atro affects wing margin development . In the wing , a line of N signaling induces the expression of wing margin markers Wingless ( Wg ) and Cut ( Ct ) ( de Celis et al . , 1996 ) . N signaling in the wing can also be visualized with a N signaling reporter ( NRE-GFP ) ( Housden et al . , 2012 ) , which expresses in a sharp line in the developing wing margin . Atro35 clones that cross the wing margin disrupt the expression pattern of the NRE-GFP reporter ( Figure 4C ) , resulting in diffuse expression of the NRE-GFP reporter . In addition , Atro35 clones that cross the wing margin cause a loss of wing margin markers ( Figure 4D and Figure 4—figure supplement 1A ) . In some cases , Atro35 clones induce ectopic Ct expression just outside of the posterior clonal border ( Figure 4D , arrow ) . In contrast to the wing , Atro35 clones do not affect Ct expression in the antennal discs , where a requirement of N signaling for Ct expression has not been shown ( Figure 4—figure supplement 1B ) . Atro binds to the fng promoter ( Figure 4A , [Yang et al . , 2013] ) , and in situ hybridization shows Atro knockdown causes increased fng transcription ( Figure 4—figure supplement 1C ) . These observations suggest that Atro may be regulating fng and/or other factors to affect N signaling . We also checked if Atro regulates N signaling in the eye . In larval eye discs , R8 photoreceptors differentiate from a three-cell equivalence group , which differentiates into R2 , R5 and R8 photoreceptors ( reviewed in [Frankfort and Mardon , 2002; Tsachaki and Sprecher , 2012] ) . N signaling is required for lateral inhibition in this equivalence group , such that one cell receives the least amount of N signaling and differentiates into R8 . Thus , N signaling loss of function mutants have extra R8’s . To see if Atro affects R8 development , we stained Atro35 clones with Senseless ( Sens ) , an early R8 marker . Atro35 clones have extra Sens positive cells ( Figure 4E ) , clustered together in groups of two to three cells , resembling the number and shape of the R2/5/8 equivalence group . This suggests that lateral inhibition is defective in Atro35 clones . However , we noticed that there is one cell that has more Sens staining than the rest within each Sens-positive cell cluster ( Figure 4E , arrowheads ) . Although Atro35 clones can induce extra Sens-positive cells , these clones do not have an excess of cells marked with a late R8 marker , Bride of sevenless ( Boss ) ( Figure 4—figure supplement 2A ) . In addition , N signaling is required for the differentiation of R7 photoreceptors and cone cells , and loss of N signaling results in a lack of R7 and cone cells in the eye ( Cooper and Bray , 2000; Flores et al . , 2000; Tomlinson et al . , 2011 ) . We found that Atro35 clones also lose R7 and cone cell markers ( Figure 4—figure supplement 3 ) . Extra macrochaetae ( emc ) , a downstream target of N signaling ( Bhattacharya and Baker , 2009 ) , is another potential direct target of Atro ( Supplementary file 1 ) . Indeed , Atro35 clones have reduced Emc protein levels anterior to the morphogenetic furrow ( Figure 4—figure supplement 2B ) . Since Atro peaks are present in/very close to multiple N signaling components and downstream targets ( fng , numb , Dl , neur , emc , Figure 4A , Supplementary file 1 ) , Atro may directly regulate multiple targets to affect N signaling in the eye . Although Atro does not bind to DNA directly , it binds with other cofactors to associate with DNA ( e . g . Tailless [Wang et al . , 2006] ) . We reasoned that analysis of the DNA sequences obtained from the Atro ChIP-seq could identify novel potential cofactors and binding motifs of Atro to gain insight into how Atro regulates developmental signaling . Therefore , we performed de novo motif analysis of our ChIP-seq data using MEME-ChIP ( Bailey et al . , 2009 ) ( Figure 5A , Figure 5—figure supplement 1 ) . The top motif discovered is a GA repeat , the binding motif of Trithorax-like ( Trl , Figure 5A ) . The Trl binding motif is enriched within Atro ChIP peak summits , indicating strongest Atro binding . Other enriched motifs include Twin of eyeless ( Toy ) and Mad ( Figure 5A ) . Next , we looked for overlap between the Atro ChIP-seq data with other ChIP datasets from S2 cells . We reasoned that we should see an increase of overlap between Atro and other ChIP data if we looked more specifically at genomic locations with higher Atro ChIP signal using Atro peaks from major chromosome arms ( excluding heterochromatin ) . Therefore , we plotted the amount of Atro overlap with other transcription factors over that expected by chance against the number of genomic locations grouped according to the amount of Atro binding ( using genomic data from modENCODE ( downloaded from http://intermine . modencode . org/ ) , CBP [Philip et al . , 2015] , and Yki [Oh et al . , 2013] ) . In this analysis , several factors , including Trl , CBP ( Nejire ) , the replication proteins Orc2 and MCM , Yorkie ( Yki ) and Polycomb proteins , showed increased overlap with increasing Atro binding ( Figure 5B ) . Interestingly , the strongest overlap was found with Trl , confirming the MEME-ChIP analysis ( Figure 5A ) . We further compared our list of Atro peaks with the published Trl ChIP-seq data ( Fuda et al . , 2015 ) and found striking overlap between the two data sets ( 1123/1377 Atro peaks overlap with Trl ChIP , Figure 5C ) . Visual inspection of Atro and Trl ChIP-seq peaks also shows binding at the same genomic regions ( Figure 6C , E ) . 10 . 7554/eLife . 23084 . 014Figure 5 . Atro and Trl bind to the same genomic locations . ( A ) The top four hits from MEME-ChIP analysis using all Atro peaks . The top hit is a ( GA ) repeat , which is the binding motif of Trl . ( B ) Atro ChIP-seq data overlap with other factors in S2 cells . The y-axis represents the overlap over that expected by chance , where the fraction of the genome covered by each factor is the overlap expected by chance . Only positive values are shown . The Atro values are selected with increasing cut-off , so that fewer but stronger Atro-binding sites are shown along the x-axis . ( C ) shows the Venn diagram of the overlap of all Atro and Trl ChIP data . All 1377 Atro ChIP peaks were used and Trl ChIP peaks are from Fuda et al . ( 2015 ) . Genomic coordinates from both data sets were used to construct this diagram . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 01410 . 7554/eLife . 23084 . 015Figure 5—source data 1 . List of Atro peaks’ average –log ( pvalue ) and fold enrichment generated from MACS2 for Figure 5—figure supplement 2 . Atro peaks that intersect Trl were compared with peaks that do not intersect Trl . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 01510 . 7554/eLife . 23084 . 016Figure 5—figure supplement 1 . Additional MEME ChIP hits of Atro ChIP-seq . These MEME ChIP results are the 5th-12th top hits of the MEME ChIP experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 01610 . 7554/eLife . 23084 . 017Figure 5—figure supplement 2 . Comparison between Atro peaks that overlap Trl and those that do not . ( A ) Box and whisker plot of Atro peaks based on MACS2 derived –log ( pvalue ) , which is an indication of the significance of called peaks . ( B ) Box and whisker plot of Atro peaks based on MACS2-derived fold enrichment values for each called peaks . Both A and B shows Atro peaks that intersect with Trl peaks are more significant ( higher –log ( pvalue ) ) and have more fold enrichment than Atro peaks that do not intersect Trl ( Wilcoxon test p-value<0 . 001 for both plots ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 01710 . 7554/eLife . 23084 . 018Figure 6 . Trl and Atro bind to the same loci simultaneously and Trl is required for the expression of en and tkv . ( A ) There is an Atro peak that overlaps with Trl peak within the sbb locus ( left peak , Atro and Trl Peak in B ) . There is an Atro peak that does not overlap with Trl about 20 kb upstream of the sbb locus ( right peak , Atro Only Peak in B ) . The red rectangle marks the region used as negative control in B . ( B ) ChIP-re-ChIP qPCR results . ChIP-re-ChIP samples are labeled as the sequence of antibody used ( e . g . IgG , Atro means rabbit IgG ChIP followed by Atro re-ChIP ) . Negative controls for the ChIP-re-ChIP are any ChIPs with IgG . Atro , Trl ChIP-re-ChIP enriches the Atro and Trl Peak ( purple bar ) but it does not enrich the Atro only peak ( same enrichment as Atro , IgG ) . Mean Ct value was used to calculate percent input and standard deviation of the Ct values was carried over in calculations and used as error bars . ( C ) Trl and Atro ChIP peaks coincide at the same loci upstream of en . ( D ) TrlR85 imaginal disc clones have decreased En levels ( arrow , wing disc clone shown here ) . ( E ) Trl and Atro ChIP peaks coincide at the tkv locus . ( F ) TrlR85 wing disc clones have decreased pMad levels ( arrow ) , possibly due to a decrease of tkv expression . All clones are marked by the absence of GFP; all figures have posterior to the left , dorsal up . Scale bars are 50 μm . ( G ) Model of how Atro and Trl function together to regulate expression of target genes . Trl is required to activate the transcription of its target genes . In the absence of Atro ( Left ) , the target gene will express at a higher level than normal . Atro binds to the same site as Trl either directly or via some unknown cofactors ( X ? , Right ) . Atro modulates the expression of its target gene by counteracting Trl; potentially Atro is doing so by recruiting Histone deacetylase 1 ( HDAC1 ) and G9a , a histone methyl transferase . Thus , target genes are expressed at the correct levels . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 01810 . 7554/eLife . 23084 . 019Figure 6—source data 1 . Source data for qPCR %Input calculations for Figure 6B and Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 01910 . 7554/eLife . 23084 . 020Figure 6—figure supplement 1 . Additional ChIP-re-ChIP qPCR results of the Trl and no ocelli ( noc ) loci . ( A ) Trl Negative and noc Negative are negative control loci for both genes . ChIP-re-ChIP samples are labeled in the sequence of antibody used ( e . g . IgG , Atro means rabbit IgG ChIP followed by Atro re-ChIP ) . Negative controls for the ChIP-re-ChIP are any ChIPs with IgG . Atro and Trl ChIP-re-ChIP enriches the Trl and noc peaks ( purple bar ) . Mean Ct values were used to calculate percent input and standard deviation was carried over in calculations to be used as error bars . ( B ) Atro ( green ) and Trl ( black ) ChIP peaks overlap at the same positions at the Trl and noc loci . The red rectangle in the Trl loci figure shows where the Trl primer set amplifies for the qPCR in A . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 02010 . 7554/eLife . 23084 . 021Figure 6—figure supplement 2 . Trl knockdown decreases Atro protein levels , coimmunoprecipitation of Trl and Atro , and pausing indices of Trl and Atro bound genes . ( A ) Trl knockdown by RNAi causes Atro protein levels to decrease ( arrow ) , while no RNAi or GFP RNAi treatments do not affect Atro protein levels . Lamin was used as a loading control . ( B ) shows the coimmunoprecipitation of Atro and Trl . Each lane is labeled with the antibody used for each IP . Blot was stained with Trl antibody . The red arrows mark the Trl bands . Trl is coimmunoprecipitated with Atro but not with IgG control . ( C ) The pausing index ( amount of Pol II in the promoter-proximal region versus the gene body calculated from PRO-seq data ( Kwak et al . , 2013 ) is higher on average for genes bound by Atro and Trl than for all other expressed genes in S2 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 02110 . 7554/eLife . 23084 . 022Figure 6—figure supplement 3 . TrlR85 clones cause an autonomous decrease Mad phosphorylation but do not affect pMad levels outside clone ( arrow ) . Clones are marked by the absence of GFP . Posterior is to the left , dorsal upwards . Scale bar is 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 23084 . 022 The above analyses suggest that Atro and Trl bind to the same sites and may form a protein complex . To determine if Atro and Trl form a protein complex , we performed coimmunoprecipitation ( coIP ) of Atro and Trl in S2 cells . Although we did not detect an interaction using standard coIP conditions , pretreating lysates with micrococcal nuclease allowed us to coimmunoprecipitate endogenous Trl with endogenous Atro ( Figure 5D ) . To directly test if Atro and Trl bind to the same loci simultaneously , we performed sequential ChIP ( ChIP-re-ChIP ) on Atro followed by Trl in S2 cells . For this analysis , we selected Atro peaks at the scribbler ( sbb ) locus because of the presence of an Atro peak that overlaps with a Trl peak as well as a nearby Atro peak that does not overlap ( Figure 6A ) . qPCR show that Atro-Trl ChIP-re-ChIP enriches for the locus with both Atro and Trl peaks , while it does not enrich for a locus with only Atro peak ( enrichment is nearly equal to Atro-IgG ChIP-re-ChIP-negative control ) ( Figure 6B ) . Additional experiments showed that Atro and Trl co-occupy all the tested loci with both Atro and Trl peaks ( Figure 6—figure supplement 1 ) . To see if Atro requires Trl to associate with DNA , we knocked down Trl using RNAi in S2 cells . Unfortunately , we could not directly test if Atro association with DNA requires Trl in cells , as knock down of Trl also leads to reduced abundance of Atro proteins ( Figure 6—figure supplement 2A ) . These findings indicate that Atro and Trl bind to the same loci and to one another , forming a complex on chromatin to regulate transcription . Both ChIP-seq and ChIP-re-ChIP data show that Atro and Trl bind to the same loci simultaneously . Interestingly , there are strong overlaps of Atro and Trl ChIP-seq peaks at the en locus ( Figure 6C ) . Since we found that Atro negatively modulates the expression of En during larval development , we wondered if Trl also regulates En expression . Therefore , we generated and analyzed Trl null mutant clones ( TrlR85 ) in imaginal discs . In contrast to Atro35 clones , TrlR85 clones have reduced En expression ( Figure 6D ) , indicating that Trl is required for the positive regulation of En expression in larval discs . Atro and Trl ChIP-seq data also indicate that Atro and Trl bind to the same region of the tkv locus ( Figure 6E ) , suggesting that Trl regulates tkv expression with Atro . Therefore , TrlR85 clones were generated to investigate the role of Trl in tkv regulation . pMad levels were used as a read out for Tkv levels . In contrast to Atro35 clones , TrlR85 clones have reduced pMad levels ( Figure 6F , Figure 6—figure supplement 3 ) . In addition , pMad staining can be found on the side of the TrlR85 clones further away from the middle of the wing ( Figure 6—figure supplement 3 , arrow ) , indicating Dpp expression was not disrupted . These changes can be explained by a decrease in Tkv levels in TrlR85 clones . In addition , we did not find any Trl or Atro peaks within 20 kb of the mad locus ( data not shown ) . Thus , the simplest model to explain our data is that the reduced pMad in TrlR85 clones is caused by a downregulation of tkv expression . Therefore , similar to Atro and Trl’s regulation of en , Trl is required for tkv expression , while Atro is required for repression of tkv .
Mutations in Atrophins lead to neurodegeneration , and loss of Atro leads to major defects in development . However , only a direct few targets of Atro are known and it is unclear how Atro regulates many genes . Here , using genome-wide ChIP-seq and in vivo analyses we show that Atro directly binds a number of developmentally critical genes; specifically regulating En expression and Dpp and Notch signaling . Trl has been proposed to activate transcription by opening and maintaining open chromatin to allow additional transcription factors to bind ( reviewed in [Granok et al . , 1995; Lehmann , 2004 ) . We find that Atro and Trl participate in a complex and bind to the same loci simultaneously . Atro does not bind to DNA directly , while Trl has a DNA binding domain ( Biggin and Tjian , 1988 ) . Thus , our results suggest a model in which Atro binds to DNA via Trl , either directly or indirectly via someunknown cofactors , and Atro modulates transcriptional activation by Trl . Our in vivo analyses of en and tkv show that Trl is required to promote their expression while Atro negatively modulates expression levels . To explain these phenotypes , we propose a model in which Trl binding leads to open chromatin to allow transcription , and Atro binding , either by directly interacting with Trl or through some unknown factors , negatively modulates Trl’s activity ( Figure 6G ) . Atro binds to Histone deacetylase 1 ( HDAC1 ) and a histone methyltransferase ( G9a ) to close chromatin and repress transcription ( Wang et al . , 2006 , 2008 ) , suggesting a mechanisms by which Atro is able to counteract Trl . Indeed , Atro- and Trl-binding sites at the en , tkv and fng loci overlap with HDAC1-binding sites ( from modENCODE , RPD3-Q3451 ) ( Celniker et al . , 2009 ) . Thus , loss of Trl leads to loss of expression of en and tkv where they are normally expressed . In the absence of Atro , Trl’s activity is no longer negatively modulated , and expression of target genes such as en and tkv are increased . However , in regions where en and tkv are not endogenously expressed , loss of Atro cannot induce ectopic expression as Atro’s function is to negatively modulate the transcription of its target genes . Interestingly , Trl is enriched at paused promoters and is involved in transcription pausing ( Lee et al . , 2008; Fuda et al . , 2015 ) . Trl recruits the negative elongation factor NELF to paused genes ( Li et al . , 2013 ) , but the precise mechanism of how Trl affects pausing is unknown . Our data suggest that Atro is present to modulate Trl’s transcription activation role . Perhaps , Atro is part of the transcriptional machinery used by Trl to pause transcription . In support of this , we noted that Atro preferentially binds to genes with higher expression ( in S2 cells , genes with high expression are likely to be paused , Figure 1C ) ( Gilchrist et al . , 2010 ) . GO analysis of Atro targets shows enrichment of signaling and patterning genes , also consistent with a role in pausing ( Figure 1F , data not shown ) . Notably , the pausing index for genes bound by both Atro and Trl is higher on average than for all expressed genes ( Figure 6—figure supplement 2C , [Kwak et al . , 2013] ) . This increase of pausing index is even higher than genes bound only by Trl ( Figure 6—figure supplement 2B ) . These data suggest that Atro may be involved in transcriptional pausing with Trl . We note that there are some Atro-binding sites that are not co-occupied by Trl . Atro also binds to other transcription factors ( such as Tailless and Cubitus Interruptus [Wang et al . , 2006; Zhang et al . , 2013] ) and Atro may interact with other factors to bind to sites not co-occupied by Trl . Visual inspection reveals strong Atro peaks that do overlap with Trl ( Figure 6A ) ; however , on a genome-wide basis , Atro peaks that do not overlap with Trl peaks are slightly weaker than the ones that do , as judged from the MACS2 derived significance ( -log ( p-value ) ) and fold enrichment values ( Figure 5—figure supplement 2 ) . Atro maternal mutants have missing , malformed and/or expanded En stripes ( Zhang et al . , 2002 ) , while in discs Atro modulates En expression . A possible explanation is that En regulation changes at different developmental times . During early embryogenesis , Pair-rule genes are required to first establish en expression ( DiNardo et al . , 1988 ) . Atro is also required for the proper expression of Pair-rule genes such as fushi tarazu and even-skipped ( Erkner et al . , 2002; Zhang et al . , 2002 ) . This could explain why loss of Atro during early embryogenesis has more severe effects on en expression . However , from late embryogenesis on , En expression no longer requires Pair-rule genes but depends on Polycomb group genes and unidentified activators ( Moazed and O'Farrell , 1992 ) . We observed reduced En expression in Trl clones in antennal and wing imaginal discs , while Atro disc clones have increased expression . These results suggest that Trl and Atro are required for regulation of En expression from late embryogenesis through larval development . Why is it important for Atro to moderate en expression ? Intriguingly , expressing higher than normal levels of En in the posterior compartment leads to lethality and anterior-posterior patterning defects of the posterior wing , suggesting that moderating En levels is required for normal development , and high levels could be toxic ( Guillén et al . , 1995; Tabata et al . , 1995 ) . Our Atro ChIP data show that Atro binds to the tkv locus and our in situ analysis reveals Atro represses tkv expression . We find that Atro35 clones have increased Dpp signaling , consistent with Atro’s role in tkv regulation . Conversely , TrlR85 clones have lower pMad levels . We reason that this is caused by decreased Tkv within the TrlR85 clones , consistent with a peak in the tkv locus . These observations mirror what we have seen with en , where Trl is required for activation and Atro is required for repression of transcription . A model for Atro and Trl regulation of Dpp signaling is shown in Figure 3—figure supplement 1 . In wild-type wings , the expression patterns of Dpp ( yellow shading ) and Tkv ( red line ) cause pMad to be found in a broad stripe ( Figure 3—figure supplement 1A , blue line ) ( Tanimoto et al . , 2000 ) . Tkv levels are increased in Atro35 clones ( Figure 3—figure supplement 1B , indicated by the shaded rectangle ) . Thus , pMad levels are increased along the interior border of the clone if the Atro35 clones are close enough to the Dpp source . Atro35 clones further away would not cause changes to pMad levels . Trl has the opposite effect where TrlR85 clones ( Figure 3—figure supplement 1C , shaded rectangle ) cause a decrease of Tkv and thus lower pMad levels . Interestingly , the Mad binding motif is enriched in the Atro ChIP , with a shift in distribution , suggesting Mad may bind adjacent to Atro , and that Atro-Mad interactions on chromatin may also affect Dpp signaling . Although Atro35 clones can lead to ectopic dpp-LacZ expression ( Erkner et al . , 2002; Zhang et al . , 2013 ) , we did not see increased Dpp in Atro35 wing clones . Additionally , loss of Atro away from the endogenous Dpp stripe does not induce ectopic pMad . Therefore , the increase of pMad staining in Atro35 clones matches the pattern that is expected if Atro regulates Dpp signaling via tkv and not via dpp regulation . Thus , we suggest that Atro regulates Dpp signaling in the wing primarily by regulating tkv transcription . To our knowledge , this report also provides the first evidence that Atro regulates N signaling . Our ChIP-seq analysis revealed Atro binding in multiple N pathway components , and genetic and molecular analysis reveals disruption of N signaling in Atro clones in eye and wing discs . Atro knockdown results in upregulation of fng transcripts in larval wings . Interestingly , patched-Gal4-driven overexpression of fng leads to an autonomous loss of wing margin marker expression as well as ectopic expression of wing margin marker on the posterior border of the Fng overexpressing region ( Panin et al . , 1997 ) . All Atro35 clones that cross the wing margin cause an autonomous loss of wing margin markers and large Atro35 clones can induce some ectopic wing margin marker expression on the posterior edge of the clone , mimicking patched-Gal4-driven Fng overexpression . Atro35 clones also disrupt N signaling reporter expression . Atro35 clones cause the N signaling reporter to express diffusely instead of in a sharp line . This diffuse expression pattern may be an indication of a loss of precise N signaling at the wing margin . Interestingly , Fng is crucial for N signaling at the margin ( de Celis and Bray , 1997; Panin et al . , 1997 ) . Thus , Atro represses fng expression , and loss of Atro and patched-Gal4-driven overexpression of Fng have similar N-related phenotype in the wing . Loss of Atro also leads to Notch loss of function in the eye . Atro35 clones have extra cells with the early R8 marker , Sens , indicating a defect in lateral inhibition . Although there are extra Sens-positive cells in Atro clones , not all these cells express the late R8 marker , Boss , thus the extra Sens-positive cells do not differentiate into R8 . Inspection of the Sens-positive cell clusters reveals there is one cell with more Sens than its neighbors in each cluster . We reason that the difference in Sens levels renders one cell with the most R8-like and this cell can express Boss . Atro35 clones also exhibit a loss of R7 and cone cell markers . Thus , Atro binds to the putative regulatory regions of genes that are connected to N signaling ( such as emc , Dl , mam , fng , neur , numb , Supplementary file 1 ) , genetically interacts with N and regulates N target expression . While the strongest overlap in our ChIP-seq peaks was seen between Atro and Trl , we also observed significant overlap of our Atro ChIP-seq data with ChIP data sets of Yorkie ( Yki ) , the key transcriptional co-activator of the Hippo pathway ( Figure 5B ) . Interestingly , the atypical cadherin Fat regulates Yki activity via the Hippo pathway ( [Bennett and Harvey , 2006; Cho et al . , 2006; Silva et al . , 2006; Willecke et al . , 2006] and reviewed in Enderle and McNeill [2013] and planar polarity via Atro ( Fanto et al . , 2003; Saburi et al . , 2012; Sharma and McNeill , 2013 ) . Significantly , neurodegeneration by Atro has been shown to be mediated in part by Yki and the Hippo pathway ( Napoletano et al . , 2011 ) . Thererfore , our finding that Yki and Atro are found at the same loci suggests a direct mechanism by which Atro may impact neurodegeneration , and suggests that Atro interactions with Yki may feed back into growth and patterning regulation by Fat cadherins . To our knowledge , this is the first genome-wide analysis of Atrophin . Our genome-wide ChIP-seq and phenotypic analyses reveal many novel direct targets of Atro , and showed that Engrailed , Notch and Dpp signaling are directly regulated by Atro . Our analyses indicate that Atro preferentially binds to Trl binding sites . Interestingly , the fraction of paused genes is significantly more correlated with sites that bind both Atro and Trl , than just Trl alone , suggesting Atro may have a function in the regulation of pausing . Significantly , ChIP-re-ChIP experiments reveal that Atro and Trl bind to the same loci simultaneously , and phenotypic analyses indicate that Atro restricts expression of genes whose expression is promoted by Trl . Taken together , our data indicate that Atro is a critical component of developmental signaling and is an important general modulator of transcription activation by Trl .
Two step crosslinking ChIP was required in order for ChIP to enrich for positive controls when using the SG2524 anti-Atro antibody ( see below ) . S2 cells ( 107 cells/mL , S2-DGRC , stock #6 from Drosophila Genomics Research Center ) were washed three times in sterile PBS . Washed cells were first fixed with 2 mM ethylene glycol bis ( succinimidyl succinate ) ( EGS ) in PBS for 45 min at room temperature followed by three PBS washes . Washed cells were crosslinked in 1% formaldehyde in PBS for 15 min at room temperature and quenched in 125 mM glycine for 5 min on ice . Cells were then washed once in ChIP Wash Buffer A ( 10 mM Hepes pH7 . 6 , 10 mM EDTA pH8 . 0 , 0 . 5 mM EGTA pH8 . 0 , 0 . 25% Triton X-100 ) and followed by a wash in ChIP Wash Buffer B ( 10 mM Hepes pH7 . 6 , 100 mM NaCl , 1 mM EDTA pH8 . 0 , 0 . 5 mM EGTA pH8 . 0 , 0 . 01% Triton X-100 ) at 4°C , 5 min each . Washed cells were resuspended in Sonication buffer ( 50 mM Hepes pH7 . 6 , 140 mM NaCl , 1 mM EDTA pH8 . 0 , 1% Triton X-100 , 0 . 1% sodium deoxycholate , 0 . 1% SDS , supplemented with proteinase inhibitors ) at a concentration of 108 cells/mL . Cells were then sonicated using a Qsonica Q700 sonicator in an ice-water bath ( until fragments were roughly 150 bp in length ) . 10 μL of 10% SDS , 100 μL of 1% sodium deoxycholate , 100 μL of 10% Triton X-100 , 28 μL of 5M NaCl were added to each mL of sonicated chromatin and incubated at 4°C for 10 min . Sonicated chromatin was then spun down at ≥20 , 000g for 5 min to remove cellular debris and the supernatant was used for ChIP . Protein G Dynabeads ( Invitrogen , Lithuania ) were blocked in 1 mg/mL BSA in sonication buffer for at least 2 hr at 4°C . Blocked beads were conjugated with the antibodies for at least 4 hr at 4°C . 5 μL anti-Atro sera ( SG2524 , raised in rabbits against Atro amino acids 121–134 , KGIDKKWTEDETKK ) , and 5 μL normal rabbit IgG ( Cell Signaling Technology , Danvers , MA ) were used for ChIP . Of the sonicated chromatin , 300 μL were incubated with conjugated beads overnight on a rotating wheel at 4°C . Beads were washed for 5 min each in Sonication buffer , Wash A ( 50 mM Hepes pH7 . 6 , 500 mM NaCl , 1 mM EDTA pH8 . 0 , 1% Triton X-100 , 0 . 1% sodium deoxycholate , 0 . 1% SDS , supplemented with proteinase inhibitors ) , Wash B ( 20 mM Tris pH8 . 0 , 1 mM EDTA pH8 . 0 , 250 mM LiCl , 0 . 5% NP-40 , 0 . 5% sodium deoxycholate ) , and TE buffer . Beads were then resuspended in Elution Buffer ( 50 mM Tris pH8 . 0 , 50 mM NaCl , 2 mM EDTA , 0 . 75% SDS , 20 ug/mL RNase A ) and incubated at 68°C overnight to remove crosslinks . Eluted chromatin was extracted by treating with Proteinase K followed by phenol chloroform DNA extraction ( with 6 μg glycogen added during DNA precipitation [Thermo Scientific , Lithuania] ) . The extracted DNA was resuspended in 50 μL Tris pH8 . 0 . Three biological replicates were used for library construction ( along with three corresponding IgG ChIP control ) . ChIP samples were treated with polynucleotide kinase and DNA polymerases for 30 min at room temperature ( 35 μL ChIP sample , 5 μL 10xNEB two buffer ( New Englands Biolabs , Ipswich , MA ) , 2 μL of 25 mM ATP , 2 μL of 10 mM dNTP , 10U T4 polynucleotide kinase , 4 . 5U T4 DNA polymerase , 1U Klenow Large Fragment DNA Polymerase , water to a volume of 50 uL ) . Afterwards , DNA was purified with PEG bead slurry ( 1M NaCl , 23% PEG , Sera-Mag Speedbeads ( Fisher Scientific , England ) with final PEG concentration of 13 . 87% ) and eluted with 35 uL Qiagen EB buffer ( Qiagen , Valencia , CA ) . Single dA overhang was added to eluted ChIP samples by incubating samples in 35 μL ChIP samples , 5 μL 10X NEB two buffer , 1 μL 10 mM dATP , 5U Klenow Fragment ( 3’−5’ exo- ) ( New England Biolab ) , water to 50 μL for 30 min at 37°C . Samples were purified with PEG bead slurry and eluted with 35 μL Qiagen EB buffer . Short adapators for sequencing were ligated to samples by incubating samples at room temperature overnight in ligation buffer ( 35 μL DNA , 12 μL 5X quick ligation buffer , 2000U quick T4 DNA ligase , 2 μL 0 . 5 μM Illumina short sequencing adaptor , water to 60 μL ) . ChIP samples were purified two times with PEG bead slurry ( started with 20% PEG instead of 23% PEG; final PEG of 8 . 89% and 10 . 91% , respectively ) and eluted with 35 μL Qiagen EB buffer . Adaptor ligated libraries were PCR amplified ( 10 cycles ) , purified with PEG bead slurry ( started with 20% PEG instead of 23%; final PEG concentration of 9 . 19% ) , and eluted with 35 μL Qiagen EB buffer . Libraries were then sequenced on Miseq ( Illumina , San Diego , CA ) using PE150V3 kit PE 85 bp . BWA program ( v0 . 6 . 1; using default parameters ) was used to align sequence reads to Drosophila genome release five ( Attrill et al . , 2016 ) . A second , independent ChIP was performed with a monoclonal antibody ( 4H6 ) raised against Atro amino acids 1369–1378 ( SRQSPLHPVP ) using Drosophila S2 cells catalog #006 from the Drosophila Genomics Resource Center . This ChIP was performed using two biological replicates ( cells grown and ChIP’ed at different times ) . The cells were grown to a density of 0 . 2–1 × 107 cells/mL and fixed in 1% formaldehyde for 15 min at ambient temperature . The reaction was quenched by 0 . 16 M glycine pH 7 . 0 for 5 min and washed in PBS . Cells were sequentially washed with ChIP Wash buffer A and ChIP Wash Buffer B for 10 min at 4°C followed by resuspension in Sonication buffer to a final concentration of 5–10 × 107 cells/mL . Nuclei were sonicated for 15 min using a Diagenode Bioruptor , rotated for 10 min followed by centrifugation for 10 min at 13 , 000 rpm at 4°C . A mix of Protein A and G Dynabeads ( Invitrogen , Lithuania ) blocked with BSA ( Sigma Aldrich , St Louis , MO ) were mixed with the antibody . Beads and antibodies were incubated for at least 2 hr followed by the addition of 0 . 5–1 × 107 cells . Chromatin and antibody bead complexes were formed during at least 2 hr followed by 5 min washes with Sonication buffer , Wash A , Wash B and TE buffer . Beads were resuspended in Elution buffer ( same as above but supplemented with 20 µg/mL glycogen ) in a new tube . Cross-linking was reversed at 68°C for at least 4 hr and proteins removed by Proteinase K digestion . DNA was purified by phenol-chloroform extraction , ethanol precipitated and finally resuspended in 200 µl 0 . 1×TE . The DNA was sequenced at the Uppsala Genome Center using SOLiD ( TM ) ChIP-Seq Library preparation , size selection ( around 150 bp + adapters 95 bp ) and sequenced using SOLiD4 75 bp fragment run . The number of mapped reads were 11270731 ( Input 1 ) , 13320338 ( Atro ChIP 1 ) , 7315911 ( Input 2 ) and 6972016 ( Atro ChIP 2 ) . S2 cells were double crosslinked , sonicated , and ChIP’ed as above , using 5 μL anti-Atro sera ( SG2524 ) , and 10 μL normal rabbit IgG ( Cell Signaling Technology ) . After the first ChIP , beads were washed and eluted for re-ChIP as described in Truax and Greer ( 2012 ) . Eluted chromatin was incubated in BSA-blocked Dynabeads for 2 hr to remove any leftover antibodies . The supernatant containing eluted chromatin was re-ChIP’ed by incubating in beads conjugated with the appropriate amount of antibodies overnight ( 5 μL anti-Atro sera ( SG2524 ) , 10 μL anti-Trl sera ( gift from K . White ) , 10 μL normal rabbit IgG ( Cell Signaling Technology ) ) . After the re-ChIP , the beads were washed and eluted as normal double crosslink ChIP samples above . qPCRs were done in technical triplicates with SYBR green PCR Master Mix ( Applied Biosystems , Canada , 20 μL reaction volume; qPCR was performed three times for each ChIP samples ) . Percent input and errors were calculated using standard percent input calculations . qPCR primer pairs are listed in Supplementary file 2 . For each biological replicate , peaks were called using MACS2 ( version 2 . 1 . 0 , FDR = 0 . 01 , genome size 1 . 2 × 108 ) . Each Atro ChIP-seq replicate was compared with its corresponding IgG ChIP-seq replicate control ( or input control for the second independent Atro ChIP-seq replicates ) in the MACS analysis . Peaks from the biological replicates were intersected and only peaks present across all replicates with summits within 100 bp were selected . Peaks were extended by 2 kb on both sides and then annotated by intersecting all Drosophila genes’ coordinates with the peaks coordinates using BEDTools ( Quinlan and Hall , 2010 ) . MEME-ChIP was carried out using first-order model , any number of repetitions , motif count of 10 , score ≥5 and an E-value threshold of ≤10 . Overlap of Atro ChIP with Trl ChIP data sets ( Figure 5C ) was done using BEDTools Intersect function . Atro-binding sites were compared to data from modENCODE ( downloaded from http://intermine . modencode . org/ ) , ( Philip et al . , 2015 ) for CBP , and ( Oh et al . , 2013 ) for Yki , and gene expression divided into three equally sized bins ( low , medium and high expressions ) from ( Cherbas et al . , 2011 ) using custom Perl scripts . The enrichment values for each factor in the Atro binding regions were calculated by taking the mean of the top three consecutive enrichment values within each region ( Philip et al . , 2015 ) . All factors were normalized so that 0 represents the genomic mean ( background levels ) and one represents the genomic maximum ( mean of top 0 . 001 percentile ) for each factor . Enriched Atro regions were used as observations and normalized enrichment values of each factor within the regions were used as variables as decribed in ( Philip et al . , 2015 ) . Hierarchical clustering was done on all significant components of the analysis using Ward clustering to calculate tree distances . The three classes of Atro-bound regions were based on hierachical clustering . Mitotic clones were generated in flies with the following genotypes: hs-flp; ; Atro35 FRT80B / ubi-GFP FRT80B hs-flp; ; TrlR85 FRT2A / hs-GFP , Minute , FRT2A hs-flp; NRE-GFP / +; Atro35 FRT80B / arm-LacZ FRT80B Mitotic clones were generated by heat shocking ( at 37°C ) larvae for 45 min at 48 hr and 72 hr after egg laying . TrlR85 clones were additionally heat shocked for 45 min about 1 . 5 hr prior to dissection to induce GFP expression . Larval tissues were dissected and prepared as in standard protocol . Sample sizes are listed in Supplementary file 3 . The following antibodies were used for immunofluorescence: mouse antibodies against En ( DSHB 4D9 , 1/400 ) , Pros ( DSHB MR1A , 1/500 ) , Ct ( DSHB 2B10 , 1/500 ) , Wg ( DSHB 4D4 , 1/500 ) , and ß-Gal ( Promega [Madison , WI] , 1/1000 ) ; rabbit antibodies against En ( Santa Cruz ( Santa Cruz , CA ) d-300 , 1/500 ) , Omb ( gift from G . Pflugfelder , 1/1000 ) , Dpp ( gift from M . Gibson , 1/100 ) , Ttk ( gift from W . Ge , 1/100 ) , and pMad ( Cell Signaling Technology ( Danvers , MA ) #9510 , 1/500 ) ; guinea pig antibodies against Sens ( gift from H . Bellen , 1/1000 ) , and Runt ( gift from C . Desplan , 1/500 ) . enGal4 / +; UAS Atro RNAi ( Bloomington stock center , line 32961 ) / + larvae were used for in situ hybridization . Larval wing discs are prepared and stained as described in Morris et al . ( 2009 ) , substituting wing discs for testes . Primers for probes are listed in Supplementary file 2 . S2 cells were purchased from Drosophila Genomics Resource Center ( S2-DGRC , stock #6 ) . S2 cells were grown in S2 media + 10% fetal bovine serum in standard conditions , and cells were kept healthy for all experiments . Double stranded RNA was made using the Megascript T7 kit ( Life Technologies , Canada ) . Primers used to make the dsRNA are listed in Supplementary file 2 . S2 cells were dsRNA treatment using standard bathing S2 knockdown protocol . 10 ug/mL of dsRNA was used for each treatment . S2 cells ( 107 cells per immunoprecipitation ) were incubated in 0 . 5 mL LPC buffer ( 5% sucrose , 35 mM Hepes pH7 . 4 , 80 mM KCl , 5 mM K2HPO4 , 5 mM MgCl2 , 5 mM CaCl2 , 0 . 01% α-lysophospatidylcholine ) for 3 min at room temperature . Cells were resuspended in 0 . 2 mL MNase buffer ( 5% Glycerol , 20 mM Tris pH7 . 4 , 60 mM KCl , 15 mM NaCl , 5 mM CaCl2 , 3 mM MgCl2 , 0 . 5% NP-40 , 1 mM DTT ) . 6 μL of micrococcal nuclease ( 40 U/μL , New England Biolabs ) were added and incubated at 25°C for 5 min . 0 . 3 mL of MNase dilution buffer ( 3 . 6 mM Tris pH8 . 8 , 12 mM EDTA , 225 mM NaCl , 60 mM KCl , 1 . 2% NP-40 , supplemented with proteinase inhibitor ) were added and incubated on shaker for 10 min at 4°C . MNase treated samples were centrifuged to remove cellular debris . Supernatant were used for immunoprecipitation using standard methods . ChIP-Sequencing data reported in this study are archived at the Gene Expression Omnibus ( https://www . ncbi . nlm . nih . gov/geo/ ) as accession numbers GSE87509 ( ChIP using Atro antibody SG2524 ) and GSE87471 ( second independent Atro ChIP with antibody 4H6 ) . | Cells with the same genetic information can look and behave differently to each other . This is because they can control the activity of their genes , changing the effects the genes have in the cell . Regulating genes in this way is important in allowing cells to adapt to their surroundings and to perform different tasks . Proteins called transcription factors control the activity of genes through other proteins called transcriptional co-activators and co-repressors . Atrophins are a group of co-repressors found in many animals including humans and fruit flies . Atrophins suppress the activity of certain genes , reducing the effects that they have in the cell . Losing Atrophin from cells can lead to severe diseases , but how Atrophin causes these effects is currently not well understood . Yeung et al . examined which genes Atrophin regulates in cells from the fruit fly Drosophila melanogaster . This investigation revealed that , amongst other genes , Atrophin controls several well-studied genes including engrailed and thickveins . These genes are important in allowing cells to communicate and co-ordinate before birth , ensuring cells work together to build complex tissues and organs . These results suggest Atrophin plays key roles in organising and shaping the body before birth . Further examination revealed that Atrophin acts in partnership with another molecule called Trithorax-like . Inside the cell many genes are protected by structures called nucleosomes that make them difficult to access , and Trithorax-like helps Atrophin to gain access to these genes . Further work will examine whether Atrophin and Trithorax-like work directly together or if other molecules bring about their interaction . It will also be important to examine how Atrophins suppress the activity of the genes they control . Errors in Atrophin1 in humans result in a nerve-damaging disease known as DRPLA; this work could also help researchers to better understand this disorder . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"developmental",
"biology"
] | 2017 | Atrophin controls developmental signaling pathways via interactions with Trithorax-like |
Telomerase synthesizes chromosome-capping telomeric repeats using an active site in telomerase reverse transcriptase ( TERT ) and an integral RNA subunit template . The fundamental question of whether human telomerase catalytic activity requires cooperation across two TERT subunits remains under debate . In this study , we describe new approaches of subunit labeling for single-molecule imaging , applied to determine the TERT content of complexes assembled in cells or cell extract . Surprisingly , telomerase reconstitutions yielded heterogeneous DNA-bound TERT monomer and dimer complexes in relative amounts that varied with assembly and purification method . Among the complexes , cellular holoenzyme and minimal recombinant enzyme monomeric for TERT had catalytic activity . Dimerization was suppressed by removing a TERT domain linker with atypical sequence bias , which did not inhibit cellular or minimal enzyme assembly or activity . Overall , this work defines human telomerase DNA binding and synthesis properties at single-molecule level and establishes conserved telomerase subunit architecture from single-celled organisms to humans .
Threats to genomic integrity occur at the ends of every linear chromosome , including incomplete DNA synthesis by the replisome and the potential for inappropriate DNA break repair . Eukaryotic cells control these reactions through the function of telomeres , typically consisting of telomeric repeat DNA bound by proteins that comprise the telosome , or in mammalian cells , shelterin ( Palm and de Lange , 2008; Stewart et al . , 2012 ) . Telomeric repeat tract maintenance depends on the specialized reverse transcriptase telomerase , which can extend a chromosome 3′ end by processive addition of single-stranded repeats ( Blackburn et al . , 2006 ) . Insufficient telomere synthesis ultimately compromises telomere function and signals a halt to cell proliferation ( O'Sullivan and Karlseder , 2010; Aubert , 2014 ) . This telomere-linked restriction of cellular renewal leads to failures of highly proliferative human tissues , with clinical manifestations including bone marrow failure , aplastic anemia , and pulmonary fibrosis ( Armanios and Blackburn , 2012 ) . The active human telomerase ribonucleoprotein ( RNP ) includes telomerase reverse transcriptase ( TERT ) , which provides the active site , and an RNA ( hTR ) containing a reiteratively copied internal template . The unique repeat addition processivity of telomerase requires conserved domains in both TERT and hTR that distinguish telomerases from other polymerase families ( Blackburn and Collins , 2011; Podlevsky and Chen , 2012 ) . The TERT N-terminal ( TEN ) domain allows retention of single-stranded DNA during the template repositioning required for tandem repeat synthesis . TEN-domain-truncated TERT , designated ‘TERT ring’ based on Tribolium TERT structure ( Gillis et al . , 2008 ) , supports only single-repeat synthesis that can be complemented to high repeat addition processivity by the TEN domain as a separate polypeptide ( Robart and Collins , 2011; Wu and Collins , 2014a ) . In addition to these and other catalytic activity requirements for TERT and hTR , a biologically functional human telomerase holoenzyme contains two sets of H/ACA proteins ( dyskerin , NHP2 , NOP10 , and GAR1 ) bound to hTR to direct RNP biogenesis and TCAB1 to redistribute the RNP from nucleoli to Cajal bodies ( Egan and Collins , 2012a; Podlevsky and Chen , 2012; Schmidt and Cech , 2015 ) . Telomerase holoenzyme must also assemble with the shelterin protein TPP1 for telomere recruitment and extension of chromosome ends ( Lue et al . , 2013; Nandakumar and Cech , 2013; Sexton et al . , 2014 ) . Endogenous human telomerase is scarce , with the number of TERT-hTR complexes per cell estimated as only ∼35 ( Cohen et al . , 2007 ) or ∼250 ( Xi and Cech , 2014 ) in even the most highly telomerase-positive tumor cell lines . Consequently , biochemical investigations of human telomerase have been greatly facilitated by enzyme reconstitution . Enzyme reconstitution in cells exploits transiently introduced plasmids to overexpress TERT and the 451-nucleotide mature hTR , which must be 3′-processed from an appropriate precursor ( Mitchell et al . , 1999; Fu and Collins , 2003 ) . Telomerase complexes reconstituted in cells have a diversity of substoichiometric-associated factors ( Egan and Collins , 2012a; Nandakumar and Cech , 2013; Schmidt and Cech , 2015 ) . As an alternative reconstitution approach , a minimal-subunit catalytically active RNP can be assembled by expressing TERT in rabbit reticulocyte lysate ( RRL ) with in vitro transcribed full-length hTR ( Weinrich et al . , 1997 ) or a half-sized RNA such as hTRmin used here ( Wu and Collins , 2014a ) , which lacks the two-hairpin H/ACA motif that assembles the holoenzyme subunits dyskerin , NHP2 , NOP10 , GAR1 , and TCAB1 . Only two hTR domains are critical for telomerase catalytic activity: a domain containing the template and adjacent pseudoknot and a branched stem-junction domain containing stem-loop P6 . 1 ( Mitchell and Collins , 2000; Chen et al . , 2002 ) . Importantly , human telomerase enzymes reconstituted in cells or in RRL can interact with the same length of single-stranded DNA , have similar specific activity , and have only minor differences in other enzyme properties such as repeat addition processivity ( Jurczyluk et al . , 2010; Zaug et al . , 2013; Wu and Collins , 2014a ) . Central to defining telomerase RNP architecture is a delineation of the number of TERT and hTR subunits that assemble together to generate an enzyme active site . RNP affinity purification and structural studies indicate a single RNA and single TERT per biologically functional telomerase holoenzyme of single-celled eukaryotes ( Livengood et al . , 2002; Witkin and Collins , 2004; Cunningham and Collins , 2005; Hong et al . , 2013; Jiang et al . , 2013; Bajon et al . , 2015 ) . This subunit stoichiometry is recapitulated by the minimal Tetrahymena telomerase RNP assembled in RRL ( Bryan et al . , 2003 ) . However , the subunit stoichiometry of an active human telomerase RNP is unresolved: some assays suggest TERT and hTR function as monomeric subunits , without dominant-negative inhibition of a wild-type ( WT ) subunit by co-expressed mutant subunit ( Errington et al . , 2008; Egan and Collins , 2010 ) , while other assays suggest obligate co-dependence of active site function across TERT and hTR subunits ( Wenz et al . , 2001; Sauerwald et al . , 2013 ) . Size fractionation of human telomerase holoenzyme has been suggested to establish TERT dimerization based on molecular mass by gel filtration of ∼600 kDa ( Wenz et al . , 2001 ) or by glycerol gradient sedimentation of 550 kDa ( Schnapp et al . , 1998 ) or 670 kDa ( Cohen et al . , 2007 ) relative to protein standards , but similar fractionation would be predicted for a holoenzyme with a single TERT , single hTR , single TCAB1 , and a complex of dyskerin , NHP2 , NOP10 , and GAR1 bound to each of two H/ACA-motif hairpin stems ( Egan and Collins , 2012a ) . Analysis using single-molecule fluorescence correlation spectroscopy detected one TERT and one hTR per RRL-reconstituted minimal RNP ( Alves et al . , 2008 ) . On the other hand , cellular subunit overexpression , purification , and crosslinking yielded particles observed by electron microscopy that were proposed to be active dimeric TERT RNPs , based on detection of two bound single-stranded DNAs ( Sauerwald et al . , 2013 ) . Unfortunately , all of the experiments above suffer from the caveat that individual complexes are inferred to have the activity measured only for a bulk population . Single-molecule fluorescence microscopy can detect the number of subunits in individual macromolecular complexes . We therefore developed a single-molecule TERT-labeling strategy to determine the TERT subunit content of human telomerase RNPs assembled and purified using methods typical in previous studies . We exploited the preserved function of N-terminally tagged human TERT to introduce the acyl carrier protein ( ACP ) tag for covalent labeling by prosthetic group transfer from derivatives of Coenzyme A ( CoA ) . ACP and ACP-based tags are well suited to the applications developed here because they are small , monomeric , and expose the conjugated prosthetic group as a conformationally dynamic extension from the protein surface ( Byers and Gong , 2007; Chan and Vogel , 2010 ) . We applied previously developed tag labeling methods ( Yin et al . , 2006; Zhou et al . , 2007 ) to investigate the TERT content of individual complexes from purifications of cellular telomerase holoenzyme reconstituted by assembly in human 293T cells and minimal recombinant RNP reconstituted by assembly in RRL . Surprisingly , different affinity purifications yielded different mixtures of complexes monomeric or variously multimeric for TERT . TERT complexes were also heterogeneous in catalytic activity and DNA-binding properties . Complexes with TERT monomer supported DNA synthesis . Apparently non-productive TERT self-association occurred through a low-complexity region of the protein dispensable for RNP catalytic activity . Overall , these studies support the function of human telomerase holoenzyme and minimal recombinant RNPs with a single subunit of TERT and demonstrate an evolutionarily conserved telomerase subunit architecture .
To quantify the TERT subunit content of reconstituted human telomerase complexes , we developed a strategy to label individual TERT molecules with a Cy3 or Cy5 fluorophore . The 8 kDa ACP and MCP tags derive from bacterial proteins that accept covalent transfer of the CoA phosphopantetheinyl ( Ppant ) group to a serine on the protein surface ( Figure 1A , left ) . In endogenous bacterial context , the Ppant group serves as a 20 Å swing-arm tether for subsequent transient attachment of the acyl groups that are the carrier proteins' cargo . For labeling ACP/MCP in vitro , the Ppant group of CoA can be pre-conjugated to diverse labels including Cy3 , Cy5 , or biotin prior to the prosthetic group transfer reaction , such that the Ppant swing-arm becomes a spacer between the label and the protein ( Belshaw et al . , 1999; George et al . , 2004; Yin et al . , 2006 ) . ACP synthase catalyzes transfer from a derivatized CoA to the ACP tag but does not label the MCP tag , while either tag is labeled by SFP synthase ( Zhou et al . , 2007 ) . Labeling of a tagged fusion protein by SFP synthase in vitro occurred with >80% efficiency ( Yin et al . , 2005 ) . 10 . 7554/eLife . 08363 . 003Figure 1 . Reconstitution , purification , and labeling of human TERT . ( A ) Left: derivatized CoA Ppant prosthetic group transfer to acyl carrier protein ( ACP ) or MCP tag by ACP or SFP synthase . The MCP tag is a modified version of the ACP tag , containing two amino acid substitutions , D36T and D39G . CoA can be modified with dye or biotin groups ( R ) for enzymatic labeling of a fusion protein . Right: schematic of two ACP- and/or MCP-telomerase reverse transcriptase ( TERT ) -labeling strategies using Cy5 ( red ) and Cy3 ( green ) . An ACP or MCP tag is N-terminal to the TERT TEN domain , which is connected to the TERT ring by a linker region ( L ) . Numbering refers to the full-length TERT amino acid sequence . A 3xFLAG tag is N-terminal to the ACP or MCP tag . ( B ) Schematic of telomerase holoenzyme reconstitution by overexpression of TERT with full-length hTR in cells ( 293T ) or minimal ribonucleoprotein ( RNP ) reconstitution by TERT expression with hTRmin in vitro ( rabbit reticulocyte lysate [RRL] ) followed by FLAG antibody purification for the TEN tag ( FLAG antibody purification , F ) or purification using a 2′OMe RNA oligonucleotide complementary to the hTR template ( Template oligo purification , O ) . Only the template of hTR or hTRmin is illustrated ( blue ) . ( C ) TERT and telomerase activity measured for O-purified , eluted complexes . Various N-terminally tagged TERT proteins were detected by TERT antibody immunoblot . The hTR ∆temp reconstitutions used template-less hTR or hTRmin with a 5′ end at hTR position 64 . Elution fractions were assayed for telomerase activity by primer extension with dTTP , ddATP , and α-32P dGTP , followed by denaturing gel electrophoresis . End-radiolabeled oligonucleotide was added prior to product precipitation to serve as a recovery control ( RC ) , here and in subsequent panels . End-radiolabeled primer is a size marker ( ▶ ) , here and in subsequent panels . Specific activity in this panel indicates product DNA normalized to amount of TERT . ( D ) SDS-PAGE analysis of RRL-expressed TERT in telomerase reconstitutions of ACP- , MCP- , or only F-TERT in the presence of hTRmin , labeled with 35S-methionine and any additional label as indicated . ACP synthase was used for ACP-TERT dye labeling and SFP synthase was used for MCP-TERT dye labeling . ( E ) Activity of telomerase reconstituted with ACP- , MCP- , or F-TERT in RRL with hTRmin and labeled as indicated . Activity was detected in reactions containing dATP , dGTP , dTTP , and α-32P dGTP , followed by denaturing gel electrophoresis . ( F ) TERT content and telomerase activity in bulk purifications of MCP-TERT reconstituted in 293T cells or RRL , assayed as described in ( C ) . TERT immunoblot with input extracts used 3% of the total purification input . Half of the post-purification sample was used for activity assays and half for TERT immunoblot . For single-molecule detection , O-purifications were diluted relative to F-purifications from the same extract . The following figure supplement is available for Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08363 . 00310 . 7554/eLife . 08363 . 004Figure 1—figure supplement 1 . Methods of human telomerase reconstitution and purification . ( A ) Telomerase reconstitutions in cells with full-length hTR or in RRL with hTRmin generate hTR-free TERT , TERT RNPs , and TERT-free hTR . These are differentially enriched by TERT-based vs template-based affinity purification . The template region is boxed as an overlay on RNA secondary structure overall ( position numbering is from full-length hTR ) . Reconstituted complexes are schematized with only part of the RNA , and proteins other than TERT are not depicted . Telomerase RNP is shown with template in the active site . Relative elution of TERT RNP vs TERT-free hTR or hTR-free TERT was not determined . ( B ) Sequence of the template affinity oligonucleotide and displacement oligonucleotide used for RNA-based purification . Only the template/pseudoknot ( t/PK ) domain of hTR is illustrated; the template region primary sequence is shown maximally base-paired to the template affinity oligonucleotide . ( C ) Activity of RRL-reconstituted , F-purified TERT RNP with displacement oligonucleotide added directly to the activity assay reaction at the concentration indicated . ( D ) Detection of overexpressed ( OE ) 293T F-TERT in unpurified cell extract by immunoblot with TERT antibody or FLAG antibody . Cell extract lacking overexpressed TERT was used as the negative control , and detection of tubulin was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 08363 . 004 Human TERT tagged at the N-terminus supports telomere elongation , whereas telomerase assembled with C-terminally tagged TERT does not ( Counter et al . , 1998; Wong and Collins , 2006 ) . Therefore , we fused the human TERT N-terminus to a triple FLAG peptide and either the ACP or MCP tag ( Figure 1A , right ) . To assemble telomerase holoenzyme , TERT was overexpressed in 293T cells along with full-length hTR overexpressed using the U3 small nucleolar RNA promoter ( Fu and Collins , 2003 ) . To reconstitute catalytically active minimal RNP , TERT was expressed in RRL pre-supplemented with vast molar excess of purified recombinant hTRmin ( Wu and Collins , 2014a ) . TERT complexes from each reconstitution method were enriched by each of two purification approaches: TERT binding to FLAG antibody resin followed by peptide elution ( F purification ) or RNA template base-pairing to a resin-immobilized 2′O-methyl RNA ( 2′OMe ) oligonucleotide followed by displacement oligonucleotide elution ( O purification; Figure 1—figure supplement 1 , panels A , B; Schnapp et al . , 1998 ) . The 3′-modified displacement oligonucleotide used in this work did not compete with DNA primer for telomerase elongation ( Figure 1—figure supplement 1 , panel C ) . 293T cell lysates or RRL expression reactions were split and purified in parallel using the F and O purification approaches ( Figure 1B ) . ACP- and MCP-tagged TERTs expressed at equivalent level and assembled active telomerase , quantified by radiolabeled dGTP incorporation in reactions also containing dTTP and ddATP ( Figure 1C ) . CoA-Cy5 labeling of ACP-TERT using ACP synthase and CoA-Cy3 labeling of MCP-TERT using SFP synthase were confirmed by SDS-PAGE and fluorescence scanning , with no labeling of TERT lacking an ACP or MCP tag ( Figure 1D ) . Importantly , the profile of telomerase product synthesis was not affected by the labeling reaction for fluorophore conjugation ( Figure 1E ) . While purification by either tagged TERT or RNA template yields active telomerase , these purification strategies also enrich either TERT not assembled with hTR or hTR without TERT , respectively . To investigate the amounts of tagged TERT vs active RNP , we compared the levels of TERT protein and enzyme activity across the four combinations of reconstitution and purification , subsequently designated 293T-F , 293T-O , RRL-F , and RRL-O ( Figure 1B , F ) . TERT was detected by an antibody raised against its C-terminal region ( Figure 1—figure supplement 1 , panel D ) . Activity was quantified from reactions with dTTP , ddATP , and radiolabeled dGTP . O-purification by the hTR template enriched more telomerase activity relative to TERT than did F-purification ( Figure 1F ) , as would be expected based on template hybridization vs antibody binding to TERT . Comparison between the pair of 293T or RRL purifications suggests that most of the TERT in 293T-F and RRL-F was not assembled as telomerase RNP . This was anticipated for the 293T-F purification , because cellular expression of hTR is limited by inefficient co-transcriptional H/ACA RNP assembly ( Darzacq et al . , 2006; Egan and Collins , 2012b ) . However , RRL reconstitution exploits the use of pre-transcribed hTR added at very high final concentration relative to TERT . Nonetheless , even optimized RRL expression produced hTR-free TERT enriched by F-purification . To investigate the TERT content of individual complexes within a bulk fraction , we used total internal reflection fluorescence microscopy to image labeled TERT complexes bound to immobilized single-stranded T15 ( T2AG3 ) 2 DNA primer . This 5′-biotinylated primer was anchored to a polyethylene glycol-coated coverslip surface via biotin–streptavidin attachment ( Figure 2A , left ) . Primers with this 3′ permutation of the telomeric repeat have exceptionally stable binding to human telomerase ( Wallweber et al . , 2003 ) due to the finely tuned recognition of template-paired primer 3′ ends in the enzyme active site ( Brown et al . , 2014; Wu and Collins , 2014b ) . Optimal DNA binding by human telomerase requires a primer length of two telomeric repeats ( Wallweber et al . , 2003 ) , which is the same length that active human telomerase protects from nuclease digestion ( Wu and Collins , 2014a ) . 10 . 7554/eLife . 08363 . 005Figure 2 . Single-molecule detection of the TERT subunit content in DNA-bound complexes . ( A ) Left: schematic for detection of TERT content by two-color co-localization . ACP-TERT was labeled with Cy5 ( red ) and MCP-TERT was labeled with Cy3 ( green ) . PEG indicates polyethylene glycol . Center: example of detection of two-color co-localization indicated by arrowheads , for a 293T-F sample . Right: percentage of two-color co-localization for DNA-bound complexes with co-expressed ACP- and MCP-TERTs , purified by the TERT tag ( F ) or template-complementary 2′OMe RNA oligonucleotide ( O ) . For this and subsequent quantifications , values are averaged from three assays using experimentally independent replicates with standard error of the mean shown . **p < 0 . 01 using one-way ANOVA , followed by Tukey's multiple comparison test; n . s . is not significant . ( B ) Left: schematic for detection of TERT content by steps of photobleaching . MCP-TERT was labeled with Cy5 ( red ) . Center: examples of photobleaching in one or two steps . Right: percentage of MCP-TERT DNA-bound complexes labeled with Cy5 that photobleached in one , two and three , or more ( 3+ ) steps . Values are the average of triplicate experimental replicates . ( C ) The predicted relationship between detections of TERT subunit co-localization and two-step photobleaching is shown as the green line ( see Materials and methods , Equation 3 ) . Data were plotted according to measured co-localization and photobleaching in two steps only . Error bars represent standard error of the mean from triplicate experimental replicates of each measured parameter . ( D ) Measured two-color co-localization and two-step photobleaching as determined by the experiments in ( A ) and ( B ) , respectively . The following figure supplement is available for Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08363 . 00510 . 7554/eLife . 08363 . 006Figure 2—figure supplement 1 . Technical robustness of the two-color co-localization assay for TERT subunit content . Quantification of two-color co-localization for DNA-bound 293T-F and 293T-O TERT complexes was consistent across different spot densities per field ( A ) and using cell extracts from independent transfections of 293T cells ( B ) . In ( A ) , light blue dots are 293T-F quantification and dark blue dots are 293T-O quantifications from imaging multiple ( ≥8 ) fields for each of triplicate experimental replicates . In ( B ) , averages are from multiple ( ≥8 ) fields counted for each sample . DOI: http://dx . doi . org/10 . 7554/eLife . 08363 . 006 We applied two parallel approaches to determine the number of TERT subunits per complex . In the first method ( Figure 2A , left ) , we assembled telomerase by co-expression of ACP-TERT and MCP-TERT and labeled the TERTs sequentially , first labeling ACP-TERT with ACP synthase and CoA-Cy5 then labeling MCP-TERT with SFP synthase and CoA-Cy3 . Fields of individual complexes were imaged to detect both dyes , and images were scored for the fraction of Cy5-labeled ACP-TERT that co-localized a Cy3-labeled MCP-TERT . In the second TERT subunit counting method ( Figure 2B , left ) , we assembled telomerase complexes containing only MCP-TERT and labeled using SFP synthase and CoA-Cy5 . Fields of individual complexes were imaged , and each Cy5 ‘spot’ in the flow cell was analyzed for the number of dye photobleaching steps that occurred before the spot vanished . In the parallel approaches , the fraction of two-color co-localized spots and the number of photobleaching steps are both readily related to the sample fractional content of TERT monomer and dimer considering all possible two-subunit combinations ( see Materials and methods , Equations 1 , 2 ) . Fluorescently labeled TERT complexes were diluted to obtain 1–4 spots per 100 µm2 of the slide surface , and unbound protein was removed before imaging . To attain similar spot count per field across samples , labeled O-purification complexes required dilution relative to F-purification complexes isolated from an equal amount of the same extract , consistent with the greater yield of active RNP for O-purification ( Figure 1F ) . Both two-color co-localization and photobleaching assays revealed the presence of more than one labeled TERT in a subset of the DNA-bound TERT complexes ( Figure 2A , B ) . In the two-color co-localization assay , there was no statistically significant difference in TERT co-localization comparing DNA-bound 293T-O , RRL-F , and RRL-O complexes ( Figure 2A; 17–23% , p = 0 . 58 ) . In contrast , 293T-F complexes had much more TERT co-localization ( 46% , p = 0 . 0015 ) . This distinction was consistent across a range of fluorescent spot density per field and different 293T cell extracts used for purifications ( Figure 2—figure supplement 1 ) . Results from the photobleaching method of TERT subunit counting also indicated no statistically significant difference in TERT subunit content across the population of DNA-bound complexes from 293T-O , RRL-F , and RRL-O ( Figure 2B; 19–27% bleaching in multiple steps , p = 0 . 33 ) . In contrast , 293T-F complexes had much more multistep photobleaching ( 43% , p = 0 . 0066 ) including a substantial fraction of complexes that photobleached in three or even more than three steps ( 12% ) . We analyzed whether the results from the methods of subunit counting were consistent with each other , assuming a mixed population of TERT monomer and TERT dimer complexes ( Materials and methods , Equation 3 , and see below ) . There is excellent correlation of two-color co-localization to two-step photobleaching results for 293T-O , RRL-F , and RRL-O but not 293T-F ( Figure 2C ) . The abundance of 293T-F TERT complexes that photobleached in three or more steps is likely responsible for this discord , as this population of complexes was distinguished from TERT dimer complexes in the count of photobleaching steps but would be lumped together with TERT dimer complexes in the count of two-color co-localization . Together , the findings above reveal a surprising diversity of TERT subunit content in DNA-bound complexes . Furthermore , it is evident that this heterogeneity varies across the methods of telomerase reconstitution and purification ( Figure 2D ) . The subunit co-localization and multistep photobleaching values measured above are related to the number of TERTs within each complex but are also influenced by the labeling efficiency of the MCP tag . Therefore , in order to calculate the fraction of complexes with TERT monomer or TERT dimer for each method of reconstitution and purification , it was necessary to establish MCP-TERT-labeling efficiency . Labeling of 293T- and RRL-reconstituted , F-purified MCP-TERT complexes was to saturation within 30 min of a reaction with CoA-Cy3 or CoA-Cy5 ( Figure 3A ) , well within the standard 2-hr labeling protocol . Also , labeling efficiency was not dependent on the reconstitution and purification method ( Figure 3B ) . We adapted a previously developed approach to quantify a minimum lower bound of labeling efficiency without assumptions from fluorescence intensity ( Yin et al . , 2005 ) . Using CoA-biotin as the synthase substrate results in covalent target protein biotinylation , which can be used as the basis for protein depletion by binding to streptavidin resin . The minimum lower bound of labeling efficiency can be calculated from the amount of protein remaining in the unbound fraction . First , we confirmed that CoA-biotin is used equivalently to CoA-fluorophore by measuring competition between CoA-biotin and CoA-Cy3 . If RRL-reconstituted F-purified MCP-TERT was labeled to saturation with CoA-Cy3 and then labeled with CoA-biotin , no biotinylation of TERT could be detected ( Figure 3C ) . Similarly , labeling with CoA-biotin drastically reduced subsequent labeling with CoA-Cy3 ( Figure 3C ) . Therefore , biotin labeling provides a surrogate for quantification of dye labeling efficiency . 10 . 7554/eLife . 08363 . 007Figure 3 . Quantification of the TERT monomer vs multimer content in purified samples based on TERT-labeling efficiency . ( A ) SDS-PAGE analysis of the kinetics of labeling F-purified 293T- or RRL-reconstituted MCP-TERT in reactions with CoA-Cy5 or CoA-Cy3 and SFP synthase . Lines within the panel indicate separate sets of gel lanes . Quantification of labeling intensity was normalized to labeling at the 4-hr time point after subtraction of background . ( B ) Cy5 labeling relative to TERT amount analyzed for telomerase reconstituted and purified as indicated . TERT was detected by TERT immunoblot . Values are the average of triplicate experimental replicates . ( C ) Validation of equivalent labeling using CoA-dye or CoA-biotin by sequential labeling of F-purified , RRL-expressed MCP-TERT with SFP synthase . Initial TERT labeling using CoA-Cy3 or CoA-biotin competes for subsequent TERT labeling by the other CoA derivative . The biotin label on MCP-TERT was detected by biotin antibody immunoblot . ( D ) Left: schematic of the biotinylated TERT depletion procedure . Right: quantification of ACP- and/or MCP-TERT remaining after streptavidin agarose depletion , following reconstitution ( RRL unless indicated otherwise ) , F-purification and labeling using CoA-biotin and ACP ( A ) or SFP ( S ) synthase . RRL-expressed TERT was 35S-methionine labeled and 293T-expressed TERT was detected by FLAG antibody immunoblot . Samples labeled in reactions lacking CoA-biotin ( not Biotin + ) were labeled with CoA and those not applied to streptavidin agarose ( not Streptavidin depletion + ) were mock-depleted on Myc antibody agarose . Lines within the panel indicate separate sets of gel lanes run in parallel . Percentage unbound was calculated as unbound signal normalized to unbound signal of the control depletion . Values are the average of triplicate experimental replicates . ( E ) Activity of the unbound fraction after streptavidin agarose depletion of biotinylated telomerase labeling using CoA-biotin and SFP synthase , under native binding conditions . Telomerase activity was assayed in reactions with dATP , dGTP , dTTP , and α-32P dGTP , followed by denaturing gel electrophoresis; number of 6-nucleotide repeats added to product DNA is indicated . Samples not depleted with streptavidin agarose were mock-depleted on Myc antibody agarose . Lines within the panel indicate separate sets of gel lanes run in parallel . Percentage unbound was normalized to unbound after control depletion . Values are the average of triplicate experimental replicates . ( F ) Left: schematic of the biotinylated TERT depletion procedure and unbound fraction analysis . Right: quantification of total TERT and biotinylated MCP-TERT in the unbound fraction of 293T-reconstituted , F-purified telomerase , following labeling using CoA-biotin or CoA and depletion by streptavidin agarose or mock-depletion on Myc antibody agarose . MCP-TERT and the biotin label on MCP-TERT were detected by immunoblot . Values are the average of triplicate experimental replicates . ( G ) Illustration of labeling efficiency determination by comparison of the percent unbound total MCP-TERT and unbound biotinylated MCP-TERT . ( H ) Calculated percentage of DNA-bound TERT monomer complexes according to fraction TERT subunit co-localization ( percentages indicated ) , assuming the TERT-labeling efficiency measured value ( 82% , blue line; bar graph at right ) , lower bound ( 51% , green line; Low L numbers at right ) , or upper bound ( 100% , purple line; High L numbers at right ) . Vertical dashed lines are the observed fraction of two-color co-localization ( from Figure 2A ) . ( I ) Calculated percentage of DNA-bound TERT monomer complexes according to fraction of one-step photobleaching ( percentages indicated ) , assuming the TERT-labeling efficiency measured value ( 82% , blue line; bar graph at right ) , lower bound ( 51% , green line; High L numbers at right ) , or upper bound ( 100% , purple line; Low L numbers at right ) . Vertical dashed lines are the observed fraction of one-step photobleaching ( from Figure 2B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08363 . 007 To preclude the depletion of unlabeled TERT as part of a TERT multimer , biotin-labeled TERT complexes were bound to streptavidin in 2 M urea . Complexes of RRL-reconstituted TERT synthesized with 35S-methionine were F-purified and labeled with biotin , biotinylated TERT was depleted using streptavidin agarose , and the fraction of unbound TERT was quantified by radiolabel detection after SDS-PAGE ( Figure 3D ) . Depletion was an indistinguishable 54 . 9 ± 0 . 7% or 50 . 6 ± 5 . 9% of ACP-TERT or MCP-TERT , respectively , quantified from the unbound ∼45% or ∼49% ( Figure 3D , lanes 3–6 ) . As a negative control , TERT from labeling reactions with underivatized CoA was not depleted ( Figure 3D , lanes 1–2 ) . Also , MCP-TERT was not depleted after a labeling reaction with ACP synthase ( Figure 3D , lanes 7–8 ) . When ACP- and MCP-TERT were co-expressed , CoA-biotin labeling of ACP-TERT by ACP synthase resulted in half the depletion attained when ACP-TERT alone was expressed ( Figure 3D , lanes 9–10 vs 3–4 ) , confirming equal co-expression of the two tagged TERTs . Similar depletion was observed for biotin-labeled F-purified MCP-TERT expressed in 293T cells , detected by immunoblot ( Figure 3D , lanes 11–12 ) . Also , ∼50% depletion was observed for the catalytic activity of MCP-TERT RNPs labeled with CoA-biotin bound to streptavidin in native rather than denaturing conditions , independent of the reconstitution or purification method or enzyme repeat addition processivity ( Figure 3E ) . To measure how efficiently biotinylated TERT was depleted by streptavidin in the 2 M urea condition that converts the entire population of protein to monomer , we determined the fraction of biotinylated TERT that was depleted compared to the fractional depletion of total TERT . For maximal immunoblot detection sensitivity , the 293T-expressed F-purified MCP-TERT was biotin labeled , allowed to bind streptavidin then analyzed by immunoblot with antibodies specific for TERT and biotin ( Figure 3F ) . Streptavidin depleted 49% of the TERT protein ( Figure 3F ) , consistent with the previous TERT depletions ( Figure 3D ) . However , 38% of biotinylated TERT remained unbound ( Figure 3F ) , revealing that streptavidin binding in 2 M urea did not completely deplete the labeled TERT . Therefore , the MCP-TERT-labeling efficiency was much higher than 51% . Correcting the quantified total TERT depletion for depletion efficiency of biotin-TERT gives an MCP-TERT-labeling efficiency of 82% ( Figure 3G ) . This matches the labeling efficiency determined for a related tag in similar reactions with SFP synthase and CoA-biotin ( Yin et al . , 2005 ) . We determined the fraction of TERT monomer vs dimer in each population using the co-localization quantifications ( Figure 2A , Materials and methods , Equation 1 ) or the photobleaching quantifications ( Figure 2B , Materials and methods , Equation 2 ) by modeling the DNA-bound complexes as having either one or two TERTs . We set labeling efficiency as 82% but also modeled a range of labeling efficiency from 51% to 100% as lower and upper bounds ( indicated as ‘Low L’ and ‘High L’ limits ) . Co-localization and photobleaching quantifications support modeling of DNA-bound 293T-O , RRL-F , and RRL-O populations as a mixture of complexes with one or two TERT subunits ( Figure 3H , I ) . The 293T-F population of DNA-bound TERT could not be modeled as a mixture of TERT monomer and dimer across the full range of labeling efficiency using the co-localization quantification ( Figure 3H ) , likely due to the substantial fraction of complexes with three or more TERT subunits ( Figure 2B ) . TERT monomer complexes exceeded TERT dimer complexes in the DNA-bound 293T-O , RRL-F , and RRL-O populations across almost the entire range of modeled labeling efficiencies ( Figure 3H , I ) . Overall , these analyses establish that TERT complexes competent for DNA binding can have a single subunit of TERT . The heterogeneity of TERT subunit content in DNA-bound complexes described above raised the question of whether only a subset of the DNA-bound complexes corresponds to active RNP . To investigate this question , we exploited the permutation-dependent telomeric-repeat DNA-binding affinity of the human telomerase active site . The single-stranded T15 ( T2AG3 ) 2 DNA primer used to bind TERT complexes to the flow cell surface has extremely slow dissociation from the telomerase holoenzyme active site ( Wallweber et al . , 2003 ) . Introducing dTTP + dATP into the imaging chamber would support primer extension to a GGGTTA-3′ end ( Figure 4A ) , which disengages from the active site with koff at least ∼100-fold greater than the TTAGGG-3′ end ( Wallweber et al . , 2003 ) . Thus , Cy5-labeled MCP-TERT complexes with DNA bound in a functional active site would exhibit activity-dependent elution in buffer with dTTP + dATP ( Figure 4A ) . Inactive RNP and hTR-free TERT would remain bound as well as some active RNP not dissociated from product ( Figure 4A ) , and also any 293T TERT bound to DNA indirectly through an associated shelterin complex . To control for activity-independent dissociation of TERT complexes from DNA , we performed parallel incubations without dTTP + dATP . We also assayed complexes reconstituted with the catalytic-dead TERT D868A ( Weinrich et al . , 1997 ) . Elution of Cy5-labeled MCP-TERT was monitored by spot count per field over 30 min in buffer with or without dNTPs ( Figure 4B ) . 10 . 7554/eLife . 08363 . 008Figure 4 . Distinct profiles of activity-dependent elution across populations of TERT complexes . ( A ) Schematic of TERT complexes' interaction with bound DNA . In the presence of dTTP and dATP , complexes bound productively to primer end GGG-3′ would elongate the primer to GGGTTA-3′ accompanied by increased likelihood of DNA release ( elution , at top ) . Non-productively bound RNP complexes and hTR-free TERT would not elongate the primer and therefore not elute by DNA synthesis , and some productively bound RNP complexes could also fail to elongate primer and/or to release from product DNA . The t1/2 values are from published studies using human telomerase holoenzyme ( Wallweber et al . , 2003 ) . ( B ) Schematic of the activity-dependent elution procedure . ( C ) Activity-dependent elution of Cy5-labeled wild-type ( WT ) or catalytic-dead ( D868A ) MCP-TERT complexes using buffer containing dATP + dTTP or buffer only . Spot count per field of labeled TERT complexes was normalized to the initial time point . Specific elution was calculated by subtracting the fraction of complexes with buffer-only elution from the fraction eluted with dNTPs . The relative count of DNA-bound complexes from sample pre-treated with RNase A is indicated by shaded gray bars . ( D ) Schematic of the procedure for post-elution counting and photobleaching of labeled complexes . ( E ) Number of MCP-TERT DNA-bound complexes labeled with Cy5 per imaging field that photobleached in one , two and three , or more steps after elution incubation with or without dNTPs . ***p < 0 . 001 by unpaired Student's t-test , n . s . is not significant . ( F ) Calculated percentage of DNA-bound TERT monomer and dimer complexes after elution according to fractional one-step photobleaching , assuming 82% TERT-labeling efficiency . DOI: http://dx . doi . org/10 . 7554/eLife . 08363 . 008 Reproducibly , more elution of WT TERT complexes occurred in the presence of buffer with dTTP + dATP vs buffer alone ( Figure 4C , compare black and gray ) . In contrast , D868A TERT complexes showed the same amount of dissociation with or without dNTPs ( Figure 4C , compare dark and light blue ) . Curiously , the fraction of WT TERT complexes with activity-dependent elution varied widely across the TERT populations from different reconstitution and purification conditions ( Figure 4C ) . RRL-F and RRL-O complexes showed predominantly activity-dependent elution . About half of the DNA-bound 293T-O complexes also showed activity-dependent elution , but a surprisingly low percentage of 293T-F complexes eluted with the opportunity for DNA synthesis . The non-eluting fraction of TERT complexes roughly correlated with the fraction of complexes that could bind slide-immobilized DNA after sample pre-treatment with RNase A ( Figure 4C , gray bars ) . Importantly , the fraction of DNA-bound 293T-O , RRL-F , and RRL-O TERT complexes with nucleotide-dependent-‘specific’ elution ( Figure 4C ) overlaps the fraction of DNA-bound complexes with monomeric TERT ( Figure 3H , I ) . This overlap establishes that at least some TERT monomer RNPs have catalytic activity . To directly measure the contribution of TERT monomer RNPs to specific elution , we used RRL-reconstituted O-purified Cy5-labeled MCP-TERT complexes to quantify TERT spot count per field and steps of photobleaching for samples after incubation in parallel for 30 min in buffer with or without dNTPs ( Figure 4D ) . Approximately one third as many labeled TERT complexes were present in samples incubated with dNTPs ( Figure 4E ) , consistent with the elution time course ( Figure 4C ) . The reduction of TERT spot count by activity-based elution occurred entirely in complexes with single-step photobleaching ( Figure 4E , p = 0 . 0006 ) . The conditions of elution altered the relative representation of TERT monomer and dimer complexes , calculated by adjusting the photobleaching step quantifications for TERT-labeling efficiency ( Figure 4F ) . Consistent with specific elution of TERT monomer complexes , the DNA-bound TERT complexes remaining after specific elution were enriched for TERT dimer . Overall , the findings above strongly suggest that human telomerase catalytic activity requires only a single TERT subunit per RNP . The heterogeneity of DNA-bound TERT complex elution was surprising . We therefore investigated whether the bulk populations of TERT complexes from different reconstitution and purification conditions had heterogeneous DNA-binding affinities as well . Towards this goal , we quantified the DNA-binding affinity of TERT complexes anchored directly to the flow cell surface . To do this , we labeled MCP-TERT with CoA-biotin , bound the biotin-labeled TERT complexes to streptavidin on the flow cell surface , and assayed the immobilized TERT complexes for retention of Cy5-labeled ( T2AG3 ) 2 ( Figure 5A ) . This direct TERT immobilization captured the full TERT heterogeneity of the bulk purification fractions , which we monitored separately by SDS-PAGE of MCP-TERT labeled with Cy5 ( Figure 5B ) . Bulk 293T-F purifications of MCP-TERT contained a large amount of a proteolytic product corresponding to the MCP-tagged TERT TEN domain alone ( Figure 5B ) . TEN domain expressed in Escherichia coli has barely detectable if any DNA-binding activity ( O'Connor et al . , 2005; Sealey et al . , 2010 ) , suggesting that it would not form a stable complex with the Cy5-labeled ( T2AG3 ) 2 . None of the other bulk purification fractions contained TEN domain alone , but curiously the 293T-F and 293T-O bulk purifications contained TERT proteolytic products corresponding to the TEN domain plus adjacent linker ( Figure 5B; see below ) . 10 . 7554/eLife . 08363 . 009Figure 5 . Direct DNA-binding affinity comparison for TERT complexes in bulk purifications . ( A ) Schematic for detection of Cy5-labeled DNA binding to biotinylated TERT complexes . ( B ) SDS-PAGE analysis of MCP-TERT complexes labeled using CoA-Cy5 . Cy5-labeled MCP-TERT proteolysis products that retain the N-terminal F-MCP tag and are enriched in the 293T-F purification are schematized in comparison to full-length TERT . ( C ) Concentration dependence of Cy5-labeled DNA retention by slide-anchored TERT complexes across a titration of 0 . 3 , 1 , 3 , 10 , and 30 nM DNA . Spot count per field was normalized to the 30 nM DNA quantification for each sample . Error bars represent standard error of the mean of spot counts of five fields per sample per DNA concentration . ( D ) Graph of the change in Cy5-labeled DNA spot count comparing assays of 30 vs 100 nM DNA , normalized to the 30 nM DNA quantifications for each sample . Error bars represent standard error of the mean of spot counts of five fields per sample per DNA concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 08363 . 009 We quantified the amount of Cy5-labeled ( T2AG3 ) 2 bound to immobilized TERT complexes using a range of DNA concentration . DNA binding across a titration from 0 . 3 to 30 nM DNA yielded Kd calculations in nM of 4 . 9 ± 0 . 7 for 293T-O , 5 . 4 ± 1 . 1 for 293T-F , 8 . 7 ± 2 . 0 for RRL-O , and 14 . 7 ± 5 . 4 for RRL-F ( Figure 5C ) . The ∼5 nM Kd of holoenzyme and ∼10 nM Kd of minimal RNP are consistent with the holoenzyme Km for elongation of similar primers measured , under different conditions , as 2 nM or 8 nM ( Wallweber et al . , 2003; Jurczyluk et al . , 2010 ) . In parallel , immobilized TERT complexes were assayed for DNA binding using 100 nM ( T2AG3 ) 2 . DNA binding by 293T-F TERT complexes increased ∼fourfold with 100 nM compared to 30 nM DNA ( Figure 5D ) . In contrast , at 100 nM compared to 30 nM DNA concentration , 293T-O TERT complexes showed no additional DNA binding and RRL-F and RRL-O complexes showed only limited additional association with DNA ( Figure 5D ) . These findings suggest that 293T-O , RRL-F , and RRL-O TERT complexes competent for DNA binding have a relatively homogeneous DNA-binding affinity matching the expectation for catalytically active human telomerase . Next , we sought to create a homogeneous pool of TERT monomer or dimer complexes . Many variations of reconstitution method had surprisingly little impact on the DNA-bound TERT monomer/dimer ratio , with one exception: elimination of the 125 amino acid linker between the TERT ring and TEN domain . Phylogenetic comparison revealed that this domain linker is particularly long in vertebrate TERTs ( Podlevsky et al . , 2008 ) , approximately 100 amino acids longer than in the ciliate and budding yeast TERTs that assemble only TERT monomer RNPs ( Livengood et al . , 2002; Bryan et al . , 2003; Witkin and Collins , 2004; Cunningham and Collins , 2005; Jiang et al . , 2013; Bajon et al . , 2015 ) . Scanning six-residue substitutions of human TERT linker sequence did not uncover any significance of the region for telomerase catalytic activity or telomere maintenance ( Armbruster et al . , 2001 ) , but this approach did not alter the atypical amino acid composition of the linker region overall . Human TERT amino acids 201–325 are 18% proline , 14% arginine , and 12% glycine . When subject to bioinformatical analysis for amino acid content ( Harbi et al . , 2011 ) , this region is identified as having high compositional bias . We also used SEG analysis ( Wootton and Federhen , 1993 ) to search for low-complexity sequence within the human TERT linker . SEG analysis identified two segments of the linker , residues 213–248 and 313–323 , as low complexity . Low-complexity regions can mediate diverse protein–protein interactions including concentration-dependent self-association ( Coletta et al . , 2010; Kato et al . , 2012 ) . Thus , the TERT low-complexity proline/arginine/glycine-rich linker ( termed the PAL ) is a candidate region for mediating self-association of overexpressed TERT . Whether the length of the human TERT PAL influences RNP assembly or activity has not been tested . Previous assays that separated the TEN domain from the TERT ring retained the PAL on either the TEN domain or TERT ring ( Robart and Collins , 2011; Wu and Collins , 2014a ) . We therefore removed TERT residues 201–325 from the N-terminally F-tagged full-length protein , either by simply deleting the region ( TERT-∆PAL; Figure 6A ) or replacing it with 5 , 10 , or 20 repeats of the sequence NAAIRS ( TERT-5N , −10N , −20N; Figure 6A ) , the six amino acid motif used previously in the non-disruptive scanning mutagenesis ( Armbruster et al . , 2001 ) . The PAL-mutant TERT proteins expressed at levels similar to WT TERT in 293T cells and in RRL ( Figure 6A ) , and binding of WT and PAL-mutant TERT complexes to FLAG antibody resin enriched similar amounts of catalytic activity ( Figure 6B ) . Although direct fusion of the TEN domain to the TERT ring did not substantially affect the quantified overall activity it appeared to reduce the amount of the longest product DNAs ( Figure 6B ) . This change in product profile was rescued by NAAIRS repeat insertion ( Figure 6B ) . Since the number of radiolabeled dGTP nucleotides incorporated into a product DNA is proportional to length , products elongated by many repeats are detected with disproportionately high sensitivity relative to their actual abundance . To more accurately profile the repeat addition processivity of the reconstituted enzymes , we assayed telomerase activity using a primer radiolabeled at its 5′ end rather than by extension with radiolabeled dNTPs . This also allowed the use of a non-limiting concentration of dGTP in the activity assay reaction ( see ‘Materials and methods’ ) . A 5-min pulse of primer extension was followed by a chase period with excess unlabeled primer to eliminate telomerase reinitiation on released product DNA . Under these conditions , primer extension was highly processive for complexes of WT TERT , TERT-∆PAL , and TERT-20N assembled in 293T cells ( Figure 6C ) . Similar results were obtained with RRL-reconstituted enzymes ( data not shown ) . We conclude that human TERT linker length and linker sequence have a very limited influence on the catalytic activity of reconstituted holoenzyme or minimal RNPs . 10 . 7554/eLife . 08363 . 010Figure 6 . Telomerase RNP assembly and activity without the TEN domain linker . ( A ) Schematic representation and expression of N-terminally F-tagged human TERT proteins with the linker replaced by 20 , 10 , or 5 repeats of the sequence NAAIRS ( TERT-20N through 5N ) or linker deleted without compensating sequence insertion ( TERT-∆PAL ) . TERTs expressed in 293T cells were detected by immunoblot with TERT antibody , and TERTs expressed in RRL were detected by 35S-methionine labeling during synthesis . ( B ) Activity and hTR content of 293T- or RRL-reconstituted , F-purified TERT RNPs with altered linker sequence , bound to FLAG antibody resin . Spot-blot hybridization was used to detect hTR . Relative activity and hTR content were normalized to the WT TERT purification after background subtraction of activity or hTR in the purification of untagged WT TERT . Specific activity was calculated from relative activity and relative hTR . ( C ) Processive extension of 5′-labeled ( T2AG3 ) 3 primer by telomerase assembled with WT , ∆PAL , or 20N TERT bound to FLAG antibody resin . The labeled primer was extended for 5 min before chase addition of unlabeled primer for a total extension time of 10 , 20 , or 40 min . ( D ) Activity and hTR content of telomerase in 293T input extracts or bound to Myc antibody resin . TPP1 OB-fold domain expression and purification were confirmed by immunoblot detection of the 3xMyc tag . Immunoblot and activity assay with whole-cell extract used 2% of the total purification input . Half of the post-purification sample was used for activity assays and half for Myc immunoblot . Spot-blot hybridization was used to detect hTR . Relative activity and hTR content were normalized to the input or bound sample for TPP1 purification of WT TERT , after bound hTR background subtraction using the purification without tagged TPP1 OB-fold domain . Relative percentage enrichment was calculated as relative bound activity adjusted for relative input activity . Specific activity was calculated from relative activity and relative hTR . ( E ) SDS-PAGE analysis of O-purified 293T MCP-TERT complexes labeled using CoA-Cy5 . MCP-TERT fragments resulting from proteolysis within the PAL of WT TERT are schematized , in comparison to full-length TERT . DOI: http://dx . doi . org/10 . 7554/eLife . 08363 . 010 Telomerase-mediated telomere synthesis is strictly dependent on TEN domain interaction with the oligonucleotide/oligosaccharide-binding ( OB ) fold domain of the shelterin component TPP1 ( Xin et al . , 2007; Schmidt et al . , 2014; Sexton et al . , 2014 ) . To test whether sequence substitutions of the PAL compromise catalytically active telomerase association with TPP1 , we co-overexpressed N-terminally 3xMyc-tagged TPP1 OB-fold domain ( TPP1 residues 88–249 ) with F-tagged WT TERT , TERT-∆PAL , or TERT-20N in 293T cells . Within a twofold difference , the TPP1 OB-fold domain co-purified active telomerase regardless of TERT linker length or sequence ( Figure 6D ) . As additional characterization prior to single-molecule imaging , we analyzed Cy5-labeled O-purified MCP-tagged WT TERT , TERT-∆PAL , and TERT-20N complexes by SDS-PAGE . Curiously , the 293T TERT-∆PAL and TERT-20N bulk purification fractions lacked the TERT proteolysis products co-enriched by WT TERT ( Figure 6E ) . The WT TERT proteolysis products correspond to the TEN domain fused to lengths of PAL ending at the two computationally identified low-complexity regions . A simple hypothesis to explain these findings is that an hTR-bound full-length TERT can co-purify a TERT fragment dimerized through the PAL . This would account for the detection of some PAL-containing TEN domain in the 293T-O bulk purification , which unlike the bulk F-purification should not directly enrich TERT fragments compromised for hTR binding . Bulk purification fractions of RRL-reconstituted WT TERT lacked the TERT proteolysis products detected in the 293T bulk purifications ( Figure 5B and data not shown ) , suggesting that the TERT PAL may be a target of protease cleavage in cells . Furthermore , this proteolysis is specific for WT PAL sequence because no TEN domain fragments were observed the TERT-∆PAL and TERT-20N purifications ( Figure 6E ) . We note that although TERT proteolysis products are present in the 293T WT TERT bulk purifications , they may not be represented in the DNA-bound subset of TERT complexes assayed by single-molecule imaging . To investigate the TERT subunit content of PAL-mutant TERT complexes bound to DNA , we first O-purified 293T- and RRL-reconstituted complexes of co-expressed ACP- and MCP-tagged WT TERT , TERT-∆PAL , or TERT-20N . A dramatic decrease in TERT co-localization was observed for TERT-∆PAL and TERT-20N relative to WT TERT ( Figure 7A; 21% vs 5% co-localization in 293T samples , p = 0 . 0008 , and 22% vs 2–3% in RRL samples , p < 0 . 0001 ) . By calculations using a value of 82% TERT-labeling efficiency , RRL-reconstituted TERT-∆PAL and TERT-20N complexes were 98% and 96% TERT monomer , respectively ( Figure 7B ) . Even by modeling using the lower-bound underestimate of TERT-labeling efficiency , TERT monomer complexes were 93–96% of the DNA-bound RRL-reconstituted TERT complex total ( Figure 7B ) . For 293T-reconstituted TERT-∆PAL and TERT-20N complexes , TERT monomers were 94% of the population with a lower-bound underestimate of 89–90% ( Figure 7B ) . 10 . 7554/eLife . 08363 . 011Figure 7 . PAL-mediated TERT dimerization . ( A ) Two-color co-localization quantification for DNA-bound O-purified complexes of coexpressed ACP- and MCP-TERTs . Values are the average of triplicate experimental replicates . ***p < 0 . 001 using one-way ANOVA , followed by Tukey's multiple comparison test . ( B ) Calculated percentage of DNA-bound TERT monomer complexes according to the fraction of two-color TERT co-localization ( percentages indicated ) , assuming the TERT-labeling efficiency measured value ( 82% , blue line; bar graph at right ) , lower bound ( 51% , green line; Low L numbers at right ) , or upper bound ( 100% , purple line; High L numbers at right ) . ( C ) Photobleaching step quantification for DNA-bound O-purified MCP-TERT complexes labeled with Cy5 . Values are the average of triplicate experimental replicates . ( D ) Calculated percentage of DNA-bound TERT monomer complexes according to the fraction of one-step photobleaching ( percentages indicated ) , assuming the TERT-labeling efficiency measured value ( 82% , blue line; bar graph at right ) , lower bound ( 51% , green line; Low L numbers at right ) , or upper bound ( 100% , purple line; High L numbers at right ) . ( E ) Activity-dependent elution of O-purified Cy5-labeled MCP-TERT complexes using buffer containing dATP + dTTP or buffer only . Spot count per field of labeled TERT complexes was normalized to the initial time point of each sample . Specific elution was calculated by subtracting the fraction of complexes with buffer-only elution from the fraction eluted with dNTPs . ( F ) Illustration presenting the hypothesis of differences in TERT PAL conformation that occur with TERT RNP assembly or dimerization . The PAL is shown with conformations that correlate with catalytically active ( red ) or inactive ( pink ) TERT complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 08363 . 011 Parallel results were obtained by quantifying TERT subunit content using Cy5-labeled MCP-TERT steps of photobleaching . MCP-tagged TERT-∆PAL and TERT-20N complexes assembled in 293T cells or RRL were dramatically depleted for multistep photobleaching compared to WT TERT complexes ( Figure 7C; 24% vs 8–9% multistep bleaching in 293T samples , p = 0 . 0036 , or 22% vs 4–7% in RRL samples , p = 0 . 0073 ) . By calculations using a value of 82% TERT-labeling efficiency , RRL-reconstituted TERT-∆PAL and TERT-20N complexes were 92% and 95% TERT monomer , respectively ( Figure 7D; 86–96% across the modeled range of TERT-labeling efficiency ) . Similarly , 293T-reconstituted TERT-∆PAL and TERT-20N complexes were 90% and 88% TERT monomer , respectively ( Figure 7D; 80–92% across the modeled range of TERT-labeling efficiency ) . To determine whether RNPs assembled with TERT-∆PAL and TERT-20N retained the characteristic permutation dependence of human telomerase DNA binding , we tested Cy5-labeled MCP-tagged TERT-∆PAL and TERT-20N complexes for activity-dependent elution . More elution of TERT-∆PAL and TERT-20N complexes occurred in buffer + dNTPs than in buffer alone ( Figure 7E ) . As observed for WT TERT complexes , the TERT-∆PAL and TERT-20N complexes assembled in 293T cells showed lower efficiency of specific elution than complexes assembled in RRL . Nevertheless , specific elution of 293T and RRL complexes of TERT-∆PAL or TERT-20N uniformly exceeded the fraction of TERT dimer complexes in each population , determined using subunit co-localization or photobleaching ( Figure 7B , D , E ) . We conclude that although PAL disruption drastically reduced TERT dimerization , RNPs assembled with PAL-mutant TERTs retained catalytic activity and even the permutation-dependent release of product DNA characteristic of the human telomerase active site . TERT dimerization was as effectively suppressed for telomerase holoenzyme assembled in cells as for minimal RNP assembled in RRL , suggesting that the TERT subunit content of reconstituted complexes has no dependence on any holoenzyme protein other than TERT . We speculate that in physiological context , protein interaction ( s ) mediated by the TERT PAL could chaperone hTR-free TERT from its synthesis in the cytoplasm to nuclear sites of RNP assembly ( Figure 7F , left ) . TERT overexpression may bypass this chaperoning requirement and promote a TERT self-association disfavorable for TEN domain positioning relative to TERT ring in an active RNP ( Figure 7F ) .
Understanding telomerase mechanism and regulation requires knowledge of the subunit stoichiometry of an active RNP . Whether assembled in vivo or in vitro , we show that human telomerase complexes monomeric for TERT are catalytically active . Monomeric TERT was abundant in the populations of DNA-bound complexes from at least three of the four bulk purification samples examined here , particularly from any purification using a template-complementary oligonucleotide . These results establish the phylogenetic conservation of a TERT-monomer telomerase active site . Also , our results support TERT haploinsufficiency rather than dominant-negative inhibition as the mechanism accounting for human disease from heterozygous TERT mutation ( Armanios et al . , 2005; Armanios and Blackburn , 2012 ) . We note that although the budding yeast telomerase holoenzyme has a TERT monomer ( Bajon et al . , 2015 ) , multiple telomerase RNPs can transiently co-localize as a cluster ( Gallardo et al . , 2011 ) . Similarly in human cells , Cajal bodies and/or shelterin interactions could dynamically cluster telomerase RNPs within a general nuclear area . The biological significance of this clustering remains to be determined ( Hockemeyer and Collins , 2015 ) . The biased amino acid composition , low-complexity sequence , and predicted lack of structure of the TERT PAL all may promote overexpressed TERT formation of dimers and aggregates . The similar TERT monomer/dimer ratio observed for DNA-bound O-purified 293T vs RRL complexes suggests that TERT self-association accounts for the vast majority of dimer formation , since 293T and RRL TERT complexes differ in all components other than TERT ( full-length hTR and H/ACA proteins vs hTRmin ) . TERT complexes with multistep photobleaching appeared to support little if any activity-dependent elution from DNA . It remains possible that active RNP dimers form under reconstitution conditions other than the standard protocols used in this work . Also , not all TERT monomer complexes had efficient activity-dependent elution: a larger fraction of 293T-O complexes than RRL-O complexes failed to elute with the opportunity for DNA synthesis , even when these complexes were converted to nearly homogeneous TERT monomer content by PAL deletion . We speculate that this difference arises from the greater heterogeneity of TERT structure , modification , and interaction partners produced by expression in cells . All of the findings above raise the need for caution in the interpretation of biochemical assays conducted using bulk purifications of TERT complexes . Surprisingly , even selection for single-stranded DNA-binding activity did not fully discriminate against inactive TERT . We pinpoint a proline/arginine/glycine-rich human TERT domain linker as the major site of TERT dimerization . Although the PAL mediates dimerization of overexpressed TERT , at lower endogenous TERT expression level , we propose that the PAL has other biological roles . To address this hypothesis , it will be important to determine PAL interaction partners using approaches that recapitulate a physiological TERT expression level . Also , it will be of interest to understand which features of the TERT PAL are functionally significant . Because the PAL is present in vertebrate but not ciliate or budding yeast TERTs , we predict that it has biological function ( s ) related to the assembly of the vertebrate telomerase holoenzyme as an H/ACA RNP .
HEK 293T cells were transiently transfected with pcDNA3 . 1 TERT expression plasmid ( s ) , the hTR expression plasmid pBS-U3-hTR-500 ( Fu and Collins , 2003 ) , and where indicated , the N-terminally triple Myc-tagged TPP1 OB-fold domain ( residues 88–249 ) expression plasmid pcDNA3 . 1-3xMyc-TPP1 ( 88–249 ) using calcium phosphate . After 48 hr , cells were resuspended in HLB buffer ( 20 mM HEPES at pH 8 , 2 mM MgCl2 , 0 . 2 mM EGTA , 10% glycerol , 0 . 1% NP-40 , 1 mM DTT , and 0 . 1 mM PMSF ) and lysed by three freeze–thaw cycles . NaCl was adjusted to 400 mM and the whole-cell extract was cleared by centrifugation . TNT T7 coupled transcription/translation reactions were assembled according to manufacturer's instructions ( Promega , Madison , WI ) with 40 ng/μl TERT expression plasmid and 100 ng/μl purified in vitro transcribed hTRmin added prior to TERT synthesis ( Wu and Collins , 2014a ) . Reactions were incubated at 30°C for 3 . 5 hr . HEK 293T cell extracts ( 200 μl per precipitation ) or RRL reconstitution reactions ( 37 . 5 μl per precipitation ) were adjusted to 150 mM NaCl and bound to 10 μl FLAG M2 monoclonal antibody resin ( Sigma–Aldrich , St . Louis , MO ) , 10 μl c-Myc antibody resin ( Sigma–Aldrich ) or 10 μl streptavidin agarose resin ( Sigma–Aldrich ) coated with 5′-biotinylated template-antisense oligonucleotide ( CTAGACCTGTCATCAGUUAGGGUUAG , where the underlined nucleotides are 2′OMe RNA; [Schnapp et al . , 1998] ) by end-over-end rotation at room temperature for 2 hr . Following binding , the resin was washed three times at room temperature with HLB containing 150 mM NaCl , 0 . 1% Triton X-100 , and 0 . 2% CHAPS . Resin-bound telomerase was then used in activity assay reactions ( see below ) . Immunoblotting for TERT detection was performed using mouse anti-TERT polyclonal primary antibody 1A4 raised against the TERT C-terminus at 1:3000 dilution . FLAG was detected using mouse anti-FLAG monoclonal primary antibody M2 ( Sigma–Aldrich ) at 1:5000 dilution . Tubulin was detected using mouse anti-alpha-tubulin monoclonal primary antibody DM1A ( Calbiochem , Billerica , MA ) at 1:500 dilution . Biotin was detected using goat anti-biotin polyclonal primary antibody ab6643 ( Abcam , Cambridge , MA ) at 1:5000 dilution . Myc was detected using rabbit anti-c-Myc polyclonal primary antibody A-14 ( Santa Cruz Biotechnology , Dallas , TX ) at 1:3000 dilution . Immunoblots using mouse primary antibodies were detected with goat anti-mouse IR 800 secondary antibody ( Rockland Immunochemicals , Limerick , PA ) at 1:20 , 000 dilution . Immunoblots using goat primary antibodies were detected with donkey anti-goat Alexa Fluor dye 680 secondary antibody ( Life Technologies , Waltham , MA ) at 1:15 , 000 dilution . Immunoblots using rabbit primary antibodies were detected with goat anti-rabbit IR 800 secondary antibody ( Rockland Immunochemicals ) at 1:20 , 000 dilution . All incubations were performed in 3% non-fat milk in Tris-buffered saline ( TBS ) buffer ( 50 mM Tris pH 7 . 5 , 150 mM NaCl ) . Membranes were washed with TBS buffer prior to visualization on a LI-COR Odyssey imager ( LI-COR Biotechnology , Lincoln , NE ) . Primer extension assays with radiolabeled nucleotide incorporation were performed in 20 μl reactions containing 10 μl resin-bound telomerase , 500 nM ( T2AG3 ) 3 telomeric primer , and >0 . 1 μM α-32P dGTP ( 3000 Ci/mmol , 10 mCi/ml , Perkin–Elmer , Waltham , MA ) in telomerase activity assay buffer ( 50 mM Tris-acetate at pH 8 , 3 mM MgCl2 , 1 mM EGTA , 1 mM spermidine , 5 mM DTT , and 5% glycerol ) with 5 μM dGTP , 250 μM dTTP , and dATP for detection of repeat addition processive synthesis or 250 μM dTTP and 500 μM ddATP for detection of single-repeat synthesis . Reactions were incubated at 30°C for 40 min . For the 5′-end labeled primer extension pulse-chase assay , 10 μl resin-bound telomerase was incubated with 20 nM 32P 5′-end labeled ( T2AG3 ) 3 telomeric primer for 30 min , then washed twice with HLB containing 150 mM NaCl and 0 . 1% NP-40 to remove unbound primer . The assay was initiated by addition of 20 μl of telomerase activity assay buffer with 250 μM dGTP , dTTP , and dATP . The reactions were incubated at 30°C for 5 min followed by addition of unlabeled ( T2AG3 ) 3 telomeric primer to a final concentration of 5 μM and further incubated at 30°C to reach the indicated total reaction time . The products of all activity assay reactions were then extracted , precipitated , and resolved on 12% polyacrylamide/7 M urea/0 . 6× Tris borate-EDTA gels . An end-labeled oligonucleotide was added prior to product precipitation to serve as a recovery control , and end-radiolabeled primer was loaded separately from product DNA as a size marker ( migration is indicated in Figures by ▶ ) . Dried gels were visualized by phosphorimaging on a Typhoon Trio system ( GE Healthcare , Piscataway , NJ ) and quantified using ImageQuant TL ( GE Healthcare ) . Activity was quantified on the combined intensities of all product DNA . Complexes bound to an affinity purification resin were washed into 50 mM HEPES at pH 8 , 1 mM DTT , and 10 mM MgCl2 . CoA-conjugated biotin was purchased ( New England Biolabs , Ipswich , MA ) and CoA-conjugated Cy3 or Cy5 was prepared as described ( Yin et al . , 2006 ) and added to a final concentration of 10 μM . Labeling reactions were carried out by addition of ACP or SFP synthase ( New England Biolabs ) to 1 μM final concentration and incubation at room temperature for 2 hr . Following the labeling reaction , the resin was washed three times at room temperature with HLB containing 150 mM NaCl , 0 . 1% Triton X-100 , and 0 . 2% CHAPS . For samples sequentially labeled with two dyes , the labeling reactions were repeated with the second dye and synthase . After the final labeling reaction and wash , complexes were eluted by incubation with 200 nM FLAG peptide or 30 μM 3′-terminal 2′ , 3′-dideoxyguanosine-modified displacement oligonucleotide ( CTAACCCTAACTGATGACAGGTCTAG; [Schnapp et al . , 1998] ) for 1 hr at room temperature . Complexes bound to FLAG antibody or 2′OMe RNA oligonucleotide resin were eluted in 14 μl or 70 μl buffer , respectively . These volumes were required to normalize activity and fluorescent spot count among preparations from the same amount of input . Labeled bulk samples were analyzed by 10% SDS-PAGE and imaged on a Typhoon Trio system ( GE Healthcare ) . Telomerase was reconstituted with ACP- and/or MCP-TERT as described above , with the RRL reaction supplemented with 35S-methionine . Following FLAG purification , complexes were labeled with CoA or CoA-biotin with ACP or SFP synthase . Samples were eluted from the affinity purification resin with 200 nM FLAG peptide and bound to streptavidin agarose or Myc antibody agarose ( Sigma–Aldrich ) for 1 hr . For depletion in denaturing conditions , samples were eluted from affinity purification resin in buffer adjusted to 2 M urea . The streptavidin-agarose unbound fraction was analyzed by 10% SDS-PAGE or activity assay . A prism-type total internal reflection fluorescence microscope was built using a Nikon Ti-E Eclipse inverted fluorescence microscope equipped with a 60× 1 . 20 N . A . Plan Apo water objective ( Nikon Instruments , Melville , NY ) . A 532-nm laser ( Coherent , Inc . , Santa Clara , CA , 350 mW ) was used for Cy3 excitation , and a 633-nm laser ( JDSU , Milpitas , CA , 35 mW ) was used for Cy5 excitation . For two-color co-localization experiments , Cy3 and Cy5 fluorescence were split into two channels and imaged separately on a single charge-coupled device ( CCD ) chip using an Optosplit II image splitter ( Cairn Instruments , Faversham , UK ) . Fluorescence signal was collected with a 512 × 512 pixel electron-multiplied CCD camera ( Andor Technology , Belfast , UK ) . All data collection was conducted at 22°C . Quartz coverslips were coated with a mixture of 99% PEG and 1% biotinylated-PEG . Airtight sample chambers were constructed by sandwiching double-sided tape between the coverslips and quartz slides ( MicroSurfaces , Inc . , Englewood , NJ ) . To prepare the slides for molecule deposition , the surface was pre-blocked by sequential 15-min incubations with 20% Biolipidure 203/206 ( NOF Corporation , White Plains , NY ) and 10 mg/ml casein ( Sigma–Aldrich ) . Following each incubation , the sample chamber was washed with telomerase slide buffer ( 50 mM Tris-HCl at pH 8 , 10% glycerol , 2 mM MgCl2 , and 0 . 2 mM EGTA ) . The surface was then incubated with 1 mg/ml streptavidin ( Sigma–Aldrich ) for 10 min and washed twice with telomerase slide buffer . Streptavidin-coated slides were incubated with 40 nM 5′-biotinylated telomeric primer ( Tel2 , T15TTAGGGTTAGGG ) in telomerase activity assay buffer for 10 min and washed with telomerase slide buffer . The slide was then incubated for 30 min with 1 μl labeled telomerase supplemented with 1 mg/ml casein followed by two washes with telomerase slide buffer to remove excess unbound sample . After washing , imaging buffer ( 1 mg/ml glucose oxidase , 0 . 34 mg/ml catalase , 0 . 8% wt/vol D-glucose , and 2 mM Trolox in telomerase slide buffer ) was flowed into the sample chamber . The fraction of two-color co-localization was experimentally determined considering only complexes with Cy5 signal and measuring the percentage of the spots that also had Cy3 signal . This was done because initial Cy5 labeling of the ACP tag by ACP synthase is selective for ACP vs MCP tag , whereas the subsequent SFP synthase labeling used to add Cy3 can label both MCP and ACP tags . By only considering complexes that labeled with Cy5 , we avoided the possibility of counting two-TERT single-color Cy3 labeled complexes as TERT monomers rather than dimers . Samples were excited with the 633-nm laser throughout the experiment and imaged at 100-ms time resolution . After the first 10–20 frames , samples were excited with the 532-nm laser for ∼20 additional frames . For photobleaching , the 633-nm laser was used for excitation and 500–1000 frames were collected at 100-ms time resolution . Tel2-bound slides were incubated with 1 μl Cy5-labeled telomerase in telomerase activity assay buffer for 30 min , and then washed twice . Antisense Tel2 oligonucleotide ( Anti-Tel2 , CCCTAACCCTAA ) was then introduced at 100 nM final concentration and incubated for 15 min to block any unbound immobilized Tel2 . The slide was washed twice , and imaging buffer was flowed into the sample chamber . The samples were excited at 633 nm to collect 30 frames at 100-ms time resolution to determine the initial number of complexes bound to immobilized Tel2 . For assays of elution , after initial imaging , the slide was washed and incubated with either 20 μl dNTP elution buffer ( 10 nM Anti-Tel2 , 500 μM dATP , and 500 μM dTTP in telomerase activity assay buffer ) or mock elution buffer ( 10 nM Anti-Tel2 in telomerase activity assay buffer ) . After 15 and 30 min , the slide was then washed with telomerase activity assay buffer , imaging buffer was flowed into the imaging chamber , and the remaining number of bound complexes was determined by collecting 30 frames at 100-ms time resolution with 633-nm excitation . For photobleaching step quantification after elution , no initial imaging or imaging at 15 min was performed . For quantification of RNase sensitivity , Cy5-labeled MCP-TERT reconstitutions were pre-incubated with 0 . 1 mg/ml RNase A at room temperature for 1 hr immediately prior to introduction to the flow cell . Streptavidin-coated slides were incubated with 1 μl biotin-labeled sample diluted in telomerase activity assay buffer for 10 min . The sample chamber was washed with telomerase slide buffer and incubated with 500 nM non-specific blocking oligonucleotide ( AAATGATAACCATCTCGC ) for 15 min , followed by two washes with telomerase slide buffer . Telomeric oligonucleotide ( TTAGGGTTAGGG ) 5′-end labeled with Cy5 was incubated for 15 min . Excess DNA was washed away and imaging buffer was flowed into the sample chamber . Bound DNA was detected by collecting 30 frames at 100-ms time resolution with 633-nm excitation . RNA was purified using TRIzol reagent ( Life Technologies ) and resuspended in 2 μl of water . The RNA was spotted onto Hybond N+ nylon membrane ( GE Healthcare ) and detected using 32P end-labeled probe complementary to hTR positions 51–72 ( Fu and Collins , 2003 ) . | Enzymes carry out the many diverse chemical reactions that support life . Some enzymes are made of just one component protein that works on its own , but others are made of multiple proteins that are all required for the enzyme to work properly . Most of what is understood about the activities of enzymes has been deduced by studying solutions containing many enzyme molecules . However , many enzymes can bind to different combinations of proteins to form groups ( or ‘complexes’ ) with a variety of three-dimensional shapes , so there may be a variety of enzyme complexes in the solution . This can lead to researchers drawing different conclusions about the same enzyme . In humans and other eukaryotic organisms , DNA is contained within structures called chromosomes . An enzyme called telomerase adds structures called telomeres to the ends of the chromosomes , which protect the DNA from damage . The center of telomerase has a protein called TERT that forms complexes with other proteins . However , it is not known how many copies of the TERT protein are present in each complex . Wu et al . studied these complexes using fluorescent tags that enabled each protein to be identified using a technique called ‘single-particle imaging’ . The experiments show that these complexes can contain either one or two TERT proteins . It had previously been suggested that TERT is only an active enzyme when it is bound to another TERT molecule , but Wu et al . show that even complexes with a single TERT are able to add telomeres to DNA . Further experiments used a mutant form of the TERT protein that cannot interact with other TERT molecules and found that complexes that contain this mutant protein still have normal enzyme activity . Large quantities of purified proteins were used in this study . Therefore , a future challenge will be to refine the method to allow experiments to use much less protein , which would more closely reflect how telomerase is produced in cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Single-molecule imaging of telomerase reverse transcriptase in human telomerase holoenzyme and minimal RNP complexes |
The great demand for long-wavelength and high signal-to-noise Ca2+ indicators has led us to develop CaRuby-Nano , a new functionalizable red calcium indicator with nanomolar affinity for use in cell biology and neuroscience research . In addition , we generated CaRuby-Nano dextran conjugates and an AM-ester variant for bulk loading of tissue . We tested the new indicator using in vitro and in vivo experiments demonstrating the high sensitivity of CaRuby-Nano as well as its power in dual color imaging experiments .
In recent years fluorescence imaging has been one of the fastest growing methods in physiology , cell biology and neuroscience , constantly driving the need for improved fluorescent probes ( Wilt et al . , 2009; Miyawaki , 2011; Looger and Griesbeck , 2012 ) . The dominance of fluorescein and eGFP in the design of such probes has resulted in an overcrowding of the green spectral band . This makes simultaneous imaging of spatially overlapping signals problematic and emphasizes the need for red-shifted probes ( Oheim et al . , 2014 ) . Many of the very favorable photophysical properties of fluorescein and eGFP are shared by the X-Rhodamine chromophore , which is finding increasing use in the development of Ca2+-indicators ( Eberhard and Erne , 1991; Egawa et al . , 2011 ) . A major problem of red-shifted fluorophores is that they are significantly more lipophilic than fluorescein-like dyes . This leads to more leakage through cell membranes as well as to intracellular compartmentalization . These effects can be minimized by using the probes as conjugates of inert hydrophilic compounds such as dextrans . This conjugation commonly uses one of the carboxylic groups of the BAPTA moiety , affecting the calcium affinity of BAPTA-based indicators ( Tsien , 1980 ) . Consequently , these indicator-dextran conjugates are strongly shifted to lower affinities making them all but useless for sensitive and quantitative [Ca2+] measurements . Importantly , this lower affinity cannot be compensated for by increasing the concentration of the probe . Such a strategy would lead to a disproportionally large increase in the calcium buffering capacity of the indicator ( Neher , 2005 ) , resulting in a stronger disruption of cellular signaling than for a low concentration of high affinity indicator ( Markram et al . , 1998 ) . We recently introduced a family of red emitting calcium indicators based on X-Rhodamine: Calcium Ruby ( CaRuby ) ( Collot et al . , 2012 ) , which bears an azido side arm for click chemistry and the resulting potential for high-yield coupling reactions ( Kolb et al . , 2001 ) . This side arm efficiently allows conjugation reactions without significant perturbation of the calcium binding affinity ( Zamaleeva et al . , 2014 ) . The dissociation constants of CaRubies ranged from 3 . 4 to 21 . 6 µM—too high for the reliable detection of small [Ca2+] transients in biological tissue ( Yasuda et al . , 2004 ) ( see Appendix 1 for details ) .
To increase the affinity of CaRuby , we modified the structure of the probe ( Figure 1A ) , focusing on the Ca2+ chelating BAPTA moiety , as increasing the electron density of BAPTA lowers its KD for calcium ( Tsien , 1980 ) . We introduced an oxygen atom on one of the aromatic rings of BAPTA by a SNAr reaction . This oxygen also serves as a link for the azido side arm , which was repositioned in the new CaRuby variant ( Figure 1—figure supplement 1 ) . Additionally , the fluorophore , which is commonly placed para to the nitrogen of the BAPTA , has an affinity-lowering effect due to its electron withdrawing nature and was therefore placed at a meta position in order to reduce its effect on the ligating nitrogen . These modifications resulted in a CaRuby variant with sub-micromolar affinity ( ‘CaRuby-Nano’ ) . In cuvette calibration experiments CaRuby-Nano was found to have a KD of 258 ± 8 nM , with a 50-fold ( ±2 ) increase of fluorescence on binding [Ca2+] ( Figure 1B–C ) and a maximum quantum yield of 0 . 45 ( Figure 1—figure supplement 2 , 3 ) . In addition to being suitable for single photon excitation , CaRuby-Nano also exhibits effective two-photon excitation over a large wavelength band ( Figure 1—figure supplement 4 ) . 10 . 7554/eLife . 05808 . 003Figure 1 . Chemical and photophysical properties of CaRuby-Nano . ( A ) Structure of CaRuby-Nano . Note the oxygen substituent and the positioning of the fluorophore-BAPTA bond . ( B ) [Ca2+]-dependent change in CaRuby-Nano fluorescence ( [Ca2+]free: 0 nM , 17 nM , 38 nM , 65 nM , 100 nM , 150 nM , 225 nM , 351 nM , 602 nM , 1 . 35 µM , 39 µM ) . ( C ) The titration curve corresponding to the spectra in ( B ) using the same color code . ( D–F ) : Climbing fiber evoked dendritic calcium signals in Purkinje cells in vitro . ( D ) Purkinje cell filled with 300 µM CaRuby-Nano dextran , with region of interest indicated by the white rectangle ( scale bar = 20 µm ) . ( E ) Region of interest with points of interest indicated . Note that many spines can be readily distinguished ( white arrow ) . Points 1–3 and 4–6 are on different spiny branchlets while points 7 and 8 are background ( scale bar = 5 µm ) . ( F ) Ca2+ transients following climbing fiber activation recorded at 2 . 8 kHz ( traces averaged over 26 trials and then averaged over the indicated spine numbers ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 00310 . 7554/eLife . 05808 . 004Figure 1—figure supplement 1 . Comparison of CaRuby structures . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 00410 . 7554/eLife . 05808 . 005Figure 1—figure supplement 2 . Absorption and emission spectra of CaRu-Nano . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 00510 . 7554/eLife . 05808 . 006Figure 1—figure supplement 3 . Determination of CaRuby-Nano fluorescence quantum yield . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 00610 . 7554/eLife . 05808 . 007Figure 1—figure supplement 4 . Two-photon excitation of CaRuby-Nano . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 00710 . 7554/eLife . 05808 . 008Figure 1—figure supplement 5 . Affinity of CaRuby-Nano and CaRuby-Nano 6 kD dextran . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 008 For verification of the new probe in biological tissue we used conjugates with 1 . 5 kD and 6 kD dextrans , which were obtained via click chemistry . As expected , this conjugation had only a small effect on the affinity of the indicator , increasing the KD from 258 nM in the free salt to 295 nM in the 6 kD dextran conjugate ( Figure 1—figure supplement 5 ) . We first tested if CaRuby-Nano performs comparably to commonly used green emitting [Ca2+] probes . For this , Purkinje cells in acute cerebellar brain slices were filled with CaRuby-Nano ( 1 . 5 kD dextran ) via a patch-clamp microelectrode ( Figure 1D ) . Climbing fiber stimulation evoked calcium signals were recorded from multiple spines ( range: 4 to 14 ) at acquisition rates between 2 . 2 and 5 . 0 kHz ( Figure 1E , F ) using random access two-photon microscopy ( Otsu et al . , 2008 ) . The rising phase time course ( 0 . 55 ms ± 0 . 13 ms; sigmoidal fit; n = 59 spines from 7 cells ) was not significantly different from that found for Fluo-5F ( 0 . 40 ± 0 . 09 ms , n = 36 spines from 4 cells , p = 0 . 37 ) under the same conditions , suggesting that CaRuby-Nano has binding kinetics comparable to established small molecule Ca2+ indicators . These fast kinetics point to a high sensitivity of CaRuby-Nano for small and fast changes in [Ca2+] , such as neuronal action potentials ( Otsu et al . , 2014 ) . Thus , we next tested the sensitivity of CaRuby-Nano using in vivo patch-clamp recordings from neocortical layer 2/3 pyramidal neurons in anesthetized mice with simultaneous two-photon [Ca2+] imaging ( Svoboda et al . , 1997 ) ( Figure 2A ) . We found that using CaRuby-Nano ( 6 kD dextran ) even single spikes resulted in reliable , easily detected fluorescence transients ( mean dR/R0 = 0 . 52 ± 0 . 19 , n = 6 cells; Figure 2B ) . For increasing spike numbers the dR/R0 vs spike number relation quickly turns sublinear and saturates as expected for high affinity indicators ( Figure 2C ) . Taken together these experiments demonstrate that CaRuby-Nano is a calcium indicator with a signal quality comparable to previously used high-affinity green emitting probes . Importantly , it is well suited for the detection of small [Ca2+] transients , setting it apart from the previous CaRuby versions . 10 . 7554/eLife . 05808 . 009Figure 2 . Spike evoked transients in layer 2/3 pyramidal neurons in vivo . ( A ) Measurement configuration ( left ) and maximum intensity projection of pyramidal neuron filled with 100 µM Alexa Fluor 488 and 200 µM CaRuby-Nano dextran ( right , at rest the fluorescence is dominated by the green dye ) . The red line indicates region imaged in line scan . Scale bar: 20 µm . ( B ) Single trial calcium signals evoked by increasing number of spikes . The corresponding membrane voltage traces are shown below . Fluorescence traces are aligned to spike onset and color-coded to match the number of APs . ( C ) The peak amplitudes ( red ) and the area under the curve ( blue ) of the fluorescence trace were plotted against the number of action potentials . While the area increases linearly , the peak amplitude saturates . The shaded regions indicate the corresponding standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 009 Having verified the suitability of CaRuby-Nano for single cell imaging experiments in vitro and in vivo , we now set out to test CaRuby-Nano for imaging neuronal network activity when applying sensory stimulation . In the past decade calcium population imaging has commonly been performed using bulk loading ( Stosiek et al . , 2003; Ohki et al . , 2005 ) of calcium indicators in the AM-ester form ( Tsien , 1981 ) . We thus synthesized an AM-ester of CaRuby-Nano and used it to load cerebellar neurons in vivo ( Figure 3A ) . We performed a series of three experiments . In all cases we found labeling identical to that commonly found in experiments using Oregon Green-488 BAPTA-1 AM ( OGB-1 AM ) to load cerebellar tissue in vivo ( Figure 3B ) ( Sullivan et al . , 2005; Ozden et al . , 2009; Schultz et al . , 2009 ) . In all experiments fluorescence traces extracted from identified Purkinje cell dendrites ( Figure 3C ) showed clear complex spike activity with a good signal-to-noise ratio ( Figure 3D , F ) . Both spontaneous activity and sensory evoked responses were again comparable to signals detected in experiments using OGB-1 AM ( Sullivan et al . , 2005; Ozden et al . , 2009; Schultz et al . , 2009 ) . These results indicate that CaRuby-Nano AM is a powerful addition to the optophysiological toolbox . 10 . 7554/eLife . 05808 . 010Figure 3 . Imaging cerebellar Purkinje cells in vivo using bulk loading of CaRuby-Nano AM . ( A ) Configuration of AM-ester injection and imaging . ( B ) Resulting staining of tissue 60 min after injection of indicator . Purkinje cells can be seen as vertical stripes with occasional brighter spots ( corresponding to dendrites; scale bar = 20 µm ) . ( C ) Active Purkinje cell dendrites identified using a spatial PCA/ICA algorithm ( Ozden et al . , 2009 ) . ( D ) Fluorescence traces from the identified dendrites , using the same color code as in ( C ) . The timing of the sensory stimulus ( foot shock ) is indicated by the underlying grey bars . ( E ) Stimulus triggered averages of 20 stimulus presentations . Note that all cells except for the third ( orange ) show a stimulus-locked response . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 010 To demonstrate the full power of CaRuby-Nano , we made use of the strong two-photon excitation spectral overlap with eGFP to conduct a set of experiments which were not possible previously: simultaneous imaging of glutamate release onto Purkinje cells using iGluSnFR ( a single-wavelength extracellular glutamate indicator constructed from the bacterial glutamate sensor Gltl and circularly permutated GFP [Marvin et al . , 2013] ) and the resulting post-synaptic [Ca2+] increase ( using CaRuby-Nano ) . Visually identified Purkinje cells showing iGluSnFR expression ( 7–9 days after viral transfection ) were filled with CaRuby-Nano via patch-clamp recording ( 6 kD dextran; Figure 4A ) . Activation of the glutamatergic climbing fiber input evoked clear fluorescence transients in both color channels ( Figure 4B ) . Glutamate signals were confined to distinct subsections of the dendritic tree ( i . e . , limited to sites of synaptic glutamate release ) , whereas the resulting [Ca2+] transients were global , with similar amplitudes throughout different regions of the dendritic tree ( Figure 4C ) ( Lev-Ram et al . , 1992 ) . The differential spatial distribution of the signals confirms that the two indicators can be spectrally isolated . 10 . 7554/eLife . 05808 . 011Figure 4 . Dual color functional imaging in vitro and in vivo . ( A–C ) Combined imaging of [glutamate] and [Ca2+] in vitro ( A ) A Purkinje cell expressing iGluSnFR was filled with 200 µM CaRuby-Nano dextran . The image shows the basal fluorescence of CaRuby-Nano ( scale bar = 5 µm ) . ( B ) Double pulse stimulation of the climbing fiber triggers spatially distinct patterns of glutamate release and Ca2+ influx ( maximum dF/F0 images; the inset at the top shows the two evoked complex spikes ) . Note the breaks between regions showing iGluSnFR activation ( indicated by white arrows ) ( C ) Fluorescence traces for CaRuby-Nano ( red ) and iGluSnFR ( green ) following single pulse climbing fiber stimulation ( top inset ) . Traces were extracted from the corresponding regions outlined in white in ( B ) . Note the absence of a fluorescent transient for iGluSnFR in the ‘Ca2+ only’ region . ( D–F ) Odor-evoked calcium responses in olfactory bulb glomeruli in vivo . ( D ) Juxtaglomerular neurons and mitral cell dendritic tufts expressing YFP demarcate glomeruli in a Kv3 . 1-eYFP mouse ( Metzger et al , 2002 ) . ( E ) Olfactory sensory neuron glutamatergic terminals , labeled with CaRuby-Nano dextran , clearly filled the inner boundaries of most glomeruli . ( F ) A 3 s application of 30% isoamyl acetate reliably triggered presynaptic calcium responses in several glomeruli . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 011 To demonstrate that dual color imaging is also possible in vivo we used CaRuby-Nano ( 6 kD dextran ) to report presynaptic activity in anesthetized Kv3 . 1-eYFP adult mice ( Metzger et al . , 2002 ) . In the olfactory bulb of these mice , mitral and tufted cells , as well as a population of periglomerular neurons , strongly express eYFP and their somata and processes clearly demarcate the external glomerular boundaries ( Figure 4D ) . Olfactory sensory neuron ( OSN ) terminals , labeled with CaRuby-Nano , filled the inner glomerular boundaries ( Figure 4E ) . In single glomeruli ( n = 8 animals ) we could record presynaptic calcium responses with excellent signal to noise ratio . Figure 4F shows a typical example in which presynaptic calcium responses were selectively evoked by odor presentation in a subset of glomeruli . These responses adapted strongly at this high odorant concentration , as reported previously ( Lecoq et al . , 2009 ) . Taken together , these last two experiments demonstrate the potential of two-channel functional imaging—both in vitro and in vivo—with the red emission and high sensitivity of CaRuby-Nano being an ideal match for numerous other indicators emitting in the green-yellow spectral band .
We have developed a high-affinity red-emitting calcium indicator . This novel indicator , CaRuby-Nano , has a KD of 295 nM ( for the dextran conjugate ) . This makes CaRuby-Nano only slightly higher affinity than the commonly used green emitting indicator Fluo-4 ( 335 nM ) . On calcium binding CaRuby-Nano shows a 50-fold fluorescence increase ( vs . 14-fold and 100-fold for OGB-1 and Fluo-4 , respectively ) . The quantum efficiency of calcium bound CaRuby-Nano ( 0 . 45 ) is lower than that of OGB-1 ( ∼0 . 7 ) but significantly higher than that of Fluo-4 ( ∼0 . 14 ) . These values classify CaRuby-Nano as an ideal indicator for the quantification of small intracellular [Ca2+] transients ( see Appendix 1 ) . Using a range of different experiments we have demonstrated that CaRuby-Nano is well suited for both in vitro and in vivo imaging experiments requiring high sensitivity to [Ca2+] changes . Finally , we show that CaRuby-Nano can be combined with activity indicators emitting in the green-yellow spectral band , to allow multiplexed imaging . The versatility of the probe is further increased by the azido function , which can easily be reduced to an amine group , thus opening the field to functionalization with numerous molecular tools such as antibodies , benzylguanine ( SNAP tag ) or peptides to facilitate specific sub-cellular targeting . CaRuby-Nano's sensitivity and potential for spectral multiplexing allows sophisticated experiments such as simultaneously measuring pre- and postsynaptic activity or imaging of different signaling modalities in the same cell , allowing previously elusive questions to be directly addressed .
All the solvents were of analytical grade . Chemicals were purchased from commercial sources . 1H-NMR and 13C-NMR were measured on a Bruker Avance III-300 MHz spectrometer ( Bruker Biospin , The Woodlands , TX , USA ) with chemical shifts reported in ppm ( TMS as internal standard ) . Mass spectra were measured on a Focus GC/DSQ II spectrometer ( ThermoScientific , Waltham , MA , USA ) for IC and an API 3000 spectrometer ( Applied Biosystems , PE Sciex ) for ES . All pH measurements were made with a Mettler Toledo pH meter . Fluorescence spectra were recorded on a JASCO FP-8300 spectrofluorometer ( JASCO , Easton , MD , USA ) . Absorption spectra were determined on a VARIAN CARY 300 Bio UV-Visible spectrophotometer . All measurements were done at a temperature of 25°C . The purity of the dyes were checked by RP-HPLC C-18 , elutant: ACN 0 . 1% TFA/Water 0 . 1% TFA , method: 20/80 to 100/0 within 20 min then 100/0 for 10 min detection at λAbs = 254 nm . The apparent dissociation constant for calcium ( KD Ca2+ ) was measured with a calcium calibration buffer kit from Invitrogen ( Lifetechnologies , USA ) . All mass spectra , NMR spectra and chromatograms are included as supplemental data . In order to develop a high affinity CaRuby , three modifications were carried out based on the first generation CaRubies . First , an oxygen atom was introduced on one of the BAPTA's cycles in order to electronically enrich the latter . Then , this oxygen atom served as an anchor to a spacer terminated by an azide function for further functionalizations either by click chemistry or by reducing it into an amine for coupling with for example , a carboxylic acid . Finally , the fluorophore moiety , an extended rhodamine which is positively charged and therefore has an electron withdrawing effect , was moved from the para position of the aniline to the meta position . As expected , these modifications lead to a significant increase of affinity towards calcium , yielding a CaRuby with a dissociation constant of 258 ± 8 nM . The synthesis pathway is displayed in Figure 5 along with the compound numbering . The NMR and mass spectra for both intermediate compounds and final products are contained in Supplementary file 1 . 10 . 7554/eLife . 05808 . 012Figure 5 . Synthesis of CaRuby-Nano and CaRuby-Nano AM ester . DOI: http://dx . doi . org/10 . 7554/eLife . 05808 . 012 To a solution of de 5-fluoro-2-nitrophenol ( 14 . 90 g , 94 . 84 mmol ) in DMF ( 75 ml ) were added dibromoethane ( 40 . 90 ml , 472 . 2 mmol , 5 eq ) and K2CO3 ( 26 . 30 g , 189 . 7 mmol , 2 eq ) , the mixture was allowed to stir at 70°C for 2 hr . The solvents were evaporated and the product was extracted with EtOAc washed with water ( three times ) and brine ( two times ) . The organic phase was dried over MgSO4 , filtered and evaporated to reach a volume of 200 ml . The symmetric dinitro compound crystallizes first and was filtered off . The filtrate was then allowed to crystallize to obtain 20 . 12 g of 1 ( 80% ) as a yellow powder . 1H-NMR ( 300 MHz , DMSO-d6 ) : δ 8 . 04 ( dd , Ja-b = 9 . 1 Hz , Ja-F = 6 . 1 Hz , 1H , Ha ) , 7 . 37 ( dd , Jc-F = 11 . 0 Hz , Jc-b = 2 . 6 Hz , 1H , Hc ) , 7 . 02 ( ddd , Jb-a = 9 . 1 , Jb-F = 7 . 8 Hz , Jb-c = 2 . 6 Hz , 1H , Hb ) , 4 . 56–4 . 53 ( m , 2H , CH2O ) , 3 . 84−3 . 81 ( m , 2H , CH2Br ) . 13C-NMR ( 75 MHz , DMSO-d6 ) : δ 164 . 82 ( d , 1JF-C = 251 Hz , CF ) , 152 . 81 ( d , 3JC-F = 12 Hz , CO ) , 136 . 17 ( d , 4JF-C = 3 Hz , CNO2 ) , 127 . 62 ( d , 3JF-C = 11 Hz , Ca ) , 108 . 01 ( d , 2JF-C = 23 Hz , Cb ) , 103 . 45 ( d , 2JF-C = 27 Hz , Cc ) , 69 . 78 ( CH2O ) , 30 . 39 ( CH2Br ) . MS ( CI ) , calculated for C8H11BrFN2O3 [M + NH4]+ 280 . 9 , found 281 . 0 . To a solution of 1 ( 19 . 79 g , 74 . 96 mmol ) in DMF ( 75 ml ) were added 2-nitrophenol ( 11 . 46 g , 82 . 45 mmol , 1 . 1 eq ) and K2CO3 ( 15 . 63 g , 112 . 4 mmol , 1 . 5 eq ) , the mixture was allowed to stir overnight at 70°C . The solvent was evaporated and the product was extracted with DCM , washed with HCl ( 1 M ) and brine ( 2 times ) . The organic phase was dried over MgSO4 , filtered and evaporated to reach a volume of 200 ml . The product crystallized and was filtered to obtain 12 . 00 g of 2 ( 50% ) as a yellow powder . 1H-NMR ( 300 MHz , DMSO-d6 ) : δ 8 . 01 ( dd , Ja-b = 9 . 1 Hz , Ja-F = 6 . 1 Hz , 1H , Ha ) , 7 . 86 ( dd , Jg-f = 8 . 1 Hz , Jg-e = 1 . 6 Hz , 1H , Hg ) , 7 . 67 ( ddd , 3J = 8 . 5 , 7 . 4 , 4Je-g = 1 . 7 Hz , 1H , He ) , 7 . 45-7 . 39 ( m , 2H , Hc , Hd ) , 7 . 15 ( ddd , 3J = 8 . 1 Hz , 7 . 4 Hz , 4J = 1 . 1 Hz , 1H , Hf ) , 7 . 01 ( ddd Jb-a = 9 . 1 , Jb-F = 7 . 8 Hz , Jb-c = 2 . 6 Hz , 1H , Hb ) , 4 . 59-4 . 54 ( m , 4H , 2CH2O ) . 13C-NMR ( 75 MHz , DMSO-d6 ) : δ 164 . 86 ( d , 1JF-C = 251 Hz , CF ) , 153 . 34 ( d , 3JC-F = 11 . 9 Hz , CO ) , 150 . 82 ( Cq Ar ) , 139 . 74 ( Cq Ar ) , 136 . 15 ( d , 4JF-C = 3 . 7 Hz , CNO2 ) , 134 . 33 ( Ce ) , 127 . 58 ( d , 3JF-C = 11 Hz , Ca ) , 124 . 85 ( Cg ) , 121 . 05 ( Cf ) , 115 . 55 ( Cd ) , 107 . 90 ( d , 2JF-C = 24 Hz , Cb ) , 103 . 57 ( d , 2JF-C = 27 . 7 Hz , Cc ) , 68 . 57 ( CH2O ) , 67 . 93 ( CH2O ) . MS ( ES+ ) , calcd for C14H11FN2O6Na [M + Na]+ 345 . 0 , found 345 . 3 . HRMS ( ES+ ) , calcd for C14H11FN2O6Na [M + Na]+ 345 . 0493 , found 345 . 0501 . To a stirred solution of 2 ( 5 . 91 g , 18 . 34 mmol ) in DMSO ( 53 ml ) was added NaOH 20% ( 11 . 5 ml ) the solution turned yellow and was allowed to stir at room temperature overnight . 50 ml of water and 10 ml HCl ( 1 M ) were then added and the product was extracted three times with EtOAc . The organic phase was washed three times with water before being dried over MgSO4 , the solution was filtered and evaporated and crystallized in EtOAc to obtain 4 . 46 g of 3 ( 76% ) as a yellow powder . 1H-NMR ( 300 MHz , DMSO-d6 ) : δ 7 . 90–7 . 85 ( m , 2H , Ha , Hg ) , 7 . 67–7 . 64 ( dd , 3J = 8 . 7 Hz , 4J = 1 . 5 Hz , 1H , He ) , 7 . 48 ( d , 3J = 8 . 4 Hz , 1H , Hd ) , 7 . 16 ( t , 3J = 7 . 7 Hz , 1H , Hf ) , 6 . 66 ( d , 4J = 2 . 2 Hz , 1H , Hc ) , 6 . 51 ( dd , 3J = 9 . 0 , 4J = 2 . 2 Hz , 1H , Hb ) , 4 . 55–4 . 54 ( m , 2H , CH2O ) , 4 . 45 ( t , J = 3 . 8 Hz , 2H , CH2O ) . 13C-NMR ( 75 MHz , DMSO-d6 ) : δ 163 . 87 ( Cq Ar ) , 154 . 51 ( Cq Ar ) , 150 . 98 ( Cq Ar ) , 139 . 79 ( Cq Ar ) , 134 . 36 ( Ce ) , 131 . 12 ( Cq Ar ) , 128 . 18 ( Ca or Cg ) , 124 . 86 ( Ca or Cg ) , 121 . 03 ( Cf ) , 115 . 75 ( Cd ) , 108 . 01 ( Cb ) , 101 . 56 ( Cc ) , 68 . 06 ( CH2O ) , 67 . 87 ( CH2O ) . MS ( CI ) , calcd for C14H16N3O7 [M + NH4]+ 338 . 0 , found 337 . 7 . HRMS ( ES+ ) , calcd for C14H13N2O7 [M + H]+ 321 . 0717 , found 321 . 0722 . To a solution of 3 ( 4 . 86 g , 15 . 19 mmol ) in DMF ( 50 ml ) were added dibromohexane ( 11 . 12 ml , 45 . 56 mmol , 3 eq ) and K2CO3 ( 3 . 16 g , 22 . 78 mmol , 1 . 5 eq ) . The mixture was allowed to stir at 70°C for 12 hr . The solvents were evaporated and the product was extracted with EtOAc washed with water ( three times ) and brine ( two times ) . The organic phase was dried over MgSO4 , filtered and evaporated . The crude was purified by column chromatography on silica gel ( Cyclohexane/EtOAc: 7/3 ) to obtain the crude 4 which was crystallized in a mixture of EtOAc and cyclohexane ( 3/7 ) to obtain 2 . 97 g of pure 4 ( 40% ) as a off white powder . Rf = 0 . 22 ( Cyclohexane/EtOAc , 7/3 ) . 1H-NMR ( 300 MHz , CDCl3 ) : δ 8 . 00 ( d , J = 9 . 1 Hz , 1H , Ha ) , 7 . 86 ( dd , J = 8 . 1 , 1 . 6 Hz , 1H , Hg ) , 7 . 64–7 . 58 ( m , 1H , He ) , 7 . 33 ( dd ( in solvent pick ) , 1H , Hd ) , 7 . 14–7 . 09 ( m , 1H , Hf ) , 6 . 66 ( d , J = 2 . 4 Hz , 1H , Hc ) , 6 . 57 ( dd , J = 9 . 1 , 2 . 4 Hz , 1H , Hb ) , 4 . 60–4 . 51 ( m , 4H , 2CH2O ) , 4 . 08 ( t , J = 6 . 4 Hz , 2H , CH2O ) , 3 . 47 ( t , J = 6 . 7 Hz , 2H , CH2Br ) , 1 . 96–1 . 85 ( m , 4H , 2CH2 ) , 1 . 56 ( dt , J = 7 . 1 , 3 . 5 Hz , 4H , 2CH2 ) . 13C-NMR ( 75 MHz , CDCl3 ) : δ 164 . 35 ( Cq ) , 154 . 63 ( Cq ) , 151 . 96 ( Cq ) , 140 . 43 ( Cq ) , 134 . 37 ( Ce ) , 133 . 34 ( Cq ) , 128 . 32 ( Ca ) , 125 . 56 ( Cg ) , 121 . 43 ( Cf ) , 116 . 14 ( Cd ) , 106 . 89 ( Cb ) , 102 . 02 ( Cc ) , 68 . 83 ( CH2O ) , 68 . 70 ( CH2O ) , 68 . 65 ( CH2O ) , 33 . 81 ( CH2Br ) , 32 . 63 ( CH2 ) , 28 . 85 ( CH2 ) , 27 . 86 ( CH2 ) , 25 . 19 ( CH2 ) . MS ( ES+ ) , calcd for C20H23BrN2O7Na [M + Na]+ 505 . 0 , found 505 . 5 . HRMS ( ES+ ) , calcd for C20H24BrN2O7 [M + H]+ 483 . 0767 , found 483 . 0772 . To a solution of 4 ( 5 . 00 g , 10 . 35 mmol ) in EtOAc ( 100 ml ) and methanol ( 30 ml ) was added Pd/C ( 1 . 10 g ) . The solution was stirred and degassed before H2 was allowed to bubble in the solution for 5 hr . The solution was then filtered off celite and rinsed with EtOAc under an atmosphere of argon . The solvents were evaporated and the crude was dissolved in acetonitrile ( 50 ml ) , to this solution were added , methyl bromoacetate ( 12 . 0 ml , 124 . 2 mmol , 12 eq ) and DIEA ( 23 . 0 ml , 124 . 2 mmol , 12 eq ) before being warmed up to 80°C . The solution was allowed to stir overnight at 80°C . The solvents were evaporated , the product was extracted with DCM and washed with water . The organic phase was dried over MgSO4 , filtered and evaporated . The crude was purified by column chromatography on silica gel ( Cyclohexane/EtOAc: 7/3 ) to obtain 3 . 71 g of 5 ( 50% ) as a yellowish syrup containing impurities ( between 2 and 3 ppm in 1H NMR ) that could not be removed . Rf = 0 . 51 ( Cyclohexane/EtOAc , 6/4 ) . 1H-NMR ( 300 MHz , CDCl3 ) : δ 6 . 85–6 . 74 ( m , 5H ) , 6 . 39 ( d , J = 2 . 7 Hz , 1H ) , 6 . 31 ( dd , J = 8 . 7 , 2 . 7 Hz , 1H ) , 4 . 20 ( m , 4H , CH2O ) , 4 . 08 ( s , 4H , 2CH2N ) , 4 . 02 ( s , 4H , 2CH2N ) , 3 . 81 ( t , J = 6 . 4 Hz , 2H , CH2O ) , 3 . 50 ( d , J = 7 . 4 Hz , 12H , 4 OMe ) , 3 . 36 ( t , J = 6 . 8 Hz , 2H , CH2Br ) , 1 . 85–1 . 80 ( m , 2H , CH2 ) , 1 . 71-1 . 67 ( m , 2H , CH2 ) , 1 . 42 ( t , J = 3 . 6 Hz , 4H , 2 CH2 ) . MS ( ES+ ) , calcd for C32H43BrN2O11Na [M + Na]+ 735 . 2 , found 735 . 8 . To a solution of 5 ( 3 . 71 g , 5 . 218 mmol ) in DMF ( 10 ml ) was added NaN3 ( 1 . 02 g , 15 . 65 mmol , 3 eq ) . The solution was stirred at 80°C overnight . The product was extracted with EtOAc and washed with water ( three times ) and brine ( two times ) , the organic phase was dried over MgSO4 , filtered and evaporated to give 3 . 52 g of 6 ( quant ) as a yellowish syrup . 1H-NMR ( 300 MHz , CDCl3 ) : δ 6 . 84–6 . 73 ( m , 5H ) , 6 . 39 ( d , J = 2 . 5 Hz , 1H ) , 6 . 31 ( dd , J = 8 . 7 , 2 . 5 Hz , 1H ) , 4 . 19 ( d , 4H , CH2O ) , 4 . 08 ( s , 4H , 2CH2N ) , 4 . 01 ( s , 4H , 2CH2N ) , 3 . 81 ( t , J = 6 . 4 Hz , 2H , CH2O ) , 3 . 48 ( d , J = 7 . 3 Hz , 12H , 4 OMe ) , 3 . 20 ( t , J = 6 . 8 Hz , 2H , CH2N3 ) , 1 . 70–1 . 64 ( m , 2H , CH2 ) , 1 . 60–1 . 51 ( m , 2H , CH2 ) , 1 . 38 ( m , 4H , 2 CH2 ) . Impurities between 2 and 3 ppm could not be removed . MS ( ES+ ) , calcd for C32H44N5O11 [M + H]+ 674 . 3 , found 674 . 3 . HRMS ( ES+ ) , calcd for C32H44N5O11 [M + H]+ 674 . 3032 , found 674 . 3054 . To a solution of 6 ( 1 . 22 g , 1 . 81 mmol ) in DMF ( 5 ml ) was added POCl3 ( 1 . 35 ml , 14 . 48 mmol , 8 eq ) dropwise without cooling . After addition the solution was allowed to stir for 40 min and then water ( 50 ml ) was added followed by slow addition of a saturated solution of NaHCO3 to reach a pH of 8 . The product was extracted with DCM and washed twice with brine before being dried over MgSO4 filtrated and evaporated . The crude was purified by column chromatography on silica gel ( Cyclohexane/EtOAc: 6/4 ) to obtain 505 mg of 7 ( 40% ) as a yellow syrup . Rf = 0 . 25 ( Cyclohexane/EtOAc , 5/5 ) . 1H-NMR ( 300 MHz , CDCl3 ) : δ 10 . 23 ( s , 1H , CHO ) , 7 . 27 ( s , 1H , Ha ) , 6 . 86–6 . 75 ( m , 4H , Hd , He , Hf , Hg ) , 6 . 39 ( s , 1H , Hc ) , 4 . 26 ( d , J = 2 . 4 Hz , 4H , 2CH2O ) , 4 . 06 ( d , J = 3 . 3 Hz , 4H , 2CH2N ) , 4 . 02 ( d , J = 5 . 8 Hz , 4H , 2CH2N ) , 3 . 96 ( t , J = 6 . 3 Hz , 2H , CH2O ) , 3 . 49 ( 2 s , 12H , 2 OMe ) , 3 . 22 ( t , J = 6 . 8 Hz , 2H , CH2N3 ) , 1 . 77 ( t , J = 7 . 1 Hz , 2H , CH2 ) , 1 . 57 ( t , J = 7 . 0 Hz , 2H , CH2 ) , 1 . 45–1 . 35 ( m , 4H , CH2 ) . 13C-NMR ( 75 MHz , CDCl3 ) : δ 187 . 99 ( CHO ) , 171 . 94 ( COOMe ) , 171 . 64 ( COOMe ) , 158 . 88 ( Cq Ar ) , 157 . 35 ( Cq Ar ) , 150 . 21 ( Cq Ar ) , 139 . 40 ( Cq Ar ) , 133 . 40 ( Cq Ar ) , 122 . 44 ( CH Ar ) , 121 . 86 ( CH Ar ) , 119 . 19 ( CH Ar ) , 118 . 28 ( Cq Ar ) , 118 . 07 ( Ca ) , 113 . 45 ( CH Ar ) , 97 . 79 ( Cc ) , 68 . 93 ( CH2O ) , 67 . 53 ( CH2O ) , 66 . 81 ( CH2O ) , 53 . 36 ( 2CH2N ) , 53 . 32 ( 2CH2N ) , 51 . 69 ( OMe ) , 51 . 65 ( OMe ) , 51 . 35 ( CH2N3 ) , 30 . 19 ( CH2 ) , 29 . 07 ( CH2 ) , 28 . 82 ( CH2 ) , 26 . 92 ( CH2 ) , 26 . 50 ( CH2 ) , 25 . 70 ( CH2 ) . MS ( ES+ ) , calcd for C33H44N5O12 [M + H]+ 702 . 3 , found 702 . 2 . HRMS ( ES+ ) , calcd for C33H44N5O12 [M + H]+ 702 . 2981 , found 702 . 3008 . The position of the carbonyl was confirmed by further NMR investigations using a Heteronuclear Multiple Bond Correlation ( Supplementary file 2 ) . To a solution of aldehyde 7 ( 300 mg , 0 . 428 mmol ) in propionic acid ( 5 ml ) was added 8-hydroxyjulolidine ( 161 mg , 0 . 856 mmol , 2 eq ) and PTSA ( 8 mg , 0 . 042 mmol , 0 . 1 eq ) . The solution was protected from light and stirred at room temperature overnight . To the brown mixture was added a solution of chloranil ( 103 mg , 0 . 428 mmol , 1 eq ) in DCM ( 10 ml ) , the reaction turned dark and was allowed to stir overnight at room temperature . The dark purple solution was evaporated to dryness . The crude was purified by column chromatography on silica gel ( gradient of 100% DCM to 9/1 DCM/Methanol ) to obtain 130 mg of 8 ( 30% ) as a purple solid after lyophilisation ( dioxane/water: 1/1 ) . Rf = 0 . 32 ( DCM/MeOH , 9/1 ) . 1H-NMR ( 300 MHz , CDCl3 ) : δ 7 . 84 ( d , J = 8 . 1 Hz , 1H , H Ar ) , 7 . 06 ( d , J = 7 . 9 , 1H , H Ar ) , 6 . 97–6 . 86 ( m , 5H , H Ar , H7 ) , 6 . 71 ( d , J = 2 . 9 Hz , 1H , H Ar ) , 4 . 47–4 . 40 ( m , 4H , CH2O ) , 4 . 21 ( s , 4H , NCH2COOMe ) , 4 . 11 ( s , 4H , NCH2COOMe ) , 3 . 87 ( t , J = 6 . 1 Hz , 2H , CH2O ) , 3 . 67 ( s , 6H , 2 OMe ) , 3 . 56 ( m , 14H , 2 OMe , H1 , H4 ) , 3 . 11 ( d , J = 7 . 0 Hz , 2H , CH2N3 ) , 3 . 04 ( t , J = 6 . 3 Hz , 4H , H6 ) , 2 . 75 ( q , J = 6 . 2 Hz , 4H , H3 ) , 2 . 13–2 . 10 ( m , 4H , H5 ) , 2 . 00 ( t , J = 5 . 5 Hz , 4H , H2 ) , 1 . 49–1 . 34 ( m , 4H , CH2 ) , 1 . 19–1 . 03 ( m , 4H , CH2 ) . 13C-NMR ( 75 MHz , CDCl3 ) : δ 171 . 97 ( CO ester ) , 171 . 56 ( CO ester ) , 153 . 04 ( C Ar ) , 152 . 74 ( C Ar ) , 152 . 31 ( C Ar ) , 152 . 09 ( C Ar ) , 151 . 02 ( C Ar ) , 150 . 43 ( C Ar ) , 144 . 79 ( C Ar ) , 139 . 41 ( C Ar ) , 138 . 16 ( C Ar ) , 132 . 61 ( C Ar ) , 128 . 20 ( CH Ar ) , 127 . 15 ( CH Ar ) , 126 . 33 ( CH Ar ) , 123 . 34 ( C Ar ) , 122 . 64 ( CH Ar ) , 122 . 61 ( CH Ar ) , 121 . 91 ( CH Ar ) , 119 . 54 ( CH Ar ) , 113 . 89 ( C Ar ) ( CH Ar ) , 113 . 43 ( C Ar ) , 113 . 35 ( C Ar ) , 105 . 16 ( C Ar ) , 69 . 10 ( CH2O ) , 67 . 70 ( CH2O ) , 67 . 19 ( CH2O ) , 53 . 66 ( NCH2COOMe ) , 53 . 52 ( NCH2COOMe ) , 51 . 73 ( 4 OMe ) , 51 . 16 ( CH2N3 ) , 50 . 97 ( C1 or C4 ) , 50 . 52 ( C1 or C4 ) , 28 . 82 ( CH2 ) , 28 . 73 ( CH2 ) , 27 . 72 ( C3 ) , 26 . 26 ( CH2 ) , 25 . 52 ( CH2 ) , 20 . 83 ( C2 ) , 20 . 00 ( C6 ) , 19 . 85 ( C5 ) . MS ( ES+ ) , calcd for C57H68N7O12 [M]+ 1042 . 5 , found 1042 . 9 . HRMS ( ES+ ) , calcd for C57H68N7O12 [M]+ 1042 . 4920 , found 1042 . 4949 . To a solution of 8 ( 100 mg , 0 . 090 mmol ) in methanol ( 6 ml ) were added , KOH ( 504 mg , 9 . 00 mmol , 100 eq ) followed by 2 ml of water , the mixture was stirred overnight . The product was washed with HCl ( 1 M ) and extracted with CHCl3 until the aqueous phase become slightly pink . The organic phase was then dried over MgSO4 , filtered and evaporated . The crude was purified on a reverse phase column C-18 using acetonitrile ( 0 . 1% TFA ) and water ( 0 . 1% TFA ) as eluant ( 20% ACN to 60% ) . The solvents were evaporated and 80 mg of CaRuby-Nano ( ∼90% ) were obtained as a purple solid after lyophilisation ( dioxane/water , 1/1 ) . MS ( ES+ ) , calcd for C53H60N7O12 [M]+ 986 . 4 , found 986 . 4 . HRMS ( ES+ ) , calcd for C53H60N7O12 [M]+ 986 . 4294 , found 1042 . 4329 . To a solution of CaRuby-Nano ( 50 mg , ∼50 µmol ) in chloroform were added bromomethyl acetate ( 80 µl , 500 µmol , 1 eq ) and NEt3 ( 60 µl , 400 µmol , 8 eq ) . The solution was protected from light and allowed to stir at room temperature overnight . The reaction was monitored by TLC ( DCM/MeOH , 9/1 ) . The solvents were evaporated and the crude was purified by column chromatography on silica gel ( gradient of 100% DCM to 9/1 DCM/Methanol ) to obtain 30 mg of CaRuby-Nano AM esters ( ∼45% ) as a purple solid after lyophilisation ( dioxane/water , 1/1 ) . Rf = 0 . 45 ( DCM/MeOH , 9/1 ) . MS ( ES+ ) , calcd for C65H76N7O20 [M]+ 1274 . 5 , found 1274 . 5 . HRMS ( ES+ ) , calcd for C65H76N7O20 [M]+ 1274 . 5140 , found 1274 . 5128 . Dextran 6000 MW ( Sigma–Aldrich , ref: 31388 ) and dextran 1500 MW ( Sigma–Aldrich , ref: 31394 ) were propargylated as described by Nielsen et al . ( 2010 ) . The 1H-NMR showed that the functionalized dextrans were propargylated once by unit . Final MW Dextran 6000: ∼9800 g . mol−1 . Final MW Dextran 1500 : ∼2400 g . mol−1 . All procedures were approved by the local ethical review committee and performed under license from the UK Home Office in accordance with the Animals ( Scientific Procedures ) Act 1986 , and in accordance with the Institut National de la Santé et de la Recherche Médicale ( INSERM ) Animal Care and Use Committee Guidelines and with Centre National de la Recherche Scientifique ( CNRS ) animal experimentation guidelines and European laws and policies , as applicable . Parasagittal cerebellar slices ( 200 μm ) were made using standard techniques ( Davie et al . , 2006 ) from C57BL6/J mice ( Harlan , UK ) at postnatal days 25–29 . Artificial CSF ( ACSF ) for both slicing and recording contained the following ( in mM ) : 125 NaCl , 2 . 5 KCl , 26 NaHCO3 , 1 . 25 NaH2PO4 , 25 glucose , 1 MgCl2 , and 2 CaCl2 , and was bubbled with 5% carbon dioxide , 95% oxygen . Slices were continuously superfused with ACSF during the experiment . For high speed imaging experiments , acute 260 µm thick slices were obtained from the cerebellar vermis of P60 CD1 mice and superfused with ACSF , as previously described ( Dugue et al . , 2009 ) . Full frame and linescan two-photon imaging was performed using microscopes optimized for in vitro ( Prairie Technologies , now Bruker Nano Surfaces , USA ) or in vivo ( MOM , Sutter , Novato , CA , USA ) experiments . Two photon excitation was provided by a pulsed Ti:Sa laser ( MaiTai HP , Spectra-Physics , Santa Clara , CA , USA ) , tuned to a central wavelength of 890–920 nm . The microscopes were controlled by ScanImage 3 . 5 and 3 . 7 . 1 ( Pologruto et al . , 2003 ) ( now Vidrio Technologies , Arlington , VA , USA ) . For two-color imaging of iGluSnFR and CaRuby-Nano the Ti:Sa was tuned to 900 nm . Fluorescence light was split into red and green channels using dichroic mirrors ( 575/DCXR , Chroma , Bellows Falls , VT , USA ) and bandpass filtered ( green: 525/50; red: 607/45; both: Semrock , Lake Forest , IL , USA ) . Patch-clamp pipettes were filled with an internal solution containing ( in mM ) : K-methanesulfonate 133 , KCl 7 , HEPES 10 , Mg-ATP 2 , Na2ATP 2 , Na2GTP 0 . 5 , EGTA 0 . 05 , 0 . 1 Alexa Fluor 488 and CaRuby-Nano dextran as indicated; pH 7 . 2 . Recordings from visually identified Purkinje cells were made using a Multiclamp 700B amplifier ( Molecular Devices , Sunny Vale , CA , USA ) . Data were lowpass filtered at 4 kHz and acquired at 20 kHz using an ITC-18 digitizer ( HEKA Intruments , Bellmore , NY , USA ) controlled by AxoGraph X ( AxoGraph Scientific , Sydney , Australia ) . Electrical stimuli were delivered via a theta-glass bipolar electrode filled with ACSF using a constant current stimulus isolator ( DS-3 , Digitimer , Letchworth Garden City , UK ) . When using electrical stimulation , 10 µM SR-95531 ( Sigma or Tocris ) was added to the perfusion medium . Climbing fiber stimulation-evoked transient [Ca2+] changes in Purkinje cell spines were recorded at high acquisition rate ( >2 kHz ) by two-photon random-access microscopy , a technique based on the use of acousto-optic deflectors ( AODs ) , which enable selective scanning of defined points ( Otsu et al . , 2008 ) . Purkinje cells were recorded in current-clamp mode , using 2–3 MΩ patch pipettes containing 300 μM CaRuby-Nano dextran . Recordings were obtained by use of a Multiclamp 700B ( Molecular Devices ) . Following the dialysis of CaRuby-Nano , Purkinje cells in slices were imaged under a 25× Leica water immersion objective ( HCX IRAPO L 25×/0 . 95 , Leica Microsystems , Wetzlar , Germany ) . Two-photon excitation was produced by a pulsed Ti:Sa laser ( Chameleon Vision Plus , Coherent , Santa Clara , CA , USA ) coupled into the transmitted light pathway of the microscope by a dichroic filter ( 740dcsx , Chroma ) and tuned to a central wavelength of 890 nm . A custom-made user interface based on National Instrument cards programmed under Labview ( both National Instruments , Austin , TX , USA ) was used to operate the AODs and coordinate the scanning protocols and signal acquisition . A multifunction card ( NI-PCI-MIO 16 E-4 ) was used to pass all the triggers necessary to synchronize the imaging and the electrophysiology and to control the piezo-electric device that moves the objective in Z . Fluorescence photons were detected by cooled AsGaP photomultipliers ( H7421-40 , Hamamatsu , Hamamatsu , Japan ) discriminated and counted on a fast digital card . Young ( P19 ) C57BL6/J mice were anesthetized using isoflurane , an incision was made into the scalp and a small ( ∼0 . 5 mm ) craniotomy was performed over lobule V of the cerebellar vermis . A wide bore ( ∼50 µm ) micropipette containing viral suspension ( AAV1 . hSyn . iGluSnFr . WPRE . SV40 , University of Pennsylvania Vector Core ) was inserted through the craniotomy and carefully lowered 1 . 0 mm into the brain . Using application of low pressure 400–800 nl viral suspension were slowly injected ( 10–20 min ) . After the injection further 5–10 min were waited before retraction of the injection pipette . The scalp was glued and sutured and the mouse left to recover . At least 7 days incubation time were allowed prior to further experiments . Kv3 . 1-eYFP mice ( Metzger et al . , 2002 ) ( 8–10 week-old ) were anesthetized with an intraperitoneal injection of ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) . CaRuby-Nano dextran ( 6 kDa ) was dissolved 2 . 5% wt/vol in a solution of aCSF ( in mM: 125 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 25 NaHCO3 , 1 MgCl2 , 2 CaCl2 and 25 glucose ) with 0 . 2% Triton X-100 ( Sigma–Aldrich ) . 8 µl of this solution was injected in the mouse naris , and mice were left on their backs to recover from anesthesia ( protocol adapted from [Wachowiak and Cohen , 2001; Lecoq et al . , 2009] ) . 7 days later , an acute craniotomy was performed over the dorsal olfactory bulb and the brain stabilized with 3 . 5% agar for imaging . To activate olfactory sensory neurons ( OSNs ) , odors were applied in a 1 ml/min flux of filtered , humidified air supplemented with 30% oxygen . eYFP and CaRuby-Nano fluorescence was collected in two separate channels ( ‘green’ and ‘red’ , respectively ) of a custom-built two-photon laser scanning microscope , with the femtosecond pulsed excitation beam set to 910 nm . Adult C57BL6 mice ( 6–9 weeks ) were anesthetized with isoflurane , supplemented with 1 mg/kg chlorprothixene . A 1 . 5–2 mm craniotomy was performed over cerebellar lobule V . Care was taken to leave the dura mater intact . CaRuby-Nano-AM was prepared and injected using standard methods ( Stosiek et al . , 2003; Sullivan et al . , 2005 ) . A 50 µg aliquot was dissolved in 20% Pluronic-127 in DMSO ( Invitrogen ) and then diluted 1:10 in saline ( 150 mM NaCl , 2 . 5 mM KCl , 10 mM HEPES , pH 7 . 4 ) . This solution was filtered and injected into the cerebellum under visual guidance using a patch-pipette and 500–750 mbar pressure for 1–3 min . After injection the preparation was left to incubate for up to 1 . 5 hr prior to imaging . This helped improve labeling and lower unspecific fluorescence . Imaging data were analyzed using ImageJ ( http://rsbweb . nih . gov/ij/ ) . Extracted fluorescence traces , linescans and electrophysiological data were analyzed using in house routines programmed in Igor Pro versions 5 or 6 . 2 ( Wavemetrics ) and in pClamp 10 ( Molecular Devices ) . Statistical analysis was performed in Matlab ( MathWorks , Natick , MA , USA ) or Igor Pro ( Wavemetrics , Portland , OR , USA ) . Experimental groups were compared using a t-test and were assumed to be significantly different if the found p-values were <0 . 05 . | The movement of calcium ions within cells controls many vital biological processes , ranging from cell growth to muscle contraction and brain activity . These calcium signals are triggered by stimuli , such as nerve impulses , which drive calcium entry into cells or release calcium from internal stores . These changes in calcium levels can span several orders of magnitude , and can be either localized to very small parts of the cell or span the entire cell . Scientists have developed numerous indicators or ‘probes’ that can detect even very low levels of calcium . One common method uses proteins that fluoresce when viewed under a fluorescence microscope each time the protein senses increases of calcium . Most of these probes fluoresce green , and so to view a second signal that occurs in the cell at the same time it's easier to use a probe that fluoresces with a different color , such as red . However , the red-shifted probes that are currently available either produce unreliable results because they tend to leak through cell membranes , or are not very sensitive to calcium ions . New types of red-shifted probes are therefore urgently needed . In 2012 , researchers developed a family of red fluorescent probes known as Calcium Ruby ( CaRuby for short ) that were more versatile than earlier red probes . Now , Collot , Wilms et al . —including several of the researchers involved in the 2012 research—have enhanced the properties of CaRuby by modifying the chemical structure of the probes . This increased the ability of CaRuby to bind calcium ions , making it more sensitive to small calcium changes . Testing the usefulness of the newly developed probes—called CaRuby Nano—in mouse nerve cells revealed the probes are highly sensitive and can even detect the calcium signal resulting from a single nerve impulse . Collot , Wilms et al . then went on to demonstrate that CaRuby-Nano can be used alongside a green-fluorescing probe to record two signals at the same time . In one experiment , the release of chemical messengers known as neurotransmitters was stimulated , which caused calcium ions to flow into the observed nerve cells . The researchers succeeded in simultaneously detecting a green signal indicating an increase in neurotransmitter levels and a red signal produced by the corresponding release of calcium . Such dual-color imaging was not possible with previous probes . Finally , it was shown that CaRuby-Nano can also be used to produce dual-color images of the brain activity of live mice . In summary , these results demonstrate that CaRuby-Nano is a highly sensitive and versatile indicator and can be used together with other probes to observe two simultaneous events in cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"tools",
"and",
"resources",
"neuroscience"
] | 2015 | CaRuby-Nano: a novel high affinity calcium probe for dual color imaging |
How the nervous system internally represents environmental food availability is poorly understood . Here , we show that quantitative information about food abundance is encoded by combinatorial neuron-specific gene-expression of conserved TGFβ and serotonin pathway components in Caenorhabditis elegans . Crosstalk and auto-regulation between these pathways alters the shape , dynamic range , and population variance of the gene-expression responses of daf-7 ( TGFβ ) and tph-1 ( tryptophan hydroxylase ) to food availability . These intricate regulatory features provide distinct mechanisms for TGFβ and serotonin signaling to tune the accuracy of this multi-neuron code: daf-7 primarily regulates gene-expression variability , while tph-1 primarily regulates the dynamic range of gene-expression responses . This code is functional because daf-7 and tph-1 mutations bidirectionally attenuate food level-dependent changes in lifespan . Our results reveal a neural code for food abundance and demonstrate that gene expression serves as an additional layer of information processing in the nervous system to control long-term physiology .
All organisms need to accurately assess their environment to respond to changes that impact their survival . Environmental changes such as food availability can lead to alterations in organismal physiology , such as stress resistance and metabolic states that have consequences for clinically important outputs such as disease progression , health , fecundity and lifespan ( Libert and Pletcher , 2007 ) . Many conserved genetic mechanisms that govern these alterations to physiology have been identified ( Libert and Pletcher , 2007; Berthoud and Morrison , 2008; Rother et al . , 2008; Alcedo et al . , 2010; Koch and Horvath , 2014 ) . Yet , how these genetic pathways encode and process information about the environment to elicit physiological outputs in vivo is unclear at a quantitative and mechanistic level , despite their importance for health and disease . In animals , the nervous system is the central site for processing sensory information and coordinates organism-wide responses to changing conditions . Food availability is a critical environmental variable that modulates metabolism and other physiological outputs via neuroendocrine circuits ( Berthoud and Morrison , 2008; Rother et al . , 2008; Alcedo et al . , 2010; Koch and Horvath , 2014 ) . In contrast to well-studied sensory modalities such as vision and olfaction ( Baier , 2013; Wilson , 2013 ) , where neural processing occurs on short timescales using electrical signals , how food availability is internally represented across a broad range of inputs to regulate long-term , food-related physiological responses remains virtually unknown . A particularly interesting food-related response is the role of dietary restriction ( DR ) in modulating lifespan in diverse species ( Bishop and Guarente , 2007; Mair and Dillin , 2008; Fontana et al . , 2010; Alic and Partridge , 2011 ) . DR occurs through changes that likely happen over long timescales ( hours to days ) , unlike fast behavioral responses to visual or olfactory cues . Neural gene expression also occurs over long timescales ( minutes to hours ) and is thus well suited for functionally encoding food abundance during DR . In Caenorhabditis elegans , daf-7 and tph-1 are conserved components of neural TGFβ and serotonin signaling pathways , respectively , and are associated with food sensing and modulation of organismal physiology . daf-7 encodes a TGFβ family member ( Ren et al . , 1996 ) , while tph-1 encodes tryptophan hydroxylase , the rate-limiting enzyme for serotonin synthesis ( Sze et al . , 2000 ) . In C . elegans , TGFβ and serotonin signaling affect lifespan and metabolism , consistent with conserved roles from invertebrates to mammals ( Sze et al . , 2000; Ashrafi , 2007; Murakami and Murakami , 2007; Petrascheck et al . , 2007; Shaw et al . , 2007; Brown and Schneyer , 2010; Oury and Karsenty , 2011 ) . tph-1 and daf-7 are expressed in an environmentally responsive manner in specific neurons with food-related functions ( Ren et al . , 1996; Schackwitz et al . , 1996; Sze et al . , 2000; Zhang et al . , 2005; Chang et al . , 2006; Liang et al . , 2006; Greer et al . , 2008; Pocock and Hobert , 2010 ) . daf-7 is expressed in the ASI sensory neurons , whose activities are responsive to bacterial food ( Ren et al . , 1996; Gallagher et al . , 2013; Zaslaver et al . , 2015 ) . Starvation reduces daf-7 expression in ASI , and laser ablations of ASI extend lifespan , consistent with the role of daf-7 and other ASI-expressed genes in modulating lifespan ( Ren et al . , 1996; Alcedo and Kenyon , 2004; Bishop and Guarente , 2007 ) . tph-1 is expressed in the NSM foregut neurons , the ADF sensory neurons , and the HSN motorneurons involved in egg-laying ( Sze et al . , 2000 ) . Both serotonin signaling mutants and NSM ablation affect food-modulated locomotion , consistent with the idea that serotonin from NSM acts in this food-related response ( Sawin et al . , 2000 ) . In the food-responsive ADF neurons ( Zaslaver et al . , 2015 ) , tph-1 expression is responsive to pathogenic bacteria and starvation , to respectively mediate aversive olfactory plasticity and stress responses ( Zhang et al . , 2005; Liang et al . , 2006 ) . daf-7 and tph-1 are therefore strong candidates for mediating the link between environmental cues and longevity . Nonetheless , how these genes cooperate to quantitatively encode a broad range of food levels to modulate lifespan is unknown . Gene-expression responses to food cues have largely been studied as ON/OFF switches to the presence or absence of food ( Zinke et al . , 2002; Baugh et al . , 2009 ) . Because food abundance is a continuous variable , we sought to understand how expression of tph-1 and daf-7 could allow animals to distinguish multiple food levels . Furthermore , gene expression is inherently variable ( Eldar and Elowitz , 2010 ) , but this property is rarely studied in vivo in multicellular animals; thus we also sought to determine how gene-expression variability affects the ability of the worm to encode its environment . Here we show that daf-7 and tph-1 expression in three pairs of neurons forms a distributed circuit that quantitatively encodes food abundance and mediates dietary effects on lifespan in C . elegans . Specific disruptions to this circuit resulted in corresponding attenuation in the ability to discriminate between food levels in both the gene-expression code and lifespan output . We found that this circuit tunes its own accuracy , largely via the regulation of the dynamic range and variability of food-responsive gene expression by tph-1 and daf-7 signalling , respectively . Our work suggests that neural regulation of gene expression in conserved pathways can couple environmental sensation to physiological output , and highlights a novel mechanism for information processing by the nervous system to impact physiology .
During DR , lifespan increases as food levels are decreased from ad libitum conditions until reaching a maximum , beyond which further food reduction lowers lifespan ( Bishop and Guarente , 2007; Mair and Dillin , 2008; Fontana et al . , 2010; Alic and Partridge , 2011 ) . To fully understand the response to food levels that C . elegans might encounter in the wild ( Felix and Duveau , 2012 ) , we modified a well-established DR protocol ( Greer et al . , 2007 ) ( Figure 1A ) to measure the lifespans of wildtype animals shifted as day 2 adults to 19 concentrations of the Escherichia coli food source across ∼11 orders of magnitude ( Figure 1B , Figure 1—figure supplement 1 and Figure 1—source data 1 ) . We inhibited progeny production with egg-5 ( RNAi ) ( Figure 1A ) to prevent matricide due to internal hatching at low food levels . This treatment does not affect the lifespan response to food; similar responses were observed in wildtype animals without egg-5 ( RNAi ) ( Figure 1—figure supplement 1 ) , and are found in the literature where similar subsets of food ranges were tested using other DR protocols ( below ) . 10 . 7554/eLife . 06259 . 003Figure 1 . Two neuronal genes , daf-7 and tph-1 , shape a complex , multiphasic relationship between lifespan and food availability . ( A ) Protocol for maintaining animals at different food levels for lifespan and imaging experiments . Effects of initiating different dietary restriction ( DR ) on different days are shown in Figure 1—figure supplement 1A , B . ( B ) Mean lifespan of wildtype worms subjected to 19 food levels ranging from 0–3 . 5 × 1010 bacterial cells/ml at 20°C . Points denoting key features in the food response and used as food conditions in subsequent experiments are highlighted . Figure 1—figure supplement 1D shows that these lifespan responses have similar shapes across different temperatures . The lifespan data are shown in Figure 1—source data 1 . ( C ) Mean lifespans of wildtype and mutant animals across the six food levels indicated in ( B ) show that loss of tph-1 and daf-7 preserves the pattern but attenuates the range of the lifespan response . Genotypes are indicated by the legends below ( E ) and ( F ) . The lifespan data are shown in Figure 1—source data 2 , and statistical comparisons between the different genotypes and food levels are shown in Figure 1—source data 3 . ( D ) tph-1 and daf-7 modulation of lifespan is bidirectional and their epistatic relationship is food-specific . The epistatic interaction between the two genes in the tph-1 ( − ) ; daf-7 ( − ) double mutant differs depending on food level . The double mutant resembled the tph-1 ( − ) single mutant in the absence of a bacterial food source and resembled the daf-7 ( − ) single mutant at a high bacterial food concentration . ( E ) Range of food-induced lifespan modulation for each genotype . Range is defined by the difference between the highest and lowest mean lifespan response across the six food levels . ( F ) Average of the mean lifespan responses across all food levels for each genotype reveals a consistent , food-independent baseline lifespan response . The schedule for transferring animals to different conditions in these lifespans are shown in Figure 1—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 00310 . 7554/eLife . 06259 . 004Figure 1—source data 1 . Summary of wild type lifespan outputs under the full range of food levels tested . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 00410 . 7554/eLife . 06259 . 005Figure 1—source data 2 . Summary of wild type and mutant lifespan outputs under six selected food levels . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 00510 . 7554/eLife . 06259 . 006Figure 1—source data 3 . Statistical significance of lifespan modulation across food levels and genetic backgrounds . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 00610 . 7554/eLife . 06259 . 007Figure 1—figure supplement 1 . Effects of DR initiation time , temperature and egg-5 ( RNAi ) on the lifespan response . ( A ) Survival curves for worms transferred from the baseline , abundant level of food ( 2 × 109 cells/ml ) to a more limited food condition ( 2 × 105 cells/ml ) at different days of adulthood . Different days of DR initiation result in substantial differences in survival trajectories . ( B ) The mean lifespan response of these worms shows lifespan extension for animals transferred on day 2 or later but a negative effect on lifespan for animals subjected to DR from day 1 of adulthood , possibly due to developmental effects . ( C ) Wildtype worms show a similar lifespan response to exposed different food levels in the absence of egg-5 ( RNAi ) . ( D ) The pattern of lifespan modulation by food abundance is maintained across the range of standard C . elegans culture temperatures . ( E ) Increased temperature consistently lowers mean lifespans across all food levels . ( F ) Increased temperature also consistently lowers the range of lifespan modulation achieved by alterations in food . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 00710 . 7554/eLife . 06259 . 008Figure 1—figure supplement 2 . Schedule of transfers for lifespans at different temperatures . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 008 We uncovered a multiphasic relationship between bacterial abundance and longevity that is more complex than previously reported with smaller concentration ranges ( Figure 1B ) ( Greer et al . , 2007; Greer and Brunet , 2009; Ching et al . , 2010 ) . We found that lifespan increased and then decreased as bacterial concentration was reduced from the highest level , forming a DR response consistent with prior reports at high food ranges ( Bishop and Guarente , 2007; Panowski et al . , 2007; Greer and Brunet , 2009; Mair et al . , 2009; Ching et al . , 2010 ) . Surprisingly , upon further reduction , lifespan increased again till a plateau was reached , suggesting that the initial decrease was not due to limiting nutrients . The longest lifespans occurred in the absence of bacteria , where the magnitudes of these effects were consistent with published dietary deprivation experiments ( Kaeberlein et al . , 2006; Lee et al . , 2006 ) . This relationship between lifespan and food abundance was maintained across temperatures ( Figure 1—figure supplement 1 ) , suggesting a robust food-sensing process . This multiphasic food response may reflect trade-offs between multiple food-regulated processes as previously discussed ( Kaeberlein et al . , 2006; Lee et al . , 2006 ) . Here we used the complex lifespan response as a functional basis for understanding how neuronal gene expression could encode food abundance . To understand how the multiphasic lifespan response to food abundance is regulated , we measured the lifespan of daf-7 and tph-1 null mutants across six bacterial concentrations that captured the complexity of broad-range DR ( circled in Figure 1B ) . Prior studies suggested that daf-7 and tph-1 mediate lifespan extension ( Murakami and Murakami , 2007; Shaw et al . , 2007; van der Goot et al . , 2012 ) . We showed that their effects were in fact bidirectional: these genes could either extend or reduce lifespan in a food-specific manner ( Figure 1C , Figure 1—source data 2 , 3 ) . Both single mutants had reduced lifespans at low food levels and increased lifespan at 6 × 108 cells/ml in comparison to wildtype; additionally , daf-7 ( − ) mutants were long-lived at the highest food level ( Figure 1C ) . The magnitude of lifespan changes we observed at high food levels ( 1 × 1010 cells/ml ) were comparable to prior studies performed at ad libitum food conditions ( Murakami and Murakami , 2007; Shaw et al . , 2007 ) . Intriguingly , tph-1 and daf-7 influenced the longevity response more strongly at low and high bacterial concentrations respectively ( Figure 1D ) , suggesting that they act at different but overlapping ranges of food . Furthermore , the double mutant resembled the tph-1 ( − ) mutant at low bacterial levels and the daf-7 ( − ) mutant at high bacterial levels ( Figure 1D ) , suggesting that these genes act in parallel rather than in a linear pathway . Together , these phenotypes indicate that daf-7 ( − ) and tph-1 ( − ) mutants were neither intrinsically long- nor short-lived; instead , their phenotypes and genetic interactions were modulated by extensive gene-environment interactions . Rather than altering the basic pattern of the lifespan response to food , loss of tph-1 or daf-7 , either alone or in combination , dampened food responsiveness by bidirectionally attenuating extension and reduction of lifespan due to DR ( Figure 1C , D ) . This effect was manifested in the diminished range of lifespans across all food levels in both the daf-7 ( − ) and tph-1 ( − ) single mutants , which was further reduced in the double mutant ( Figure 1E ) . This result also supports the idea that these genes act in parallel pathways . Furthermore , the mean lifespan across all food levels were similar in all the genotypes tested ( Figure 1F ) , suggesting that mutations in tph-1 and daf-7 lowered the food-responsive component of longevity around a consistent , food-independent lifespan that may be specified by other environmental parameters such as temperature ( Figure 1E and Figure 1—figure supplement 1 ) . This bidirectional dampening of the food response and preservation of an underlying lifespan differs from previously described DR regulators , such as aak-2 , daf-16 , pha-4 and skn-1 , whose mutants abolish DR-mediated lifespan extension ( Bishop and Guarente , 2007; Greer et al . , 2007; Panowski et al . , 2007 ) . Thus , tph-1 and daf-7 mutants reveal a previously unobserved DR phenotype , and our results suggest that these genes mediate a bidirectional lifespan response to DR . tph-1 is expressed in the ADF sensory neurons , the NSM neurons within the foregut , and the hermaphrodite-specific HSN motor neurons ( Sze et al . , 2000 ) . daf-7 is expressed in a single pair of ASI sensory neurons ( Ren et al . , 1996; Schackwitz et al . , 1996 ) . To determine whether tph-1 and daf-7 act in these neurons to modulate lifespan , we expressed these genes in specific neurons and tested their ability to rescue the lifespan phenotypes in the tph-1 ( − ) ; daf-7 ( − ) double mutant . Expression of tph-1 in either ADF or NSM neurons ( Figure 2A–C ) or of daf-7 in ASI neurons ( Figure 2D , E ) could rescue the lifespan phenotypes at low and high food levels , respectively . These results indicate that the activity of tph-1 and daf-7 in these respective neurons are relevant to lifespan modulation . 10 . 7554/eLife . 06259 . 009Figure 2 . Neuron-specific rescue of lifespan phenotypes . ( A ) Lifespan outcomes of wildtype , tph-1 ( − ) and daf-7 ( − ) single mutants and the tph-1 ( − ) ; daf-7 ( − ) double mutant indicates that the double mutant closely resembles the tph-1 ( − ) mutant in the absence of a bacterial food source . ( B ) In the absence of bacterial food , restoration of tph-1 activity in the NSM neurons via the expression of a tph-1 cDNA driven by the ceh-2 promoter rescues the lifespan reduction observed in the tph-1 ( − ) ; daf-7 ( − ) double mutants . ( C ) Restoration of tph-1 expression in the ADF neurons via the srh-142 promoter also shows reversal of the lifespan reduction . ( D ) Lifespan outcomes of wildtype , tph-1 ( − ) and daf-7 ( − ) single mutants and the tph-1 ( − ) ; daf-7 ( − ) double mutant indicates that the double mutant closely resembles the daf-7 ( − ) mutant at a high concentration of the bacterial food source . ( E ) At high food level , restoration of daf-7 expression in the ASI neurons via the expression of daf-7 under the gpa-4 promoter reverses the lifespan extension observed in the tph-1 ( − ) ; daf-7 ( − ) double mutants . All comparisons are drawn against non-transgenic siblings of animals bearing the extrachromosomal array of interest . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 009 Previous studies showed that daf-7 and tph-1 expression are regulated by environmental cues ( Ren et al . , 1996; Schackwitz et al . , 1996; Sze et al . , 2000; Zhang et al . , 2005; Chang et al . , 2006; Liang et al . , 2006; Greer et al . , 2008; Pocock and Hobert , 2010 ) . However , their expression profiles over a broad range of inputs remain unknown because manual studies limit the number of animals and environmental conditions that can be feasibly studied in a consistent way . To overcome these limitations , we used an automated , high-throughput microfluidic-based platform ( Figure 3A and Figure 3—figure supplement 1 ) ( Chung et al . , 2008; Crane et al . , 2012 ) for quantitative large-scale imaging of individual worms bearing single-copy fluorescent transcriptional reporters for both tph-1 and daf-7 ( Ptph-1::mCherry and Pdaf-7::Venus ) across different food levels ( Figure 3B ) . For brevity , we refer to these reporter activities as tph-1 and daf-7 expression . Our reporters contain the same regulatory regions as published reporters that have been well validated , and show identical expression patterns ( Ren et al . , 1996; Schackwitz et al . , 1996; Sze et al . , 2000; Zhang et al . , 2005; Chang et al . , 2006; Liang et al . , 2006; Greer et al . , 2008; Pocock and Hobert , 2010 ) ( Figure 3B ) . Starvation , hypoxia , or pathogenic bacteria alter both tph-1 reporter expression and serotonin levels ( Zhang et al . , 2005; Liang et al . , 2006; Pocock and Hobert , 2010 ) , while corresponding changes occur in daf-7 RNA levels and daf-7 reporter expression ( Ren et al . , 1996 ) . These published results indicate that tph-1 and daf-7 reporters are faithful readouts for the expression of their respective genes ( see ‘Materials and methods’ for additional details on reporter validation ) . 10 . 7554/eLife . 06259 . 010Figure 3 . High-throughput quantitative imaging of tph-1 and daf-7 fluorescent reporters reveals neuron-specific , graded expression responses to food level . ( A ) Microfluidic system enabling high-throughput , neuron-specific quantitative imaging of gene expression in a large number of individual animals . Animals are transferred from culture plates to a liquid suspension at day 6 of adulthood and then loaded into the device for imaging . Figure 3—figure supplement 1 shows an overview of this imaging system . ( B ) Representative merged fluorescent image of transgenic worm with red and green fluorescent reporters for tph-1 and daf-7 transcriptional activity . Shapes indicate locations and identities of specific neurons . ( C ) Mean expression profiles of tph-1 in NSM ( Ptph-1NSM ) and ADF ( Ptph-1ADF ) , and daf-7 in ASI ( Pdaf-7ASI ) across six different food levels are neuron-specific and largely non-monotonic . Measurements are normalized to the highest mean expression response observed in each respective neuron; error bars are SEM . ( D ) Distribution of the expression responses of tph-1 in NSM and ADF and daf-7 in ASI at different food levels . Means are indicated by the solid lighter-shade lines behind the distributions . Dashed line denotes the highest mean expression for each neuron , which was used for normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 01010 . 7554/eLife . 06259 . 011Figure 3—figure supplement 1 . Overview of the high-throughput quantitative imaging system for single-cell fluorescent intensity measurements in C . elegans . ( A ) The system incorporates a microfluidic chip with pressure controlled valves for worm handling , a custom-built pneumatic control system to control on-chip valves , a standard wide-field epifluorescent microscope and camera system and custom LabVIEW software for integration and control . ( B ) The microfluidic chip consists of worm and fluid inlets and outlets , two control channels for valve activation and a second layer cooling channel for immobilization . Sequential activation of the valves , worm and fluid inlets permit sequential loading of individual worms for imaging . ( C ) The LabVIEW software interacts with the pneumatic valve control system ( green blocks ) and camera ( blue blocks ) to automate the imaging process by sequential loading and imaging . ( D ) Image of the LabVIEW user interface . ( E ) Block diagram of steps used to calculate single-cell fluorescent intensities from raw z-stacks . Cell identifications and locations are calculated from 2-D maximum projections . Locations are then mapped onto the z-stacks for 3-D volume integration . ( F ) Maximum projections and binary images from a thresholding procedure are used to ascertain cellular locations . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 011 We measured tph-1 expression levels in both NSM and ADF , and daf-7 in ASI , in animals exposed to the same six food levels that define our complex DR response ( Figure 3C ) . Remarkably , we found that each neuron type had a specific pattern of activity across the six food levels ( Figure 3C ) . Even with respect to a single gene , tph-1 , the expression response in NSM differed from that in ADF , suggesting non-redundant roles of NSM and ADF in encoding bacterial abundance . Consistent with low tph-1 expression in the absence of food ( Figure 3C ) , serotonin levels were reduced in NSM and ADF after starvation ( ‘Materials and methods’ and Liang et al . , 2006 ) . Notably , the responses of tph-1 in NSM and daf-7 in ASI were non-monotonic , prohibiting unique representation of the food level using either of these readouts alone . The dynamic range and inter-individual variability of the three expression patterns also differed , suggesting different sensitivities towards the sensory input from food in different neurons ( Figure 3C , D ) . Moreover , the distributions of reporter intensity for each neuron type shifted in a graded manner across the six bacterial concentrations ( Figure 3D ) indicating that the expression of tph-1 and daf-7 could provide information about a continuous range of environmental inputs for individual animals . This graded response contrasts with many switch-like pathways , such as MAP kinase signalling , that produce outputs with a finite number of ( usually two ) stable states , where the population responds to changing environmental conditions by shifting sub-populations of cells from one state to another ( Ferrell , 1996 ) . Since gene expression may be used to internally represent environmental conditions and mediate long-term physiological outputs such as longevity , we next asked to what extent expression in individual neurons correlates with lifespan across food levels . We found that the average expression levels of the individual genes in each neuron alone were insufficient to either represent food inputs or uniquely specify lifespan outputs ( Figure 4A ) , largely due to non-monotonic expression and lifespan responses ( Figures 1 , 3C ) . However , combining non-monotonic encoding across multiple neurons can increase the representational capability , by allowing animals to use a combinatorial scheme to internally represent environmental conditions with higher resolution ( Figure 4B ) . 10 . 7554/eLife . 06259 . 012Figure 4 . The combination of all neuronal gene-expression readouts generates a unique internal representation of food levels . ( A ) Relationship between individual gene-expression profiles and the lifespan responses across the six food levels indicates that the individual readouts are insufficient to uniquely specify lifespan responses . ( B ) The combination of tph-1 in NSM ( Ptph-1NSM ) and ADF ( Ptph-1ADF ) and daf-7 in ASI ( Pdaf-7ASI ) create a non-monotonic multivariate encoding scheme capable of both uniquely representing food inputs and potentially specifying lifespan outputs . ( C ) The ability of expression and lifespan readouts to respond to and represent ( encode ) food conditions can be estimated by using the readouts of interest to infer the true food conditions . The results can be visually represented by matrices where the squares in each column indicate the frequency with which particular inferences are made for a given true food level . Distinct , non-overlapping readouts result in high discriminatory power , represented by a highly diagonalized matrix ( top ) . Indistinct , overlapping response profiles result in low ( random ) discriminatory power , represented by a uniform matrix ( bottom ) . ( D ) Matrices indicating the representational capability of tph-1 and daf-7 readouts individually or in combination exhibit similar total encoding fidelity to that of lifespan outputs in wildtype animals . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 012 To assess the accuracy of the internal representation of food levels based on the graded combinatorial expression of tph-1 and daf-7 in wildtype animals , we next applied a decoding analysis ( Figure 4C , D ) ( Dayan and Abbott , 2005 ) . Using only the gene-expression data , we applied this algorithm to infer the food level that individual animals were exposed to . This approach takes into account the entire distribution of gene expression , and thus involves both means and variance to estimate accuracy . Intuitively , decreased overlap between the distributions from different food levels leads to increased accuracy ( Figure 4C ) . We modeled neuronal encoding with a maximum likelihood estimator applied to the population distributions of expression data at each food level . We next inferred food exposure based on expression in individuals and then compared them to the actual food level to determine the accuracy of the gene-expression responses . The results were summarized as matrices showing the frequency of each inferred food stimuli vs the actual food stimuli for each population ( Figure 4C , D ) . If the expression distributions were distinct ( non-overlapping ) across food levels , a correct inference was always obtained , resulting in a strongly diagonalized matrix ( Figure 4C , top ) . If the distributions overlapped entirely , then the food level inference was random and the values in the matrix were inversely related to the number of food conditions tested ( Figure 4C , bottom ) . Using expression data from NSM , ADF or ASI individually , we showed that each neuronal readout had some discriminatory power ( Figure 4D ) . When neuron pairs were combined , and particularly when all three neuron pairs were used , the accuracy improved ( Figure 4D ) . This result suggests that non-redundant encoding by each neuron pair improves the accuracy of the overall representation . Generally , this coding system was best at distinguishing the highest and lowest food levels , consistent with the boom and bust lifestyle of C . elegans in the wild ( Felix and Duveau , 2012 ) . Interestingly , the intermediate bacterial level , 6 × 108 cells/ml , where we observed the lowest lifespan , also showed a heightened accuracy . While the graded and non-monotonic encoding scheme may allow discrimination among certain intermediate food levels , the discriminatory ability was also limited by variation in the responses . To determine if the accuracy of gene expression was suitable for modulating lifespan , we next assessed how accurately the lifespan distributions could be used to infer the food level using a similar decoding procedure . This analysis is important because the accuracy of the output sets constraints for the accuracy required in the internal representation . Ideally , the accuracy of both the representation and output should be similar , so that the representation carries sufficient information for the output . In this decoding analysis , we assessed whether lifespan could also be used to infer the food level experienced by an individual , using estimators based on the Weibull distribution of our survival data . Remarkably , the inferential accuracy of lifespan was similar to that of gene expression ( Figure 4D ) , indicating that neuronal gene expression captured sufficient information about food levels to mediate corresponding lifespan outputs . Prior work suggested that tph-1 and daf-7 cross-regulate ( Sze et al . , 2000; Chang et al . , 2006 ) . Thus , we explored the possibility that the internal representation based on tph-1 and daf-7 expression in wildtype animals emerges from their mutual regulation . Because our reporter transgenes were distinct from the endogenous genes , we were able to measure reporter expression in tph-1 ( − ) and daf-7 ( − ) single and double mutants , which could reveal self- and cross-regulatory interactions . These measurements of reporter expression across six food levels uncovered new and extensive regulatory relationships between tph-1 and daf-7 that were neuron-specific and food-dependent ( Figure 5A ) . First , daf-7 ( − ) and tph-1 ( − ) mutants had non-redundant phenotypes , indicating that they acted in parallel pathways ( Figure 5A ) . For example , compared to the single mutants , the tph-1 ( − ) ; daf-7 ( − ) double mutant showed a greater increase in the expression of tph-1 in NSM at the higher food levels ( Figure 5A , left ) . Second , we observed crosstalk between tph-1 and daf-7 . Loss of daf-7 signalling affected tph-1 expression in both NSM and ADF neurons ( Figure 5A , left and middle ) while loss of tph-1 affected daf-7 expression in the ASI neuron at higher bacterial concentrations ( Figure 5A , right ) . Third , both tph-1 and daf-7 regulated their own expression . tph-1 expression in NSM and ADF was altered in tph-1 ( − ) mutants ( Figure 5A , left and middle ) and daf-7 expression in ASI was altered in daf-7 ( − ) mutants ( Figure 5A , right ) . The self-regulation of daf-7 and tph-1 was not mediated via their crosstalk: daf-7 influenced its own expression in the absence of tph-1 activity ( Figure 5A , right ) and tph-1 self-regulated in the absence of daf-7 activity ( Figure 5A , left and middle ) . 10 . 7554/eLife . 06259 . 013Figure 5 . Crosstalk and self-regulation of tph-1 and daf-7 are important in shaping the pattern , range and variability of food-induced gene-expression responses . ( A ) Food-responsive expression profiles of tph-1 in NSM ( Ptph-1NSM ) and ADF ( Ptph-1ADF ) , and daf-7 in in ASI ( Pdaf-7ASI ) for different genetic backgrounds , as indicated in the legend . As in Figure 3 , all values are normalized to the highest wildtype mean expression response observed in the respective neuron ( dotted line ) . ( B ) Effects of tph-1 and daf-7 signalling on the dynamic range of food-induced expression modulation for tph-1 in NSM and ADF and daf-7 in ASI . The dynamic range ( ΔF ) is defined by the difference between the highest and lowest mean expression responses across the six food levels for each genotype . ( C ) Effects of tph-1 and daf-7 signalling on the inter-individual variability of expression responses as measured by the standard deviation of gene-expression distributions in each neuron . ( D ) Schematic of the total regulatory effects of tph-1 and daf-7 on both the dynamic range ( ΔF ) and variability and uncertainty ( σ ) of expression readouts to induce opposing effects on representational capability . Figure 5—figure supplement 1 shows the distributions of gene expression for Ptph-1NSM , Ptph-1ADF , and Pdaf-7ASI across all genotypes tested . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 01310 . 7554/eLife . 06259 . 014Figure 5—figure supplement 1 . Neuron-specific expression distributions in wild type and mutants . Distributions of NSM , ADF and ASI expression intensities in wild type , tph-1 ( − ) mutants , daf-7 ( − ) mutants and tph-1 ( − ) ; daf-7 ( − ) double mutants . Values represent the sum of the expression intensities in both cells of the neuron-pair . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 014 This intricate regulation had several profound effects on food encoding . First , the regulation between daf-7 and tph-1 shaped their responses to food . The relatively monotonic increase in tph-1 expression in ADF with bacterial concentration in wildtype animals was specified by daf-7 regulation; in daf-7 ( − ) mutants , the response of tph-1 in ADF became more non-monotonic ( Figure 5A , middle ) . The non-monotonic response of daf-7 expression in ASI was also altered by the loss of daf-7 auto-regulation ( Figure 5A , right ) . Since a non-monotonic encoding scheme provides theoretical advantages in representing the range of inputs ( above ) , changes in the shape of these responses could impact how efficiently food is represented . Second , these regulatory interactions also affected the dynamic range of the gene-expression profiles . Amplification of dynamic range may contribute to increased encoding capability by increasing separation of gene-expression signals . Compared to wildtype , mutations in tph-1 and daf-7 increased the dynamic range of tph-1 expression in NSM but not in ADF , while the daf-7 ( − ) mutation dramatically decreased the dynamic range of daf-7 in ASI ( Figure 5B , D ) . Third , besides affecting average gene expression , the daf-7 ( − ) mutation also had a strong effect on the inter-individual variance of each gene-expression distribution ( Figure 5C , D and Figure 5—figure supplement 1 ) . In contrast to the effects of dynamic range , increasing inter-individual variation degrades encoding capability by increasing overlap between gene-expression signals . At several food levels , daf-7 ( − ) mutants increased the variance of tph-1 expression compared to the wildtype or the tph-1 ( − ) single mutant , whilst the daf-7 ( − ) mutation reduced the variance of daf-7 expression in ASI ( Figure 5C ) . This result provided an interesting example of regulated inter-individual variability in gene expression that is associated with a physiological output in a metazoan . Thus , the shape , dynamic range and inter-individual variance of the gene-expression responses could all be regulated by interactions between daf-7 and tph-1 . To understand the multifaceted effects of daf-7 and tph-1 signalling on fidelity , we computed the encoding accuracy of neuronal gene expression in daf-7 ( − ) and tph-1 ( − ) mutants using the same maximum likelihood framework as applied to the wildtype animals ( Figure 4C , D , and Figure 6—figure supplement 1 ) . tph-1 ( − ) mutants showed increased encoding capability in all readouts while daf-7 ( − ) mutants demonstrated consistently diminished fidelity ( Figure 6A ) , consistent with their respective effects on dynamic range and the variability of gene expression ( Figure 5B , D ) . The intermediate encoding capability of the double mutants reflects the additive effects of the mutations with regard to encoding ( Figures 5B , C , 6A ) . As with wildtype animals , combining information from multiple neurons tended to increase the accuracy in these mutants and tph-1 expression in the NSM and ADF neurons accounted for the majority of this representational capability ( Figure 6B ) . These results showed that tph-1 and daf-7 served mechanistically distinct and opposing roles in modulating the fidelity of this food representation ( Figure 6C , D ) . Whilst the transcriptional responses of daf-7 itself added little encoding capability overall ( Figures 4D , 6B ) , its regulatory role in reducing gene-expression variability facilitated the encoding power of tph-1 ( Figure 6B , C ) . Conversely , tph-1 outputs served a major encoding role in the system ( Figure 6B ) , but its regulatory effect decreased the dynamic range of transcriptional responses ( Figure 5B ) and limited the encoding accuracy of the system in wildtype animals ( Figure 6C ) . Together , our results reveal the roles of cross- and self-regulation within this circuit , and the mechanisms by which tph-1 and daf-7 tune performance . 10 . 7554/eLife . 06259 . 015Figure 6 . Cross- and self-regulation of tph-1 and daf-7 control the accuracy of internal representation of food levels . ( A ) Encoding accuracy of individual neuron-specific expression readouts in wildtype and mutant populations , as indicated by the legend . Dotted line indicates the lower bound for encoding accuracy due to chance . ( B ) Functional combinations of the neuron-specific expression readouts increase encoding accuracy in both wildtype and mutant populations . ( C ) Matrices indicating the full encoding accuracy of the combination of all gene-expression readouts in wildtype and mutant animals reveal a surprising increase in accuracy with the loss of tph-1 . ( D ) Schematic indicating the distinct mechanisms by which tph-1 and daf-7 control the representational capabilities of the system . tph-1 and daf-7 exert their effects largely via modulating dynamic range or variability in gene expression , respectively . ( E ) Mutant animals show diminished encoding accuracy relative to wildtype when only functional expression readouts ( filled symbols ) are considered . For example , only daf-7 expression in ASI ( Pdaf-7ASI ) is a functional readout in the tph-1 ( − ) mutant . ( F ) Encoding accuracy of lifespan responses in the mutants exhibit decreases that are consistent with the loss of representational capability in ( E ) . Figure 6—figure supplement 1 shows the decoding analysis for all neuron combinations across all genotypes using both maximum likelihood-based and probability-based decoders . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 01510 . 7554/eLife . 06259 . 016Figure 6—figure supplement 1 . Food encoding accuracy of all neuron combinations and lifespan outputs for wildtype and mutant animals . ( A , C , E , G ) Matrices of probability-based food encoding fidelity for single and combined neuronal outputs and lifespan outputs in the wildtype and mutant animals . These results are based on distributing the calculated raw probabilities of animals being from each food condition before assigning animals to particular food levels based on maximum likelihood . ( B , D , F , H ) Matrices of maximum likelihood-based food encoding fidelity for single and combined neuronal outputs and lifespan outputs in the wild type and mutant animals . Genotype is denoted by the colored borders around the matrices and indicated in the legend below ( A and B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06259 . 016 While our reporter system allowed us to ascertain all transcriptional outputs in all genetic backgrounds , not all of these readouts resulted in functional outputs in the mutants . For example , in daf-7 ( − ) mutants , daf-7 expression in ASI did not yield a functional protein that could mediate physiological outputs . When we accounted for the loss of the corresponding functional genetic outputs in the mutants , the overall functional representational capability was consistently diminished in comparison to wildtype ( Figure 6E ) . The loss of representational capability in either tph-1 ( − ) or daf-7 ( − ) mutants confirms that these genes have non-redundant encoding functions via two distinct mechanisms . tph-1 ( − ) mutants showed diminished functional representation primarily due to loss of tph-1 as a functional output whereas daf-7 ( − ) mutants showed diminished functional representation primarily due to increased variability in tph-1 expression ( Figure 6A , B , E ) . Remarkably , we found corresponding decreases in the accuracy of both the functional representations and lifespan outputs when daf-7 and tph-1 were mutated ( compare Figure 6E , F ) . This result is consistent with the disruption in internal representation and the reduction in lifespan-responsiveness to food in the mutants ( Figure 1C , E ) . We note that tph-1 ( − ) ; daf-7 ( − ) double mutants retained some response to food levels ( Figures 1C , 6F ) indicating the existence of additional unknown gene ( s ) that encode food . Nonetheless , the significant and correlated effects of mutating daf-7 and tph-1 on the accuracy of both gene-expression and lifespan readouts ( Figure 6E , F ) strongly indicate that tph-1 and daf-7 act as a substantial part of a multi-neuron system linking environment to lifespan .
We have shown that food abundance is functionally encoded in vivo in C . elegans by the combinatorial expression of tph-1 and daf-7 in three neuron pairs . Lifespan increases and decreases as a function of food abundance . The phenotypes of tph-1 ( − ) and daf-7 ( − ) mutants reveal that rather than serving intrinsic roles in lifespan extension , serotonin and TGFβ are signals required for bidirectional food-dependent lifespan changes ( Figure 1D ) . This bidirectional phenotype was only revealed by analysis of lifespan across a very broad range of food levels . Many genes have been implicated in lifespan modulation or DR-related responses; a systematic analysis under additional conditions may provide new insights on their roles in food-dependent responses . That daf-7 and tph-1 act in food-responsive sensory neurons , together with the bidirectional attenuation of lifespan responses in the single and double mutants suggests that loss of tph-1 and daf-7 impairs the ability to sense and/or convey information about food levels , leading to a flattened response ( Figure 1C ) . This interpretation is further supported by the observations that loss of daf-7 and tph-1 reduced the dynamic range of lifespan responses ( Figure 1E ) , and lowered the accuracy of food level representation ( Figure 6E , F ) . tph-1 and daf-7 have established functional connections to conserved pathways for regulating metabolism , cellular maintenance , reproduction and longevity , such as the insulin/IGF pathway ( Liang et al . , 2006; Shaw et al . , 2007; Narasimhan et al . , 2011 ) . Thus , this internal perception of food availability may facilitate the coordination of energy utilization and appropriate physiological and behavioral responses to optimize survival in variable environments . How much information is conveyed by daf-7 and tph-1 ? Several results indicate that daf-7 and tph-1 account for a substantial if not the majority of the food response in lifespan modulation . First , loss of both tph-1 and daf-7 reduced the dynamic range of the lifespan response to food by ∼60% ( Figure 1E ) . Second , the decoding analysis , which systematically considers the overlap between multiple response distributions ( Figure 4 ) , showed that much of the information about food levels is lost in the tph-1 ( − ) ; daf-7 ( − ) double mutant: the predictive accuracy of the lifespan response drops to 5% above random ( Figure 6—figure supplement 1H ) . Third , the predictive accuracy of gene expression and lifespan are heterogeneous—certain food levels were correctly predicted with far higher accuracy ( >75% ) than the average accuracy ( see decoding matrices in Figures 4D , 6C and Figure 6—figure supplement 1 ) . This heterogeneous performance is due in part to non-monotonic expression responses ( Figure 3C ) , as some food levels lead to similar responses . Fourth , it is important to note that an encoding system only needs to match the accuracy of the output to be good enough for regulating a biological process . The mean accuracy of lifespan responses to food thus provides a biologically relevant limit on the required performance for encoding food levels . In this light , the comparable mean accuracy between daf-7/tph-1 gene expression and lifespan indicates that the performance of the coding system is well suited to the accuracy of the process it modulates . Fifth , we found corresponding drops in the predictive accuracy of both gene expression and lifespan across daf-7 ( − ) and tph-1 ( − ) mutants ( compare Figure 6E , F ) . This result is particularly remarkable because the predictive accuracy depends on not just the means but also variances . While these results support the idea that daf-7 and tph-1 functionally encode significant information about food levels , the tph-1 ( − ) ; daf-7 ( − ) double mutant still retains some information about food , as shown by its lifespan response to food and its predictive accuracy of food ( Figures 1C , 6F ) . Therefore additional molecule ( s ) exist that cooperate with daf-7 and tph-1 in food encoding . These may include previously characterized DR regulators or undiscovered genes . Our findings provide a more nuanced view of the interactions between the serotoninergic and TGFβ neuroendocrine systems . Genetic epistasis between tph-1 and daf-7 had suggested that these two genes functioned in the same pathway ( Sze et al . , 2000; Chang et al . , 2006 ) . Our results indicate that tph-1 and daf-7 are likely to act in parallel to modulate lifespan in response to food , and that their mutual regulation represents crosstalk between these two pathways . First , we found that their epistatic relationships are food-dependent rather than hard-wired . This result is reminiscent of another food-dependent epistasis between neuropeptide and catecholamine receptors ( npr-1 and tyra-3 respectively ) that act in C . elegans sensory neurons ( Bendesky et al . , 2011 ) . Perhaps crosstalk during sensory processing leads to more complex genetic interactions . Second , tph-1 and daf-7 play distinct regulatory roles in shaping the food representation and tuning discriminatory power ( Figures 5D , 6D; discussed below ) . Third , daf-7 and tph-1 show additive phenotypes in both gene-expression responses and predictive accuracy of lifespan ( Figures 5A , 6F ) . Since the predictive accuracy reflects the information content on food , the additive loss indicates that daf-7 and tph-1 carry non-redundant information about food abundance . Together , these results point to parallel rather than sequential signalling for serotonin and TGFβ . Conceptually , the genetic regulation between serotonin and TGFβ is similar to feed-forward and feedback regulatory architectures that occur in other neural circuits operating on a shorter time-scale ( Harris and Mrsic-Flogel , 2013; Wilson , 2013 ) , and analogous mechanisms to regulate information transduction may be used for different sensory modalities . Different regulatory interactions between daf-7 and tph-1 modulate encoding capacity in specific ways , revealing novel system-level roles for these interactions . tph-1 activity limits the dynamic range , and consequently the representative capacity of tph-1 and daf-7 expression in response to food . It is possible that the increased dynamic range of daf-7 expression in tph-1 ( − ) animals may partially compensate for the loss of tph-1 as a major functional encoder . Surprisingly , tph-1 self-regulation indicates that tph-1 and serotonin signaling is likely to self-limit representative capacity by reducing the dynamic range . In contrast , the cross-regulatory effect of daf-7 is necessary to reduce variability in tph-1 expression , which represents a new component of tph-1 expression that is regulated by daf-7 . The parity in encoding performance between tph-1 and daf-7 expression and lifespan ( Figure 6E , F ) , along with the attenuation of lifespan response in the mutants , suggests that this neuronal representation has major functional information-carrying capacity . Distributing a combinatorial code across multiple neurons with distinct responses ensures robustness of signal fidelity against variability in each neuron-type , analogous to population codes in certain sensory circuits ( Singer and Gray , 1995 ) . Furthermore , a distributed code enables multicellular organisms to implement additional layers of regulation unavailable to unicellular organisms . Inter-individual variability in gene expression has been extensively studied in single cell systems where it plays key roles in diverse processes ( Eldar and Elowitz , 2010 ) . By measuring this variation in neurons using high-throughput quantitative microscopy , we highlight implications of this variability in encoding food levels in nervous system , which have not been extensively explored in intact animals and are likely to be generalizable to other neural processes . First , gene-expression variability limits the accuracy of neural food representations . Second , this variability can be regulated , as shown by the role of TGFβ in reducing variability in serotonin signaling to increase fidelity . Third , in this TGFβ and serotonin circuit , the effects of variability are tempered by the dynamic range of responses ( which can reduce the overlap between broad distributions ) and by the use of multiple neurons . Thus , the overall fidelity of information processing is shaped at least in part by an interplay of variability and other parameters . Neural encoding has traditionally been studied in the context of short-term electrical activity . However , activity-dependent transcription has been studied in diverse forms of neural plasticity that impact behavior , learning and memory ( Zhang et al . , 2005; Chang et al . , 2006; Flavell and Greenberg , 2008; Pocock and Hobert , 2010 ) . Our results suggest that gene expression in a multi-neuron system can also encode food abundance to tune physiological states at longer time scales . In general , the complexity of gene-expression responses , such as those observed in the songbird forebrain during song communication , suggest that such responses can perform computations ( Clayton , 2013 ) . It is likely that additional layers of computation in the nervous system , such as the gene-expression encoding shown here , are necessary to mediate longer-term behavior and physiology .
The following parental strains were used to generate all other used in this study: N2 ( wildtype ) , QL101 tph-1 ( n4622 ) II , QL282 daf-7 ( ok3125 ) III , EG6701 ttTi4348 I; unc-119 ( ed3 ) III; oxEx1580 , EG6699 ttTi5605 II; unc-119 ( ed3 ) III; oxEx1578 . Worms were cultured according to standard protocols ( Stiernagle , 2006 ) . The reporters for daf-7 and tph-1 expression were generated from fosmid clones carrying the wildtype tph-1 and daf-7 genomic sequences ( Source BioScience Lifesciences , UK ) . We used a recombineering pipeline ( Sarov et al . , 2006 ) to replace the coding sequence of the target gene in the fosmid with either mCherry or Venus . We then subcloned the reporter constructs flanked by the native 5′ and 3′ intergenic regions of daf-7 or tph-1 respectively into pCFJ352 and pCFJ151 for integration into defined positions of the C . elegans genome using MosSCI ( Frokjaer-Jensen et al . , 2008 ) . The tph-1::mCherry reporter ( drcSi61 ) was integrated on chromosome I at the ttTi4348 Mos locus . The daf-7::Venus reporter ( drcSi7 ) was integrated on chromosome II at the ttTi5605 Mos locus . We used standard genetic techniques to construct strains carrying both fluorescent reporters in various genetic backgrounds . Cell-specific rescue lines used in Figure 3 were generated in two ways . The tph-1 ADF- and NSM-expression constructs were from Zhang et al . ( 2005 ) . The ASI-specific daf-7 construct was generated in two steps; first we constructed a GFP reporter driven by the promoter of gpa-4 , which is specifically expressed in ASI ( Jansen et al . , 1999 ) , using the recombineering pipeline ( Sarov et al . , 2006 ) above . This was then subcloned into the pCFJ151 vector for MosSCI targeting of ttTi5605 locus ( Frokjaer-Jensen et al . , 2008 ) . This recombinant plasmid was then digested with NheI and AvrII to linearize it for use as a vector backbone . Using a daf-7-containing fosmid as template , we amplified a PCR product that contained the entire daf-7 gene from start to stop codon including its native introns and added overlapping linkers to the gpa-4 5′ and 3′ UTRs at the respective ends of the PCR product ( Primers DP101 , 5-TCAACAAGCTCAGGAGGTAGCGGCGATGTTCATGGCATCTTCAC-3 and DP102 5-TCTAGTTAAACGTTTATGAGCAACCGCATTTC-3 , underlined sequences represent gpa-4 linkers ) . A second PCR product was then amplified from the undigested recombinant gpa-4::GFP reporter plasmid above , which contained the entire gpa-4 3′ intergenic sequence and added linkers to the daf-7 gene and the multiple cloning site ( MCS ) of the pCFJ151 vector ( Primers DP103 5-CGGTTGCTCATAAACGTTTAACTAGATTAGCACAAATG-3 and DP104 5-TTGACTAGAGGGTACCAGAGCTCACTAAATAATAAAAAACCTAAAATTGTTTC-3 , underlined sequences represent daf-7 and MCS linkers , respectively ) . These fragments were then seamlessly joined into the digested vector backbone using Gibson assembly ( New England Biolabs , Ipswich , MA ) to yield the final vector , which contained the daf-7 gene flanked by all the regulatory elements of gpa-4 . To generate transgenic animals carrying these cell-specific rescue constructs we injected them into both N2 and tph-1 ( n4622 ) ; daf-7 ( ok3125 ) double mutants at a concentration of 10 ng/μl along with an ofm-1::GFP reporter ( Miyabayashi et al . , 1999 ) at 25 ng/μl . Transgenic animals were maintained by selecting for GFP-positive animals under a fluorescent dissecting scope . Our daf-7 and tph-1 reporters contain similar regulatory sequences , and are expressed in the same patterns , as reporters that have been extensively validated in the literature . tph-1 reporters have been validated by experiments showing corresponding changes under a variety of environmental conditions , including starvation ( Liang et al . , 2006; Cunningham et al . , 2012 ) , exposure to pathogenic bacteria ( Zhang et al . , 2005 ) , and hypoxia ( Pocock and Hobert , 2010 ) . The supplemental information in Liang et al . ( 2006 ) showed that serotonin levels in ADF were reduced in animals starved for 8 hr . We further verified that serotonin levels in NSM of wild type adults were also reduced after food deprivation , in agreement with the behavior of the Ptph-1::mCherry reporter . The serotonin levels were determined by anti-serotonin immunofluorescence staining in 4 day old adult worms . Worms subjected to our DR protocol at two food levels ( no food and 6 × 108 cells/ml ) were used for immunostaining using an anti-serotonin primary antibody ( Immunostar Inc . , Hudson , WI ) and Alexa594 coupled secondary antibody as previously described ( Loer and Kenyon , 1993 ) . Day 4 adults deprived of bacteria showed less serotonin staining ( 1 . 19 ± 0 . 29 AU; n = 26 ) than those exposed to 6 × 108 cells/ml of bacteria ( 1 . 94 ± 0 . 28 AU; n = 21 ) ; these differences were significant ( p = 1 . 38 × 10−11; Students T-Test ) . daf-7 reporters have been previously validated by experiments showing corresponding changes in both daf-7::GFP and daf-7 mRNA when switching from starved to fed states , under pheromone induction of dauer , and during exit of dauer ( Ren et al . , 1996 ) . In particular , daf-7 is down-regulated upon starvation , and both daf-7::GFP and daf-7 mRNA levels increase when starved animals are fed ( Ren et al . , 1996 ) , consistent with our observations ( Figure 3 ) . For expression studies in middle-aged adults used in this study , we found that using fluorescent reporters was the most reliable method . daf-7 and tph-1 are expressed in a small number of cells , resulting in very low mRNA levels that were not reliably measured by real-time quantitative PCR because the signals were close to background ( data not shown ) . Furthermore , procedures for whole-mount immunohistochemistry ( Loer and Kenyon , 1993 ) and single-molecule fluorescence in situ hybridization ( Raj et al . , 2008 ) resulted in high background fluorescence that precluded accurate measurements in older adults using these methods . Our DR protocol was derived from two previously published methods for DR on agar plates ( Greer et al . , 2007; Ching and Hsu , 2011 ) . After growing of E . coli ( OP50 ) overnight in LB at 37°C shaking , the bacterial cultures were shocked with streptomycin at 50 μg/ml for 30 min whilst shaking and then chilled on ice for 15 min before being centrifuged at 4500×g for 25 min . The supernatant was then decanted and the bacteria were resuspended in S Basal containing streptomycin ( 50 μg/ml ) . The volume of S Basal + streptomycin required for resuspension was determined by measuring the OD600 of a 10-fold dilution of the overnight culture prior to centrifugation and then calculating the volume required to give the resuspended culture a theoretical OD600 of 56 , which corresponds to a total cell count ( live + dead ) of ∼1 × 1010 cells/ml . Serial dilutions of this high-density stock were used to derive all subsequent bacterial concentrations . Stocks were stored at 4°C until plates were ready to seed . We used NGM plates supplemented with both streptomycin and carbenicillin , each at 50 μg/ml . The use of dual antibiotics ensured that bacteria were unable to proliferate on the agar plates . Plates were seeded with a dispensing pipette to ensure that all plates received an equal volume of liquid . Plates were typically seeded 2–3 days in advance to allow the bacterial lawns to dry out before worms were put on the plates . Many C . elegans ageing studies use fluoro-2′-deoxyuridine ( FuDR ) to inhibit the germline of experimental animals in order to eliminate progeny that may otherwise confound analyses . However , the use of FuDR is problematic as the germline in C . elegans is a major regulator of longevity ( Hsin and Kenyon , 1999; Lin et al . , 2001 ) and its use can cause gene-specific effects on lifespan ( Aitlhadj and Sturzenbaum , 2010 ) . Our protocol uses a novel method to eliminate production of progeny by inhibiting formation of the eggshell of fertilized C . elegans embryos through RNAi of egg-5 ( Cheng et al . , 2009; Parry et al . , 2009 ) resulting in their death . Several results indicate that egg-5 ( RNAi ) does not affect food-dependent lifespan responses . We observed similar lifespan responses for wildtype animals in experiments where egg-5 ( RNAi ) was omitted ( Figure 1—figure supplement 1 ) . Moreover , we also observed similar lifespan responses under dietary deprivation ( Kaeberlein et al . , 2006; Lee et al . , 2006 ) and at the intermediate to high food levels where lifespans have been measured in other publications ( Greer and Brunet , 2009; Greer et al . , 2009 ) . Lifespan assays were performed on 6 cm CellStar ( Greiner Bio-One , UK ) plates at density of 15 worms that were passaged to fresh plates by manual transfer using a platinum wire pick . To avoid physical damage to the worms during transfer , animals were floated off the pick by immersing it into a 10 μl droplet of S Basal + streptomycin on the surface of the new plate . NGM agar plates containing carbenicillin and streptomycin were always seeded with a single 225 μl aliquot of the appropriate bacterial concentration resuspended in S Basal + streptomycin , while plates with 0 cells/ml , equivalent to dietary deprivation , were seeded with S Basal + streptomycin as a control . Animals for all strains were raised on live OP50 bacteria for two generations at 20°C . Synchronized L4-stage progeny of the F2 parents were manually transferred to NGM plates supplemented with 1 mM IPTG and 50 μg/ml carbenicillin that were seeded with HT115 bacteria expressing dsRNA targeting egg-5 . Animals were exposed to egg-5 RNAi for 24 hr before being transferred to NGM + streptomycin + carbenicillin plates seeded with our baseline food concentration of 2 × 109 cells/ml . On day 2 of adulthood , animals were shifted to the desired DR food level and temperature and then transferred at regular intervals to fresh plates until day 14 of adulthood according to a defined scheme ( Figure 1—figure supplement 2 ) . We initiated DR on day 2 adults because it had an intermediate effect on lifespan as compared to initiation on day 1 , 3 , and 4 ( Figure 1—figure supplement 1 ) . We reasoned this intermediate effect served as a potentially more sensitive window to detect factors that modify food responses , as the effects of food were non-saturating ( unlike the effect of day 3 DR initiation , which was indistinguishable from day 4 DR initiation ) . Animals were scored for movement upon gentle prodding with a wire pick; failure of response was scored as death . Animals were scored for death at every transfer point and then daily after the last transfer point . Lifespans were subjected to Kaplan–Meier analysis and significance was assessed using both Log-Rank and Wilcoxon tests . Animals in imaging experiments were cultured on 10 cm CellStar ( Greiner Bio-One ) to allow for a higher number of worms ( ∼100 ) than lifespan assays . For quantitative imaging , animals were grown and subjected to DR under identical conditions to those assayed for lifespan , except that animals were washed from plate to plate with the same schedule instead of manual transfers . Due to the greater number of worms in these assays , animals were transferred between plates by washing with S Basal + streptomycin . 10 cm plates were seeded with five 225 μl aliquots of bacteria or S Basal + streptomycin in a cross-like formation . In our initial imaging experiments using the daf-7 and tph-1 reporters in a wildtype background , L1-stage larvae were synchronized collecting animals that hatched in a 2-hr window and transferring them to fresh 10 cm NGM plates seeded with OP50 at 20°C . These animals were then harvested 36 hr later once they had reached the L4 stage and washed on to plates for egg-5 RNAi . After 24 hr on the RNAi plates the animals were washed to the baseline food level of 2 × 109 cells/ml for 1 day before the initiation of DR on the second day of adulthood . Animals were transferred to fresh plates on the third and fifth days of adulthood and imaged on the sixth day . Imaging experiments involving comparisons to strains containing the daf-7 ( ok3125 ) mutation could not be treated as above , as this mutation causes a severe egg-laying defect . Instead , strains for these experiments were grown for 2 generations as before , and then the gravid F2 adults were collected and treated with Sodium Hypochlorite to break open the animals and liberate their eggs ( Stiernagle , 2006 ) . The eggs were then deposited onto NGM plates seeded with OP50 for either 72 hr , in the case of strains in a daf-7 ( ok3125 ) -containing background , or 48 hr for non-daf-7 ( ok3125 ) -containing strains . L4-stage animals were then harvested after these respective intervals and then treated exactly as above . For our quantitative imaging studies , animals subjected to our DR protocol were imaged at day 6 of adulthood using a custom microfluidic platform ( Chung et al . , 2008; Crane et al . , 2012 ) . Animals were suspended in S Basal + streptomycin on day 6 of adulthood and introduced into our custom microfluidic device via pressure driven flow ( Chung et al . , 2008; Crane et al . , 2012 ) . Briefly , microfluidic devices were manufactured in polydimethylsiloxane using standard multilayer soft lithographic techniques ( Unger et al . , 2000 ) and covalently bonded to a glass coverslip via oxygen plasma treatment . On the device , individual animals were sequentially directed into and trapped within an imaging channel gated by pressure-driven on-chip valves ( Unger et al . , 2000 ) under the control of custom LabVIEW software ( Figure 3—figure supplement 1 ) . Dense , 2-micron fluorescent z-stacks through head of each worm were collected using a standard epifluorescence microscope ( Nikon Ti-E inverted microscope ) with a 40× oil objective ( 1 . 3 NA ) and a Hamamatsu Orca R2 camera . Red and green fluorescent intensities for each of our fluorescent reporters were collected simultaneously using an Optosplit II emission splitter and stored for analysis . We automated the image acquisition and image processing using custom LabVIEW and MATLAB scripts; the analysis software is deposited at https://github . com/meizhan/SVMelegans . Z stacks from our quantitative imaging studies were loaded into MATLAB to be analyzed for single-cell expression ( Figure 3—figure supplement 1 ) . To identify neuron-pairs and their locations within the imaging plane , maximum projections were computed and a thresholding algorithm was utilized to locate individual fluorescent cells . Identifications of the cells were then computed based on relative distances and locations within the worm head . For quantification , the three dimensional volume around each cellular location was extracted from the full z-stack and intensity was integrated over a consistent number of the brightest pixels , which fully encapsulate the entire cell in all cases . To avoid potential interference from condition-specific changes in the gut auto-fluorescence , the background intensity was calculated for cell pairs near the gut ( ADF , ASI ) , via estimation of the mode of the intensity distribution in a volume around the neuron . This background intensity value was subtracted from the integrated fluorescence to obtain the final output . To estimate the ability of our gene-expression readouts and lifespan responses to encode information about the food inputs , we applied a Bayes classifier with fivefold cross validation ( Dayan and Abbott , 2005 ) . To do this we randomly segmented the data into five test groups . For each test group , we computed expression or lifespan probability distributions for each food level based on the remaining data ( the training set ) . We employed a multivariate Gaussian distribution to fit the expression data and a Weibull distribution for lifespan data . Using these probability distributions , we then calculated the conditional probability that each member of the test group had their particular expression or lifespan output if they had been subjected to each of the food levels . The accuracy of these raw probability values can be found in Figure 6—figure supplement 1 panels A , C , E and G . To make a final inference about the food level each worm was subjected to , the animal was assumed to have come from the food level with the highest conditional probability . The encoding accuracy of this maximum likelihood model can be found in Figure 6—figure supplement 1 panels B , D , F , and H . As with all measurements of variance , these accuracy estimates should be considered as lower bounds due to experimental noise . In the decoding analysis for expression data , we limited the data from Figure 5—figure supplement 1 to animals with corresponding data from all three neuron pairs ( N ≥ 726 for WT , N ≥ 110 for tph-1 ( − ) , N ≥ 83 for daf-7 ( − ) and N ≥ 54 for tph-1 ( − ) ;daf-7 ( − ) for each food level ) . For the lifespan data , we used the pooled aggregate of all of our lifespan data ( highlighted rows in Figure 1—source data 1 for wildtype and pooled mutant data in Figure 1—source data 2 ) . | To maximize their chances of survival , animals need to be able to sense changes in the abundance of food in their environment and respond in an appropriate manner . The nervous system is able to sense cues from the environment and coordinate responses in the whole organism , but it is not clear how this leads to long-term changes in the organism's biology . In nematode worms , two genes called daf-7 and tph-1 appear to be involved in connecting the sensing of food availability with changes in the biology of the organism . The daf-7 gene encodes a hormone , while tph-1 encodes an enzyme that makes a neurochemical called serotonin . Here , Entchev , Patel , Zhan et al . found that daf-7 and tph-1 genes are active in three pairs of neurons in nematode worms . The experiments show that these neurons collectively form a circuit that carries information about the abundance of food , which leads to changes in how long the worms live . When this circuit was disrupted by removing these genes , the worms' ability to adjust their lifespan in response to changes in the availability of food was weakened , likely because they were unable to sense food . The experiments also show that the circuit regulates itself , largely because daf-7 and tph-1 are able to control each-other's activity . Together , these results suggest that changing the activity of certain genes in neurons enables nematode worms to alter their biology in response to changes in the availability of food . Neurons in the brain use electrical activity to communicate and process information and Entchev , Patel , Zhan et al . 's findings imply that gene activity can also perform a similar role . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | A gene-expression-based neural code for food abundance that modulates lifespan |
The requirement for vernalization , a need for prolonged cold to trigger flowering , aligns reproductive development with favorable spring conditions . In Arabidopsis thaliana vernalization depends on the cold-induced epigenetic silencing of the floral repressor locus FLC . Extensive natural variation in vernalization response is associated with A . thaliana accessions collected from different geographical regions . Here , we analyse natural variation for vernalization temperature requirement in accessions , including those from the northern limit of the A . thaliana range . Vernalization required temperatures above 0°C and was still relatively effective at 14°C in all the accessions . The different accessions had characteristic vernalization temperature profiles . One Northern Swedish accession showed maximum vernalization at 8°C , both at the level of flowering time and FLC chromatin silencing . Historical temperature records predicted all accessions would vernalize in autumn in N . Sweden , a prediction we validated in field transplantation experiments . The vernalization response of the different accessions was monitored over three intervals in the field and found to match that when the average field temperature was given as a constant condition . The vernalization temperature range of 0–14°C meant all accessions fully vernalized before snowfall in N . Sweden . These findings have important implications for understanding the molecular basis of adaptation and for predicting the consequences of climate change on flowering time .
The sessile nature of plants necessitates that they modulate most aspects of their growth and development in response to external conditions . One aspect of this is the alignment of developmental transitions with seasonal cues . A major seasonal cue is temperature and plants have evolved the ability to integrate daily fluctuations in external temperature in order to monitor long-term trends ( Aikawa et al . , 2010 ) . Exposure to weeks of cold temperature accelerates the transition to flowering in a process called vernalization . In Arabidopsis thaliana vernalization involves the quantitative epigenetic silencing of FLC ( Michaels and Amasino , 1999; Sheldon et al . , 1999 ) . Cold exposure promotes a cell-autonomous epigenetic switch at FLC in an increasing proportion of cells ( Angel et al . , 2011 , 2015 ) . This epigenetic switching mechanism requires a Polycomb complex associated with PHD proteins ( De Lucia et al . , 2008 ) , including the cold-induced VIN3 ( Sung and Amasino , 2004 ) . This enables activation of FT , is a potent activator of flowering in A . thaliana ( Searle et al . , 2006 ) . At a standard vernalization temperature of 5°C , the length of cold required to achieve complete epigenetic silencing varies between A . thaliana accessions and this maps to non-coding cis polymorphisms in FLC ( Coustham et al . , 2012; Li et al . , 2014 ) . Accessions collected from northerly latitudes typically require longer vernalization , for example , the accession Lov-1 originates from near the northerly limit of the Arabidopsis range in Lövvik , North Sweden ( 62 . 5°N ) and requires three months of vernalization to fully accelerate flowering ( Shindo et al . , 2006; Coustham et al . , 2012 ) . Effective temperature ranges for vernalization have been determined empirically for different plant species , many of which have been incorporated into chilling unit models that are widely used in agriculture ( Byrne and Bacon , 1992 ) . A genetically informed photothermal model for flowering in A . thaliana has assumed that vernalization occurs when daytime hourly temperatures are higher than 0°C and lower than 6°C ( Wilczek et al . , 2009; Chew et al . , 2012 , 2014 ) . However , accessions from southern Europe have been found to vernalize at constant temperatures significantly higher than 6°C ( Wollenberg and Amasino , 2012 ) , suggesting that accessions from northerly latitudes might vernalize most efficiently at relatively low temperatures . Here , we show this is not the case and find that vernalization in a range of A . thaliana accessions is most effective across a relatively high temperature range with the N . Swedish accession Lov-1 , showing maximal vernalization at 8°C . We show that vernalization is complete before snowfall in N . Sweden with the plants flowering immediately upon snowmelt . Vernalization responsiveness in the field matched that when the average monthly temperature was given as constant conditions . Our work has important implications for modeling flowering time and predicting the impact of climate change .
In order to investigate natural variation for vernalization temperature requirement in A . thaliana accessions we selected several genotypes that represent most of the major FLC haplotypes ( Li et al . , 2014 ) : Lov-1 ( Lövvik , N . Sweden—latitude 62 . 5°N ) , Var2-6 ( Vårhallen , S . Sweden—latitude 55 . 58°N ) , Ull2-5 ( Ullstorp , S . Sweden—latitude 56 . 06°N ) , Edi-0 ( Edinburgh , UK—latitude 55 . 95°N ) and the reference Columbia line containing FRIGIDA ( Col FRISf2 , [Michaels and Amasino , 1999] ) ( Figure 1—figure supplement 1 ) . All genotypes were vernalized for varying periods at different constant temperatures between 0°C and 14°C and the efficiency of vernalization assayed by determining flowering time ( Figure 1A-E ) . All the genotypes showed limited vernalization after 4 and 6 weeks exposure to 0°C and vernalized more efficiently at all other temperatures . Col FRISf2 and Edi-0 were most effectively vernalized after 4 , 6 or 12 weeks at 2°C , 5°C and 8°C and still vernalized relatively efficiently at 12°C and 14°C ( Figure 1A and C ) . Even after 2 weeks of cold at 2°C and 8°C the flowering of Col FRISf2 plants was similar , so lack of any difference was not due to vernalization being close to saturation ( Figure 1—figure supplement 2 ) . Ull2-5 showed similar temperature sensitivity to Col FRISf2 and Edi-0 , but required 12 weeks for vernalization to be fully effective ( Figure 1D ) . In contrast , Lov-1 and Var2-6 plants showed a differential response to temperature with 2 and 12°C less effective than 5 and 8°C after 6-weeks vernalization ( Figure 1B and E ) . For Lov-1 the only temperature that resulted in flowering after 4 weeks exposure was 8°C and although the enhanced effect of this temperature diminished over time , 8°C consistently resulted in the most effective vernalization ( Figure 1B ) . Thus , the different accessions show distinct temperature profiles for vernalization and all require temperatures higher than 0°C . 10 . 7554/eLife . 06620 . 003Figure 1 . Vernalization responses at a range of constant temperatures . Days to flower were recorded for five genotypes after 0 ( crosses ) , 4 ( red squares ) , 6 ( blue triangles ) and 12 ( green circles ) weeks vernalization at a range of temperatures , n ≥ 10 . NV = non-vernalized . Error bars = ±S . D . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 00310 . 7554/eLife . 06620 . 004Figure 1—figure supplement 1 . Map showing accession collection sites . Image generated using Google Maps ( GeoBasis-DE/BKG 2009 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 00410 . 7554/eLife . 06620 . 005Figure 1—figure supplement 2 . 2 week vernalization of Col FRISf2 does not reveal differential response to 2 and 5°C treatments . Days to flowering recorded after 2 weeks vernalization at a range of temperatures . n = 12 . NV = nonvernalized . Error bars = ±S . D . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 005 The requirement for longer cold for effective vernalization in the Lov-1 accession has previously been shown to involve quantitative variation in accumulation of epigenetic silencing of FLC ( Shindo et al . , 2006; Coustham et al . , 2012 ) . We compared this quantitative variation in the silencing of Col FRISf2 and Lov-1 FLC alleles after 4 weeks cold exposure at 2 , 5 , 8 , 12 and 14°C ( Figure 2—figure supplement 1A and B ) . In contrast to Col FRISf2 , the Lov-1 allele re-activated after 30 days in the warm after vernalization at all the tested temperatures . However , the degree of re-activation was lowest after vernalization at 8°C , consistent with vernalization being most effective at this temperature . Similarly , 6 weeks vernalization at 8°C resulted in lower FLC re-activation post-cold and higher levels of FT induction than 5°C , with similar VIN3 expression ( Figure 2A–C ) . Epigenetic silencing of FLC is associated with Polycomb silencing and accumulation of H3K27me3 over the gene body ( Angel et al . , 2011; Yang et al . , 2014 ) . In Lov-1 it takes longer to accumulate the H3K27me3 , mainly due to lower starting levels ( Coustham et al . , 2012 ) . We found similar accumulation of gene body H3K27me3 in the Col FRISf2 FLC allele at 5 , 8 or 14°C , but differential H3K27me3 accumulation in the Lov-1 allele ( Figure 2D , Figure 2—source data 1 ) . Vernalization at 8°C resulted in higher levels of H3K27me3 compared to 5 or 14°C ( Figure 2D ) , suggesting that the Polycomb silencing is most effective at 8°C for the Lov-1 FLC allele . 10 . 7554/eLife . 06620 . 006Figure 2 . Quantitative PCR and ChIP analyses of plants vernalized at 5°C , 8°C and 14°C . Changes in FLC ( A ) , VIN3 ( B ) and ( C ) FT expression were determined directly after 6 weeks of cold exposure ( T0 ) and again after 10 ( T10 ) and 30 ( T30 ) days subsequent growth at 20°C . Two-tailed Student's t-test results: *p < 0 . 05 , ***p < 0 . 005 . n = 3 . Error bars = ±S . D . ( D ) H3K27me3 levels over the FLC locus were higher for Lov-1 after 6 weeks vernalization at 8°C than 14°C or 5°C ( samples were harvested 30 days post cold ) . **** p < 0 . 0001 , Wilcoxon matched-pairs signed rank test on measurements for 12 primer pairs . Error bars = ±S . E . M . NV = nonvernalized , DNF = did not flower and ns = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 00610 . 7554/eLife . 06620 . 007Figure 2—source data 1 . Primers used for qPCR ChIP . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 00710 . 7554/eLife . 06620 . 008Figure 2—figure supplement 1 . FLC expression determined after 4 weeks of vernalization at a range of temperatures . Quantitative PCR ( qRT-PCR ) analysis showing FLC expression levels before cold ( hatched ) , after 4 weeks vernalization ( white ) and after 10 days ( grey ) and 30 days ( black ) subsequent growth at 20°C . *p = 0 . 0038 two-tailed Student's t-test . n = 3 . NV = non-vernalized . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 008 The relatively high temperature range for vernalization of the N . Swedish accession was surprising given that flower buds appeared within 2 weeks of the snowmelt on native Arabidopsis at the N . Swedish Lövvik site ( Figure 3—figure supplement 1 ) . This early flowering may limit herbivory and help in the competition for nutrients ( Kawagoe and Kudoh , 2010; Akiyama and Agren , 2012 ) . A long-term ( >5 year study ) of the natural populations at several N . Swedish sites throughout the High Coast region , showed most populations behaved as winter annuals with germination occurring predominantly in August and September with no spring germination ( Figure 3—source data 1 ) . The rapid flowering after snowmelt suggests that vernalization must have occurred before the end of November given the recurrent snow cover and low temperatures at the Lövvik site over the winter months ( Figure 3—figure supplement 2 , Figure 3—figure supplement 3A ) . Hourly climate data collected near Lövvik between 1st August ( the earliest germination date observed for natural populations ) until snow cover between 2008 and 2013 show an average air temperature of ~8°C ( Figure 3—figure supplement 3A ) . Analysis of national data ( 1st August—30th November ) also revealed an overall average autumn daily average temperature of 8 . 86°C between 1961 and 2008 ( SD = 0 . 63 ) with over 86% of days falling within the range identified as being effective for Lov-1 vernalization , ( 0°C , 15°C ) ( Figure 3—figure supplement 3B ) . The agreement of average autumn temperatures with the effective vernalization temperatures identified for the Lov-1 reinforced the view that epigenetic silencing of FLC would occur before snowfall . We tested the hypothesis of a seasonal shift in the timing of vernalization in N . Sweden by setting up field experiments close to the Lövvik site in autumn 2011 and 2012 ( locations shown in Figure 3—figure supplement 4 ) . Seedlings were transplanted into the field at the beginning of September and then transferred to a warmed greenhouse at three time points during autumn ( Figure 3A , Figure 3—figure supplement 5A ) . This enabled us to explicitly test whether 12 weeks of growth preceding winter would be sufficient to fully vernalize Lov-1 . Flowering time of the different cohorts showed that vernalization was complete by the end of November in both 2011 ( Figure 3B ) and 2012 ( Figure 3—figure supplement 5B ) . Furthermore , plants left to overwinter in the field flowered rapidly at snowmelt , at the same time as the native A . thaliana population ( Figure 3—figure supplement 6 ) . 10 . 7554/eLife . 06620 . 009Figure 3 . Field experiments reveal vernalization occurs in autumn in northern Sweden . ( A ) Date of sowing and plant transfers to the greenhouse are shown with hourly soil surface temperatures recorded during autumn 2011 . ( B ) Days to flower recorded after plants were transferred to a warmed greenhouse at three time points during autumn: Transfer 1 ( black ) , Transfer 2 ( grey ) and Transfer 3 ( white ) . n ≥ 10 . Error bars = ±S . D . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 00910 . 7554/eLife . 06620 . 010Figure 3—source data 1 . Developmental stage of natural Arabidopsis thaliana populations in spring in the High Coast area of N . Sweden ( 62 . 5°N ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 01010 . 7554/eLife . 06620 . 011Figure 3—figure supplement 1 . The Lov-1 natural population flowers rapidly after snowmelt in spring . Photographs of representative Lov-1 rosettes taken ( A ) before snow cover and ( B ) immediately after snowmelt ( green markers indicate rosette size ) . ( C ) Evidence of stem elongation was apparent 16 days post snowmelt . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 01110 . 7554/eLife . 06620 . 012Figure 3—figure supplement 2 . Snow consistently covers and protects plants from subzero air temperatures during winter . ( A ) Snow cover and melt dates recorded over 47 years . ( B ) Box plots of average snow depth recorded through the year . ( C ) Air and soil temperatures recorded simultaneously during winter 2008/2009 . Green and grey boxes = median to 1st and 3rd quartile , respectively . Upper and lower whiskers represent 1 . 5* interquartile range ( IQR ) or highest/lowest values . Blue crosses = outlier values greater than 1 . 5*IQR . Blue crosses = outlier values . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 01210 . 7554/eLife . 06620 . 013Figure 3—figure supplement 3 . Temperature records from N . Sweden near Lövvik . ( A ) Hourly air temperature collected between 2008 and 2013 . Grey shading highlights the temperatures used to calculate the mean values shown for five consecutive Lov-1 autumn growing seasons . ( B ) Box plots of mean average daily temperatures recorded during autumn ( 1st August—30th November ) over 47 years . Dashed red lines indicate 0°C and 15°C—the upper and lower temperature thresholds identified for Lov-1 vernalization . Green and grey boxes = median to 1st and 3rd quartiles , respectively . Upper and lower whiskers represent 1 . 5* IQR or highest/lowest values . Blue crosses = outlier values greater than 1 . 5*IQR . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 01310 . 7554/eLife . 06620 . 014Figure 3—figure supplement 4 . Field locations and climate data collection sites in Sweden . Hourly temperature data were collected in Eden . Swedish climate data were provided by Swedish Hydrological and Meteorological Institute weather stations located in Härnösand . Plants for the 2011 and 2012 field experiments were germinated in Sundsvall and transferred to Ramsta . ( Map courtesy of Google , GeoBasics-DE/BKG 2009 . ) DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 01410 . 7554/eLife . 06620 . 015Figure 3—figure supplement 5 . Sweden field experiments results 2012 . ( A ) Date of sowing and plant transfers to the greenhouse are shown with hourly soil surface temperatures recorded during autumn 2012 . ( B ) Days to flower recorded after plants were transferred to a warmed greenhouse at three time points during autumn: Transfer 1 ( black ) , Transfer 2 ( grey ) and Transfer 3 ( white ) . n ≥ 10 . Error bars represent ±S . D . Mann–Whitney U test results: ****p < 0 . 0001 ( U value: 10 . 50 ) , ***p = 0 . 0009 ( U value: 34 ) . DNF = did not flower . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 01510 . 7554/eLife . 06620 . 016Figure 3—figure supplement 6 . Plants flowered synchronously with natural populations after 5 months of continuous snow cover . ( A ) Surface temperature recorded at Ramsta indicating that overwintered plants were continuously covered by snow during winter 2012 . ( B ) Representative images of the overwintered cohort with floral buds visible . ( C ) Percentage plants with visible buds on 26th April 2013 , 5 days after snowmelt and ( D ) Image of natural population taken 26th April 2013 . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 01610 . 7554/eLife . 06620 . 017Figure 3—figure supplement 7 . Genetic map showing Lov-1 introgressed region on chromosome 5 . ( A ) Vertical lines indicate PCR-based markers used to distinguish between Col-0 ( light grey ) and Lov-1 ( dark grey ) regions . NILLov-1 introgression line contains the Lov-1 FLC locus ( region outlined in green ) . ( B ) Markers used to map the introgressed regions on Chromosome 5 . The positions correspond to AGI coordinates . Where the marker is a simple sequence length polymorphism ( SSLP ) , the product size is shown for Col-0/Lov-1 . Where the marker is a Cleaved Amplified Polymorphic sequence ( CAPS ) , the enzyme required to digest the PCR product of the specified accession is given . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 017 In order to link the flowering time changes with the changed epigenetic silencing at FLC we included a near isogenic line carrying the Lov-1 FLC allele ( NILLov-1 ) in the genetic background of Col FRISf2 in the field experiments . This line was generated through six generations of introgression and had been genotyped with markers to define the introgressed region ( Figure 3—figure supplement 7 ) . NILLov-1 took longer to flower than Col FRISf2 after the first two transfers in 2012 ( Figure 3—figure supplement 5B ) . This revealed the clear contribution of the Lov-1 FLC allele to differential vernalization response under field conditions , which likely involves the four non-coding polymorphisms in FLC close to the nucleation site of the PHD-PRC2 previously defined as underpinning the molecular variation in FLC epigenetic silencing between Lov-1 and Col FRISf2 ( Coustham et al . , 2012 ) . Expression analysis in the perennial species Arabidopsis halleri growing under natural field conditions has shown that plants average temperature over long-term scales ( Aikawa et al . , 2010 ) . It was therefore interesting that the optimal vernalizing temperature for Lov-1 matched the average temperature over the 3-month season when vernalization occurred ( Figure 3B ) . We therefore compared vernalization response in the different transplant intervals with vernalization in constant temperatures equivalent to the average temperature of the field conditions ( Figure 4—source data 1 ) . The different genotypes showed temporal differences in vernalization responsiveness over the three transplant periods in the field . Remarkably , vernalization responsiveness was very similar when the average field temperature was given as a constant temperature , with each genotype showing a different overall profile ( Figure 4 ) . Indeed , the match is remarkable given the daily oscillations in temperature especially in the transplant 1 period ( Figure 3A , Figure 3—figure supplement 5A ) . How plants integrate these fluctuating temperatures over such long timescales is an important area for future molecular dissection . 10 . 7554/eLife . 06620 . 018Figure 4 . Prediction of vernalization response under field conditions . Days to flower recorded after the three transplants during field experiments in 2011 and 2012 are shown in grey . Red dashed lines indicate changes in flowering time estimated by flowering time results observed after vernalization at constant temperatures . Error bars represent ±S . D . n ≥ 10 , DNF = did not flower . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 01810 . 7554/eLife . 06620 . 019Figure 4—source data 1 . Cabinet flowering time data were selected where conditions most closely matched mean temperatures recorded during 2011 and 2012 field experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 01910 . 7554/eLife . 06620 . 020Figure 4—figure supplement 1 . Accumulation of temperatures within different effective vernalization ranges . ( A ) Predicted accumulation of effective vernalization weeks during 2011 and ( B ) 2012 field experiments . Red and blue lines indicate accumulated hours ( 0°C , 6°C ) and daily average temperatures 0°C , 15°C ) respectively . The green line indicates maximal temperature accumulation under constant 8°C growing conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 06620 . 020 All the genotypes analysed were found to vernalize effectively during autumn ( Figure 3B , Figure 3—figure supplement 5B ) , however they have been shown to differ in their seed dormancy ( Atwell et al . , 2010 ) ; accessions from N . Sweden generally have much lower seed dormancy requirement than those from further south ( Debieu et al . , 2013 ) . Thus , the low seed dormancy of Lov-1 would enable germination to occur early enough to exploit the whole of the N . Swedish autumn for vernalization . The increased seed dormancy of S . Swedish accessions ( e . g . , Ull2-5 ) is likely to delay germination leading to the necessity for vernalization in some years to extend into winter in S . Sweden . It is interesting to speculate that the reduced effectiveness of temperatures below 5°C for other Swedish accessions Lov-1 and Var2-6 ( Figure 1B , E ) could prevent premature vernalization occurring during unseasonal cool periods in early autumn . Our data also show that the ( 0°C , 6°C ) temperature range widely used to estimate vernalization in A . thaliana ( Wilczek et al . , 2009 ) would only predict partial vernalization of later flowering accessions during our field experiments ( Figure 4—figure supplement 1 ) . Our data suggest that raising the upper threshold temperature to 15°C would improve estimates of vernalization progress for later flowering accessions under natural field conditions . In summary , we have employed a combination of molecular and ecological approaches to connect temperature-induced molecular changes at FLC with ecologically significant effects in the field . We show that growth at the northern limit of the A . thaliana species range has involved a seasonal shift in the timing of vernalization . Perhaps as a response to selection in these extreme conditions one N . Swedish accession , Lov-1 , shows a more distinct vernalization temperature optimum that matches the average historical temperature for August-November in that geographical region ( Figure 1 , Figure 2 , Figure 3—figure supplement 3A ) . Early germination enables vernalization to complete before snowfall and allows flowering to occur directly after snowmelt when the photoperiod and ambient temperatures increase . Rises in global temperature have already reduced vernalization periods to an extent that has impacted the phenology of a range of plant species ( Fitter and Fitter , 2002; Cook et al . , 2012 ) . Studies such as this are therefore important to understand how rapidly populations might adapt under future climate scenarios .
Genotypes used , standard growth and vernalization conditions have been described previously ( Shindo et al . , 2006 ) . Briefly , plants were sown in a randomized design and stratified for 3 days at 4°C . Seedlings were grown for 7 days at 22°C and then vernalized in cabinets at 14°C , 12°C , 10°C , 8°C ( all in Sanyo ( Moriguchi , Japan ) MLR-351H cabinets ) , 5°C ( walk-in vernalization room ) , 2°C ( modified Liebherr ( Kirchdorf , Germany ) KP3120 ) or 0°C ( Johnson Controls , Milwaukee , WI ) . All temperatures were recorded as ± ≤1 . 5°C , 70% ± ≤10% RH . An 8hr photoperiod was provided by fluorescent tubes for temperatures ≥8°C and LEDs for temperatures ≤2°C . Plants were transferred to random locations in a controlled environment room ( 16 hr light , 22°C ± 2°C ) and flowering time was scored as the number of days of growth until floral buds became visible . Total RNA was extracted as described previously ( Box et al . , 2011 ) . cDNA was synthesized using Precision nano-script reverse transcription ( Primerdesign ) with oligo d ( T ) and analysed by qPCR on a LightCycler 480 II intrument ( Roche , Basel , Switzerland ) , using LightCycler 480 Probes Master mix ( Roche ) . FLC mRNA was assayed using Roche Universal Probe Library ( UPL ) #65 ( 5′-ctggagga-3′ ) with primers sFLC_UPL_F ( 5′-gtgggatcaaatgtcaaaaatg-3′ ) and sFLC_UPL_R ( 5′-ggagagggcagtctcaaggt-3′ ) . VIN3 mRNA was assayed using UPL#67 ( 5′-tggtggat-3′ ) with primers VIN3_UPL_F ( 5′-cgcgtattgcggtaaagataa-3′ ) and VIN3_UPL_R ( 5′-tctctttcgccaccttcact-3′ ) . FT mRNA was assayed using UPL#138 ( 5′-tggtggat-3′ ) with primers FT_UPL_#138_F ( 5′-ggtggagaagacctcaggaa-3′ ) and FT_UPL_#138_R ( 5′-ggttgctaggacttggaacatc-3′ ) . Expression of each gene was normalized to UBC ( At5g25760 ) with primers UBC_UPL_F ( 5′-tcctcttaactgcgactcagg-3 ) , UBC_UPL_R ( 5′-gcgaggcgtgtatacatttg-3 ) and UPL#9 ( 5′-tggtgatg-3′ ) . Statistical analyses of logged expression data were performed using GraphPad Prism version 6 software ( La Jolla , CA ) . ChIP assays were performed as previously described ( Sun et al . , 2013 ) using H3K27me3 and H3 antibodies cited by Angel et al . ( 2011 ) . Primers used in this analysis are shown in Figure 2—source data 1 . SHOOT MERISTEMLESS ( STM ) was used as the internal control and data are represented as the ratio of ( H3K27me3FLC/H3 FLC ) to ( H3K27me3 STM/H3 STM ) . Statistical analysis of ChIP data was performed using GraphPad Prism version 5 software for Mac . Hourly temperatures were recorded using Tinytag data-loggers ( Chichester , UK ) . Historical climate data were obtained from Swedish Meteorological and Hydrological Institute . Three temperature and snow-depth readings taken at 0600 hr , 1200 hr and 1800 hr were used to calculate daily means . Boxplots graphs were created using QI Macros add-ins for Excel ( Denver , CO ) . Statistical analyses of climate data were performed using GraphPad Prism version 6 software . Seeds were stratified for 4 days at 5°C , sown into trays using a randomized block design and placed outside ( 62° 23 . 463´N , 17° 18 . 272´E ) . Seedlings were thinned to one plant per cell after 7 days and then transferred to Ramsta ( 62° 50 . 988´N , 18° 11 . 570´E ) 1 week later . At each transfer date , plants were returned to a greenhouse in Mid-Sweden University , Sundsvall ( 16 hr light , 22°C ± 2°C ) where flowering time was determined as the number of days growth until floral buds became visible . | Plants are not able to move around and so they need to be able to adapt their growth and development to seasonal changes in their environment . For example , prolonged exposure to cold temperatures during winter can prime some plants to flower when temperatures increase in the spring—a process called vernalization . In these plants , extended periods of cold temperatures lead to lower activity of a gene called FLC , which normally inhibits flowering . In the plant Arabidopsis thaliana , vernalization requires several months of exposure to temperatures between 0–6°C . Recently , A . thaliana plants from southern Europe were found to vary in the temperature requirements for vernalization , responding to temperatures higher than 6°C . This suggested that plants from northern Europe might vernalize preferentially at lower temperatures . Here , Duncan et al . compared vernalization in a collection of A . thaliana plants ( or ‘accessions’ ) sampled from different regions of Sweden and the UK . The experiments show that all the accessions needed temperatures above 0°C to vernalize and that vernalization still worked relatively well at temperatures as high as 14°C . The optimal temperature range for vernalization differed between the accessions , but plants from more northern areas did not necessarily vernalize at lower temperatures . For example , for one particular accession from northern Sweden , the temperature that is optimum for vernalization was 8°C , a notably higher temperature than expected . Historical local climate records suggested that this accession would vernalize before the first snowfall of the winter in North Sweden . Duncan et al . confirmed this proposal with field experiments . Plants were grown in natural field sites in September and then moved into a greenhouse . The experiments show that the plants complete vernalization by November , which strongly suggests that FLC is silenced during autumn rather than during winter , as previously thought . This changed temperature response is due , in part , to a small number of tiny genetic differences in regions of the FLC gene that do not code for protein . These findings have important implications for future studies of vernalization and flowering time , and for understanding how plants will adapt to on going and future climate change . The next step is to understand what causes these changed temperature responses at a molecular level , which should enable selective breeding for flowering and harvest date in a range of crops . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"short",
"report",
"plant",
"biology"
] | 2015 | Seasonal shift in timing of vernalization as an adaptation to extreme winter |
Pathways modulating glucose homeostasis independently of insulin would open new avenues to combat insulin resistance and diabetes . Here , we report the establishment , characterization , and use of a vertebrate ‘insulin-free’ model to identify insulin-independent modulators of glucose metabolism . insulin knockout zebrafish recapitulate core characteristics of diabetes and survive only up to larval stages . Utilizing a highly efficient endoderm transplant technique , we generated viable chimeric adults that provide the large numbers of insulin mutant larvae required for our screening platform . Using glucose as a disease-relevant readout , we screened 2233 molecules and identified three that consistently reduced glucose levels in insulin mutants . Most significantly , we uncovered an insulin-independent beneficial role for androgen receptor antagonism in hyperglycemia , mostly by reducing fasting glucose levels . Our study proposes therapeutic roles for androgen signaling in diabetes and , more broadly , offers a novel in vivo model for rapid screening and decoupling of insulin-dependent and -independent mechanisms .
Characterized by the inability to control blood glucose levels , diabetes is a metabolic disease of major socio-economic concern . Blood glucose levels are regulated by multiple tissues including the pancreas , muscle , liver , adipocytes , gut and kidney ( Defronzo , 2009 ) . Signals from endocrine hormones are integrated by each tissue to effectively maintain glucose homeostasis , and aberrations in this interplay underlie the pathogenesis of diabetes . Currently , seven classes of antidiabetic drugs exist , of which only three function without increasing circulating insulin levels and only one that definitively functions independently of insulin ( Chaudhury et al . , 2017 ) . Restoring normoglycemia independently of insulin secretion or action could delay disease progression as an improved glycemic status can restore β-cell mass and function ( Wang et al . , 2014 ) . Lower dependence on insulin-stimulating therapies can also prevent hyperinsulinemia-driven insulin resistance ( Shanik et al . , 2008 ) and obesity ( Mehran et al . , 2012 ) . In contrast to insulin stimulators , Biguanides ( e . g . , Metformin ) and Thiazolidinediones ( e . g . , Pioglitazone ) are effective antidiabetic agents that primarily sensitize tissues to insulin ( reviewed by ( Soccio et al . , 2014; Rena et al . , 2017 ) ) . Likewise , sodium-glucose transporter two inhibitors ( e . g . , Dapagliflozin ) have a complementary mechanism of reducing glucose reabsorption in the kidney ( Bailey et al . , 2013 ) . Increasing evidence points to additional molecular pathways that can improve metabolic homeostasis independently of insulin , for instance , using leptin therapy ( Neumann et al . , 2016 ) or exercise ( Stanford and Goodyear , 2014 ) . Interestingly , currently prescribed drugs were discovered from their historical use in herbal medicine ( Ehrenkranz et al . , 2005; Bailey , 2017 ) or from screens directed against hyperlipidemia ( Fujita et al . , 1983 ) . However , so far , an unbiased search for insulin-independent pathways controlling glucose metabolism has remained elusive , primarily due to the lack of a disease-relevant animal model for rapid screening . Due to its high fecundity and amenability to chemical screening , the zebrafish serves as an excellent platform to study diabetes , and it has been successfully used to study β-cell mass and activity , as well as glucose metabolism ( Andersson et al . , 2012; Gut et al . , 2013; Tsuji et al . , 2014; Nath et al . , 2015; Li et al . , 2016; White et al . , 2016; Gut et al . , 2017; Matsuda et al . , 2018 ) . Here , using the zebrafish model , we generated an innovative drug discovery strategy , screened chemical libraries and specifically identified insulin-independent effects of androgen signaling on glucose homeostasis .
Insulin plays a central role in glucose homeostasis by increasing glucose uptake in peripheral tissues , promoting glycogenesis in the liver and decreasing glucose production by inhibiting glucagon secretion ( Aronoff et al . , 2004 ) . We generated zebrafish devoid of insulin signaling and determined the degree to which these mutants recapitulate core features of diabetic metabolism observed in mammals . The zebrafish genome contains two insulin genes – insulin ( ins ) and insulinb ( insb ) . Using CRISPR/Cas9 mutagenesis , we generated a 16 bp deletion allele of ins ( Figure 1A ) and a 10 bp insertion allele of insb . Although ins and insb mutant embryos appear morphologically unaffected ( Figure 1—figure supplement 1A ) , Insulin was entirely absent in pancreatic islets of ins mutants ( Figure 1B ) , whereas there was no observable change in insb mutant islets ( Figure 1—figure supplement 1B and C ) . ins mutants exhibit a drastic increase in total glucose levels ( up to 10-fold ) , measured from 1 to 6 days post fertilization ( dpf ) ( Figure 1C ) . Additionally , staining for lipid content using Nile Red revealed large unused yolk reserves ( Figure 1D ) , suggesting defects in lipid absorption and processing . Due to a combination of these metabolic defects , ins mutants do not survive beyond 12 dpf ( Figure 1E ) . Moreover , although 3 month old ( adult ) ins +/- animals are normoglycemic ( Figure 1—figure supplement 1D ) , 50 dpf ( juvenile ) ins +/- animals are noticeably smaller ( Figure 1—figure supplement 1E ) , consistent with a role for Insulin in growth control ( Nakae et al . , 2001 ) . insb mutants , on the other hand , are viable and fertile . During WT development , insb expression is minimal beyond 48 hpf ( Papasani et al . , 2006; White et al . , 2017 ) ( Figure 1—figure supplement 1F ) , and it is undetectable in adult β-cells ( Tarifeño-Saldivia et al . , 2017 ) . To assess whether it is capable of function , we overexpressed insb under the ins promoter . Under the hyperglycemic conditions resulting from morpholino ( MO ) -mediated ins knockdown , insb overexpression successfully lowered glucose levels , thus indicating that insb is functional ( Figure 1—figure supplement 1G ) . However , due to the post-embryonic expression of ins , survival and metabolic homeostasis in zebrafish depends primarily on ins . This predominant role of ins distinguishes the zebrafish insulins from the redundant metabolic roles of mouse Ins1 and Ins2 ( Duvillié et al . , 1997 ) . To further explore the nature of the metabolic defects in zebrafish ins mutants , we probed the proteome of 108 hpf ins mutants compared to their WT siblings and assessed this dataset relative to seven other proteomes from diabetic tissues of rodent or human origin ( Hwang et al . , 2010; Mullen and Ohlendieck , 2010; Giebelstein et al . , 2012; Valle et al . , 2012 ) ; S . -J . Kim et al . , 2014a; Capuani et al . , 2015; Braga et al . , 2016; Zabielski et al . , 2016 ) . Strikingly , pathways like gluconeogenesis , mitochondrial dysfunction , sirtuin signaling , and oxidative phosphorylation , which were affected in diabetic conditions across these studies , were similarly disrupted in zebrafish ins mutants ( Figure 1F and supplementary file 1 ) . Together , these findings indicate that zebrafish ins is crucial for metabolic homeostasis and survival , and that its absence causes core features of diabetic metabolism already at larval stages . Screening of small molecules in ins mutants requires large numbers of mutant animals . However , the early lethality of ins mutants did not allow the generation of adult animals that can be incrossed . To overcome this obstacle , we used an efficient endoderm induction ( Kikuchi et al . , 2001 ) and transplantation technique ( Stafford et al . , 2006 ) ( Figure 2A ) to selectively modify endodermal tissues without altering the germline . Tg ( ins:DsRed ) ; ins +/+ embryos were injected with sox32 mRNA at the one-cell stage , conferring an endodermal fate on all cells . Between 3 to 4 hpf , cells were transplanted from these embryos to the mesendoderm of similarly staged embryos from Tg ( ins:RasGFP ) ; ins +/- incrosses . This transplantation procedure was remarkably efficient at contributing to host endoderm ( Figure 2—figure supplement 1A–A’’ ) , and the pancreatic islet of nearly every host embryo contained both donor derived ( i . e . , ins +/+ ) as well as host β-cells ( Figure 2B–B’’ ) . These chimeric animals were raised to adulthood and genotyping ( Figure 2—figure supplement 1B–C ) revealed a near Mendelian ratio of mutant animals ( Figure 2C ) . In summary , these chimeric animals contain ins +/+ endodermal tissues but retain an ins -/- germline , thereby allowing an all-mutant progeny to be obtained by incrossing . With the ability to obtain large numbers of ins mutant embryos , we next aimed to analyze the effect of known glucose homeostasis modulators and also to screen for novel ones . We tested the effects of molecules that have been proposed to help normalize glucose levels in an insulin-independent manner as well as others that do so in an insulin-dependent manner . Anti-diabetics such as metformin , pioglitazone and dapagliflozin , as well as the Lyn kinase activator MLR1023 ( Saporito et al . , 2012 ) , were tested . We also tested fraxidin , identified in a screen for molecules that increase glucose uptake in zebrafish ( Lee et al . , 2013 ) . Surprisingly , metformin and MLR1023 exhibited no glucose-lowering effect in ins mutants , suggesting that they act more as sensitizers of insulin signaling rather than independently of insulin ( Figure 3A ) . In ins mutants , Pck1 levels are higher compared to wild types ( Supplementary file 1 ) , suggesting increased gluconeogenesis in the absence of the inhibitory action of insulin . This observation is also supported by the drastic reduction of glucose levels in ins mutants treated with the Pck1 inhibitor , 3 MPA ( Figure 3A ) . Metformin has well-known abilities to reduce hepatic gluconeogenesis , and it also reduces glucose levels in isoprenaline-treated wild-type zebrafish ( Gut et al . , 2013 ) . However , metformin’s inability to reduce glucose levels in zebrafish ins mutants reveals metformin’s dependence on insulin signaling for its action . MLR1023’s glucose level lowering effects have been proposed to be insulin-dependent ( Ochman et al . , 2012 ) , and thus , its inability to lower glucose levels in zebrafish ins mutants further suggests the lack of any insulin signaling in these animals . On the other hand , fraxidin , dapagliflozin , and pioglitazone reduced glucose levels by 5 , 12 , and 11% respectively ( Figure 3A ) , thus attributing part of their glucose lowering effect to an insulin-independent mechanism . The lack of adipose tissue ( Minchin and Rawls , 2017 ) and the primitive nature of kidney ( pronephros ) function at these developmental stages ( Elmonem et al . , 2018 ) may result in an incomplete recapitulation of adipose signaling and renal function on glucose homeostasis ( Defronzo , 2009 ) . This limitation could also explain the small magnitude of glucose level reduction observed with the PPARγ agonist or SGLT2 inhibitor treatments in zebrafish ins mutants . Based on these data with known glucose level lowering drugs , we decided to screen chemical libraries to identify molecules that could reduce glucose levels by more than 10% in zebrafish ins mutants . To rapidly measure free glucose levels in a 96-well plate format , we adapted a kit-based protocol that is sensitive to endogenous changes in larval glucose levels ( Figure 3—figure supplement 1A–C ) , and established a screening pipeline ( Figure 3B ) . We screened 2233 molecules in 2 replicates at 10 μM concentration and found three hits ( Figure 3C ) that reproducibly reduced glucose levels upon retesting with independent chemical stocks and the unmodified standard glucose measurement kit . These three hits - flutamide ( androgen receptor antagonist ) , ODQ ( soluble guanylyl cyclase ( sGC ) inhibitor ( Boulton et al . , 1995 ) ) and vorinostat ( broad HDAC inhibitor ( Finnin et al . , 1999 ) ) were found , upon retesting multiple times , to reduce glucose levels by 40 , 22% and 19% , respectively ( Figure 3D , Figure 3—figure supplement 1D ) . sGC inhibition by ODQ has been previously reported to increase net hepatic glucose uptake and shift the balance towards glycogenesis ( An et al . , 2010 ) . Contrary to our findings , clinical use of vorinostat has been associated with hyperglycemia as a side effect ( Mann et al . , 2007 ) . This difference could be due to the broad nature of Vorinostat’s HDAC inhibition properties including anti-proliferative effects ( Richon , 2006 ) , which are likely to affect developmental processes in zebrafish ins mutants . These three drugs did not prolong the survival of zebrafish ins mutants , likely because lowering glucose levels alone was not sufficient to normalize all the metabolic , growth , and differentiation processes ( Taniguchi et al . , 2006 ) dysregulated in these animals . Given the strong reduction in glucose levels observed after flutamide treatments , we further tested the hypothesis that glucose levels in ins mutants were being reduced through androgen receptor antagonism . First , flutamide caused a dose-dependent decrease in glucose levels in ins mutants ( Figure 4—figure supplement 1A ) . Second , we treated ins mutants with AR antagonists of two types: ( i ) steroidal ( Cyproterone ) and ( ii ) non-steroidal ( nilutamide , hydroxyflutamide , bicalutamide , enzalutamide ) , and observed a consistent decrease in glucose levels across all treatments , albeit at varying efficiency ( Figure 4A ) , possibly reflecting the different efficacy of these antagonists towards zebrafish AR ( Raynaud et al . , 1979; Teutsch et al . , 1994; Tran et al . , 2009 ) . Finally , to modulate AR protein levels , we injected 1 ng of control or ar MO into one-cell stage embryos and observed a reduction of glucose levels in ar MO injected ins mutants ( Figure 4B ) but not in ar MO injected wild-type animals ( Figure 4C ) . These data support a role , and a benefit , for antagonizing AR specifically in hyperglycemic conditions . A number of mechanisms have been proposed to explain the predisposition of women with androgen excess to diabetes , including insulin resistance , visceral adiposity , and β-cell dysfunction ( Navarro et al . , 2015 ) . Under high fat diet , a combination of neuronal and pancreatic β-cell specific roles for AR have been proposed to predispose female mice with androgen excess to diabetes ( Navarro et al . , 2018 ) . Supporting this role , ar gene expression was observed in the zebrafish central nervous system and , additionally , in the liver ( Gorelick et al . , 2008 ) ( Figure 4—figure supplement 1B ) . To investigate how AR antagonism mediates glucose level reduction in zebrafish ins mutants , we used a transcriptomic approach . RNA-Seq analyses of 120 hpf ins mutants treated with flutamide or cyproterone revealed 504 and 476 differentially expressed genes ( DEGs ) compared to vehicle treated mutants ( Figure 4D ) , respectively . Of these DEGs , 40 were regulated in parallel ( both up or both down ) for both AR antagonists tested , likely highlighting the common AR-specific effects . Cross-referencing these 40 genes with a transcriptomic comparison of ins mutants to phenotypically wild-type siblings , led to 12 genes ( Figure 4E ) that were differentially expressed upon loss of ins , and were partially or fully restored to wild-type levels upon treatment with AR antagonists ( Figure 4—figure supplement 1C ) . Amongst these 12 genes , btg2 and insig1 have been reported to play crucial roles in controlling liver gluconeogenesis ( Carobbio et al . , 2013; Kim et al . , 2014b ) , and they also contain two androgen response elements ( AREs ) close to their transcription start site ( Figure 4—figure supplement 1D ) . Additionally , upon intraperitoneal injections of flutamide in hyperglycemic adult animals ( Figure 4—figure supplement 1E ) , we observed 19% lower fasting plasma glucose levels ( Figure 4F ) , likely due to reduced hepatic glucose production . Our findings corroborate the observations of better anthropometric indices previously observed with flutamide ( Sahin et al . , 2004 ) or metformin +flutamide combination therapies ( Gambineri et al . , 2004; Amiri et al . , 2014 ) , and attribute a part of this beneficial effect to flutamide’s insulin-independent action through AR antagonism . In conclusion , to the best of our knowledge , ours is the first study to report the generation and use of a rapid screening strategy to identify insulin-independent pathways modulating metabolism in vertebrates . Given the recent success of SGLT2 inhibitors as combination therapy in diabetes ( Bailey et al . , 2013 ) , our study is an important step towards identifying more insulin-independent mechanisms governing glucose homeostasis . One of the limitations of our screen is the relatively low size of the chemical library screened . However , as the endoderm transplant technique reported here can be combined with several genetic or metabolic readouts , future studies with larger chemical libraries should unveil mechanisms governing other disease-relevant phenomena as well . Such comprehensive insight into insulin-independent mechanisms and their interactions with insulin signaling in homeostasis and disease will open new avenues for designing therapies to treat metabolic disorders .
Zebrafish husbandry was performed under standard conditions in accordance with institutional ( MPG ) and national ethical and animal welfare guidelines . The transgenic and mutant lines used in this study are Tg ( ins:DsRed ) m1018 ( Anderson et al . , 2009 ) , Tg ( ‐4 . 0ins:GFP ) zf5 ( Huang et al . , 2001 ) , Tg ( sox17:GFP ) s870 ( Sakaguchi et al . , 2006 ) , Tg ( ins:Flag-NTR , cryaa:mCherry ) s950 ( Andersson et al . , 2012 ) , Tg ( ins:EGFP-HRas , cryaa:mCherry ) bns294 , Tg ( ins:TagRFPt-P2A-insB ) bns285 , insbns102 ( ins mutants ) , and insbbns295 ( insb mutants ) . 1 , 2 , 3 , 4 , 5 , and 6 days post fertilization ( dpf ) correspond to 24 , 48 , 72 , 96 , 120 , and 144 hr post fertilization ( hpf ) , and 3 months post fertilization ( mpf ) corresponds to 90 dpf . CRISPR design platform ( http://crispr . mit . edu ) was used to design sgRNAs against ins ( targeting sequence: TCCAGTGTAAGCACTAACCCAGG ) and insb ( targeting sequence: GGATCGCAGTCTTCTCC ) genes and constructs were assembled as described previously ( Jao et al . , 2013; Varshney et al . , 2015 ) . Briefly , a mixture of 25 pg gRNA with 300 pg Cas9 mRNA was injected into one cell stage wild-type embryos . High-resolution melt analysis ( HRMA ) ( Eco-Illumina ) was used to determine efficiency of sgRNAs and genotype animals with ins primers 5’-GTGCTCTGTTGGTCCTGTTGG-3’ and 5’-CATCGACCAGATGAGATCCACAC-3’ , and insb primers: 5’-AGTATTAATCCTGCTGCTGGCG-3’and 5’-GTGTAGAAGAAACCTCTAGGC-3’ . Immunostaining and imaging was performed as described previously ( Yang et al . , 2018 ) . Briefly , zebrafish larvae were euthanized and fixed overnight at 4°C with 4% paraformaldehyde ( dissolved in buffer with composition: 22 . 6 mM NaH2PO4 , 77 mM Na2HPO4 , 120 μM CaCl2 , 117 mM sucrose , pH 7 . 35 ) . After two PBS washes , the larvae were deskinned , and permeabilized using PBS containing 0 . 5% TritonX-100% and 1% DMSO for 1 hr . Larvae were then incubated in blocking buffer ( Dako ) containing 5% goat serum for 2 hr , and incubated with primary antibody overnight at 4°C . Next , samples were washed 3 × 10 min with PBS containing 0 . 1% TritonX-100 , incubated overnight at 4°C with secondary antibody and DAPI ( 10 µg/ml ) , washed 3 × 10 min and mounted in agarose . Antibody dilutions used are as follows: guinea pig anti-Insulin polyclonal ( 1:100 , Thermo ) , mouse anti-Glucagon ( 1:300 , Sigma ) , chicken anti-GFP ( 1:300 , Aves ) , goat anti-guinea pig AlexaFluor568 ( 1:300 , Thermo ) , goat anti-mouse AlexaFluor647 ( 1:300 , Thermo ) , goat anti-chicken AlexaFluor488 ( 1:500 , Thermo ) . Zeiss LSM700 ( 10X ) and LSM800 ( 25X ) were used to acquire data , and Imaris ( Bitplane ) was used to analyze data and to create maximum intensity projection images . Neutral lipid staining using Nile Red dye was performed at a working concentration of 0 . 5 μg/mL for 30 min in the dark , followed by acquisition of fluorescent images using an LP490 filter on a Nikon SMZ25 stereomicroscope . For knockdown of gene expression , the following splice-blocking antisense morpholinos ( Gene Tools , LLC ) were injected into one-cell embryos at the indicated dosage per embryo: insa MO ( 4 ng , 5′-CCTCTACTTGACTTTCTTACCCAGA-3’ ) ( Ye et al . , 2016 ) ar MO ( 1 ng , 5'-AGCAGAGCCGCCTCTTACCTGCCAT-3' ) ( Peal et al . , 2011 ) standard control MO ( 4 or 1 ng , 5'-CCTCTTACCTCAGTTACAATTTATA-3' ) . Intraperitoneal injections and glucose level measurement in 6-month-old adult zebrafish was performed as described previously ( Curado et al . , 2008; Moss et al . , 2009; Eames et al . , 2010 ) . Briefly , ablation of β-cells in Tg ( ins:Flag-NTR , cryaa:mCherry ) s950 ( Andersson et al . , 2012 ) animals was performed by injecting 0 . 25 gm MTZ/kg body weight twice – on day 0 and day 1 – injecting twice improved the consistency of ablation . Flutamide ( 10 mg/kg ) or vehicle ( DMSO ) was injected on days 2 , 3 and 4 . For injections , animals were anaesthetized using 0 . 02% Tricaine . On day 4 , animals were euthanized and blood glucose was measured using a FreeStyle Freedom Lite Glucose Meter ( Abbott ) . Free glucose level measurements were performed as described previously ( Jurczyk et al . , 2011 ) , with minor modifications . After desired treatment conditions , pools of 10 animals were collected in 1 . 5 mL Eppendorf tubes and frozen at −80°C after complete removal of water . For analysis , pools of wild-type embryos were resuspended in PBS . Samples were homogenized using a tissue homogenizer ( Bullet Blender Gold , Next Advance ) . A Glucose Assay Kit ( CBA086 , Merck ) was used for glucose detection . Different volumes were used for resuspension and glucose detection between wild types and ins mutants: wild-type samples were resuspended in a volume corresponding to 8 μl/animal and 8 μl was used for the glucose detection reaction . ins mutant embryos , due to their much higher glucose content , were resuspended in a volume corresponding to 16 μl/animal and only 2 μl was used for glucose detection . For the endoderm transplant experiment , sox32 mRNA was transcribed using an Sp6 mMessage mMachine kit ( Ambion ) . Using a micro-injector , 100 pg of sox32 mRNA was injected into Tg ( ins:DsRed ) embryos , which served as donors . Embryos from an ins +/- incross served as hosts . Between the 1 k-cell and sphere stages ( 3–4 hpf ) , 15–20 cells from donor embryos were transplanted to host embryos , targeting the host mesendoderm at the margin of the blastoderm . Larvae were collected at 120 hpf and fixed with 4% paraformaldehyde in PBS overnight at 4°C . In situ hybridization was performed as described previously ( Thisse and Thisse , 2008 ) . ar digoxigenin-labelled anti-sense probe was synthesized using T7 polymerase ( Roche ) and DIG RNA labelling kit ( Roche ) . The probe template was amplified using the following primers: ar ISH-forward 5′‐TGGAGTTTTTCCTTCCTCCA-3’ and ar ISH-reverse 5’- TAATACGACTCACTATAGGGTCATTTGTGGAACAGGATT- 3’ , obtaining a 1100 bp probe as described previously ( Gorelick et al . , 2008 ) . Embryos were imaged on a Nikon SMZ25 stereomicroscope . Wild-type and mutant larvae were processed in the same tube and genotyped after the images were taken . 3-mercaptopicolinic acid ( 3 MPA ) treatment was performed at 1 . 5 mM concentration , metformin treatment at 250 μM concentration , and enzalutamide and bicalutamide treatments at 20 μM concentration . All other drug treatments were performed at 10 μM . For plate based screening , three 84 hpf ins mutant larvae were placed in each well of a 96-well plate in 200 μl of egg water buffered with 10 mM HEPES . All drug treatments were performed at 10 μM with 1% DMSO , unless otherwise stated . Drug treatment was performed from 84 to 120 hpf , after which each well was visually analyzed to assess toxicity . Subsequently , 100 μl of egg water was removed and 25 μl of 5X cell culture lysis buffer ( Promega ) was added . The plate was left shaking for 1 min at 750 rpm , and after gentle shaking at 150 rpm for 30 min , another round of vigorous shaking was performed for 1 min at 750 rpm . 8 μl from each well was used for the glucose detection reaction in a new 96-well plate using the Glucose Assay Kit ( CBA086 , Merck ) . Drug libraries used in this screen are: For RNA-seq analysis , total RNA was isolated from 120 hpf zebrafish using the RNA Clean and Concentrator kit ( Zymo Research ) , and samples were treated with DNase ( RNase-free DNase Set , Promega ) to avoid contamination by genomic DNA . Integrities of the isolated RNA and library preparation were verified with LabChip Gx Touch 24 ( Perkin Elmer ) . 3 µg of total RNA was used as input for Truseq Stranded mRNA Library preparation following manufacturer’s ‘low sample’ protocol ( Illumina ) . Sequencing was performed on NextSeq500 instrument ( Illumina ) using v2 chemistry , resulting in a minimum of 23M reads per library with 1 × 75 bp single end setup . The resulting raw reads were assessed for quality , adapter content and duplication rates with FastQC ( Andrews , 2010 ) . Trimmomatic ( version 0 . 33 ) was employed to trim reads after a quality drop below a mean of Q20 in a window of 5 nucleotides ( Bolger et al . , 2014 ) . Only reads between 30 and 150 nucleotides were cleared for further analyses . Trimmed and filtered reads were aligned with the Ensembl Zebrafish genome version DanRer10 ( GRCz10 . 90 ) , using STAR 2 . 4 . 0a with the parameter ‘--outFilterMismatchNoverLmax 0 . 1’ to increase the maximum ratio of mismatches to mapped length to 10% ( Dobin et al . , 2013 ) . The number of reads aligning to genes was counted with featureCounts 1 . 4 . 5-p1 tool from the Subread package ( Liao et al . , 2014 ) . Only the reads that mapped , at least partially , to within exons were admitted and aggregated for each gene . Reads that overlapped multiple genes or aligned to multiple regions were excluded . Differentially expressed genes were identified using DESeq2 version 1 . 62 ( Love et al . , 2014 ) . Maximum Benjamini-Hochberg corrected p-value of 0 . 05 , along with a minimum combined mean of 5 reads , were set as inclusion criteria . The Ensembl annotation was enriched with UniProt data ( release 06 . 06 . 2014 ) based on Ensembl gene identifiers ( ‘Activities at the Universal Protein Resource ( UniProt ) , " 2014 ) . RNA-seq data have been deposited in the ArrayExpress database at EMBL-EBI ( www . ebi . ac . uk/arrayexpress ) under accession number E-MTAB-7283 . From this dataset , normalized read counts for ins and insb expression are: ConditioninsinsbDMSO1831Flutamide1762Cyproterone1732 Average normalized counts for ins and insb from transcriptomic data from zebrafish adult β-cells ( Tarifeño-Saldivia et al . , 2017 ) are 4324351 and 0 respectively . For microarray expression profiling , RNA was isolated from pooled 108 hpf zebrafish larvae using the RNA Clean and Concentrator kit ( Zymo Research ) combined with DNase digestion ( RNase-free DNase Set , Promega ) . 10 animals were used for each pooled sample . Sample quality was tested using a Bioanalyzer and microarray analysis was performed by Oak Labs ( Germany ) . Microarray data have been deposited in the ArrayExpress database at EMBL-EBI ( www . ebi . ac . uk/arrayexpress ) under accession number E-MTAB-7282 . For each of the three biological replicates within a genotype , protein was extracted from pools of 600 larvae at 5 dpf using 4% SDS in 0 . 1 M Tris/HCl , pH 7 . 6 and a tissue disrupting sterile pestle ( Axygen ) for lysis . After heating to 70°C at 800 rpm for 10 min and DNA shearing by sonication , cell debris was removed by centrifugation at 14 . 000 x g for 10 min and retaining the supernatant . Using a DC protein assay ( BioRad ) , 7 mg of solubilized proteins per sample were acetone precipitated at −20°C overnight , followed by centrifugation at 14 . 000 x g for 10 min and washing the pellet using 90% acetone . After evaporation of residual acetone , samples were dissolved in urea buffer ( 6 M urea , 2 M thiourea , 10 mM HEPES , pH 8 . 0 ) , followed by enzymatic peptidolysis as described ( Graumann et al . , 2008; Kim et al . , 2018 ) with the following modifications: 10 mM dithiothreitol , 55 mM iodoacetamide and 100:1 protein to enzyme ratio of the proteolytic enzymes were used . Subsequent sample processing and data analyses were performed as described previously ( Sokol et al . , 2018 ) . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD012027 . Canonical pathway analysis was performed using Ingenuity Pathway Analysis ( IPA ) ( Qiagen ) . Differentially expressed proteins from our study ( log2FC ± 1 . 5 ) and from previously published datasets were subjected to a Comparison analysis in IPA . P-value maximum cut-off was set at 0 . 05 and the processes are listed according to those affected across most studies . | Diabetes is a disease that affects the ability of the body to control the level of sugar in the blood . Individuals with diabetes are unable to make a hormone called insulin – which normally stimulates certain cells to absorb sugar from the blood – or their cells are less able to respond to this hormone . Most treatments for diabetes involve replacing the lost insulin or boosting the hormone’s activity in the body . However , these treatments can also cause individuals to gain weight or become more resistant to insulin , making it harder to control blood sugar levels . In addition to insulin , several other factors regulate the levels of sugar in the blood and some of them may operate independently of insulin . However , little is known about such factors because it is impractical to carry out large-scale screens to identify drugs that target them in humans or mice , which are often used as experimental models for human biology . To overcome this challenge , Mullapudi et al . turned to another animal known as the zebrafish and generated mutant fish that lack insulin . The mutant zebrafish had similar problems with regulating sugar levels as those observed in humans and mice with diabetes . This observation suggests that insulin is just as important in zebrafish as it is in humans and other mammals . The mutant zebrafish did not survive into adulthood , and so Mullapudi et al . transplanted healthy tissue into the zebrafish to allow them to produce enough insulin to survive . These adult zebrafish produced many offspring that still carried the insulin mutation . Mullapudi et al . used these mutant offspring to screen over 2 , 000 drugs for their ability to decrease blood sugar levels in the absence of insulin . The screen identified three promising candidate drugs , including a molecule that interferes with a receptor for a signal known as androgen . These findings will help researchers investigate new ways to treat diabetes . In the future , the screening approach developed by Mullapudi et al . could be adapted to search for new drugs to treat other human metabolic conditions . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"short",
"report"
] | 2018 | Screening for insulin-independent pathways that modulate glucose homeostasis identifies androgen receptor antagonists |
In neural circuits , individual neurons often make projections onto multiple postsynaptic partners . Here , we investigate molecular mechanisms by which these divergent connections are generated , using dyadic synapses in C . elegans as a model . We report that C . elegans nrx-1/neurexin directs divergent connectivity through differential actions at synapses with partnering neurons and muscles . We show that cholinergic outputs onto neurons are , unexpectedly , located at previously undefined spine-like protrusions from GABAergic dendrites . Both these spine-like features and cholinergic receptor clustering are strikingly disrupted in the absence of nrx-1 . Excitatory transmission onto GABAergic neurons , but not neuromuscular transmission , is also disrupted . Our data indicate that NRX-1 located at presynaptic sites specifically directs postsynaptic development in GABAergic neurons . Our findings provide evidence that individual neurons can direct differential patterns of connectivity with their post-synaptic partners through partner-specific utilization of synaptic organizers , offering a novel view into molecular control of divergent connectivity .
Neurons are typically wired into discrete circuits through stereotyped patterns of synaptic connections geared to perform specific functions . Individual neurons within circuits may receive convergent synaptic inputs from multiple classes of presynaptic partnering neurons , and likewise , make divergent synaptic outputs onto distinct postsynaptic targets . We have gained an understanding of some of the core mechanisms that sculpt convergent connectivity through studies of developmental processes such as activity-dependent synapse elimination ( Brown et al . , 1976; Campbell and Shatz , 1992; Okawa et al . , 2014a; Sanes and Lichtman , 1999; Shatz and Kirkwood , 1984; Walsh and Lichtman , 2003 ) . In contrast , the molecular processes controlling the establishment of divergent synaptic connections ( between a single presynaptic partner and multiple postsynaptic target cells ) are not clearly defined ( Okawa et al . , 2014b ) . Neural circuit models often represent divergent connections as a means for enabling the same signal from an individual presynaptic neuron to reach many different postsynaptic target cells . However , the strength of connections with postsynaptic partners can vary widely , strongly suggesting that presynaptic neurons have the capacity to establish and regulate connections with each postsynaptic target independently . While molecular guidance cues directing axon outgrowth have been well-studied , an understanding of the molecular mechanisms responsible for directing target-specific connectivity has remained elusive . A primary mechanism for establishing nascent synapses is through the actions of synaptic adhesion molecules , also known as synaptic organizers ( de Wit and Ghosh , 2016; Missler et al . , 2012 ) . These organizers are often anchored to the pre- and post-synaptic membranes ( e . g . neurexins , neuroligins , leucine-rich repeat transmembrane proteins/LRRTMs ) and promote synapse formation through trans-synaptic adhesion and signaling . The importance of these processes in establishing proper neural circuit connectivity is highlighted by the links between mutations in genes encoding these synaptic adhesion/organizing molecules and neuropsychiatric and neurodevelopmental disorders , such as autism spectrum disorder and schizophrenia ( Kim et al . , 2008; Reichelt et al . , 2012; Rujescu et al . , 2009 ) . Intriguingly , synaptic organizers are capable of acting in a cell-specific manner to promote synapse formation ( Chen et al . , 2017; Siddiqui et al . , 2013; Zhang et al . , 2015 ) . Thus , an exciting possibility is that individual neurons could encode connections with alternate synaptic partners through differential deployment of synaptic organizers . We have investigated this possibility in the motor circuit of the nematode Caenorhabditis elegans where individual excitatory cholinergic motor neurons form synapses with both body wall muscles and GABAergic motor neurons . Through a screen for genes that govern the formation of these divergent synaptic connections , we demonstrate that the synaptic organizer nrx-1/neurexin directs the outgrowth of previously uncharacterized dendritic spine-like structures and the formation of synaptic connections with GABAergic neurons , but is not required for synaptic connectivity with muscles . Conversely , genes previously shown to be required for cholinergic connectivity with muscles ( Francis et al . , 2005; Gally et al . , 2004; Pinan-Lucarré et al . , 2014 ) are not required for the formation of synapses onto GABAergic neurons . Our findings demonstrate that cholinergic neurons utilize distinct molecular signals to establish synapses with GABAergic motor neurons versus body wall muscles , thus revealing that a single presynaptic neuron establishes divergent connections by employing parallel molecular strategies for the formation of connections with each postsynaptic partner .
To establish a system to investigate mechanisms instructing synaptic connectivity , we labeled post-synaptic specializations on dorsally directed GABAergic DD neurons using cell-specific expression ( flp-13 promoter ) of the GFP-tagged acetylcholine receptor subunit ACR-12 . Prior work showed that ACR-12 receptors in GABAergic motor neurons are clustered opposite cholinergic terminals and GABAergic expression of ACR-12::GFP rescues acr-12 mutant phenotypes ( Barbagallo et al . , 2017; Petrash et al . , 2013 ) . Moreover , postsynaptic ACR-12::GFP clusters relocate appropriately during developmental synaptic remodeling of the DD neurons , suggesting these clusters faithfully report synaptic inputs ( He et al . , 2015; Howell et al . , 2015 ) . The morphology of DD neurons is highly polarized , facilitating clear visualization of the axonal and dendritic neuronal compartments . In the present work , we focus much of our analysis on the spatially isolated neurites of the DD1 neuron ( Figure 1A–C ) . In adults , the anterior DD1 process extends from the soma to enter the ventral nerve cord fascicle ( the dendritic compartment ) , where prior EM studies show that approximately 26 synaptic inputs from cholinergic neurons are concentrated ( the synaptic region , Figure 1C , D ) ( White et al . , 1978 , 1976 ) . The process then crosses the longitudinal midline of the worm via a commissural connection and enters the dorsal nerve cord where it forms en passant synaptic outputs onto body wall muscles ( the axonal compartment ) ( Figure 1A , B ) ( White et al . , 1976 ) . We find that ACR-12::GFP receptor clusters in DD1 are confined to the synaptic region of the ventral dendritic process in the mature animal ( Figure 1C ) . As C . elegans synapses are formed en passant , pre- and post-synaptic specializations typically appear , at the light level , to be localized along the main shafts of neuronal processes . Surprisingly , we noted that the majority of ACR-12::GFP clusters do not appear localized to the shaft of the primary DD1 dendritic process , instead appearing to protrude from the primary DD1 dendrite shaft ( Figure 1C ) . To investigate this finding in more detail , we examined morphological features in the synaptic region of the DD1 dendrite ( Figure 1D ) . Intriguingly , we noted finger-like structures ( ~0 . 3–1 μm in length ) projecting outward from the DD1 dendrite in this region ( Figure 1D , Figure 1—figure supplement 1A ) . In contrast , these structures are not present in the asynaptic region of the process immediately adjacent to the cell soma ( Figure 1D ) . These protrusions are obscured by the processes of other ventral cord neurons when using promoters that provide for more broad expression ( e . g . unc-47 ) , and are therefore most clearly identifiable with specific labeling of DD neurons ( Figure 1—figure supplement 1B ) . Spine-like protrusions are also clearly identifiable in the dendrites of more posterior DD neurons ( Figure 1—figure supplement 1C ) , but are not apparent in a related class of post-embryonic born , ventrally directed GABAergic ( VD ) neurons ( Figure 1—figure supplement 1D ) , although the density of VD processes may complicate their detection . The dendritic protrusions concentrate clusters of ACR-12 receptors at their tips ( Figure 1E ) , and over 60% of ACR-12::GFP clusters appear localized to protrusions ( Figure 1F ) . The dendritic protrusions increase in abundance through larval development , and this increase correlates well with a similar developmental increase in ACR-12::GFP receptor clusters ( Figure 1—figure supplement 2 ) . Together , our results provide evidence that cholinergic receptors cluster at morphologically distinct finger-like structures present on DD neuron dendrites , raising the interesting possibility that these structures serve similar roles to dendritic spines in the mammalian nervous system . To explore the above possibility further , we evaluated the spatial relationship between ACR-12 clusters located on these spine-like structures and cholinergic release sites . We found that dendritic protrusions and ACR-12 receptor clusters are both located opposite clusters of cholinergic synaptic vesicles ( Figure 2A , B ) , indicating that these likely represent mature synapses . We therefore next investigated whether these clusters indicate post-synaptic receptors residing at the cell surface . To address this question , we inserted an HA epitope tag into the extracellular C-terminus of ACR-12::GFP ( diagram in Figure 2C ) ( Gottschalk and Schafer , 2006 ) . Injection of Alexa594 conjugated anti-HA antibody into live transgenic animals expressing this construct produces specific labeling in the DD1 synaptic region , and is also evident in coelomocytes ( scavenger cells that take up excess antibody ) , confirming successful injection ( Figure 2C ) . The anti-HA signal colocalizes with ACR-12::GFP clusters in the synaptic region of DD1 , but is not evident in the cell soma . In contrast , the intracellular GFP moiety produces fluorescence that is evident in both the soma and the synaptic region of the dendrite , representing both synaptic and internal receptor pools ( Figure 2C ) . Injection of anti-GFP antibody did not produce specific labeling , confirming that the intracellularly positioned GFP is not accessible to antibody ( Figure 2—figure supplement 1A ) . Our analysis of ACR-12 localization in DD1 indicates that ACR-12 is incorporated into mature receptor complexes that are specifically targeted for transport to DD neuron dendrites , and reside on the cell surface at post-synaptic sites . We next sought to gain an understanding of the subunit composition of ACR-12 receptors in GABAergic neurons . Most acetylcholine receptors are formed as heteromeric combinations of five subunits . Prior work has demonstrated that partially or improperly assembled acetylcholine receptor intermediates are not transported out of the ER and are instead targeted for degradation ( Blount and Merlie , 1990; Merlie and Lindstrom , 1983 ) . We therefore reasoned that genetic ablation of obligate ACR-12 partnering subunits , by interfering with assembly , transport , and synaptic targeting of ACR-12 receptor complexes , could provide an efficient strategy for identifying subunit partners . We found that single mutations in the acetylcholine receptor subunit genes unc-38 , unc-63 , lev-1 or unc-29 strongly decrease ACR-12::GFP clustering . In wild type , we observe approximately 15 receptor clusters within the DD1 synaptic region . These clusters are eliminated almost completely with mutation of these subunit genes ( Figure 2—figure supplement 1B , C ) . In contrast , mutations in other nAChR subunit genes with previously reported neuronal expression , such as acr-9 and acr-14 ( Cinar et al . , 2005; Fox et al . , 2005 ) , do not disrupt synaptic ACR-12 clustering ( Figure 2—figure supplement 1B ) . Mutations in genes important for AChR assembly and trafficking ( unc-50 , unc-74 , and ric-3 ) ( Boulin et al . , 2008; Eimer et al . , 2007; Halevi et al . , 2002; Haugstetter et al . , 2005 ) also abolish synaptic clusters of ACR-12::GFP ( Figure 2—figure supplement 1B , C ) . GABA neuron-specific expression of wild type cDNAs encoding individual AChR subunits ( UNC-63 or UNC-38 ) or accessory proteins in the respective mutants is sufficient to restore ACR-12 clustering to wild type levels , providing support that both gene classes act cell-autonomously in GABA neurons to promote receptor assembly , maturation and synaptic delivery ( Figure 2—figure supplement 1B , C ) . Cell-specific expression of either UNC-29::GFP or UNC-63::GFP in DD neurons produces punctate labeling in the DD1 synaptic region that closely resembles ACR-12::GFP clustering ( Figure 2D ) . Mutation of acr-12 in these transgenic animals significantly reduces UNC-29::GFP or UNC-63::GFP receptor clusters and fluorescence signal in the DD1 synaptic region ( Figure 2D , Figure 2—figure supplement 1D , E ) , providing further evidence that they coassemble with ACR-12 . Together , our results indicate UNC-38 , UNC-63 , LEV-1 , UNC-29 and ACR-12 subunits coassemble into a pentameric acetylcholine receptor in GABAergic neurons . To decipher molecular mechanisms by which these newly defined post-synaptic structures develop , we examined ACR-12::GFP labeling in DD neurons of 28 strains carrying mutations in 36 candidate genes . These candidates predominantly encode scaffold and cell-cell interaction proteins previously implicated in synapse formation , many of which have previously demonstrated expression in GABAergic neurons ( Cinar et al . , 2005 ) ( Figure 3—figure supplement 1 ) . Mutations in most genes tested ( 75% ) produce no significant disruption in ACR-12 receptor clustering ( Figure 3A , light blue ) . A second group , comprising 8 of the 36 genes analyzed ( Figure 3A , green ) , produces mild to moderate ( 26–39% ) decreases in ACR-12 clustering . Many of the mutants in this second group identify genes ( e . g . lev-10 , madd-4 ) that perform previously characterized functions in neuromuscular synapse development and muscle AChR clustering ( Gally et al . , 2004; Pinan-Lucarré et al . , 2014 ) ( Figure 3—figure supplement 2A , B ) , but appear to play comparatively minor roles in establishing synaptic connectivity with GABAergic neurons . One of the 36 mutations tested is clearly distinguishable by a striking decrease in ACR-12 clustering . Mutation of the nrx-1 gene ( orange , Figure 3A ) does not significantly affect transgene expression in GABAergic neurons ( Figure 3—figure supplement 2C , D ) , but produces a ~70% reduction in ACR-12 receptor clustering ( nrx-1 ( ok1649 ) , p<0 . 0001 ) ( Figure 3A , B ) . nrx-1 encodes the sole C . elegans ortholog of the synaptic organizer neurexin . Neurexin has been well documented to play roles in mammalian synapse formation and function ( Chen et al . , 2017; Dean et al . , 2003; Graf et al . , 2004; Missler et al . , 2003 ) . Roles for NRX-1 in C . elegans synapse formation remain , by comparison , less well defined . Moreover , roles for neurexin in establishing divergent connectivity have not been previously addressed in any system . Importantly , we do not observe appreciable alterations in the clustering of muscle AChRs in nrx-1 mutants ( Figure 3C ) , similar to previously reported findings ( Hu et al . , 2012 ) . The profound alterations in ACR-12 localization described above , coupled with the lack of effect on muscle AChRs , therefore warranted an in-depth analysis of cholinergic synapses with GABAergic neurons in nrx-1 mutants . We first sought to distinguish whether nrx-1 performs a specific role in synapse formation or serves more generalized functions in the developmental maturation of DD neurons . We studied the developmental remodeling of DD neurons , characterized by the dorsoventral repositioning of synaptic markers , which occurs at the L1/L2 transition in wild type animals ( Jin and Qi , 2018; White et al . , 1978 ) . In particular , we examined the repositioning of pre- and post-synaptic markers expressed specifically in DD motor neurons ( Figure 3—figure supplement 3A ) . Presynaptic remodeling , as measured using the synaptic vesicle marker mCherry::RAB-3 , proceeds normally in nrx-1 mutants , indicating that this process does not require neurexin expression . The clustering of ACR-12 receptors , however , is impaired both prior to and following remodeling in nrx-1 mutants , suggesting that NRX-1 is required for receptor clustering in each of these developmental stages . As is the case for mammals , the C . elegans nrx-1 locus encodes both long ( nrx-1L ) and short ( nrx-1S ) neurexin isoforms ( Figure 3—figure supplement 3B ) . The long isoform encodes a single pass transmembrane protein harboring intracellular PDZ binding and interleaved extracellular LNS ( laminin-neurexin-sex hormone-binding globulin ) and EGF-like domains ( Figure 3D ) . The ok1649 allele generates an in-frame deletion , eliminating 861 bp predicted to encode an extracellular LNS domain , raising the possibility that partial NRX-1 function may still be present in this strain . We therefore expanded our analysis to include additional nrx-1 deletions , three of which ( ds1 , tm1961 , and nu485 ) are predicted to primarily impact the long isoform , while another ( wy778 ) removes the transmembrane and cytoplasmic domains shared by all NRX-1 isoforms ( Calahorro and Ruiz-Rubio , 2013; Maro et al . , 2015; Tong et al . , 2015 ) . All of the deletions tested disrupt ACR-12 receptor clustering in DD GABAergic neurons , with the most severe disruptions occurring in nrx-1 ( wy778 ) and nrx-1 ( nu485 ) mutants ( Figure 3E ) . Strikingly , we also observe significant disruption of ACR-12 receptor clustering in the VD GABAergic neurons ( Figure 3F , G ) of nrx-1 mutants , indicating a requirement for nrx-1 at synapses onto both GABAergic neuron classes . In some instances ( 14 of 34 animals for nrx-1 ( wy778 ) ) , we noted that a few ACR-12 receptor clusters remain detectable . We investigated whether these residual ACR-12 receptors are localized to the cell surface by injecting Alexa594 conjugated anti-HA antibody into nrx-1 mutants expressing ACR-12::GFP::3xHA . Antibody fluorescent signal clearly colocalizes with ACR-12::GFP fluorescence , providing evidence that the few remaining receptor clusters in nrx-1 mutants are present at the cell surface ( Figure 3—figure supplement 3C ) . We interpret this result to indicate that neurexin is not essential for membrane insertion , although we can’t rule out that this process may occur less efficiently in nrx-1 mutants . We next examined whether the localization of putative ACR-12 partnering subunits is also disrupted by mutation of nrx-1 . As noted above , ACR-12 receptors are formed as heteromeric complexes in GABAergic neurons that likely incorporate the UNC-29 and UNC-63 AChR subunits . Neurexin deletion ( wy778 ) reduces UNC-29::GFP and UNC-63::GFP clusters in DD1 by 60% , consistent with a requirement for neurexin in the proper localization of mature , heteromeric receptor complexes ( Figure 3—figure supplement 3D , E ) . In contrast to our findings for GABAergic neurons , nrx-1 deletion does not appreciably disrupt the clustering of AMPA-type glutamate receptors in interneurons ( Figure 3—figure supplement 4A ) , consistent with the idea that neurexin is not globally required for the establishment of synaptic connectivity in worms . Additionally , we do not observe an appreciable decrease in cholinergic synaptic vesicle clusters ( acr-2::SNB-1::GFP ) , although this analysis does not rule out all potential presynaptic defects ( Figure 3—figure supplement 4B ) . To elucidate mechanisms by which nrx-1 may instruct the formation of synapses between cholinergic and GABAergic motor neurons , we evaluated mutations in the nlg-1 gene . nlg-1 encodes the sole C . elegans ortholog of neuroligin , a well characterized binding partner of neurexin ( Banerjee et al . , 2017; Boucard et al . , 2005; Hu et al . , 2012; Ichtchenko et al . , 1995 , Ichtchenko et al . , 1996 ) . We find that mutation of nlg-1 produces no appreciable defects in ACR-12 receptor clustering ( Figure 3A , B ) , indicating , surprisingly , that NRX-1 operates independently of NLG-1 to direct post-synaptic development in GABAergic neurons . We next investigated whether neurexin is required for the outgrowth or stabilization of the spine-like processes we observe in DD dendrites . Wild type animals ( at the L4 stage ) have an average of 7–8 of these spiny protrusions within the synaptic region of the anterior DD1 dendrite ( Figure 4A , B ) . In contrast , nrx-1 mutants have strikingly reduced numbers of spiny protrusions . For example , only two spiny protrusions are visible on average in DD1 dendrites of nrx-1 ( ok1649 ) , and nrx-1 ( wy778 ) mutants show a near complete absence of spines ( Figure 4A , B ) . nrx-1 deletion disrupts both ACR-12 receptor clustering and spine outgrowth in posterior DD neurons to a similar extent as observed for DD1 ( Figure 4C , Figure 4—figure supplement 1A ) , indicating that the requirement for nrx-1 is shared across DD neurons . Roughly 15% of the receptor clusters remaining in nrx-1 ( ok1649 ) mutants are associated with the remaining spines ( compared with 67% in wild type ) . In contrast , the number of dendritic receptor clusters is not appreciably altered by nrx-1 deletion ( p=0 . 67 , student’s t test ) , suggesting NRX-1 may preferentially regulate spine-associated receptor clusters in DD neurons . nrx-1 deletion also significantly reduces spiny protrusions in L2 animals ( the earliest stage at which they are visible ) ( Figure 4—figure supplement 1B ) , suggesting that neurexin is required for initial spine outgrowth , although an additional role in maintenance is also possible . In accordance with our finding that nlg-1 is not required for ACR-12::GFP localization , mutation of nlg-1 does not alter spine number ( Figure 4A , B ) , arguing against an essential role for neuroligin in the formation of cholinergic synapses with GABAergic neurons . Likewise , spine number is not appreciably altered in either acr-12 or unc-63 mutants ( Figure 4A , B ) , indicating that spine outgrowth proceeds normally in the absence of functional ACR-12 receptors . To understand how NRX-1 regulates the formation of cholinergic synapses with GABAergic neurons , we sought to define the requirements for nrx-1 expression . We first examined expression of a nrx-1L::GFP transcriptional reporter incorporating ~4 . 8 kb of sequence upstream of the nrx-1L start site , and found that this reporter is strongly expressed in cholinergic motor neurons ( Figure 5A ) . We next asked whether specific nrx-1 expression in cholinergic neurons is sufficient to rescue the post-synaptic defects of nrx-1 mutants . We found that cholinergic nrx-1L expression reverses the ACR-12 clustering defects of nrx-1 ( wy778 ) mutants , while expression in either GABAergic neurons or muscles fails to rescue ( Figure 5B–D ) . Notably , cholinergic expression of nrx-1L also restores spine-like protrusions in nrx-1 mutants ( Figure 5—figure supplement 1A , B ) . Thus , our results suggest that presynaptic NRX-1 acts non-autonomously to direct post-synaptic assembly in GABAergic neurons . To investigate this possibility in more detail , we examined the subcellular localization of NRX-1 by expressing GFP-tagged NRX-1L ( NRX-1L::GFP ) ( Maro et al . , 2015 ) in a subset of cholinergic neurons ( DA/DB ) . Expression of unc-129::NRX-1L::GFP produces discrete puncta along the dorsal nerve cord where the synaptic outputs of DA/DB neurons are located ( Figure 5E ) . NRX-1L::GFP clusters colocalize with clusters of mCherry::RAB-3 fluorescence , providing evidence that NRX-1L is preferentially localized to cholinergic presynaptic sites ( Figure 5E , F ) . Intriguingly , cholinergic expression of a rescuing transgene lacking the PDZ binding motif located at the intracellular NRX-1 C-terminus ( nrx-1LΔPDZ ) in nrx-1 mutants can also rescue ACR-12 clustering defects ( Figure 5—figure supplement 2A , B ) . NRX-1LΔPDZ::GFP is exclusively expressed in cholinergic axons and partially colocalizes with mCherry::RAB-3 fluorescence , indicating that NRX-1 presynaptic localization and function is possible without PDZ protein binding via this motif ( Figure 5—figure supplement 2C–E ) . Loss of nlg-1 function does not affect the axonal localization of NRX-1L::GFP ( Figure 5—figure supplement 3A , C–D ) . NRX-1L::GFP localization is also not appreciably altered by acr-12 deletion , suggesting that the positioning of NRX-1 at presynaptic terminals occurs independently of post-synaptic receptor clustering ( Figure 5—figure supplement 3B–D ) . Our results indicate NRX-1 positioning at presynaptic sites occurs independently of post-synaptic receptor localization , and raise the intriguing possibility that NRX-1 localization to the presynaptic domain may serve as an initiation signal for developmental maturation of post-synaptic specializations in GABAergic neurons . We have previously demonstrated that mutation of the COE-type ( Collier/Olf/Ebf ) transcription factor unc-3 disrupts ACR-12 clustering in VD GABAergic neurons ( Barbagallo et al . , 2017 ) . In light of our findings here that nrx-1 deletion similarly disrupts ACR-12 clustering and spine-like protrusion outgrowth in DD neurons ( Figure 6A , B ) , we next investigated the role of UNC-3 transcriptional regulation in the development of cholinergic connectivity with GABAergic DD neurons . Prior work has shown that activity of UNC-3 is essential for the specification of cholinergic neurotransmitter identity ( Kratsios et al . , 2015 , 2011 ) . To investigate the requirement for unc-3 in ACR-12 clustering , we first evaluated whether cholinergic transmission itself is critical for the development of post-synaptic specializations on DD neurons . We found that mutations in the unc-17 cholinergic vesicular ACh transporter produces no appreciable changes in spine-like protrusion number or ACR-12 clusters ( Figure 6B ) , arguing against a strong requirement for cholinergic transmission in the formation of these structures . We next asked whether UNC-3 transcriptional regulation of nrx-1 expression is critical for the development of cholinergic connectivity with GABAergic neurons . To address this question , we first tested whether unc-3 is required for expression of the nrx-1L::GFP transcriptional reporter described above . We found that mutation of unc-3 significantly reduces nrx-1L::GFP fluorescence in motor neuron cell bodies of the ventral nerve cord , as well as the majority of fluorescence in ventral cord processes ( Figure 6C , D ) . The remaining ventral cord GFP fluorescence is associated with the processes of head neurons that project into the nerve cord , which are presumably not subject to unc-3 regulation . We next used fluorescent in situ hybridization ( FISH ) to determine the effects of unc-3 mutation on nrx-1 mRNA abundance . Fluorescent signals indicating nrx-1 mRNA are clearly associated with cholinergic motor neuron cell bodies ( co-labeled with unc-17::GFP ) in wild type animals ( Figure 6E ) , consistent with our prior analysis using the nrx-1L::GFP transcriptional reporter . The smFISH signals are strongly diminished in nrx-1 ( nu485 ) deletion mutants ( Figure 6E , F ) , confirming they accurately report nrx-1 mRNA abundance . Labeled nrx-1 mRNA signals in cholinergic motor neurons are strikingly reduced by mutation of unc-3 , consistent with the possibility that nrx-1 is a transcriptional target of unc-3 ( Figure 6E , F ) . We noted that a second nrx-1L transcriptional reporter incorporating only ~2 kb of nrx-1 regulatory sequence did not produce strong fluorescence in ventral cord motor neurons ( Figure 6C , D ) . We reasoned that regulatory elements required for nrx-1 expression in these neurons may be present in the sequence that differs across these two transcriptional reporters ( 2 versus 4 . 8 kb ) ( Figure 6C , D ) . As both mammalian COE transcription factors and UNC-3 bind a conserved COE binding motif ( TCCCNNG/AG/AG/AG/A ) to regulate transcription of target genes ( Kim et al . , 2005; Kratsios et al . , 2011; Wang et al . , 2015 , 1993 ) , we searched for COE binding motifs within this region . We identified a potential COE motif ( TCCCAAAGGG ) located approximately 20 bp from the 5’ end of the 4 . 8 kb nrx-1L::GFP transcriptional reporter . Mutation of this site ( TCCCAAAGGG >>TAAAAAAGGG ) within the 4 . 8 kb nrx-1L::GFP transcriptional reporter eliminates all fluorescence from ventral cord motor neurons , while fluorescence in the processes extending from head neurons remains visible ( Figure 6C , D ) , offering evidence that UNC-3 directly regulates nrx-1 transcription in ventral cord neurons . We reasoned that forced expression of nrx-1L in cholinergic neurons using a promoter not subject to unc-3 regulation may allow NRX-1 to coordinate synapse development independently of UNC-3 transcriptional regulation . We expressed the nrx-1L isoform using a regulatory region of the unc-3 gene that drives expression in ventral cord cholinergic neurons ( Barbagallo et al . , 2017 ) . We found that cholinergic-specific expression of the nrx-1L isoform significantly restored receptor clusters in the dendritic region of unc-3 animals ( Figure 6G ) , indicating that the lack of ACR-12 receptor clusters in unc-3 mutants is largely driven by the absence of nrx-1 expression , although additional phenotypes associated with mutation of unc-3 may contribute ( e . g . variable nerve cord defasciculation ) ( Barbagallo et al . , 2017 ) . These findings define the gene regulatory mechanisms controlling neurexin expression in presynaptic neurons and illustrate their involvement in the establishment of synaptic connectivity . To investigate how nrx-1 deletion impacts the spatial arrangement of pre- and post-synaptic specializations , we examined strains coexpressing ACR-12::GFP in GABAergic neurons with the synaptic vesicle marker mCherry::RAB-3 in cholinergic neurons . In the wild type , ACR-12 receptor clusters at the tips of spiny protrusions are submerged within the presynaptic domains of cholinergic axons , where synaptic contacts are presumably located ( Figure 7A , left ) . In nrx-1 mutants , however , we noted a gap between the neurites of the pre- and post-synaptic neurons ( Figure 7A , right ) , suggesting that nrx-1 coordinates the extension of receptor-bearing spiny protrusions to presynaptic domains of cholinergic axons . These results , in combination with the receptor clustering defects in VD GABAergic neurons and the lack of an appreciable effect on muscle synapses described above , predict that nrx-1 deletion would impair cholinergic synaptic activation of GABAergic neurons , while cholinergic transmission onto muscles would remain unaffected . We recorded Ca2+ transients from either GABAergic motor neurons or muscles immediately following presynaptic cholinergic depolarization in order to address this question ( Figure 7—figure supplement 1A ) . We used combined cell-specific expression of Chrimson for cholinergic depolarization ( Klapoetke et al . , 2014; Larsch et al . , 2015 ) , and GCaMP6s for monitoring [Ca2+] changes ( Chen et al . , 2013 ) in either post-synaptic GABAergic motor neurons or body wall muscles ( Figure 7B , E ) . Strikingly , we found that nrx-1 deletion disrupts GABA neuron Ca2+ transients in response to cholinergic stimulation , but produces no appreciable effect on muscle Ca2+ transients , consistent with a specific requirement for nrx-1 in the development of functional connectivity between cholinergic and GABAergic neuron , but not muscle , synaptic partners ( Figure 7B–G ) . Cholinergic depolarization ( 5 s ) evokes robust stimulus-coupled Ca2+ transients in both GABAergic neurons ( 67% of stimuli ) and muscles ( 88% of stimuli ) that occur within 250 ms of stimulus onset ( average response latency: 0 . 22 ± 0 . 06 s in motor neurons and 0 . 25 ± 0 . 02 s in muscles ) . These transients are not observed in the absence of cholinergic Chrimson expression or in the absence of retinal ( Figure 7—figure supplement 1B , C ) , consistent with a requirement for presynaptic Chrimson-mediated depolarization . In both cell types , evoked Ca2+ transients rise rapidly following stimulation ( τrise: 0 . 48 ± 0 . 08 s in GABA neurons; 0 . 74 ± 0 . 06 s in muscles ) , and persist throughout the duration of stimuli before decaying to baseline . Motor neuron transients are typically shorter in duration ( mean duration: 7 . 7 ± 1 s in GABA neurons; 10 . 8 ± 0 . 5 s in muscles ) and decay more rapidly ( τdecay: 1 . 1 ± 0 . 4 s in GABA neurons; 4 . 6 ± 0 . 6 s in muscles ) compared with muscle transients , likely reflecting differences in both synaptic connectivity and physiology across the two cell types . For both GABA neurons and muscles , evoked Ca2+ responses are eliminated almost completely by mutations that impair post-synaptic AChR function in the respective cell types ( acr-12 or unc-29;acr-16 , respectively ) . Specifically , acr-12 deletion reduces the mean peak amplitude of GABA neuron calcium responses to cholinergic stimulation by 60% ( Figure 7C–D , Figure 7—figure supplement 1D ) and increases the failure rate ( no response to stimulation ) by 53% compared to wild type , consistent with prior electrophysiology studies ( Petrash et al . , 2013 ) . Similarly , for muscles , the mean peak amplitude of calcium responses to cholinergic stimulation is reduced by 95% in unc-29;acr-16 double mutants ( Figure 7F–G , Figure 7—figure supplement 1E ) , and the failure rate is increased roughly 15-fold to 77% , consistent with prior electrophysiology studies of evoked synaptic responses in these double mutants ( Francis et al . , 2005 ) . nrx-1 deletion reduces the mean peak amplitude of GABA neuron calcium responses to cholinergic stimulation by roughly 71% ( Figure 7C–D , Figure 7—figure supplement 1D ) , and increases the failure rate for GABA neuron recordings by 47% . By comparison , nrx-1 deletion does not produce a significant decrease in either mean peak fluorescence ( Figure 7F–G , Figure 7—figure supplement 1E ) or the failure rate in recordings of evoked muscle activity . Together , these findings support a specific requirement for nrx-1 in cholinergic transmission onto GABA neurons , while nrx-1 appears dispensable for transmission onto muscles under our recording conditions . Notably , both the failure rate and mean peak amplitude of evoked GABA neuron Ca2+ responses are restored to wild type levels with expression of a rescuing nrx-1L transgene in nrx-1 mutants using a cholinergic neuron-specific promoter ( Figure 7C–D , Figure 7—figure supplement 1D ) . Consistent with a requirement for NRX-1 in cholinergic synaptic connectivity with GABAergic motor neurons , automated worm track analysis showed that nrx-1 mutants display defects in the amplitude of dorsoventral bending , a feature of worm movement previously associated with GABAergic function ( McIntire et al . , 1993; Petrash et al . , 2013 ) . These effects are rescued with cell-specific expression of nrx-1L in cholinergic neurons ( Figure 7—figure supplement 1F ) . Thus , presynaptic nrx-1 expression in cholinergic neurons is required for synaptic connectivity between cholinergic and GABAergic motor neurons , and deficits in these connections alter motor performance . Together , our data indicate that NRX-1 located in presynaptic cholinergic neurons is required for establishing synaptic connectivity with partnering GABAergic neurons , but not muscle cells . NRX-1 signaling promotes both receptor clustering and the outgrowth of post-synaptic spine-like morphological features in GABAergic dendrites . Our findings support a model where distinct synaptic organizers , acting on specific post-synaptic targets , are coordinately regulated with neuronal identity , perhaps offering a mechanism for independent developmental regulation of synaptic outputs across alternate partners ( Figure 8 ) .
Using candidate deletion analysis and cell-specific rescue , we identified AChR subunits and accessory proteins required for the assembly and localization of ACR-12 receptors in GABAergic neurons , defining the subunit composition of this neuronal receptor . Our findings implicate four additional receptor subunits ( UNC-38 , UNC-63 , UNC-29 , LEV-1 ) that co-assemble with ACR-12 to form pentameric receptor complexes in GABAergic neurons , and demonstrate that three accessory proteins ( UNC-74 , UNC-50 , RIC-3 ) with more generalized roles in AChR assembly and maturation ( Boulin et al . , 2008; Halevi et al . , 2002; Jospin et al . , 2009 ) are also required . These findings provide evidence that ACR-12 receptors in GABA neurons are similar in subunit composition to muscle L-AChRs , differing only in the inclusion of the ACR-12 subunit in GABA neurons , whereas muscle L-AChRs incorporate the LEV-8 subunit ( Boulin et al . , 2008; Towers et al . , 2005 ) . Consistent with our analysis , a prior study showed that ACR-12 can be co-purified with the UNC-29 or LEV-1 subunits ( Gottschalk et al . , 2005 ) . We have previously shown that unc-29 is expressed in GABAergic motor neurons , and UNC-29::GFP localizes similarly to ACR-12::GFP in DD neurons , both in the mature animal and during developmental remodeling of these neurons ( He et al . , 2015 ) . We show that ACR-12 receptor complexes are concentrated at the tips of spine-like dendritic protrusions in GABAergic DD neurons . Spiny processes associated with D-type GABAergic neurons had been noted in prior electron microscopy studies ( White et al . , 1976; 1986 ) , but , to our knowledge , were not characterized further . We find that these AChR-containing dendritic protrusions are apposed by presynaptic clusters of cholinergic vesicles and increase in number during the course of larval development , perhaps representing new synaptic connections formed with post-embryonic born cholinergic neurons that are integrated into the circuit following the L1/L2 transition ( White et al . , 1978 ) . While further investigation of these structures will undoubtedly reveal additional insights , the characteristics we define here raise the interesting possibility that these dendritic protrusions are structural specializations for housing neurotransmitter receptors and other proteins required for post-synaptic signaling , perhaps representing an evolutionary precursor to mammalian dendritic spines . nrx-1 deletion impairs both AChR localization and spiny outgrowths in DD neurons , creating a gap between the pre- and post-synaptic neurons . Similarly , in Drosophila , mutation of the single neurexin gene dnrx causes disorganization of synaptic structure at the neuromuscular junction ( Li et al . , 2007 ) . Knockout of two out of the three mouse alpha neurexins reduces dendrite length and total spine number in the cortex ( Dudanova et al . , 2007 ) , though significant numbers of dendritic spines remain detectable . In our studies , we find that nrx-1 is required for ACR-12 receptor localization in GABAergic DD and VD neurons , although we observe spine-like protrusions only in DD neurons . These findings may point toward the idea that NRX-1 is not solely involved in directing spine development , but serves an additional role in receptor clustering . Further , we show that dendritic protrusions form independently of a requirement for AChRs containing either ACR-12 or UNC-63 , offering additional support that spine outgrowth and receptor clustering may be independently regulated by NRX-1 . Similarly , dendritic spines on mouse CA1 pyramidal neurons form normally in the absence of functional glutamate receptors ( Lu et al . , 2013 ) . Although GABA neuron ACR-12 AChRs and muscle L-AChRs share very similar subunit composition ( Boulin et al . , 2008; Lewis et al . , 1980 ) , we find that genes required for proper localization of muscle L-AChRs ( e . g . madd-4 , lev-10 ) play comparatively minor roles at synapses onto GABA neurons . Conversely , loss of nrx-1 function specifically affects GABAergic , but not muscle , AChR clustering , and cholinergic transmission onto muscles appears largely unaffected . Our cell-specific rescue experiments indicate nrx-1 acts in cholinergic neurons to coordinate post-synaptic development in GABAergic neurons via trans-synaptic signaling . Numerous studies support neuroligin as a primary trans-synaptic binding partner with neurexin ( Boucard et al . , 2005; Comoletti et al . , 2006; Ichtchenko et al . , 1995 , Ichtchenko et al . , 1996; Nguyen and Südhof , 1997 ) . Indeed , C . elegans NRX-1/neurexin function has been characterized almost exclusively in the context of its partnership with NLG-1/neuroligin . A retrograde neurexin-neuroligin signaling pathway that regulates neurotransmitter release has been described , involving signaling through the Ca2+ channel auxiliary subunit UNC-36/α2δ ( Hu et al . , 2012; Tong et al . , 2017 ) . Consistent with this representing a distinct mechanism from that described in our work , mutation of unc-36 has no appreciable effect on post-synaptic development in our experiments . Postsynaptic expression of NLG-1/neuroligin is required at GABAergic synapses . However , MADD-4/Punctin likely acts as a major presynaptic partner for NLG-1 in this case , with NRX-1 playing a comparatively minor role ( Maro et al . , 2015; Tong et al . , 2015; Tu et al . , 2015 ) . Finally , recent work also implicates neurexin-neuroligin signaling in sexually dimorphic neurite plasticity ( Hart and Hobert , 2018 ) . In contrast to these studies , we find that NRX-1 operates independently of NLG-1 to direct the formation of cholinergic synapses with GABAergic neurons . How therefore might presynaptic neurexin direct postsynaptic maturation ? Our analysis shows that presynaptic NRX-1 localizes properly in the absence of acr-12 , arguing against a requirement for direct binding of neurexin to the postsynaptic receptor . Prior studies offer strong evidence for alternate neurexin binding partners that support trans-synaptic signaling ( Boucard et al . , 2012; de Wit et al . , 2009; Ko et al . , 2009; Missler et al . , 1998; Petrenko et al . , 1996; Pettem et al . , 2013; Sugita et al . , 2001; Uemura et al . , 2010 ) . For several of these gene families ( e . g . neurexophilins , cerebellins ) , clear C . elegans orthologs are not present . Others are included in our candidate analysis ( e . g . casy-1/calsyntenin , lat-2/latrophilin ) , but single gene mutations do not produce post-synaptic defects comparable to mutation of nrx-1 , suggesting either the possibility of a novel post-synaptic nrx-1 binding partner or redundant post-synaptic mechanisms . Additional genetic or proteomic studies will be required to distinguish between these possibilities and address this important question . Transcriptional regulators of neurexin expression are only beginning to be elucidated ( Runkel et al . , 2013 ) . Previously , we found that mutation of the COE-type transcription factor unc-3 disrupts AChR clustering in GABAergic dendrites ( Barbagallo et al . , 2017 ) . These disrupted clusters are unlikely to reflect a requirement for acetylcholine release in AChR clustering , as neither tetanus toxin expression nor mutation of unc-17/vAChT produces appreciable defects in ACR-12 clustering ( Barbagallo et al . , 2017 ) ( this study ) . Here , we find that neurexin is a transcriptional target of UNC-3 , and we propose that UNC-3 regulation of nrx-1 expression directs development of postsynaptic specializations in GABAergic neurons . Prior work indicates that UNC-3 transcriptional regulation of MADD-4/Punctin is essential for proper development of cholinergic synapses with muscles ( Kratsios et al . , 2015 ) . However , madd-4 is dispensable for cholinergic receptor clustering in neighboring GABAergic neurons ( Barbagallo et al . , 2017 ) ( this study ) . Thus , cholinergic connectivity with distinct synaptic targets–muscles and GABAergic neurons–is coordinately regulated with cholinergic neuronal identity by unc-3 transcriptional control of alternate synaptic organizers . Our work here defines a novel yet essential role for NRX-1 during synapse formation . Neurexin acts in a target-specific manner to coordinate postsynaptic development . Presynaptic neurexin instructs receptor localization and the development of spine-like processes in postsynaptic GABAergic neurons , and nrx-1 expression is critical for cholinergic transmission onto GABAergic neurons , while not required for signaling onto neighboring muscle . Our work suggests synaptic target-specific utilization of organizers such as neurexin may specify divergent connectivity , and provide a molecular mechanism for target-specific regulation of synapse development .
C . elegans strains were maintained at room temperature ( 22–24°C ) on nematode growth media plates ( NGM ) seeded with the Escherichia coli strain OP50 . All strains are derivatives of the N2 Bristol strain ( wild type ) . Transgenic strains were obtained by microinjection to achieve germline transformation ( Mello et al . , 1991 ) and identified with co-injection markers as previously ( Barbagallo et al . , 2017 ) . Integrated lines were produced by X-ray irradiation and outcrossed to wild type . A complete list of all strains used in this work is included in Supplementary file 1 . Plasmids were constructed using the two-slot Gateway Cloning system ( Invitrogen ) as described previously ( Bhattacharya et al . , 2014 ) and confirmed by restriction digest and/or sequencing as appropriate . All plasmids and primers used in the study are described in Supplementary files 2 and 3 respectively . Tagged receptor constructs: To generate flp-13::ACR-12::GFP::3xHA , ACR-12::GFP was amplified from pDEST-38 , removing the original stop codon and adding a 3X HA tag . The product ( 3195 bp ) was ligated into a destination vector to generate pDEST-113 . pDEST-113 was recombined with pENTR-3’-flp-13 to create pAP138 ( flp-13::ACR-12::GFP::3xHA ) . To generate flp-13::UNC-63::GFP , 5’ and 3’ fragments of unc-63 cDNA were PCR amplified from pDEST-57 , ligated into pPD117 . 01 ( mec-7::GFP ) , converted into the destination vector pDEST-79 , and recombined with pENTR-5’-flp-13 to create pAP84 ( flp-13::UNC-63::GFP ) , where GFP is inserted into the intracellular loop of UNC-63 . AChR subunit and accessory rescue constructs: Wild type unc-38 ( 1536 bp ) , unc-63 ( 1509 bp ) , unc-74 ( 1344 bp ) , unc-50 ( 906 bp ) , and ric-3 ( 1137 bp ) rescue constructs were PCR amplified and ligated into destination vectors to generate pDEST-51 , pDEST-57 , pDEST-58 , pDEST-56 , and pDEST-59 , respectively . Each of these was recombined with pENTR-unc-47 to create pAP45 ( unc-47::unc-38 cDNA ) , pAP59 ( unc-47::unc-63 cDNA ) , pAP53 ( unc-47::unc-74 cDNA ) , pAP57 ( unc-47::unc-50 cDNA ) , and pAP55 ( unc-47::ric-3 cDNA ) . nrx-1 reporter and rescue constructs: To generate the 5 kb nrx-1L::GFP transcriptional reporter , the nrx-1L promoter was amplified from wild type genomic DNA ( −4786 bp relative to start ) and cloned into pENTR-D-TOPO to generate pENTR-5’-nrx-1L . pENTR-5’-nrx-1L was then recombined with pDEST-93 ( GFP ) to generate pAP156 ( nrx-1L::GFP ) . The 2 kb nrx-1L promoter::GFP fusion construct ( −2033 bp relative to start ) was created using the same strategy , recombining pENTR-5’-nrx1L2kb with pDEST-93 to create pAP118 ( nrx-1L2kb::GFP ) . nrx-1LCOEΔ::GFP ( pAP178 ) was generated by PCR amplification using mutant primers that disrupt the COE motif ( TCCCAAAGGG >TAAAAAAGGG ) . Rescuing NRX-1L minigene constructs were generated by ligation of a 10 , 598 bp NheI fragment of the nrx-1 genomic locus extending from the nrx-1L start to exon 21 ( amplified from cosmid C29A12 ) with a 715 bp fragment of the nrx-1 cDNA ( isoform A ) amplified from a plasmid containing the nrx-1 coding sequence ( GM470 , provided by Kang Shen [Maro et al . , 2015] ) , and converted into the destination vector pDEST-143 . For cell-specific rescue , pDEST-143 was recombined with pENTR-3’-unc17β , pENTR-3’-unc47 , pENTR-3’-myo3 , and pENTR-3’-unc3 to create minigene plasmids pAP204 ( unc-17β::nrx-1L ) , pAP206 ( unc-47::nrx-1L ) , pAP208 ( myo-3::nrx-1L ) , and pAP202 ( unc-3::nrx-1L ) , respectively . A nrx-1L gene fragment ( Integrated DNA Technologies ) lacking sequence encoding the PDZ binding domain was synthesized , digested and ligated into pAP204 ( unc-17β::nrx-1L ) to generate the pCL83 rescuing construct ( unc-17β::nrx-1LΔPDZ ) . unc-129::NRX-1L::GFP was generated by conversion of the NRX-1L::GFP plasmid GM477 ( provided by Kang Shen [Maro et al . , 2015] ) into the destination vector pDEST-99 and recombination with pENTR-3’-unc129 to create pAP120 ( unc-129::NRX-1L::GFP ) . unc-129::NRX-1LΔPDZ::GFP was generated by amplifying nrx-1L ( from start to PDZ binding domain ) and GFP from pAP120 and ligating into a construct containing unc-129 promoter to generate pAP199 ( unc-129::NRX-1LΔPDZ::GFP ) . Chrimson and GCaMP constructs: To generate acr-2::Chrimson , Chrimson coding sequence was amplified from odr-7::Chrimson ( construct provided by Dirk Albrecht [Larsch et al . , 2015] ) and ligated into a destination vector to create pDEST-104 . pDEST-104 was recombined with pENTR-5’-acr2 to create pRB2 ( acr-2::Chrimson ) . To generate ttr-39::GCaMP6s::SL2::mCherry , GCaMP6s was amplified from pGP-CMV-GCaMP6s ( Addgene ) and ligated into a destination vector to create pDEST-95 . pDEST-95 was recombined with pENTR-5’-ttr39 ( promoter provided by David Miller [Petersen et al . , 2011] ) to create pAP130 ( ttr-39::GCaMP6s::SL2::mCherry ) . For all imaging , nematodes were immobilized with sodium azide ( 0 . 3 M ) on a 2 or 5% agarose pad . Each n represents analysis of the nerve cord from an independent animal . Images were obtained using either a 3i ( Intelligent Imaging Innovations ) Everest spinning-disk confocal microscope or Olympus BX51WI spinning disk confocal equipped with a 63x objective . All DD1 confocal images were obtained by imaging L4 hermaphrodites of similar size in the region near the pharynx using identical image and laser settings for each marker , and receptor clusters were quantified in a region from the DD1 cell body to connecting commissure ( i . e . the synaptic region ) . Analysis of synapse number/fluorescence intensity was conducted using either Volocity 6 . 3 or ImageJ software ( open source ) using defined intensity threshold values acquired from control experiments for each fluorescent marker . Specifically , the ‘find objects’ function in Volocity was used , excluding objects > 10 μm2 and <0 . 2 μm2 . Alternatively , the ‘analyze particles’ function of ImageJ was used . For ImageJ analyses , background fluorescence was first subtracted by calculating the average intensity of each image in a region devoid of puncta . In some cases , fluorescence intensity within a region of interest was also measured and normalized to wild type control as indicated . Confocal montages of the nerve cord were assembled by using the ‘straighten to line’ function in ImageJ . Only images where the DD1 neuron was clearly distinguishable from neighboring cells were included in analyses . For measurements of dendritic morphology , post-synaptic protrusions in the ventral dendritic region anterior to the DD1 soma ( i . e . the synaptic region ) were quantified in L4 hermaphrodites . For posterior DD neurons , post-synaptic protrusions were quantified anterior to the DD3 soma ( 25 μm region ) . All protrusions ≥ 0 . 3 μm were analyzed , measuring from the base of the main dendritic process to the tip of the protrusion . For imaging and quantification of flp-13::ACR-12::GFP in Figure 3A and Figure 2—figure supplement 1B , strains IZ1458 ( ufIs126 ) and IZ1557 ( ufIs126; acr-12 ( ok367 ) ) were included in the analysis as wild type control . There was no appreciable difference in the number of receptor clusters between the two strains ( ufIs126; acr-12 ( ok367 ) 14 . 4 ± 0 . 6 , n = 49; ufIs126 14 . 8 ± 0 . 6 , n = 48 ) . ACR-12 receptor clusters positioned within 0 . 5 μm of the dendritic shaft were scored as dendritic , while receptors outside this region were scored as associated with spiny protrusions . Spine and receptor cluster number in the DD1 synaptic region were analyzed in synchronized animals using strains IZ1458 and IZ1464 as described previously ( He et al . , 2015 ) . Briefly , embryos for each strain were picked to separate 60 mm unseeded plates and allowed to hatch for 40 min . Newly hatched L1 larvae were moved to freshly seeded plates , and the midpoint of the 40 min in which the embryos hatched was considered t = 0 . Plates were incubated at 25°C for 28 , 34 , 46 , and 52 hr . Developing protrusions ≥ 0 . 2 μm in the synaptic region were analyzed , and ACR-12 receptor clusters were quantified as above . Custom Stellaris FISH probes against nrx-1 mRNA were obtained from Biosearch Technologies as a mix of 48 probes conjugated to CAL Fluor Red 590 Dye . Experiments were performed using wild type , unc-3 ( e151 ) mutants , and nrx-1 ( nu485 ) mutants expressing either unc-47::GFP ( oxIs12 ) , unc-4::GFP ( wdIs5 ) , or unc-17::GFP ( vsIs48 ) markers to label populations of GABA and ACh motor neurons . Synchronized populations of L3-L4 larval animals were fixed and hybridized as described previously ( Ji and van Oudenaarden , 2012; Raj et al . , 2008 ) . Images were obtained using spinning disk microscopy as above . Z-projections were analyzed in ImageJ using the ‘analyze particles’ function . Following background subtraction , the total number of nrx-1 mRNA molecules was calculated for a 45 x 5 . 5 μm straightened region of the anterior ventral nerve cord using a defined intensity threshold across all images . For staining of ACR-12 receptors at the cell surface , mouse monoclonal α-HA antibodies ( 16B12 ) coupled to Alexa594 were diluted in injection buffer ( 20 mM K3PO4 , 3 mM K citrate , 2% PEG 6000 , pH 7 . 5 ) . Antibody was injected into the pseudocoelom of early L4 stage wild type or nrx-1 ( wy778 ) animals as described previously ( Gottschalk and Schafer , 2006 ) . Animals were allowed to recover for six hours on seeded NGM plates . Only animals in which fluorescence was observed in coelomocytes ( indicating uptake of excess antibody and successful injection ) were included in the analysis . Injections of anti-GFP Alexa594 antibody followed the same protocol . One-day old adults were placed on thinly seeded NGM plates and tracked for a period of 5 min using Single Worm Tracker 2 . 0 ( WT2 ) ( Yemini et al . , 2011 ) . Worm tracker software version 2 . 0 . 3 . 1 , created by Eviatar Yemini and Tadas Jucikas ( Schafer lab , MRC , Cambridge , UK ) , was used to analyze movement . Transgenic animals expressing ttr-39::GCaMP6s::SL2::mCherry ( GABA neurons ) or myo-3::NLSwCherry::SL2::GCaMP6s ( muscle , from M . Alkema ) along with acr-2::Chrimson ( cholinergic neurons ) were placed on plates seeded with OP50 containing 2 . 75 mM All-Trans Retinal ( ATR ) for 24 hr prior to experiments . Young adults were immobilized on 5% agarose pads in 2 , 3-Butanedione monoxime ( BDM ) ( 30 mg/ml ) . For all genotypes , control animals grown in the absence of ATR were imaged . Imaging was carried out on a Yokogawa CSU-X1-A1N spinning disk confocal system equipped with EM-CCD camera ( Hammamatsu , C9100-50 ) and 40X C-Apochromat 1 . 2 NA water immersion objective . Optogenetic stimulation experiments employed a 625 nm ( 40 W ) LED ( Mightex Systems ) . Optical output through the objective was 0 . 3 mW/mm2 at the focal plane of the specimen . Simultaneous GCaMP excitation ( 488 nm ) and emission ( 525 nm ) acquisition and Chrimson activation were achieved using a 556 nm edge BrightLine single-edge short-pass dichroic beam splitter positioned in the light path ( Semrock ) . Data were acquired using Volocity software . Images were binned at 4 × 4 during acquisition and sampled at 10 Hz . GABA motor neuron and muscle ROIs in respective experiments were identified by mCherry fluorescence . Recordings from motor neuron cell bodies were obtained systematically , beginning at the anterior end of the ventral nerve cord and moving in a posterior direction . Each field typically contained 1–5 GABA motor neurons . Only recordings of neurons located anterior to the vulva were included in the analysis . Muscle recordings were obtained either directly anterior or posterior to the vulva . Photobleaching correction was carried out by fitting an exponential function to the data ( CorrectBleach plugin , ImageJ ) . A linear fit ( Igor Pro , Wavemetrics ) of the background fluorescence was subtracted from the cell body fluorescence across all time points . Pre-stimulus baseline fluorescence ( F0 ) was calculated as the average of the corrected background-subtracted data points in the first 4 s of the recording and the corrected fluorescence data was normalized to prestimulus baseline as ∆F/F0 , where ∆F = F – F0 . Peak ∆F/F0 was determined by fitting a Gaussian function to the ∆F/F0 time sequence using Multi peak 2 . 0 ( Igor Pro , WaveMetrics ) . All data collected were analyzed , including failures ( no response to stimulation ) . Peak ∆F/F0 values were calculated from recordings of >10 animals per genotype . Mean peaks ± SEM were calculated from all peak ∆F/F0 data values and normalized to the wild type mean . Latency was calculated as the time required from stimulus onset for fluorescence ( ∆F/F0 ) to reach two times the pre-stimulus baseline standard deviation ( Larsch et al . , 2015 ) . Duration was measured as the time between the onset of the transient and the completion of the decay back to the baseline . Rise and decay time constants were determined from the time constant of exponential fits between the baseline and peak fluorescence as appropriate . | Nervous systems are complex networks of interconnected cells called neurons . These networks vary in size from a few hundred cells in worms , to tens of billions in the human brain . Within these networks , each individual neuron forms connections – called synapses – with many others . But these partner neurons are not necessarily alike . In fact , they may be different cell types . How neurons form distinct connections with different partner cells remains unclear . Part of the answer may lie in specialized proteins called cell adhesion molecules . These proteins occur on the cell surface and enable neurons to recognize one another . This helps ensure that the cells form appropriate connections via synapses . Cell adhesion molecules are therefore also known as synaptic organizers . Philbrook et al . have now examined the role of synaptic organizers in wiring up the nervous system of the nematode worm and model organism Caenorhabditis elegans . Motor neurons form connections with two types of partner cell: muscle cells and neurons . Philbrook et al . screened C . elegans that have mutations in genes encoding various synaptic organizers . This revealed that a protein called neurexin must be present for motor neurons to form synapses with other neurons . By contrast , neurexin is not required for the same neurons to establish synapses with muscles . Philbrook et al . found that neuron-to-neuron synapses arise at specialized finger-like projections . These resemble the dendritic spines at which synapses form in the brains of mammals , and had not been previously identified in C . elegans . In worms that lack neurexin , these spine-like structures do not form correctly , disrupting the formation of neuron-to-neuron connections . Previous work has implicated neurexin in synapse formation in the mammalian brain . But this is the first study to reveal a role for neurexin in establishing partner-specific synaptic connections . Mutations in synaptic organizers , including neurexin , contribute to disorders of brain development . These include schizophrenia and autism spectrum disorders . Learning more about how neurexin helps establish specific synaptic connections may help us understand how these disorders arise . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Neurexin directs partner-specific synaptic connectivity in C. elegans |
Inflammatory bowel disease ( IBD ) is driven by dysfunction between host genetics , the microbiota , and immune system . Knowledge gaps remain regarding how IBD genetic risk loci drive gut microbiota changes . The Crohn’s disease risk allele ATG16L1 T300A results in abnormal Paneth cells due to decreased selective autophagy , increased cytokine release , and decreased intracellular bacterial clearance . To unravel the effects of ATG16L1 T300A on the microbiota and immune system , we employed a gnotobiotic model using human fecal transfers into ATG16L1 T300A knock-in mice . We observed increases in Bacteroides ovatus and Th1 and Th17 cells in ATG16L1 T300A mice . Association of altered Schaedler flora mice with B . ovatus specifically increased Th17 cells selectively in ATG16L1 T300A knock-in mice . Changes occur before disease onset , suggesting that ATG16L1 T300A contributes to dysbiosis and immune infiltration prior to disease symptoms . Our work provides insight for future studies on IBD subtypes , IBD patient treatment and diagnostics .
Crohn’s disease ( CD ) and ulcerative colitis ( UC ) , the two main forms of inflammatory bowel disease ( IBD ) , are characterized by chronic relapsing inflammation of the gastrointestinal tract ( Podolsky , 2002; Turpin et al . , 2018 ) . The etiology of IBD is complex , as host genetics , the gut microbiota and environmental exposures all contribute to disease pathogenesis ( Xavier and Podolsky , 2007; Garrett et al . , 2010a ) . A breakdown in the ability of a genetically susceptible host to respond appropriately to the gut microbiota may lead to an overactive local immune response ( Sartor , 2008; Eckburg and Relman , 2007 ) initiating the chronic cycle of intestinal inflammation core to IBD . Many genes within IBD loci are directly involved in pathways controlling the sensing and innate responses to bacteria ( Xavier and Podolsky , 2007; Jostins et al . , 2012 ) . The relatively longstanding observation that there is an absence of intestinal inflammation in several gnotobiotic mouse models of spontaneous colitis maintained under germ-free housing conditions supports this idea ( Elson et al . , 2005; Sellon et al . , 1998 ) . Furthermore , data from IBD patients demonstrating that diversion of the fecal stream greatly improves symptoms ( Rutgeerts et al . , 1991; McIlrath , 1971 ) as well as reduces inflammatory cytokine levels ( Daferera et al . , 2015 ) also lends plausibility to this concept . Dysbiosis of the gut microbiota , including alterations in frequency , diversity and richness of microbial populations ( Manichanh et al . , 2006; Ott et al . , 2004 ) , has been associated with IBD ( Morgan et al . , 2012; Frank et al . , 2007; Willing et al . , 2009 ) . For example , a reduction in the abundance of the phylum Firmicutes , including the genus Clostridium ( Rajilić-Stojanović et al . , 2013 ) as well as Proteobacteria and Actinobacteria , has been associated with IBD ( Frank et al . , 2007 ) . In contrast , there is variation in the abundance of Bacteroides in IBD ( Swidsinski et al . , 2002; Swidsinski et al . , 2005; Gophna et al . , 2006; Rehman et al . , 2010; Chu et al . , 2016 ) , including an increase in Bacteroides fragilis ( Swidsinski et al . , 2005; Gophna et al . , 2006 ) . Additionally , certain microbes , including Ruminococcus gnavus , are able to thrive in the high levels of oxidative stress in the inflamed gut and are associated with increased disease activity ( Hall et al . , 2017a ) . A population-based metagenomics analysis found associations between fecal levels of secreted proteins and microbiome composition ( Zhernakova et al . , 2016 ) . Additionally , the unique ileal transcriptome of CD patients is associated with alterations in Firmicutes ( Haberman et al . , 2014 ) , implicating a role for genetics in shaping the microbiota . Genome wide association studies have shown that host genetic variants play a role in microbial dysbiosis ( Hall et al . , 2017b ) . Further supporting the idea that these microbial alterations could be the driving force behind the immune response , microbial changes have been observed prior to onset of disease symptoms and specific microbiota signatures are associated with disease location ( Imhann et al . , 2018 ) . While data suggest a dynamic interplay between the microbiota , the immune system and disease pathophysiology ( Palm et al . , 2015 ) , how dysbiosis manifests and more specifically how individual species are affected by host genetics is less well understood . Because there is substantial diversity in disease location , symptoms and genetic susceptibility in IBD , understanding how specific microbial alterations occur will be important for identifying disease subtypes as well as developing precision medicine treatment options . Genetic diversity among patients is a compounding factor with respect to disease phenotype , suggesting that specific risk loci may play a role in underlying microbial community makeup . Genome-wide association studies have linked over 230 genetic risk loci to increased IBD risk ( Jostins et al . , 2012; Liu et al . , 2015; de Lange et al . , 2017; Luo et al . , 2017 ) . The microbiota and disease severity have also been linked to genetic risk ( Moustafa et al . , 2018 ) . Genetics can affect the composition of the human gut microbiome ( Hall et al . , 2017b ) . Additionally , IBD is associated with a variety of risk alleles that affect immune activation in response to microbial recognition and handling ( Davies and Abreu , 2015 ) . Many studies have linked IBD to genetic risk variants in NOD2 ( Turpin et al . , 2018 ) , which is involved in recognition of bacterial muramyl dipeptide , but one study has linked the presence of multiple NOD2 variants to increased levels of Enterobacteriaceae ( Knights et al . , 2014 ) . A single nucleotide polymorphism ( SNP ) in ATG16L1 ( re2241880 , Thr300Ala ) is associated with an increased risk of CD ( Hampe et al . , 2007; Rioux et al . , 2007 ) . ATG16L1 functions in autophagy , a cellular recycling system that aids in the sequestration of intracellular bacteria ( Cooney et al . , 2010; Travassos et al . , 2010 ) . ATG16L1 hypomorphic mice have defects in anti-microbial peptide secretion by Paneth cells ( Cadwell et al . , 2008 ) and ATG16L1-deficient macrophages show increased secretion of pro-inflammatory cytokines ( Saitoh et al . , 2008 ) . Conditional deletion of ATG16L1 specifically in epithelial cells reduces autophagy and renders mice more susceptible to Salmonella enterica serovar Typhimurium infection and systemic translocation ( Conway et al . , 2013 ) . The Thr300Ala ( T300A ) variant displays enhanced degradation via active caspase 3 resulting in a reduction in the levels of ATG16L1 protein and reduced levels of autophagic flux ( Murthy et al . , 2014; Lassen et al . , 2014 ) . Moreover , microbial signatures are associated with variants in the gene for NOD2 ( Xavier and Podolsky , 2007 ) , which interacts with ATG16L1 ( Sartor , 2008 ) . T300A knock-in mice display defects in Paneth cells and goblet cells ( Lassen et al . , 2014 ) . More recently , T300A has been shown to directly affect lysozyme secretion from Paneth cells in mice in the context of infection ( Bel et al . , 2017 ) . These data suggest that T300A may play a role in shaping the response to gut microbes . Specifically , this may occur by loss of autophagy due to enhanced degradation of ATG16L1 T300A and altered anti-microbial peptide secretion . These effects in turn may affect microbial composition . While many of these studies have observed differences in mice in the context of infection , the effect of T300A on microbial composition has not been addressed . Furthermore , while these data suggest that alterations in the makeup of gut microbial communities may arise from the presence of certain risk alleles and may shape the downstream immune response; this has yet to be formally demonstrated . A dysfunctional response to the microbiota can lead to adaptive immune cell infiltration in the gut during colitis ( Feng et al . , 2010; Lodes et al . , 2004 ) and is a key contributor to disease pathology . Initially , studies suggested a role for Th1 cells in CD ( Parronchi et al . , 1997 ) whereas Th2 cells were viewed as responsible for inflammation in UC ( Inoue et al . , 1999 ) . However , a more nuanced understanding of the role of T cells in IBD has lead to the implication of Th17 cells in CD pathogenesis as well . Th17 cells are characterized by the expression of the transcription factor RORγt+ and produce high levels of the cytokines IL-17 and IL-22 . High levels of these cytokines have been found in the gut mucosa of CD patients as compared to healthy controls ( Fujino et al . , 2003; Andoh et al . , 2005 ) . More recently , IFN-γ , a Th1 cell cytokine , IL-17 and IL-22 have been shown to be increased in the affected ileum compared to the uninfected ileum of CD patients ( Li et al . , 2017 ) . Because it is still unclear whether dysbiosis precedes inflammation , the initiation of an adaptive response to the gut microbiota in IBD still needs to be fully elucidated . Herein , we have utilized 16S rRNA gene amplicon sequencing , shotgun metagenomic sequencing , and analysis of immune cell populations in the gut to demonstrate that the risk allele T300A affects gut microbial composition in mice , both during inflammation under conventional housing conditions , and also in gnotobiotically-housed mice that harbor microbial communities derived from stool specimens from IBD patients , or in mice associated with Bacteroides ovatus with a limited , defined microbial community . Furthermore , we have shown that these differences can affect the immune response in the gut lamina propria ( LP ) prior to the manifestation of disease .
Previous studies have shown that IBD risk alleles can affect microbial composition ( Knights et al . , 2014 ) . Because ATG16L1 is involved in the handling of intracellular bacteria , we asked whether the IBD risk allele T300A could affect the gut microbiota in conventionally-housed SPF wild type ( WT ) vs . T300A knockin mice ( Lassen et al . , 2014 ) at steady state . We conducted 16S rRNA gene amplicon surveys on stool samples from WT and T300A mice ( Figure 1 ) . We observed overall alterations in the microbiota of T300A mice compared to WT mice ( Figure 1a ) . WT and T300A mice showed distinct separation by t-distributed stochastic neighbor embedding ( t-SNE ) analysis ( Figure 1b ) and we saw an increase in the order Bacteroidales ( Figure 1c ) . Overall , these data suggest that T300A mice have alterations in the gut microbiota at steady state and that the abundance of the order Bacteroidales may be influenced by the presence of the T300A IBD risk allele in mice . Given that the alterations observed in the human gut IBD microbiota are often studied in the context of gut inflammation and injury , we sought to determine whether the T300A allele would influence the gut microbiota in mice during a state of large intestinal injury and inflammation modeled using a chemical-based perturbation , as T300A mice do not spontaneously develop intestinal inflammation . We treated WT and T300A mice with 2 . 5% dextran sulfate sodium ( DSS ) in the drinking water for 7 days followed by 7 days of regular drinking water . We conducted 16S rRNA gene amplicon surveys on stool specimens after this two-week intervention from WT and T300A mice ( Figure 2 ) . We observed a decrease in the phylum Firmicutes and an increase in the phyla Bacteroidetes , Proteobacteria , and Cyanobacteria in T300A mice as compared to WT mice ( Figure 2a ) . The microbiota of WT mice separated from cage-matched T300A mice by t-distributed stochastic neighbor embedding ( t-SNE ) analysis ( Figure 2b ) . These data support that T300A plays a role in shaping the gut microbiota in the context of intestinal injury and recovery . When we looked specifically at the phyla Bacteroidetes , there was an increased abundance in T300A mice compared to WT mice ( Figure 2c ) suggesting that the T300A risk allele may specifically enhance Bacteroides spp . In particular , B . ovatus was increased significantly in T300A mice ( Figure 2d ) . We analyzed the data using PICRUSt in order to infer metabolic function and found increased glycosaminoglycan degradation in samples from T300A mice ( Figure 2—figure supplement 1 ) , a process which is uniquely attributed to Bacteroidetes ( Koropatkin et al . , 2012 ) . The degradation of glycosaminoglycans by Bacteroidetes has the potential to affect gut barrier function by reducing the thickness and integrity of the mucus layer and in turn contributes to bacterial translocation and host immune system activation . This observation has implications for the mechanism by which increased abundance of Bacteroidetes can contribute to disease pathogenesis in IBD . Additionally , we also observed increased weight loss in T300A mice from day 8 through 13 following treatment with DSS compared to WT mice , suggesting that these alterations in the microbiota are associated with heightened disease susceptibility ( Figure 2e ) . Next , we sought to determine whether the differences observed in the microbial community in T300A conventionally-housed mice during inflammation were specific to the mouse microbiome or whether the host T300A genotype could also influence the composition of donor human stool . We obtained a limited number of human stool samples from patients with a genotype of WT or T300A and with active or inactive disease status to associate germ-free ( GF ) mice . It should be noted that the T300A allele is a relatively common mutation in healthy individuals ( Hampe et al . , 2007 ) , thus our healthy control is heterozygous for T300A based on the availability of stool samples . First , to verify that patient stool samples would maintain a composition based on human donor disease status within the mouse gut , we orally inoculated WT and/or T300A gnotobiotically-housed mice with stool samples from two CD patients in remission ( donor genotype T300A or WT , termed ‘inactive CD’ ) , a CD patient with active inflammation ( genotype T300A , termed ‘active CD’ ) , an ulcerative colitis ( UC ) patient with active inflammation ( genotype T300A , termed ‘active UC’ ) , or a healthy volunteer ( genotype heterozygous T300A ( AG ) , termed ‘healthy control’ ) ( Figure 3a ) . We compared the phylogenetic compositions from 16S rRNA gene amplicon sequencing data from the original human donor stool samples with phylogenetic composition analyses from our shotgun metagenomic data using MetaPhlAn from stool samples from GF mice associated with the indicated human stool samples . We observed that samples from these mice maintained diversity based on patient disease status by PCA analysis ( Figure 3a ) . Individual mice associated with the same patient stool clustered together with the original samples from human donor samples used for the GF associations . This observation suggested that the differences in the human microbiome from patient samples were maintained within the mouse gut . To determine whether the presence of T300A could alter the gut microbiota , we orally inoculated human stool from individual IBD patients with active disease ( one with CD or one with UC ) into gnotobiotically-housed WT or T300A mice . Four weeks after oral inoculation , we assessed the stool microbial composition by metagenomic sequencing and MetaPhlAn followed by PCA analysis ( Figure 3b ) . We observed a difference in WT vs . T300A mice by PCA in mice that received stool from a CD patient with active disease but not in mice inoculated with stool from a UC patient with active disease ( Figure 3b ) . These data suggest that differences in the active CD microbiota in mice are due to the presence of the risk allele T300A as opposed to microbial population drift due to human stool transfer into mice alone . When we examined the microbial composition of WT vs . T300A mice associated with stool from IBD patients with active disease , we observed an increase in the genus Bacteroides in T300A mice that received stool from patients with active CD as compared to WT mice ( Figure 4a ) . B . ovatus was also increased in T300A mice that received stool from patients with active CD ( Figure 4b ) , further supporting the idea that T300A has the potential to drive increases in Bacteroides spp . We did observe high levels of Bacteroides in mice that received stool from patients with active UC but no significant difference between WT and T300A mice ( Figure 4b ) . We did not see major significant changes to the microbiota in WT vs . T300A mice associated with active UC stool . These data suggest that the human CD microbiota may be more susceptible to the presence of T300A , driving larger differences in microbial populations compared to mice associated with stool from patients with active UC . Previous reports on the abundance of Bacteroidetes show conflicting results in humans ( Hold et al . , 2014 ) , with some studies reporting an increased abundance while others report a decreased abundance in patients with IBD . Therefore , we investigated whether the increase in Bacteroides in the microbiota of T300A mice was also seen in IBD patients with the risk allele T300A ( Figure 5 ) . We utilized patient genotype data and stool microbiome data available from a previous study ( Jostins et al . , 2012 ) . When we compared the microbiota from IBD patients with WT ( AA ) genotype versus heterozygous patients ( AG ) versus T300A ( GG ) genotype , we saw a trend toward an increase in the genus Bacteroides ( Figure 5a ) . We also observed a trend toward an increase in the B . fragilis group ( Figure 5b ) and a significant increase in the species Bacteroides caccae ( Figure 5c ) . In contrast , we observed a decrease in Clostridia and an increase in Gammaproteobacteria ( Figure 5—figure supplement 1 ) both of which have been observed in numerous human IBD studies ( Frank et al . , 2007; Gophna et al . , 2006; Rehman et al . , 2010 ) and pre-clinical colitis models ( Rooks et al . , 2014; Garrett et al . , 2010b ) . These data show that the presence of T300A has the potential to reconfigure the human microbiota and that the presence of a heterozygous genotype displays an intermediate phenotype with Bacteroides spp , suggesting that even a mild alteration in ATG16L1 can have an effect on bacterial abundance . This association of the risk allele T300A and alterations in the microbiota have not previously been described . This link between genotype and gut microbial phenotype in humans is important for furthering our understanding of the basis of IBD pathogenesis . Whether or not genotype and the microbiota are also associated with differences in immune populations in humans , prior to IBD disease onset or recrudescence is yet to be determined . We sought to discern whether the alterations we observed in the gut microbiota of T300A mice were also associated with changes in immune cell populations in the gut of gnotobiotic WT versus T300A mice associated with patient stool . Crohn’s disease has been linked to an increase in both Th17 cells and Th1 cells in the gut ( Parronchi et al . , 1997; Fujino et al . , 2003; Andoh et al . , 2005; Li et al . , 2017 ) . Th17 and Th1 cells are characterized by the expression of the transcription factors RORγt and T-bet respectively . Therefore , we analyzed RORγt+ ( Th17 ) and T-bet+ ( Th1 ) T cells in addition to Foxp3+ regulatory T cells ( Treg ) , as well as GATA-3+ ( Th2 ) cells by flow cytometry from mice associated with stool from patients with active CD or active UC ( Figure 6 ) . T cells were gated as the frequency of CD3+CD4+ , live , CD45+ , lymphocytes ( Figure 6—figure supplement 1 ) . Myeloid cells can also play a role in inflammation during IBD , and loss of ATG16L1 in myeloid and dendritic cells leads to enhanced colitis and increased production of inflammatory cytokines ( Zhang et al . , 2017 ) . Thus , we analyzed the following myeloid cell populations: CD11c+MHCII+ conventional dendritic cells ( DCs ) , CD11b-CD103+ tolerogenic DCs , CD11b+GR-1int neutrophils , and CD11b+GR-1hi inflammatory monocytes by flow cytometry ( Figure 6 ) . Conventional DCs , neutrophils and inflammatory monocytes were gated out of live , CD45+ cells and tolerogenic DCs were gated out of conventional DCs ( Figure 6—figure supplement 2 ) . We examined the frequency of T cell populations and myeloid cell populations in the lamina propria ( LP ) of the colon and ileum from WT and T300A mice associated with stool from patients with either active CD or active UC ( Figure 6 ) . We observed increased frequencies of Th17 and Th1 T cells exclusively in the LP of the colon and ileum from T300A mice that received stool from patients with active CD ( Figure 6a and c ) and not in the LP of the colon or ileum of mice that received stool from patients with active UC ( Figure 6b and d ) . These data support an association between alterations in the gut microbiota driven by the risk allele T300A and increased frequencies of Th17 and Th1 T cells in the gut LP . We also found an increase in the frequency of Tregs in the LP of the colon from T300A mice associated with stool from patients with CD ( Figure 6a ) and a minor increase in the ileum in T300A mice associated with stool from patients with UC ( Figure 6d ) . The only differences we observed in myeloid cells occurred in the LP of the colon from mice associated with stool from patients with active CD ( Figure 6a ) . We saw a reduced frequency of conventional DCs and a slight increase in the frequency of tolerogenic DCs along with a decrease in neutrophils and inflammatory monocytes in T300A mice ( Figure 6a ) . Why some of these populations are reduced is yet to be determined . Overall these data suggest that the presence of the T300A allele in mice can enhance Th17 and Th1 cells in the LP of the colon and ileum and alter the frequency of DCs , neutrophils , and inflammatory monocytes in the presence of stool from patients with active CD . It should be noted that histology-based grading of the colon and small intestine from mice associated with human IBD stool showed no signs of colitis , enteritis , or any intestinal histopathology ( Figure 6—figure supplement 3 ) . This suggests that an alteration in the microbiota composition and alterations in gut immune cells occur before the onset of disease as assessed by histology . Whether specific microbes are responsible for differences in individual subsets of immune cells in the absence of disease is not well understood . Previous studies support that Bacteroides spp . can drive a Th17 response in the colon ( Tan et al . , 2016 ) , suggesting that the increase we observe in Th17 cells in the LP of the colon from T300A mice could be driven by the increase in Bacteroides in these mice ( see Figure 4 ) . Therefore , we assessed whether the presence of B . ovatus , a species that we observed increased in abundance in T300A mice associated with active CD stool , would selectively alter gut immune populations in T300A mice . To test this hypothesis we associated GF and altered Schaedler flora ( ASF ) , a restricted , defined microbial community of eight bacterial members , WT and T300A mice with ~109 CFU of B . ovatus 8483 for 3 weeks . Because it has been shown that housing of mice in separate cages can affect the composition of the gut microbiota , WT and T300A mice were co-housed during the association of all studies performed with B . ovatus 8483 . Information on caging , age , and sex of mice for all studies can be found in Supplementary file 1 . We found no difference in any immune populations in WT or T300A GF mice or GF mice associated solely with B . ovatus 8483 ( Figure 7 ) . However , T300A ASF mice showed an increase only in Th17 cells in the LP of the colon and ileum when B . ovatus 8483 was present ( Figure 8b and d ) , but there was no difference in ASF mice in the absence of B . ovatus 8483 ( Figure 8a and c ) . Besides a slight decrease in Th1 cells in the ileum of T300A ASF mice in the presence of B . ovatus 8483 ( Figure 8d ) , there were no other changes in the immune populations analyzed . These data suggest that the presence of the T300A allele and additional members of the microbiota can increase the presence of Th17 cells in the gut . This increase in frequency of Th17 cells in T300A ASF mice with B . ovatus 8483 also occurred in the absence of colitis , enteritis , or any intestinal histopathology ( Figure 8—figure supplement 1 ) , or differences in the level of colonization of WT and T300A mice with B . ovatus 8483 ( Figure 8—figure supplement 2 ) . This demonstrates that the differences in Th17 cells in ASF T300A B . ovatus 8483 mice were not due to an increase in colonization of B . ovatus but rather were most likely due to the presence of the T300A allele alone . The increase in Th17 cells in the gut of T300A ASF mice associated with B . ovatus prompted us to look at expression levels of the IL-23 specific subunit , IL-23p19 , in the LP of the colon and ileum from T300A mice ( Figure 9 ) . IL-23 induces IL-17 producing cells that can induce inflammation ( Langrish et al . , 2005 ) and is essential for the production of IL-17 in T-cell mediated colitis ( Yen et al . , 2006 ) . Additionally , antibodies targeting IL-23 have been utilized as a treatment for IBD and antibodies specific for IL-23p19 show promising clinical trial results ( Moschen et al . , 2018 ) , suggesting that IL-23 plays an important role in IBD pathogenesis . We found that in T300A ASF mice , the presence of B . ovatus 8483 increased expression levels of Il23p19 in the LP of both the colon ( Figure 9a ) and ileum ( Figure 9b ) . Overall , these results demonstrate that the presence of B . ovatus 8483 in the context of the T300A allele and a minimal microbiota can enhance both IL-23 expression and Th17 cell infiltration into the gut before the onset of disease symptoms in mice .
The role of the gut microbiota in IBD pathogenesis is well established , but how and when microbial dysbiosis occurs during disease progression is less well understood . The identification of IBD genetic risk loci involved in the processing and handling of microbes provides more definitive clues into disease pathogenesis . However , we still do not fully understand how host genotype or particular SNPs influence microbial community assembly , community composition , or function . The reduced barrier integrity associated with defects in Paneth cell anti-microbial peptide secretion in T300A hypomorphic mice ( Cadwell et al . , 2009 ) and T300A knock-in mice ( Conway et al . , 2013 ) , and increased production of inflammatory cytokines by ATG16L1 deficient innate immune cells in response to infiltrating bacteria ( Saitoh et al . , 2008 ) can activate an adaptive immune response to the gut microbiota . However , whether genetic risk loci directly affect specific gut microbial population frequency is not known . Here we show that the presence of the CD risk allele ATG16L1 T300A in both mice and humans can shape the gut microbiota . More specifically , it can enhance the abundance of Bacteroides , including B . ovatus , as well as increase the local Th1 and Th17 response in the gut LP in T300A mice associated with stool from a patient with active CD but not stool from a patient with active UC . We found that minimal microbiota ASF T300A mice associated with B . ovatus 8483 , specifically increased Th17 cells , suggesting that the other differences we saw in Treg and myeloid cell populations in T300A mice associated with stool from a patient with active CD may be due to the presence of other species in the CD microbiota . While we observe increases in other microbial populations , including Proteobacteria and Cyanobacteria in conventional-housed T300A subjected to chemically-induced inflammatory conditions , the increase in Bacteroidetes was common between both DSS-treated conventionally-housed mice and gnotobiotically-reared human stool associated mice , suggesting that T300A could potentially be a driver of increased Th17 and Th1 T cell populations in the gut LP via increased Bacteroides spp . When T300A ASF mice were associated with B . ovatus 8483 , we did not observe differences in levels of colonization , suggesting that the increase in the frequency of Th17 T cells in the gut LP was independent of the level of colonization . Why we did not find differences in immune populations in GF WT and T300A mice monocolonized with B . ovatus 8483 is intriguing as it suggests that this increase in Th17 cells requires the presence of a more complex microbiota . Future work will be needed to elucidate how the immune response to B . ovatus changes in the presence of additional microbial species and in the context of distinct host genotypes associated with IBD . Specifically , what is the nature of the co-occurrence relationships between B . ovatus and other members of a microbiota necessary for Th17 T cell expansion ? Alternatively , are there a series of one-on-one host-microbe relationships in tandem or sequentially that are required for Th17 T cell expansion in the LP of the colon and ileum ? Previous studies have demonstrated the ability of individual Bacteroides spp . to induce IBD in genetically susceptible mice ( Bloom et al . , 2011 ) and others have shown that Bacteroides spp . can induce IL-17 response in the gut ( Tan et al . , 2016 ) . Exactly how T300A could enhance Bacteroides specifically is yet to be determined . We found that expression of a subunit of IL-23 , Il23p19 , a cytokine known to increase IL-17 production from T cells , was increased in both the LP of the colon and the ileum from T300A mice associated with B . ovatus 8483 . Other studies have shown that B . fragilis enterotoxin can induce formation of autophagosomes ( Ko et al . , 2017 ) . Another report demonstrated that ATG16L1 is required to generate a regulatory T cell response to B . fragilis outer membrane vesicles ( Chu et al . , 2016 ) . Because T300A is susceptible to caspase 3 cleavage and is preferentially targeted for degradation compared to WT ATG16L1 ( Turpin et al . , 2018 ) , reduced xenophagy levels could allow for an outgrowth of Bacteroides , and an insufficient regulatory T cell response necessary to dampen Th17 population expansion . ATG16L1 and Nod2 ( another CD risk allele ) have been shown to interact in an autophagy-dependent antibacterial pathway ( Homer et al . , 2010 ) , suggesting that defects in either pathway could affect Bacteroides abundance . For example , Nod2 is essential for preventing the expansion of Bacteroides vulgatus in mice ( Ramanan et al . , 2014 ) . We also saw that T300A in humans was associated with a trend in increased levels of Bacteroides including the B . fragilis group and a significant increase in B . caccae which is closely related to B . ovatus ( Sakamoto and Ohkuma , 2011 ) . An intermediate phenotype was seen in people with the heterozygous ( AG ) genotype . More research , especially in a prospective cohort of at-risk individuals , will be needed to determine whether IBD patients with T300A have an increase in B . ovatus prior to disease onset and whether this contributes to disease pathogenesis . IBD symptoms can vary in nature and severity as well as disease location . While some patients will respond to treatment , others will not , suggesting that subtypes of the disease may be dependent on both genetics and the microbiota . Healthy individuals and CD patients may harbor multiple risk alleles , compounding the effect of specific alleles on the microbiota , and further studies will be needed to address the complexity of genetic associations with gut microbial composition . However , understanding how risk alleles can shape both the microbiota and the underlying immune response , in tandem , can provide insight into identifying subsets of the disease as well as inform diagnosis and treatment decisions . We have shown that the risk allele ATG16L1 T300A , a SNP associated with an increased risk of CD , contributes to dysbiosis in mice , specifically an increase in Bacteroides among other alterations , and correlates with an enhanced Th1 and Th17 immune response in the gut LP . These changes precede the onset of disease in human stool microbiome associated mice , suggesting that microbiota changes induce inflammatory cell population shifts in the gut . These results shed light on the etiology of CD and provide insight into the relationship between SNPs , dysbiosis and the immune response in the gut .
All mouse strains employed are on the C57BL/6J background and were sacrificed between 8–13 weeks of age . Generation of the ATG16L1 T300A knock-in mice has been previously described by Haberman et al . ( 2014 ) . Conventionally-housed mice were kept at the Massachusetts General Hospital , and all procedures were performed in accordance with the Institutional Animal Care and Use Committee at Massachusetts General Hospital . ATG16L1 T300A knock-in mice were aseptically rederived as gnotobiotic animals and maintained in semi-rigid gnotobiotic isolators under a strict 12 hr light cycle in the Harvard T . H . Chan Gnotobiotic Center for Mechanistic Microbiome Studies . All gnotobiotic experiments were performed at the Harvard T . H . Chan Gnotobiotic Center for Mechanistic Microbiome Studies and were approved and carried out in accordance with Harvard Medical School's Standing Committee on Animals and the National Institutes of Health guidelines for animal use and care . Mice gavaged with different stool samples from human donors were maintained in separate isolator cages . For all mouse experiments a minimum of two experiments were conducted throughout . Conventionally-housed mice were fed 2 . 5% ( w/v ) dextran sulfate sodium ( DSS , MP Biomedicals; MW = 36 , 000–50 , 000 ) dissolved in sterile distilled drinking water ad libitum for 7 days , followed by 7 days of regular drinking water . Body weights were monitored daily . Stool samples were collected into RNAlater after the last day of regular drinking water and stored at −80°C until DNA was extracted for 16S rRNA gene amplicon sequencing . Stool samples for 16S rRNA amplicon surveys were collected into RNAlater and stored at −80°C prior to DNA extraction . The samples were then stored at −80°C until shipping to the Broad Institute for DNA extraction . For DNA extraction , a combination of the QIAamp 96 PowerFecal Qiacube HT Kit ( Qiagen Cat No . /ID: 51531 ) , the Allprep DNA/RNA 96 Kit ( Qiagen Cat No . /ID: 80311 ) , and IRS solution ( Qiagen Cat No . /ID: 26000-50-2 ) were used with a custom protocol . One mouse stool pellet per sample was transferred into individual wells of the PowerBead plate , with 0 . 1 mm glass beads ( Cat No . /ID: 27500–4-EP-BP ) on dry ice . 650 μl of 56°C heated PW1 buffer with 1:100 vol addition of 1M DTT was added directly to each sample well before lysis by bead beating on a TissueLyzer II at 20 Hz for a total of 10 min . Samples pelleted by centrifugation for 6 min at 4500 x g . Supernatants transferred to a new S block ( supplied in PowerFecal Kit ) and combined with 150 μl of IRS solution and vortexed briefly before 1 min incubation . Sealed samples centrifuged again for 6 min , 4500 x g and up to 450 μl of supernatant were transferred to new S block , combined with 600 μl of Buffer C4 ( PowerFecal Kit ) , mixed by pipetting 10x and incubated at room temperature for 1 min . Samples were transferred into AllPrep 96 DNA plate on top of clean S blocks and centrifuged for 3 min at 4500 x g . Step was repeated until all sample passed through . Allprep DNA plate was placed on top of a 2 mL waste block . 500 μl AW1 buffer was added to the DNA plate which was sealed and centrifuged for 4 min at 4500 x g . 500 μl AW2 buffer was added to the DNA plate , repeating centrifuge step . Allprep DNA plate was placed on top of elution plate . 100 μl of 70°C heated EB Buffer was added to each sample column and incubated for 5 min and centrifuged for 4 min at 4500 x g to elute . Purified DNA was stored at −20°C . Human stool samples for association with gnotobiotic mice were acquired from patients from Massachusetts General Hospital according to institutional review board #2004P001067 . Patients produced the sample at the hospital visit , and it was immediately stored at −80°C by the staff . Participants provided informed consent for the study and are part of the PRISM cohort study . Frozen human stool samples ( 50 mg/mouse ) were reconstituted in sterile PBS 0 . 05% Cysteine ( 250 mg/ml ) . 25 mg ( 100 ul ) was orally instilled followed by 12 . 5 mg ( 50 ul ) placed on the anus and 12 . 5 mg ( 50 ul ) placed on the back fur of each mouse . Stool samples and gut tissues were collected at 4 weeks post human stool association . Bacteroides ovatus 8483 ( ATCC ) was grown in liquid culture with Bacteroides basal medium as described in Pantosti et al ( Pantosti et al . , 1991 ) . GF or ASF mice were gavaged with ~109 CFU in 100 μl . Mice were sacrificed after 3 weeks . At sacrifice , stool samples were collected and weighed and CFU/g of stool was calculated after growth on Brucella agar plates with 5% sheep blood , hemin , and vitamin K1 ( Hardy Diagnostics A30 ) . 16S rRNA gene libraries targeting the V4 region of the 16S rRNA gene were prepared by first normalizing template concentrations and determining optimal cycle number by way of qPCR . Two 25 μl reactions for each sample were amplified with 0 . 5 units of Phusion with 1X High Fidelity buffer , 200 μM of each dNTP , 0 . 3 μM of 515F ( 5’-AATGATACGGCGACCACCGAGATCTACACTATGGTAATTGTGTGCCAGCMGCCGCGGTAA-3’ ) and 806rcbc0 ( 5’CAAGCAGAAGACGGCATACGAGATTCCCTTGTCTCCAGTCAGTCAGCCGGACTACHVGGGTWTCTAAT-3’ ) . 0 . 25 μl 100x SYBR were added to each reaction and samples were quantified using the formula 1 . 75^ ( deltaCt ) . To ensure minimal over-amplification , each sample was diluted to the lowest concentration sample , amplifying with this sample optimal cycle number for the library construction PCR . Four 25 μl reactions were prepared per sample with master mix conditions listed above , without SYBR . Each sample was given a unique reverse barcode primer from the Golay primer set ( Bloom et al . , 2011 and Koropatkin et al . , 2012 ) . Replicates were then pooled and cleaned via Agencourt AMPure XP-PCR purification system . Purified libraries were diluted 1:100 and quantified again via qPCR ( Two 25 μl reactions , 2x iQ SYBR SUPERMix ( Bio-Rad , REF: 1708880 with Read 1 ( 5’-TATGGTAATT GT GTGYCAGCMGCCGCGGTAA-3’ ) , Read 2 ( 5’-AGTCAGTCAG CC GGACTACNVGGGTWTCTAAT-3’ ) . Undiluted samples were normalized by way of pooling using the formula mentioned above . Pools were quantified by Qubit ( Life Technologies , Inc . ) and normalized into a final pool by Qubit concentration and number of samples . Final pools were sequenced on an Illumina MiSeq 300 using custom index 5’-ATTAGAWACCCBDGTAGTCC GG CTGACTGACT-3’ and custom Read one and Read two mentioned above . 16S data were processed using QIIME ( RRID:SCR_008249 ) , and taxonomy was assigned using the Greengenes predefined taxonomy map of reference sequence OTUs to taxonomy ( McDonald et al . , 2012 ) . The resulting OTU tables were checked for mislabeling71and contamination ( Knights et al . , 2011b ) . A median sequence depth of 24 , 583 per sample was obtained , and samples with fewer than 5000 filtered sequences were excluded from analysis . 16S rRNA gene libraries were constructed as previously described ( Kostic et al . , 2015 ) DNA samples were quantified by Quant-iT PicoGreen dsDNA Assay ( Life Technologies ) and normalized to a concentration of 50 pg/μl . Illumina sequencing libraries were prepared from 100 to 250 pg of DNA using the Nextera XT DNA Library Preparation kit ( Illumina ) according to the manufacturer’s recommended protocol , with reaction volumes scaled accordingly . Prior to sequencing , libraries were pooled by collecting equal volumes ( 200 nl ) of each library from batches of 96 samples . Insert sizes and concentrations for each pooled library were determined using an Agilent Bioanalyzer DNA 1000 kit ( Agilent Technologies ) . Libraries were sequenced on HiSeq 2 × 101 to yield ~10 million PE reads . Post-sequencing de-multiplexing and generation of BAM and Fastq files are generated using the Picard suite ( RRID:SCR_006525 ) ( https://broadinstitute . github . io/picard/command-line-overview . html ) . Metagenomic data were analyzed using MetaPhlAn ( RRID:SCR_004915 ) ( v . 2 . 2 ) ( Truong et al . , 2015 ) for taxonomic profiling and HUMAnN2 ( RRID:SCR_016280 ( http://huttenhower . sph . harvard . edu/humann2 ) for functional profiling . All sequencing data generated for these studies has been deposited at SRA #SUB4222585 Principal coordinate plots were generated using t-Distributed Stochastic Neighbor Embedding ( t-SNE ) as implemented in the R package Rtsne ( RRID:SCR_016342 ) . Bray-Curtis dissimilarity , where xsi denotes the abundance of strain s in sample i , was used as the distance measure . We followed the guidelines given by the authors ( FAQ at http://lvdmaaten . github . io/tsne/ ) and selected the free parameter , perplexity , by generating mappings with perplexity values between 5 and 50 in increments of 5 . Mappings with lowest error and best visual properties were obtained using perplexity = 50 . We set the absent proportion data to zeros and remapped the proportions ( 0 , 1 ) to the interval ( 1e-7 , 1 - 1e-7 ) . We transformed the proportions using logit transformation , after which the data fit a normal distribution . Microbial functional modules were inferred from the 16S rRNA gene amplicon sequencing using PICRUSt ( Langille et al . , 2013 ) ( Phylogenetic Investigation of Communities by Reconstruction of Unobserved States ) and functional modules were reconstructed using HUMAnN ( Abubucker et al . , 2012 ) ( RRID:SCR_014620 ) . Four weeks post association with human stool , the colon and distal 10 centimeters of the small intestine ( SI ) were removed , opened longitudinally , and rinsed in PBS ( Dulbecco’s – calcium and magnesium free ) to remove fecal contents . The epithelial layer was removed using two rounds of 5 mM EDTA ( 10 ml/colon 25 ml/SI ) rotating at 37C ( round 1 includes 1 mM DTT ) . After removing the epithelial layer , the lamina propria was washed in PBS and chopped into ~1 mM pieces . The tissue was digested in RPMI with glutamine ( Sigma ) with the following: 10% FBS , 1% penicillin/streptomycin ( Corning ) , 0 . 5 mg/ml Dispase ( STEM cell tech ) , 1 mg/ml collagenase D ( Roche ) , 50 µg/ml DNAse ( Roche ) in two rounds of 30 min rotating at 37C . Isolated single cells were filtered over 40 µM filter and resuspended in 5 mM EDTA followed by a wash with FACS buffer ( 2% FBS 1 mM EDTA ) . Cells were counted on the hemocytometer followed by flow cytometry analysis . 1 × 106 isolated colon lamina propria or small intestine lamina propria cells were stained with LIVE/DEADTM fixable yellow dead cell stain kit ( Thermo Fisher Scientific , L34959 ) for 15 min at room temperature . Cells were washed with FACS buffer and then stained with FC block ( RRID:AB_2103871 , Biolegend , 101310 ) for 10 min on ice followed by mouse extracellular fluorochrome-conjugated antibodies against mouse: CD45 ( RRID:AB_312979 and RRID:AB_493535 ) , CD3 ( RRID:AB_2028475 ) , CD4 ( RRID:AB_312981 ) , CD11c ( RRID:AB_313775 ) , CD11b ( RRID:AB_893232 ) , MHCII ( RRID:AB_2069376 ) , GR-1 ( RRID:AB_313377 ) , and CD103 ( RRID:AB_465799 ) . Cells were permeabilized and fixed using the Foxp3 Fix/Perm kit ( Biolegend , 421403 ) and stained at room temperature for 45 min with the following intracellular antibodies against mouse: RORγt ( RRID:AB_2573254 ) , T-bet ( RRID:AB_1595488 ) , Foxp3 ( RRID:AB_492981 ) , or GATA-3 ( RRID:AB_1645303 ) . Flow cytometry was conducted using a BD LSR II . Gnotobiotically-housed mice associated with stool from human donors were sacrificed and colons were removed , opened longitudinally , placed in cassettes and fixed in 4% paraformaldehyde ( Sigma-Aldrich ) and embedded in paraffin . Sections ( 5 μM ) were H and E-stained and evaluated in a blinded fashion for epithelial hyperplasia ( 0–3 ) , epithelial injury ( 0–3 ) , polymorphonuclear infiltration ( 0–3 ) and mononuclear infiltration ( 0–3 ) . Statistical analysis for flow cytometry data was performed in Prism 7 . 0b ( RRID:SCR_002798 ) . Data are plotted as mean ± SEM and either Mann-Whitney U test or two-way ANOVA as described in figure legend . p < 0 . 05 was considered significant . ns = not significant . To discover association of bacterial abundance and the mouse genotype , we use limma ( RRID:SCR_010943 ) , an R package for linear modeling that powers differential expression analyses ( Ritchie et al . , 2015 ) . We adjusted for age , cage , and sex in the analysis . Corrected P-values were generated using Benjamani-Hochberg false discovery rate . Expression of Il23p19 was analyzed from total cells isolated from the colon and ileum lamina propria . Epithelial cells were removed as described in the lamina propria immune cell isolation methods described above . Lamina propria cells were resuspended in RLT buffer and frozen at −80C until further analysis . RNA was extracted using the RNeasy Mini kit ( QIAGEN ) and cDNA was generated using iScript cDNA synthesis Kit ( Bio-Rad , 1708891 ) . Quantitative PCR was conducted using the following primers for Il23p19: forward primer: AGCGGGACATATGAATCTACTAAGAGA , reverse primer: GTCCTAGTAGGGAGGTGTGAAGTTG . | Trillions of bacteria live inside the human gut . From helping to digest our food to producing vitamins , these bacteria can have a big impact on our health , yet some people tolerate these bacteria better than others . In some cases , the body reacts badly to its own bacteria , stimulating an over-exuberant immune response . The gut becomes too inflamed , causing pain and diarrhoea , which could lead to an inflammatory bowel disease such as Crohn’s disease or ulcerative colitis . The symptoms of inflammatory bowel disease can vary from person to person , and how someone responds to treatment can be as individual as the symptoms as well . The causes of inflammatory bowel disease are complex; our genes , immune system and gut bacteria all play a role . Previous research has found hundreds of mutations in our genes that increase a person’s risk of developing inflammatory bowel disease . No single mutation is the root cause for every one person with inflammatory bowel disease , and individuals with these mutations may not even develop the condition . For those who do develop the disease , certain immune cells can be found in high numbers , such as the white blood cell known as Th17 cells . People with inflammatory bowel disease may also house different bacteria compared to someone with a healthy gut . In some cases of inflammatory bowel disease , there are elevated amounts of one type of bacteria called Bacteriodes . It is not clear how these mutated genes , the types of bacteria that live inside our gut , and the immune response are all related . Lavoie et al . focused on a mutated gene , known as ATG16L1T300A , which increases risk of Crohn’s disease in humans , and the experiments compared mice that had this human mutated gene with those that did not . The mice started off germ-free , meaning that they did not have any gut bacteria . Lavoie et al . then exposed the mice to samples of human stools , which contain gut bacteria , and after a month they analysed the guts of the mice . On average , the mutant mice had more Bacteriodes and Th17 cells in their guts than the normal mice . However , none of the mice developed inflammatory bowel disease , suggesting that changes to gut bacteria and immune cells may occur before the disease can be diagnosed . Together these findings show how just one mutated gene affects the bacteria and immune cells in the gut; but there are hundreds of other known mutations linked with inflammatory bowel disease . By unravelling the effects of more of these mutations , scientists could begin to learn more about the causes of this condition , and potentially improve its treatment options . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2019 | The Crohn’s disease polymorphism, ATG16L1 T300A, alters the gut microbiota and enhances the local Th1/Th17 response |
The X-chromosome gene regulatory process called dosage compensation ensures that males ( 1X ) and females ( 2X ) express equal levels of X-chromosome transcripts . The mechanism in Caenorhabditis elegans has been elusive due to improperly annotated transcription start sites ( TSSs ) . Here we define TSSs and the distribution of transcriptionally engaged RNA polymerase II ( Pol II ) genome-wide in wild-type and dosage-compensation-defective animals to dissect this regulatory mechanism . Our TSS-mapping strategy integrates GRO-seq , which tracks nascent transcription , with a new derivative of this method , called GRO-cap , which recovers nascent RNAs with 5′ caps prior to their removal by co-transcriptional processing . Our analyses reveal that promoter-proximal pausing is rare , unlike in other metazoans , and promoters are unexpectedly far upstream from the 5′ ends of mature mRNAs . We find that C . elegans equalizes X-chromosome expression between the sexes , to a level equivalent to autosomes , by reducing Pol II recruitment to promoters of hermaphrodite X-linked genes using a chromosome-restructuring condensin complex .
The essential , X-chromosome-wide regulatory process called dosage compensation ensures that males ( XO or XY ) and females ( XX ) , from worms to humans , express equivalent levels of X-chromosome products despite their unequal dose of X chromosomes ( Gelbart and Kuroda , 2009; Meyer , 2010; Conrad and Akhtar , 2012; Jeon et al . , 2012 ) . The failure to dosage compensate is lethal . Dosage compensation strategies differ across species , but invariably a regulatory complex is targeted to the X chromosomes of one sex to modulate transcription along the entire X . The molecular mechanisms by which these complexes regulate gene expression remain elusive . Here we analyzed X-chromosome dosage compensation in the nematode Caenorhabditis elegans to determine the step of transcription controlled by its dosage compensation complex ( DCC ) . The DCC binds to both X chromosomes of hermaphrodites to reduce transcription by half ( Meyer , 2010; Pferdehirt et al . , 2011 ) . Sequence-specific DNA binding sites recruit the DCC to X and facilitate its spreading along X ( Ercan et al . , 2009; Jans et al . , 2009; Pferdehirt et al . , 2011 ) . The DCC shares subunits with condensin ( Csankovszki et al . , 2009; Mets and Meyer , 2009 ) , a protein complex required for the compaction , resolution , and segregation of mitotic and meiotic chromosomes ( Wood et al . , 2010 ) , suggesting that DCC-dependent changes in chromosome structure facilitate transcription regulation . In principle , the DCC could control any step of transcription: recruitment of RNA polymerase II ( Pol II ) to the gene promoter , initiation of transcription , escape of Pol II from the promoter or proximal pause sites , elongation of RNA transcripts , or termination of transcription . To understand the mechanism of C . elegans dosage compensation , we first developed a procedure to map the position , density , and orientation of transcriptionally engaged Pol II genome-wide in C . elegans and then devised a strategy to identify the transcription start sites ( TSSs ) . Nascent RNA transcripts from approximately 70% of C . elegans genes undergo a rapid co-transcriptional processing event in which the 5′ end is replaced by a common 22-nucleotide leader RNA ( SL1 ) through a trans-splicing mechanism ( Blumenthal , 2012 ) . Because trans-splicing removes information about Pol II initiation from nascent RNAs , TSSs have been difficult to identify from accumulated mRNAs ( Morton and Blumenthal , 2011 ) . The paucity of annotated promoters has made transcription regulation a challenge to study in C . elegans . By comparing the quantity , location , and direction of engaged Pol II from wild-type and dosage-compensation-defective embryos relative to TSSs , we determined the step of transcription controlled by the DCC . Our work establishes a general strategy for TSS mapping in any organism and provides an invaluable TSS data set for dissecting C . elegans gene regulation . We show that C . elegans equalizes X-chromosome-wide gene expression between the sexes by reducing Pol II recruitment to the promoters of X-linked genes in XX embryos via a mechanism that utilizes a chromosome-restructuring complex . We also show that a separate regulatory mechanism functions in C . elegans to elevate the intrinsic level of transcription from the X chromosomes of both sexes , so that after dosage compensation , X chromosomes and the two sets of autosomes have equivalent expression .
To map the distribution of transcriptionally engaged Pol II genome-wide , we performed global run-on sequencing ( GRO-seq ) experiments using nuclei from three stages of wild-type animals ( embryos , starved L1 larvae , and L3 larvae ) and dosage-compensation-defective embryos . In GRO-seq reactions , engaged polymerases are allowed to transcribe ( run-on ) short distances ( 100 nucleotides ) and incorporate affinity tags into their nascent RNAs under conditions that prohibit new initiation ( Core et al . , 2008 ) . Tagged transcripts are affinity purified , amplified , sequenced , and aligned to the genome to map engaged Pol II ( Figure 1A–E ) . 10 . 7554/eLife . 00808 . 003Figure 1 . Genome-wide annotation of Caenorhabditis elegans transcription start sites . ( A ) – ( E ) Examples of newly annotated transcription start sites ( TSSs ) for protein-coding genes , non-coding RNA genes , and multigenic transcription units called operons identified using the combination of GRO-seq and GRO-cap . Red arrows demark the WormBase ( WB ) gene models . Dashed vertical lines show the WB gene starts . The GRO-seq signal is in reads per kilobase per million ( RPKM ) . For protein coding genes , the GRO-seq signal was averaged across 25 bp windows with a 25 bp step . The GRO-cap signal is in reads per million ( RPM ) . TAP+ is the signal from capped mRNAs , and TAP− is the background . For ( D ) and ( E ) , the GRO-cap signal is the TAP+ signal after subtracting the TAP− signal . ( A ) TSS for a trans-spliced gene . The TSS maps 981 bp upstream of the WB start , with a continuous intervening GRO-seq signal . ( B ) TSS for the polycistronc microRNA cluster mir-54-56 maps 158 bp upstream of the primary transcript start . ( C ) TSS for a non-trans-spliced gene . The TSS from GRO-cap and GRO-seq aligns with the WB start site . ( D ) Identification of the operon TSS shows the operon includes an additional gene , tag-175 . The TSS for the operon maps 781 bp upstream of tag-175 . ( E ) TSSs for genes in operons that also use independent promoters , including the TSS for a snoRNA gene within the intron of a gene . vha-10 mRNA is trans-spliced with an SL2 RNA , indicating processing from a polycistronic RNA and an SL1 RNA , indicating transcription from an independent promoter . ( F ) Heat maps show that TSSs vastly improve gene models . The GRO-seq signal from embryos was plotted , one gene per row , for each of 4246 genes relative to the WB start ( left ) or the new TSS ( right ) . The genes were ordered with increasing distance between the TSS and WB start . The light line moving rightward in the right panel does not represent TSSs . It reflects reduced GRO-seq signal immediately downstream of the trans-splice acceptor site that has been commonly annotated as the WB start site . ( G ) Heat maps showing the GRO-cap signal from embryos that was plotted for each of 4246 genes relative to the WB start ( left ) or the new TSS ( right ) . The genes were ordered with increasing distance between the TSS and WB start . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 00310 . 7554/eLife . 00808 . 004Figure 1—source data 1 . Aligned reads for GRO-seq , GRO-cap , and ChIP-seq experiments . The number of reads from each replicate of GRO-seq , GRO-cap , and ChIP-seq that uniquely aligned to the Caenorhabditis elegans genome are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 00410 . 7554/eLife . 00808 . 005Figure 1—source data 2 . Annotation of transcription start sites for protein-coding genes . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 00510 . 7554/eLife . 00808 . 006Figure 1—source data 3 . Annotation of transcription start sites for non-coding RNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 00610 . 7554/eLife . 00808 . 007Figure 1—figure supplement 1 . GRO-seq profiles are reproducible between replicates . The GRO-seq profiles of a select X-chromosome genomic region from two biological replicates of control RNAi embryos and their average GRO-seq profile are shown along with the unique mappability of GRO-seq data in the region . Red arrows show the location and direction of transcription for each protein-coding gene in the region , which are dnj-7 , C55B6 . 1 , ZK867 . 2 , spp-22 , syd-9 , F46H5 . 2 , and K03A1 . 2 , from left to right . Gene models are from the WormBase WS230 release . The level of GRO-seq signal is provided in RPKM ( reads per kilobase per million ) . Throughout this manuscript , the average GRO-seq signal of two biological replicates for each developmental stage or condition is used . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 00710 . 7554/eLife . 00808 . 008Figure 1—figure supplement 2 . Genome-wide GRO-seq signal is highly correlated between replicates . ( A ) , ( C ) – ( E ) Scatter plots comparing GRO-seq signal between biological replicates . ( B ) Scatter plot comparing GRO-seq signal between averaged replicates of wild-type embryos vs control RNAi embryos . Average GRO-seq signal was calculated in 500 bp windows genome-wide . Pair-wise comparisons were performed between samples using windows with at least one read in both replicates . The average GRO-seq signal within the window is shown in RPKM ( reads per kilobase per million ) . The red line represents a theoretical 1:1 fit . The statistical relationship between the replicates is indicated by the Spearman correlation coefficient ρ . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 00810 . 7554/eLife . 00808 . 009Figure 1—figure supplement 3 . GRO-seq signal within protein coding genes is highly correlated between replicates . ( A ) – ( E ) Average GRO-seq expression within the gene bodies was calculated using gene models from the WormBase WS230 release . For all genes greater than 1 . 1 kb , the GRO-seq signal was totaled within the gene body , excluding the first and last 300 bp . The total GRO-seq signal was divided by the total number of uniquely mappable base pairs within the same region to generate the average expression . The average gene body expression level in RPKM ( reads per kilobase per million ) is plotted on the axes . The statistical relationship between the replicates is indicated by the Spearman correlation coefficient ρ . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 00910 . 7554/eLife . 00808 . 010Figure 1—figure supplement 4 . GRO-seq expression is correlated with gene expression from microarray and RNA-seq experiments . ( A ) GRO-seq experiments have a higher dynamic range than microarray experiments . Scatter plots are shown of gene expression levels determined by GRO-seq vs microarray experiments from control RNAi embryos . Average GRO-seq expression was calculated as in Figure 1—figure supplement 3 . Microarray data were obtained from Jans et al . ( 2009 ) . The GRO-seq signal is shown as the log10 of the average RPKM ( reads per kilobase per million ) , and microarray data are shown as the log10 of expression values . ( B ) GRO-seq and RNA-seq data are correlated . Scatter plots are shown of gene expression levels determined by GRO-seq vs RNA-seq experiments from starved L1s . Average GRO-seq expression was calculated as in Figure 1—figure supplement 3 , and RNA-seq data were obtained from Maxwell et al . ( 2012 ) . The GRO-seq signal is the log10 of the average RPKM , and RNA-seq reads are the log10 of FPKM ( fragments per kilobase per million ) . The two samples show a Spearman correlation coefficient of 0 . 788 . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 01010 . 7554/eLife . 00808 . 011Figure 1—figure supplement 5 . Genome-wide annotation of TSSs improves gene models . ( A ) – ( D ) To gauge the improvement of our new transcription start site ( TSS ) calls on gene model accuracy , we plotted the average GRO-seq signal across a 2 kb window centered on the WormBase ( WB ) starts or TSSs for genes having TSSs identified in the same developmental stage . For example , ( A ) shows a plot of the average GRO-seq signal from 4246 genes of control RNAi embryos around the WB starts or our TSSs called from embryos . The GRO-seq signal is averaged at each bp and then averaged across 25 bp windows . Plotting the GRO-seq signal against real TSSs reduces the upstream signal due to incorrectly annotated gene starts , indicating a dramatic improvement in gene models . n represents the number of genes in each stage having a TSS identified in that stage . RPKM: reads per kilobase per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 01110 . 7554/eLife . 00808 . 012Figure 1—figure supplement 6 . GRO-cap signal is strong at newly annotated TSSs . ( A ) – ( D ) Because the corrected GRO-cap signal ( TAP+ signal after subtracting the TAP− signal ) was used to annotate transcription start sites ( TSSs ) , we assessed whether our TSS calls coincided with the spike of the GRO-cap signal , as would be expected . To do so , we averaged the corrected GRO-cap signal across a 2 kb window centered on the TSSs or WormBase ( WB ) starts for genes having TSS annotated in the same developmental stage . For example , ( A ) shows the GRO-cap signal from control RNAi embryos plotted for genes with a TSS call in wild-type embryos . Each plot shows increased GRO-cap signal at the TSSs compared to the WB starts , indicating a vast improvement in gene models . The GRO-cap signal was averaged over 25 bp windows . n represents the number of genes in each stage having a TSS identified in that stage . RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 01210 . 7554/eLife . 00808 . 013Figure 1—figure supplement 7 . Heat maps showing GRO-seq and GRO-cap signal relative to either WB starts or TSSs for developmental stages reveal improvements in gene models . ( A ) – ( C ) The GRO-seq signal was plotted , one gene per row , for each gene of a specific developmental stage relative to the WormBase ( WB ) starts . The genes were ordered from top to bottom with increasing distance between the transcription start site ( TSS ) and WB start . The GRO-seq signal was averaged across 15 bp windows . Darker red indicates more transcription . ( D ) Heat maps showing the GRO-cap signal from sdc-2 mutant embryos plotted against either WB starts ( left ) or TSSs ( right ) called in wild-type embryos for 4246 genes . The genes were ordered with increasing distance between the TSS and WB start . RPKM: reads per kilobase per million; RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 01310 . 7554/eLife . 00808 . 014Figure 1—figure supplement 8 . TSSs can be far upstream of the previously annotated WB starts . ( A ) – ( C ) Shown are GRO-seq signals , GRO-cap signals ( TAP+ and TAP− or TAP− subtracted from TAP+ [E–F] ) , ChIP-chip signals of phospho ser 2 Pol II ( from Pferdehirt et al . , 2011 ) , and ChIP-chip signals of hypo-phosphorylated Pol II ( 8WG16 antibody , modENCODE_3545 ) for genes whose transcription start sites ( TSSs ) are far upstream of the WormBase ( WB ) starts . ( A ) The TSS for ZK1073 . 1 is 5051 bp upstream of the WB start and 5101 bp upstream of the trans-splice acceptor site . ( B ) The TSS for ztf-22 is 8927 bp upstream of the WB start and 8917 bp upstream of the trans-splice acceptor site . ( C ) The TSS for operon CEOP4336 is 6693 bp upstream of both the WB start and the first trans-splice acceptor site . The combination of continuous Pol II signal in the upstream regions and the lack of 3′ UTRs or polyA signals ( Mangone et al . , 2010 ) in the upstream regions implies that transcription within the outron is not from sources other than the designated TSS . These results strongly support the argument that the GRO-cap signal paired with the continuous GRO-seq signal from the WB start defines true TSSs . RPKM: reads per kilobase per million; RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 01410 . 7554/eLife . 00808 . 015Figure 1—figure supplement 9 . GRO-cap revealed that 21 U-RNAs have a TSS 2 bp upstream of the mature RNA . GRO-cap readily identified the transcription start sites ( TSSs ) of 21 U-RNAs from L3 larvae . To map TSSs , we determined the highest GRO-cap signal ( TAP+ minus TAP− ) within 10 bp of the 5′ end of 9148 mature , non-overlapping 21 U-RNAs . Of these RNAs , 5260 ( 57 . 5% ) had a putative TSS with a GRO-cap Z-score greater than 3 in the 10 bp interval ( p<0 . 01 ) . The TSSs for 4783 ( 91% ) of 21 U-RNAs RNAs with a called TSS were precisely 2 bp upstream of the mature RNA , indicating that 21 U-RNAs receive a 5′ cap and are processed to the mature sequence by removing the two 5′-most base pairs . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 01510 . 7554/eLife . 00808 . 016Figure 1—figure supplement 10 . Features of promoters and TSSs . ( A ) A transcription start site ( TSS ) can be far downstream of the WormBase ( WB ) start . The TSS for MO3C11 . 3 is 2510 bp downstream of the WB start in all developmental stages examined . The TSS was identified for ymel-1 , a downstream gene in the operon known to have either SL1 or SL2 RNA leaders on its mRNA . ( B ) WB gene models can be based on inaccurately predicted transcript isoforms . rpt-4 , the first gene in an operon , has two annotated RNA isoforms in WB . Isoform a has a small annotated exon followed by an intron of greater than 2 kb . However , the only TSS in the region identified by GRO-cap in all developmental stages assayed is just upstream of the SL1 splice acceptor site for isoform b , implying that isoform a is incorrect or expressed in a stage not analyzed . ( C ) Identification of the TSS for a gene within an operon . tag-30 mRNA is trans-spliced with an SL2 RNA , indicating it is processed from a polycistronic message , and it is trans-spliced with an SL1 RNA , indicating it also transcribed from an independent promoter . GRO-cap identified the internal TSS for tag-30 . ( D ) Genes can have two or more different RNA isoforms that share the same 5′ end but different 3′ ends . GRO-seq identifies the 3′ accumulation of Pol II corresponding to both 3′ ends . GRO-seq and corrected GRO-cap ( TAP+ signal after subtracting TAP− signal ) signals are shown for unc-84 . RPKM: reads per kilobase per million; RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 01610 . 7554/eLife . 00808 . 017Figure 1—figure supplement 11 . Distances between the TSS and WB starts of the trans-splicing acceptor site . ( A ) and ( B ) For all genes with a transcription start site ( TSS ) called in wild-type embryos , the difference between the TSS and WormBase ( WB ) start or SL1 trans-splice acceptor site was grouped in bins of 25 bp and plotted as a histogram . Positive distances mean that the TSS is upstream of the annotated WB start or trans-splice acceptor site , while negative distances mean the TSS is downstream . The dotted red line demarks the position where the TSS calls are the same as the WB starts or trans-splicing acceptor site . ( A ) Plot of distance between TSS and WB start . The prevalence of distances near zero suggests that many WB start positions are correct and likely reflects non-trans-spliced genes . ( B ) Plot of outron length , the distance between the TSS and site of SL1 attachment to RNAs ( Allen et al . , 2011 ) . The SL1 trans-splice acceptor sites correspond to the site with the highest number of SL1 reads . Genes with multiple isoforms were eliminated from the analysis if the most 5′ part of the isoform differed . Many outrons are in the 50–500 bp range , consistent with previous estimations , but our data show that outron length is often significantly longer , up to 14 kb . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 01710 . 7554/eLife . 00808 . 018Figure 1—figure supplement 12 . Comparison of enhancers in Chen et al . ( 2013 ) and our annotated TSSs . ( A ) – ( C ) Comparison of our GRO-seq and GRO-cap profiles to enhancer regions from Chen et al . ( 2013 ) . ( A ) Chen et al . ( 2013 ) analyzed scRNA sequencing data from mixed-stage embryos to annotate transcription start sites ( TSSs ) . They required that clusters of scRNA signal ( labeled as ‘TICs’ and shown in red or blue ) be within 200 bp of the WormBase ( WB ) start to be annotated as a new TSS . They then classified as enhancers ( shown in yellow ) the scRNA clusters that are not associated with a gene , have specific chromatin modifications not associated with promoters , and overlap with transcription factor binding sites . Shown is a panel from their Figure 6A ( © Genome Research , Cold Spring Harbor Laboratory Press ) . Our GRO-seq and GRO-cap ( TAP+ minus TAP− ) data from mixed-stage embryos are shown for the same genomic region upstream of the tol-1 gene as in Figure 6A . Our data and their data are precisely aligned . In this example , spikes of GRO-cap signal correspond with TICs , and several GRO-cap spikes correspond to their newly annotated enhancers . Continuous ChIP-chip signal for hypo-phosphorylated Pol II ( 8WG16 antibody , modENCODE_3545 ) and continuous GRO-seq signal occur from the most upstream enhancer to the WB start . The GRO-seq signal increases in intensity as it passes GRO-cap spikes and TICs , suggesting that each transcription initiation event contributes to the cumulative Pol II signal , which stops increasing in intensity once it reaches the WB start . In addition , no 3′ UTRs or polyA sites were found in this tol-1 upstream region from the data sets of Jan et al . ( 2011 ) and Mangone et al . ( 2010 ) , implying no tol-1-independent polyadenylated transcription units in the upstream region . This analysis suggests that for this region some enhancers are likely to be TSSs that give rise to full-length transcripts . ( B ) The TSS we called for agmo-1 from GRO-cap and GRO-seq data corresponds to the enhancer called by Chen et al . ( 2013 ) . ChIP-chip data from modENCODE for hypo-phosphorylated Pol II antibody and the lack of 3′ UTRs support the TSS call . The distance between the TSS and both the WB start and the trans-splice acceptor site is 2534 bp . This example further supports the proposal that a proportion of the enhancers are outron TSSs . ( C ) The TSS we called for pat-2 from GRO-cap and GRO-seq data corresponds to one of the two enhancers called by Chen et al . ( 2013 ) upstream of pat-2 . ChIP-chip data from modENCODE for hypo-phosphorylated Pol II antibody and the lack of 3′ UTRs support the TSS call . The distance between the TSS and the WB start is 2878 bp , while the distance from the TSS to the trans-splice acceptor site is 2875 bp . This example further supports the proposal that a proportion of the enhancers are outron TSSs . The ChIP-chip signal 3′ of the pat-2 3′ UTR is from the genes on the opposite strand . RPKM: reads per kilobase per million; RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 018 The two GRO-seq biological replicates for each stage had high statistical correlation throughout the genome ( Spearman correlation , ρ > 0 . 94 ) and across gene bodies ( Spearman correlation , ρ > 0 . 98 ) ( Figure 1—figure supplements 1–3 and Figure 1—source data 1 ) . Gene expression levels calculated from GRO-seq data correlated well with expression data from microarrays and RNA-seq experiments ( Figure 1—figure supplement 4 ) . For the majority of expressed genes , we found continuous GRO-seq signal upstream of the WormBase ( WB ) -annotated transcription starts ( Figure 1A , B ) , suggesting that GRO-seq reactions contain nascent RNAs with true 5′ ends . To map TSSs unambiguously , we performed a series of enzymatic selections on our GRO-seq run-on RNAs to capture only those RNAs with a 5′ cap , the 7-methyl guanosine residue added shortly after transcription initiation ( Rasmussen and Lis , 1993 ) . This modified GRO-seq procedure , called GRO-cap ( Figure 2 ) , enabled us to map TSSs with nucleotide resolution by tracking nascent RNAs prior to trans-splicing . Use of nascent RNAs without size selection also reduces the background from RNAs that are capped post-transcriptionally and increases the probability of identifying TSSs from promoters with low Pol II occupancy . 10 . 7554/eLife . 00808 . 019Figure 2 . GRO-cap strategy for identifying TSSs . GRO-cap is a modified form of GRO-seq that utilizes the tagging and extensive purification of nascent RNAs from GRO-seq ( Core et al . , 2008 ) and then employs redundant enzymatic steps to enrich for RNAs with 5′ caps . Of particular importance to this study , GRO-cap permits analysis of RNAs prior to their co-transcriptional processing , which replaces true transcription start sites ( TSSs ) with trans-spliced leader RNAs in Caenorhabditis elegans . GRO-seq run-ons have been tuned to extend the length of nascent RNAs by only 100 nucleotides on average , thus minimizing any possibility that independent transcription units might been artifactually linked . In GRO-cap , nuclei are isolated and RNA polymerases are allowed to transcribe briefly in a run-on reaction in the presence of Br-UTP , as in GRO-seq . RNA is isolated but the base-hydrolysis step of GRO-seq is omitted to increase the probability of capturing nascent RNA molecules with a 5′ 7-methyl-GTP cap . BrU-RNAs made during the run-on reaction are enriched by selection with anti-BrdU beads to ensure the identification of true TSSs from capped nascent RNAs rather than 5′ ends from RNAs that received post-transcriptional capping ( Fejes-Toth et al . , 2009 ) . A 3′ RNA adapter ( red ) is ligated to the RNAs , followed by another round of bead enrichment . Selection against 5′ mono-phosphate RNAs that do not represent capped RNAs ( and any carry-through 5′ RNA adapters ) is achieved by sequential enzymatic treatment with Terminator exonuclease to degrade 5′ mono-phosphate RNAs and then alkaline phosphatase to remove 5′ phosphates from 5′ mono-phosphate RNAs resistant to the exonuclease . Half of the nuclear run-on ( NRO ) RNA pool is treated with tobacco acid pyrophosphatase ( TAP+ ) to remove the 5′ cap from the RNA , thereby exposing a 5′ mono-phosphate . The other half is left untreated ( TAP− ) to provide a control population of residual 5′ mono-phosphate RNAs that never had 5′ caps . The 5′ mono-phosphate RNAs are ligated to 5′ RNA adapters ( blue ) . The TAP+ and TAP− samples are prepared for Illumina sequencing as in GRO-seq by reverse transcription of RNA into DNA and then amplification of DNA from 5′ and 3′ adapter regions . We note that transcripts <500 bp are captured most efficiently on Illumina sequencing platforms . The enriched TSS regions are identified by mapping the 5′ ends of the sequence reads back to the genome and comparing the TAP+ and TAP− sites to eliminate false TSSs . Comparing the GRO-cap candidate TSSs to the 5′ ends of transcription units defined by GRO-seq permits reliable assignment of TSSs to transcription units . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 019 For genes that lack trans-splicing , the site of maximum GRO-cap signal was coincident with both the 5′-most GRO-seq signal and the WB-annotated start ( Figure 1C ) , confirming that GRO-cap and GRO-seq together permit high-confidence mapping of TSSs . For trans-spliced genes with no previously identified TSSs , strong GRO-cap and GRO-seq signals were found upstream of the WB-annotated starts , and TSS calls were supported by uninterrupted GRO-seq signal between the maximum GRO-cap signal and the WB start ( Figure 1A ) . In total , a TSS was identified for 31 . 7% ( 6353 genes ) of all C . elegans protein-coding genes from at least one of the three developmental stages examined ( Figure 1—source data 2 ) . Of our TSS calls , 77% are for genes shown previously to be trans-spliced ( Figure 1—source data 2 ) ( Allen et al . , 2011 ) . Plotting the average GRO-seq and GRO-cap signals from each developmental stage relative to the TSSs from the same stage revealed the vast improvement in gene models ( Figure 1F , G and Figure 1—figure supplements 5 , 6 and 7A–C ) . Independent GRO-cap reactions from the same developmental stage gave very similar results ( Figure 1G and Figure 1—figure supplement 7D ) . TSSs were also annotated for the majority of genes encoding short non-coding RNAs such as snoRNAs , 21 U-RNAs , and microRNAs , including TSSs for the five polycistronic microRNA clusters ( Figure 1C , E and Figure 1—source data 3 ) ( see ‘Materials and methods’ ) . The TSSs for the 21 U-RNAs were 2 bp upstream from the mature RNA ( Figure 1—figure supplement 9 ) . In addition , the primary TSSs and gene composition were determined for many multigenic transcription units called operons in which the polycistronic pre-mRNAs are processed to monocistronic mRNAs through 3′ end formation and trans-splicing using SL2 RNA ( Figure 1D , E ) ( Blumenthal , 2012 ) . Identification of the TSS for one operon showed it to include a 5′ gene not previously ascribed to it ( Figure 1D ) . TSSs for several operons showed the TSS to be far downstream ( >2 kb ) of the WB start , a result we also found for genes not included in operons ( Figure 1—figure supplement 10A , B ) . Within operons some genes have independent promoters to transcribe their pre-mRNAs ( Allen et al . , 2011 ) , and their TSSs were determined , including snoRNA genes within introns of internal genes ( Figure 1E and Figure 1—figure supplement 10A , C ) . In general , we used conservative statistical criteria for calling TSSs , and additional TSSs can be identified by visual inspection of the data . The new TSS calls revealed promoters to be further upstream from the 5′ ends of mature mRNAs than previously thought , as demonstrated by heat maps showing the GRO-seq signal or GRO-cap signal of re-annotated genes relative to WB starts or TSSs ( Figure 1F , G and Figure 1—figure supplement 7A–D ) and histograms showing distances between TSSs and WB starts or SL1 trans-splice acceptor sites ( TSA ) ( Figure 1—figure supplement 11A , B ) . The TSS-to-TSA , called the outron , was previously thought to be 50–500 bp . We found instead that outrons can be as long as 14 kb and have a median of 260 bp and mean of 753 bp ( Figure 1—figure supplement 11B ) . Fully 59% of outrons are longer than 200 bp , 21% are longer than 1 kb , and 2 . 3% are longer than 5 kb ( e . g . , Figure 1—figure supplement 8A–C ) . Multiple lines of evidence indicate that the GRO-seq signal between newly called TSSs and previously identified TSAs reflects legitimate outrons rather than independent overlapping upstream transcripts . First , 3′ UTRs or polyA signals are rare in outrons of >1 kb in length , indicating the engaged Pol II is not from independent polyadenylated transcripts . From 565 such outrons , only 1 . 4% had an identified 3′ UTR in the Jan et al . ( 2011 ) study , and 0 . 7% had a 3′ UTR in the Mangone et al . ( 2010 ) study . Furthermore , only 3 . 5% had a polyA site ( Mangone et al . , 2010 ) . Second , regions corresponding to long outrons have a continuous ChIP-chip signal from antibodies enriched for either the ser2 phosphorylated form of Pol II or the hypo-phosphorylated form of Pol II ( Figure 1—figure supplement 8A–C ) . These results and the restricted run-on length of ∼100 nucleotides ( Core et al . , 2008 ) indicate that the GRO-seq signal corresponds to bound Pol II in vivo and is not an artifact of the nuclear run-on ( NRO ) reactions in vitro extending beyond the 3’ ends defined in vivo ( Figure 1—figure supplement 8A–C and Figure 3A ) . Third , a heat map of individual genes showing the GRO-cap signal relative to TSSs reveals that a dominant TSS contributes the majority of the vast GRO-cap signal ( Figure 1G ) . Together these observations strongly support the argument that the GRO-cap signal paired with the continuous GRO-seq signal from WB starts defines true TSSs . Recently published studies of TSSs in C . elegans that used cutoffs for TSS calls of either 1 kb upstream ( Gu et al . , 2012 ) or 200 bp upstream ( Chen et al . , 2013 ) of WB starts identified some outron TSSs but could not identify TSSs for a large class of genes with longer outrons ( see ‘Discussion’ ) . 10 . 7554/eLife . 00808 . 020Figure 3 . Features of promoters and TSSs . ( A ) Trans-spliced genes can have multiple transcription start sites ( TSSs ) , suggesting that trans-splicing eliminates the pressure to have only one precise TSS per gene . Shown are GRO-seq and corrected GRO-cap ( TAP+ signal after subtracting TAP− signal ) signals with ChIP-seq data of hypo-phosphorylated Pol II ( 8WG16 antibody , modENCODE_2439 ) for the trans-spliced gene sca-1 expressed in the L3 larval stage . The total GRO-seq signal becomes more intense as additional TSSs ( from 5′ to 3′ ) contribute to the pool of engaged Pol II molecules that transcribe through the upstream regulatory region of sca-1 . The combination of continuous Pol II signal in the upstream region and the lack of 3′ UTRs or polyA signals ( Mangone et al . , 2010 ) strengthens the interpretation that the GRO-cap signal combined with the continuous GRO-seq signal identified true TSSs for sca-1 . From left to right , the TSSs reside upstream of the WormBase ( WB ) gene model by 5728 bp , 4582 bp , 3044 bp , 1669 bp , and 159 bp . The ChIP-seq signal 3′ of the sca-1 3′ UTR is from the klp-7 gene on the opposite strand . ( B ) A gene can use different primary TSSs in different developmental states . The primary TSS for tag-294 in embryos and L3 larvae is 1529 bp upstream of the WB start , while the primary TSS in starved L1 larvae is 656 bp upstream . DNA sequences flanking newly annotated TSSs have evolutionarily conserved core promoter elements , including ( C ) TATA-box elements and ( D ) initiator elements ( Inr ) . Of 4547 embryo genes with TSSs , 162 genes ( 3 . 6% ) have a TATA element with a perfect match to the consensus 15–45 bp upstream of it , and 745 genes ( 16 . 4% ) have an Inr with the adenine residing at the TSS ( +1 bp ) . Consensus sequences for TATA elements and the Inr are above the graphs . RPKM: reads per kilobase per million; RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 02010 . 7554/eLife . 00808 . 021Figure 3—figure supplement 1 . Evolutionarily conserved promoter elements . ( A ) A TATA box element with one or no mismatch from the consensus ( TATAWAWR ) is highly enriched 15–45 bp upstream of transcription start sites ( TSSs ) for 391 of 4547 Caenorhabditis elegans genes . The consensus derived from the 578 elements in this region is above the histogram . ( B ) The core promoter region is highly conserved across nematode species . The UCSC Genome Browser uses phastCons ( http://compgen . bscb . cornell . edu/phast/ ) to investigate DNA sequence conservation across seven Caenorhabditis species . Values range from 0 ( no conservation ) to 1 ( highest conservation ) for each base pair . We calculated the average DNA conservation in a 2 kb window surrounding the new TSSs ( top ) and WormBase ( WB ) starts ( bottom ) and found substantial conservation in each location , likely for different reasons . The conservation at TSSs likely reflects conservation of core promoter elements , and the conservation near WB starts likely reflects conservation at the junction between trans-splice acceptor site and first exon . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 02110 . 7554/eLife . 00808 . 022Figure 3—figure supplement 2 . Conserved core promoter elements in promoters of microRNA genes . ( A ) A TATA box element with a perfect match to the consensus derived for genes encoding microRNAs was highly enriched 29–32 bp upstream of transcription start sites ( TSSs ) for 15 of 57 microRNA genes . The consensus is shown above the histogram . The distance between the TATA element and the TSS was calculated from the 3′-most base of the TATA element . ( B ) A TATA element with one or no mismatch from the consensus is highly enriched 15–45 bp upstream of TSSs for 24 of 57 microRNA genes . The consensus derived from the 31 elements in this region is above the histogram . ( C ) Inr elements are enriched in promoters of microRNA genes . Of 57 microRNA genes , 21 have an Inr with an adenine at the +1 position of the TSS . The consensus derived from the 38 Inr motifs in the region is above the histogram . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 022 Several other noteworthy features of promoters emerged from this comprehensive mapping of C . elegans transcription units and TSSs . ( 1 ) Many trans-spliced genes have multiple TSSs ( Figure 3A ) , suggesting that trans-splicing has removed the selective pressure to form promoters with a single TSS . ( 2 ) Genes can use different TSSs across developmental stages , indicating developmental stage-specific regulation of transcription initiation ( Figure 3B ) . ( 3 ) DNA sequences flanking the newly annotated TSSs have strong sequence conservation across nematode species ( Figure 3—figure supplement 1B ) and also have evolutionarily conserved core promoter elements , including the TATA-box ( worm consensus TATAWAWR ) ( Figure 3C and Figure 3—figure supplements 1A and 2A , B ) and the initiator element ( Inr ) ( worm consensus YCAYTY ) ( Figure 3D and Figure 3—figure supplement 2C ) , both of which facilitate formation of the Pol II pre-initiation complex ( Juven-Gershon and Kadonaga , 2010 ) . Three prominent features of transcription emerged . Accumulation of Pol II at 3′ ends of genes is abundant ( Figure 4A–C and Figure 4—figure supplement 1A , B ) , as is divergent transcription from promoters lacking upstream divergent genes ( Figure 4D–F ) . In contrast , promoter-proximal RNA Pol II pausing in C . elegans is rare under normal growth conditions , unlike in other metazoans , as shown later in ‘Results’ . 10 . 7554/eLife . 00808 . 023Figure 4 . Features of Caenorhabditis elegans transcription: 3′ Pol II pausing and divergent transcription . ( A ) – ( C ) Pol II 3′ accumulation is prevalent in worms . 3′ End pausing ratios were calculated by dividing the highest average GRO-seq signal at the 3′ end by the average GRO-seq signal in the gene body . ( A ) A histogram of the 3′ end pausing ratios shows 3′ accumulation of Pol II is more extensive in Caenorhabditis elegans than in Drosophila . The histogram compares 3′ accumulation for 3984 genes expressed in C . elegans embryos with 6107 genes expressed in Drosophila cell lines . ( B ) The GRO-seq signal surrounding the 3′ end ( cleavage and polyadenylation site [CPS] ) was averaged for genes in or not in operons . Genes at the beginning ( n = 430 ) and middle ( n = 276 ) of operons were more highly expressed and had higher 3′ accumulation than genes at the end ( n = 474 ) of operons or not in operons ( n = 5048 ) . Genes plotted had to be greater than 3 kb in length . ( C ) 3′ End pausing ratios were calculated for all classes of genes in ( B ) and plotted as boxplots . For this analysis , genes had to be greater than 3 kb in length and the gene body RPKM ( reads per kilobase per million ) had to be ≥1 . Genes found at the beginning ( n = 415 ) and middle ( n = 275 ) of operons had higher 3′ pausing than genes at the end ( n = 467 ) of operons . Genes lacking a downstream gene had similar 3′ pausing ratios whether or not ( n = 3670 ) they were in operons . The 3′ pausing ratios for genes in all classes were greater than for Drosophila genes ( n = 3260 ) . ( D ) – ( F ) Upstream divergent transcription is common at promoters of C . elegans genes . GRO-seq and GRO-cap profiles show transcription of a divergent gene pair ( D ) or divergent transcription from a promoter without an upstream divergent gene partner ( E ) . Gene on plus strand ( red gene and signal ) . Gene on minus strand ( blue gene and signal ) . ( F ) Upstream divergent transcription from C . elegans promoters is intermediate between that in humans and Drosophila . Plot compares the log2 ( sense/antisense ) transcription ratio of human and fly promoters to C . elegans promoters without divergent gene pairs . The median log2 ratios are 0 . 3 for humans , 2 . 3 for C . elegans , and 5 . 0 for Drosophila . RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 02310 . 7554/eLife . 00808 . 024Figure 4—source data 1 . Gene expression , and 5′ pausing and 3′ pausing data for protein-coding genes . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 02410 . 7554/eLife . 00808 . 025Figure 4—figure supplement 1 . The 3′ accumulation of RNA polymerase II . ( A ) The 3′ accumulation of RNA Pol II is positively correlated with gene expression . The maximum average GRO-seq signal from control RNAi embryos was calculated across a 200 bp window at the 3′ end of each gene greater than 1 . 1 kb with an average expression of greater than 1 RPKM ( reads per kilobase per million ) in the gene body . This 3′ end GRO-seq average was compared with the average GRO-seq expression of the same genes in a scatter plot . The positive correlation of 3′ RNA Pol II accumulation and level of gene expression is reflected by the Spearman correlation of 0 . 776 . ( B ) The 3′ pausing ratios are greater in Caenorhabditis elegans than in Drosophila . To calculate 3′ pausing ratios , the maximum average GRO-seq signal across a 200 bp window near the 3′ end was divided by the average GRO-seq signal in the gene body . The distribution of ratios from embryos , starved L1s , L3s , and Drosophila S2 cells was plotted . Although the distributions of all C . elegans states were significantly different from Drosophila S2 cells ( Mann–Whitney U test , p<0 . 0001 for all comparisons to S2 cells ) , the 3′ pausing ratios got progressively smaller for the later C . elegans developmental stages . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 02510 . 7554/eLife . 00808 . 026Figure 4—figure supplement 2 . RNA polymerase II accumulation at 3′ ends compared against trans-spliced genes , non-trans-spliced genes , and U-rich regions at 3′ ends . ( A ) Shown are the 3′ pausing ratios for the first and middle genes in operons ( n = 949 ) , last genes in operons ( n = 625 ) , monocistronic genes with trans-splicing ( n = 603 ) , monocistronic genes without trans-splicing ( n = 603 ) , and Drosophila genes ( n = 3942 ) plotted as boxplots . All genes in this analysis had to be ≥2 kb and have a gene body RPKM ( reads per kilobase per million ) ≥5 . For comparison of monocistronic gene sets , each monocistronic gene with trans-splicing had a monocistronic gene without trans-splicing of equivalent expression level . The first and middle genes in operons had the highest 3′ pausing ratio . Monocistronic genes with trans-splicing had a slightly higher 3′ pausing ratio than last genes in operons . Monocistronic genes with trans-splicing had a higher 3′ pausing ratio than monocistronic genes lacking trans-splicing ( Mann–Whitney U p<10−10 ) . The 3′ pausing ratios for genes in all classes were greater than for Drosophila genes . Monocistronic genes were classified as trans-spliced or not trans-spliced depending on whether an SL leader was identified in Allen et al . ( 2011 ) . ( B ) and ( C ) Accumulation of Pol II GRO-seq signal at 3′ ends ( cleavage and polyadenylation site [CPS] ) does not overlap with U-rich regions . ( B ) Accumulation of Pol II at 3′ ends of the first and middle genes in operons ( n = 706 ) , last genes in operons ( n = 474 ) , and genes not in operons ( n = 5048 ) . ( C ) Plot of relative U-richness for the gene sets in ( B ) . The proportion of genes with a U at each base pair was calculated and then averaged over 25 bp windows . The vertical green line in ( B ) is aligned with the peak of 3′ GRO-seq signal and is drawn at the same position in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 02610 . 7554/eLife . 00808 . 027Figure 4—figure supplement 3 . Divergent transcription in Caenorhabditis elegans . ( A ) and ( B ) Comparison of average GRO-seq signal around promoters with a divergent gene pair vs promoters not associated with a divergent gene partner . The genes analyzed had a log2 ( sense/antisense ) of ≤1 . 5 . The upstream divergent transcripts from a promoter without a divergent gene partner are shorter and less abundant than those from a promoter of a divergent gene pair . ( C ) Comparison of the GRO-cap signal around promoters of divergent gene pairs ( blue ) and promoters not associated with a divergent gene partner . Upstream divergent transcription ( also referred to as antisense ) begins at approximately the same distance from the transcription start site ( TSS ) of a gene whether or not a divergent gene partner is present . The GRO-cap signal was only evaluated in this analysis for genes having a log2 ( sense/antisense ) ratio ≤1 . 5 . Distance was calculated between the TSS and the maximum antisense GRO-cap signal within 500 bp . ( D ) and ( E ) Promoters having a TATA box matching the consensus or having a TATA with no or one mismatch preferentially transcribe in a single direction when no divergent gene is upstream . Comparison of the average GRO-seq signal from control RNAi embryos in sense and antisense directions for a 3 kb window surrounding the TSS . RPKM: reads per kilobase per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 027 Pol II 3′ accumulation , likely caused by slow 3′ end formation and RNA processing ( Gromak et al . , 2006 ) , is positively correlated in C . elegans with the expression level of the gene ( Figure 4—figure supplement 1A and Figure 4—source data 1 ) and is more extensive in C . elegans than in Drosophila ( Figure 4A and Figure 4—figure supplement 1B ) . Multiple peaks of 3′ accumulation within a gene help identify genes having several isoforms with distinct 3′ ends ( Figure 1—figure supplement 10D ) . The prevalence in C . elegans of polycistronic operons requiring extensive RNA processing to produce monocistronic mRNAs caused us to ask whether 3′ Pol II accumulation was positively correlated with the gene’s position in an operon and hence the level of RNA processing . First and middle genes in an operon require two forms of co-transcriptional RNA processing at the 3′ end to generate monocistronic mRNAs , polyadenylation of the upstream gene , and trans-splicing of the downstream gene , while the terminal gene requires only polyadenylation at the 3′ end . First and middle genes showed more 3′ accumulation than terminal genes or genes not in operons ( Figure 4B , C ) . In other words , 3′ pausing is longer at genes with another gene to process just downstream . The 3′ Pol II accumulation at terminal genes and genes not in operons was equivalent , yet greater than in Drosophila genes . Lastly , 3′ accumulation was similar between terminal genes of operons and monocistronic genes undergoing trans-splicing , and both gene sets had greater 3′ accumulation than monocistronic genes lacking trans-splicing , which nonetheless had more 3′ accumulation than Drosophila genes ( Figure 4—figure supplement 2A ) . Cues triggering trans-splicing , particularly operon-specific trans-splicing , appear to facilitate 3′ accumulation and perhaps predispose C . elegans Pol II to greater 3′ pausing genome-wide , whether or not a gene resides in an operon . It should be noted that Pol II accumulation at 3′ ends does not overlap with U-rich regions at 3′ ends . Therefore , the high GRO-seq signal is not due to selective enrichment of U-rich RNAs ( Figure 4—figure supplement 2B , C ) . GRO-seq readily detects divergent transcription from promoters . In C . elegans , divergent transcripts are short and initiated 75–150 bp upstream from TSSs of promoters lacking upstream divergent genes ( Figure 4E and Figure 4—figure supplement 3A–C ) . The frequency of upstream divergent transcription in C . elegans appears to be intermediate in degree between that in mammals , where it occurs at the majority of active promoters ( Kapranov et al . , 2007; Core et al . , 2008; Seila et al . , 2008 ) , and that in Drosophila , where it occurs only rarely ( Nechaev et al . , 2010; Core et al . , 2012 ) ( Figure 4F ) . For both mammals and worms , promoters with TATA elements rarely support divergent transcription ( Figure 4—figure supplement 3D , E ) ( Core et al . , 2012 ) . These results underscore fundamental similarities and differences in the architecture , evolution , and function of eukaryotic promoters . Analysis of transcriptionally engaged Pol II in wild-type vs dosage-compensation-defective embryos provided a robust assessment of the genome-wide impact of disrupting dosage compensation . To assess dosage compensation , we disrupted sdc-2 ( sex determination and dosage compensation ) , the central XX-specific factor that triggers assembly of all DCC components onto X and induces hermaphrodite sexual differentiation by repressing the autosomal male sex-determining gene her-1 ( Figure 5A ) ( Dawes et al . , 1999 ) . Without sdc-2 , DCC subunits fail to bind to X , and her-1 is expressed , causing XX embryos to become severely masculinized and die from overexpression of X-chromosome genes ( Figure 5A ) . The pivotal role of sdc-2 in dosage compensation made its depletion the most effective way to disrupt dosage compensation , although depletion of any DCC condensin subunit causes similar XX lethality and elevation of X gene expression , as shown previously by our genome-wide measurements of gene expression ( Jans et al . , 2009 ) . 10 . 7554/eLife . 00808 . 028Figure 5 . GRO-seq analysis of dosage-compensation . ( A ) Genetic hierarchy for coordinate control of sex determination and dosage compensation . sdc-2 is expressed solely in XX embryos and triggers the hermaphrodite fate . sdc-2 acts together with sdc-1 and sdc-3 , both zinc finger proteins , to induce hermaphrodite sexual development by repressing transcription of the male sex-determining gene her-1 . sdc-2 acts together with sdc-3 and dpy-30 , also a member of the MLL/COMPASS gene activating complex , to load the DCC onto X and thereby turn dosage compensation on . sdc-2 is the single gene required for all DCC components to assemble onto X . Without sdc-2 , her-1 is expressed , causing sexual transformation of XX embryos to the male fate , and the DCC fails to assemble onto X , causing severe dosage compensation disruption and the death of all XX embryos . The DCC contains not only the X loaders ( red , orange ) but also five homologs of the mitotic condensin complex ( yellow , blue , green ) . The DCC binds to the X chromosomes of only XX animals to reduce transcription by half , thereby equalizing X-chromosome gene expression between males ( XO ) and hermaphrodites ( XX ) . ( B ) GRO-seq shows that both RNA isoforms of her-1 are elevated in sdc-2 XX mutants . her-1 is expressed at such a low level in XX embryos that the two transcription start sites ( TSSs ) are only evident with GRO-cap in the sdc-2 mutants . The gene model ( red arrow ) incorporates 3′ end data from Jan et al . ( 2011 ) . ( C ) and ( D ) The X-linked protein coding gene pdi-2 is elevated in expression in sdc-2 mutants as is the gene encoding the mir-62 microRNA . For pdi-2 , the elevation starts at the TSS and is evident throughout the gene . Red arrows show our re-annotated gene models . ( E ) The DCC subunit DPY-27 binds just upstream of the TSS . Comparison of the average DPY-27 ChIP-seq signal relative to WormBase ( WB ) starts and TSSs of X-linked genes . RPKM: reads per kilobase per million; RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 02810 . 7554/eLife . 00808 . 029Figure 5—source data 1 . Genome-wide changes in gene expression caused by the disruption of dosage compensation . Genes were separated into different gene sets based on their length , origin , and chromosome ( X vs autosome ) to compare GRO-seq gene expression between the sdc-2 mutant and control RNAi embryos . Shown are the total number of genes , the median sdc-2 mutant/control RNAi expression ratio , the average sdc-2 mutant/control RNAi expression ratio together with the standard error of the mean , and the number of genes in each set that are more highly expressed in each condition . Across numerous protein-coding gene sets , the X chromosome is more highly expressed and the autosomes are slightly less expressed in the sdc-2 mutant . Furthermore , when X-linked genes are significantly changed in expression , they are almost exclusively increased in expression . Each gene set is separated into two sets , one containing all genes and the other containing genes that are significantly changed in expression as determined by analysis with DESeq ( p<0 . 05 ) ( Anders and Huber , 2010 ) . For the first two lists ( labeled with ‘≥250 bp’ ) , average GRO-seq gene expression was calculated from the beginning to the end of either the WormBase ( WB ) model or the newly annotated transcription start site ( TSS ) gene model . WB genes had to be expressed at greater than 1 RPKM ( reads per kilobase per million ) , and have at least 250 uniquely mappable bases in both sets . For the next set ( labeled with ‘WormBase WS230 Genes ≥1 . 1 kb’ ) , average GRO-seq expression was calculated for genes greater than 1 . 1 kb , with the first and last 300 bp of the gene excluded . The level of expression had to be ≥1 RPKM for WB genes , and have at least 250 uniquely mappable bases for a gene to be included . For the final two sets of genes ( labeled with ‘≥1 . 1 kb’ and ‘≥1 . 5 kb’ ) , expression was calculated for genes of the indicated length that have a newly annotated TSS , with the first and last 300 bp of the gene excluded . A gene had to have at least 250 uniquely mappable bases for it to be included . RNA polymerase II transcribed microRNAs are controlled by dosage compensation , while RNA polymerase III transcribed tRNAs are not . Average GRO-seq gene expression from sdc-2 mutant and control RNAi embryos was compared across ncRNAs . For microRNAs , expression values were calculated from the full length of the WB ‘primary transcript’ or re-annotated TSS gene models . For tRNAs , expression values were calculated from the beginning of the ‘mature transcript’ to 50 bp downstream of the stop . Because tRNAs are highly repetitive and transcription of highly transcribed tRNAs continues downstream of the stop , the extra 50 bp was included to increase the unique mappability of each tRNA . For a gene to be considered for analysis , it had to have at least 25 bp of uniquely mappable DNA , and to have an average expression of at least 1 RPKM in both control RNAi and sdc-2 mutant embryos . The median and mean sdc-2/control expression levels show that X-linked microRNAs are more susceptible to dosage compensation than autosomal microRNAs . X-linked tRNAs are decreased slightly in expression in the sdc-2 mutant , suggesting that its expression is not controlled by dosage compensation . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 02910 . 7554/eLife . 00808 . 030Figure 5—figure supplement 1 . Western blot shows the reduction in SDC-2 protein levels in sdc-2 ( y93 , RNAi ) animals . The sdc-2 ( y93 ) partial-loss-of function mutant was treated with RNAi against sdc-2 to reduce its gene activity . Extracts from wild-type and sdc-2 mutant embryos were fractionated on an SDS-PAGE gel , transferred to a membrane , and probed with antibodies to SDC-2 and Α-tubulin as a loading control . SDC-2 is less abundant in the RNAi-treated mutant than wild-type embryos . The mutant SDC-2 protein has a lower molecular weight than the wild-type protein because the y93 allele is an in-frame deletion . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 03010 . 7554/eLife . 00808 . 031Figure 5—figure supplement 2 . X-linked gene expression is selectively increased in sdc-2 mutants . ( A ) – ( D ) Scatter plots of the average GRO-seq signal ( log2 RPKM ) from RNAi control and sdc-2 mutant embryos from the bodies of genes on X or autosomes . The genes had to be ≥1 . 1 kb and had to have at least 250 uniquely mappable bp in the gene body . ( A ) – ( B ) Whether the group of genes on X contains the subset of genes with transcription start sites ( TSSs ) or the larger set of genes annotated in WormBase ( WB ) , the genes are more highly expressed in the sdc-2 mutants . ( C ) – ( D ) Whether the group of genes on autosomes contains the subset of genes with TSSs or the larger set of genes annotated in WB , the genes are fairly equivalently expressed in sdc-2 mutant embryos and control RNAi embryos . RPKM: reads per kilobase per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 03110 . 7554/eLife . 00808 . 032Figure 5—figure supplement 3 . Occupancy of the DCC subunit DPY-27 in the promoter of a gene is correlated with the gene’s expression level but not its dosage compensation status . The total DPY-27 ChIP-seq signal was calculated for the 500 bp window upstream of the transcription start sites ( TSSs ) for X-linked genes and compared with the expression level of the genes in control RNAi embryos ( A ) and the change in expression levels between sdc-2 mutant and control RNAi embryos ( B ) . ( A ) DPY-27 has greater occupancy in the promoters of the more highly expressed genes . The scatter plot compares the log2 RPM of the DPY-27 ChIP-seq signal with the log2 ( RPKM ) ( reads per kilobase per million ) of the GRO-seq signal from control RNAi embryos . The two are positively correlated ( Spearman correlation coefficient of 0 . 428 ) . ( B ) The level of DPY-27 occupancy in the promoter of the gene does not predict whether the expression of the gene will change in response to the disruption of dosage compensation . The scatter plot compares the log2 ( RPM ) of the DPY-27 ChIP-seq signal with the log2 ( sdc-2 mutant/control RNAi ) expression difference . No correlation was found ( Spearman correlation coefficient of −0 . 003 ) . RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 032 Severe disruption of sdc-2 function was achieved by treating an sdc-2 partial loss-of-function XX mutant with sdc-2 RNAi [sdc-2 ( y93 , RNAi ) ] ( Figure 5—figure supplement 1 ) . GRO-seq data showed that expression of her-1 ( Figure 5B ) and protein-coding genes on X ( Figure 5C ) were elevated in sdc-2 mutants , while expression of genes on autosomes was mildly reduced ( Figure 5—figure supplement 2A–D ) . her-1 was de-repressed 12 . 7-fold , and the increase in GRO-seq signal was uniform from the promoter to the 3′ end , suggesting that SDC-2 controls sex determination by reducing Pol II recruitment or initiation at the her-1 promoter in XX embryos ( Figure 5B ) . The median change of gene expression in sdc-2 mutant vs control embryos was an increase of 1 . 63- to 1 . 67-fold for X-linked genes of all lengths and a slight reduction of 0 . 79- to 0 . 81-fold for autosome-linked genes ( Figure 5—source data 1 and Figure 4—source data 1 ) . For the subset of X and autosomal genes whose expression was statistically different ( p≤0 . 05 ) between mutant and control embryos , 99% of X-linked genes had elevated expression in the mutants , while 63–66% of autosomal genes had reduced expression ( Figure 5—source data 1 and Figure 4—source data 1 ) . The reduced autosomal gene expression was robust to normalization procedures that counteract potential complications from increased X expression ( see ‘Materials and methods’ ) . Our GRO-seq experiments also provided the first indication that dosage compensation controls the expression of small non-coding RNAs . We found most embryonically expressed X-linked microRNAs to be dosage compensated ( Figure 5D and Figure 5—source data 1 , and ‘Materials and methods’ ) , while X-linked tRNAs were not , implying that RNA polymerase II is broadly sensitive to dosage compensation , and RNA polymerase III is insensitive ( Figure 5—source data 1 ) . Previous studies showed the DCC binds to sequence-specific DNA recruitment sites on X and disperses to promoter regions of actively transcribed genes ( Ercan et al . , 2009; Jans et al . , 2009; Pferdehirt et al . , 2011 ) , but the lack of precise TSS calls had prevented an accurate alignment of DCC binding sites with promoters . Mapping of new TSSs relative to DCC binding sites called from our ChIP-seq experiments performed for this comparison showed the peak of DCC binding to be immediately upstream of the TSS ( Figure 5E ) . Although the DCC binds to promoter regions , prior studies showed that DCC binding to a gene is neither necessary nor sufficient for the compensation of that gene , and not all genes are dosage compensated ( Jans et al . , 2009 ) . We assessed the generality of that conclusion by comparing DCC binding upstream of a TSS to the increase in gene expression caused by disrupting sdc-2 . Our results confirmed and extended the original conclusion ( Figure 5—figure supplement 3A , B ) . Thus , the DCC can act at a distance to control gene expression , and DCC binding intensity is not a proxy for the dosage compensation status of the gene . To determine the step of transcription controlled by the DCC , we compared the distribution of the GRO-seq signal along X and autosomal genes between control and dosage-compensation-defective embryos . The change in Pol II distribution expected from disrupting dosage compensation differs according to the step of transcription affected by the DCC . If the DCC restricts Pol II recruitment , a uniform increase in engaged Pol II would be expected from the promoter to the 3′ end of genes of mutants . If the DCC reduces transcription by preventing the release of Pol II promoter-proximal pausing , or by reducing transcription elongation at stages downstream of the pause site , an increase in the level of engaged Pol II would be expected in the gene body and 3′ end in mutants , with the increase beginning more promoter proximal for a mechanism that controls pausing . Promoter-proximal pausing of transcriptionally engaged Pol II is a rate-limiting step of transcription in metazoans ( Adelman and Lis , 2012 ) observed at ∼ 40% of active genes in mammalian cells ( Core et al . , 2008 ) and more than 60% of active genes in Drosophila ( Core et al . , 2012 ) . To determine whether control of 5′ pausing was a plausible mechanism for dosage compensation , we calculated 5′ pausing ratios ( GRO-seq signal for 5′ end/gene body ) for genes in each developmental stage using TSSs calls for the respective stage ( Figure 4—source data 1 ) . In contrast to Drosophila and mammalian genomes , very few genes ( 0 . 38% , 15 of 3975 ) of wild-type C . elegans embryos showed evidence of 5′ pausing ( Figure 4—source data 1 , Figure 6A , B , and Figure 6—figure supplement 1A–D ) . The paused genes were not enriched on X ( Figure 6C ) . Moreover , the number of 5′ paused genes was not decreased in sdc-2 mutants ( Figure 4—source data 1 and Figure 6D , E ) . 10 . 7554/eLife . 00808 . 033Figure 6 . Promoter-proximal RNA Pol II pausing is rare in Caenorhabditis elegans and is not the target of dosage compensation . ( A ) GRO-seq and GRO-cap signals show that a gene not paused in embryos becomes paused in L1 larvae deprived of food . ( B ) Comparison of average GRO-seq signal from embryos , starved L1 larvae , and L3 larvae within 2 kb of transcription start sites ( TSSs ) called in all three stages shows that promoter-proximal pausing in embryos and L3s is rare compared to that in starved L1 larvae . ( C ) Promoter-proximal pausing is not enriched on the X chromosome relative to autosomes in embryos . If dosage compensation prevented the release of Pol II from promoter-proximal pause sites , there should be higher levels of pausing on the X chromosome . ( D ) and ( E ) The level of promoter-proximal pausing is not decreased in sdc-2 mutants compared to control embryos . If dosage compensation reduced gene expression by preventing the release of Pol II from promoter-proximal pause sites , the sdc-2 mutant should exhibit lower levels of pausing . ( D ) Although X-linked genes have increased expression in sdc-2 mutants , their level of pausing is not decreased . ( E ) The level of pausing displayed by autosomal genes is unchanged in sdc-2 mutants . RPKM: reads per kilobase per million; RPM: reads per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 03310 . 7554/eLife . 00808 . 034Figure 6—figure supplement 1 . The dosage compensation process does not control promoter-proximal pausing of Pol II . ( A ) – ( C ) Shown are plots of the average GRO-seq signal from different developmental stages of wild-type embryos plotted across a 2 kb window centered on the transcription start sites ( TSSs ) identified in the listed developmental stage . The GRO-seq signal is averaged over 25 bp windows . n represents the number of genes in a stage having a TSS identified in that stage . For example , ( A ) shows the average GRO-seq signal for control RNAi embryos , starved L1s , and L3s plotted relative to the distance from the embryo-derived TSSs . Embryos and L3 larvae exhibit lower levels of promoter-proximal pausing than L1 larvae deprived of food , and ( D ) the RNAi process does not change the level of pausing . Promoter-proximal pausing is rare in embryos and therefore unlikely to be the target for the dosage compensation process . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 034 Nonetheless , the GRO-seq assay is capable of detecting 5′ pausing in C . elegans , since genes not paused in embryos become paused in L1s deprived of food ( Figure 6A and Figure 6—figure supplement 1A–C ) . In starved L1s , we found 7 . 7% of genes ( 166 of 2133 ) to exhibit 5′ pausing , and most were on autosomes ( Figure 4—source data 1 ) . A prior genome-wide analysis of Pol II binding in L1 animals cultured with and without food discovered an enrichment of Pol II only at some promoters of food-deprived larvae , consistent with 5′ proximal pausing ( Baugh et al . , 2009 ) . Our results confirm their proposal that Pol II pausing can be induced by food deprivation . These cumulative results make it highly implausible that 5′ pausing control is the mechanism of X-chromosome dosage compensation in embryos , and they are consistent with the lack of a C . elegans negative elongation factor complex ( NELF ) , which contributes , in part , to pausing in other organisms ( Yamaguchi et al . , 1999 ) . Analysis of GRO-seq data comparing transcriptionally engaged Pol II in control vs dosage-compensation-defective XX embryos revealed a uniform increase in the level of engaged Pol II across the entire length of X-linked genes , from TSSs to 3′ ends , in the sdc-2 mutant . This conclusion was reached both by metagene analysis ( Figure 7A ) and by the analysis of individual genes exhibiting a range of overexpression in the mutant ( Figures 5C and 7C , and Figure 7—figure supplement 1 ) . Analysis of individual genes averts complications in interpretations that might arise from averaging of data . 10 . 7554/eLife . 00808 . 035Figure 7 . The DCC condensin complex reduces X-chromosome gene expression in XX embryos by restricting Pol II recruitment to promoters . ( A ) Uniform increase in GRO-seq signal across the length of X-linked genes results from disrupting dosage compensation . Metagene analysis comparing the average GRO-seq signal from 665 X-linked genes ≥1 . 5 kb in control RNAi or sdc-2 mutant embryos . Genes were scaled to the same length as follows: 5′ ends ( −1 kb to +500 bp of the transcription start site [TSS] ) and 3′ ends ( 500 bp upstream to 1 kb downstream of 3′ end ) were not scaled , and the gene body was scaled to 2 kb . The signal was averaged at each base pair and then averaged across 25 bp windows . The GRO-seq signal is elevated approximately 1 . 6-fold across genes in sdc-2 mutant versus control RNAi embryos ( below ) . ( B ) The GRO-seq signal is decreased slightly across autosomal genes in sdc-2 mutant versus RNAi control embryos . Metagene analysis of 2949 autosomal genes ≥1 . 5 kb performed as in ( D ) . The ratio of the GRO-seq signal in mutant versus control embryos is about 0 . 9 . ( C ) Heat map shows that the GRO-seq signal is increased along the length of each X-linked gene in sdc-2 mutants . For each of 665 genes , the GRO-seq signal from mutant or control embryos was summed across 100 bp windows and the sdc-2/control ratio was calculated for each window . The log2 ( sdc-2/control ) value was plotted across the scaled gene . ( D ) Heat map shows that the GRO-seq signal is moderately decreased along the length of individual autosomal genes in sdc-2 mutants . For each of 2949 autosomal genes , the log2 ( sdc-2/control ) value was plotted across the scaled gene as in ( F ) . ( E ) Dosage compensation does not specifically affect Pol II elongation . An elongation density index was calculated for each gene greater than 2 kb in length that did not have another gene on the same strand within 1 kb of the TSS . After excluding the first and last 500 bp of the gene , the average signal across the last 75% of the remaining gene was divided by the average signal across the first 25% of the remaining gene . Ratios of the indices between the sdc-2 mutant and control RNAi embryos are not significantly different for genes on the X compared to the autosomes . Error bars represent a 95% confidence interval for the mean indices . n = 481 ( X ) ; n = 1861 ( autosomes ) . ( F ) Occupancy of hypo-phosphorylated Pol II at the promoters of X-linked genes is increased in dosage compensation mutants , showing greater Pol II recruitment . Comparison of normalized Pol II ChIP-chip signal from control RNAi or sdc-2 mutant embryos relative to newly annotated TSSs of X-linked genes . ( G ) Sense and upstream divergent transcription are coordinately increased for X-linked genes in sdc-2 mutants . Comparison of average sense or antisense GRO-seq signal from sdc-2 mutant and control RNAi embryos for a 3 kb window surrounding TSSs for genes with no divergent gene partner . The GRO-cap signal was only evaluated in this analysis for genes having a log2 ( sense/antisense ) ratio ≤1 . 5 in control RNAi embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 03510 . 7554/eLife . 00808 . 036Figure 7—figure supplement 1 . GRO-seq signal is increased along the length of individual X-linked genes when dosage compensation is disrupted . The GRO-seq signal from sdc-2 mutant vs control RNAi embryos is shown for a representative set of 27 X-linked genes . The first vertical line in each gene shows the location of the transcription start site ( TSS ) , and the second vertical line shows the location of the 3′ end . The GRO-seq signal has been averaged across 100 bp windows . The ratio of sdc-2 expression vs control expression and the gene name are shown above each plot . Previously , we found that not all genes on X are dosage compensated ( Jans et al . , 2009 ) , and these GRO-seq data are consistent with that finding . RPKM: reads per kilobase per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 03610 . 7554/eLife . 00808 . 037Figure 7—figure supplement 2 . Disruption of dosage compensation causes a uniform increase in GRO-seq signal across the length of X-linked genes in different quartiles of gene expression determined from control RNAi samples . ( A ) – ( D ) Metagene analyses comparing the average GRO-seq signal from X-linked genes ≥1 . 5 kb in control RNAi and sdc-2 mutant embryos . The 665 X-linked genes have been split into four quartiles of gene expression and plotted independently . Genes were scaled to the same length as follows: the 5′ end ( 1 kb upstream to 500 bp downstream of the TSS ) and the 3′ end ( 500 bp upstream to 1 kb downstream of the termination site ) were not scaled , and the remainder of the gene was scaled to a length of 2 kb . Signal was averaged across the genes of each group at each base pair and then averaged across 25 bp windows . The ratio of GRO-seq signal in mutant vs control embryos is plotted below each metagene analysis . RPKM: reads per kilobase per million; RPM: reads per million; TSS: transcription start site . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 03710 . 7554/eLife . 00808 . 038Figure 7—figure supplement 3 . Disruption of dosage compensation causes a uniform increase in GRO-seq signal across the length of X-linked genes of different size ranges . ( A ) – ( C ) Shown are metagene analyses comparing average GRO-seq signal from X-linked genes ≥1 . 5 kb in control RNAi or sdc-2 mutant embryos . The 665 X-linked genes have been split into three size ranges ( 1 . 5–3 . 0 kb , 3 . 0–6 . 0 kb , and >6 kb ) and plotted independently as in Figure 7A . RPKM: reads per kilobase per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 03810 . 7554/eLife . 00808 . 039Figure 7—figure supplement 4 . The level of antisense transcription is unaffected by dosage compensation . ( A ) Sense and upstream divergent transcription are coordinately increased for X-linked genes in sdc-2 mutants . Scatter plot comparing the log2 ( sense/antisense ) ratios for sdc-2 mutant and control RNAi embryos shows that the ratios are similar , indicating that sense and antisense transcription is coordinately increased in sdc-2 mutants . In combination with GRO-seq and ChIP-chip data , this result supports the view that recruitment of Pol II is elevated in sdc-2 mutants . The statistical relationship between the replicates is indicated by the Spearman correlation coefficient ρ . The red line depicts a 1:1 relationship between the ratios . ( B ) Occupancy of hypo-phosphorylated Pol II at the promoters of X-linked genes that do not have a divergent gene pair yet have antisense transcription is increased in dosage compensation mutants , showing greater Pol II recruitment . Comparison of normalized Pol II ChIP-chip signal from control RNAi or sdc-2 mutant embryos relative to newly annotated transcription start sites ( TSSs ) of X-linked genes . The ChIP-chip signal was only evaluated in this analysis for genes having a log2 ( sense/antisense ) ratio ≤1 . 5 in control RNAi embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 039 For the metagene analysis , the ratio of engaged polymerase in mutant vs control was consistently elevated by at least 1 . 6-fold across scaled X-linked genes , from TSSs to 3′ ends ( Figure 7A ) . This elevation in the GRO-seq signal was also readily apparent in metagene analyses of X-linked genes subdivided by quartiles of expression ( Figure 7—figure supplement 2 ) or by gene length ( Figure 7—figure supplement 3 ) . In contrast , metagene analysis of autosomal genes from the same data sets showed only a slight decrease in engaged Pol II across their lengths ( Figure 7B ) , consistent with the slight decrease in autosomal gene expression in mutants ( Jans et al . , 2009 ) . Heat maps displaying the log2 ratio of mutant to control GRO-seq signal across individual X-linked ( Figure 7C ) and autosomal ( Figure 7D ) genes confirmed the uniform elevation of transcriptionally engaged Pol II across X-linked genes and the reduction across individual autosomal genes of sdc-2 mutants . Together , these results demonstrate that the dosage compensation mechanism reduces X-linked gene expression in C . elegans hermaphrodites by restricting either the recruitment or initiation of Pol II . This conclusion is strongly supported by the analysis of elongation density indices ( average GRO-seq signal across the last 75% of genes/average signal across the first 25% ) in mutant vs control embryos for genes on X and autosomes . The ratios of indices between control and mutant embryos were not significantly different for genes on X compared to autosomes , indicating that dosage compensation does not preferentially affect Pol II elongation ( Figure 7E ) . Two further lines of evidence show Pol II recruitment to be the predominant step of transcription targeted by the dosage compensation process . First , re-analysis of our previous genome-wide ChIP-chip analysis of Pol II binding ( Pferdehirt et al . , 2011 ) relative to the new TSSs showed an approximately twofold increase in the occupancy of hypo-phosphorylated Pol II at the promoters of X-linked genes in sdc-2 mutant vs control embryos ( Figure 7F ) . DNA-bound hypo-phosphorylated Pol II at promoters is enriched for non-initiated Pol II . The increase in Pol II occupancy at promoters of mutants ( measured by ChIP ) was nearly equivalent to the increase in post-initiated Pol II ( measured by GRO-seq ) , indicating that Pol II promoter recruitment is rate-limiting when dosage compensation is active . Second , in sdc-2 mutants , the levels of sense and upstream divergent transcription were elevated coordinately at promoters on X lacking upstream divergent genes ( Figure 7G and Figure 7—figure supplement 4A ) . At those promoters , the level of hypo-phosphorylated Pol II was also elevated about twofold in mutant vs control embryos , as assayed by ChIP ( Figure 7—figure supplement 4B ) . Since we found increased Pol II occupancy in mutants and observed virtually no promoter-proximal pausing in control embryos , escape from pausing into productive elongation , or any other form of post-initiation regulation , cannot be the cause of elevated transcription at the divergent promoters in sdc-2 mutants . Furthermore , an increase in transcription initiation by bound polymerases cannot account for these results , but an increase in Pol II recruitment can . Our combined experiments demonstrate that C . elegans dosage compensation controls X-chromosome-wide gene expression predominantly by reducing Pol II recruitment to promoters of hermaphrodites . In organisms that equalize X-chromosome gene expression between the sexes by a dosage compensation process , the question arises as to whether the compensated level of X-chromosome expression is equivalent to or half of the expression from the two sets of autosomes . The answer to this question has been controversial , although evidence has mounted in favor of a mechanism to balance total expression between X chromosomes and the two sets of autosomes based on measurements of accumulated transcripts ( Xiong et al . , 2010; Deng et al . , 2011; Disteche , 2012; Lin et al . , 2012; Deng et al . , 2013; Jue et al . , 2013 ) . Our GRO-seq experiments have addressed this question in the most definitive way to date by measuring the levels of nascent transcripts prior to co-transcriptional processing . We show that in wild-type C . elegans embryos , X and autosomes have nearly equivalent levels of total gene expression , and that levels of transcribing Pol II are uniform across X and autosomal genes ( Figure 8A ) . In dosage-compensation-defective mutants , the level of X expression and engaged Pol II exceeds that of autosomes by 1 . 7-fold , from the TSSs to the 3′ ends ( Figure 8B ) . These results demonstrate the existence of a separate mechanism in C . elegans to elevate the intrinsic rate of transcription from the X chromosomes of both sexes , so that after dosage compensation , X chromosomes and the two sets of autosomes have equivalent expression . Our experiments provide evidence that it , like dosage compensation , works at the level of controlling Pol II recruitment . 10 . 7554/eLife . 00808 . 040Figure 8 . Gene expression is balanced between X chromosomes and autosomes . ( A ) Caenorhabditis elegans has a mechanism to equalize expression between X chromosomes and autosomes . Metagene analysis comparing the average GRO-seq signal from X-linked and autosome-linked genes of control RNAi embryos . The X to autosome expression ratio is 0 . 9 . ( B ) In dosage-compensation-defective mutants , the level of X-chromosome expression exceeds that of autosomes by 1 . 7-fold . Metagene analysis comparing the average GRO-seq signal from X-linked and autosome-linked genes of sdc-2 mutant embryos . RPKM: reads per kilobase per million . DOI: http://dx . doi . org/10 . 7554/eLife . 00808 . 040
GRO-cap has advantages over TSS mapping strategies that use total accumulated RNA or short-capped ( sc ) RNA as the starting material . The use of nascent RNAs without base hydrolysis or size selection , as in GRO-cap , enriches the proportion of 5′ capped RNAs within the starting RNA population , reduces the level of false TSS calls from RNAs that are capped post-transcriptionally ( Fejes-Toth et al . , 2009 ) , and increases the probability of identifying TSSs from promoters that are transcribed at low levels . Most importantly , the GRO-cap strategy permits reliable assignment of TSSs by pairing TSS calls with uninterrupted GRO-seq signal for transcriptionally engaged Pol II between the GRO-cap TSS and the previously annotated 5′ end . Multiple lines of evidence showed that GRO-cap signal combined with continuous GRO-seq signal from WB starts defines true TSSs . Two reports of nematode TSSs have recently been published . The first used two approaches to identify the transcription starts ( Gu et al . , 2012 ) . The first , called CapSeq , used total accumulated RNA as the starting material for the enzymatic enrichment of RNAs with 5′ caps and set the cutoff for TSS calls to be within 1 kb upstream and 100 bp downstream from previously annotated 5′ ends . As a consequence , CapSeq did not identify TSSs for a large class of protein-coding genes . TSSs of small processed non-coding RNAs were also difficult to identify by CapSeq . The second approach , called CIP-TAP , used scRNAs and was equivalently effective as GRO-cap for identifying TSSs of small non-coding RNAs . The second report also used scRNAs to map TSSs and set the cutoff for TSS calls to be no greater then 200 bp upstream of the previously annotated 5′ ends , thereby also not defining TSSs for a large class of genes ( Chen et al . , 2013 ) . The authors found numerous clusters of Pol II initiation upstream of their calls and classified many of these initiation events as enhancer-like chromatin signatures based on overlap with bound transcription factors . Based on comparison with our data ( e . g . , Figure 1—figure supplement 12A , B ) , we propose that a proportion of the upstream enhancer-like signatures are TSSs giving rise to full-length transcripts . Two clear examples of genes are shown in Figure 1—figure supplement 12B , C , where the single TSS ( either 2534 bp or 2878 bp upstream of the WB start ) called from GRO-cap and GRO-seq data was classified as an enhancer by Chen et al . ( 2013 ) . Although 5′ promoter-proximal pausing in metazoans is a common rate-limiting step of transcription that is highly regulated to control gene expression ( Adelman and Lis , 2012 ) , we found it to be rare in C . elegans under normal growth conditions . However , in L1 larvae deprived of food , we found 5′ pausing at 7 . 7% of genes with TSS calls . A prior study using ChIP-seq discovered the accumulation of Pol II at promoters of starved larvae and proposed that 5′ promoter-proximal pausing was responsible for Pol II accumulation ( Baugh et al . , 2009 ) . Our results confirm that interpretation . Two factors promote 5′ pausing in metazoans , negative elongation factor ( NELF ) and DRB sensitivity-inducing factor ( DSIF ) , although the relative contribution of each is not fully understood ( Adelman and Lis , 2012; Yamaguchi et al . , 2013 ) . C . elegans appears to lack NELF , suggesting either that DSIF is sufficient in the infrequent cases of pausing or that the core promoter complex ( Kwak et al . , 2013 ) or another not-yet identified negative elongation factor participates . The 3′ accumulation of Pol II has been documented previously in flies and humans using GRO-seq ( Core et al . , 2008 , 2012 ) . We found 3′ Pol II accumulation to be more extensive in nematodes than flies . Pol II 3′ accumulation is likely caused by slow 3′ end formation and RNA processing ( Gromak et al . , 2006 ) . Because C . elegans has numerous polycistronic operons requiring extensive RNA processing to produce monocistronic mRNAs ( Blumenthal , 2012 ) , we were able to discover a strong positive correlation between the amount of 3′ Pol II accumulation and the amount of RNA processing . Curiously , though , even genes that were not part of operons exhibited greater 3′ accumulation than genes in flies . This observation raises the question of whether RNA processing in C . elegans is generally slower than that in other organisms to accommodate extensive trans-splicing , or whether C . elegans accommodates operons by imposing additional regulation on Pol II to enhance its 3′ pausing at all genes to assess whether to halt transcription or continue elongation . Gene expression in metazoans is controlled by diverse regulatory mechanisms that function over widely different distances . Some mechanisms act locally on individual genes , while others such as dosage compensation function across large chromosomal territories or along entire chromosomes to regulate a large set of genes coordinately . In general , the step of transcription controlled by long-range mechanisms is not understood . In our study , multiple lines of evidence supported the conclusion that C . elegans dosage compensation regulates gene expression along X primarily by reducing the recruitment of RNA Pol II to the promoters of hermaphrodite X-linked genes . First , regulation of 5′ promoter-proximal pausing cannot be the mechanism underlying C . elegans dosage compensation . If the DCC reduced X-chromosome gene expression by increasing 5′ promoter-proximal pausing , numerous genes on X would be paused in wild-type embryos , and disruption of dosage compensation would reduce the level of pausing . GRO-seq experiments showed instead that 5′ promoter-proximal pausing is rare in wild-type XX embryos . The few genes that exhibited 5′ pausing were not enriched on X chromosomes relative to autosomes , and the level of pausing was not decreased in dosage-compensation defective mutants . Second , Pol II-mediated transcription elongation is not preferentially affected by the DCC . The ratios of elongation density indices ( average GRO-seq signal in last 75% of a gene/average GRO-seq signal in first 25% of a gene ) calculated for genes on X chromosomes and autosomes in control vs dosage-compensation-defective embryos were not significantly different between X-linked and autosomal genes , indicating that Pol II-mediated transcription elongation is not selectively changed on X chromosomes of mutants . Third , the level of transcriptionally engaged Pol II assayed genome-wide by GRO-seq in control vs dosage-compensation-defective XX embryos revealed a uniform increase in engaged Pol II across the entire length of X-chromosome genes , but not autosomal genes , in mutants . Hence , the DCC controls X-chromosome gene expression in XX animals by reducing either the recruitment or initiation of Pol II . This conclusion was validated both by metagene analysis and by analysis of hundreds of individual X-linked genes showing different levels of de-repression in dosage-compensation-defective mutants . Fourth , genome-wide quantification of Pol II occupancy by ChIP plotted relative to the new TSSs showed an increase in the hypo-phosphorylated form of Pol II at promoters of dosage-compensation-defective embryos vs control embryos that was equivalent to the increase in post-initiated Pol II measured by GRO-seq . Since DNA-bound hypo-phosphorylated Pol II at promoters is enriched for non-initiated Pol II , these results indicate that Pol II promoter recruitment is rate limiting when dosage compensation is active . Our combined experiments reveal that the primary mechanism by which the C . elegans dosage compensation process reduces X-chromosome gene expression by half in XX embryos is to limit Pol II recruitment to promoters of X-linked genes . Our study does not eliminate the possibility of a minor repressive influence acting through another step of transcription or through a post-transcriptional mechanism such as RNA stability . How might condensin reduce Pol II recruitment to the promoters of X-linked genes by approximately twofold ? Our current and prior ( Jans et al . , 2009 ) studies showed that DCC binding to the promoter of a gene is neither necessary nor sufficient to elicit repression of the gene . Hence , the DCC influences gene expression over long distance , likely by imposing changes in higher-order chromosome structure . Clues to such a DCC function were suggested originally by the simultaneous discovery of the DCC and condensin’s biochemical properties in vitro as an ATPase that alters DNA topology ( Chuang et al . , 1994; Kimura and Hirano , 1997; Hagstrom et al . , 2002; Hirano , 2012; Piazza et al . , 2013 ) and its canonical roles in vivo of compacting and resolving mitotic and meiotic chromosomes for proper chromosome segregation ( Hirano and Mitchison , 1994; Hagstrom et al . , 2002; Chan et al . , 2004; Hirano , 2012; Piazza et al . , 2013 ) . However , mechanisms of DCC function are perhaps best deduced from its non-canonical roles in vivo of regulating interphase chromosome structure ( Bauer et al . , 2012 ) and meiotic crossover recombination ( Mets and Meyer , 2009; Wood et al . , 2010; Aragon et al . , 2013 ) . Condensin II in Drosophila induces axial compaction of interphase chromosomes , globally disrupts inter-chromosomal interactions , and promotes dispersal of peri-centric heterochromatin ( Bauer et al . , 2012 ) . These activities serve to compartmentalize the interphase nucleus into discrete chromosomal territories . Furthermore , nematodes carrying mutations that disrupt condensin I , which shares four subunits with the DCC , display elongated chromosomal axes during meiotic prophase and exhibit chromosome-wide changes in the distribution of double strand breaks and crossovers ( Mets and Meyer , 2009 ) , providing strong evidence that DCC subunits control intra-chromosomal structure . These cytological , biochemical , and genetic observations of condensin function suggest that chromosome structure might affect Pol II recruitment in several ways , which are not mutually exclusive . Chromosome compaction and associated topological changes in DNA could broadly affect promoter accessibility of Pol II and the regulatory factors that recruit or stabilize it , thereby reducing Pol II recruitment in a quantitatively similar manner at different sites . Compaction of interphase chromosomal territories could reduce the local concentration of bound transcription activators and hence bound Pol II . Changes in intra-chromosomal interactions could alter the relationships between distal regulatory regions and their target promoters , thereby limiting Pol II recruitment . These and other models are topics of future studies . Dosage compensation strategies differ across species . Mammals inactivate one of the two female X chromosomes , flies double expression of the single male X chromosome , and nematodes halve expression of both hermaphrodite X chromosomes . A central question is whether the molecular mechanisms underlying these diverse forms of chromosome-wide transcriptional regulation are the same or different . In Drosophila melanogaster , dosage compensation is achieved by the male-specific lethal ( MSL ) complex that binds along the single X chromosome of males to double transcription of X-linked genes ( Gelbart and Kuroda , 2009; Conrad and Akhtar , 2012 ) . The complex contains two long non-coding RNAs and five proteins , including the H4K16 histone acetyltransferase MOF ( Hilfiker et al . , 1997; Conrad et al . , 2012a ) and the H2BK34 ubiquitin ligase MSL2 ( Wu et al . , 2011 ) . The MSL complex was proposed to regulate gene expression by controlling transcription elongation ( Smith et al . , 2001 ) . This model was subsequently supported by genome-wide mapping of MSL proteins and MSL-dependent H4K16ac to the bodies of male X-linked genes , with a bias toward 3′ ends ( Alekseyenko et al . , 2006; Gilfillan et al . , 2006; Legube et al . , 2006 ) , and GRO-seq experiments in S2 cells with or without RNAi of the msl2 gene ( Larschan et al . , 2011 ) . While a recent study analyzing genome-wide Pol II occupancy in males and females by ChIP-seq experiments suggested an alternative view that fly dosage compensation operates at the level of Pol II recruitment or initiation ( Conrad et al . , 2012b ) , a mathematical computation error discovered in this study ( Ferrari et al . , 2013; Straub and Becker , 2013 ) rendered the results insufficient to distinguish between the competing models . The elongation model received recent further support from the discovery that dosage compensation is disrupted by impairing the function of SPT5 , a transcription elongation factor that co-localizes with the MSL complex on male X chromosomes and interacts physically with the MSL complex ( Prabhakaran and Kelley , 2012 ) . Thus , the weight of evidence strongly favors enhancement of transcription elongation as the primary mechanism of fly dosage compensation . Not only does the overall dosage compensation strategy differ between worms and flies , the underlying molecular mechanism appears to differ as well . In worms , reduction in X-chromosome gene expression is primarily achieved by reducing recruitment of Pol II to promoters , while in flies , elevation in X-chromosome gene expression is primarily achieved by facilitating Pol II transcription elongation . Multiple solutions have evolved to coordinately control gene expression across an entire chromosome . In many species , the evolution of sex chromosomes to be the primary determinants of sexual fate resulted in males having one X chromosome and females having two . Such chromosome sex-determining mechanisms had the potential to cause two problems in gene expression , an imbalance in X-chromosome gene expression between the sexes and an imbalance in gene expression between the male X chromosome and his two sets of autosomes ( Disteche , 2012 ) . X-chromosome dosage compensation strategies such as the Drosophila strategy solved both problems by doubling transcription of the single male X chromosome . An alternative solution would be to co-evolve two mechanisms , one to increase expression of X chromosomes in both sexes , thereby preventing hypo X expression in males , and a second to decrease total expression from female X chromosomes to prevent hyper X expression relative to female autosomes and male X chromosomes . Nematodes and mammals evolved the second strategy for equalizing X-chromosome expression between the sexes . It has been controversial whether these organisms also evolved a strategy to elevate X expression in both sexes and thereby balance expression between X chromosomes and autosomes ( Xiong et al . , 2010; Deng et al . , 2011; Disteche , 2012; Lin et al . , 2012; Deng et al . , 2013; Jue et al . , 2013 ) . The bulk of evidence now favors the presence of a mechanism to up-regulate X-chromosome expression in males and females of both organisms ( Deng et al . , 2011 , 2013; Jue et al . , 2013 ) , and our results provide the most compelling evidence to date for C . elegans . Our approach of examining nascent transcripts in mixed-stage embryos by GRO-seq conferred three advantages . First , it enabled us to quantify transcription specifically from somatic embryonic cells and thereby avert the complication of quantifying maternally contributed germline transcripts that contaminate mature embryo mRNA . Second , we could examine nascent transcription at a stage for which most germline-specific mechanisms of gene regulation would have been erased . Third , we could quantify transcriptionally engaged Pol II across the entire length of a gene , starting from the bona fide TSS , and assess the step of transcription controlled by the process . We found that a major part of the mechanism to increase X expression in C . elegans is to increase the level of Pol II recruitment to promoters , as was shown recently for mammals ( Deng et al . , 2013 ) . Together , these results reinforce the evolutionary importance of balancing gene expression across all chromosomes of a genome .
The following nematode strains were used:Wild-type strain N2sdc-2 mutant , sdc-2 ( y93 , RNAi ) Control RNAi , N2 ( L4440 RNAi ) . Bacteria for use in feeding RNAi were prepared by growing HT115 bacteria bearing an RNAi plasmid ( sdc-2 or the L4440 negative control ) overnight in TB and ampicillin , inducing for 2 hr with 1 mM IPTG , pelleting , and resuspending in 1 vol ( wt/vol ) of LB with 20% glycerol . For RNAi treatment , embryos were harvested from gravid hermaphrodites and allowed to hatch off for 24 hr . NG agar plates supplemented with 1 mM IPTG and 100 μg/ml carbenicillin were spotted with super-induced RNAi bacteria and allowed to induce further overnight at 25°C . Hatched off L1 larvae were spotted on the RNAi plates and grown at 20°C until gravid . For wild-type samples , embryos were harvested from gravid hermaphrodites grown at 20°C on concentrated HB101 . Embryos were allowed to hatch off and starve for 24 hr at 20°C . Worms were fed and grown to L3 stage under liquid culture ( 1 worm/μl; 10 mg/ml HB101 ) for 34 hr at 20°C . Animals were collected from whichever stage was desired . After washing twice with M9 buffer , animals were washed with cold nuclear isolation buffer ( 250 mM sucrose , 10 mM Tris-HCl ( pH 7 . 9 ) , 10 mM MgCl2 , 1 mM EGTA , 0 . 25% NP-40 , 1 mM DTT , protease inhibitors , 4 U/ml SUPERaseIn [AM2696; Ambion , Grand Island , NY] ) . Animals were resuspended in nuclear isolation buffer ( embryos and starved L1 in 3 vol , L3 in 1 vol ) , and dripped into liquid nitrogen to freeze . Starved L1 and L3 samples were ground under liquid nitrogen by mortar and pestle . Larval samples , post-grinding , and embryo samples were dounced with a Kontes 2 ml glass dounce to release nuclei . Douncing and collection of nuclei was performed for up to six rounds as follows: dounce with 10X pestle A , 10X pestle B , 5 min centrifugation at 100×g , removal of nuclei-containing supernatant , and addition of an equal volume of nuclear isolation buffer to the pellet . Nuclear isolation was monitored each round to determine effectiveness and when it was complete . The pooled supernatant was centrifuged for 5 min at 1000×g to pellet nuclei . The nuclear pellet was washed with nuclear freezing buffer ( 40% glycerol , 50 mM Tris-HCl ( pH 8 . 3 ) , 0 . 1 mM EDTA , 5 mM MgCl2 , 1 mM DTT , protease inhibitors , 4 U/ml SUPERaseIn ) . Approximately 1 × 108 nuclei were resuspended in 100 µl nuclear freezing buffer and stored at −80°C until GRO-seq reactions were performed . All NRO reactions and bead enrichment steps for GRO-cap were carried out as described above , with the exception that the RNA for GRO-cap was not base hydrolyzed . After the first bead binding , the TruSeq RNA 3′ Adapter ( RA3 ) ( Illumina part # 15013207 ) was ligated to the 3′ end of the NRO RNA . First , 50 pmol of the 3′ adapter were mixed with NRO RNA and 2 μl 50% PEG 8000 , brought to 14 μl with DEPC water , incubated at 70°C for 3 min and put on ice for 2 min . Then , 2 μl of 10× T4 RNA Ligase I buffer , 2 μl 10 mM ATP , 1 μl SUPERaseIN , and 1 . 5 μl T4 RNA Ligase I ( M0204; NEB ) were added and reaction incubated at 22°C for 4–6 hr . The reaction was then brought to 100 μl with binding buffer and subjected to a second round of bead enrichment . After the second bead enrichment , 5′ mono-phosphate RNAs were selected against in two successive steps . First , to selectively degrade RNAs with a 5′ mono-phosphate , NRO RNA was resuspended in 16 . 5 μl DEPC water , mixed with 0 . 5 μl SUPERaseIN , 2 μl 10× Terminator reaction buffer A , and 1 μl of Terminator 5′ phosphate-dependent exonuclease ( TER51020; Epicentre ) , and incubated at 30°C for 1 hr . The reaction was extracted and precipitated using the standard method ( above ) , and resuspended in 10 μl DEPC water . Second , 5′ mono-phosphate RNAs were dephosphorylated to prevent their participation in subsequent ligation reactions . For this , the RNA was then mixed with 1 μl SUPERaseIN , 14 . 5 μl DEPC water , 10× Antarctic Phosphatase buffer , 1 . 5 μl of Antarctic Phosphatase ( M0289S; NEB ) , and incubated at 37°C for 30 min . The reaction was brought to 200 μl with 10 mM Tris-HCl ( pH 7 . 5 ) , 5 mM EDTA , and heat inactivated at 65°C for 5 min . The reaction was then extracted and precipitated using the standard extraction method and resuspended in 10 μl DEPC water . The 5′ capped RNAs then were prepared for ligation and the final library preparation steps . The NRO RNAs were then split in half and the experimental sample was treated with tobacco acid pyrophosphatase ( TAP ) : 1 μl SUPERaseIN , 15 μl DEPC water , 3 μl 10× TAP buffer , and 1 μl TAP ( T19500; Epicentre ) , with incubation at 37°C for 1 hr . The control reaction was treated identically except for the addition of TAP . The reaction was brought to 200 μl and then extracted and precipitated using the standard method . The TruSeq RNA 5′ Adapter ( RA5 ) ( Illumina part # 15013205 ) was ligated to the 5′ end of the NRO RNA . First , 50 pmol of the 5′ adapter were mixed with NRO RNA , and 2 μl 50% PEG 8000 , brought to 14 μl with DEPC water , incubated at 70°C for 3 min and put on ice for 2 min . Then , 2 μl of 10× T4 RNA Ligase I buffer , 2 μl 10 mM ATP , 1 μl SUPERaseIN , and 1 . 5 μl T4 RNA Ligase I ( M0204; NEB ) were added and reaction incubated at 22°C for 4–6 hr . The reaction was then brought to 100 μl with binding buffer and subjected to a third round of bead enrichment . After the third enrichment , samples were reverse transcribed , amplified and PAGE purified as described ( Core et al . , 2008 ) , and quantified before submission for sequencing . Libraries were sequenced with Illumina’s HiSeq 2000 platform . Reads were required to have passed the CASAVA 1 . 8 quality filtering to be considered further . To remove reads containing the RT-PCR adapter , we used cutadapt version 0 . 9 . 5 ( http://code . google . com/p/cutadapt/ ) with the following command ( cutadapt -a TGGAATTCTCGGGTGCCAAGG -a AAAAAAAAAAAAAAAAAAAA -z -O 15 -e . 1 --minimum-length=30 ) . The remaining reads were trimmed to 30 bp in length and aligned uniquely to the C . elegans WS230 genome using bowtie’s default settings ( version 0 . 12 . 7 ) , which permit two mismatches in the first 28 bp . Because the 5’ base in each read most closely identifies the location of transcriptionally engaged RNA polymerase prior to the run-on , we created pile-ups using only the first base of each read . For GRO-seq , the pile-up of reads mapping to both strands was normalized by the number of millions of reads that mapped uniquely to the genome and was multiplied by 1000 to obtain RPKM ( reads per kilobase per million ) . For GRO-cap , reads were normalized by the same method but were not multiplied by 1000 , hence RPM ( reads per million ) . As expected , the GRO-cap signal is enriched at TSSs compared to the GRO-seq signal . Further analysis of our GRO-seq data using an alternative normalization strategy confirmed that X-linked genes are increased in expression , and autosomal genes are decreased in expression in the sdc-2 mutant . This new analysis evaluated the original normalization strategy . Because expression of genes on the X chromosome is elevated in sdc-2 mutants , it is difficult to determine whether the autosomes are changed in expression in the sdc-2 mutant . The proportion of autosomal reads relative to total reads per experiment is lower in the sdc-2 mutant ( 75 . 9% ) than in the control RNAi ( 85 . 7% ) . To normalize the autosomal expression between the two conditions , we divided the average expression of all genes in the sdc-2 mutant by the scaling factor of this difference ( 75 . 9%/84 . 715% = 0 . 885 ) . We then investigated changes in the sdc-2 mutant compared to the control for our most inclusive gene set: WB WS230 genes that are >250 bp in length ( see Figure 5—source data 1 ) . With this alternative normalization , the median sdc-2/control gene expression ratios increase for both X ( 1 . 67 becomes 1 . 88 ) and autosomes ( 0 . 81 becomes 0 . 91 ) . This procedure changed the proportion of X-linked genes that are more highly expressed in the mutant ( genes increased in expression by 1 . 5-fold or greater ) from 64 . 9% to 78 . 4% . The original normalization showed that fewer autosomal genes increase by 1 . 5-fold than decrease by 1 . 5-fold ( 5 . 2% compared to 26 . 7% , respectively ) , and the alternative normalization shows the same trend ( 8 . 4% compared to 16 . 5% ) . Therefore , the same conclusions about X and autosomal gene expression can be made with either normalization strategy . We computationally identified all 30mers in the WS230 genome . After passing these sequences through the cutadapt parameters outlined above , we aligned the remaining sequences uniquely using bowtie . We mapped the 5′ base to determine whether a read beginning at that base pair can be aligned uniquely . Genome annotation files were downloaded from WB . Protein coding genes that were labeled as ‘Coding_transcript’ and ‘gene’ were extracted from the WS230 annotation file . Genes encoding microRNAs and snoRNAs that were labeled as ‘miRNA_primary_transcript’ and ‘snoRNA_mature_transcript’ , respectively , were extracted from the WS225 annotation file . tRNAs predicted by ‘tRNAscan-SE-1 . 23’ were extracted from the WS230 annotation file . The WB remap and unmap tools were used to convert DNA coordinates between releases , thereby ensuring that all analyses matched the correct genome versions . To determine whether dosage compensation specifically affects transcription elongation across the X chromosome , we determined an elongation density index for each gene with a newly annotated TSS . We calculated the average GRO-seq signal across the last 75% of the gene and divided it by the average GRO-seq signal across the first 25% of the gene , excluding the first and last 500 bp of the gene from this calculation . To ensure that we averaged the GRO-seq signal over a sufficiently large number of bases , we required that genes be ≥2 kb in length . To avoid outlier ratios that can result from a low number of reads , genes with an average RPKM <1 in the first 500 bp , or the first 25% or last 75% of the remaining gene were excluded from the analysis . To reduce the possibility that the elongation density index was influenced by the 3′ accumulation of Pol II of an upstream gene , we required that genes lack another gene on the same strand within 1 kb of the TSS . We analyzed 481 X-linked genes and 1861 autosomal genes . To compare GRO-seq signal across genes , we scaled genes to be the same length , allowing us to average the GRO-seq signal across them . To avoid small genes that could affect the sensitivity of our analyses , we required that genes be ≥1 . 5 kb in length . These genes were scaled to the same length as follows: the 5′ end ( 1000 bp upstream to 500 bp downstream of the TSS ) and the 3′ end ( 500 bp upstream to 1000 bp downstream of the WB stop site ) were not scaled , and the remainder of the gene was scaled to a length of 2 kb . We predicted that leaving the ends of the gene unscaled might allow us to better identify any effects that occurred at the ends of genes . To investigate GRO-seq trends surrounding the TSS or across scaled metagenes , we plotted the average GRO-seq signal across these regions of interest . To do so , we totaled the strand-specific GRO-seq signal for every gene in the gene list at each base pair in the region . We then divided the total GRO-seq signal at each base pair by the number of genes that are uniquely mappable at that base pair . We then took a 25 bp moving window average of this average GRO-seq signal . We used the Python package matplotlib to produce heat maps showing the GRO-seq expression of the TSSs of all genes , and the difference in expression between DCC mutant and control RNAi embryos . To show the GRO-seq signal at either the empirically determined TSS or WB start , the signal from each gene with a TSS call was plotted , one gene per row , and the GRO-seq signal was averaged across 15 bp windows . Genes were ordered from top to bottom with increasing distance between the WB start and the TSS called from GRO-cap . To show expression changes in the DCC mutant , we scaled each gene ≥1 . 5 kb to the same length as described above . We then totaled the GRO-seq signal from the DCC mutant embryos and separately from control RNAi embryos across 100 bp windows and calculated an average . The log2 of the ratio was plotted for every 100 bp bin across the length of the metagene . To calculate the relative level of upstream divergent transcription for each gene , we determined the maximum sense and antisense transcription in a 150 bp window within 500 bp of the TSS . Upstream divergent transcript data from Human and Drosophila samples were obtained ( Core et al . , 2012 ) . We used R to plot kernel density estimations of the log2 ( sense/antisense ) ratio for C . elegans , Human , and Drosophila . To determine how far upstream the closest upstream divergent gene was to each TSS , we searched for the closest transcript ( protein coding , non-coding RNA , tRNA , or rRNA ) upstream and antisense to the TSS . Prior to the search , the 6353 protein coding genes with new TSS calls were re-annotated with the most upstream TSS identified . To determine the upstream distance between the TSS and the start of upstream divergent transcription , we searched 500 bp upstream of the TSS to identify the position with the highest level of GRO-cap TAP+ minus the TAP− signal . Wild-type N2 animals were grown on NG agar plates with HB101 bacteria . Mixed-stage embryos were harvested from gravid hermaphrodites , and cross-linked with 2% formaldehyde for 30 min . Cross-linked embryos were resuspended in 3 ml of FA buffer ( 150 mM NaCl , 50 mM HEPES-KOH ( pH 7 . 6 ) , 1 mM EDTA , 1% Triton X-100 , 0 . 1% sodium deoxycholate , 1 mM DTT , protease inhibitor cocktail ) for every 1 g of embryos . This mixture was frozen on liquid nitrogen , then ground under liquid nitrogen by mortar and pestle . Chromatin was sheared by the Covaris S2 sonicator ( 20% duty factor , power level 8 , 200 cycles per burst ) for a total of 30 min processing time ( 60 s ON , 45 s OFF , 30 cycles ) . To perform the ChIP reactions , extract containing approximately 2 mg of protein was incubated in a microfuge tube with 6 . 6 μg of anti-DPY-27 or random IgG antibodies overnight at 4°C . A 25 μl bed volume of protein A Sepharose beads was added to the ChIP for 2 hr . ChIPs were washed for 5 min at room temperature twice with FA buffer ( 150 mM NaCl ) , once with FA buffer ( 1 M NaCl ) , once with FA buffer ( 500 mM NaCl ) , once with TEL buffer ( 10 mM Tris-HCl ( pH 8 . 0 ) , 250 mM LiCl , 1% NP-40 , 1% sodium deoxycholate , 1 mM EDTA ) , and twice with TE buffer . Protein and DNA were eluted twice with 1% SDS , 250 mM NaCl , 1 mM EDTA at 65°C for 15 min . After elution , sequencing libraries were prepared as published ( Zhong et al . , 2010 ) with minor changes: sequencing adapters were as described ( Lefrancois et al . , 2009 ) and adapters were ligated using the NEB Quick Ligation Kit ( M2200 ) . Libraries were sequenced on the Illumina GA2 platform . After barcode removal , 32 bp reads were aligned uniquely to the C . elegans WS190 genome using bowtie . MACS ( version 1 . 4 ) was used to call peaks and create pileups with DPY-27 ChIP as the treatment and random IgG ChIP as the control . To account for read depth , the ChIP signal was normalized to the total number of millions of reads that uniquely aligned to the genome . To correct for non-specific binding , the IgG signal was subtracted from the DPY-27 signal . The resulting ChIP-seq signal from two biological replicates was averaged at each base pair genome-wide . Raw RNA Pol II ChIP-chip data from experiments using 8WG16 antibody ( raised against the hypo-phosphorylated Pol II C-terminal domain ) ( Pferdehirt et al . , 2011 ) was obtained from GEO ( accession numbers GSM634580 and GSM634582 ) . The ChIP signal was normalized to the GC content of individual probes using MA2C ( Song et al . , 2007 ) . The average ChIP-chip signal surrounding the TSS was calculated using the sitepro script within the CEAS package version 1 . 0 . 2 ( http://liulab . dfci . harvard . edu/CEAS ) . To determine whether C . elegans genes contain known core promoter motifs such as the TATA-box and Initiator element ( Inr ) , we performed motif searches using MEME ( http://meme . ebi . edu . au/meme/intro . html ) . To identify a worm TATA-box , we searched for strand-specific motifs within a region 15–45 bp upstream of the TSS . To identify a worm Inr , we searched for strand-specific motifs within 10 bp of the TSS in either direction . These searches identified a TATA consensus of TATAWAWR , compared to TATAWAWR for yeast ( Rhee and Pugh , 2012 ) , and an Inr consensus of YCAYTY , compared to YYANWYY in humans and TCAYTY in Drosophila ( Juven-Gershon and Kadonaga , 2010 ) . To determine where these motifs lie , we calculated their distance from the TSS . For TATA , we calculated how far upstream the most 5′ base lies . For the Inr , we calculated how far the adenine lies from the TSS . In other organisms the adenine has been shown to be the +1 nucleotide in transcription; the location of the worm Inr relative to the TSS suggests that this is true in C . elegans . | In many species , including humans , females have two X chromosomes whereas males have only one . To ensure that females do not end up with a double dose of the proteins encoded by genes on the X chromosome , animals employ a strategy called dosage compensation to control the expression of X-linked genes . The mechanisms underlying dosage compensation vary between species , but they typically involve a regulatory complex that binds to the X chromosomes of one sex to modify gene expression . In the nematode worm Caenorhabditis elegans—which consists of hermaphrodites ( XX ) and males ( XO ) —this regulatory complex , called the dosage compensation complex ( DCC ) , binds to both X chromosomes of XX individuals , reducing gene expression from each by 50% . DCC shares many subunits with a protein complex called condensin , which regulates the structure of chromosomes to achieve proper chromosome segregation . However , it is unclear exactly how the DCC controls the expression of X-linked genes . For a gene to be expressed , an enzyme called RNA polymerase II must bind to the gene’s promoter—a stretch of DNA upstream of the protein-coding part of the gene—so that it can begin transcribing the DNA into RNA . Promoters have been difficult to define in C . elegans , but Kruesi et al . devised a strategy to map transcription start sites , and hence promoters , throughout the worm genome . The strategy integrates the results of two methods: One measures the extent and orientation of each gene’s transcribed region , and the other locates the distinctive cap structures that mark the true 5′ ends of newly made RNAs . Using this new promoter information , coupled with genome-wide measurements of the levels of newly synthesized transcripts from wild-type and dosage-compensation-defective animals , they showed that C . elegans achieves dosage compensation by reducing the recruitment of RNA polymerase II to the promoters of X-linked genes in XX individuals . Kruesi et al . also identified a second regulatory mechanism that acts in both sexes to increase the level of transcription of genes on the X chromosome . This ensures that after dosage compensation , genes on the X chromosome are expressed at a similar level to those on the autosomes ( all chromosomes other than X and Y ) . As well as shedding light on the mechanism by which dosage compensation occurs in C . elegans , the study by Kruesi et al . provides a valuable data set on transcription start sites in the worm , and puts forward a general strategy that could be used to map these sites in other species . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2013 | Condensin controls recruitment of RNA polymerase II to achieve nematode X-chromosome dosage compensation |
A prerequisite for the design of an HIV vaccine that elicits protective antibodies is understanding the developmental pathways that result in desirable antibody features . The development of antibodies that mediate antibody-dependent cellular cytotoxicity ( ADCC ) is particularly relevant because such antibodies have been associated with HIV protection in humans . We reconstructed the developmental pathways of six human HIV-specific ADCC antibodies using longitudinal antibody sequencing data . Most of the inferred naive antibodies did not mediate detectable ADCC . Gain of antigen binding and ADCC function typically required mutations in complementarity determining regions of one or both chains . Enhancement of ADCC potency often required additional mutations in framework regions . Antigen binding affinity and ADCC activity were correlated , but affinity alone was not sufficient to predict ADCC potency . Thus , elicitation of broadly active ADCC antibodies may require mutations that enable high-affinity antigen recognition along with mutations that optimize factors contributing to functional ADCC activity .
A major concept underlying HIV vaccine research is that we can learn from the processes that lead to potent antibody responses in natural infection to inform immunogen design . For this reason , there have been several detailed studies of the evolutionary pathways of potent and broad neutralizing antibodies ( bnAbs ) to HIV ( Bonsignori et al . , 2017; Bonsignori et al . , 2016; Doria-Rose et al . , 2014; Landais et al . , 2017; Liao et al . , 2013; MacLeod et al . , 2016; Simonich et al . , 2019; Wu et al . , 2015 ) . These studies highlight the critical role of somatic hypermutation ( SHM ) , particularly in complementary determining regions ( CDRs ) of antibody heavy chains , in driving the breadth and potency of HIV bnAbs ( Kwong and Mascola , 2018 ) . One recent study also highlighted that maturation of the antibody light chain , too , can be critical for breadth ( Simonich et al . , 2019 ) . Collectively , these studies have provided important insights into the specific mutations that drive maturation of antibody lineages and specificities of bnAbs . The keen interest in HIV bnAbs stems in part from a plethora of compelling studies that demonstrate the efficacy of bnAbs in protection of macaques from experimental SHIV infection ( Pegu et al . , 2017 ) . The data are less convincing for sterilizing protection by HIV-specific non-neutralizing antibodies that mediate antibody-dependent cell cytotoxicity ( ADCC ) in the same experimental animal models ( Fouts et al . , 2015 ) , where the functional interactions of antibodies are harder to test due to species differences ( Bournazos et al . , 2017 ) . However , they have been implicated in protection from HIV-infection in humans in several settings . In the only partially effective HIV vaccine trial to date , antibodies that mediate ADCC were associated with protection , whereas neutralizing antibodies were not ( Haynes et al . , 2012 ) . Moreover , in the setting of mother-to-child transmission , ADCC antibodies have been associated with both decreased transmission risk and disease progression in infants ( Mabuka et al . , 2012; Milligan et al . , 2015 ) . For these reasons , ADCC antibodies , which have the potential to eliminate infected cells , have been increasingly recognized as important to consider for vaccine design and prevention efforts ( Lewis et al . , 2017 ) . Yet , nothing is known about the process of SHM required for HIV specificity and ADCC function of antibodies . In light of the relevance of ADCC antibodies in human infections , we deep sequenced the antibody repertoire of an individual who developed potent ADCC antibodies to the HIV envelope ( Env ) gp120 and gp41 proteins . We used computational methods that were specifically developed to study antibody evolutionary pathways to infer the naïve precursors and resolve chronological lineage intermediates leading to six different mature ADCC-mediating IgG1 antibodies that have been previously described ( Williams et al . , 2015; Williams et al . , 2019 ) . Among the six lineages , most inferred naïve antibodies did not mediate ADCC . To develop ADCC functionality , inferred naive antibodies required CDR-localized SHM in one or both chains . In five of six cases , enhancement of ADCC potency to mature levels required mutations in framework regions ( FWR ) , either alone or alongside additional CDR mutations . Interestingly , in one lineage , the developing antibodies first gained the capacity to bind HIV Env and then subsequently acquired ADCC activity due to additional SHM . Overall , binding affinity for Env and facilitation of ADCC were correlated , but we observed cases where binding affinity was similar amongst two antibodies within the same lineage but they differed in their ADCC capacity , suggesting that affinity alone is not sufficient for an antibody to mediate ADCC .
We previously reported the isolation of six monoclonal antibodies ( mAbs ) , each with potent antibody-dependent cell cytotoxicity ( ADCC ) activity , from a sample collected 914 days post-infection ( D914 ) from a clade A-infected Kenyan woman ( Williams et al . , 2015; Williams et al . , 2019 ) . All these mAbs derived from distinct B cell lineages , with four targeting two distinct gp41 Env epitopes ( 006 , 016 , 067 , 072 ) , and the other two targeting distinct gp120 Env epitopes ( 105 , 157 ) . One of the gp120 targeting mAbs ( 105 ) also had modest capacity to neutralize cell-free virus ( Williams et al . , 2015 ) , whereas the others were non-neutralizing . To infer the ontogeny of these six mAbs , we used next-generation sequencing ( NGS ) to sequence full-length antibody variable region genes from five longitudinal blood samples collected from the subject , spanning from pre-infection ( D-119 ) to over 4 years post-infection ( D1512 ) . Amongst the different antibody chains and timepoints sequenced , the sequencing libraries ranged widely in their sample coverage , between 6 and 409% with an average of 100% coverage , based on estimated B cell frequencies within the available peripheral blood mononuclear cell ( PBMC ) samples ( Table 1 ) . It is important to note that , despite the name ‘deep sequencing’ , NGS-sampled datasets are relatively shallow compared to the entire repertoire . This is for two reasons: ( 1 ) each 10 mL PBMC sample is only ~0 . 22% of the subject’s whole-body circulating blood volume ( ~4 . 5 L ) , not counting lymphoid organs or peripheral tissue and ( 2 ) circulating human memory B cell repertoires fluctuate over time ( Horns et al . , 2019; Laserson et al . , 2014 ) , making it difficult to accurately track clonal B cell families . Despite these limitations inherent to studies of this type , our sequencing efforts successfully returned clonal sequences for all six lineages of interest . We identified sequences clonally related to the gamma heavy ( VH ) and lambda ( VL ) or kappa ( VK ) light chains of the six mature mAbs ( Table 2; Figure 1A–B ) , inferred the most likely naive ancestors of each mAb chain , annotated clonal family V , ( D ) , and J gene usage ( Table 2 ) , and calculated the degree of mutation in each lineage over time ( Table 2; Figure 1C ) . Overall , the level of SHM in the mature D914 mAbs ranged from 3 . 3–11 . 5% , similar to our previous reporting of these lineages using different annotation methods ( Williams et al . , 2015; Williams et al . , 2019 ) . Incidentally , consensus among clonal sequences within families revealed potentially artifactual cloning mutations at the 5’ and 3’ ends of the variable regions in some of the previously characterized mature mAbs ( Williams et al . , 2015; Williams et al . , 2019 ) and suggested , overall , that the consensus sequences were the most likely sequences for the antibodies . When directly compared , the sequence-corrected mAbs demonstrated equivalent epitope binding strength and ADCC activity to those previously studied ( data not shown ) . Thus , we corrected the 5’ and 3’ ends of the mature D914 mAb variable regions to consensus sequences among each clonal family for use in this study ( Supplementary file 1 ) . Each lineage varied in the time point ( s ) from which we identified clonal sequences , with most clonal VH sequences existing in the D462 datasets for four of the six lineages ( Table 2A ) . The NGS sequences within each lineage that shared the highest nucleotide ( nt ) identity with the D914 mature VH variable regions , ranging from 88–99% , were found within the D462 or D791 timepoints ( Supplementary file 2 ) . As test cases , we synthesized antibodies with VH sequences derived from the D462 NGS results that were similar to the mature 006 and 016 mAbs ( 88 . 5% and 97 . 5% nt identity , respectively ) . The 006-NGSVH was paired with 006-NGSVL light chain that had 95 . 8% nt identity to the mature 006 VL ( Figure 2—figure supplement 1A ) and , since this option was not available for 016 VL , the 016-NGSVH was paired with a computationally-inferred light chain ( 016-2VL ) that preceded the mature in development ( Figure 2—figure supplement 1B ) . mAb functionalities were then assessed using the Rapid and Fluorometric ADCC assay , which uses primary PBMCs as target and effector cells ( Gómez-Román et al . , 2006 ) ; in this method , mAbs bind to antigen-coated target cells and facilitate lysis by effector cells . Both NGS-based mAbs demonstrated full ADCC function ( Figure 2 ) , suggesting that ADCC function likely developed within these lineages by D462 post-infection . Inferred naïve ancestor mAbs across lineages varied in their ability to bind HIV Env and facilitate ADCC function . Of the gp41-targeted lineages , only the inferred naive precursor of the 072 lineage bound gp41 ( KD = 34 . 4 nM ) and mediated detectable ADCC , although the activity was very low ( Figure 3A , C ) . Both naive mAbs from the gp120-targeted lineages bound monomeric gp120 ( KD = 23 . 4 nM for 105-0VH0VK and 41 . 1 µM for 157-0VH0VK; Figure 3B ) . Interestingly , despite the 105 naïve mAb having higher binding affinity compared to 157 naïve mAb , it did not mediate ADCC , while the 157 naive mAb mediated potent ADCC ( Figure 3C ) . Typically , the uncertainty in the computational inference of antibody lineage naive precursors is largely ignored , even though single amino acid differences in the naive mAb can result in differences in HIV Env binding and functional properties ( Yuan et al . , 2011 ) . To account for inference uncertainty , we also tested ‘alternative naive’ mAbs for each lineage that were inferred not by partis ( Ralph and Matsen , 2016b ) , but by linearham ( Dhar et al . , 2020 ) , a second computational method that jointly models V ( D ) J recombination and evolutionary history . For 7 of 12 antibody chains , this alternative approach generated naïve sequences that differed from the original naïve sequences by one to four amino acids within the CDR3 and/or FWR4 regions ( Supplementary file 1 ) . For the remaining five , the alternative method inferred the same naive sequence as the original method . When tested , the alternative naïve sequences did not alter antigen binding or function in five of six antibody lineages when compared to the original naïve mAbs ( Figure 3—figure supplement 1A ) . However , in the 157 lineage we found that the alternative naïve VK chain ( 157-AVK ) differed from the original naïve ( 157-0VK ) by two adjacent amino acids in the CDRL3 ( Figure 3—figure supplement 1D ) . When 157-AVK was paired with either the original ( 157-0VH ) or alternative ( 157-AVH ) VH chains , it abolished gp120 binding and ADCC functionality ( Figure 3—figure supplement 1B , C ) . Thus , there is some uncertainty about whether the true 157 lineage naïve precursor was specific for HIV and/or capable of mediating ADCC . To reconstruct probable developmental routes that led to the mature D914 mAb heavy and light chains , we used Bayesian phylogenetic lineage inference , which emphasizes only the inferred ancestral intermediates that have high relative statistical confidence and infers chronological ordering of these likely ancestral sequences . This method was specifically developed to reconstruct antibody evolution using sparse data ( Simonich et al . , 2019 ) . Among the 12 heavy and light chain lineages for the six mAbs , our methods resolved anywhere from 1 to 14 probable intermediate sequences that lay between the naive and the mature sequences ( Table 2 ) . Figure 4 , showing 105VK development , exemplifies one pathway , although all pathways may be viewed in Figure 4—figure supplements 1–3 . Of note , 8 of 12 of these computationally-inferred lineages were validated by the existence of NGS-sampled sequences that were identical either in nucleotide or amino acid sequence to at least one of the inferred intermediates ( Figure 4 , Figure 4—figure supplements 1–3 ) . To focus our subsequent lineage studies on determining which heavy chain and/or light chain mutations enabled HIV Env binding and ADCC gain-of-function , we employed the following strategy to select inferred intermediates of interest ( detailed fully in Materials and methods ) . For each chain , we chose lineage intermediates with high relative confidence that were chronologically near the middle of the inferred lineage , paired them together ( see Materials and methods ) , and performed preliminary experiments to determine whether antigen binding or ADCC gain-of-function occurred before or after these ‘middle’ intermediates ( Figure 4 , Figure 4—figure supplements 1–3 ) . If the middle intermediate had function , we focused on pre-middle inferred intermediates , if available within the computationally-inferred lineages . If the middle intermediate lacked function , we focused on post-middle inferred intermediates . Selected heavy and light chain sequences , were paired together in all possible combinations , roughly chronologically ( i . e . increasing SHM and within per-chain resolved chronologies ) , to reveal per-chain mutations contributing to mAb function . Because we did not use paired sequencing methods , these pairings do not necessarily reflect true biological intermediates , but , instead , allowed us to identify ordered steps within each chain’s maturation that conferred HIV binding and/or ADCC activity . For clarity , we are reporting only the pairings that revealed steps in gain-of-function for the six lineages . Pertinent mAbs thus defined the mutations that conferred HIV binding and ADCC activity in each lineage as follows: gp41 antibody 006: mAb 006 ( VH3-23 , VL2-11 ) targets a discontinuous epitope that includes the C-terminal heptad repeat ( Williams et al . , 2019 ) . Mutations in CDRH2 and CDRH3 , totaling 1 . 4% VH SHM , were minimally required for antigen binding and ADCC activity by 006-1VH0VL in the 006 lineage ( Figure 5A ) . The addition of six light chain mutations among CDRL1 , CDRL3 , and FWRL2 augmented gp41 binding and ADCC potency in 006-1VH1VL . gp41 antibody 016: mAb 016 ( VH4-34 , VL1-51 ) targets a similar gp41 epitope as 006 ( Williams et al . , 2019 ) . Detectable gp41 binding and ADCC function was demonstrated by the 016-1VH1VL mAb which contains mutations in heavy and light chain CDRs and FWRs ( 5 . 6% VH SHM and 3 . 6% VL SHM ) ( Figure 5B ) . A subsequent CDRL3 insertion of residue S94 ( see details on chronology resolution of this insertion in Methods ) allowed for augmented ADCC capacity ( 016-1VH2VL ) . Additional FWR and CDR mutations in the VH chain ( 016-2VH2VL ) further strengthened gp41 binding and ADCC functionality . gp41 antibody 067: mAb 067 ( VH1-69 , VL2-11 ) targets the C-C’ loop ( Williams et al . , 2019 ) . For the 067 lineage , the 067-0VH1VL mAb incorporated a single CDRL3 mutation ( 0 . 3% VL SHM ) and displayed detectable antigen binding; we could not determine whether there was meaningful ADCC function or not as levels were just above background ( Figure 5C ) . To achieve more convincing ADCC function , two CDRH2 mutations were also required ( 1 . 0% VH SHM ) . Further mutations in either the heavy or light chain FWRs modestly increased ADCC potency to >50% that of the 067 mature mAb . gp41 antibody 072: mAb 072 ( VH1-69 , VK1-27 ) targets an overlapping but distinct epitope as 067 ( Williams et al . , 2019 ) . Unlike the other gp41-targeting lineages , the 072 lineage inferred naive mAb bound gp41 antigen and mediated weak ADCC , as aforementioned ( Figures 3A , C and 5D ) . To gain ADCC potency , only mutations in the VH chain were necessary; mAbs 072-1VH0VK ( 0 . 8% VH SHM ) and 072-2VH0VK ( 0 . 5% VH SHM ) demonstrate that two substitutions in either the CDRH1 or CDRH2 increased activity , but , when combined in 072-3VH0VK , they were not additive or synergistic ( Figure 5D ) . Instead , FWRH3 and FWRH4 mutations in mAb 072-4VH0VK were required for ADCC activity comparable to that of 072 mature mAb . gp120 antibody 105: mAb 105 ( VH3-15 , VK3-20 ) targets a V3 linear epitope ( Williams et al . , 2015 ) . The inferred 105 lineage naive and 105-0VH1VK mAbs bound gp120 but displayed only indeterminate levels of ADCC activity , as defined in Materials and methods ( Figure 6A ) . 105-1VH1VK mAb demonstrates that heavy chain CDRH1 and CDRH2 mutations ( 1 . 9% VH SHM ) were required for ADCC gain of function , even though these mutations did not affect binding affinity ( Figure 6A ) . mAb 105-1VH1VK lacked full breadth , however , in that it mediated ADCC against clade A HIV strain BG505 . W6M . B1 , but not clade C CAP210 . 2 . 00 . E8 ( Figure 6—figure supplement 1A ) . ADCC breadth matching that of the 105 mature was achieved , instead , by 105-1VH2VK . Lastly , we note that the FWR4 mutation in the 105-1VH chain was likely a computational artifact and unlikely to affect mAb function , as this mutation is absent in the mature 105 mAb . gp120 antibody 157: mAb 157 ( VH1-69 , VK3-11 ) targets a CD4-induced , C11-like epitope ( Williams et al . , 2015 ) . As detailed in Figure 3—figure supplement 1 and further shown in Figure 6B , inferred naives 157-0VHAVK and 157-0VH0VK differed in their binding and ADCC capabilities , leaving us uncertain about whether this lineage was able to bind gp120 and facilitate ADCC since inception or gained these capacities following two CDRL3 mutations ( Figure 6B ) . The 157-0VH0VK naive lacked full ADCC breadth , however . While this mAb demonstrated high ( >50% ) ADCC potency against two clade A HIV strains ( Figure 6B and Figure 6—figure supplement 1B ) , it had low functionality against autologous QA255 transmitted founder virus and a clade C strain , and failed to mediate any ADCC against two other strains of clades C/D and B ( Figure 6—figure supplement 1B ) . Overall , the majority of lineages , regardless of epitope specificity , ultimately required substitutions in both the heavy and light chains to develop ADCC function >50% as potent as their respective mature mAb ( Figure 7 ) . In five of six lineages , detectable binding accompanied ADCC capacity , with lineage 105 being the exception that bound prior to developing ADCC function . Most notably , substitutions in CDRs were typically required for ADCC gain-of-function , while subsequent FWR substitutions , either alone or alongside additional CDR mutations , augmented this activity ( Figure 7 ) . Since the chronology of inferred ancestral sequences was only resolved on a per-chain basis , we note that heavy and light chain mutations likely co-occurred in an interlaced fashion . However , in two cases ( 016 and 105 ) , boosted ADCC potency was conferred by specific mutations in a single chain that were resolved to have occurred subsequent to mutations already incorporated in the prior intermediate . In other cases , namely 006 , 067 , and 072 , where we cannot know if FWR mutations occurred before or after critical CDR mutations , we can only conclude that the FWR mutations are required in addition to CDR mutations to boost ADCC potency . A graphical summary of the key developmental steps for each lineage is presented in Figure 7—figure supplement 1 . In our detailed analyses of binding and ADCC activity , there were three lineages ( 067 , 072 , and 105 ) where we observed two mAbs from the same lineage having similar binding affinities but different ADCC capabilities . To visualize the relationship between binding affinity and ADCC potency , we plotted mAb binding affinities ( KD ) against corresponding RFADCC potencies and found that these functionalities were positively correlated in all lineages ( Figure 8 ) . It is notable that the antibodies in the 157 lineage that mediate potent ADCC ( naive 157-0VH0VK and mature 157-mVHmVK ) bound gp120 antigen with much lower affinity than antibodies with similar ADCC potency in the other five gp41- or gp120-targeted lineages . These data indicate that binding affinity alone does not dictate ADCC potency . Lineages 067 , 072 , and 105 target defined linear epitopes and longitudinal sequence data from subject QA255 was available to evaluate evolution within these epitopes . The 067 and 072 lineages target overlapping C-C’ loop epitopes ( Williams et al . , 2019 ) , and , amongst 28 longitudinal QA255 Env sequences ( Bosch et al . , 2010 ) , these epitopes remain stable between 21 dpi and 1729 dpi ( Env sequences GenBank accessions MW383929-MW383956 ) . Lineage 105 targets the V3 loop and there were changes in this region among the QA255 Env sequences , although it was unknown if the sequence changes would impact binding of 105 to its epitope . To address this , we finely mapped the epitope and sites of escape for both the mature and naive mAbs using a phage-display deep mutational scanning ( Phage-DMS ) approach ( Garrett et al . , 2020 ) . The mAbs were screened using a library displaying gp41 and V3 peptides from HIV strains BG505 . W6M . C2 ( clade A ) , BF520 . W14M . C2 ( clade A ) , and C . ZA . 1197MB ( clade C ) . Both naive and mature 105 lineage mAbs enriched ( i . e . bound to ) wild-type peptides generated in the backgrounds of BG505 . W6M . C2 and C . ZA . 1197MB Env , but not peptides of the BF520 . W14M . C2 Env background ( Figure 9—figure supplement 1 ) . The differences in binding are likely explained by sequence differences at sites 307–309 in the V3 loop: IRI ( bound ) vs . VHL ( not bound ) . The phage display library also contained mutant peptides representing every possible single amino acid mutation in the context of all three envelope variants , allowing us to map mutations that enable escape from antibody binding . The main sites of selection indicative of escape centered around residues 308–316 spanning the sequence RIGPGQA ( Figure 9A–B , Figure 9—figure supplement 2 ) . The footprints between the naïve and mature mAbs were similar , with perhaps stronger selection observed in the naïve mAb at some positions such as 309 and 315 . However , overall the data suggest that SHM and affinity maturation in this lineage did not affect the major epitope binding footprint . We compared the longitudinal V3 sequence data ( Bosch et al . , 2010 ) with the detailed information on the amino acids important for binding defined by Phage-DMS . There was little evidence of strong selection pressure on the 105 epitope of Env between 21 dpi and 1729 dpi . The only sweeping mutation ( H308R ) emerged at 189 dpi ( 50% frequency ) and swept the population by 560 dpi ( Figure 9C ) . We infer that this substitution enabled the initial activation of the 105 lineage because this lineage prefers R308 over H308 , as aforementioned , and it is known that 105 mAbs cannot mediate ADCC against autologous transmitted founder QA255 . 21P . A17 gp120 ( Williams et al . , 2015 ) , which contains H308 . Except for a subset of Q315R viruses present at 189 dpi and 560 dpi , all 308–316 residues remain stable until 1729 dpi , when substitutions are observed in 7/10 clones at either residue 308 or 316 , with A316T present in 50% of 1729 dpi clones ( Figure 9C ) . Overall , analyses of sequence variation in the 067 , 072 , and 105 lineages suggest limited selection pressure for escape from ADCC antibodies in this subject .
Renewed interest in ADCC-capable antibodies as important to HIV vaccine responses has highlighted the need for a better understanding of their natural development ( Forthal and Finzi , 2018 ) . Here , we report the evolution of six ADCC antibody lineages within a single HIV-infected individual . The inferred naïve precursors of these ADCC lineages varied in their abilities to bind HIV antigen , a finding that agrees with studies of HIV-neutralizing antibody lineages ( Stamatatos et al . , 2017 ) . To achieve potent ADCC activity , most lineages required mutations in both their heavy and light chains . There was also evidence that some of these changes contributed to greater breadth . Generally , ADCC function was achieved through mutations in CDRs , while increased ADCC potency required additional mutations in FWR . ADCC activity accompanied antigen binding in all lineages except one; in the exception , the V3/gp120-targeting 105 lineage demonstrated binding capacity prior to developing ADCC functionality . Although there was some evidence that binding affinity was not solely responsible for ADCC capability or potency , binding affinity and ADCC activity were largely correlated . In sum , this study presents six examples of developmental pathways taken by ADCC-mediating antibodies , highlighting that there are common themes in the ontogeny of HIV-specific ADCC antibodies , but that each evolutionary pathway has unique features . Although not widely applied , antibody lineage studies require robust inference methods that account for statistical uncertainty if we are to be confident in their conclusions . This is highlighted in this study by the different properties displayed by two very similar inferred naïve precursors of the 157 lineage . The development of the 157 lineage was ambiguous: computational uncertainty in two CDRL3 residues resulted in multiple probable naive antibodies with different properties . In these different scenarios , the 157 naïve Ab was either fully capable of both antigen binding and potent ADCC , or it required two CDRL3 substitutions to achieve both functions . In the other five lineages , our use of multiple , tailored computational approaches allowed us to report results with high confidence , since the uncertainties present in computational inference of those lineages did not affect functional characteristics . Three gp41-specific antibody lineages were not capable of binding HIV Env by their inferred naïve precursors , as measured by BLI; they developed this ability through mutation . The lack of detectable antigen binding by inferred naïve precursors is not altogether surprising as this has been observed for a number of HIV bnAbs ( Stamatatos et al . , 2017 ) . Binding capacity was gained in two of the gp41-directed lineages ( 006 and 067 ) through limited mutations localized in the CDRs of single antibody chains ( either heavy or light , respectively ) . The other gp41-directed lineage , 016 , which targets a similar epitope as the 006 lineage , required many mutations amongst the CDRs and FWRs of both chains , highlighting that the gain of activity by different lineages can vary even if they are targeting the same epitope . As is true for all studies of this type , it is possible that our methods were not sensitive enough to detect low level binding or that , despite our best efforts , the inferred naïve sequences were inaccurate . Gp41-directed antibodies have been found to recognize self-antigens ( Verkoczy and Diaz , 2014; Verkoczy et al . , 2014 ) and this , or its response to another antigen , could explain how the 016 lineage emerged before gaining HIV-specificity . We chose the RFADCC assay because RFADCC signal has been repeatedly correlated with disease outcome in humans ( Dhande et al . , 2018; Lewis et al . , 2019; Mabuka et al . , 2012; Madhavi et al . , 2014; Milligan et al . , 2015; Ruiz et al . , 2017 ) , thus making it relevant to understanding the development of potentially protective responses . Additionally , we previously demonstrated that the BL035 . W6M . C1 strain of gp120 used here yields RFADCC results representative of activity against seven different strains of various clades ( Milligan et al . , 2015 ) . There was evidence that the 072 lineage had ADCC function from inception , although the amount of ADCC was low . The remaining lineages required an average of 1 . 2% SHM to gain detectable ADCC function ( ranging 0 . 1–4 . 3% SHM ) . For augmentation of ADCC function to >50% of mature activity , the lineages required an average of 2 . 2% SHM ( range 0 . 6–4 . 8% SHM ) . NGS-sampled sequences that were present at 462 dpi displayed high function with 5 . 3% SHM ( 006 lineage ) , and developmental inferences for the 006-VH , 067-VH and 105-VH chains were informed by clonal sequences from only the 462 dpi time point , suggesting that the fully functional VH intermediates inferred for these lineages were likely present prior to 462 dpi . This is not surprising , as ADCC responses to HIV can develop within the first few months of infection ( Dugast et al . , 2014 ) . Low mutation levels to achieve potent ADCC are promising for ADCC vaccine efforts , especially since these six characterized lineages have desirable characteristics for combating HIV , such as potent ADCC activity , cross-clade ADCC breadth , highly conserved epitopes , and potential to help clear infected cells ( Williams et al . , 2015; Williams et al . , 2019 ) . Moreover , these levels of mutation are easily attainable by vaccines ( Schramm and Douek , 2018 ) . Importantly , most lineages required FWR mutations in one or both chains to achieve potent ADCC functionality , with or without additional CDR mutations . Although they can cost thermostability ( Henderson et al . , 2019 ) , FWR mutations are known to contribute to the properties of HIV-specific bnAbs , predominantly by altering Ab conformation in ways that allow conserved epitopes to be bound with high affinity ( Henderson et al . , 2019; Kepler and Wiehe , 2017; Ovchinnikov et al . , 2018 ) . FWR residues contribute to the structure of the Ab and , though they do not usually directly contact the antigen , they can affect binding affinity indirectly ( Foote and Winter , 1992 ) . Furthermore , since they indeed affect Ab structure , FWR mutations could confer favorable Ab flexibility or an orientation that mediates better Fc receptor dimerization and ADCC potency ( Acharya et al . , 2014 ) . FWR mutations in Fab ‘elbow’ regions can affect conformational flexibility and paratope plasticity during bnAb development ( Henderson et al . , 2019 ) . Here , the 006-1VH lineage intermediate caused ADCC gain-of-function and it contained , among other mutations , a CDRH3 elbow mutation Y111F . Future structural studies are warranted to explore how FWR mutations affect Ab structure and function in these lineages . Binding affinity has been implicated in the regulation of ADCC in several studies on tumor-specific Abs ( Mazor et al . , 2016; Tang et al . , 2007 ) . Here , while binding affinity correlated with ADCC function , it was not solely responsible for ADCC potency . In one lineage , binding capacity preceded detectable ADCC function . Notably , we found that half of the lineages contained clonally-related antibodies with similar binding affinity to one another but markedly different ADCC potencies . Conversely , mAbs within the two gp120-targeted lineages exemplify that there can be dramatic differences in binding affinity among mAbs that mediate equivalent , potent ADCC activity . Thus , ADCC function cannot be predicted based on affinity alone . These data support the notion that factors such as epitope specificity , binding mode , and other antibody-antigen interaction dynamics contribute to ADCC potency , as other studies have emphasized ( Acharya et al . , 2014; Orlandi et al . , 2020; Tolbert et al . , 2020 ) . Amongst Env sequences collected over ~4 . 5 years of HIV infection in subject QA255 , we observed little directional escape in the C-C’ loop and V3 epitopes targeted by the 067 , 072 , and 105 antibody lineages , and only at later stages of infection . This finding could indicate that selective pressure exerted by ADCC antibodies is often not enough to drive escape , although there is evidence that escape can occur ( Isitman et al . , 2012 ) . There are at least two possible explanations for limited escape from ADCC antibodies in an infected individual: one is that the activity of ADCC antibodies is insufficient to drive escape because they have limited impact on infection dynamics; the other is that the targeted killing of infected cells leads to less pressure for escape compared to the process of blocking virus entry mediated by neutralizing antibodies . If the latter is true , ADCC could be particularly useful components of vaccines responses and therapeutic tools . Our study offers a granular understanding of how ADCC functionality developed in six human Ab lineages . The results of this study are relevant to rational vaccine design , especially as technologies for guiding Ab development continue to emerge through reverse vaccinology efforts ( Burton , 2017; Rappuoli et al . , 2016; Schramm and Douek , 2018 ) . Additionally , the reliance on FWR mutations to achieve high ADCC functionality could inform mAb optimization strategies for therapeutic uses .
Peripheral blood mononuclear cell ( PBMC ) samples were obtained between 1997 and 2002 from a female HIV-1 seroconverted subject , QA255 , who was enrolled in a prospective cohort of HIV-1-negative high-risk women in Mombasa , Kenya ( Lavreys et al . , 2002 ) . QA255 was 40 years old at the time of HIV infection ( D0 ) . Study participants were treated according to Kenyan National Guidelines; QA255 did not receive antiretrovirals at any point during the period in which samples were analyzed for this study . Antiretroviral therapy was offered to all participants in the Mombasa Cohort beginning in March 2004 , with support from the President’s Emergency Plan for AIDS Relief . The infecting virus was clade A based on envelope sequence ( Bosch et al . , 2010 ) . Approval to conduct this study was provided by the ethical review committees of the University of Nairobi Institutional Review Board , the Fred Hutchinson Cancer Research Center Institutional Review Board , and the University of Washington Institutional Review Board . Study participants provided written informed consent prior to enrollment . PBMCs stored in liquid nitrogen were thawed at 37°C , diluted 10-fold in pre-warmed RPMI and centrifuged for 10 min at 300 x g . Cells were washed once in phosphate-buffered saline , counted with trypan blue , centrifuged again , and total RNA was extracted from PBMCs using the AllPrep DNA/RNA Mini Kit ( Qiagen , Germantown , MD ) , according to the manufacturer’s recommended protocol . RNA was stored at −80°C . Antibody sequencing was performed as previously described ( Simonich et al . , 2019; Vigdorovich et al . , 2016 ) . Library preparation was performed in technical replicate , as indicated in Table 1 , using the same RNA isolated from each timepoint: D-119 ( 119 days prior to HIV infection ) , D462 , D791 , D1174 , and D1512 post-infection . Briefly , RACE-ready cDNA synthesis was performed using the SMARTer RACE 5’/3’ Kit ( Takara Bio USA , Inc , Mountain View , CA ) using primers with specificity to IgM , IgG , IgK , and IgL , as previously reported ( Simonich et al . , 2019 ) . One replicate each for D791 and D1174 were prepared using template switch adaptor primers that included unique molecular identifiers in the cDNA synthesis step: SmartNNNa 5’ AAGCAGUGGTAUCAACGCAGAGUNNNNUNNNNUNNNNUCTTrGrGrGrGrG 3’ ( Turchaninova et al . , 2016 ) , where ‘rG’ indicates ( ribonucleoside ) guanosine bases . cDNA was diluted in Tricine-EDTA according to the manufacturer’s recommended protocol . First-round Ig-encoding sequence amplification ( 20 cycles ) was performed using Q5 High-Fidelity Master Mix ( New England BioLabs , Ipswich , MA ) and nested gene-specific primers . Amplicons were directly used as templates for MiSeq adaption by second-round PCR amplification ( 10–20 cycles ) , purified and analyzed by gel electrophoresis , and indexed using Nextera XT P5 and P7 index sequences ( Illumina , San Diego , CA ) for Illumina sequencing according to the manufacturer’s instructions ( 10 cycles ) . Gel-purified , indexed libraries were quantitated using the KAPA library quantification kit ( Kapa Biosystems , Wilmington , MA ) performed on an Applied Biosystems 7500 Fast real-time PCR machine . Libraries were denatured and loaded onto Illumina MiSeq 600-cycle V3 cartridges , according to the manufacturer’s suggested workflow . Sequences were preprocessed using FLASH , cutadapt , and FASTX-toolkit as previously described ( Simonich et al . , 2019; Vigdorovich et al . , 2016 ) . The sequences from our NGS replicates were merged either cumulatively or on a per-timepoint basis , as appropriate , to achieve the highest depth possible for each analysis . Sequences were then deduplicated and annotated with partis ( https://github . com/psathyrella/partis ) using default options including per-sample germline inference ( Ralph and Matsen , 2016a; Ralph and Matsen , 2016b; Ralph and Matsen , 2019; ) . Sequences with internal stop codons or CDR3 regions that were out-of-frame or had mutated codons at the start or end were removed . Indel events were identified , tracked , and reversed for alignment purposes . We did not exclude singletons in an attempt to retain very rare or undersampled sequences . Sequencing run statistics are detailed in Table 1 . Cumulative and per-timepoint datasets underwent clonal family clustering using both the partis unseeded and seeded clustering methods ( Ralph and Matsen , 2016b ) . Inference of unmutated common ancestor sequences and simultaneous clonal family clustering was performed using the seed clustering method along with previously-identified QA255 mature ADCC antibody sequences ( Williams et al . , 2015; Williams et al . , 2019 ) as ‘seeds’ . As an additional measure to ensure highest accuracy for inferred naïve sequences , the uncertainty on each inferred naive sequence was visualized both with the partis --view-alternative-annotations option and by comparing these results to the most likely naive sequences inferred by linearham software ( see next section ) . The comparison revealed only the minor differences that would be expected based on linearham’s method , which uses a more detailed model to infer more accurate naive sequences for individual clonal families . For the unseeded clustering , each dataset was subsampled to 50–150K sequences for computational efficiency . For each dataset , three random subsamples were analyzed and compared to ensure that our subsampling was sufficient to minimize statistical uncertainties . Seeded analyses were not subsampled . Due to the lack of D genes and much shorter non-templated regions in light-chain rearrangements , computational clustering analyses artifactually overestimate clonality in light chain families , that is they cluster together sequences that did not originate from the same rearrangement event , but which come from almost identical naïve rearrangements . This caveat , affecting all unpaired antibody chain sequencing studies , ultimately reduces accuracy in inferring true light chain clonal families and their maturation pathways . Alternative naïve sequences were inferred by applying a Bayesian phylogenetic Hidden Markov Model approach using the linearham software ( https://github . com/matsengrp/linearham ) , using the partis-inferred clonal family clusters for each lineage as input ( Dhar et al . , 2020 ) . Linearham samples naive sequences from their posterior distribution rather than providing a single naive sequence estimate , like other software programs do . The most probable linearham-predicted naïve sequences were compared to the most probable partis-predicted naïve sequences . In families with largely symmetric phylogenetic trees , the two methods return similar results , whereas with highly imbalanced trees , linearham is much more accurate . Antibody lineages were inferred as previously described ( Simonich et al . , 2019 ) . In order to subsample large clonal families according to phylogenetic relatedness to the antibody chain of interest , initial phylogenetic trees of each QA255 clonal family was inferred using FastTree 2 ( Price et al . , 2010 ) . This allowed clonal families to be reduced to the 75–100 sequences most relevant to inferring the lineage history of the antibody chain of interest ( https://github . com/matsengrp/cft/blob/master/bin/prune . py ) , which was then analyzed with RevBayes ( Höhna et al . , 2016 ) using an unrooted tree model with the general time-reversible ( GTR ) substitution model . All settings for RevBayes runs were customized to ensure likelihood and estimated sample sizes were >100 for each lineage . MCMC iterations ranged from 10 , 000 to 200 , 000 , with thinning frequencies between 10 and 200 iterations and number of burn-in samples between 10 and 190 . All RevBayes runs were done in technical duplicate ( i . e . specifying different starting trees ) and duplicates were all confirmed to agree on lineage chronology . RevBayes output was summarized for internal node sequences ( https://github . com/matsengrp/ecgtheow ) , resulting in summary graphics where relative confidence in unique inferred sequences and amino acid substitutions are represented by color intensity ( Figure 4—figure supplements 1–3 ) . For each lineage , inferred intermediate sequences found on the most probable lineage paths were selected for study ( Supplementary file 1 , Figure 4—figure supplements 1–3 ) . In the 016-VL lineage , we determined that a 3-nt insertion event occurred prior to the inferred_7_712 intermediate within this antibody’s most likely developmental pathway ( Figure 4—figure supplement 1 ) . Since most insertion-containing clonal NGS sequences ( 136 of 142 ) encoded serine at the insertion site ( amino acid position 94 ) , we inserted S94 into 016-2VL instead of N94 that would correspond to the mature 016-VL sequence . For thoroughness , the inferred_5_789 016-VL intermediate ( which precedes inferred_7_712 ) was synthesized both without ( 016-1VL ) and with ( 016-2VL ) the S94 insertion . To determine if the computationally-inferred naïve and ancestor sequences were observed in the NGS data , we performed the following procedure for each lineage ( implemented in a script found here: https://git . io/Je7Zp ) . A local BLAST database was created for each seeded clonal family and queried for sampled sequences that had high nucleotide sequence identity to lineage members using the ‘blastn’ command ( Biopython package ) . E-value of 0 . 001 was used; other settings were default . Blastn matches for each lineage member were sorted according to their percent nt identity and alignment length . Sampled sequences with 100% nt or 100% aa identity in common with lineage members were noted ( Supplementary file 1 , Figure 4—figure supplements 1–3 ) . To identify the closest sampled sequences to each VH mature sequence , blastn results per VH mature query were viewed and the highest percent aa identity match was selected . As outlined in our Results , we first selected ‘middle’ lineage intermediates for studying ADCC gain-of-function based on their moderate inclusion of amino acid substitutions ( between naïve and mature sequences; Supplementary file 1 ) and high statistical confidence ( Figure 4—figure supplements 1–3 ) , along with , in some cases , their validation by sampled NGS sequences ( Figure 4—figure supplements 1–3 ) , and/or concentration of substitutions in complementarity-determining regions ( CDRs ) ( Supplementary file 1 ) . Middle intermediate chains were paired with partner naïve , middle , and mature chains and tested for antigen binding and ADCC function . Based on preliminary results using middle intermediates , ADCC-focused lineages were chosen from remaining pre-middle inferred intermediates or post-middle intermediates depending on whether the middle intermediate chain contributed to ADCC activity . If multiple relevant intermediate choices were available within the pertinent ( early or late ) portion of an inferred lineage ( Figure 4—figure supplements 1–3 ) , we selected sequences that had amino acid substitutions that were concentrated in CDRs ( Supplementary file 1 ) and , whenever possible , we selected intermediates that were validated by NGS-sampled sequences , as aforementioned ( Figure 4—figure supplements 1–3 ) . For 067-VH , 067-VK , 072-VH , and 105-VH lineages , early mutations were implicated in ADCC gain-of-function , but we lacked early lineage resolution and therefore could not select early inferred intermediates to study . Instead , we selectively incorporated CDR substitutions from the middle intermediate sequence into the inferred naïve sequence . Non-conservative FWR3 substitutions were also incorporated into the 067-VH early lineage . For each lineage , intermediate sequences were numbered consecutively based on their position in the developmental pathway between naïve ( 0 ) and mature ( m ) . Heavy and light chain pairings for lineage intermediates were based on chronology , which reflected levels of mutation . Each chain was paired with several partner chains for functional assessment . Ultimately , developmental lineages featuring the minimal number of mAbs to study ADCC development were defined based on levels of mAb mutation and ADCC activity . For each lineage , the minimal antibodies included ( 1 ) the inferred naïve , ( 2 ) the most mutated intermediate ( s ) that lacked ADCC activity , if available , ( 3 ) the least mutated intermediate ( s ) that gained detectable ADCC function , and ( 4 ) the least mutated intermediate ( s ) that demonstrated ADCC activity comparable to the mature antibodies ( >50% ) . Mature antibodies were always included as benchmark positive controls . For antibody production: HEK 293 F cells ( RRID:CVCL_D603; originally derived from female human embryonic kidney cells ) were obtained from Invitrogen ( Thermo Fisher Scientific , Waltham , MA , catalog #R790-07 ) and grown at 37°C in Freestyle 293 Expression Medium ( Thermo Fisher Scientific , catalog #12338002 ) in baffle-bottomed flasks orbiting at 135 rpm . These cells were not further authenticated in our hands . For RFADCC: CEM . NKR cells ( RRID:CVCL_X622; originally derived from female human T-lymphoblastoid cells ) were obtained from NIH AIDS Reagent Program ( Germantown , MD , catalog #458 ) and grown at 37°C in RPMI 1640 media with added penicillin ( 100 U/mL ) , streptomycin ( 100 µg/mL ) , Amphotericin B ( 250 ng/mL ) , L-glutamine ( 2 mM ) , and fetal bovine serum ( 10% ) ( all from Thermo Fisher Scientific ) . These cells were not further authenticated in our hands . Antibody heavy and lightchain variable regions were synthesized as FragmentGENES ( GENEWIZ , South Plainfield , NJ ) and subsequently cloned into corresponding Igγ1 , Igκ or Igλ expression vectors ( Tiller et al . , 2008 ) . Equal ratios of heavy and light chain plasmids were co-transfected into HEK 293 F cells using FreeStyle MAX ( Thermo Fisher Scientific , catalog #16447100 ) according to the manufacturer’s instructions . Column-based Pierce protein G ( Thermo Fisher Scientific , catalog #20397 ) purification of IgG was done according to the manufacturer’s instructions . The RFADCC assay was performed as described ( Gómez-Román et al . , 2006; Williams et al . , 2019 ) . In short , CEM-NKr cells ( NIH AIDS Reagent Program , catalog #458 ) were double-labeled with PKH-26-cell membrane dye ( Sigma-Aldrich , St . Louis , MO ) and a cytoplasmic-staining dye ( Vybrant CFDA SE Cell Tracer Kit , Thermo Fisher Scientific ) . The double-labeled cells were coated with clade A gp120 ( BL035 . W6M . Env . C1 , Wu et al . , 2006 ) or clade C gp41 ectodomain ( C . ZA . 1197MB ) ( Rousseau et al . , 2006 ) for 1 hr at room temperature at a ratio of 1 . 5 μg protein: 1 × 105 double-stained target cells . For RFADCC breadth assays , additional HIV gp120 monomers of the following strains were: QA255 . 22P . A17 , BG505 . W6M . B1 , CAP210 . 2 . 00 . E8 , MK184 . W0M . G3 , and SF162 . All gp120 and gp41 proteins were sourced from Immune Technology Corp , New York , NY , or Cambridge Biologics , Brookline , MA , as specified in the Key Resources Table . Coated targets were washed once with complete RMPI media ( RPMI supplemented with 10% FBS , 4 . 0 mM Glutamax , and 1% antibiotic-antimycotic , all from Thermo Fisher Scientific ) . Monoclonal antibodies were diluted in complete RPMI media to a concentration of 100–500 ng/mL and mixed with 5 × 103 coated target cells for 10 min at room temperature . PBMCs ( peripheral blood mononuclear cells; Bloodworks Northwest , Seattle , WA ) from an HIV-negative donor were added at a ratio of 50 effector cells: one target cell . The coated target cells , antibodies , and effector cells were co-cultured for 4 hr at 37°C then fixed in 1% paraformaldehyde ( Affymetrix , Santa Clara , CA ) . Cells were analyzed by flow cytometry ( Symphony I/II , BD Biosciences , San Jose , CA ) and ADCC activity was defined as the percent of PKH-26+ CFDA- cells after background subtraction , where background ( antibody-mediated killing of uncoated cells ) was standardized to be 3–5% as analyzed using FlowJo software ( FlowJo LLC , Ashland , OR ) . To mitigate differences in activity observed with different PBMC donors and between experiments , ADCC activity for each sample was normalized to monoclonal positive control mAbs: 167-D ( NIH AIDS Reagent Program , Cat #11681 ) for gp41-targeted Abs and C11 ( NIH AIDS Reagent Program , Cat #7374 ) for gp120-targeted Abs . The activity of an unrelated antibody , FI6v3 , that recognizes influenza hemagglutinin protein was used to define the limit of detection . For each replicate experiment , samples were categorized in the following manner: positive ( sample >2*FI6v3 ) , indeterminate ( 1*FI6v3 ≤ sample ≤2*FI6v3 ) , negative ( sample <1*FI6v3 ) . Experiments were excluded if FI6v3 signal was >10% of monoclonal positive control signal . Background ( uncoated cells ) and negative-control ( 1*FI6v3 ) signal was subtracted from each sample’s activity and the resultant values were averaged across experimental replicates , normalized to the respective lineage’s mature mAb activity , and plotted in Prism v8 . 0c ( GraphPad , San Diego , CA ) . A designation of indeterminate was also assigned in cases where samples were negative in the majority of experimental replicates , but indeterminate or positive in any replicate ( s ) . In such cases , average activity was reported as usual . QA255 monoclonal antibody binding to monomeric gp120 or gp41 was measured using biolayer interferometry on an Octet RED instrument ( ForteBio , Fremont , CA ) . Antibodies diluted to 8 µg mL−1 in a filtered buffer solution of 1X PBS containing 0 . 1% BSA , 0 . 005% Tween-20 , and 0 . 02% sodium azide were immobilized onto anti-human IgG Fc capture biosensors ( ForteBio ) . C . ZA . 1197MB or BL035 . W6M . C1 gp41 was diluted to 250 nM , or as indicated , in the same buffer ( above ) and a series of up to six , two-fold dilutions were tested as analytes in solution at 30°C . The kinetics of mAb binding were measured as follows: association was monitored for 180 s , dissociation was monitored for 180 s , and regeneration was performed in 10 mM Glycine HCl ( pH 1 . 5 ) . For experiments in which dilution series were run , binding-affinity constants ( KD; on-rate , Kon; off-rate , Kdis ) were calculated using ForteBio’s Data Analysis Software 7 . 0 . Responses ( nanometer shift ) were calculated and background-subtracted using double referencing against the buffer reference signal and non-specific binding of biosensor to analyte . Data were processed by Savitzky-Golay filtering prior to fitting using a 1:1 model of binding kinetics . Profiling of escape mutations was done as previously described ( Garrett et al . , 2020 ) . We utilized a deep mutational scanning ( DMS ) phage display library containing wildtype and mutant peptides that tile across the gp41 and V3 portions of three Envelope strains ( BG505 . W6M . C2 , BF520 . W14M . C2 , and C . ZA . 1197MB ) . Peptides in the library are 31 amino acids long and contain every possible single amino acid mutation at the central position , with peptides overlapping by 30 amino acids . Immunoprecipitation was performed in technical duplicate by incubating 10 ng of antibody sample with 1 mL of gp41/V3 Phage-DMS library at a concentration representing 200 , 000 pfu/mL of each unique clone . Antibody and phage were allowed to form complexes by rocking overnight at 4°C in 96-deep-well plates that had been pre-blocked with 3% BSA in TBST . To isolate phage-antibody complexes , 20 uL each of Protein A and Protein G Dynabeads were added and incubated at 4°C for 4 hr . Beads were magnetically separated , washed 3x with 400 µL wash buffer ( 150 mM NaCl , 50 mM Tris-HCl , 0 . 1% [vol/vol] NP-40 , pH 7 . 5 ) , resuspended in 40 µL of dH2O before lysis ( 95°C for 10 min ) , and stored at −20°C . PCR and deep sequencing of the enriched phage was done as previously described . Briefly , sequences were amplified to add Illumina barcodes ( two rounds ) and then pooled and sequenced on an Illumina MiSeq with 1 × 125 bp single-end reads ( BioProject accession PRJNA685289 ) . Enrichment and scaled differential selection were calculated as previously described , with analysis performed and plots generated using RStudio ( Figure 9—source code 1 , Figure 9—source datas 1–3 ) . Two biological replicates were run , using the same antibody preps but distinct peptide library preps . The QA255 longitudinal antibody deep sequencing datasets ( Table 1 ) and Phage-DMS sequencing data sets generated during this study are publicly available at BioProject SRA , accessions PRJNA639297 [https://www . ncbi . nlm . nih . gov/bioproject/PRJNA639297/] and PRJNA685289 [https://www . ncbi . nlm . nih . gov/bioproject/PRJNA685289] . The inferred antibody variable region sequences generated in this study have not been deposited in GenBank because computationally-inferred sequences are not accepted , but they are available in Supplementary file 1 . GenBank accession numbers for mature QA255 antibody are MT791224-MT791235 and QA255 Envelope sequences are MW383929-MW383956 . GenBank accession numbers for HIV Env variants are available in the Key Resources Table . The custom code generated or used in this study for antibody lineage determination is publicly available on GitHub: prune . py ( https://github . com/matsengrp/cft/blob/master/bin/prune . py ) , ecgtheow ( https://github . com/matsengrp/ecgtheow ) , CFT ( https://github . com/matsengrp/cft ) , and Blast validation ( https://git . io/Je7Zp ) . Code for Phage-DMS analysis is provided as Figure 9—source code 1 . Where applicable , raw data were normalized and/or averaged across replicates using Microsoft Excel . Plots were generated using GraphPad Prism version 8 . 0 c . Relevant experimental details , such as use of biological and technical replicates , can be found in figure legends . For RFADCC plots , bars with error bars represent mean and SD . RFADCC experimental exclusion criteria are detailed within the RFADCC method section above . For Phage-DMS , calculations were performed as previously described ( Garrett et al . , 2020 ) , using RStudio; Phage-DMS analysis source code is available in Figure 9—source code 1 . | Nearly four decades after the human immunodeficiency virus ( HIV for short ) was first identified , the search for a vaccine still continues . An effective immunisation would require elements that coax the human immune system into making HIV-specific antibodies – the proteins that can recognise , bind to and deactivate the virus . Crucially , antibodies can also help white blood cells to target and destroy cells infected with HIV . This ‘antibody-dependent cellular cytotoxicity’ could be a key element of a successful vaccine , yet it has received less attention than the ability for antibodies to directly neutralize the virus . In particular , it is still unclear how antibodies develop the ability to flag HIV-infected cells for killing . Indeed , over the course of an HIV infection , an immune cell goes through genetic changes that tweak the 3D structure of the antibodies it manufactures . This process can improve the antibodies' ability to fight off the virus , but it was still unclear how it would shape antibody-dependent cellular cytotoxicity . To investigate this question , Doepker et al . retraced how the genes coding for six antibody families changed over time in an HIV-carrying individual . This revealed that antibodies could not initially trigger antibody-dependent cellular cytotoxicity . The property emerged and improved thanks to two types of alterations in the genetic sequences . One set of changes increased how tightly the antibodies could bind to the virus , targeting sections of the antibodies that can often vary . The second set likely altered the 3D structure in others ways , potentially affecting how antibodies bind the virus or how they interact with components of the immune system that help to kill HIV-infected cells . These alterations took place in segments of the antibodies that undergo less change over time . Ultimately , the findings by Doepker et al . suggest that an efficient HIV vaccine may rely on helping antibodies to evolve so they can bind more tightly to the virus and trigger cellular cytotoxicity more strongly . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2021 | Development of antibody-dependent cell cytotoxicity function in HIV-1 antibodies |
Within the primate visual system , areas at lower levels of the cortical hierarchy process basic visual features , whereas those at higher levels , such as the frontal eye fields ( FEF ) , are thought to modulate sensory processes via feedback connections . Despite these functional exchanges during perception , there is little shared activity between early and late visual regions at rest . How interactions emerge between regions encompassing distinct levels of the visual hierarchy remains unknown . Here we combined neuroimaging , non-invasive cortical stimulation and computational modelling to characterize changes in functional interactions across widespread neural networks before and after local inhibition of primary visual cortex or FEF . We found that stimulation of early visual cortex selectively increased feedforward interactions with FEF and extrastriate visual areas , whereas identical stimulation of the FEF decreased feedback interactions with early visual areas . Computational modelling suggests that these opposing effects reflect a fast-slow timescale hierarchy from sensory to association areas .
The primate visual system is a hierarchical network of feedforward and feedback connections that supports visual perception , object recognition and selective attention ( Gilbert , 2013 ) . At the earliest level of the visual hierarchy , the primary visual cortex receives input from the lateral geniculate nucleus and propagates this information to higher association areas including the posterior parietal cortex , ventral temporal cortex and frontal eye fields ( FEF ) ( Hubel and Wiesel , 1962; Felleman and Van Essen , 1991; Van Essen and Maunsell , 1983; Maunsell and van Essen , 1983 ) . During active perception , neural signals originating in these later cortical areas provide top-down modulation of activity in early visual areas , depending on factors such as expectation and context ( Gilbert , 2013; Felleman and Van Essen , 1991; Moore and Armstrong , 2003; Bastos et al . , 2015 ) . Functional interactions between early sensory regions and later association areas have been investigated in humans using task-based functional magnetic resonance imaging ( fMRI ) ( Bressler et al . , 2008; Vossel et al . , 2012 ) . This work has demonstrated that task demands play an important role in determining the nature of the interactions between early and late visual cortex ( Gilbert , 2013 ) . For example , studies combining fMRI with transcranial magnetic stimulation ( TMS ) during performance of visual tasks have shown that stimulation of the FEF can modulate neural activity in several visual areas , including the primary visual cortex ( area V1 ) ( Bressler et al . , 2008; Ruff et al . , 2006 , 2008 ) . Increments in the strength of excitatory TMS over the right FEF have been shown to increase neural activity in early retinotopic cortex representing the peripheral visual field , and to reduce activity in central visual-field representations ( Ruff et al . , 2006 ) . This neurophysiological effect confers a perceptual advantage for detection of stimuli in the visual periphery relative to the fovea ( Ruff et al . , 2006 ) . Despite evidence for dynamic bidirectional interactions between FEF and early visual cortex during task-related visual processing , numerous findings from resting-state fMRI ( rsfMRI ) studies , in both human and non-human primates , suggest there is little or no functional coupling between these regions in the absence of active task demands ( i . e . , in the resting-state ) ( Vincent et al . , 2007; Belcher et al . , 2013; Damoiseaux et al . , 2006; Yeo et al . , 2011; Power et al . , 2011; Mantini et al . , 2013; Gordon et al . , 2016 ) . At rest , FEF belongs to extensive networks of fronto-parietal regions ( Yeo et al . , 2011; Power et al . , 2011 ) , whereas the early visual cortex , including area V1 , belongs to a primary visual network encompassing occipital and inferior parietal regions ( Yeo et al . , 2011; Power et al . , 2011; Gordon et al . , 2016 ) . Despite this apparent functional segregation between visual networks encompassing early visual cortex and FEF at rest , the nature of any latent interactions between them remains unknown . Likewise , the neural principles facilitating the emergence of integration between segregated regions at opposing ends of the visual cortical hierarchy remain unclear . Here we combined rsfMRI and non-invasive brain stimulation to examine the causal influence of perturbations of local neural activity within early and late visual areas – specifically areas V1/V2 and FEF – in the absence of visual task demands . Across two separate imaging sessions , we employed continuous theta-burst TMS ( Huang et al . , 2005 ) to inhibit intrinsic neural activity either within the right occipital pole or within the right FEF . We recorded resting-state brain activity immediately before and after TMS , and examined the influence of this perturbation on functional and effective connectivity between the targeted regions and the rest of the brain . We also employed computational modelling to provide candidate mechanisms for expected changes in interactions between cortical regions following local stimulation . Based on recent empirical ( Hasson et al . , 2008; Murray et al . , 2014; Bassett et al . , 2013; Lerner et al . , 2011; Honey et al . , 2012; Gauthier et al . , 2012 ) and computational ( Gollo et al . , 2015; Chaudhuri et al . , 2015 ) work we tested the hypothesis that a temporal hierarchy of timescales , recapitulating the hierarchical organization of neuronal receptive fields , can explain the effects of local stimulation on inter-regional connectivity . According to the model , activity in higher regions such as FEF fluctuates at a slower temporal scale than activity in early sensory regions such as V1/V2 ( Murray et al . , 2014; Honey et al . , 2012; Chaudhuri et al . , 2015 ) . We simulated the effects of local inhibition of early and late visual areas to test whether the proposed temporal gradient from sensory areas ( fast ) to association areas ( slow ) matched the observed changes in connectivity following local cortical perturbations in human participants . Our experimental and modelling results suggest that local inhibition of early visual cortex reduces the discrepancy in endogenous synchronization between lower and higher levels of the visual cortical hierarchy , whereas inhibition of FEF increases the discrepancy . Our work provides novel insights into the neural mechanisms that underlie the effects of local inhibition on large-scale brain dynamics .
To characterize the intrinsic connectivity profiles of the two seed regions-of-interest ( ROIs; 7 . 5 mm radius ) at baseline , we conducted seed-to-whole brain functional connectivity analyses of the resting-state data acquired prior to focal stimulation . These analyses quantified the correlation between the blood-oxygen-level dependent ( BOLD ) signal timecourse extracted from the seed ROIs and the timecourses extracted from all other brain voxels ( see Materials and methods ) . The ROIs were centered on coordinates within the right occipital pole and the right FEF , as specified above . As shown in Figure 2 , activity within the right occipital seed ( V1/V2 ) was positively correlated with activity in other nodes of the visual network , including the lingual gyri , fusiform gyri , the lateral occipital cortex , and the cuneus bilaterally ( Yeo et al . , 2011; Power et al . , 2011; Gordon et al . , 2016 ) ( p<0 . 05 familywise error corrected at cluster level ( FWE ) , Figure 2a – red ) . By contrast , activity in the same occipital seed was negatively correlated ( i . e . , anticorrelated ) with activity in bilateral FEF , the left supramarginal gyrus , the left inferior post-central gyrus , and the left posterior insula ( Figure 2a – blue , p<0 . 05 FWE ) . These results confirm and extend previous findings by showing that in the absence of active visual task demands , interactions between early visual cortex and the right FEF can be anticorrelated ( Yeo et al . , 2011; Power et al . , 2011; Gordon et al . , 2016; Ekstrom et al . , 2008 ) . 10 . 7554/eLife . 15252 . 005Figure 2 . Baseline connectivity between TMS-targeted regions and the rest of the brain before stimulation . ( a ) Regions with higher functional connectivity with early visual cortex at the right occipital pole ( site of subsequent inhibitory TMS ) in the baseline resting-state ( i . e . , before TMS ) . At baseline , activity in the frontal eye fields ( FEF ) , left supramarginal gyrus , left inferior postcentral gyrus , and left insula was anticorrelated with activity in right early visual cortex ( corresponding to the to-be-targeted region of the occipital pole; pink circle ) . ( b ) Regions with positive ( red – yellow ) and negative ( blue – light blue ) functional connectivity with the right FEF at baseline . The right FEF showed a diffuse pattern of connectivity encompassing frontal , parietal , and temporal cortical areas . Regions known to be part of the default-mode network ( medial prefrontal cortex , posterior cingulate , angular gyrus , and medial temporal gyrus ) were significantly anticorrelated with the right FEF at baseline . All results are p<0 . 05 FWE corrected at cluster level . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 00510 . 7554/eLife . 15252 . 006Figure 2—figure supplement 1 . Effects of seed ROI size on baseline connectivity between TMS-targeted regions and the rest of the brain before stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 00610 . 7554/eLife . 15252 . 007Figure 2—figure supplement 2 . Control analyses on baseline connectivity between TMS-targeted regions and the rest of the brain . Seed to voxel connectivity was performed using baseline images from the other experimental session ( i . e . , V1/V2 seed on FEF baseline data and vice versa ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 007 Neural activity in the right FEF seed was positively correlated with activity in bilateral supplementary motor areas , medial dorsal cingulate , bilateral dorsolateral prefrontal cortex , bilateral superior parietal cortex , and extrastriate visual areas including bilateral precuneus , middle occipital gyri , and inferior temporal gyri ( Figure 2b – red , p<0 . 05 FWE ) . Activity within the right FEF seed was also negatively correlated with activity in default mode regions encompassing the medial prefrontal cortex , posterior cingulate cortex , angular gyrus and bilateral medial temporal gyri ( Power et al . , 2011 ) ( Figure 2b – blue , p<0 . 05 FWE ) . These findings are consistent with those of previous neuroimaging studies of intrinsic functional connectivity profiles of human FEF ( Yeo et al . , 2011; Hutchison et al . , 2012 ) , and related findings in macaque ( Vincent et al . , 2007; Hutchison et al . , 2012 ) . Specifically , our results support the idea that the FEF is a functional brain hub , with widespread connections to a variety of resting-state networks including the fronto-parietal and default-mode systems ( Power et al . , 2013; Fornito et al . , 2016; van den Heuvel , 2013a , 2013b ) . Note that the aforementioned baseline effects were replicated using different sized TMS-seed regions ( see Figure 2—figure supplement 1 ) , baseline data ( Figure 2—figure supplement 2 ) and preprocessing procedures ( including global signal regression; see Materials and methods ) . Having established baseline patterns of whole-brain connectivity for the two seed regions at rest , we next examined the influence of local inhibitory TMS on this network activity . To this end , we compared patterns of functional connectivity before and immediately after application of TMS over the right occipital pole and FEF target sites ( see Materials and methods ) . Inhibitory TMS of right visual cortex ( V1/V2 ) resulted in the emergence of positive correlations between this region and bilateral FEF ( Figure 3—figure supplement 1 and Supplementary file 2 ) . TMS of the visual cortex also resulted in the emergence/increase of positive correlations between V1/V2 and extrastriate visual regions including the lingual gyri , the lateral occipital cortex and the parietal cortex ( p<0 . 05 FWE; Figure 3a – red , details in Supplementary files 1 and 2 ) . These results were replicated when we adjusted the radius of the seed regions to 10 mm and 15 mm , and across different data preprocessing pipelines ( Figure 3—figure supplement 2 and Materials and methods ) . On the other hand , inhibitory TMS of right FEF resulted in a reduction in positive correlations between the targeted FEF region and visual areas encompassing the bilateral fusiform and occipital gyri ( Figure 3b – blue , details in Supplementary files 1 and 2 ) . Overall , these effects were replicated using TMS-seed regions of 10 mm and 15 mm radius ( see Figure 3—figure supplement 2 ) and using different preprocessing procedures ( including global signal regression; see Materials and methods ) . Note that the individual sites of stimulation could not be unequivocally linked with specific effects on functional connectivity ( Figure 3—figure supplement 3 ) . 10 . 7554/eLife . 15252 . 008Figure 3 . Distinct effects of local inhibitory TMS over early visual cortex and FEF . ( a ) Inhibitory TMS of early visual cortex ( right occipital pole; encircled lightning symbol ) was associated with the emergence of positive correlations between BOLD signals in V1/V2 and bilateral FEF , and the emergence/increase of positive correlations between the V1/V2 seed region and bilateral occipital and parietal cortices ( see Supplementary file 1 and 2 for details , p<0 . 05 FWE corrected at cluster level ) . ( b ) Inhibitory TMS of the right FEF ( encircled lightning symbol ) resulted in the reduction of positive correlations between this target region and bilateral occipital visual areas ( p<0 . 05 FWE corrected at cluster level ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 00810 . 7554/eLife . 15252 . 009Figure 3—figure supplement 1 . Overlap of baseline and post-TMS connectivity between the early visual cortex seed and FEF , bilaterally . Analysis revealed frontal regions that were anti-correlated with BOLD signal activity in the right occipital pole at baseline ( illustrated in blue ) , regions that increased their connectivity ( i . e . , emergence of positive correlations ) with right early visual cortex following inhibitory TMS ( red-orange ) , and regions that were both anti-correlated with the right occipital pole and increased their functional connectivity with this region following TMS ( pink ) . Note that the FEF cluster showing positive correlations with V1/V2 following V1/V2 stimulation extends slightly inferiorly ( up to Z = 36 ) . All p< 0 . 05 FWE , cluster level . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 00910 . 7554/eLife . 15252 . 010Figure 3—figure supplement 2 . Impact of seed ROI size on effects of local inhibitory TMS over early visual cortex ( V1/V2 ) and FEF . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 01010 . 7554/eLife . 15252 . 011Figure 3—figure supplement 3 . Relationship between TMS sites and changes in brain activity . Assessment of possible relationships between the locations of TMS target sites , changes in the local amplitude of fluctuations in neural activity ( ALFF ) and changes in functional connectivity with cortical areas that showed a significant variation in functional connectivity following TMS ( Figure 3 ) . Blue spheres represent sites showing the highest changes in ALFF and functional connectivity following TMS ( top 33% of changes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 01110 . 7554/eLife . 15252 . 012Figure 3—figure supplement 4 . Relationship between local TMS-induced changes in neural activity and widespread modulation of functional connectivity . The relationship between TMS-induced changes in the local amplitude of slow fluctuations in neural activity ( ALFF , 0 . 01–0 . 1 Hz ) and TMS-induced changes in functional connectivity between the right FEF target site and left/right visual clusters ( as shown in Figure 3 , details in the Supplementary file 1 ) , respectively . ( a ) Reduction in ALFF following TMS was associated with greater anticorrelation ( i . e . , negative functional connectivity ) between FEF and early visual cortices . ( b ) Reduction in ALFF over right early visual cortex following TMS was associated with increased positive functional connectivity between this region and occipito-parietal visual areas . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 01210 . 7554/eLife . 15252 . 013Figure 3—figure supplement 5 . Dynamic Causal Modelling used to determine the direction of changes in functional connectivity following TMS of V1/V2 and FEF . Each model encompassed bidirectional connections between the TMS target region ( i . e . , V1/V2 or FEF , purple spheres ) and core cortical areas ( black spheres ) that showed a significant change in functional connectivity following focal inhibitory TMS ( Figure 3 , details in Supplementary file 1 ) . Each region consisted of a sphere of 7 . 5 mm of radius centered on the depicted co-ordinates ( MNI space ) . Connections that showed changes in the subject-specific parameters between the pre- and post-TMS sessions are shown in red ( feedforward connectivity ) and blue ( feedback connectivity ) ( paired t-test , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 013 We also assessed changes in functional connectivity pre- and post-TMS ( for both V1/V2 and FEF sessions ) using a control region ipsilateral to the TMS stimulation site but outside the networks of interest ( i . e . , the inferior portion of the right motor cortex , x = 51 , y = −10 , z = 18 ) . These analyses revealed no significant changes in functional connectivity between the pre- and post-TMS scans , for both V1/V2 and FEF sessions . It is well known that the local effects of TMS can vary considerably across participants ( Hamada et al . , 2013 ) . To investigate this issue , we examined whether TMS-induced changes in intrinsic connectivity were related to individual differences in local changes in BOLD signal fluctuations at the site of stimulation . We first extracted the mean amplitude of low-frequency fluctuations in BOLD ( ALFF [Yang et al . , 2007] ) from the TMS-target site for each individual participant ( see Materials and methods; details in Supplementary file 3 ) . We then determined the correlation between TMS-induced changes in local ALFF and the change in functional connectivity between the targeted ROIs for each participant , with the significant clusters identified at the group level ( Figure 3 ) . There was a significant correlation between TMS-induced modulation of the amplitude of slow signal fluctuations ( ALFF ) and TMS-induced changes in functional connectivity for the right FEF ( right FEF ROI to left visual cluster r = 0 . 6 , p<0 . 01; right FEF ROI to right visual cluster r = 0 . 6 , p<0 . 01 ) and for the right occipital pole ( right occipital ROI to left visual cluster r = 0 . 4 , trend-level p=0 . 05 ) ( Figure 3—figure supplement 4 ) . This analysis suggests that local TMS-induced reductions in neural activity , as indexed by the amplitude of slow fluctuations in the local BOLD signal , were associated with changes in functional connectivity across visual areas . Once again , these significant correlations were robust across different preprocessing procedures , including deletion ( scrubbing ) of motion-contaminated volumes ( Power et al . , 2012 ) . Moreover , the effects remained after controlling for changes in mean BOLD signal between baseline and post-TMS sessions ( FEF target seed to left visual cluster r = 0 . 5 , p=0 . 01; FEF target seed to right visual cluster r = 0 . 5 , p=0 . 01; occipital target seed to left visual cluster r = 0 . 4 , p=0 . 05 ) . Importantly , there was no significant relationship between TMS-induced changes in functional connectivity and ALFF values in the baseline and post-TMS sessions . As a further control , we also examined activity within homologous visual areas in the opposite ( non-stimulated ) hemisphere . There were no significant correlations between changes in the amplitudes of slow signal fluctuations ( ALFF values ) extracted from homologous visual areas in the left hemisphere ( i . e . , contralateral to the TMS-target sites in the right hemisphere ) and changes in connectivity ( left FEF seed to left visual cluster r = 0 . 2 , p=0 . 42; left FEF seed to right visual cluster r = 0 . 1 , p=0 . 71; left occipital seed to left visual cluster r = 0 . 2 , p=0 . 12; left occipital seed to right visual cluster r = 0 . 2 , p=0 . 51 ) . Finally , there were no clear associations between changes in ALFF and individual TMS target sites ( Figure 3—figure supplement 3 ) . To further explore the nature and directionality of TMS-induced perturbations in functional connectivity , we employed stochastic dynamic causal modelling ( DCM ) ( Daunizeau et al . , 2012 ) . DCM contributes to the analysis by modelling how one region exerts influence over another ( i . e . , effective connectivity , [Friston and Harrison , 2003] ) , which is not possible using traditional functional connectivity analysis . To this end , we generated two models comprising bidirectional links between the V1/V2 or FEF seed regions and the clusters showing significant changes in connectivity following TMS ( Figure 3 , details in the Materials and methods ) . DCM analyses of model parameters ( pre- versus post-TMS ) suggested that stimulation of V1/V2 altered the feedforward influence of V1/V2 on the right occipito-parietal cortex ( details in Figure 3—figure supplement 5 and Supplementary file 1 , paired t-test p=0 . 04 ) . Conversely , stimulation of FEF significantly decreased the feedback influence of this region upon V1/V2 ( Figure 3—figure supplement 5 , p=0 . 0026 for the right cluster and p=0 . 0012 for the left cluster ) . Thus , for both V1/V2 and FEF , modulation of effective connectivity induced by focal TMS spread from the stimulation site to distant cortical regions . Results from functional and effective connectivity analyses suggest that V1/V2 stimulation ( Figure 3 , Figure 3—figure supplement 5 ) increased feedforward interactions between this cortical area and higher visual cortical areas . Conversely , reduced positive correlations following FEF stimulation were driven by a reduction in feedback connectivity ( i . e . , a reduced influence of FEF on V1/V2 ) . Based on evidence from empirical and computational work ( Hasson et al . , 2008; Murray et al . , 2014; Lerner et al . , 2011; Honey et al . , 2012; Gauthier et al . , 2012; Gollo et al . , 2015 ) , we hypothesized that these opposing effects of inhibitory TMS on widespread connectivity might be accounted for by the different timescales at which sensory brain regions and network hubs fluctuate in their levels of activity . To test this hypothesis , we used the Kuramoto model of coupled heterogeneous oscillators , constrained by knowledge of whole brain anatomical connectivity and then adjusted to maximally match the acquired resting-state data at baseline ( see Materials and methods for details ) . The effect of inhibitory TMS was simulated by slowing down ( by a variable amount ) the intrinsic frequency of the target region . This decision was motivated by the observation that changes in connectivity induced by inhibitory TMS were related to a reduction in the power of the local BOLD signal . Such an energy decrease is embodied in our oscillatory model by a slowdown in the intrinsic oscillatory frequency . The model simulates the dynamics of the whole brain at a relatively high resolution ( 513 uniformly sized cortical and sub-cortical regions ) . We focused on the differences in functional connectivity between the two regions targeted by TMS ( V1/V2 and FEF ) and the rest of the brain . Simulation results showed that virtual inhibition of right V1/V2 within the model increased the positive correlations between this region and the rest of the brain ( red in Figure 4a ) , consistent with the group effects we observed following actual TMS over this region in our participant group . This increase in connectivity was robust across a broad range of parameters ( i . e . , slowing of intrinsic frequencies , omega , Figure 4a ) . Importantly , we found that the simulated effect of local V1/V2 inhibition was due mainly to a significant increase in connectivity between this node and other nodes encompassing occipito-temporal and frontal areas of the right hemisphere ( Figure 4b ) . Conversely , simulated inhibition of FEF , again by slowing its intrinsic frequency , was associated with the reduction of correlations in this region’s connectivity with the rest of the brain ( blue in Figure 4a and Figure 4—figure supplement 1 ) . This significant effect involved right frontal areas surrounding FEF , and occipito-temporal cortices ( Figure 4c ) . Overall , the effects observed within the model were consistent with the effects observed following actual TMS of V1/V2 and FEF in the experimental participants . 10 . 7554/eLife . 15252 . 014Figure 4 . Modelling the effects of local inhibitory TMS on diffuse network connectivity . ( a ) In line with the experimental results , computational modelling showed that inhibition of the intrinsic frequencies of V1/V2 and FEF ( see Materials and methods ) had opposite effects on how these regions were connected with the rest of the brain . Specifically , increasing inhibition of V1/V2 enhanced functional connectivity ( i . e . , positive correlations ) with the rest of the brain until the natural frequency of V1/V2 matched the mean frequency of the whole brain ( gray dot ) . Conversely , inhibition of FEF reduced its positive correlations , or enhanced its anticorrelations , with other brain regions ( see also Figure 4—figure supplement 1 ) . s . e . m . = standard error of the mean . ( b ) Following a reduction in its natural frequency , V1/V2 significantly increased its connectivity with other cortical regions comprising occipito-temporal and frontal areas ( including FEF; Wilcoxon signed-rank test Bonferroni corrected for multiple comparisons , p<2 x 10−6 ) . ( c ) A significant reduction in correlations and increase in anticorrelations between FEF and surrounding frontal areas , as well as visual occipito-temporal areas , was observed after slowing of the natural frequency of FEF ( Wilcoxon signed-rank test , p<2 x 10−6 ) . The reduction in omega for panels ( b ) and ( c ) was 0 . 0008 . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 01410 . 7554/eLife . 15252 . 015Figure 4—figure supplement 1 . Effects of virtual inhibitory TMS on simulated functional connectivity between FEF and V1/V2 . The figure shows that virtual inhibition of V1/V2 ( reduction in omega ) results in increased functional connectivity between this area and FEF ( red line ) . By contrast , reductions in omega greater than 0 . 0008 result in the emergence of anticorrelations between the two regions of interest ( FEF and V1/V2; blue line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 01510 . 7554/eLife . 15252 . 016Figure 4—figure supplement 2 . Effects of simulated TMS over two control regions - right postcentral gyrus ( part of the sensorimotor network ) and left superior temporal gyrus ( part of the auditory network ) – on functional connectivity . Simulated inhibition of the right postcentral gyrus caused significant changes in connectivity with ipsilateral parietal and frontal regions . Conversely , inhibition of the left superior temporal gyrus caused a specific change in connectivity between this region and a region in the left temporal pole . Notably , functional connectivity between V1/V2 and FEF was not modulated following simulated inhibition of either of the two control sites ( bar charts in the right panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 01610 . 7554/eLife . 15252 . 017Figure 4—figure supplement 3 . Baseline distribution of natural frequencies ( omega ) in the 513 nodes . The x-axis represents the 513 nodes comprising the adopted brain parcellation . Baseline ( i . e . , pre-stimulation ) values of omega ( natural frequency expressed in Hz ) are presented on the y-axis . Note that at baseline the natural frequency of FEF is slower than the natural frequency of V1/V2 . Likewise , the value of omega for FEF is below the global mean whereas that for V1/V2 is above the global mean . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 01710 . 7554/eLife . 15252 . 018Figure 4—figure supplement 4 . Effects of local inhibition on actual and simulated functional connectivity between the target region ( V1/V2 or FEF ) and the rest of the brain . ( a ) Actual and simulated effects of stimulation of V1/V2 . ( b ) Actual and simulated effects of stimulation of FEF . Raw ( i . e . , unthresholded ) results are shown . Experimental results ( left side of the figure ) were generated using data from one participant who showed a consistent decrease in local BOLD signal amplitude ( ALFF; see Results and Materials and methods ) following TMS of V1/V2 and FEF . The simulation illustrates a representative trial . Overall , the figure shows that the model reproduces the spatial structure of the global changes in functional connectivity induced by local TMS . The reduction in omega for panel a was 0 . 0003 and for panel b was 0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 15252 . 018 To test whether the opposing changes in functional connectivity revealed by the simulations were driven by non-specific effects of local stimulation , we performed several additional simulations . First , we assessed whether inhibition ( delta ω of 0 . 001 ) of cortical regions not relevant for our hypotheses – in this instance areas in the vicinity of the right postcentral gyrus and left superior temporal gyrus – could also produce effects on functional connectivity similar to those observed after simulated inhibition of V1/V2 or FEF . Second , we tested whether inhibition of these two control regions would cause specific changes in functional connectivity between early visual cortex and FEF . Simulated inhibition of either control region yielded a markedly different change in connectivity than that observed following simulated inhibition of V1/V2 or FEF ( see Figure 4—figure supplement 2 ) . Moreover , connectivity between V1/V2 and FEF was not affected by simulated inhibition of either of the two control regions .
We examined dynamic reconfigurations of intrinsic interactions between cortical areas at lower and higher ends of the visual hierarchy in humans following local stimulation with TMS . We applied inhibitory theta-burst stimulation via TMS to one of two regions – early visual cortex and FEF – within the right hemisphere , and examined changes in functional and effective connectivity in the resting-state . Local inhibitory TMS of these regions generated opposing patterns of connectivity between early and late areas within the cortical hierarchy . These distinct effects are consistent with an intrinsic timescale-based cortical hierarchy ( Murray et al . , 2014; Bassett et al . , 2013; Lerner et al . , 2011; Honey et al . , 2012; Gauthier et al . , 2012; Gollo et al . , 2015; Chaudhuri et al . , 2015 ) , with early visual areas ( V1/V2 ) exhibiting shorter timescales than those of higher areas ( such as FEF ) . By incorporating these timescale effects , the computational model provides supporting evidence for opposing changes in diffuse connectivity following focal inhibitory TMS in the human brain . This finding highlights that inter-regional effects of focal TMS can be predicted by the hierarchical organization of timescales across the cerebral cortex . At baseline , prior to inhibitory stimulation , neural activity in early visual cortex was anticorrelated with activity in the FEF bilaterally . Previous neuroimaging work in humans and monkey has suggested little or no functional coupling between FEF and V1/V2 in the resting state ( Vincent et al . , 2007; Belcher et al . , 2013; Damoiseaux et al . , 2006; Yeo et al . , 2011; Power et al . , 2011; Mantini et al . , 2013; Gordon et al . , 2016 ) . For instance , areas V1 and V2 lie within a primary visual network that does not include FEF ( Yeo et al . , 2011 ) . Our results support previous work by showing a functional segregation between early and late areas within the visual cortical hierarchy at rest . This finding is consistent with the notion that task-induced changes in sensory visual areas may drive the emergence of functional integration between levels of the visual cortical hierarchy ( Ekstrom et al . , 2008; Roelfsema , 2006 ) . A novel finding of our study is that inhibition of early visual cortex resulted in the emergence of positive correlations between this region and FEF , along with other extrastriate and parietal visual areas . Previous studies have investigated intrinsic reconfigurations of large-scale neural systems following focal perturbations of peripheral sensorimotor regions ( e . g . , Cocchi et al . , 2015; O'Shea et al . , 2007; Grefkes and Fink , 2011 ) . In line with our results , fMRI studies have shown that local perturbation of primary motor areas can lead to recruitment of related areas ( O'Shea et al . , 2007; Grefkes and Fink , 2011; Cocchi et al . , 2015 ) . For example , inhibitory TMS of the primary motor cortex in healthy adults has recently been shown to increase the strength of resting-state functional connectivity within the sensorimotor system ( Cocchi et al . , 2015 ) . Our finding that unilateral inhibition of early visual cortex increases the positive coupling with FEF and other visual areas adds to a growing literature suggesting that local perturbations of neural activity in peripheral nodes recruit functionally related brain regions ( Cocchi et al . , 2015; O'Shea et al . , 2007; Grefkes and Fink , 2011; Bestmann et al . , 2010 ) . Specifically , our computational modelling suggests that the effect of inhibitory TMS of early visual cortex might be due to slowing of intrinsic timescales within V1/V2 , so that they become closer to the slow intrinsic fluctuations of higher cortical regions such as the parietal cortex and FEF ( Murray et al . , 2014; Honey et al . , 2012; Gollo et al . , 2015; Chaudhuri et al . , 2015 ) . In contrast to the increased positive coupling apparent after inhibition of early visual cortex , an identical inhibitory TMS protocol delivered over the right FEF led to decoupling between the target region and bilateral early visual cortex . Results from the DCM analysis suggest that such functional decoupling was related to a significant decrease in feedback signals from FEF to early visual cortex . In keeping with the neural mechanism proposed to explain increased connectivity following the inhibition of V1/V2 , results from our computational model suggest that the observed reduction in feedback modulation from FEF to V1/V2 can be explained by a further slowing of local FEF dynamics by TMS . The observed changes in connectivity following inhibitory TMS may therefore be explained in terms of a reduction in synchronization discrepancy following local stimulation of V1/V2 and an increase in synchronization discrepancy following stimulation of FEF . Within this broad context , preliminary analyses suggest that there are a host of topographical specificities and nuances that accompany virtual simulations . Further work is needed to address the specificity of our model and the source of possible discrepancies between virtual and empirical findings . Our findings also reveal a relationship between the magnitude of local signal change and remote modulations in intrinsic functional connectivity . We found that TMS-induced changes in local BOLD signal amplitude at the stimulation site were related to remote modulations in functional connectivity . This suggests that variations in ongoing low-frequency ( <0 . 1 Hz ) fluctuations may be a marker of TMS-induced modulation of widespread cortical connectivity . The dynamic integration of information between sensory and association regions of the cortex is essential for normal brain function . Here we combined functional brain imaging , neural stimulation and computational modelling to elucidate the neural mechanisms that support the emergence and dissolution of interactions between cortical regions within the human visual system following local perturbations in neural activity . Our results suggest that the selective effect of local inhibitory TMS on diffuse patterns of connectivity can be accounted for by an intrinsic hierarchical ordering of cortical timescales ( Murray et al . , 2014; Honey et al . , 2012; Gollo et al . , 2015 ) .
Target regions for inhibitory theta-burst TMS were defined using high-resolution structural T1 3D images obtained for each participant and loaded into an ANT Visor Neuro-navigation system with NDI Polaris Spectra infrared camera . The two TMS target sites were identified prior to the first experimental session ( Figure 1 ) . The anatomical locations of the TMS target coordinates were manually refined according to each individual participant's anatomy . Specifically , if the target region was located in a sulcus , the location for TMS was moved to the gyrus closest to the centroid coordinate ( V1/V2: MNI centroid x = 25 , y = −92 , z = −9; FEF: MNI centroid x = 31 y = −2 , z = 47 ) . The early visual cortex target was located anatomically within the occipital pole , posterior to the descending occipital gyrus laterally and the lingual gyrus medially ( corresponding to areas V1/V2 [Thiebaut de Schotten et al . , 2014] ) . In line with previous studies , the FEF target was located anatomically within the posterior middle frontal gyrus , immediately ventral to the junction of the superior frontal sulcus and ascending limb of the pre-central sulcus ( Ruff et al . , 2006; Heinen et al . , 2014 ) . A continuous theta-burst TMS protocol was utilized to induce local inhibition of cortical activity using a previously validated protocol ( Huang et al . , 2005 ) . The inhibitory TMS protocol involved uninterrupted ( 40 seconds ) bursts of 3 TMS pulses delivered at 50 Hz , repeated at 200 ms intervals . TMS was administered using a figure-of-eight coil ( 70 mm diameter ) . For FEF stimulation , the TMS coil handle was held at a 45-degree angle to the sagittal plane ( Nyffeler et al . , 2006 ) . For stimulation of the right occipital pole , the coil was oriented with the handle pointing to the right ( Kammer et al . , 2001 ) . The intensity of the inhibitory stimulation of the two cortical ROIs was set to 80% of the active motor threshold ( Huang et al . , 2005; Cocchi et al . , 2015 ) . The active motor threshold was defined as the minimum TMS intensity required to trigger a motor evoked potential ( MEP ) > 200 µV in at least three out of five consecutive trials while participants were actively contracting their hand muscle ( using a pincer grip ) at a level equivalent to ~20% of their maximum voluntary contraction ( Huang et al . , 2005 ) . The location used to establish the active motor threshold was identified with single-pulses of TMS over the right hemisphere . The TMS coil was systematically moved until the optimal cortical site to induce the largest and most reliable motor response ( motor evoked potential , MEP ) in a muscle of the left hand ( the abductor pollicis brevis ( APB ) muscle ) was established . MEPs were recorded using surface electromyography ( EMG ) electrodes ( Ag-AgCl ) from the left APB . Electromyography signals were amplified ( x1000 ) and filtered ( 5–500 Hz ) using a Neurolog system ( Digitimer , UK ) , and digitised ( 20 kHz ) using a data acquisition interface ( BNC-2110; National Instruments ) and custom Matlab software ( MathWorks [Natick , Massachusetts] , see Source code 1 ) . During resting state fMRI scans , participants were instructed to keep their eyes open and to fixate on a central white cross on a uniform black background . Participants were instructed to let their minds wander freely during the scan . Eye tracking video software was employed to ensure that participants kept their eyes open and looked straight ahead throughout the sessions of resting-state fMRI data acquisition . The eye-open resting-state protocol was preferred over the eye-closed protocol due to recent concerns about controlling wakefulness when participants have their eyes closed ( Tagliazucchi and Laufs , 2014 ) . Neuroimaging data were acquired using a 3T Siemens Trio scanner equipped with a 32-channel head coil at The University of Queensland’s Centre for Advanced Imaging ( Australia ) . Whole brain T2* images were acquired using an echo-planar imaging sequence ( 38 axial slices , 320 volumes , gap = 10% , slice thickness = 3 mm , in-plane resolution = 64 × 64 , time repetition = 2250 ms , time echo = 28 ms , flip angle = 90° , FOV= 210 × 210 mm , descending slice acquisition ) . T1 3D images were acquired using the following parameters: 192 axial slices , slice thickness = 0 . 9 mm; in-plane resolution = 64 × 64 , time repetition = 1900 ms , flip angle= 9° , time echo = 2 . 32 ms , FOV = 230 × 230 mm . Preprocessing of the resting-state fMRI data was performed using the Data Processing Assistant for Resting-State fMRI 3 ( Chao-Gan and Yu-Feng , 2010 ) . The first 10 image volumes were discarded to allow tissue magnetization to reach a steady-state and to provide participants with an opportunity to adapt to the MR scanner environment . DICOM images were converted to Nifti format and underwent slice time correction . To improve normalization , individual participant structural images ( T1 ) were coregistered to functional images using the DARTEL algorithm implemented within the Matlab toolbox SPM8 . Tissue segmentation ( gray matter , white matter and cerebrospinal fluid ) was performed to improve the characterization of non-neural signals in subject space . The following nuisance covariates were regressed from each voxel’s time series: six head motion parameters , linear trends , volume-level mean of frame-to-frame displacements greater than 0 . 4 mm ( including the preceding and two subsequent frames [Power et al . , 2014] ) and signals related to time-series unlikely to be modulated by neural activity ( CompCor method [Behzadi et al . , 2007] ) . After covariate regression , images were normalized to standard MNI space , smoothed using a Gaussian function with a 6 mm full width at half maximum kernel . Finally , a temporal bandpass filter was applied retaining frequencies between 0 . 01–0 . 1 Hz . Supplementary analyses demonstrated that both the mean frame-wise displacement ( Power et al . , 2014 ) and the number of scrubbed volumes were not significantly different for the pre- and post-TMS scans , in either experimental session . Note that after scrubbing motion-contaminated volumes ( Power et al . , 2014 ) , the number of remaining volumes for each functional scanning session exceeded 8 . 8 min , which is sufficient time to capture stable correlation coefficients ( Van Dijk et al . , 2010 ) . Thus , the results reported reflect the outcomes following motion correction ( i . e . , re-alignment , regression of the six head motion parameters and scrubbing ) . Given recent concerns in the literature regarding the use of global signal regression , we did not regress the mean global signal during pre-processing . The mean global signal was not significantly different across pre- and post-TMS sessions , for either stimulation site ( occipital TMS session: t20 = 1 . 22 , p=0 . 23 , FEF TMS session: t20 = 0 . 941 , p=0 . 35 ) . Nevertheless , to ensure that this methodological decision did not alter the main results reported here , we re-ran the analysis incorporating regression of the mean BOLD signal , and obtained similar results . | In humans , the parts of the brain involved in vision are organized into distinct regions that are arranged into a hierarchy . Each of these regions contains neurons that are specialized for a particular role , such as responding to shape , color or motion . To actually ‘see’ an object , these different regions must communicate with each other and exchange information via connections between lower and higher levels of the hierarchy . However , it remains unclear how these connections work . A brain region called the primary visual cortex is the lowest level of the visual cortical hierarchy as it is the first area to receive information from the eye . This region then passes information to higher regions in the hierarchy including the frontal eye fields ( FEF ) , which help to control visual attention and eye movements . In turn , the FEF is thought to provide ‘feedback’ to the primary visual cortex . Cocchi et al . examined how the FEF and primary visual cortex communicate with the rest of the brain by temporarily inhibiting the activity of these regions in human volunteers . The experiments show that inhibiting the primary visual cortex increased communication between this region and higher level visual areas . On the other hand , inhibiting the FEF reduced communication between this region and lower visual areas . Computer simulations revealed that inhibiting particular brain regions alters communication between visual regions by changing the timing of local neural activity . In the simulations , inhibiting the primary visual cortex slows down neural activity in that region , leading to better communication with higher regions , which already operate on slower timescales . By contrast , inhibition of the FEF reduces its influence on lower visual regions by increasing the difference in timescales of neural activity between these regions . The next step is to determine whether similar mechanisms regulate changes in the activity of neural networks outside of the visual system . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | A hierarchy of timescales explains distinct effects of local inhibition of primary visual cortex and frontal eye fields |
Fast synaptic inhibition in the nervous system depends on the transmembrane flux of Cl- ions based on the neuronal Cl- driving force . Established theories regarding the determinants of Cl- driving force have recently been questioned . Here , we present biophysical models of Cl- homeostasis using the pump-leak model . Using numerical and novel analytic solutions , we demonstrate that the Na+/K+-ATPase , ion conductances , impermeant anions , electrodiffusion , water fluxes and cation-chloride cotransporters ( CCCs ) play roles in setting the Cl- driving force . Our models , together with experimental validation , show that while impermeant anions can contribute to setting [Cl-]i in neurons , they have a negligible effect on the driving force for Cl- locally and cell-wide . In contrast , we demonstrate that CCCs are well-suited for modulating Cl- driving force and hence inhibitory signaling in neurons . Our findings reconcile recent experimental findings and provide a framework for understanding the interplay of different chloride regulatory processes in neurons .
Fast synaptic inhibition in the nervous system is mediated by type A γ-aminobutyric acid receptors ( GABAARs ) and glycine receptors ( GlyRs ) , which are primarily permeable to chloride ( Cl- ) ( Farrant and Kaila , 2007 ) . Together with the neuronal membrane potential , the transmembrane gradient for Cl- sets the driving force for Cl- flux across these receptors , controlling the properties of inhibitory signaling . Modification of neuronal intracellular Cl- concentration ( [Cl-]i ) has been shown to play a causative role in multiple neurological diseases including epilepsy , chronic pain , schizophrenia and autism ( Rivera et al . , 2004; Huberfeld et al . , 2007; Price et al . , 2009; Hyde et al . , 2011; Tyzio et al . , 2014 ) . Similarly , intracellular Cl- is thought to be modulated during brain development so that GABAergic transmission contributes optimally to the construction of neural circuits ( Ben-Ari , 2002 ) . Given the importance of Cl- for brain function and dysfunction , the cellular mechanisms that control its transmembrane gradient and driving force are of considerable interest . Plasmalemmal Cl- transporters , in particular cation-chloride cotransporters ( CCCs ) , are understood to be the major mechanism by which neurons regulate the driving force for Cl- permeable anion channels ( Kaila et al . , 2014 ) . Recently , it has been suggested that in fact impermeant anions control local [Cl-]i and driving force ( Glykys et al . , 2014 ) , rather than the CCCs . The majority of intracellular anions are impermeant to the neuronal membrane; these include ribo- and deoxynucleotides , intracellular proteins and metabolites ( Burton , 1983 ) . Impermeant anions induce what is known as the Donnan ( or Gibbs-Donnan ) effect ( Hill , 1956; Sperelakis , 2012 ) – an uneven distribution of impermeant molecules across the membrane which is osmotically unstable . Without active ion transport to counter this effect , neurons would swell and burst ( Kay , 2017 ) . Animal cells , including neurons , maintain cell volume in the presence of impermeant anions by using the Na+/K+-ATPase to pump Na+ out of the cell and K+ in , along with the passive movement of water and other ions ( Tosteson and Hoffman , 1960; Armstrong , 2003; Liang et al . , 2007; Kay , 2017 ) . This pump-leak mechanism , whilst stabilizing cell volume , also establishes the negative resting membrane potential and transmembrane Na+ and K+ gradients , which serve as energy sources for the coupled transport of other molecules , including Cl- by CCCs . In the absence of active Cl- transport , [Cl-]i is set by the membrane potential; i . e . the Nernst potential of Cl- ( ECl=RTF lnCliClo ) equals the transmembrane potential ( Vm ) . The transmembrane Cl- gradient and the driving force ( DF = Vm−ECl ) for Cl- permeable ion channels are therefore the outcome of multiple , dynamically interacting mechanisms . This makes experimental investigation of the determinants of Cl- driving force difficult , particularly at a local level . Computational models based on established biophysical first principles are a productive means for exploring the roles of cellular mechanisms in generating local Cl- driving force . Here , we establish numerical and novel analytic solutions for an inclusive model of Cl- homeostasis to elucidate the determinants of the neuronal driving force for Cl- . We demonstrate that baseline [Cl-]i is a product of the interaction of Na+/K+-ATPase activity , the mean charge of impermeant anions , ion conductances , CCCs and water permeability . Consistent with recent experimental reports ( Glykys et al . , 2014 ) , and our own experimental validation using electroporation of anionic dextrans and optogenetic probing of EGABA , we find that impermeant anions can contribute to setting [Cl-]i . However , we find that they can only affect the Cl- driving force by modifying active transport mechanisms , and then only negligibly . Impermeant anions therefore do not appreciably modify synaptic signaling properties , contrary to the interpretation of recent experiments ( Glykys et al . , 2014 ) . In contrast , we demonstrate using biophysical models and gramicidin perforated-patch clamp recordings that CCCs selectively regulate substantial changes in the Cl- driving force . This is consistent with a meta-analysis of experimental data from the field , which shows a strong correlation between the activity of the specific CCC , KCC2 , and Cl- driving force . The ability of CCCs to specifically modulate Cl- at a local level depends on the characteristics of Cl- electrodiffusion in the structure concerned , demonstrated using multicompartment modeling . Together , our models provide a theoretical framework for understanding the interplay of chloride regulatory processes in neurons and interpreting experimental findings .
To compare the effects of impermeant anions and Cl- cotransport on Cl- homeostasis we first developed a single compartment model based on the pump-leak formulation ( Tosteson and Hoffman , 1960; Kay , 2017 ) ( Figure 1A ) . This model , defined by a set of differential equations , incorporated mathematical representations of the three major permeable ion species Cl- , K+ , Na+ as well as impermeant ions ( Xz ) with mean charge z . Permeable ions could move across the cellular membrane via passive conductances according to each ion’s respective electrochemical gradient . Further , active transport of Na+ and K+ by the Na+/K+-ATPase ( with a 3:2 stoichiometry ) and cotransport of Cl- and K+ ( 1:1 stoichiometry ) by the cation-chloride cotransporter KCC2 with a non-zero conductance gKCC2 of 20 µS/cm2 unless otherwise stated were included . Finally , our formulation accounted for the dynamics of cell volume ( w ) , intracellular osmolarity ( Π ) and transmembrane voltage ( Vm ) ( see Materials and methods ) . Importantly , regardless of initial starting concentrations of permeant or impermeant ions , cell volume or Vm , the model converged to stable fixed points without needing to include any means for ‘sensing’ ion concentration , volume or voltage . Initial permeable ion concentrations also did not influence the final cellular volume . For example , despite initiating the model with different starting concentrations of Cl- ( 1 , 15 , 40 and 60 mM , respectively , Figure 1B ) , [Cl-]i always converged to the same stable concentration of 5 . 2 mM , a typical baseline [Cl-]i for adult neurons , and volume always converged to 2 . 0 pL , a typical volume for hippocampal neurons ( Ambros-Ingerson and Holmes , 2005 ) . The model is robust in the sense that its convergence to a stable steady state does not depend on a narrow set of parameters and initial values . Trying an alternative model of the Na+/K+-ATPase ( Hamada et al . , 2003 ) produced similar results ( Figure 1—figure supplement 1A-B ) . Consistent with previous results ( Xiao et al . , 2002; Dierkes et al . , 2006; Dijkstra et al . , 2016 ) , ‘turning off’ the activity of the Na+/K+-ATPase in our model led to a progressive collapse of transmembrane ion gradients , progressive membrane depolarization and continuous and unstable cell swelling . Such effects could be reversed by reactivation of the pump ( Figure 1C ) . The relative activity of the Na+/K+-ATPase sets the final stable values for ion concentrations ( of both permeable and impermeant ions ) , Vm and volume ( Figure 1C ) . When we increased the activity of the Na+/K+-ATPase in our model , the final steady-state concentration for K+ increased , whilst Na+ and Cl- dropped to levels that approximate those observed in mature neurons ( Figure 1D ) . Note that although sufficient Na+/K+-ATPase activity is critical for steady state ionic gradients including that of Cl- , these are relatively stable near the default pump rate ( Figure 1—figure supplement 2 ) . At the same time , the final , stable-state membrane potential and cell volume also decreased . Interestingly , as has been observed previously ( Fraser and Huang , 2004 ) , beyond a certain level , further increases in Na+/K+-ATPase activity have negligible effects on cell volume and transmembrane voltage . For subsequent analysis , we chose a ‘default’ effective pump rate for the Na+/K+-ATPase of approximately 1 . 0 × 10−2 C/ ( dm2 . s ) , that is a pump rate constant of 10−1 C/ ( dm2 . s ) ( equation ( 2 ) ) , and mean intracellular impermeant anion charge ( z ) of −0 . 85 as extrapolated from reasonable cellular ionic concentrations and osmolarity ( Lodish et al . , 2009; Raimondo et al . , 2015 ) . This resulted in steady-state ion concentrations and membrane potentials that approximate those experimentally observed in mature neurons: Cl- 5 mM; K+ 123 mM; Na+ 14 mM; Xz 155 mM; and a Vm of −72 . 6 mV ( Jiang and Haddad , 1991; Diarra et al . , 2001; Tyzio et al . , 2008 ) . We were able to corroborate the numerical solutions for final steady-state values by developing a parametric-analytic solution ( Supplementary file 1 ) . We observed exact correspondence between the numerical and analytic solutions within our model ( Figure 1D ) . In subsequent analyses , this novel analytic solution allowed us to explore rapidly a large parameter space to determine how various cellular attributes might affect Cl- homeostasis . Using the analytic solution , we investigated how changes in baseline ion conductance for the major ions in our model ( gK , gNa and gCl ) affected Cl- homeostasis . We calculated the steady-state values for the Cl- reversal potential ( ECl ) and K+ reversal potential ( EK ) , resting membrane potential ( Vm ) and volume ( w ) whilst independently manipulating the conductance for each ion ( Figure 2 ) . Increasing the baseline K+ conductance ( gK ) resulted in ECl , EK and Vm converging to similar steady-state values ( Figure 2A ) without significantly affecting cell volume . We were also able to replicate the classic dependence of membrane potential on log ( [K+]o ) ( Figure 2—figure supplement 1 ) . In contrast , increasing the baseline Na+ conductance ( gNa ) beyond 20 µS/cm2 resulted in a steady increase of ECl , Vm and volume with a minimal increase of EK ( Figure 2B ) . EK is maintained in the face of increased passive K+ efflux accompanying membrane depolarization due to increased active influx of K+ by the Na+-dependent ATPase which increases its effective pump rate due to increased intracellular Na+ concentration with larger gNa ( Figure 2—figure supplement 2 ) . The effect of manipulating Cl- conductance ( gCl ) depended on the activity of concurrent cation-chloride cotransport by KCC2 ( Figure 2C ) . In the presence of active KCC2 at very low values of gCl , the steady state [Cl-]i is such that ECl approaches EK . This follows because in the absence of alternative Cl- fluxes , KCC2 utilizes the transmembrane K+ gradient to transport Cl- until ECl equals EK . With increasing gCl however , ECl increases , moving away from EK toward Vm , and at very high Cl- conductances ECl and Vm approached similar values in our model . Without the activity of KCC2 , any non-zero gCl had no effect on steady state ECl , EK , Vm or volume ( Figure 2D ) . In this instance ECl always equals Vm as the movement of Cl- across the membrane is purely passive . Without the activity of KCC2 , there can be no driving force for Cl- flux at steady state ( Vm-ECl = 0 ) . Our model therefore behaved in a manner consistent with established theoretical predictions ( Kaila et al . , 2014 ) . Next , we used our single-cell unified model to explore how the activity of cation-chloride cotransport affects Cl- homeostasis . In our model , the activity of KCC2 is set by the conductance of KCC2 ( gKCC2 ) . Using the numerical formulation with the default values described in Figures 1 and 2 , we steadily increased gKCC2 from 20 µS/cm2 to 370 µS/cm2 and tracked changes to ECl , EK , Vm and volume . Increasing KCC2 activity over time caused a steady decrease in [Cl-]i reflected by a hyperpolarization of ECl ( Figure 3A ) . Vm decreased only modestly , resulting in an increase in the driving force for Cl- flux that tracks the increase in gKCC2 . This effect saturates as EK constitutes a lower bound on ECl . Importantly , increases in gKCC2 resulted in persistent changes to ECl and the driving force for Cl- . Employing alternate models for KCC2 ( Fraser and Huang , 2004; Lewin et al . , 2012; Raimondo et al . , 2012 ) and the Na+/K+-ATPase ( Hamada et al . , 2003 ) did not change this result when compensation for parameterization was given , although different KCC2 models result in different kinetic rates for Cl- and K+ transport ( Figure 3—figure supplements 1 and 2 ) . Using the analytic solution to our model , we calculated how KCC2 activity affects steady state values of ECl , EK , Vm , volume and Cl- driving force ( Figure 3B ) . In confirmation of our findings in Figure 2D , with no KCC2 activity ( gKCC2 = 0 ) , ECl equaled Vm and the Cl- driving force was zero . As we increased gKCC2 , steady state ECl pulled away from Vm and approached EK . This resulted in an increase in Cl- driving force ( Vm-ECl ) with steady state values of 11 . 3 mV at our chosen default value of gKCC2 . The results obtained with our model are therefore fully consistent with the view that CCCs , in this case KCC2 , establish the driving force for Cl- . To test this theoretical finding , we performed gramicidin perforated patch-clamp recordings from CA3 hippocampal neurons in rat organotypic brain slices whilst activating Cl- permeable GABAA receptors with muscimol ( 10 µM ) , in order to measure the GABAAR driving force , which approximates Cl- driving force ( Figure 3C ) . We then tested our model predications by applying the CCC blocker , furosemide ( 1 mM ) ( Figure 3D , E ) . We noted that after furosemide was introduced the EGABA became significantly more depolarized ( baseline: −78 . 8 ± 2 . 8 mV vs furosemide: −71 . 6 ± 3 . 0 mV , n = 10 , p=0 . 01 , paired t-test ) , whilst there was no significant difference in Vm ( −69 . 9 ± 1 . 8 mV vs −71 . 5 ± 2 . 6 mV , n = 10 , p=0 . 36 , paired t-test ) ( Figure 3E ) . This reflects a significant change in the GABAAR driving force ( 8 . 9 ± 3 . 4 mV vs 0 . 1 ± 4 . 6 mV , n = 10 , p=0 . 04 , paired t-test ) . Values were stable prior to baseline recordings with EGABA 5 min prior to furosemide application at −78 . 4 ± 8 . 1 mV , Vm−70 . 1 ± 5 . 6 mV and DF at 9 . 1 ± 3 . 2 ( see Figure 3E ) . We then noted that this change in driving force persisted for at least 15 min post-application of furosemide ( Figure 3E ) . These results are consistent with our model predictions and demonstrate how the application of a CCC blocker reduces the GABAAR driving force ( and hence Cl- driving force ) by selectively depolarizing EGABA with negligible effects on Vm . In addition , we sought experimental data from the literature to determine whether changes in KCC2 activity correlate with alterations to steady-state [Cl-]i . We focused on changes in KCC2 expression level , as this is likely to be a strong predictor of changes in KCC2 activity . Indeed , in a meta-analysis of seven studies and eight experiments from our review of 26 studies , weighted for methodological biases and data quality , we observed a significant correlation ( R2 = 0 . 796 , p<0 . 001 ) between the change in KCC2 expression and Cl- driving force ( Figure 3F ) . Absolute changes in Vm were less than 2 mV in all but one study , meaning that the change in driving force could be ascribed to significant shifts in EGABA ( R2 = 0 . 045 , p<0 . 001 ) . The outlier data point ( showing a 8 . 45 mV change ) was from a study into the effects of acute stress , where other factors could have transiently influenced Vm ( MacKenzie and Maguire , 2015 ) . The meta-analysis supports the prediction that cation-chloride cotransport by KCC2 is an important determinant of [Cl-]i ( R2 = 0 . 83 , p<0 . 001 , nine studies ) and driving force ( see Supplementary file 2 Table S2-1 for raw data , and the scoring table for weighting in Table S2-2 ) . To determine the effect of impermeant anions on Cl- homeostasis , we first explored whether adjusting the concentration of impermeant anions ( [X]i ) , while maintaining a constant mean impermeant ion charge ( z ) , had any impact on ECl , EK , Vm or volume . The mean charge ( z ) is the mean of the charge of all the different species of impermeant molecules in the cell , including uncharged ones , where charge is the difference between the number of protons and electrons of a molecule . Impermeant anions are more abundant than impermeant cations , and so in this manuscript we often refer to the group as impermeant anions rather than impermeant ions or impermeant molecules . For example , were there α impermeant molecules of charge −1 and β impermeant molecules of charge 0 , then z would be −αα+β . We initiated the full single-compartment model with different starting concentrations of impermeant anions all with the same mean charge , z = −0 . 85 , and observed that regardless of the initial concentration of impermeant anions , over a period of minutes , the cell adjusted its volume to give an identical steady-state impermeant anion concentration ( Figure 4A , [A]i = 155 mM ) . Analytically , it can be shown that the number of moles of X determines completely the volume of the compartment , while the permeant ions alone cannot be used to predict steady state volume ( Kay , 2017 ) . Similarly , all initial impermeant anion concentrations resulted in identical steady state values of ECl ( −83 . 8 mV ) , EK ( −95 . 1 mV ) and Vm ( −72 . 6 mV ) ( Figure 4B ) . This shows that simply adjusting the amount of impermeant anions within a cell has no persistent effect on [Cl-]i . We then tested the effect of dynamically adding impermeant anions with the default mean charge either intracellularly ( Figure 4C ) or extracellularly ( Figure 4D ) . While impermeant anions are being added to the cell , the membrane potential hyperpolarizes and ECl decreases . However , following the cessation of impermeant anion influx , ECl , EK , Vm and [X]i return to steady state values due to compensatory changes to cell volume ( Figure 4C ) . There are transient transmembrane fluxes of all ions while anions are added into the cell , and in particular the inward flux of the cations Na+ and K+ , such that the sum [X]i + [Cl-]i is not necessarily kept constant during impermeant anion addition ( Figure 4—figure supplement 1 ) . Impermeant ions ( with the default mean charge ) were added to the extracellular space , which is effectively an infinite bath in the model , while proportional decreases in [Cl-]o were applied to correct for the changes to charge and osmotic balance . Additions in the extracellular space , similarly , resulted in a temporary depolarization of ECl and Vm , but no persistent shift in these parameters ( Figure 4D ) . The addition of extracellular impermeant anions did however result in a small compensatory decrease in cell volume secondary to the large shifts in [Cl-]i required to maintain the proportion of [Cl-]i to [Cl-]o according to the Nernst potential . In summary , there is no lasting effect on the reversal potential or driving force for Cl- if only the concentration of a neuron’s intracellular or extracellular impermeant anions is altered . This is because concentration changes alone modulate only osmoneutrality , whereas changes to intracellular charge balance affect electroneutrality and therefore the membrane and ionic potentials , which we tested next . We next sought to determine how changes in the mean charge of the impermeant ions ( z ) might influence the driving force for Cl- . Such changes in z could be associated with various cellular processes , including post-translational modifications of proteins that decrease their charge without changing the absolute number of protein molecules . To investigate this parameter , we modified the mean charge ( z ) of intracellular impermeant anions from −0 . 85 to −1 whilst measuring accompanying changes in ECl , EK , Vm and cell ( Figure 5A ) . We found that this shift to a more negative z resulted in both a transient and persistent decrease in ECl , EK and V . Importantly , the shifts in ECl were accompanied by broadly matching shifts in EK and V , which resulted in a small change in the driving force for Cl- of <0 . 2 mV . Both numerical and analytic calculation of steady state values for ECl , EK and Vm in our model showed that changing the mean charge of impermeant anions , while substantially affecting ECl , had very small effects on the driving force for Cl- ( Figure 5B ) . By shifting z within reasonable ranges for mammalian neurons ( Lodish et al . , 2009; Raimondo et al . , 2015 ) , and assuming osmo- and electro-neutrality , only shifts of <1 mV could be generated . In addition , although the absolute number of impermeant anions ( moles ) remained constant throughout the process of modifying z , cell volume shifted , and as a consequence modest alterations to the concentration of impermeant anions occurred as well . Next , instead of adjusting the charge of some of the intracellular impermeant anions as described above , we directly added new impermeant anions to the cell , which had a more negative charge than the previous mean charge ( Figure 5C and D ) . This had the effect of both increasing the absolute quantity of impermeant anions and adjusting the mean charge of impermeant anions . The ‘addition’ of impermeant anions in this way models the de novo synthesis of impermeant anion species , or their active transport into the cell . This process also resulted in both transient and persistent changes to ECl , EK and V , which was dependent on the extent that z was altered . Again , whilst the large additions of impermeant anions could substantially alter the Cl- reversal potential , this had a negligible effect on the driving force for Cl- due to matching shifts in Vm . Driving shifts in ECl in this manner also resulted in changes to cell volume ( Figure 5—figure supplement 1 ) . To experimentally test our biophysical modeling predictions , we used photo-activation of ChR2 expressing GABAergic interneurons and whole-cell patch clamp recordings of mouse organotypic CA3 hippocampal pyramidal neurons to measure Vm , EGABA , and GABA driving force ( which approximates Cl- driving force ) before and after addition of impermeant anions ( Figure 5D ) . To add impermeant anions to the recorded cell , we used single-cell electroporation of fluorescently tagged anionic dextrans ( Alexa Flour 488 , see Materials and methods ) . Successful addition of impermeant anions could be confirmed visually by observing strong and stable fluorescence of the anionic dextran restricted to the recorded cell ( Figure 5D , E ) . To drive impermeant anions into the cell , negative voltage steps ( 20 ms , 0 . 5–1 V ) were applied to the electroporation pipette necessarily resulting in direct membrane depolarization , which recovered over a period of 1–5 min ( Figure 5D ) . Once this acute perturbation had settled , as predicted by our model , changing the mean charge of impermeant anions , following the addition of highly negatively charged dextrans to the cell , resulted in a stable , mean negative shift in Vm from a baseline of −67 . 0 ± 4 . 0 mV to −74 . 0 ± 3 . 1 mV ( n = 6 , p=0 . 03 , Wilcoxon test , Figure 5E , F ) . Again in line with our predictions , addition of impermeant anions also resulted in a significant reduction of resting EGABA ( a proxy for ECl ) from baseline values of −72 . 5 ± 2 . 0 mV to −77 . 3 ± 1 . 4 mV ( p=0 . 03 , Wilcoxon test ) . Importantly , however , similar shifts in Vm and EGABA resulted in an undetectable shift in GABA and hence Cl- driving force ( 5 . 5 ± 4 . 4 mV vs 3 . 3 ± 3 . 5 mV , n = 6 , p=0 . 22 , Wilcoxon test , Figure 5E , F ) . Our single-compartment model of Cl- homeostasis , in conjunction with experimental validation , demonstrates that whilst the adjustment of mean impermeant anion charge can significantly affect ECl , this results in negligible changes to the driving force for Cl- . This contrasts with the results shown earlier , where adjusting the activity of cation-chloride cotransport modulates both ECl and the driving force for Cl- substantially . We next set out to determine how , and under what conditions , the modification of impermeant anions could potentially generate the very small persistent shifts in Cl- driving force we observed in our models . Due to their small size ( <1 mV ) these were not detectable during the experimental validation . First , we repeated the simulation performed in Figure 5A by changing the mean charge of impermeant anions in the cell , but under conditions where the Na+/K+-ATPase effective pump rate ( Jp ) was either a cubic function of the transmembrane Na+ gradient ( default condition ) or was fixed at a constant value ( Figure 6A ) . In the case where the pump rate was fixed , adjusting the mean charge of impermeant anions generated no persistent change in Cl- driving force ( Figure 6A ) . Modifying impermeant anions caused a significant change in steady-state intracellular Na+ concentration when Jp was kept constant . However , small shifts in Cl- driving force occurred only when the effective pump rate was variable , in which minor changes to [Na+]i caused significant changes to Jp , which in turn resulted in a small shift in Cl- driving force . There is a direct relationship between the mean charge of impermeant anions ( z ) , [Na+]i , the effective Na+/K+-ATPase pump rate and Cl- driving force . This relationship was abolished when the effective Na+/K+-ATPase pump rate was held constant by removing its dependence on Na+ ( Figure 6B ) . In addition , even large variations in effective pump rate near the default value caused negligible shifts in Cl- driving force of <1 mV . These results were similar when using the experimentally matched ATPase model by Hamada et al . ( 2003 ) , with slight differences in final values ( Figure 6—figure supplement 1A ) . These small , impermeant anion driven , Na+/K+-ATPase pump-dependent shifts in Cl- driving force are completely dependent on the presence of cation-chloride cotransport . In the absence of KCC2 , there is no Cl- driving force as ECl = Vm ( Figure 6F ) . We then tested whether relaxing the condition of transmembrane osmoneutrality might also alter impermeant anion induced effects on Cl- driving force . We modeled a situation where increases in cell surface area beyond a certain ‘resting’ surface area generated a hydrostatic pressure ( membrane tension ) , which could balance an osmotic pressure difference of 10 mM between the intra- and extracellular compartments ( see Figure 6C , schematic in Figure 6E and Materials and methods ) . In this case , adding impermeant anions of default charge z resulted in constrained increases in cell volume , which were accompanied by persistent transmembrane differences in osmolarity and intracellular Na+ concentration . This was sufficient to generate small differences in driving force for Cl- of <0 . 2 mV for reasonable increases in cell surface area ( Nichol and Hutter , 1996; Dai et al . , 1998 ) . Again , this was entirely due to Na+ driven shifts in the Na+/K+-ATPase effective pump rate . By removing the dependence of Na+/K+-ATPase activity on Na+ concentration , addition of impermeant anions no longer generated persistent shifts in Cl- driving force ( Figure 6C ) . Using the experimentally-matched ATPase model by Hamada et al . ( 2003 ) generated similar results ( Figure 6—figure supplement 1B ) because the model is also directly dependent on [Na+]i . We observed a direct relationship between transmembrane osmotic gradient , [Na+]i , the effective Na+/K+-ATPase pump rate and Cl- driving force . This relationship was removed when the effective Na+/K+-ATPase pump rate was held constant , with no changes in Cl- driving force seen despite the generation of the same shift in the transmembrane osmotic gradient ( Figure 6D ) . In summary , changes in Cl- driving force generated by changing the ionic contributions to cellular charge ( by altering the mean charge of impermeant anions ) or osmoneutrality ( by increasing the contribution of hydrostatic pressure ) are due to the alteration of the dynamics of active ion transport mechanisms in the cell . However , these effects are negligible in magnitude and cannot contribute significantly to setting physiologically observed Cl- driving forces . It is worth reiterating that any non-zero Cl- driving force is entirely dependent on the presence of active Cl- cotransport . In our model , in the absence of KCC2 , neither the Na+/K+-ATPase nor impermeable anions can shift Cl- out of equilibrium ( Figure 6F ) . An important functional question is how Cl- driving force might be modified at a local level within a neuron . We considered local persistent changes of Cl- driving force for the case of active transmembrane Cl- fluxes ( Figure 7 ) and impermeant anions ( Figure 8 ) by extending the single-compartment model described above into a multi-compartment model or ‘virtual dendrite . ’ This dendrite was 100 μm in length and consisted of 10 compartments , each of 10 μm length and 1 μm diameter . The compartments contained the same mechanisms and default parameterization as the single compartment model described above . Compartmental volume was changed by altering the radius , while holding the length constant . In addition , all ions , except impermeant anions , could move between compartments by electrodiffusion ( Figure 7A and Materials and methods ) . To explore the local effects of CCC activity , we increased gKCC2 from our default value of 20 µS/cm2 to 600 µS/cm2 in the second distal compartment of the virtual dendrite exclusively . This resulted in a persistent decrease in ECl , concurrent with a modest decrease in Vm , resulting in a permanent increase in Cl- driving force and minimal change in compartment volume ( Figure 7B ) . The spatial precision of this alteration depended strongly on the diffusion constant for Cl- . With a Cl- diffusion constant of 2 . 03 × 10−7 dm2 . s−1 , these alterations spread widely through the virtual dendrite . For example , the change in Cl- driving force was 4 . 8 mV in the furthermost compartment ( 90 μm apart ) as compared to 5 . 9 mV in the compartment manipulated . When we decreased the Cl- diffusion constant by one order of magnitude , the change in Cl- driving force was 7 . 3 mV in the compartment in which KCC2 was adjusted , but only 1 . 8 mV in the furthermost compartment from the site of manipulation ( Figure 7C ) . These findings suggest that local differences in cation-chloride cotransport activity can drive spatially restricted differences in Cl- driving force under conditions of constrained Cl- diffusion; however , under conditions of typical ionic diffusion the effect of Cl- transport by KCC2 is relatively widespread . Following the last result , we considered whether changing impermeant anions in part of a dendrite could create a local area with a different Cl- driving force compared to the rest of the cell . We first added impermeant anions of the default charge ( z = −0 . 85 ) exclusively to the second-most distal compartment of the virtual dendrite while measuring the Cl- reversal , Vm and Cl- driving force in all compartments . During addition of the impermeant anions , ECl and Vm decreased with an accompanying decrease in Cl- driving force . However , following cessation of impermeant anion influx , all parameters returned to baseline levels , except for the volume of that specific compartment , which showed a modest increase ( Figure 8B ) . This suggests that local addition of impermeant anions of mean charge has no local effect on Cl- homeostasis but can affect the volume of the compartment concerned . Next , we again added impermeant anions to the second-last compartment of the virtual dendrite , but this time we added impermeant anions with a more negative charge than ( z = −1 ) than the current mean . This resulted in the mean charge of impermeant anions in that compartment becoming more negative ( Figure 8C ) . During the addition of the impermeant anions , ECl and Vm decreased across the dendrite , but with small accompanying shifts in Cl- driving force . Following cessation of local impermeant anion influx , a persistent shift in ECl and Vm was observed specifically in the compartment manipulated . However , this generated a negligible , persistent change in Cl- driving force ( <0 . 01 mV change in driving force for a compartment specific change in z from −0 . 85 to −0 . 93 ) , only within that specific compartment of the virtual dendrite . Again , impermeant anion addition resulted in a permanent increase in the volume of the compartment concerned . This finding suggests that local impermeant anions can adjust the Cl- reversal potential locally , but are not well-placed to cause significant , permanent shifts in the driving force for Cl- . Indeed , electrodiffusion may further limit the degree to which local changes in impermeant ion charge can modify driving forces through alterations in active ionic transport: the resulting permanent Cl- driving force changes in the multi-compartment model are many times smaller than the shifts in the single compartment version ( as compared to Figure 6B ) .
The driving force for Cl- is a fundamental parameter affecting the excitability of neuronal networks ( Raimondo et al . , 2017 ) . Recently , impermeant anions , rather than CCCs , have been suggested as the primary determinants of the neuronal driving force for Cl- ( Glykys et al . , 2014 ) . Here , we have explored the determinants of the Cl- driving force in neurons by deriving theoretical models based on biophysical first principles . We show that the Na+/K+-ATPase , baseline K+ , Na+ and Cl- conductances , mean charge of impermeant anions , water permeability and CCCs , likely all play roles in setting neuronal [Cl-]i . However , our findings suggest that while impermeant anions can contribute to setting the [Cl-]i in neurons , they can only affect Cl- driving force by modifying the activity of active transport mechanisms ( i . e . the Na+/K+-ATPase ) . Our modelling and experimental data demonstrate that under physiologically relevant conditions , impermeant anions do not alter the Cl- driving force significantly . In contrast , CCCs are well placed to modulate Cl- driving force and hence inhibitory signaling . Previous theoretical models , which account for the dynamics of Cl- ions , have been useful in determining how changes to the driving force for Cl- are critical for controlling the effect of synaptic inhibition in the brain ( Qian and Sejnowski , 1990; Staley and Proctor , 1999; Doyon et al . , 2011; Jedlicka et al . , 2011; Lewin et al . , 2012; Mohapatra et al . , 2016 ) . Whilst these models have included the Na+/K+-ATPase , the interacting dynamics of several ion species , CCCs ( Doyon et al . , 2011; Krishnan and Bazhenov , 2011 ) , electrodiffusion ( Qian and Sejnowski , 1989 ) and impermeant anions and volume regulation ( Dijkstra et al . , 2016 ) , none have combined all these mechanisms to explore how their combination determines the local driving force for Cl- . Our theoretical approach is based on the pump-leak formulation ( Tosteson and Hoffman , 1960 ) . It suggests that mammalian cells maintain their volume under osmotic stress generated by impermeant anions and the Donnan effect by employing active transport of Na+ and K+ using the Na+/K+-ATPase ( Armstrong , 2003; Kay , 2017 ) . A Donnan equilibrium , a true thermodynamic equilibrium requiring no energy to maintain it , is not possible in cells with pliant membranes like neurons ( Sperelakis , 2012 ) . Our model conforms to the pump-leak formulation: abolishing the activity of the Na+/K+-ATPase leads to cell swelling , progressive membrane depolarization and rundown of ionic gradients , including that of Cl- . Therefore , the Na+/K+ ATPase is a fundamental cellular parameter that stabilizes cell volume and determines all ionic gradients including that of Cl- and hence must be considered in any attempt to model ion homeostasis . Interestingly however , we demonstrate that above a certain level of Na+/K+-ATPase activity , even many fold changes in pump rate have minimal effects on volume , ECl and Vm . This might explain recent experimental findings in which periods of Na+/K+ ATPase inhibition using ouabain caused modest changes to cell volume ( Glykys et al . , 2014 ) . It therefore seems unlikely that neurons adjust the Na+/K+-ATPase as a means for modulating Cl- driving force . Baseline ion conductances are another important factor that affect Cl- driving force . Our model is consistent with recent experimental results that demonstrate that increased neuronal Na+ conductance ( for example by activating NMDA receptors , or preventing closure of voltage-gated Na+ channels ) , leads to progressive neuronal swelling , membrane depolarization and Cl- accumulation ( Rungta et al . , 2015 ) – the primary pathological process in cytotoxic edema ( Liang et al . , 2007 ) . We also show that tonic neuronal Cl- conductance only affects baseline [Cl-]i and driving force in the presence of CCCs . Without active Cl- flux , which CCCs provide , there is no driving force for passive Cl- flux and hence no mechanism for [Cl-]i changes resulting from selective modification of a Cl- conductance . This is consistent with both classic ( Misgeld et al . , 1986; Thompson and Gähwiler , 1989 ) and recent experimental findings ( Berglund et al . , 2016 ) . In our model , we find that elevating the activity of KCC2 , the most active CCC in mature neurons ( Ben-Ari , 2002 ) , increases the driving force for Cl- by shifting the reversal potential for Cl- closer to that of K+ . Interestingly , large shifts ( ~7 mV ) in driving force were associated with very minor ( 1% ) changes in volume or membrane potential . As such , modulating KCC2 represents a specific means for manipulating the neuronal Cl- driving force . This is consistent with traditional dogma , recent and previous experimental results ( Kaila et al . , 2014; Klein et al . , 2018 ) as well as our own experimental validation using furosemide to block the activity of KCC2 , which drove significant changes in driving force with little effect on Vm . In further support of this , our meta-analysis of numerous experimental studies showed a strong correlation between change in KCC2 expression and Cl- driving force , but not between KCC2 expression and Vm . There is an ongoing debate as to whether some cotransporters , including CCCs , might also couple water transport to ion transport ( Zeuthen , 1994; MacAulay et al . , 2002; Gagnon et al . , 2004; Charron et al . , 2006 ) . Although we did not model the active movement of water by KCC2 , this scenario would not alter the central importance of CCCs for setting the Cl- driving force . Using our multi-compartment model , which incorporated electrodiffusion , we found that local modification of KCC2 activity has a specific local effect on Cl- driving force that is dependent on the characteristics of intracellular Cl- diffusion . Cytoplasmic Cl- diffusion rates had to be reduced substantially before we observed local changes in Cl- driving force driven by KCC2 ( Qian and Sejnowski , 1989; Kuner and Augustine , 2000 ) . Whilst differences in KCC2 activity might generate a gradient in Cl- driving force between large subcellular structures ( i . e . dendrites versus soma ) , our modeling results call into question the idea of synapse-specific regulation of Cl- driving force within the same cellular domain ( Földy et al . , 2010 ) . Glykys et al . ( 2014 ) used Cl- imaging and various experimental manipulations to claim that intracellular and extracellular concentrations of impermeant anions ( [X]i and [X]o ) set [Cl-]i and the Cl- driving force . From our theoretical analysis , we find that modifying the amount of impermeant anions inside or outside neurons has no persistent effect on [Cl-]i or Cl- driving force , unless we include a mechanism that allows a transmembrane osmotic pressure differential to develop that indirectly affects active transport mechanisms . Even in this case , under transmembrane pressure differentials that do not lyse the membrane ( Nichol and Hutter , 1996 ) , Cl- driving force changes are negligible ( <1 mV ) . Recently , it has been suggested that the viscoelastic properties of the cellular cytoskeleton could allow it to take up osmotic shifts created by impermeant anion movement like a sponge ( Sachs and Sivaselvan , 2015 ) . This would mean that one would not see as large a volume shifts as predicted by our models . In our model , we have assumed that water can pass through the neuronal membrane to equalize osmotic differences . Although it is thought that some neurons do not express aquaporin channels ( Andrew et al . , 2007 ) , water can permeate the phospholipid bilayer ( Fettiplace and Haydon , 1980 ) . Therefore , whilst differences in neuronal water permeability might affect the time taken to reach steady-state , the steady state values themselves are unlikely to be affected . We conclude that [Cl-]i and the Cl- driving force are not determined by the concentration of impermeant anions . However , our theoretical findings offer a potential explanation for recent experimental observations . We show that modifying the mean charge of impermeant anions ( i . e . z in [Xz]i ) , rather than their concentration , can affect [Cl-]i and ECl . Relating this to prior experimental observations , Glykys et al . ( 2014 ) used SYTO64 staining of nucleic acids and perfusion of weak organic acids in conjunction with Cl- imaging to suggest that [Cl-]i depends upon internal impermeant anions ( [X]i ) . If such a manipulation modifies the mean charge of internal impermeant anions , and not concentration per se , this could account for the observed changes in [Cl-]i . Glykys et al . ( 2014 ) did not measure Vm or the Cl- driving force in these experiments . The clear prediction from our model is that any manipulation , which changes the mean charge of impermeant anions would not appreciably affect the Cl- driving force because any impermeant anion driven change on ECl- is matched by an equivalent effect on Vm due to accompanying shifts in cation concentrations . We have provided experimental support for this prediction by showing that whilst EGABA ( and ECl ) can be shifted by addition of impermeant anions using electroporation of membrane impermeant anionic dextrans , Vm is shifted in a similar direction resulting in an undetectable change in Cl- driving force . Future experiments could further test our model by electroporating positively charged dextrans which would be predicted to depolarize both Vm and ECl , again with minimal effects on Cl- driving force . Given prior theoretical predictions ( Kaila et al . , 2014; Voipio et al . , 2014; Savtchenko et al . , 2017 ) , it is interesting that our model reveals that changing impermeant anions could affect the Cl- driving force at all . We found that the small ( <1 mV ) impermeant anion-driven changes in Cl- driving force observed in our model were caused by indirect effects on Na+ concentration and hence Na+/K+-ATPase activity . The impermeant anion-driven changes in Cl- driving force are even smaller in the multi-compartment model ( <0 . 1 mV ) , in which electrodiffusion allows local changes in Na+ to dissipate . When Na+/K+-ATPase activity was decoupled from the transmembrane Na+ gradient , we found that impermeant anions were unable to cause persistent shifts in Cl- driving force as predicted theoretically ( Kaila et al . , 2014; Voipio et al . , 2014; Savtchenko et al . , 2017 ) . It is important to note that these small , impermeant anion-Na+/K+-ATPase-driven shifts in Cl- driving force are dependent on the presence of cation-chloride cotransport in the form of KCC2 and would entail changes in energy use by the Na+/K+-ATPase . In other words , active transport mechanisms are again required to drive changes in Cl- homeostasis . In summary , our theoretical models , which are derived from well-established physical principles , are consistent with our own experimental data and that of others ( Glykys et al . , 2014; Kaila et al . , 2014; Klein et al . , 2018 ) , and suggest that impermeant anions alone cannot shift Cl- out of equilibrium across the neuronal membrane . Were neurons to alter impermeant anion concentration or charge , the resting membrane potential would be modified with little effect on the Cl- driving force . Our work confirms the central importance of CCC activity in determining the effects of inhibitory synaptic transmission in the nervous system .
The single-compartment model consisted of a cylindrical semipermeable membrane separating the extracellular solution from the intracellular milieu with variable volume ( Figure 1 ) . The extracellular ionic concentrations were assumed constant ( Table 1 ) . Permeable ions in the model were K+ , Na+ and Cl- with their usual charges , while impermeant anions X were assumed to be a heterogeneous group of impermeant chemical species with a mean intracellular charge z and a mean extracellular charge -1 . The default z ( -0 . 85 ) was chosen on the basis of known resting intracellular ion concentrations ( Lodish et al . , 2009; Raimondo et al . , 2015 ) and osmolarity ( Π ) . Bicarbonate ions were not included in our model as a permeant anion as they were assumed to be important for acute depolarizing effects ( via GABAARs ) rather than the chronic shifts in Cl- driving force , which are the focus of this work ( Staley and Proctor , 1999 ) . The model included ionic leak currents for the permeable ions , Na+/K+-ATPase transporters and a CCC , in this case the K+-Cl- cotransporter ( KCC2 ) . KCC2 and not NKCC1 is thought to be the most active CCC in mature neurons ( Ben-Ari , 2002 ) , therefore , to maintain conceptual simplicity only KCC2 was modeled . Cell volume ( w ) change was based on osmotic water flux and incorporated a membrane surface area scaling mechanism . An analytical solution to the model at steady state was derived using standard techniques and can be found in Supplementary file 1 . The numerical model was initialized assuming conditions close to electroneutrality and an osmotic equilibrium between the intracellular and extracellular compartments . A forward Euler approach was used to update variables at each time step ( dt ) of 1 ms . Using a smaller dt did not influence the results in Figure 1–5 . Code was written in Python 2 and is available on GitHub ( Düsterwald and Currin , 2018; copy archived at https://github . com/elifesciences-publications/model-of-neuronal-chloride-homeostasis ) . The GitHub repository includes a file of Supplementary figures , in which we display transmembrane fluxes of all ions and water for relevant simulations . An example figure displaying ionic flux for all ions is available for Figure 4C in Figure 4—figure supplement 1 . The membrane potential Vm was based on the ‘Charge Difference’ approach ( Rybak et al . , 1997; Fraser and Huang , 2004 ) as follows: ( 1 ) Vm= F ( [Na+]i+[K+]i-[Cl-]i+z[Xz]i ) CmAmwhere F is Faraday’s constant , Cm is the unit membrane capacitance and Am is calculated as the ratio of the surface area ( of the cylinder ) to cell volume . The term in brackets is the sum of all ionic charges within the cell . This approach has the advantage that the initial voltage can be calculated without needing to assume a steady state as is required for by the Goldman-Hodgkin-Katz ( GHK ) equation . Intracellular concentrations of the permeable ions Na+ , K+ and Cl- were updated individually by summing trans-membrane fluxes . Leak currents were calculated using Ohm's Law , I=g ( Vm-E ) . In addition , Na+ and K+ were transported actively by the Na+/K+-ATPase , with pump rate Jp , which was approximated by a cubic function dependent on the transmembrane sodium gradient , following ( Keener and Sneyd , 1998 ) : ( 2 ) Jp=P[Na+]i[Na+]o3 , where P is the pump rate constant . Because it is a function of the sodium gradient , Jp decreases as [Na+]i depletes . This formulation has been shown to be similar to more accurate kinetic models reliant on both the Na+ gradient and ATP concentration ( Keener and Sneyd , 1998 ) . To switch the ATPase pump on or off ( Figure 1C ) , P was decreased/increased exponentially over 10–20 min , consistent with previous reports of the dynamics of inhibition of the ATPase by ouabain and in turn the inhibition of ouabain’s effects by potassium canrenoate ( Baker and Willis , 1972; Yeh and Lazzara , 1973 ) . The ATPase pumps 2 K+ ions into the cell for every 3 Na+ ions out and these constants must be multiplied by Jp for each ion , respectively . K+ and Cl- were also modified by flux through the type 2 K-Cl cotransporter ( KCC2 ) , which has a stoichiometry of 1:1 and transports both ions in the same direction . Flux though KCC2 , JKCC2 ( Doyon et al . , 2016 ) , was modeled as follows: ( 3 ) JKCC2=gKCC2EK-ECl , where gKCC2 is a fixed conductance and EK and ECl are the Nernst potentials for K+ and Cl- respectively . JKCC2 is 0 when EK=ECl . The rate of change of the intracellular concentration of the three permeant ions was given by the following equations , with the Nernst potentials for each ion given by Eion=RTzFln ( [ion]o[ion]i ) , w indicating the cell volume , and dwdt as described in Equation 7 ( Figure 1–5 , 6A , B , 7 and 8 or 9 or Equation 9 ( Figure 6C–E ) : ( 4 ) d[Na+]idt=-AmFgNaVm-ENa+3Jp-1wdwdt[Na+]i ( 5 ) d[K+]idt=-AmFgKVm-EK-2Jp-JKCC2-1wdwdt[K+]i ( 6 ) d[Cl-]idt=AmFgClVm-ECl+JKCC2-1wdwdt[Cl-]i . In most calculations , because the osmotic flux of water is expected to be faster than ion fluxes , the volume of the cell ( w ) was adjusted to reduce the difference between Πi ( intracellular osmolarity ) and Πo ( extracellular osmolarity ) at each time step by explicitly modelling water flux , where vw is the partial molar volume of water , pw the osmotic permeability of a biological membrane and SA the surface area ( Hernández and Cristina , 1998 ) :dwdt=vw∙pw∙SA∙ ( Πi-Πo ) . In some calculations ( Figure 6C–E ) in which we allowed transmembrane differences in osmolarity to develop , we assumed that at rest the cylindrical cell had a radius of ra and zero pressure across the membrane , and that the tension ( T ) in the membrane followed Hooke’s law such that the tension was proportional to the difference between the dynamic circumference of the cell and that of the resting state . From Laplace’s law the hydrostatic pressure in the cell was given by: ( 8 ) Hp={4πkm ( 1−rar ) r>ra0 , otherwisewhere km is the spring constant of the membrane ( Sachs and Sivaselvan , 2015 ) . Equation 7 was thus reformulated: ( 9 ) dwdt=vw∙pw∙SA∙Πi-Πo-HpRT . In order to simulate extreme conditions of constrained volume , a larger km was employed than is realistic ( Dai et al . , 1998 ) . Intracellular ion concentrations were updated again after volume change at each time step . Volume changes were manifested in the cylindrical compartment as change in the radius . In Figures 1–6 , the cell was initialised with diameter 10 μm and length 25 μm . Impermeant anions were manipulated in the compartment in Figures 4–6 and Figure 8 through several mechanisms . Anions could be added to the compartment at a constant rate and have either the default intracellular X charge z = −0 . 85 ( Figures 4C , 6C and 8B ) , or a different charge ( Figures 5C , 6A and 8C ) . In these cases , the number of moles of X in the compartment was increased . Alternatively , the charge of a species of intracellular X was slowly changed imitating a charge-carrying trans-membrane reaction ( Figures 5A and 6A ) . In this case , the number of moles of intracellular X did not change and it was assumed charge imbalance was compensated by the extracellular milieu . Finally , extracellular X- was changed in Figure 4D by removing as much Cl- as X- was added , thus maintaining osmolarity and electroneutrality in the extracellular space . The single-compartment dendrite model was incorporated in a multi-compartment model by allowing electrodiffusion to occur between individual compartments operating as described above . Compartments were initialised with a radius of 0 . 5 μm and length of 10 μm . Compartments were linked linearly without branching; 10 connected compartments in total were used . The time step dt was decreased to 10−3 ms for simulations in multiple compartments . Code was written in Python 3 and is available on GitHub ( Düsterwald and Currin , 2018 ) . Electrodiffusion . The Nernst-Planck equation ( NPE ) was used to model one-dimensional electrodiffusion , based on Qian and Sejnowski ( 1989 ) . The NPE incorporates fluxes because of diffusion and drift ( i . e . the movement of ions driven by an electric field ) . It has been shown to be more accurate than using Jdiffusion alone in small structures like dendrites ( Qian and Sejnowski , 1989 ) . The NPE for J the flux density of ion C is calculated as: ( 10 ) J=-DzF2RTCdVmdx-DFdCdx , where D is the diffusion constant of ion C ( Table 1 ) , z is its charge , [C] is its concentration and x is the distance along the longitudinal axis over which electrodiffusion occurs . The NPE was implemented between compartments i and i + 1 , assuming the i→i + 1 direction was positive , using a forward Euler approach . The midpoints of the compartments were used to calculate dx , i . e . dx=hi+hi+12 , and the concentrations of C in each compartment were averaged to obtain Jdrift , ensuring that Ji→i+1 = Ji+1→i , where the fluxes had units of mol/ ( s . dm2 ) : ( 11 ) Ji→i+1=-D zFRTCi+Ci+12dVmdx+d[C]dx . The flux was multiplied by the surface area between compartments and then divided by compartment volume to determine the flux in terms of molar concentration ( M/s ) , i . e . πr2πr2hi=1hi , and finally implemented numerically with a forward Euler approach . The implementation mirrored that in Qian and Sejnowski ( 1989 ) for non-branching dendrites , but was adjusted for compartments whose volumes can change: ( 12 ) Ci→i+1=-dthiD zFRTCi+Ci+12Vmi-Vmi+1dx+Ci-Ci+1dx . A literature search was performed to identify experimental studies that aimed to correlate a change in KCC2 expression with changes in [Cl-]i . The MEDLINE database was used and accessed via the PubMed online platform . Search terms included ‘chloride’ , ‘Cl’ , ‘intracellular’ , ‘KCC2’ , ‘cotransporter’ , ‘neuronal’ , ‘GABA’ using appropriate Boolean operators . All 26 studies that demonstrated changes in KCC2 expression and EGABA were considered for the meta-analysis . As there is a well-described differential expression of KCC2 and NKCC1 at different stages of development , with KCC2 expression increasing and NKCC1 expression decreasing across development , only studies that used tissue older than postnatal day seven were included ( eight included data from younger animals ) . Other exclusion criteria included: reporting a significant change in NKCC1 ( five studies ) ; use of non-rodent tissue ( two studies ) ; no quantification of the change in KCC2 ( two studies ) . Nine experiments from eight studies met all criteria and were included ( Coull et al . , 2003; Lagostena et al . , 2010; Lee et al . , 2011; Ferrini et al . , 2013; Campbell et al . , 2015; MacKenzie and Maguire , 2015; Mahadevan et al . , 2015; Tang et al . , 2015 ) . However , one study did not report the change in Vm and hence was excluded from the figure ( Mahadevan et al . , 2015 ) . Data used in regression can be seen in Supplementary file 2 ( Table S2-1 ) and includes the analysis for regression against change in [Cl-]i . To accommodate varied experimental preparations and techniques influencing data quality and biases , a 34-point scoring system was designed to weight the studies ( Table S2-2 ) . A weighted least squares regression model was then used to correlate the percentage ( % ) change in KCC2 expression versus change in Cl-d driving force . For all experiments , organotypic slices were prepared from rodent brain tissue . Wistar rats were used for the experiments testing the effects of CCC blockade . However , to allow for optogenetic manipulation , a crossed mouse strain on a C57BL/6 background was used . This mouse strain was a cross between mice expressing cre-recombinase in glutamic acid decarboxylase 2 ( GAD2 ) positive interneurons ( GAD2-IRES-cre , Jax Lab strain 010802 ) and mice with a loxP-flanked STOP cassette preventing transcription of the red-fluorescent protein tdTomato ( a cre-reporter strain , Ai4 , Jax Lab strain 007914 ) . This created the GAD2-cre-tdTomato strain which resulted in cre-recombinase and tdTomato expression in all GABAergic interneurons ( Taniguchi et al . , 2011 ) . The use of all animals was approved by either the University of Oxford ( rat ) or the University of Cape Town ( mouse ) animal ethics committees . Organotypic brain slices were prepared using 7 day old Wistar rats ( CCC experiment ) or crossed GAD2-IRES-cre ( RRID:IMSR_JAX:010802 ) and Ai14 tdtomato reporter ( RRID:IMSR_JAX:007914 ) mice ( impermeant anion experiment ) and followed the protocol originally described by Stoppini et al . , 1991 . Briefly , brains were extracted and swiftly placed in cold ( 4°C ) dissection media consisting of Gey’s Balanced Salt Solution ( GBSS #G9779 , Sigma-Aldrich , USA ) supplemented with D-glucose ( #G5767 , Sigma-Aldrich , USA ) . The hemispheres were separated and individual hippocampi were removed and immediately cut into 350 μm slices using a Mcllwain tissue chopper ( Mickle , UK ) . Cold dissection media was used to rinse the slices before placing them onto Millicell-CM membranes ( #Z354988 , Sigma-Aldrich , USA ) . Slices were maintained in culture medium consisting of 25% ( vol/vol ) Earle’s balanced salt solution ( #E2888 , Sigma-Aldrich , USA ) ; 49% ( vol/vol ) minimum essential medium ( #M2279 , Sigma-Aldrich , USA ) ; 25% ( vol/vol ) heat-inactivated horse serum ( #H1138 , Sigma-Aldrich , USA ) ; 1% ( vol/vol ) B27 ( #17504044 , Invitrogen , Life Technologies , USA ) and 6 . 2 g/l D-glucose . Slices were incubated in a 5% carbon dioxide ( CO2 ) humidified incubator at between 35–37°C . Recordings were made after 6–14 days in culture . Previous studies have shown that after 7 days in culture ( equivalent to postnatal day 14 ) GABAergic signaling and EGABA has sufficiently developed to a level resembling mature nervous tissue ( Streit et al . , 1989; Wright et al . , 2017 ) . For the impermeant anion experiment , the mouse organotypic brain slices were injected 1 day after culture with adeno-associated vector serotype 1 ( AAV1 ) containing a double-floxed sequence for channelrhodopsin ( ChR2 ) linked to a yellow fluorescent protein ( YFP ) tag driven by the elongation factor one promoter ( UNC Vector Core , USA ) . The vector was diffusely injected into slices using a custom-built Openspritzer pressurized ejection device ( Forman et al . , 2017 ) . Slices were left for 6 days in culture to allow for robust expression of ChR2-YFP in GAD2+ interneurons . For recordings , slices were transferred to a submerged chamber which was perfused with artificial cerebro-spinal fluid ( aCSF ) bubbled with carbogen gas ( 95% oxgen:5% carbon dioxide ) . aCSF was composed of ( in mM ) : NaCl ( 120 ) ; KCl ( 3 ) ; MgCl2 ( 2 ) ; CaCl2 ( 2 ) ; NaH2PO4 ( 1 . 2 ) ; NaHCO3 ( 23 ) ; D-Glucose ( 11 ) with pH adjusted to be between 7 . 35–7 . 40 using 0 . 1 mM NaOH . Neurons were visualized using a 20x or 60x water-immersion objective ( Olympus , Japan ) on a BX51WI upright microscope ( Olympus , Japan ) . Widefield images were obtained using a Mightex CCE-B013-U CCD camera . For the optogenetic experiments , we used epifluorescence microscopy to select slices that exhibited strong ChR2-YFP expression throughout the hippocampal region . Micropipettes were prepared from borosilicate glass capillaries with an outer diameter of 1 . 20 mm and inner diameter of 0 . 69 mm ( Warner Instruments , USA ) , using a horizontal puller ( Sutter , USA ) . Recordings were made using Axopatch 1D and Axopatch 200B amplifiers and acquired using pClamp9 ( Molecular Devices ) or WinWCP software ( University of Strathclyde , UK ) . For the CCC blockade experiment , gramicidin perforated patch recordings ( Kyrozis and Reichling , 1995 ) were performed using glass pipettes containing ( in mM ) : 135 KCl , 4 Na2ATP , 0 . 3 Na3GTP , 2 MgCl2 , 10 HEPES and 80 μg/ml gramicidin ( Calbiochem; pH 7 . 38; osmolarity , 290 mosmol/l ) . After obtaining a cell-attached patch , the gramicidin perforation process was evaluated by continuously monitoring the decrease in access resistance . Recordings were started once the access resistance had stabilized between 20–80 MΩ , which usually occurred 20–40 min following gigaseal formation . Rupture of the gramicidin patch , referred to as a break through , induces a large influx of the high Cl- internal solution into the cell . This causes a significant and permanent increase in the EGABA at which time recordings would be discarded . GABAAR activation was achieved by pressure application of muscimol ( 10 μM , Tocris , UK ) , a selective GABAAR agonist , via a picospritzer . To calculate EGABA , GABAAR currents were elicited at different command voltages . These were a series of 10 mV steps above and below a holding voltage of −60 mV . Reported membrane potentials were corrected for the voltage drop across the series resistance for each neuron . Holding current ( reflecting membrane current ) and total current ( reflecting membrane current plus the GABAAR-evoked current ) were plotted against the corrected holding potential to generate a current-voltage ( I-V ) curve . Using this graph , the EGABA was defined as the potential where the total current equals the holding current . Some of the data used for calculating EGABA values in Figure 3E was also used in a previous study ( Wright et al . , 2017 ) . Vm was defined as the x-intercept of the holding current , and the driving force as the difference between the two . To block CCC function , the KCC2 blocker furosemide ( 1 mM , Sigma , USA ) was applied . For the impermeant anion experiments , whole-cell recordings were utilized as electroporation consistently ruptured the gramicidin patches . Pipettes were filled with internal solution composed of ( in mM ) : K-Gluconate ( 120 ) ; KCl ( 10 ) ; Na2ATP ( 4 ) ; NaGTP ( 0 . 3 ) ; Na2-phosphocreatinine ( 10 ) and HEPES ( 10 ) . To test the effect of introducing impermeant anions , we electroporated the anionic 10 000 MW Dextran Alexa-Flour 488 ( Thermo Fisher , USA ) using a separate pipette positioned near the soma of the patched cell . This molecule is a hydrophilic polysaccharide , which is both membrane impermeant and highly negatively charged . Pipettes were filled with a 5% dextran solution in phosphate buffered saline and voltage pulses ( 5–10 of 20 ms duration , 0 . 5–1 V ) were applied using a stimulus isolator . Successful electroporation of the anionic dextran was confirmed visually by observing the cell fill with the fluorescent dye . Electroporation also resulted in immediate membrane depolarization , which recovered over a period of 1–5 mins . EGABA , Vm and driving force were calculated during voltage steps in voltage clamp mode ( as above ) with GABAARs activated via endogenous synaptic release of GABA using photo-activation of GAD2+ interneurons expressing ChR2-YFP with 100 ms pulses of blue light using a 470 nm LED ( Thorlabs ) , in the presence of 5 μM CGP-35348 ( Torcis Bioscience , UK ) to block GABABR activation . EGABA , Vm and driving force were calculated before and atleast five mins following electroporation to allow for stabilization of Vm , EGABA and driving force . The image of the dextran filled cell was acquired using a confocal microscope ( LSM510 Meta NLO , Car Zeiss , Jena , Germany ) . Analysis was performed using custom written scripts in the MATLAB environment ( MathWorks ) . Results are presented as mean ± SEM . | Cells called neurons in the brain communicate by triggering or inhibiting electrical activity in other neurons . To inhibit electrical activity , a signal from one neuron usually triggers specific receptors on the second neuron to open , which allows particles called chloride ions to flow into or out of the neuron . The force that moves chloride ions ( the so-called ‘chloride driving force’ ) depends on two main factors . Firstly , chloride ions , like other particles , tend to move from an area where they are plentiful to areas where they are less abundant . Secondly , chloride ions are negatively charged and are therefore attracted to areas where the net charge ( determined by the mix of positively and negatively charged particles ) is more positive than their current position . It was previously believed that a group of proteins known as CCCs , which transport chloride ions and positive ions together across the membranes surrounding cells , sets the chloride driving force . However , it has recently been suggested that negatively charged ions that are unable to cross the membrane ( or ‘impermeant anions’ for short ) may set the driving force instead by contributing to the net charge across the membrane . Düsterwald et al . used a computational model of the neuron to explore these two possibilities . In the simulations , altering the activity of the CCCs led to big changes in the chloride driving force . Changing the levels of impermeant anions altered the volume of cells , but did not drive changes in the chloride driving force . This was because the flow of chloride ions across the membrane led to a compensatory change in the net charge across the membrane . Düsterwald et al . then used an experimental technique called patch-clamping in mice and rats to confirm the model’s predictions . Defects in controlling the chloride driving force in brain cells have been linked with epilepsy , stroke and other neurological diseases . Therefore , a better knowledge of these mechanisms may in future help to identify the best targets for drugs to treat such conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2018 | Biophysical models reveal the relative importance of transporter proteins and impermeant anions in chloride homeostasis |
Salicylate and acetylsalicylic acid are potent and widely used anti-inflammatory drugs . They are thought to exert their therapeutic effects through multiple mechanisms , including the inhibition of cyclo-oxygenases , modulation of NF-κB activity , and direct activation of AMPK . However , the full spectrum of their activities is incompletely understood . Here we show that salicylate specifically inhibits CBP and p300 lysine acetyltransferase activity in vitro by direct competition with acetyl-Coenzyme A at the catalytic site . We used a chemical structure-similarity search to identify another anti-inflammatory drug , diflunisal , that inhibits p300 more potently than salicylate . At concentrations attainable in human plasma after oral administration , both salicylate and diflunisal blocked the acetylation of lysine residues on histone and non-histone proteins in cells . Finally , we found that diflunisal suppressed the growth of p300-dependent leukemia cell lines expressing AML1-ETO fusion protein in vitro and in vivo . These results highlight a novel epigenetic regulatory mechanism of action for salicylate and derivative drugs .
The anti-inflammatory activity of salicylate was first described by the Greek physician Hippocrates . One of its widely used derivatives , acetylsalicylic acid ( Aspirin ) , inhibits prostaglandin biosynthesis by irreversibly inactivating cyclooxygenases via non-enzymatic acetylation of a single serine residue ( Warner et al . , 1999 ) . Interestingly , salicylic acid does not possess this acetylating activity ( since it is lacking the acetyl group ) and does not inhibit cyclooxygenase in vitro . However , salicylic acid blocks cyclooxygenase expression at the transcriptional level thereby explaining its anti-inflammatory properties ( Xu et al . , 1999 ) . In addition , both salicylic acid and aspirin inhibit nuclear factor kappa B ( NF-κB ) activity ( Kopp and Ghosh , 1994 ) by inhibiting IκB kinase β ( IKKβ ) ( Yin et al . , 1998 ) . Other possible mechanisms of action have been proposed that include JNK pathway inhibition ( Schwenger et al . , 1997 ) and direct allosteric activation of AMP kinase ( AMPK ) ( Hawley et al . , 2012 ) . However , the pleiotropic effects of salicylate treatment on different cell types remain incompletely understood . Salsalate , a salicylate precursor , is an effective therapy for type 2 diabetes ( Goldfine et al . , 2010 ) , a metabolic disorder associated with insulin resistance and a strong pro-inflammatory component dependent on NF-κB ( Donath and Shoelson , 2011 ) . The efficacy of salicylates on insulin resistance is thought to reflect its anti-inflammatory activity and to be mediated by IKKβ inhibition ( Yin et al . , 1998 ) . Interestingly , examination of the chemical structure of anacardic acid , a previously reported p300 inhibitor , revealed that it contained a salicylic acid moiety linked to a long alkyl chain ( Balasubramanyam et al . , 2003; Sung et al . , 2008 ) . Previously , we showed that full activation of NF-κB activity requires the reversible acetylation of NF-κB by CBP/p300 histone acetyltransferases ( HATs ) ( Chen et al . , 2001 ) . Here , we have tested the possibility that salicylic acid might exert its transcriptional inhibitory activity by directly affecting NF-κB acetylation via inhibition of CBP/p300 acetyltransferase activity .
To determine whether salicylate inhibits p300 and other acetyltransferases , we used in vitro acetylation assays with purified histones and a recombinant p300 catalytic domain . Salicylate effectively inhibited p300 dependent acetyltransferase activity ( IC50 = 10 . 2 mM ) and CBP-mediated acetyltransferase activity ( IC50 = 5 . 7 mM ) in vitro , but did not detectably inhibit PCAF or GCN5 acetyltransferases ( Figure 1A ) in vitro . 10 . 7554/eLife . 11156 . 003Figure 1 . Salicylate inhibits CBP/p300 in vitro . ( A ) Recombinant p300 , CBP , GCN5 , or PCAF and histones were incubated with 14C-labeled acetyl-CoA with or without sodium salicylate , separated by SDS-PAGE , analyzed by autoradiography , and quantified with Image J software . Acetylation levels are relative to those in untreated controls . ( B ) Thermal stability assay for sodium salicylate binding to the p300 HAT domain . Tm , melting temperature . ( C ) and ( D ) Lineweaver-Burk plots showing kinetic analysis of p300 acetyltransferase inhibition by sodium salicylate . Histone acetylation was measured with several concentrations of acetyl-CoA ( C ) or histone ( D ) in the presence or absence of sodium salicylate . DOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 00310 . 7554/eLife . 11156 . 004Figure 1—figure supplement 1 . CoA metabolites of salicylate and diflunisal are more potent inhibitors of p300 . ( A ) Dose-response data for inhibition of p300 by salicylate and salicyl-CoA . Acetylation of an H4 ( 3–14 ) peptide was monitored using direct microfluidic mobility shift analysis as previously described ( Montgomery et al . , 2014; Fanslau et al . , 2010 ) . Error is given as the 95% confidence interval . ( B ) Dose-response data for inhibition of p300 by diflunisal and diflunisal-CoA . Error is given as the 95% confidence interval . DOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 004 To confirm that salicylate binds to p300 , we used thermal stability assays . A p300 HAT domain construct ( residues 1279–1666 ) bearing an inactivating Tyr1467Phe mutation to facilitate purification of homogeneously hypoacetylated p300 was expressed and purified with an N-terminal 6-His tag from E . coli cells . The protein was further purified by chromatography and incubated with increasing concentrations of sodium salicylate for 30 min and with SYPRO orange dye ( Invitrogen ) . Thermal melt curves were obtained by heating the protein from 20–95°C and monitoring fluorescence at 590 nm . This experiment revealed that the thermal unfolding temperature of p300/acetyl-CoA was 48 . 6°C , while treatment with 10 and 25 mM salicylate reduced the unfolding temperature to 46 . 1°C and 40 . 8°C , respectively ( Figure 1B ) . Kinetic analysis of p300 acetyltransferase activity with various concentrations of acetyl-CoA ( Figure 1C ) and histone ( Figure 1D ) substrates revealed that salicylate exhibits direct competitive p300 inhibition against acetyl-CoA and noncompetitive inhibition against histones . Taking this data together , we surmised that salicylate inhibits p300 acetyltransferase activity by directly competing with acetyl-CoA binding near its binding site on CBP and p300 . To determine whether salicylate induces histone deacetylation directly in cells , we treated HEK293T cells with various concentrations of salicylate . Western blot analysis with antibodies against various specific acetyl-lysine modifications of histone H2A , H2B , H3 , and H4 showed that addition of salicylate correlated with the deacetylation of H2AK5/K9 , H2BK12/K15 , and H3K56 in a dose-dependent manner ( Figure 2A and Figure 2—figure supplement 1 ) . Other histone residues , including H3K9 , K14 , K27 , K36 and H4K5 , K8 , K12 , K16 , have also been reported to be acetylated by CBP/p300 ( Schiltz et al . , 1999; Kouzarides , 2007 ) , but their acetylation state did not change in response to salicylate , possibly as a consequence of redundant activity of other acetyltransferases in the cellular environment ( Kouzarides , 2007 ) or opposing effects caused by inhibition of its previously characterized targets . The IC50 for salicylate-mediated inhibition of H2B acetylation ( 4 . 8 mM ) was close to the IC50 of CBP measured in vitro and to the plasma concentrations of salicylate ( 1–3 mM ) in humans after oral administration ( Goldfine et al . , 2010; 2013 ) . 10 . 7554/eLife . 11156 . 005Figure 2 . Salicylate inhibits specific lysine acetylation of histone and nonhistone proteins independently of AMPK activation . ( A ) Decreased acetylation of specific lysines in histones in the presence of salicylate . HEK293T cells were treated with the indicated concentrations of sodium salicylate for 24 hr . Site-specific histone acetylation was detected by Western blot with specific antisera . Bands were quantified with Image J software . Acetylation was normalized to that of untreated cells and plotted . Representative results are shown in Supplementary Figure 1 . Experiments are repeated and error bars indicate SEM . ( B–D ) Salicylate-induced hypoacetylation of histone H2B was rescued by overexpression of p300 ( B ) but not by the catalytically inactive p300 mutant F1504A ( C ) , or PCAF ( D ) . HEK293T cells were transfected with increasing amounts of expression vectors for p300 or F1504A or PCAF , treated with sodium salicylate for 24 hr , and analyzed by Western blotting analysis with an antiserum specific for acetyl histone H2BK12/K15 . Bands were quantified with Image J software . Acetylation was normalized to that of untreated control . Average levels of relative acetylation are plotted and error bars indicate SEM . Representative results are shown in Supplementary Figure 2—figure supplement 2 . ( E ) IC50 values generated from all curves in panel ( B ) and ( D ) were plotted against the amount of plasmid transfected ( p300 or PCAF ) . ( F ) , ( G ) HEK293 T cells were transfected with expression vectors for p300 and NF-κB p65 ( F ) or p53 ( G ) , treated with salicylate for 24 hr , and analyzed by Western blot with specific antibodies against acetyl NF-κBK310 ( F ) or acetyl p53K382 and acetyl H2BK12/15 ( G ) . Compound C ( 10 μM ) , a specific AMPK inhibitor , was added to salicylate-treated cells for 24 hr before Western blot ( G ) . KR , p65 K310R mutantDOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 00510 . 7554/eLife . 11156 . 006Figure 2—figure supplement 1 . Salicylate induces histone deacetylation in HEK293T cells HEK293T cells were treated with sodium salicylate as indicated for 24 hr , immediately lysed in Laemmli buffer , and then subjected to western blot analysis with the indicated antibodies . Histones H2A , H2B , H3 and H4 were used as input loading controls . Experiments are repeated five times and representative data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 00610 . 7554/eLife . 11156 . 007Figure 2—figure supplement 2 . Salicylate-induced deacetylation of histone H2B can be rescued by overexpression of p300 , but not PCAF , in a dose-dependent manner . ( A ) Overexpression of p300 but not PCAF specifically leads to hyperacetylation of histone H2B . Expression plasmids for p300 WT , catalytically inactive ( Y1503A or F1504A ) p300 , or PCAF were transfected into HEK293T cells by calcium phosphate . H2B acetylation was measured by western blot and the specified antibodies . ( B ) Overexpression of p300 but not PCAF rescued salicylate-induced H2B deacetylation in a dose-dependent manner . Transfected cells were prepared as above and treated with sodium salicylate as indicated for 24 hr . H2B acetylation was measured by Western Blot and the specified antibodies . Experiments are repeated five times and representative data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 007 To further test the hypothesis that p300 is a relevant target of salicylate in vivo , we overexpressed exogenous p300 at different levels and determined whether it suppresses the effect of salicylate on histone H2B acetylation . HEK293T cells were transfected with wild type ( WT ) p300 , catalytically inactive p300 Y1503A and F1504A mutants ( Suzuki et al . , 2000 ) , or PCAF , and the acetylation state of H2BK12/15 was assessed after salicylate treatment . Overexpression of WT p300 suppressed the effect of salicylate in a dose-dependent manner and increased H2BK12/K15 acetylation ( Figure 2B ) , but overexpression of catalytically inactive p300 mutants ( Figure 2C ) and PCAF did not ( Figure 2D and Figure 2—figure supplement 2 ) . Furthermore , the IC50 of salicylate strongly correlated with the amount of transfected p300 but not PCAF ( Figure 2E ) . These findings support the hypothesis that salicylate-mediated H2B deacetylation is specifically due to inhibition of p300 acetyltransferase activity . To determine whether salicylate down-regulates the acetylation of non-histone proteins , we overexpressed NF-κB and p53 in 293T cells , treated the cells with salicylate , and assessed acetylation of these proteins with specific antibody against acetyl NF-κBK310 and acetyl p53K382 ( Figure 2F and G ) . Salicylate decreased acetylation of both NF-κB and p53 in a dose-dependent manner . These findings strongly support the hypothesis that p300 acetyltransferase activity is a biologically relevant target for salicylate in vivo in cultured cells . Recently , salicylate was reported to activate AMPK by allosteric binding to its AMP binding site . ( Hawley et al . , 2012 ) To confirm this finding , we treated HEK293T cells with various doses of salicylate . The levels of a phosphorylated form of acetyl-CoA carboxylase ( ACC ) , an established AMPK target , increased in a dose-dependent manner ( Figure 2G ) . Compound C , an AMPK inhibitor , suppressed p-ACC accumulation in response to salicylate , but did not inhibit deacetylation of acetylated H2BK12/K15 or acetyl-p53K382 . This experiment demonstrates that salicylate-mediated protein deacetylation is not dependent on AMPK activation and activity ( Figure 2G ) . Next , we tested a series of other drugs that contain a salicylic acid moiety . A substructural homology search of the DrugBank database ( www . drugbank . ca ) ( Wishart et al . , 2006; 2008 ) identified five additional FDA-approved drugs: 4-aminosalicylic acid , 5-aminosalicylic acid , diflunisal , mycophenolic acid , and repaglinide that contain salicylic acid . We tested their ability to inhibit CBP/p300 acetyltransferase activity in in vitro HAT assays . All five drugs inhibited p300 with different IC50 ( Figure 3A and B ) . Three of the drugs inhibited p300 more potently than salicylate: the antidiabetes drug repaglinide ( IC50 = 374 μM ) , the immunosuppressant mycophenolic acid ( IC50 = 664 μM ) , and diflunisal , an older nonsteroidal anti-inflammatory drug ( IC50 = 996 μM ) ( Figure 3 , A and B ) . 10 . 7554/eLife . 11156 . 008Figure 3 . Structural homology search identifies diflunisal as a potent p300 inhibitor . ( A ) FDA-approved drugs that contain a structure similar to that of salicylate are shown in red . Numbers below the structures are IC50 of each drug , measured by in vitro p300 HAT assays . ( B ) Relative HAT activities are plotted by in vitro HAT assays using recombinant p300 and histones with increasing amount of various FDA approved drugs . Acetylation levels are relative to those in untreated controls . ( C ) Relative levels of histone acetylation in response to diflunisal . HEK293T cells were treated with various amount of diflunisal , followed by Western blotting with specific acetyl histone antibodies , as indicated . Bands were quantified with Image J software and plotted . Experiments are repeated two to five times . Error bars indicate SEM . Representative results are shown in Figure 2—figure supplement 1 . ( D ) and ( E ) Diflunisal-induced hypoacetylation of histone H2B was rescued by overexpression of p300 ( D ) but not by the catalytically inactive p300 mutants Y1503A and F1504A ( E ) ( F ) Diflunisal inhibits acetylation of NF-κB p65 ( G ) and p53 ( G ) independently of slight AMPK activation ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 00810 . 7554/eLife . 11156 . 009Figure 3—figure supplement 1 . Diflunisal induces histone deacetylation in HEK293T cells HEK293T cells were treated with diflunisal as indicated for 24 hr , immediately lysed in Laemmli buffer , and then subjected to western blot analysis with the indicated antibodies . Histones H2A , H2B , H3 and H4 were used as input loading controls . Experiments are repeated five times and representative data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 00910 . 7554/eLife . 11156 . 010Figure 3—figure supplement 2 . Diflunisal-induced deacetylation of p300 is rescued by overexpression of p300 in a dose-dependent manner , but not inactive p300 mutants . HEK293T cells were transfected with expression vectors for WT p300 or catalitycally inactive mutant . 24 hr after transfection , cells were treated with diflunisal as indicated for 24 hr . H2B acetylation was measured by Western Blot using specified antibodies . Experiments are repeated four times and representative data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 010 Since repaglinide induces insulin secretion at nanomolar concentration , it is likely that its ability to inhibit p300 with an IC50 of 374 μM is irrelevant to its antidiabetic activity . We therefore selected diflunisal for further analysis . Diflunisal induced deacetylation of specific histone residues , H2AK5 , K9 , H2BK12/K15 , H3K56 ( Figure 3C and Figure 3—figure supplement 1 ) , a pattern of histone acetylation similar to that induced by salicylate . Importantly , the IC50 for H2BK12/K15 inhibition in cells was 160 μM , which is within the range of plasma concentrations of diflunisal ( 150–350 μM ) after daily oral administration ( Nuernberg et al . , 1991; Mano et al . , 2006 ) . Overexpression of WT p300 suppressed the effect of diflunisal in a dose-dependent manner and increased H2BK12/K15 acetylation ( Figure 3D , Figure 3—figure supplement 2 ) , but overexpression of catalytically inactive p300 mutants did not ( Figure 3E , Figure 3—figure supplement 2 ) . Diflunisal also suppressed acetylation of the nonhistone proteins NF-κB p65K310 ( Figure 3F ) and p53K382 ( Figure 3G ) . Here also , Compound C , an AMPK inhibitor , suppressed p-ACC accumulation in response to diflunisal , but did not inhibit deacetylation of acetylated H2BK12/K15 or acetyl-p53K382 , indicating that AMPK is not necessary for these effects ( Figure 3G ) . These findings support the model that diflunisal also targets p300 acetyltransferase activity independently of AMPK . Previously , we reported that the leukemogenicity of the AML1-ETO fusion protein , generated by a t ( 8;21 ) translocation in acute myelogenous leukemia , is regulated by p300-mediated acetylation of lysine 43 of the fusion protein ( Wang et al . , 2011 ) . To investigate a potential application of our newly characterized salicylate- and diflunisal-mediated inhibition of CBP/p300 activity , we tested the effects of various doses of salicylate and diflunisal on two AML1-ETO expressing cancer cell lines ( human Kasumi-1 cells and a mouse AE9a-driven AML cell line that we generated ) . In support of our model , K24 and K43 acetylation of AML1-ETO were decreased in a dose-dependent manner by salicylate ( Figure 4A , left ) and by diflunisal ( Figure 4A , right ) . Salicylate inhibited cell proliferation at concentrations as low as 1 mM ( Figure 4B ) . This growth inhibition was caused in part by increased apoptosis , as shown by annexin V/7AAD double staining ( Figure 4C ) . Diflunisal also increased apoptosis in a dose-dependent manner ( Figure 4D ) . Additional measurements of nuclear DNA distribution showed a dose-dependent increase of the sub-G1 cell fraction , highly suggestive of apoptotic fragmentation by salicylate ( Figure 4E ) and diflunisal ( Figure 4F ) . We also noted an increased fraction of G1 cells and a decreased fraction of S and G2/M cells after salicylate treatment , consistent with reports that CBP/p300 is required for the G1/S transition ( data not shown ) ( Ait-Si-Ali et al . , 2000; Iyer et al . , 2007 ) . Salicylate did not affect the surface expression of differentiation-related antigens ( CD11b and CD34 in Kasumi-1 , Mac-1 and C-kit in AE9a ) ( data not shown ) , in accordance with our previous finding that acetylation of AML1-ETO is required for self-renewal and leukemogenesis but not for its ability to block cell differentiation ( Wang et al . , 2011 ) . To further test whether p300 is the relevant target of diflunisal in Kasumi-1 cells , we transduced lentiviral expression vectors for p300 or empty control into Kasumi-1 cells ( Figure 4G ) . Cells transduced with the empty vector showed inhibition of growth by diflunisal , similar to untransduced cells ( Figure 4H ) . In contrast , p300-transduced cells were significantly more resistant to diflunisal ( Figure 4H and I ) , and exhibit less apoptotis measured by annexin V positive cells ( Figure 4J ) and sub-G1 fraction ( Figure 4K ) . These results support the model that diflunisal kills Kasumi-1 cells by apoptosis due to p300 inhibition . 10 . 7554/eLife . 11156 . 011Figure 4 . Sodium salicylate and diflunisal decrease acetylation of AML1-ETOK43/K24 and block the growth of t ( 8;21 ) leukemia cells by inducing apoptosis . ( A ) Kasumi-1 cells expressing the AML1-ETO fusion protein were treated with sodium salicylate ( 4 or 8 mM , left ) or diflunisal ( 100 or 200 μM , right ) for 24 hr , followed by immunoprecipitation of AML1-ETO and analysis by Western blotting with an anti-acetyl lysine antiserum . ( B ) Kasumi-1 and AE9a cells treated or not with salicylate were counted by trypan-blue exclusion under light microscopy . ( C ) and ( D ) Annexin-V/7AAD staining of Kasumi-1 and AE9a cells after 24 hr of treatment with salicylate ( C ) or diflunisal ( D ) . ( E ) and ( F ) Kasumi-1 and AE9a cells treated or not with salicylate ( E ) or diflunisal ( F ) were collected , and DNA content was measured by propidium iodide staining after overnight fixation in 75% ethanol . The percentage of sub-G1 cells is shown . ( G ) p300 overexpression in the p300 lentiviral tranduced cells were confirmed by Western Blotting . ( H ) and ( I ) Kasumi-1 cells were transduced with p300 or empty lentiviral vector and treated with or not diflunisal . Cells were counted by trypan-blue exclusion under light microscopy . ( J ) and ( K ) Annexin-V/7AAD staining ( J ) and subG1 population analyzed by PI staining ( K ) of p300 transduced Kasumi-1 cells after 6 hr of treatment of diflunisal . DOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 011 Finally , to examine whether diflunisal inhibits leukemia development in vivo , we inoculated SCID mice with Kasumi-1 cells . Starting 3 weeks after inoculation , the mice were treated daily with diflunisal ( 50 or 100 mg/kg orally ) or vehicle . Diflunisal reduced tumor volumes in a dose-dependent manner ( Figure 5A ) and had minimal effects on body weight ( Figure 5B ) . After 3 weeks of treatment , the tumors were significantly smaller in diflunisal-treated mice than in vehicle-treated controls ( Figure 5C ) , and most of the tumors had disappeared in mice treated with the higher dose of diflunisal ( Figure 5D ) . 10 . 7554/eLife . 11156 . 012Figure 5 . Sodium salicylate and diflunisal inhibit the growth of AML1-ETO leukemia cells in SCID mice . The mice were inoculated with Kasumi-1 cells ( 3x107 ) and , starting 3 weeks later , were treated daily with oral diflunisal ( 50 or 100 mg/kg ) or vehicle for 3 weeks . ( A ) and ( B ) Plots of tumor volume ( A ) and body weight ( B ) . ( C ) and ( D ) Tumor size ( C ) and body weight ( D ) at sacrifice after 3 weeks of treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 11156 . 012 This study shows that salicylate inhibits CBP/p300 acetyltransferase activity by directly competing with acetyl-CoA , and it down-regulates the specific acetylation of histones and non-histone proteins in cells . We also found that diflunisal , an FDA-approved drug containing a salicylic acid substructure , inhibited CBP/p300 more potently than salicylate . Both drugs inhibited p300-dependent AML-ETO leukemic cell growth in vitro and in vivo . Thus , diflunisal and salicylate have promise as a oral therapy for patients with acute myelogenous leukemia associated with a t ( 8;21 ) translocation , an exciting potential application of our observations . . In plants , from where it was originally isolated , salicylate acts as an immune signal to induce systemic acquired resistance . It specifically activates the transcription cofactor NPR1 ( nonexpressor of PR genes 1 ) by binding to its paralogs NPR3 and NPR4 ( Fu et al . , 2012 ) . Since plants contain an ortholog of p300/CBP ( Bordoli et al . , 2001 ) , some of its activities in plants could also be mediated by inhibition of the plant ortholog of p300 or CBP . In animals , salicylate is an extensively studied small compound widely used as an anti-inflammatory drug . Many mechanisms of action have been proposed for the anti-inflammatory effects of salicylate in mammalian cells , including weak inhibition of cyclooxygenase ( Warner et al . , 1999 ) , inhibition of IKKβ inhibition ( Yin et al . , 1998 ) and topoisomerase II ( Cox et al . , 2011 ) , modulation of NF-κB ( Kopp and Ghosh , 1994 ) , and activation of AMPK ( Hawley et al . , 2012 ) . The simple structure of salicylate might enable it to interact with different affinities with many cellular proteins , which could explain its pleiotropic effects . While we cannot completely rule out the possibility that the effects of salicylate on acetylation may derive in part from off-target effects , in addition to the direct interaction with CBP/p300 reported here , our study demonstrate a direct inhibition of p300 and CBP by salicylate , diflunisal and their metabolites ( see discussion below ) . We observed that the acetylation of both histones and non-histone proteins ( NF-κB ) is suppressed in cells treated with either salicylate or diflunisal . We identified histone AcH3K56 as the most sensitive histone acetyl mark to inhibition by both drugs . This results is highly consistent with the literature showing that CBP ( also known as Nejire ) in flies and CBP and p300 in humans acetylate H3K56 ( Das et al . , 2009 ) . We also note that the pattern of histone marks inhibition are remarkably similar between the two drugs , but at different concentrations in agreement with their relative abilities to inhibit p300/CBP in vitro . In terms of what is observed for the other histone aceylated sites , the situation is more complex . Indeed , many histone modifications are regulated by multiple HAT enzymes . For example , acetyl H4K5 is regulated by HAT1 , CBP , p300 , Tip60 , HB01 whereas acetyl H3 K14 is regulated by CBP , p300 , PCAF , gcn5 , ScSAS3 ( Kouzarides , 2007 ) . We therefore interpret the observed lack of inhibition of some histone H3 or H4 acetylation sites by salicylate or diflunisal to reflect the compensating activities or other histone acetyltransferases that target the same sites . We also observed inhibition of NF-κB acetylation by both salicylate and diflunisal ( Figures 2D and 3D ) . The NF-κB subunit RelA is acetylated on lysines 218 , 221 , and 310 and these modifications are required for full NF-κB activation ( Chen et al . , 2001; 2002 ) . Therefore , salicylate’s NF-κB inhibitory effect can be at least partly explained by p300 inhibition . AMPK is reported to inhibit p300 acetyltransferase activity by phosphorylating p300 at serine 89 ( Zhang et al . , 2011 ) , suggesting that AMPK activation by allosteric salicylate binding might inhibit p300 indirectly . However , our findings clearly showed that salicylate does not inhibit p300 by activating AMPK since the p300 inhibition is insensitive to an AMPK inhibitor ( Figure 2G ) . Overall , our findings suggest that many of the pleiotropic effects of salicylate on different cell types and in diseases , including leukemia , are mediated by specific inhibition of CBP/p300 acetyltransferase activity , leading to deacetylation of histones and non-histone proteins . Interestingly , the IC50 for p300 inhibition by both salicylic acid and diflunisal was significantly lower in HEK293T cells ( Figure 2A ) than in HAT assays in vitro ( Figure 1A ) . A number of mechanisms discussed below could account for these differences . First , AMPK activation in cells might enhance the p300 inhibitory effect of salicylate in cells ( Zhang et al . , 2011 ) . Second , we have also observed that short-term treatment of cells with salicylate or diflunisal is associated with a decreased in the expression p300 ( see Figure 2F and 3D ) . This phenomenon is accentuated when cells are treated longer with salicylate or diflunisal ( data not shown ) . Other molecules have been shown to induce p300 degradation through the activation of different signaling transduction cascades ( Chen and Li , 2011 ) and the autoacetylation of p300 is important for its enzymatic activity ( Thompson et al . , 2004 ) . While we have not further explored this interesting observation here , we could envisage a mechanism by which inhibition of p300 autoacetylation would both contribute to the inactivation of the enzyme but also to a change in its stability . Third , metabolism of salicylic acid and diflunisal may also contribute to increased cellular potency in vivo . Indeed , we have found that salicyl-CoA , a known major intermediate of salicylate metabolism ( Knights et al . , 2007 ) , inhibits CBP/p300 with 28-fold increased potency in comparison with salicylate: IC50=220 μM for salicyl-CoA vs 6 . 12 mM for salicylate ( Figure 1—figure supplement 1 ) . A similar 52-fold increase in potency is observed with diflunisal-CoA in comparison to diflunisal: IC50=20 μM for diflunisal-CoA vs 1 . 05 mM for diflunisal ( Figure 1—figure supplement 1 ) . Although further investigation will be required to understand the relative contribution of both salicylic acid , diflunisal and their metabolites to the novel in vivo effects of salicylate reported here , these observations provide a potential mechanistic basis for the potent cell-based effects of these compounds . Fourth , it should be noted that a fragment of p300 consisting of its HAT domain is used in our in vitro experiments and in all other published studies using recombinant p300 protein . It is possible that the sensitivity of this subdomain to salicylate inhibition might be significantly different from the full-length protein present in cells . Most of the non steroidal anti-inflammatory drugs ( NSAIDs ) inhibit both COX-1 and COX-2 , although they vary in their relative potencies against the two isozymes ( Patrignani et al . , 1997 ) . However , salicylate does not , unlike its acetylated derivative aspirin , inhibit COX-1 and COX-2 activity in vitro ( Vane , 1971 ) ( Vargaftig , 1978; Mitchell et al . , 1993 ) ; ( Cromlish and Kennedy , 1996 ) . However , salicylate has been shown to exert a comparable analgesic and anti-inflammatory action as aspirin . Salicylates have been proposed to exert their pharmacological effects via inhibition of the transcription factor nuclear factor NF-|B and other targets . In these experimental systems , the concentrations used were in the same range as used in the experiments described in our paper ( 5–20 mM ) ( Kopp and Ghosh , 1994 ) ( Pierce et al . , 1996; Oeth and Mackman , 1995; Schwenger et al . , 1996; 1997; 1999 ) . Further , it should also be noted that salicylate plasma concentrations in patients taking salicylic acid ( 3–4 g/day ) range between 1–3 mM , a concentration at which partial inhibition of p300/CBP is observed . These data are therefore consistent with the proposed model that partial or complete inhibition of p300 by salicylate represents one of its relevant biological targets . It is important to note that a small molecule the size of salicylic acid used at such high concentrations is expected to interact with a number of cellular proteins and that our discovery that salicylic acid targets p300 does not imply that previous targets are not also part of the cellular response to these drugs . Several other FDA-approved drugs with substructures similar to that of salicylate also directly inhibited p300 acetyltransferase activity ( Figure 3A ) . The salicylate substructure is now identified in three HAT inhibitors , including anacardic acid ( 6-nonadecyl salicylic acid ) ( Balasubramanyam et al . , 2003 ) , salicylic acid and diflunisal and might therefore represent an important scaffold for developing new p300 inhibitors . Recently , p300 has emerged as a potential therapeutic target for respiratory diseases , HIV infection , metabolic diseases , and cancer ( Dekker and Haisma , 2009 ) . Indeed , our findings show that salicylate and the related compound diflunisal exhibit anti-tumor activity against a specific leukemia carrying a t ( 8;21 ) translocation , a tumor previously reported to be dependent on p300 in vitro and in vivo ( Wang et al . , 2011 ) . We have tested whether other NSAIDs , including acetaminophen and indomethacin , also inhibit p300 acetyltransferase in vitro , but did not detect any inhibitory activity ( data not shown ) . Importantly , NSAIDs that lack p300 inhibitory activity failed to inhibit Kasumi-1 cells growth ( data not shown ) . These results identify a novel epigenetic therapeutic target for salicylate , the epigenetic regulator CBP/p300 . Further efforts will focus on unraveling the relative roles of different cellular targets of salicylate , such as CBP/p300 , cyclooxygenases , IKKβ , and AMPK . Our results also suggest that salicylate may be useful for treating inflammation , diabetes , neurodegenerative disease , and other pathologies in which CBP/p300 has a critical role .
Recombinant HAT ( 1 mg ) , either p300 , CBP , PCAF , or GCN5 ( Enzo Life Sciences ) , and 10 µg of histones ( Sigma ) were incubated with sodium salicylate ( Sigma ) in reaction buffer ( 50 mM HEPES , pH 8 . 0 , 10% glycerol , 1 mM ) at 30°C for 30 min and then with 0 . 1 mCi of 14C acetyl-CoA at 30°C for 60 min . Reactions were stopped by adding 6x sample buffer and analyzed by SDS-PAGE . The gels were dried , and signals were obtained by autoradiography and quantified with Image J software . To quantify acetylated histone levels , we generated a standard curve from signals of lanes loaded with 2 . 5 , 5 , and 10 g of 14C labeled histones . A thermal stability assay was used to assess the binding of salicylate to the p300 HAT domain . A p300 HAT domain construct ( residues 1279–1666 ) bearing an inactivating Tyr1467Phe mutation to facilitate purification of homogeneously hypoacetyalted p300 was cloned into a pET-DUET vector with an N-terminal 6-His tag and expressed in BL21 ( DE3 ) E . coli cells . Cells were grown at 37°C until they reached an optical density ( 600 nm ) of 0 . 8 , incubated with 0 . 5 mM IPTG ( Isopropyl β-D-1-thiogalactopyranoside ) overnight at 18°C to induce protein expression , harvested , and lysed by sonication in lysis buffer ( 25 mM HEPES , pH 7 . 5 , 500 mM NaCl , and 5 mM ®-mercaptoethanol ) . The lysate was cleared by centrifugation and applied to a Ni-NTA affinity column . The protein was eluted from the column with increasing concentrations of imidazole in lysis buffer ( 20–250 mM ) and treated overnight with TEV protease to cleave the 6-His tag . The protein was further purified by passage through a HiTrap SP HP ion-exchange column and a size-exclusion Superdex 200 column equilibrated with 25 mM HEPES , pH 7 . 5 , 150 mM NaCl , and 5 mM -mercaptoethanol . X-ray crystallography showed that the purified p300 HAT domain protein binds to acetyl-CoA or CoA ( apparently from the bacterial cell ) ( Maksimoska et al . , 2014 ) In thermal stability experiments in a 384-well ABI plate ( Applied Biosystems ) , the p300 HAT domain bound to acetyl-CoA/CoA was incubated with increasing concentrations of sodium salicylate for 30 min . The final concentration of p300 was 2 μM in reaction buffer ( 0 . 1 M HEPES , pH 7 . 5 , 150 mM NaCl , and 5 mM -mercaptoethanol ) . Then , 4 μl of a 1:200 dilution of stock SYPRO orange dye ( Invitrogen ) in reaction buffer was added to achieve a total reaction volume of 20 μl . Thermal melt curves were obtained by heating the protein from 20–95°C and monitoring fluorescence at 590 nm with a 7900HT Fast Real Time PCR System ( Applied Biosystems ) . All curves were obtained in triplicate and averaged . Salicyl-CoA and diflunisal-CoA were synthesized from their parent carboxylic acids and HPLC purified according to previously reported methods ( Padmakumar et al . , 1997 ) . The purity of each acyl-CoA was confirmed by analytical HPLC prior immediately prior to utilization ( Montgomery et al . , 2014; Fanslau et al . , 2010 ) . P300 inhibition assays were performed using direct microfluidic mobility shift analysis as previously described . Briefly , p300 reaction mixture ( 50 mM HEPES , pH 7 . 5 , 50 mM NaCl , 2 mM EDTA , 2 mM DTT , 0 . 05% Triton-X-100 , 50 nM p300 , 2 μM FITC-histone H4 peptide ) was plated in 384-well plates and allowed to equilibrate at room temperature for 10 min . Reactions were initiated by addition of acetyl-CoA ( final concentration = 5 μM ) , bringing the final assay volume to 30 μL . Assays were quenched after 10 min ( <15% product accumulation ) by addition of 5 μL of 0 . 5 M neutral hydroxylamine and transferred to a Perkin-Elmer Lab-Chip EZ-Reader instrument for analysis . Separation conditions were: downstream voltage of −500 V , upstream voltage of −2500 V , and a pressure of −1 . 5 psi . Percent conversion was calculated by ratiometric measurement of substrate/product peak heights . Percent activity represents the percent conversion of KAT reactions treated with inhibitors relative to untreated control KAT reactions , and corrected for nonenzymatic acetylation . Dose-response analysis of p300 inhibition was performed in triplicate and analyzed by nonlinear least-squares regression fit to Y = 100/ ( 1 + 10∧ ( Log IC50 – X ) *H ) , where H = Hill slope ( variable ) . IC50 values represent the concentration that inhibits 50% of KAT activity . Calculations were performed using Prism 6 ( GraphPad ) software . HEK293T cells ( ATCC ) were maintained in DMEM supplemented with 10% FCS . The viability of Kasumi-1 cells and AE9a mouse leukemia cells was assessed in triplicate by trypan blue exclusion . The Kasumi cell line was isolated and characterized by one of the coauthors ( S . Nimer ) ( Becker et al . , 2008 ) . All cell lines were tested annually for mycoplasma contamination . Only negative mycoplasma cultures were used during the conduct of these experiments . pCi-p300 and pCi-PCAF are described elsewhere ( Boyes et al . , 1998 ) . pcDNA3/myc-p300 and pcDNA3/T7-p65 were described previously ( Chen et al . , 2001 ) . pCi-p300 Y1503A , F1504A and pcDNA3/T7-p65 K310R were constructed by using the QuickChange site-directed mutagenesis kit ( Promega ) . Lentiviral plasmids , pCSII-CMV-MCS , pCAG-HIVgp , pCMV-VSV-G-RSV-Rev are kindly provided by H Miyoshi , RIKEN BioResource Center , Tsukuba , Japan ( Bai et al . , 2003 ) . p300 CDS was cloned into XhoI/NotI site of pCSII-CMV-MCS . HEK293T cells were treated with sodium salicylate for 24 hr and lysed in lysis buffer ( 25 mM Tris , pH 6 . 8 , 2% SDS , and 8% glycerol ) . For Western blot analysis , we used antibodies against acetyl histone H2AK5 ( ab1764 , Abcam ) , acetyl histone H2AK9 ( ab47816 , Abcam ) , acetyl histone H2BK12/K15 ( ab1759 , Abcam ) , acetyl histone H3K9 ( 06–942 , Millipore ) , acetyl histone H3K14 ( 12–359 , Millipore ) , acetyl histone H3K27 ( 07–360 , Millipore ) , acetyl histone H3K36 ( 07–540 , Millipore ) , acetyl histone H3K56 ( 2134–1 , Epitomics ) , acetyl histone H4K5 ( 06-759-MN , Millipore ) , acetyl histone H4K8 ( 06-760-MN , Millipore ) , acetyl histone H4K12 ( 6-761-MN , Millipore ) , acetyl histone H4K16 ( 06-762-MN , Millipore ) , histone H2A ( 07–146 , Millipore ) , histone H2B ( ab1790 , Abcam ) , histone H3 ( 07–690 , Millipore ) , histone H4 ( 07–108 , Millipore ) , p300 ( ab3164 , Abcam ) , PCAF ( ab96510 , Abcam ) , tubulin ( T6074 , Sigma ) , acetyl NF-κB p65K310 ( 3045 , Cell Signaling ) , NF-κB p65 ( sc-372 , Santa Cruz Biotechnology ) , and acetyl lysine ( 9441 , Cell Signaling ) . pCSII-CMV-MCS vectors and packaging plasmids were transfected to HEK293T cells , supernatant were collected and ultracentrifuged 48 hr after transfection . Same amount as p24 levels of lentiviruses contain empty or p300 expression vectors were transduced to Kasumi-1 cells . Kasumi-1 cells were treated with sodium salicylate or diflunisal for 24 hr and lysed in RIPA buffer . AML1-ETO protein in the lysate was immunoprecipitated with an anti-ETO antibody ( Santa Cruz Biotechnology ) . Antibodies against AML1 and acetylated AML1-ETO K24/K43 ( generated in the Nimer lab ) were used for Western blotting . Apoptosis was analyzed with an Annexin V-APC/7AAD Apoptosis kit ( Becton-Dickinson ) according to manufacturer’s instructions . To assess the distribution of nuclear DNA content in the cell-cycle analysis , cells were collected , washed in PBS , fixed overnight in 75% ethanol at –20°C , treated with 1% RNase A for at least 15 min at 37°C , and stained with 50 mg/ml propidium iodide . To monitor CD34 expression , the cells were stained with an allophycocyanin-conjugated anti-CD34 antibody ( Becton Dickinson ) . To monitor CD11b expression , the cells were stained with a phycoerythrin-conjugated anti-CD11b antibody ( Beckman-Coulter ) . To monitor C-kit and Mac-1 expression , the cells were stained with allophycocyanin-conjugated anti-C-kit and phycoerythrin-conjugated anti-Mac-1 antibodies ( Becton-Dickinson ) . Cells were sorted with a Becton-Dickinson FACSCalibur , and the data were analyzed with FlowJo software . Severe combined immunodeficiency ( SCID ) mice were injected with 30 million Kasumi-1 cells in 100 μl of PBS and 100 μl of Matrigel . Three weeks after inoculation , when tumors can be detected , mice were treated daily with oral diflunisal ( 50 or 100 μg/kg ) or vehicle . Tumor volume and body weight were recorded every 2 days . After 3 weeks of treatment , the mice were killed and tumor size and body weight were recorded . All mice were maintained according to National Institutes of Health guidelines and all animal use protocols were approved by the Institutional Animal Care and Use Committee . | People have been using a chemical called salicylate , which was once extracted from willow tree bark , as medicine for pain , fever and inflammation since ancient Greece . Aspirin is derived from salicylate but is a more potent drug . Aspirin exerts its anti-inflammatory effect by shutting down the activity of proteins that would otherwise boost inflammation . Aspirin achieves this by releasing a chemical marker , called an acetyl group , to be added to these proteins via a process known as protein acetylation . However , salicylate cannot trigger protein acetylation and so it was not clear how it reduces inflammation . An anti-diabetes drug that is converted into salicylate in the body reduces inflammation by inhibiting a protein called NF-κB . In 2001 , a group of researchers reported that NF-κB becomes active when an enzyme called p300 adds an acetyl group to it . This raised the question: does salicylate reduce inflammation by blocking , instead of triggering , protein acetylation . Now , Shirakawa et al . – who include a researcher involved in the 2001 study – show that salicylate does indeed block the activity of the p300 enzyme . Shirakawa et al . then searched a database looking for drugs that have salicylate as part of their molecular structure . The search led to a drug called diflunisal , which was even more effective at blocking p300 in laboratory tests . Some cancers , including a blood cancer , rely on p300 to grow; diflunisal was shown to stop this kind of cancer cell from growing , both in the laboratory and in mice . Together , the experiments suggest that salicylate and drugs that share some of its structure might represent useful treatments for certain cancers , as well as other diseases that involve the p300 enzyme . | [
"Abstract",
"Introduction",
"Results",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2016 | Salicylate, diflunisal and their metabolites inhibit CBP/p300 and exhibit anticancer activity |
The earliest developmental origins of dysmorphologies are poorly understood in many congenital diseases . They often remain elusive because the first signs of genetic misregulation may initiate as subtle changes in gene expression , which are hard to detect and can be obscured later in development by secondary effects . Here , we develop a method to trace back the origins of phenotypic abnormalities by accurately quantifying the 3D spatial distribution of gene expression domains in developing organs . By applying Geometric Morphometrics to 3D gene expression data obtained by Optical Projection Tomography , we determined that our approach is sensitive enough to find regulatory abnormalities that have never been detected previously . We identified subtle but significant differences in the gene expression of a downstream target of a Fgfr2 mutation associated with Apert syndrome , demonstrating that these mouse models can further our understanding of limb defects in the human condition . Our method can be applied to different organ systems and models to investigate the etiology of malformations .
Morphogenesis is guided by dynamic spatio-temporal regulation of gene expression patterns ( Chan et al . , 2017; Andrey and Mundlos , 2017 ) , the zones within tissues where genes are expressed during specific periods in development . Critical changes in the space , time or intensity of gene expression patterns can result in organ malformation , with reduced or even loss of function . These errors of morphogenesis , which occur in approximately 3% of live births ( Toxicology NRC C on D , 2000 ) , are induced by environmental and/or genetic insults that alter the normal process of development . The common approach to revealing the origin of these alterations is to look for phenotypic abnormalities in animal models and to visually assess the overall patterns of gene expression . This qualitative approach has not , however , been able to identify the earliest signs of dysmorphogenesis in many diseases because the first changes in the gene expression patterns may be subtle , limited in time , and later abnormalities may obscure the original genetic cause . To reveal the primary etiology of congenital malformations , more rigorous methods are needed to quantify the three-dimensional ( 3D ) phenotypes of gene expression patterns . A useful tool for tracing back development should be able to perform a quantitative statistical comparison of normal and disease-altered embryogenesis and to detect the earliest signs of genetic misregulation leading to organ malformation . The shape of developing organs is regulated by gene activity in space and time ( Andrey and Mundlos , 2017 ) . Cascades of gene regulatory networks provide the detailed instructions necessary to organize cell behavior and to orchestrate tissue growth and differentiation . Gene expression patterns can be readily mapped within tissues in a true 3D framework combining whole-mount-in-situ hybridization ( WMISH ) ( Rosen and Beddington , 1993 ) with Optical Projection tomography ( OPT ) ( Sharpe et al . , 2002 ) . WMISH is a standard molecular technique for detecting the expression of a specific gene using a labeled complementary RNA probe ( de la Pompa et al . , 1997; Correia and Conlon , 2001 ) , and OPT is a mesoscopic imaging procedure that can produce high-resolution 3D reconstructions of whole developing embryos processed by WMISH ( Sharpe , 2003; Boot et al . , 2008 ) . These technologies represented breakthroughs in developmental biology and have provided invaluable qualitative insights into gene function and development ( Sharpe , 2003 ) . However , methods for quantifying the 3D distributions of gene expression in a systematic , objective manner are still lacking . Expanding the potential of OPT from qualitative to quantitative analysis of gene expression patterns is challenging . Gene expression is characterized by highly dynamic patterns , with fast rates of change and fuzzy boundaries that usually do not correspond to well-defined anatomical structures but rather to tissue regions where cells have dynamically up- and downregulated genes . Gene expression patterns have rarely been quantified ( Jernvall et al . , 2000; Airey et al . , 2006; Salazar-Ciudad and Jernvall , 2010; Mayer et al . , 2014; Hu et al . , 2015; Xu et al . , 2015; Martínez-Abadías et al . , 2016 ) . We propose to quantify the shape of developing organs in association with their underlying gene expression patterns by applying Geometric Morphometrics ( GM ) , a set of statistical tools for measuring and comparing shapes with increased precision and efficiency ( James Rohlf and Marcus , 1993; Klingenberg , 2002; Klingenberg , 2010; Adams et al . , 2013; Hallgrimsson et al . , 2015 ) . Previous attempts had analyzed the phenotypic and gene expression patterns of variation independently ( Jernvall et al . , 2000 ) , using different morphometric methods ( Hu et al . , 2015; Xu et al . , 2015 ) or were restricted to two-dimensional analyses ( Martínez-Abadías et al . , 2016 ) . Therefore , the current study is the first to combine OPT and GM to characterize the shape of gene expression patterns quantitatively in 3D and to associate these changes to phenotypic changes . Thus , this approach provides the ability to replace qualitative observations with the quantification of subtle yet significant biological differences that underlie the processes through which morphogenesis is altered by disease . Here , we illustrate how our method can reveal the genetic origin of developmental defects by investigating limb malformations in Apert syndrome [OMIM 101200] . Apert syndrome is a rare congenital disease , with a disease prevalence of 15–16 per million live births , that is characterized by cranial , neural , limb , and visceral malformations ( Cohen and MacLean , 2000 ) . Over 99% of Apert cases are associated with one of two missense mutations , S252W or P253R , in the Fibroblast Growth Factor Receptor 2 ( FGFR2 ) ( Wilkie et al . , 1995; Park et al . , 1995 ) . The mutations occur on neighboring amino acids on the linker region between the second and third extracellular immunoglobulin domains of FGFR2 , and alter the ligand-binding specificity of the receptors ( Yu et al . , 2000; Yu and Ornitz , 2001 ) . Thus , the FGF receptors are activated inappropriately , altering the entire FGF/FGFR signaling pathway and causing dysmorphologies of different organs and systems ( McIntosh et al . , 2000 ) . Apert syndrome shares craniofacial dysmorphologies with other craniosynostosis syndromes but is differentiated on the basis of limb defects of the fore- and hindlimb digits . The craniofacial dysmorphology of Apert syndrome ( i . e . premature closure of cranial sutures and patent anterior fontanelle associated with atypical head shape , midfacial retrusion and palatal defects ) ( Cohen and MacLean , 2000 ) has been intensively investigated , especially because mouse models show cranial phenotypes that correspond with the human condition ( Holmes et al . , 2009; Martínez-Abadías et al . , 2010; Holmes and Basilico , 2012; Hill et al . , 2013; Heuzé et al . , 2014 ) . The associated limb defects are less well studied , however , in part because mouse models for Apert syndrome present only subtle limb anomalies ( Chen et al . , 2003; Wang et al . , 2005; Wang et al . , 2010 ) , even in mice carrying the P253R mutation ( Wang et al . , 2010 ) which is associated with the more severe limb malformations in Apert syndrome ( Slaney et al . , 1996; von Gernet et al . , 2000 ) . Here , we present precise phenotyping of the limbs of newborn and embryonic specimens of the Fgfr2+/P253R Apert syndrome mouse model , and reveal significant differences that can be traced to as early as one day after the initiation of limb development . To explore the molecular basis of these initial signs of limb dysmorphology , we applied our method combining OPT and GM to assess the expression pattern of a downstream target of Fgf signaling , Dusp6 . We chose to assess Dusp6 because it is well-documented as a direct target of Fgf signaling ( Kawakami et al . , 2003; Li et al . , 2007; Ekerot et al . , 2008 ) , and because , in addition to being a relevant gene for limb morphogenesis , it is also important for facial , brain and heart development ( Kawakami et al . , 2003; Li et al . , 2007; Maillet et al . , 2008 ) . Therefore , Dusp6 was an ideal candidate for our proof-of-concept study . Our quantitative analyses demonstrate that the Apert syndrome Fgfr2 P253R mutation induces changes in the expression pattern of Dusp6 and that these genetic changes are associated with significant phenotypic alterations . These results provide insight into the origins of limb malformations in Apert syndrome .
Previous studies have reported that most Apert syndrome mice do not show obvious abnormalities of the limbs ( Chen et al . , 2003 ) , and thus focused their molecular analyses on the skull ( Wang et al . , 2010 ) . Histopathological analyses in Apert syndrome mice revealed overall limb shortening resulting from abnormal osteogenic differentiation , but no signs of limb disproportion or syndactyly ( Wang et al . , 2005; Wang et al . , 2010 ) . Syndactyly has only been reported in three specimens of an outbred knock-in Fgfr2+/P253R model ( Yin et al . , 2008 ) , but no experimental analyses were performed to further explain why the FGFR2 mutation did not affect limb development in all the specimens within the sample . As a further test of whether or not the Fgfr2 P253R mutation affects limb development in mice , we first performed an extensive quantitative analysis of the size and shape of individual forelimb bones using data from high resolution microCT images of newborn ( P0 ) mutant and unaffected littermates ( Figure 1A–F ) . Our results revealed many more significant differences between P0 unaffected and Fgfr2+/P253R mutant littermates than previously reported . We found that the humerus , radius and ulna were statistically significantly shorter in length but had increased bone volumes in Fgfr2+/P253R mutant mice in comparison to unaffected littermates ( Table 1 and Figure 1G ) . More localized size differences were detected in the bones derived from the autopod that give rise to the hands . The distal phalanx of digit I , the proximal phalanx of digit V , and metacarpals II , III and IV were significantly longer in Fgfr2+/P253R mutant mice ( Table 1 and Figure 1G ) . In contrast , the proximal phalanx of digit III was shorter and lower in bone volume in Fgfr2+/P253R Apert syndrome mice relative to unaffected littermates ( Table 1 and Figure 1G ) . The scapula and the clavicle , the bones that form the shoulder girdle , were also significantly affected: the scapula was longer , the clavicle was shorter and both bones showed increased bone volumes in Fgfr2+/P253R Apert syndrome mice ( Table 1 and Figure 1G ) compared to unaffected littermates . The Principal Components Analysis ( PCA ) based on the shape of the humerus did not show marked shape differences between unaffected and Fgfr2+/P253R Apert syndrome mice ( Figure 1H ) . However , the PCA of the scapula indicated a clear morphological differentiation between these two groups ( Figure 1I ) . The scapula of Fgfr2+/P253R Apert syndrome mice presented a more robust phenotype , with wider and longer scapulae , in comparison to that of their unaffected littermates . Overall , these size and shape differences demonstrate that Fgfr2+/P253R Apert syndrome mice present widespread and significant limb dysmorphologies at P0 that were not previously reported and would not have been revealed without microCT scanning and quantitative statistical testing . Some defects , such as shoulder anomalies and short humeri , have a direct correspondence with the human phenotype ( Park et al . , 1995 ) . In newborn mice , however , we did not detect any clear sign of syndactyly , which is the most prominent limb defect in people with Apert syndrome ( Cohen and Kreiborg , 1995; Holten et al . , 1997 ) . As the forelimb of mice is not yet completely ossified at P0 and because Fgfr2+/P253R mutant littermates die shortly after birth , we could not assess whether other limb abnormalities might appear later in development . To determine the earliest developmental basis of the limb anomalies quantified in newborn mice , we developed a quantitative method to explore early embryonic limb development . First , to visualize the expression pattern of a downstream target of Fgfr2 , we obtained OPT scans of Fgfr2+/P253R Apert syndrome mouse embryos that had been analyzed with WMISH to reveal Dusp6 expression ( Figure 2 ) . Qualitative assessment of the 3D reconstructions showed that Dusp6 was widely expressed throughout the embryo from embryonic day ( E ) 10 . 5 to E11 . 5 , with highest intensity in the limbs , the head and the somites ( Figure 2 ) . Dusp6 was also expressed in the heart with moderate intensity . By comparing the distribution of the Dusp6 gene expression pattern visually , it was possible to distinguish between embryos at the E10 . 5 and the E11 . 5 stages of development . At E10 . 5 , Dusp6 was prominently expressed in the facial prominences and in the forming somites along most of the craniocaudal segment , whereas at E11 . 5 , the expression of Dusp6 was more widespread in the brain and restricted to the caudal somites . Focusing on the limbs , the expression of Dusp6 at the two different stages was also readily distinguishable , with Dusp6 expression domains thinning into a more extended domain along the limb outline as the limb buds grow from E10 . 5 to E11 . 5 ( Figure 2 ) . The limb bud expression patterns of Fgfr2+/P253R mutant and unaffected littermates were not distinguishable because of the large amount of developmental variation within litters ( Figure 2 ) . Quantitative testing was thus required to allow the more accurate evaluation of limb alterations that are potentially associated with Apert syndrome but visually undetectable . We developed a method for 3D shape analysis of the limb and the associated gene expression pattern of Dusp6 ( Figure 3 ) . This protocol enabled us to determine differences in limb size and shape between genotype groups and to assess whether these phenotypic differences are associated with altered gene expression patterns ( Figure 3 ) . Our approach uses GM methods to measure directly the limb anatomy and gene expression domains segmented from the 3D reconstructions of the embryo OPT scans . As expression of Dusp6 showed a fuzzy spatial gradient , multiple thresholding was used to define a consistently high gene expression pattern ( Figure 3 , steps from 1 to 5 ) . After manual and semiautomatic recording of the 3D coordinates of landmarks on the surfaces of the limb and the gene expression domains blinded to group allocation ( Figure 3 , step 6; Video 1 ) , multivariate statistical analyses were performed to explore shape and size variation and covariation patterns for the limb morphology and the Dusp6 domain ( Figure 3 , steps 7 and 8 ) . As gene expression patterns are highly dynamic and rapidly change in size , shape and position within a few hours as development progresses ( Martínez-Abadías et al . , 2016 ) , individual limb buds from Fgfr2+/P253R mutant embryos and their unaffected littermates aged between E10 . 5 and E11 . 5 were staged using a fine-resolution staging system ( https://limbstaging . embl . es ) ( Musy et al . , 2018 ) . The staging results showed that the analyzed limbs represent a temporal continuum throughout development , with no significant differences between the staging of unaffected and mutant littermates of the same litter ( Figure 4 ) . We partitioned the time span from E10 to E11 . 5 into four periods , each one approximately representing 12 hr of development ( see Table 2 and 'Materials and methods' for further details on sample composition ) . We first focused on analyzing limb dysmorphology , aiming to determine the youngest stage at which there were morphological differences between mutant and unaffected limbs . To find the earliest moment when differences arise , we traced limb development backwards in time using Geometric Morphometric methods , starting first with embryos from the oldest period ( as the differences would be easier to find ) and from there proceeding towards the earlier ( younger ) periods . In this way , we should be able to identify confidently the initiation of limb dysmorphogenesis associated with the Fgfr2 P253R mutation . During the 'Late' period , we detected that Fgfr2+/P253R limb buds were already clearly separated from those of their unaffected littermates in the morphospace defined by the Principal Component Analysis ( PCA ) ( Figure 5A , Figure 5—figure supplement 1 ) . Relative to those of their unaffected littermates , the limbs of Fgfr2+/P253R mice presented subtle phenotypic limb differences: limbs were shorter and thicker , with limited development of the wrist ( Figure 5A ) . Quantitative comparison of limb size showed that the limbs of mutant mice were also significantly smaller than those of their unaffected littermates ( Table 3 and Figure 6A ) . Interestingly , in this 'Late' period the limbs of mutant mice were smaller than unaffected littermates ( although this was not the case in earlier periods ) . Overall , these results confirmed that the Fgfr2 P253R Apert syndrome mutation has an effect on limb development , altering both the size and shape of the limbs . These subtle but significant phenotypic differences would most probably have remained undetected by a qualitative approach . Our quantitative approach revealed their statistical significance and pointed to the origin of the Apert syndrome limb malformation prior to E11 . 5 , before the 'Late' period . During the 'Mid late' period , the limbs of Fgfr2+/P253R mutant mice were still distinguishable from those of unaffected mice in the morphospace of the PCA ( Figure 5B ) . At this period , the limbs of Fgfr2+/P253R mice lacked the antero-posterior asymmetry and the narrowing of the wrist region more typical of unaffected littermates . Instead , Fgfr2+/P253R mice showed a limb phenotype that was elongated in the proximo-distal axis and thickened in the dorso-ventral axis ( Figure 5B ) , resembling the limb shape of younger unaffected embryos . This shape difference coincided with reduced growth in Fgfr2+/P253R mice , as the limbs of Fgfr2+/P253R mice tended to be smaller than unaffected limbs ( Table 3 and Figure 6A ) . Therefore , significant differences between unaffected and mutant limbs could still be detected during the 'Mid late' period of development and the origins of limb defects associated with Apert syndrome should be sought earlier in development . The first period during which no significant differences could be detected between unaffected and Fgfr2+/P253R Apert syndrome mice was at the 'Mid early' period ( Figure 5C ) . Despite internal variation in limb shape , with Fgfr2+/P253R mice spreading throughout the morphospace and unaffected littermates concentrated on one region , mutant and unaffected littermates completely overlapped . Therefore , limb shape differences could no longer be detected between groups . Limb size differences were not significant either ( Table 3 and Figure 6A ) . Therefore , our results suggest that the critical time point of limb dysmorphogenesis associated with Apert syndrome occurred between the 'Mid late' and 'Mid early' periods , corresponding to the transition period from E10 . 5 to E11 ( Figure 5B–C ) . Consistent with the above , no further sign of limb shape dysmorphology was detected during the 'Early' period of development ( Figure 5D ) . During this period , there was a great range of developmental variation , with unaffected and Fgfr2+/P253R mutant mice completely overlapping in the morphospace and with all limbs displaying similar incipient bud shapes ( Figure 5D ) . The limbs of Fgfr2+/P253R mice were significantly larger than the limbs of their unaffected littermates ( Table 3 and Figure 6A ) , suggesting that at this early time point , there is a significant effect of the Fgfr2 P253R mutation on limb size but not on limb shape ( Figure 6A ) . We next sought to obtain direct evidence of altered genetic regulation that could explain the observed limb phenotype by analyzing the shape dynamics of the expression of Dusp6 , a direct target gene of Fgf signaling . As in the limb shape analysis described above , we chose to work backwards in developmental time , first examining the gene expression of Dusp6 in the embryos from the latest period . We found that in the 'Late' period , Dusp6 expression was already different in Fgfr2+/P253R mutant mice and unaffected littermates . The differences were significant both in shape ( Figure 5E ) and size ( Table 3 and Figure 6B ) . In the limbs of unaffected mice , the Dusp6 expression domain appeared as a thin domain underlying the apical ectodermal ridge , whereas in Fgfr2+/P253R mutant mice , the shape of the Dusp6 domain was expanded in all directions ( Figure 5E ) . Accordingly , the volume of the Dusp6 expression domain was significantly larger in Apert syndrome mice ( Table 3 and Figure 6B ) , even when these mice had significantly smaller limbs ( Figure 6A ) . The Dusp6 expression domain thus grew disproportionately in the limbs of Fgfr2+/P253R mutant mice in the latest period of development ( Figure 5E ) , which is consistent with reports of whole-body size reduction in Fgfr2+/P253R Apert mice and of the over-activation of Fgfr2 signaling by the Apert syndrome mutation ( Yu and Ornitz , 2001; Ibrahimi et al . , 2001 ) . During the 'Mid late' period , the shapes of expression patterns were still distinct — the expanded Dusp6 expression domain persisted on the dorsal and ventral sides of mutant limbs , but was reduced on the anterior and posterior sides ( Figure 5F ) . The overall volume of the Dusp6 expression domains remained larger in Fgfr2+/P253R mutant mice , but the difference was no longer statistically significant ( Table 3 and Figure 6B ) . During the 'Mid early' period , the separation between unaffected and Fgfr2+/P253R mutant mice was maintained in the PCA analysis ( Figure 5G ) . Unaffected mice showed a Dusp6 expression domain that was expanded towards the anterior and posterior edges of the expression domain ( Figure 5G ) . By contrast , Fgfr2+/P253R mutant mice did not show the extension and the posterior asymmetry of the Dusp6 expression domain typical of normal limb development , suggesting a lack of differentiation in the Dusp6 expression of mutant limbs ( Figure 5G ) . The 'Early' period was the only time point when we did not detect a significant difference between unaffected and Fgfr2+/P253R mice ( Figure 5H ) . The PCA showed variation in the expression domains of Dusp6 , with similar gene expression patterns in terms of both shape ( Figure 5H ) and size ( Table 3 and Figure 6B ) across all mice . Therefore , the first observation of an alteration in the gene expression pattern ( Figure 5G ) occurred earlier than the change in the limb shape ( Figure 5B ) . Our analyses provide evidence that differences in the Dusp6 gene expression pattern first occurred during the 'Mid early period' , preceding the phenotypic limb differentiation , which occurred a few hours later during the 'Mid late period' . Overall , the time courses showing the dynamics of limb shape and gene expression pattern changes throughout development ( Figure 5 and Figure 6 ) confirmed that the Fgfr2 Apert syndrome mutation causes an aberrant overexpression of Dusp6 early in development , which could later lead to significant limb malformations . Finally , we explored the patterns of correlation between the limb phenotype and gene expression to further explore how altered Fgf signaling relates to the limb malformations induced by the Apert syndrome Fgfr2 P253R mutation . First , we assessed the relationship between the size of the limbs and the volume of the Dusp6 expression domain , pooling all the forelimbs and hindlimbs and assessing the correlation between these two traits ( Figure 6C , D ) . The trend line showed that for the same limb size , Fgfr2+/P253R mutant mice showed larger Dusp6 expression domains , both in forelimbs ( R2 = 0 . 4 ) and hindlimbs ( R2 = 0 . 6 ) . If the extension of the Dusp6 expression depended only on limb growth , and the gene domain passively became larger by just following limb tissue growth , a higher correlation between the size of the limb and the gene expression would be expected . However , the moderate correlation found here suggests that the size of the Dusp6 gene expression zone is not solely dependent on limb growth but might be controlled by further genetic regulatory factors . Second , we assessed the morphological integration between the shape of the limbs and the shape of the Dusp6 expression domain . The statistical analysis of the covariance pattern between these shapes can reflect the interaction of the phenotype and the gene expression pattern during limb development . As shown by analysis of additional genes expressed during limb development ( Martínez-Abadías et al . , 2016 ) , even when a gene is expressed within the limb , the shape of the limb and the shape of the gene expression domain are not correlated by definition , and the integration pattern can change from a strong association to no significant correlation within few hours as the limb develops ( Martínez-Abadías et al . , 2016 ) . The dynamics of the integration pattern can identify the key periods during which the expression of a gene is relevant for determining the shape of the limb . If the morphological integration is low , the expression of the gene will not be as relevant for determining the shape of the limb as it would be when the integration is high and the impact of the altered gene expression on the phenotype is minimal . If the morphological integration is high , the impact of the genetic mutation will be maximized ( i . e . changes in the gene expression pattern will produce changes in the limb shape ) . Our results showed that the shape of the limb and the shape of the Dusp6 expression domain were indeed highly correlated ( RV = 0 . 88 in forelimbs; RV = 0 . 91 in hindlimbs ) . This is evidence that altered Fgf signaling , induced by the Fgfr2 P253R Apert syndrome mutation , can be associated with limb dysmorphology . By comparing the morphological integration patterns in mutant and unaffected littermates during different periods , we can test whether this interaction is maintained or disrupted by the disease during development . If the pattern or magnitude of morphological integration is different in mutant mice , it should reveal further mechanisms underlying the etiology of the disease . Our analyses showed that the pattern of morphological integration of the shape of the limbs and the shape of the Dusp6 expression domains was similar in unaffected and Fgfr2+/P253R mutant mice ( Figure 6E , Figure 6—figure supplement 2 ) . Our results confirmed that the Fgfr2 P253R mutation does not disrupt the strong association between limb shape and the Dusp6 expression domain . Therefore , the alteration of the Dusp6 expression pattern between E10 and E11 . 5 , caused by the Fgfr2 mutation , will correlate with the limb dysmorphologies associated with Apert syndrome . Overall , the high correlation between the shape of the limb and the Dusp6 expression domain provides further evidence that altered Fgf expression resulting from the Fgfr2 mutation is strongly associated with limb defects in Apert syndrome .
By definition , revealing the primary etiology of an abnormality requires going back in time to the earliest moment when abnormal development can be found . Typically , the earliest changes will be the most subtle , and so the most statistically sensitive techniques are necessary to reveal these changes . The methods that are currently used to assess gene expression patterns are mainly qualitative and focus only on shape and size differences that can be simply detected by eye . Therefore , slight changes in gene expression domains , even if they have large effects on the phenotype ( Honeycutt , 2008 ) , can remain undetected . To reveal these subtle changes , we have developed a precise method combining OPT and GM to quantify embryo morphology and the underlying 3D gene expression patterns in a systematic , objective manner . This enables the visualization and quantification of how the genotype translates into the phenotype during embryonic development , the comparison of normal and disease-altered patterns of genetic and phenotypic variation and , eventually , the identification of the origins of abnormal morphogenesis . This approach can further our understanding of the etiology of genetic diseases in research using animal models ( Spradling et al . , 2006; Rosenthal and Brown , 2007 ) , even in models that do not seem to recapitulate the human disease faithfully ( Guénet , 2011 ) . Our study of the Fgfr2 P253R mouse model for Apert syndrome is an exemplary case demonstrating how quantitative assessment can overcome the shortcomings of traditional qualitative morphological assessment and lead to new discoveries . To date , the molecular and developmental mechanisms underlying the limb defects associated with Apert syndrome have remained obscure , even when these limb abnormalities clinically differentiate Apert syndrome from other craniosynostosis syndromes ( such as Pfeiffer , Crouzon and Saethre-Chotzen syndromes ) ( Park et al . , 1995; Holten et al . , 1997; Cohen and MacLean , 2000 ) . Most Apert syndrome research has focused on premature fusion of cranial sutures and craniofacial malformations ( Cohen and MacLean , 2000; Holmes et al . , 2009; Martínez-Abadías et al . , 2010; Holmes and Basilico , 2012; Hill et al . , 2013; Heuzé et al . , 2014 ) , clinical traits that are consistently phenocopied in mouse models . However , little research has been carried out on the limb defects in Apert syndrome using the same animal models , mainly because most previous research reported that malformations of the limbs are subtle or absent in the different mouse models for Apert syndrome ( Chen et al . , 2003; Wang et al . , 2005; Wang et al . , 2010 ) . Contrary to these previous results , our quantitative morphometric analyses demonstrate that the limbs of Fgfr2+/P253R Apert syndrome mice have significant defects that are detectable in newborn mice and that can be traced back to early embryogenesis ( Figure 1 , Figure 5 and Figure 6 ) . Our analyses provide insight into the genetic origins of these limb defects , showing that altered expression patterns of genes in the Fgf signaling pathway precede and contribute to limb dysmorphogenesis in Fgfr2+/P253R Apert syndrome mice . In fact , we detected that Dusp6 expression patterns were different in unaffected and mutant littermates a few hours before the first limb dysmorphologies appeared ( Figure 5 ) , and confirmed that limb shape and Dusp6 expression patterns were highly correlated ( Figure 6E , Figure 6—figure supplement 2 ) . The altered Fgf signaling that was observed was due to the Fgfr2 P253R Apert syndrome mutation , which causes loss of ligand specificity of the receptor and increased affinity of ligands for Fgfr2 . As the Apert syndrome mutation is located in the linker region between the second and the third Ig-like domains of the receptor , and as alternative splicing of the protein occurs in the carboxy-terminal half of IgIII domain , the mutation affects both isoforms of the receptor , Fgfr2IIIb and Fgfr2IIIc ( Ibrahimi et al . , 2001 ) . Fgfr2IIIb is the isoform that is predominantly expressed in epithelial cells which normally binds to ligands produced in mesenchymal cells , whereas Fgfr2IIIc is expressed in mesenchymal cells and binds to ligands produced in epithelial cells . The available evidence indicates that Apert mutations alter this ligand specificity and thus raise the level of Fgfr2 signaling — either through the aberrant interaction of mutant receptors with inappropriate ligands or through the upregulation of Fgfr2 ( Hajihosseini , 2008 ) and unpublished data ) . Our analyses showed that over-activation of the Fgf signaling pathway results in more expanded ( Figure 5E–G ) and larger ( Figure 6B ) expression domains for Dusp6 , a gene that acts as a negative-feedback control of Fgf signaling ( Ekerot et al . , 2008 ) . A delay in misregulation of the expression of Dusp6 may explain the lack of antero-posterior asymmetry and the shape deficiencies observed in Fgfr2+/P253R mutant mice ( Figure 5 ) . We found evidence that limb shape and Dusp6 expression were highly associated ( Figure 6C–E ) , but it is likely that other downstream genes of the Fgf signaling pathway also contribute to the limb shape malformations associated with Apert syndrome . Our analyses also demonstrated that the embryonic limb defects persisted until birth ( Figure 1 ) and thus did not disappear during development . For instance , we found that Fgfr2+/P253R Apert syndrome mice presented postnatal limb malformations involving the shape , length and volume of many bones of the forelimb , including the scapula , humerus , ulna , radius , metacarpals and phalanges ( Table 1 and Figure 1 ) . It is not possible , however , to correlate early changes in limb bud shape directly to the defects in chondrogenesis and long bone length that appear later in development . The skeletal defects associated with Apert syndrome are the result of continuous altered Fgf signaling throughout growth and development . Overall , activated Fgf signaling may lead to decreased bone growth ( Li et al . , 2007 ) . Our results demonstrate that Dusp6 participates in this process , beginning very early in development , as do many other downstream targets of Fgf signaling that have different functions and spatio-temporal patterns , all of which are involved in the complex process that regulates limb morphogenesis . Although subtle , the first malformations detected in this study suggest that further research into the origins and causes of limb dysmorphologies in Apert syndrome using these and other mouse models is warranted . Our quantitative approach could be similarly applied to investigate other developmental defects and dysmorphologies ( Winter and Baraitser , 1987 ) . This is relevant as major developmental defects represent a leading cause of infant mortality and affect a small but relevant percentage of the population , severely compromising their quality of life ( Toxicology NRC C on D , 2000 ) . OPT and GM can potentially be used to analyze any organ and animal model that can be visualized during development using light microscopy , and any gene whose gene expression pattern can be detected by WMISH and shows a continuous expression domain over development ( Martínez-Abadías et al . , 2016 ) . Our study is a proof of concept that demonstrates that this approach is feasible and develops a protocol that could be adjusted for the study of different genes and organs . With the current pipeline , the implementation of the approach should be straightforward and no more time- or resource-consuming than other cellular or molecular experiments . WMISH labeling is a standard technique in most molecular and developmental laboratories . OPT imaging is increasingly accessible and software for performing GM is freely available . We exemplified the method by analyzing limb defects in mouse models , but it could be applied to malformations affecting the face , the brain , the tail , or any other organ that involves complex 3D shapes that are quantifiable with GM . Organs with morphologies that are devoid of reliable anatomical landmarks , such as the developing heart , should be analyzed with alternative non-landmark-based methods , such as spherical harmonics ( Shen et al . , 2009 ) . Moreover , the OPT-GM approach could be applied to other vertebrate animal models , such as zebrafish , chicken and Xenopus . Finally , our approach could also be extended to analyze protein expression patterns . Immunostaining labeling with a secondary antibody coupled to a BAAB-resistant fluorophore , such as Alexa , will reveal the expression pattern of proteins and this information can be precisely captured and quantified in 3D using the OPT-GM approach . The limitation of the approach is the requirement to generate large samples of wildtype and mutant mice per gene and time point , since several WMISH labeling and OPT scanning activities cannot be performed within the same embryos . For high-throughput analyses assessing the expression of multiple genes , such as transcriptomic and microarray assays ( Yeh et al . , 2013 ) , our OPT-GM approach is a complementary technique that could confirm whether the differences in the level of gene expression of candidate genes are associated with changes in the patterns of this expression ( i . e . the regions within a tissue where the genes are expressed ) and further phenotypic malformations . This type of quantitative analyses will lead to a deeper understanding of how development translates genetic into phenotypic variation . Through its increased quantitative sensitivity , our method has allowed us to reveal that the mouse model for Apert syndrome does indeed show very early abnormalities in limb development . We detected that dysregulation of an Fgfr2 target gene precedes measurable changes in limb bud morphology , thus identifying a relevant component of its genetic etiology . Quantitative evaluation of size and shape should thus be performed before discarding any animal models as useful for investigating human congenital malformations ( Zuniga et al . , 2012 ) . Our method has the potential to become a useful tool for biomedical research , providing insight into the processes that cause malformations and lead to malfunction , which is essential for understanding diseases and discovering potential therapies .
We analyzed the Fgfr2+/P253R Apert syndrome mouse model , an inbred model backcrossed on the C57BL/6J genetic background for more than ten generations , which carries a mutation that in humans with Apert syndrome is associated with more severe syndactyly . This gain-of-function mutation , which is embryonically lethal in homozygosis , involves a proline to arginine amino acid change at position 253 of the Fgfr2 protein , which alters the ligand-binding specificity of the receptor and causes stronger receptor signaling . Further details of the mouse model and on the generation of targeting construct can be found elsewhere ( Wang et al . , 2010 ) . All the experiments were performed in compliance with the animal welfare guidelines approved by the Pennsylvania State University Animal Care and Use Committees ( IACUC46558 , IBC46590 ) . High-resolution micro-computed tomography ( μCT ) scans were acquired by the Center for Quantitative Imaging at the Pennsylvania State University ( www . cqi . psu . edu ) using the HD-600 OMNI-X high-resolution X-ray computed tomography system ( Varian Medical Systems , Palo Alto , CA ) . Pixel sizes ranged from 0 . 01487 to 0 . 01503 mm , and all slice thicknesses were 0 . 016 mm . Image data were reconstructed on a 1024 × 1024 pixel grid as a 16-bit TIFF . To reconstruct forelimb morphology from the μCT images , isosurfaces were produced with median image filter using the software package Avizo 6 . 3 ( Visualization Sciences Group , VSG ) ( Figure 1A–F ) . We assessed forelimb morphology at P0 using unaffected ( N = 10 ) and mutant ( N = 12 ) littermates of Apert syndrome mouse models . A selection of 54 landmarks was collected on the left forelimb of all P0 mice , as shown in Figure 1B–F , following anatomical criteria selected to best describe the size and shape of forelimb bones in a reliable and reproducible manner . Landmarks at the proximal and distal tips of the phalanges , metacarpals , radius , ulna and clavicle were collected to represent the simple tubular shape of these bones . Additional landmarks were collected in anatomical structures of the humerus and scapula to better represent their complex 3D shapes ( Figure 1B–F ) . To minimize measurement error , each landmark was collected twice by the same observer , restricting the deviations between the two trials to 0 . 05 mm . At P0 , we estimated the dimensions of the long bones of the forelimbs using the 3D coordinates of the landmarks located at the proximal and distal ends of the bones ( Figure 1B–F ) . We also estimated the bone volumes from volume data collected from the microCT scans . To assess size differences in bone length and bone volume between mutant and unaffected P0 mice of the Fgfr2+/P253R Apert syndrome mouse model , we performed a two-tailed one-way ANOVA on those variables showing a normal distribution , and the non-parametric Mann-Whitney U-Test on those variables that deviated from a normal distribution . The shape of the humerus and the scapula was comparatively assessed in unaffected and Fgfr2+/P253R Apert syndrome littermates using Geometric Morphometrics . Shape information was extracted using a Generalized Procrustes Analysis ( GPA ) ( Rohlf and Slice , 1990 ) , in which configurations of landmarks are superimposed by shifting them to a common position , rotating and scaling them to a standard size until a best fit of corresponding landmarks is achieved ( Dryden and Mardia , 1998 ) . The resulting Procrustes coordinates from the GPA were the input for further statistical analysis to compare the shape of the bones in unaffected and Fgfr2+/P253R mice . A Principal Component Analysis ( PCA ) was used to explore the morphological variation within each bone . PCA performs an orthogonal decomposition of the data and transforms variance covariance matrices into a smaller number of uncorrelated variables called Principal Components ( PCs ) , which successively account for the largest amount of variation in the data ( Hallgrimsson et al . , 2015 ) . Each specimen is scored for every Principal Component and the specimens can be plotted using these scores along the morphospace defined by the principal axes . To examine early embryonic mouse limb development in Apert syndrome mice , we bred four litters of the Fgfr2+/P253R Apert syndrome mouse model and collected them between E10 . 5 and E11 . 5 . In total , 32 mouse embryos were harvested and classified by PCR genotyping into unaffected ( N = 16 ) and mutant ( N = 16 ) littermates ( see Table 2 and Figure 4 for further details on sample size and composition ) . We analyzed the maximum number of samples that we could process simultaneously within the same WMISH batch , as explained next . Dusp6 gene expression was assessed by whole-mount-in-situ hybridization ( WMISH ) . Mouse embryos were dissected in cold phosphate-buffered saline , 0 . 1% tween 20 ( PBST ) , fixed overnight in 4% paraformaldehyde ( Sigma ) , dehydrated in a graded PBST/methanol series and stored at –20°C in methanol . The mouse embryos recovered their original size after rehydration in decreasing series of methanol/PBST . WMISH was carried out using Dusp6 antisense RNA probes labeled with digoxigenin-UTP ( Roche ) , following standard protocols ( de la Pompa et al . , 1997 ) . Alkaline phosphatase coupled anti-digoxigenin ( anti-DIG-AP , Roche ) and NBT/BCIP staining ( Roche ) were used to reveal the expression pattern for Dusp6 . To minimize variations during experimental procedures , all embryos were processed systematically within the same batch , processing unaffected and mutant littermates from different litters in separate tubes , but simultaneously using the same probe , timings and concentration reagents . After embedding in agarose , dehydrating in methanol and chemically clearing the samples with benzyl alcohol and benzyl benzoate ( BABB ) , the stained whole embryos were scanned with both fluorescence and transmission light with a cyan fluorescent protein ( CFP ) filter using our home-built OPT imaging system mounted on a Leica MZ 16 FA microscope ( Sharpe et al . , 2002 ) . The embryos were 3D-reconstructed from the resulting 2D images using Matlab ( The MathWorks , Inc . ) and visualized using Amira 6 . 3 ( Visualization Sciences Group , FEI ) . From the OPT fluorescence scans , we produced 3D reconstructions of the embryo surface and we dissected the available right and left fore- and hindlimbs of each specimen , resulting in a sample of 100 embryonic limbs ( Table 2 ) . From the OPT transmission scans , we recovered the Dusp6 expression domain . As Dusp6 is expressed in a fuzzy spatial gradient , as already shown by other genes ( Martínez-Abadías et al . , 2016 ) , we used 3D multiple thresholding to visualize the gene expression domain at different intensities ( Figure 3 ) . To analyze the gene expression domains comparatively across the samples , we inspected the whole range of threshold values under which the gene expression could be visualized for each limb , from the threshold showing its first appearance to the threshold under which it disappeared and was no longer detectable . We analyzed the 3D reconstruction on the basis of a threshold computed as 1/3 that of the last grey value showing the Dusp6 expression domain , which displayed a Dusp6 domain at high gene expression ( Figure 3 , step 3 ) . Finally , we obtained 80 limbs ( 46 forelimbs and 34 hindlimbs ) with associated gene expression patterns for Dusp6 . Limbs or gene expression domains that were damaged during experimental manipulation or which showed obvious artifacts resulting from altered staining or imaging were discarded from the analyses . Outlier points were detected following standard protocols in GM that are based on squared Procrustes distance , and landmark misplacements were fixed whenever possible . To account for breeding and developmental variation , individual limb buds were staged using our publicly available web-based staging system ( https://limbstaging . embl . es ) ( Musy et al . , 2018 ) . Considering the spline curve along the outline of the limb , this tool provides a stage estimate with a reproducibility of ±2 hr . According to the staging results , the different mouse litters were ordered following a continuous temporal sequence from E10 to E11 . 5 ( Table 2 and Figure 4 ) . To minimize high developmental variation within and among litters of mice , hindlimbs and forelimbs from each litter were analyzed separately , except for those from two litters from the earliest stage that were pooled into the same group because their temporal distribution completely overlapped , as shown in Figure 4 . To capture the size and shape of the limbs and the expression domains of Dusp6 , we collected a set of anatomical landmarks as well as curve and surface semi-landmarks ( Figure 5—figure supplement 1 ) , as recommended in structures devoid of homologous landmarks . Semi-landmarks are mathematical points located along a curve ( Bookstein , 1997 ) or a surface ( Gunz et al . , 2005 ) within the same object that can be slid to corresponding equally spaced locations across the sample . Only five anatomical landmarks were discernible in the limb , and four in the Dusp6 expression pattern ( Figure 5—figure supplement 1 ) . We defined several curves and surfaces over the limb as well as gene expression structures with a relatively low number of semi-landmarks in order to minimize the high dimensionality of the data while providing an adequate shape coverage that precisely captured the subtle morphological changes associated with the Apert syndrome mutation . We used Amira 6 . 3 ( Visualization Sciences Group , FEI ) to record the anatomical landmarks and Viewbox 4 ( dHAL software , Kifissia , Greece ) to construct a limb template of surface and curve semi-landmarks and to interpolate them onto each target shape ( Figure 5—figure supplement 1 ) . The 3D landmark coordinates defining the shape of the limb and the Dusp6 expression domain were analyzed using Procrustes-based landmark analysis ( Bookstein , 1997 ) . Semi-landmarks were allowed to slide in the GPA by minimizing the bending energy ( Bookstein , 1997; Gunz et al . , 2005; Mitteroecker and Gunz , 2009 ) . Quantitative shape analyses based on PCA were performed as explained above . We estimated the size of the limb as the centroid size , calculated as the square root of the summed squared distances between each landmark coordinate and the centroid of the limb configuration of landmarks ( Dryden and Mardia , 1998 ) . The volumes of the Dusp6 domains were estimated from the 3D reconstructions of the OPT scans . Differences in limb size and gene expression volume between mutant and unaffected embryonic mice were tested for statistical significance using a two-tailed Welch Two Sample t-test . We quantified the integration between the limb and the Dusp6 expression pattern and produced visualizations of the patterns of associated shape changes between them using two-block Partial Least Squares analysis ( PLS ) ( Rohlf and Corti , 2000 ) . This method performs a singular value decomposition of the covariance matrix between the two blocks of shape data ( i . e . , the limb and the Dusp6 expression domain ) . Uncorrelated pairs of new axes are derived as linear combinations of the original variables , with the first pair accounting for the largest amount of inter-block covariation , the second pair for the next largest amount and so on . The amount of covariation is measured by the RV coefficient , which is a multivariate analogue of the squared correlation ( Klingenberg , 2009 ) . Statistical significance was tested using permutation tests under the null hypothesis of complete independence between the two blocks of variables . Separate analyses for each developmental period , as well as for forelimbs and hindlimbs of all stages , were computed . All of the analyses were performed using R ( R Development Core Team , 2014; http://www . R-project . org ) ; the R package geomorph ( Adams and Otárola-Castillo , 2013; http://cran . r-project . org/web/packages/geomorph ) , SPSS Statistics 22 ( IMB , 2013 ) and MorphoJ ( Klingenberg , 2011 ) . | Our development in the womb is complex . Genes need to switch on and off in a precise order , controlling the activity of millions of cells as they work together to form different tissues . For everything to happen smoothly , cells must use instructions provided by each gene exactly at the correct moment and in the correct place . In this biological assembly line , the slightest change can lead to a defect . Certain genetic mutations can change when and where cells use particular genes , and this can cause errors in development . These kinds of mutations are a common cause of birth defects , but we cannot always pinpoint how they begin . For example , a single mutation in a gene called FGFR2 causes malformations in the head , the heart and the limbs in a rare disease called Apert syndrome . The first signs that development has gone wrong can be subtle changes in the use of certain genes , impossible to detect with standard methods . As development continues , other processes can mask the impact of problems with certain genes . Ultimately , changes alter the shape of the developing embryo . Genetically engineered mouse models can mimic the gene defects that cause disease in humans . But current methods are not sensitive enough to detect the very first signs of defects . Now , Martínez-Abadías et al . developed a new method to detect these subtle changes and reveal the precise moment when development starts to go wrong . In mice , a specific mutation in the FGFR2 gene affects the activity of a series of other genes . To track the levels of one of these genes , Martínez-Abadías et al . marked mouse embryos using a chemical label . Scanning the embryos then revealed the pattern of the cells using the gene during the earliest stages of development . In mice carrying a mutation in the FGFR2 gene , subtle changes in gene expression began just a few hours after their limbs start to develop . But it took another half a day to see the effects of these changes on the shape and size of the growing limbs . This approach revealed changes in gene expression before any problems with development were visible by eye . Tracking subtle changes in the way cells use genes could allow us to detect the origins of embryo malformations before they appear , pointing at the best moment to start a treatment . With further development , the model could extend to other genes , proteins , animal models and diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2018 | Quantification of gene expression patterns to reveal the origins of abnormal morphogenesis |
Whether somatic mutations contribute functional diversity to brain cells is a long-standing question . Single-neuron genomics enables direct measurement of somatic mutation rates in human brain and promises to answer this question . A recent study ( Upton et al . , 2015 ) reported high rates of somatic LINE-1 element ( L1 ) retrotransposition in the hippocampus and cerebral cortex that would have major implications for normal brain function , and suggested that these events preferentially impact genes important for neuronal function . We identify aspects of the single-cell sequencing approach , bioinformatic analysis , and validation methods that led to thousands of artifacts being interpreted as somatic mutation events . Our reanalysis supports a mutation frequency of approximately 0 . 2 events per cell , which is about fifty-fold lower than reported , confirming that L1 elements mobilize in some human neurons but indicating that L1 mosaicism is not ubiquitous . Through consideration of the challenges identified , we provide a foundation and framework for designing single-cell genomics studies .
The mechanisms that generate the immense morphological and functional diversity of neurons in the human brain have long been a subject of speculation and controversy . The immune system , with its systematic genomic rearrangements such as V ( D ) J recombination , and the ordered generation of random somatic mutation coupled with a selection process , have suggested appealing analogies for generating the cellular diversity of the nervous system , and have led to searches for analogous genomic diversity in the brain ( Muotri and Gage , 2006 ) . LINE-1 ( L1 ) elements are endogenous retrotransposons that transcribe an RNA copy that is reverse-transcribed into a DNA copy that can then insert into a novel site in the genome , creating mutations capable of disrupting or modifying the expression of genes in which they insert or neighboring genes ( Goodier and Kazazian , 2008 ) . Evolutionarily , transposon mobilization is an essential cause of the generation of species diversity ( Cordaux and Batzer , 2009 ) , so interest in possible L1 activity during brain development was spurred by the discovery that these elements can mobilize in neuronal progenitor cells ( Coufal et al . , 2009; Muotri et al . , 2005 ) . The importance of any mutation process , such as retrotransposon mobilization , in generating neuronal diversity is constrained by the rate at which mutation takes place , since if a given type of mutation occurs infrequently , it is unlikely to be a useful generator of diversity . Single-cell sequencing is a powerful technology that has revealed and quantified previously unknown mechanisms of somatic mutation in the human brain , providing a first proof of principle for the systematic measurement of somatic mutation rates in any normal human tissue ( Evrony et al . , 2012; McConnell et al . , 2013; Cai et al . , 2014; Evrony et al . , 2015; Lodato et al . , 2015 ) . Single-cell sequencing can therefore determine the extent to which somatic mutations diversify the genomes of cells in the brain , which is foundational to understanding their potential functional impact in normal brains and possible roles in neuropsychiatric diseases of unknown etiology ( Poduri et al . , 2013 ) . L1 mobilization has been observed at low rates using indirect genetic techniques such as a transgenic L1 reporter in rodent brain in vivo ( Muotri et al . , 2005; 2010 ) and human progenitor cells in vitro ( Coufal et al . , 2009 ) , while studies profiling human brain bulk DNA suggested much higher rates ( Baillie et al . , 2011; Bundo et al . , 2014; Coufal et al . , 2009 ) . Single-cell sequencing has been proposed as the definitive method to resolve these disparate estimates ( Erwin et al . , 2014; Goodier , 2014 ) . A recent single-cell sequencing study ( Upton et al . , 2015 ) reported high rates of somatic L1 retrotransposition in the hippocampus ( 13 . 7 per neuron on average ) and cerebral cortex ( 16 . 3 per neuron ) , and suggested that L1 retrotransposition was "ubiquitous" . Such a high rate of retrotransposition could present it as a possibly essential event in neurogenesis and would have major implications for brain function . Here we describe experimental artifacts that elevated the study's apparent rate of somatic retrotransposition by >50-fold . Reanalysis of their data while filtering these artifacts generates a consensus that retrotransposition does occur in developing brain but at a much lower rate consistent with prior single-cell studies ( Evrony et al . , 2012; Evrony et al . , 2015 ) , thereby constraining the range of possible functional roles for retrotransposition in the brain . Our discussion of the challenges in single-cell sequencing may provide a useful framework for the design and analysis of single-cell genomics studies .
Upton et al . isolated single neuronal cells from postmortem human brains and amplified their genomes using the MALBAC method ( Zong et al . , 2012 ) . They then profiled human-specific L1 elements ( L1Hs ) using their RC-seq method ( Shukla et al . , 2013; Upton et al . , 2015 ) that captures and amplifies both the 5' and 3' ends of L1Hs elements via oligonucleotide hybridization and PCR , hence providing sequence data that identify L1Hs insertion sites in the genome . Although whole genome amplification methods have remarkable abilities to amplify picogram quantities of DNA from a single cell into microgram quantities , chimeric DNA molecules falsely linking unrelated DNA fragments are well known to arise during single-cell genome amplification ( Macaulay and Voet , 2014 ) . Chimera artifacts also arise during ligation and PCR steps of sequencing library preparation ( Kircher et al . , 2012; Quail et al . , 2008 ) , processes integral to the RC-seq method . Chimeric sequences can create a DNA fragment connecting a LINE element to an unrelated portion of the genome , creating the appearance of biological LINE mobilization ( Figure 1—figure supplement 1A–B ) . Sequence analysis of Upton et al . 's putative PCR-validated candidates ( Table S2 of Upton et al . ) demonstrates that more than half of them ( 7 of 13 ) are chimera artifacts that could not have been generated by the process of L1 mobilization ( Supplementary file 1 ) . Some chimeras originated immediately downstream of germline L1Hs/L1Pa elements that are incapable of retrotransposition , because the L1 elements are truncated , are from an old , inactive L1 subfamily , or contain numerous inactivating mutations ( Figure 1B; Supplementary file 1 ) . In other cases , L1 5' and 3' junction chimeras originating from distinct L1 elements were misinterpreted as two ends of the same L1 ( Figures 1B–C; Supplementary file 1 ) . Some candidates lack poly-A tails ( Figure 1C; Supplementary file 1 ) , a key feature of retrotransposon insertions . Although the remaining 6/13 PCR-validated insertions lack clear evidence of being chimeras , the possibility cannot be excluded based on the limited PCR validation performed , and they are likely also chimeras because each is supported by only 1 or 2 sequencing reads ( see below ) . The presence of chimeric artifacts among a set of insertions passing limited PCR validation supports the importance of additional careful analysis of candidate L1 sequences to help define more accurate rates of retrotransposition . 10 . 7554/eLife . 12966 . 003Figure 1 . Chimera artifacts in RC-seq . ( A ) Full-length ( FL ) PCR using primers flanking the insertion site is necessary for definitive validation of somatic insertions in single cells in the setting of chimeras . One breakpoint per chimera DNA molecule refers to the breakpoint of the candidate insertion being analyzed since a DNA molecule can in principle have multiple different chimera events each involving different loci ( which would be unlikely to create a structure that would validate by FL-PCR ) . For most RC-seq candidates , Upton et al . did not attempt 3’ or 5’ PCR for the computationally identified junction and only performed this for the opposite junction . ( B ) Top schematic illustrates one of several methods for identifying L1 chimeras in next-generation sequencing data such as RC-seq . Bottom schematic illustrates how two independent chimeras aligning to the same locus appear to have a TSD . ( C ) An example somatic insertion candidate that passed Upton et al . single-junction PCR validation but derived from two independent chimera artifacts . Yellow region is non-L1 sequence from chromosome 3 that allows tracing of the chimera to its source . L1Pa4 is an inactive L1 subfamily ( Hancks and Kazazian , 2012 ) . See Supplementary file 1 ( 'RC-seq | Somatic L1 PCR' sheet ) for analyses of all somatic candidates passing Upton et al . PCR validation . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 00310 . 7554/eLife . 12966 . 004Figure 1—figure supplement 1 . Overview of single-cell L1 profiling methods and chimeras in the context of genome amplification , analysis , and PCR validation ( A ) MDA and MALBAC amplification of single-cell genomes and downstream L1 profiling steps . MALBAC generates shorter amplicons than MDA as well as peaks and troughs in genome coverage ( see Figure 4 ) . ( B ) Schematic of sequencing reads in WGS , L1-IP , and RC-seq L1 profiling methods . WGS and L1-IP filter most chimeras using a score model based on read count and other parameters , while RC-seq does not employ a read count filter , leading to over-calling of false-positive chimeras . ( C ) PCR validation approach in each L1 profiling method . Schematic on right illustrates short amplicon length of MALBAC , which precludes definitive validation of L1 insertions longer than the MALBAC amplicon length . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 004 Upton et al . performed 3' junction PCR validation for 10 RC-seq candidates for which they had detected the 5’end , and while 3’ junction validation is technically straightforward , it failed for all 10 . Whereas the authors attribute their validation failure to poly-A tails obstructing PCR amplification , poly-A tails do not obstruct PCR: the single-cell RC-seq method used by Upton et al . itself entails three PCR steps in which L1 3' junctions are amplified , and 3' junction PCR is the standard validation approach used in L1 studies ( Ewing and Kazazian , 2010; Grandi et al . , 2012; Huang et al . , 2010; Iskow et al . , 2010 ) . The failure of all 10/10 3' junction validation attempts suggests a high prevalence of false-positives among RC-seq candidates detected with only 5' junction reads , which represent 27% of all candidate insertions . “Full-length” validation is the most accurate method to screen out false-positive candidate somatic insertions . A bona fide L1 insertion creates two genome breakpoints at the insertion site , one on each end of the insertion ( 5' and 3' ) , while a chimera has only one breakpoint at the called insertion site ( Figure 1A ) . Full-length cloning validation , in which the entire L1 insertion is amplified in a single DNA molecule spanning both breakpoints ( using primers based in the genomic sequence flanking the L1 ) , is therefore the only way to confirm that both breakpoints are present in the same DNA molecule as a bona fide insertion , as opposed to two different chimeric molecules ( Figure 1A; Figure 1—figure supplement 1 ) . However , Upton et al . did not perform full-length cloning validation on any insertion . Two independent chimeras can even occur by chance in two different DNA molecules/copies whose non-L1 sequences overlap the same genomic locus , giving the false appearance of a target site duplication ( TSD ) ( Figures 1B–C; Supplementary file 1 ) . In single-cell sequencing , especially when read count is not used to filter candidates ( see below ) , full-length cloning of at least some candidate insertions is important to exclude chimera artifacts . Instead of full-length cloning validation ( Evrony et al . , 2015; Stewart et al . , 2011 ) , Upton et al . carried out multiple 5' junction PCR reactions per candidate using primers spaced every 500 bp along the ~6000 bp L1 . Multiple PCR reactions , each with an L1 primer matching hundreds to thousands of genomic loci , introduces additional mechanisms for generating false-positives . Some candidates required nested PCR with 62 PCR cycles , an extremely high level of amplification , suggesting that the targets are chimeras present at very low level in the single-cell amplified DNA . The MALBAC method employed by Upton et al . for single-cell genome amplification probably precludes definitive full-length cloning validation of some insertions , suggesting it is not an ideal method for studying retrotransposition . MALBAC produces short amplicons ( 0 . 5–1 . 5kb ) compared to multiple displacement amplification ( MDA ) ( 10–50 kb amplicons ) ( Dean et al . , 2002; Zhang et al . , 2015; Zong et al . , 2012 ) , so insertions longer than ~1 . 5 kb ( ~15–30% of somatic L1 insertions in human cancer studies [Helman et al . , 2014; Lee et al . , 2012; Tubio et al . , 2014] ) would not be efficiently validated in MALBAC single cells ( Figure 1—figure supplements 1A and C ) . The many chimeras among ‘validated’ somatic insertions come about because Upton et al . relied solely on computational analysis of contig sequences ( assembled from short sequencing reads ) to filter chimeras and labeled most remaining candidates as putative somatic insertions . However , sequence analysis of short contigs that do not span the full length of insertions can identify chimeras but cannot rule them out . For example , 5' junction chimeras originating from inside an L1 element or 3' junction chimeras originating from within the poly-A tail cannot be distinguished from true insertion breakpoints by sequence analysis alone and require further experiments ( i . e . , full-length cloning ) ( Figure 1A ) . Furthermore , even some chimeras that can be identified computationally were not filtered: one recurrent type of chimera artifact accounts for 16% of their candidates detected only at a 5' junction ( Supplementary file 1; see example candidate chr11:112602973 ) . Further analyses , as well as long-read sequencing ( e . g . , PacBio ) , may reveal additional ways to remove chimeras computationally by sequence analysis alone; but with short-read sequencing , even ideal sequence-based filtering algorithms cannot filter chimeras originating from within L1 . A core principle of next-generation sequencing analysis is the use of read counts to distinguish true mutations from artifacts that inevitably arise during DNA sequencing ( Robasky et al . , 2014; Sims et al . , 2014 ) . Multiple reads supporting a mutation serves the same role as replication does in any scientific experiment , increasing the confidence that the finding is not an artifact . This is especially important in single-cell sequencing where chimeric DNA artifacts are more prevalent than in standard sequencing ( Macaulay and Voet , 2014 ) . Essentially all major mutation-detection algorithms use the signal strength ( read count and often other parameters ) of known true mutations and false-positive events to predict the likelihood that individual candidate mutations are real and to determine a signal cutoff ( Chen et al . , 2009; Cibulskis et al . , 2013; DePristo et al . , 2011; Mills et al . , 2011; 1000 Genomes Project Consortium , 2012 ) . However , Upton et al . did not employ a read count filter or signal model and therefore considered candidate insertions supported by only a single read as equivalent to the smaller number of candidates with higher read support . As a result , 97% ( 4634/4759 ) of their single-cell insertion calls were supported by a single Illumina sequencing read and 99 . 6% by 1 or 2 reads; 94% of their >320 , 000 candidates from 'bulk' DNA were also supported by only 1 read ( Figure 2A ) . 10 . 7554/eLife . 12966 . 005Figure 2 . RC-seq read count distributions and junction detection rates of somatic insertion candidates are inconsistent with known true germline insertions . ( A ) RC-seq read count distributions of bulk and single-cell germline known non-reference ( KNR ) insertion and somatic candidate calls ( see 'Materials and methods' ) . Inset schematics and labels 'a' to 'd' illustrate key findings . Label 'a' ( inset schematic ) points to the subset of KNR insertions appearing at low read counts in single-cells , distinct from the distribution of KNR insertions in bulk samples , due to dropout/non-uniformity at length scales < 30 kb inherent to MALBAC amplification ( Appendix 2 ) . These factors are also responsible for the broader distribution of higher read count KNR insertions ( label 'b' ) in single-cell versus bulk samples . Areas labeled 'c' in the top and bottom graphs highlight the population of single-cell KNR insertions at high read counts that is absent from single-cell somatic candidates . KNR insertions present in a single copy per cell ( chrX insertions in male samples ) show the same pattern ( Figure 2—figure supplement 1A ) . Instead , single-cell somatic candidates appear at very low read counts ( label 'd' , inset schematic ) . The likely bona fide insertion detected in two single cells on chromosome 6 is labeled and appears at high read count relative to other somatic candidates . Purple dashed line indicates threshold of > 2 reads used for calculation of somatic retrotransposition rates . See also Figure 2—figure supplement 1 . ( B ) L1 junction detection rates in bulk and single-cell RC-seq ( see 'Materials and methods' ) . Fewer KNR insertions are detected at both ( 5' and 3' ) junctions in single-cell versus bulk samples due to MALBAC amplification dropout/non-uniformity . A significantly lower fraction of single-cell somatic candidates are detected at both junctions relative to single-cell KNR insertions , confirming the vast majority of somatic candidates are false-positives . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 00510 . 7554/eLife . 12966 . 006Figure 2—figure supplement 1 . Read count distributions of known germline insertions in different L1 profiling methods . ( A ) Read count distributions of chrX ( hemizygous ) KNR insertions , showing that most germline insertions present in a single copy per cell are still detected by multiple reads in single-cell RC-seq . KNR insertions are defined as in Figure 2A . Note that most KNR insertions in any individual are in the heterozygous state ( i . e . single copy per cell ) such that the read count distribution of all genome-wide KNR insertions ( Figure 2A ) still provides a good approximation of somatic insertions ( see 'Materials and methods' ) . ( B ) Read count distribution of germline gold-standard KNR insertions in bulk and single-cell samples from Evrony et al . ( 2015 ) detected by the high-coverage whole-genome sequencing ( WGS ) L1 profiling method , illustrating that true insertions appear at high read counts . See Figure S4B in Evrony et al . ( 2015 ) 2B and S11 for histograms of insertion confidence scores . WGS gold-standard KNR insertions are defined as insertions identified in both bulk samples of the individual and detected in prior independent L1 profiling studies from other laboratories ( see 'Materials and methods' for details ) . ( C ) Read count distribution of germline gold-standard KNR insertions in bulk and single-cell samples from Evrony et al . ( 2012 ) detected by the targeted L1-insertion profiling ( L1-IP ) method , illustrating that true insertions appear at high read counts . Raw ( top ) and normalized ( bottom ) read counts are shown . See Figure S4B in Evrony et al . ( 2012 ) for further read count and score histograms . The minimum number of raw reads for calling insertions in the pipeline is 10 , as described in Evrony et al . ( 2012 ) . L1-IP gold-standard KNR insertions are defined as insertions identified with a confidence score ≥0 . 5 in at least half of the bulk samples of the individual and detected in prior independent L1 profiling studies from other laboratories ( see 'Materials and methods' for details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 006 Upton et al . 's rationale for not using read counts in their analysis is their suggestion that in their single-cell RC-seq method , chimeras appear at higher read counts than true insertions such that nearly all true insertions would be detected by only 1 read . This proposal can be tested using the read count distribution of a 'gold standard' mutation set . In single-cell samples , somatic insertions should appear at the same signal level distribution as germline known non-reference L1 insertions ( KNR ) , which are population-polymorphic L1 insertions absent from the reference human genome but identified in prior L1 studies . Germline KNR insertions share the same sequence characteristics as somatic insertions ( Helman et al . , 2014; Lee et al . , 2012; Tubio et al . , 2014 ) and bear no distinguishing feature that would lead to different read counts . Therefore , KNR insertions can be used to directly test Upton et al . 's model that true insertions preferentially appear at lower read counts than chimeras . Using RC-seq single-cell germline KNR insertion data provided by the authors upon request , we found that KNR insertions were detected by much higher read counts than candidate somatic insertions . In single-cell RC-seq samples , 53% , 24% and 20% of the 4049 calls of high-confidence gold-standard KNR insertions were detected with ≥3 , ≥20 and ≥40 reads per sample , respectively; only 32% were detected with only 1 read ( Figure 2A; Figure 2—figure supplement 1A ) . In contrast , 97% ( 4634/4759 ) of single-cell somatic insertion candidates were detected with only 1 read and only 0 . 4% ( 20/4759 ) with ≥3 reads ( Figure 2A ) . The strikingly higher read depths of gold-standard germline KNR L1 insertions relative to somatic insertion candidates in the same experiment is consistent with the vast majority of claimed somatic insertions not corresponding to bona fide insertions . Analysis of RC-seq L1 junction detection rates provides additional evidence that nearly all somatic candidates are false-positives ( Figure 2B ) . 11% of single-cell KNR insertion calls were detected at both L1 ( 5' and 3' ) junctions , whereas >250-fold less— only 0 . 04% ( 2/4682 ) — of single-cell somatic insertion candidates were detected at both junctions . Sequence analysis shows 8 of the 12 hippocampal single-cell somatic candidates detected at both junctions ( including candidates in which each junction was detected in a different sample ) are chimera artifacts ( Supplementary file 1 ) . The remainder cannot be excluded as chimeras without full-length PCR validation . Furthermore , RC-seq bulk somatic candidates have a non-canonical distribution of large TSD sizes , inconsistent with nearly all prior L1 research ( Appendix 1 ) . Analysis of 10 randomly selected candidates with large ( >50 bp ) TSDs found all were chimera artifacts ( Supplementary file 1 ) . A more plausible RC-seq somatic insertion rate can be calculated using a read count threshold calibrated to germline high-confidence KNR insertions as a gold standard . A read count threshold of > 2 optimizes sensitivity and specificity , maintaining detection of 53% of true positive KNR insertion calls across all single cells ( Figure 2A ) and a per-cell KNR detection sensitivity of 24% ( Figure 3A ) , while excluding ~99 . 6% of false-positive calls ( see 'Materials and methods' ) . Only 20 somatic insertion candidates supported by >2 reads were detected across all 170 cells , and 12 of these were chimeras upon further sequence analysis ( Supplementary file 1 ) . The remaining 8 candidates yield a sensitivity-corrected somatic insertion rate estimate of 0 . 19 per cell , with no significant difference in rates between cell types ( hippocampal neurons and glia , cortical neurons , and AGS hippocampal neurons ) ( p = 0 . 98 , ANOVA ) ( Figure 3B ) . 95% of single cells did not have any somatic insertion candidates ( excluding chimeras ) supported by >2 sequencing reads . These RC-seq somatic insertion rates are quite consistent with rates previously estimated by L1 insertion profiling ( L1-IP ) in single cortical and caudate neurons ( 0 . 07 ± 0 . 15 ( SD ) ; p = 0 . 54 , ANOVA ) ( Evrony et al . , 2012 ) , and using single-neuron whole-genome sequencing ( 0 . 18 ± 0 . 47 ( SD ) ; p = 0 . 37 , ANOVA ) ( Evrony et al . , 2015 ) , suggesting a notable consensus by three methods confirming that somatic L1 insertions are present in human brain , but fewer than one per average genome . 10 . 7554/eLife . 12966 . 007Figure 3 . RC-seq sensitivity for gold-standard true insertions and corrected RC-seq somatic insertion rates . ( A ) Average sensitivity of single-cell RC-seq for gold-standard KNR insertions at different read count thresholds . Sensitivity of single-cell L1-IP ( Evrony et al . , 2012 ) and single-cell WGS ( Evrony et al . , 2015 ) are shown for comparison . Note , the average number of uniquely mapped reads in the targeted enrichment methods of L1-IP and RC-seq are 3 . 2 and 16 . 7 million reads , respectively , so L1-IP achieves higher sensitivity than RC-seq with fewer reads even with a more liberal read count threshold for RC-seq . Gold-standard KNR insertions are defined for each single-cell method as in Figures 2A and Figure 2—figure supplement 1B–C . Error bars ± SD . As illustrated in the schematic on the left , Σm is the number of single cells in the study ( i . e . mA+mB+ . . . ) , and Σn is the number of gold-standard KNR insertions used to calculate sensitivity across the profiled individuals ( i . e . nA+nB+ . . . ; as seen in the schematic , Σn increases as more individuals are profiled ) . See also Figure 3—figure supplement 1 . ( B ) Average RC-seq somatic insertion rates per cell . These are pre-PCR validation rates , since Upton et al . did not attempt PCR validation for these somatic candidates . The percentage of cells without any candidates ( above the threshold of >2 reads and after excluding chimeras ) is shown . See Supplementary file 1 ( "RC-seq | Somatic L1 >2 reads" sheet ) for analysis of all somatic candidate sequences . Error bars ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 00710 . 7554/eLife . 12966 . 008Figure 3—figure supplement 1 . Single-cell sensitivity of L1-profilng methods for gold-standard germline KNR insertions . Single-cell sensitivity of L1-profiling methods for gold-standard germline KNR insertions at additional read count and score thresholds not shown in Figure 3A , and separately for single-cell samples from each individual . RC-seq Aicardi-Goutieres ( AGS ) single cells have lower sensitivity ( higher dropout ) than single cells from other individuals , indicating lower quality tissue/single cells from this individual . Gold-standard KNR insertions are defined for each single-cell method as in Figures 2A and Figure 2—figure supplement 1B–C . Error bars ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 008 Notably , 2 of the 8 somatic insertion candidates detected following read count filtering correspond to a single , likely bona fide L1Hs insertion in neuron #11 ( 12 reads ) and glial cell #2 ( 6 reads ) from the hippocampus of individual 45 ( Supplementary file 1 ) . This intergenic insertion shows RC-seq reads capturing both 5' and 3' junctions bearing all the hallmarks of a true retrotransposition event: a TSD , poly-A tail , and a 3' transduction that traces its source to a population-polymorphic ( KNR ) L1 on chromosome 2 that was identified in a prior L1 profiling study ( Iskow et al . , 2010 ) . This same somatic insertion was also detected in glial cells #7 and #8 of the individual , each with 2 reads . Upton et al . highlighted this insertion for its detection in multiple cells but did not note its high-signal level—this candidate had the 5th and 9th highest read counts of all 4759 somatic candidate calls ( Figure 2A ) . This clonal retrotransposon event also showed >1 read in all 4 cells in which it was detected and was detected at both 5' and 3' junctions . The basic signal characteristics of this one clear somatic insertion event make it dramatically different from those of the thousands of other somatic insertions proposed by Upton et al . ( Figures 2A–B ) . MALBAC-amplified single cells profiled by RC-seq had reduced performance relative to bulk RC-seq in terms of gold-standard KNR insertion read counts and junction detection rates ( Figures 2A–B ) , and had significantly lower sensitivity for KNR insertions ( higher dropout ) than L1 profiling of MDA-amplified single cells ( Figure 3A; Figure 3—figure supplement 1 ) . We therefore further studied the quality of Upton et al . 's single cells and the performance of the MALBAC method ( Zong et al . , 2012 ) that the authors used for single-cell genome amplification . Analysis of Upton et al . 's pre-RC-seq whole-genome sequencing of MALBAC-amplified single cells shows that at genomic scales < 50 kb ( high-resolution view ) , which includes the size range of retrotransposons and single-nucleotide variants ( SNV ) , there are systematic ~1 kb peaks of high genome amplification separated by troughs of low amplification or complete dropout ( Figure 4A ) . These peaks and troughs often occur in the same locations as in MALBAC single cells from an unrelated study by Zong et al . ( 2012 ) ( Figure 4A ) , suggesting that this non-uniformity in genome amplification is inherent to MALBAC . In contrast , MDA single cells show significantly better uniformity of genome amplification at these size scales ( Figure 4A ) . The non-uniformity of MALBAC at genomic scales encompassing the size range of retrotransposon elements likely explains the subset of true KNR insertions appearing at low read counts ( Figure 2A ) and the low sensitivity ( high allelic dropout ) of single-cell RC-seq ( Figure 3A ) . It also explains MALBAC's lower overall breadth of genome-wide coverage at nucleotide resolution ( i . e . higher locus dropout ) relative to MDA ( Figure 4—figure supplement 2A ) . 10 . 7554/eLife . 12966 . 009Figure 4 . MDA and MALBAC single-cell genome amplification uniformity . ( A ) High-resolution coverage plots of MDA single neurons ( Evrony et al . , 2015 ) and MALBAC single cells from Zong et al . ( 2012 ) and Upton et al . ( 2015 ) . MALBAC samples show significant non-uniformity with systematic high peaks ( stars ) and troughs of genome amplification . MALBAC single neurons from Upton et al . were pooled from hippocampus ( n = 92 cells ) and cortex ( n = 35 cells ) of normal individuals to produce high-coverage samples for the plots . Pooling eliminates stochastic noise of individual cells but preserves systematic non-uniformity inherent to MALBAC . Area shown is chr2:155 , 815 , 550–155 , 848 , 725 encompassing the region of one of the single-cell RC-seq somatic L1 candidates detected with both 5' and 3' junctions ( chr2:155 , 823 , 436 ) . Red lines mark off-scale peaks . ( B ) Low-resolution ( ~500 kb bin ) genome-wide coverage plots of representative single cells from the above studies . MALBAC single cells from Zong et al . have significantly better uniformity at these scales than MDA single neurons as measured by median absolute pairwise deviation ( MAPD ) and median absolute deviation from the median ( MDAD ) scores ( lower scores indicate higher uniformity ) ( Cai et al . , 2014; Evrony et al . , 2015 ) . In contrast , individual MALBAC single cells from Upton et al . have significantly lower quality than both MALBAC single cells from Zong et al . and MDA single neurons . Pooling of all 92 normal hippocampus single neurons from Upton et al . achieves high uniformity ( low MAPD/MDAD scores ) , indicating the low quality of individual single cells from Upton et al . is due to stochastic noise , likely from factors preceding MALBAC amplification . Note , high-coverage MALBAC and MDA samples from Zong et al . and Evrony et al . were subsampled to a lower read depth similar to read depth of Upton et al . samples , confirming prior analyses showing uniformity quality metrics are not affected by sequencing depth in low resolution analyses ( Evrony et al . , 2015 ) . ( C ) Power spectral density ( y-axis ) , which reflects the degree of read depth variability ( uniformity ) as a function of genomic spatial frequency ( x-axis ) . Higher spatial frequencies ( right side of x-axis ) reflect smaller genomic scales ( i . e . higher resolution , as in Figure 4A ) , and lower spatial frequencies ( left side of x-axis ) reflect larger genomic scales ( i . e . lower resolution , as in Figure 4B ) . Plots show differences in MDA and MALBAC genome amplification uniformity across genomic scales: MDA single cells have greater read depth variability at larger genomic scales than MALBAC single cells [label I] , while MALBAC has greater read depth variability at smaller genomic scales < 30 kb [label II] ( i . e . scale of SNVs , small indels , retrotransposons; frequencies > ~3 . 5·10–5 bp ) , consistent with high-resolution coverage plots ( Figure 4A ) . MALBAC single cells from Upton et al . were pooled to obtain high-coverage samples for the analysis . Plots for individual 1465 and SW480 samples were calculated in Evrony et al . ( 2015 ) and are presented again for comparison to Upton et al . samples . Additional unrelated bulk sample NA12877 is plotted for comparison . See Appendix 2 for additional details , Figure 4—figure supplement 1 for average MAPD/MDAD scores of single cells and additional coverage plots , and Figure 4—figure supplement 2 for basic genome coverage statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 00910 . 7554/eLife . 12966 . 010Figure 4—figure supplement 1 . MDA and MALBAC single-cell quality and low-resolution genome-wide amplification uniformity . ( A ) MAPD and MDAD statistics of copy number ( coverage ) variability calculated in ~500 kb bins spanning the genome . Lower MAPD and MDAD scores reflect higher quality and uniformity of single-cell amplification . References used for normalization of genome coverage are listed . MALBAC single cells from Upton et al . have higher variability in quality/amplification uniformity among single cells and lower average quality/amplification uniformity than both MALBAC single cells from Zong et al . ( 2012 ) and MDA single neurons from Evrony et al . ( 2015 ) . However , pooling of MALBAC single cells from Upton et al . achieves high quality and uniformity ( low MAPD and MDAD scores ) better than single-cell samples from other studies , indicating the low quality of individual single cells from Upton et al . is due to stochastic noise in each single cell likely from factors preceding MALBAC amplification such as poor tissue quality . The higher sequencing depth of MDA single neurons from Evrony et al . and MALBAC single cells from Zong et al . relative to MALBAC single cells from Upton et al . has minimal effect on MAPD/MDAD statistics since Poisson error of read counts in low-resolution 500 kb bins is well below the noise levels due to single-cell amplification even at low sequencing depths . This is confirmed by MAPD/MDAD statistics after subsampling high-coverage samples to lower read depths and by prior analyses in Evrony et al . ( 2015 ) . ( B ) Representative genome-wide copy number ( coverage ) plots in ~500 kb equal-read bins for unamplified bulk DNA and MDA single-neuron samples from Evrony et al . ( 2015 ) ( top row ) , MALBAC single cells from Zong et al . ( middle row ) , and MALBAC single cells from Upton et al . ( bottom row ) . Bin copy numbers are relative to the specified reference sample . These genome-wide copy number plots were used to calculate MAPD and MDAD statistics shown in panel A . MAPD and MDAD dispersion statistics are shown for each sample . Orange lines denote ± 1 copy . Purple points are off scale . Note the high quality of MALBAC single cells from Zong et al . at this resolution that achieve uniformity similar to the MDA 100-neuron sample . In contrast , the highest quality MALBAC hippocampus single neuron from Upton et al . ( neuron 45–15 ) has uniformity similar to an average MDA single neuron and an average quality MALBAC neuron from Upton et al . ( neuron 42–1 ) has significantly less uniformity . Although pooling of all Upton et al . normal hippocampus single neurons has uniformity similar to the highest quality MALBAC single cell from Zong et al . , this reflects only systematic MALBAC noise since stochastic single-cell noise is mostly removed by pooling . On the other hand , individual MALBAC single cells from Zong et al . achieve high quality with both stochastic single-cell noise and MALBAC noise . This reflects the fact that the low quality of MALBAC single cells from Upton et al . is due to high stochastic noise present in each individual single cell that was likely present prior to MALBAC amplification . The bottom right corner of each plot shows the total sequencing depth of the sample . Plots after subsampling to lower sequencing depth shows that total sequencing depth has only a small effect on low-resolution coverage plots and statistics . Therefore , differences in sequencing depth between samples do not alter the conclusions regarding the quality of single cells from each study . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 01010 . 7554/eLife . 12966 . 011Figure 4—figure supplement 2 . MDA and MALBAC genome coverage . ( A ) Fraction of genome covered at nucleotide resolution at ≥1x and ≥10x read depth at different subsampled read depths . Data for 1465 and SW480 samples from Evrony et al . ( 2015 ) and Zong et al . ( 2012 ) , respectively , is included for comparison and were previously shown in Evrony et al . ( 2015 ) . Reads were randomly subsampled to obtain different total subsampled read depths ( i . e . different subsampled total sequenced bases normalized to genome size ) , allowing comparison between samples regardless of their original total read depths . Upton et al . MALBAC hippocampus single neurons from normal individuals were pooled into one sample to achieve coverage necessary for the analysis . Plots show the average coverage across samples of each sample type ( shading ± SD ) . Plots for each sample set are shown up to the maximum read depth that was possible to subsample equivalently across all samples in a sample set . Note the slightly improved coverage at ≥10x read depth of pooled MALBAC single neurons from Upton et al . relative to MALBAC single cells from Zong et al . However , this is a comparison between a pooled sample , which cancels stochastic noise present in individual single cells , versus single cells analyzed individually that were not pooled and still contain stochastic noise present in each cell . This suggests that the MALBAC implementation in Upton et al . was commensurate or slightly better than that in Zong et al . , but individual cells in Upton et al . are still of significantly lower quality than individual cells in Zong et al . ( Figure 4B and Figure 4—figure supplement 1 ) and if sequenced to high coverage would very likely have significantly greater dropout at nucleotide resolution . ( B ) Lorenz curves as in Evrony et al . ( 2015 ) showing the cumulative fraction of reads as a function of cumulative fraction of the genome , averaged across samples of each sample type . These curves reflect the uniformity of genome coverage . A sample with perfectly even genome coverage would appear on the diagonal y = x line . The left panel is plotted using all sequencing reads of each sample . The right panel is plotted after subsampling to the same total read depth across all samples ( i . e . the total read depth of MALBAC single cell SRX202787 , which has the lowest total read depth of the graphed samples ) , showing the same trends as the left panel . Note the better performance of pooled MALBAC single neurons from Upton et al . compared to individual MALBAC single cells from Zong et al . This is a result of pooling Upton et al . single neurons to obtain a high coverage sample for analysis , which cancels stochastic noise present in individual neurons as described in panel A . Plots for individual 1465 and SW480 samples were calculated in Evrony et al . ( 2015 ) and are presented again for comparison to the Upton et al . sample . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 011 At larger genomic scales of ~500 kb bins ( low-resolution view ) , MALBAC single cells from Upton et al . show significantly lower quality and higher variability among individual cells than both MALBAC-amplified single cells from Zong et al . and MDA-amplified single neurons ( Figure 4B; Figure 4—figure supplement 1A–B ) . Pooling of all 92 normal hippocampus single neurons from Upton et al . shows performance commensurate with MALBAC single cells from Zong et al . ( Figure 4B; Figure 4—figure supplement 1A–B ) , indicating that the low quality of Upton et al . 's single cells may be due to stochastic factors preceding MALBAC , such as poor tissue quality , rather than MALBAC itself . Of note , at these genomic scales , MALBAC single cells from Zong et al . have high reproducibility and better uniformity of genome coverage than MDA ( Figure 4B; Figure 4—figure supplement 1A–B ) ( Evrony et al . , 2015 ) , enabling MALBAC's better performance in detection of large copy number variants ( Hou et al . , 2013 ) . Power spectral density measuring amplification uniformity across all genomic scales confirmed better uniformity of MALBAC at large genomic scales ( > 30 kb ) and better uniformity of MDA at small genomic scales ( < 30 kb ) ( Figure 4C ) ( Evrony et al . , 2015; Zhang et al . , 2015 ) . The above and additional analyses are discussed further in Appendix 2 . Altogether , these results: a ) suggest MALBAC and low quality single cells as contributors to single-cell RC-seq sensitivity loss; b ) emphasize the importance of single-cell quality control at genomic scales relevant to the studied mutation type; and c ) confirm that MALBAC and MDA each have advantages at different genomic scales and for different mutation types but that MALBAC is not especially well-suited for retrotransposon studies .
To justify not using a read count filter , Upton et al . state that “in single-cell RC-seq libraries , putative chimeras are disproportionately likely to amplify efficiently and accrue high read depth” ( Upton et al . , 2015 ) . In other words , they are suggesting that their method preferentially amplifies noise ( chimeric sequences ) instead of signal ( true insertions ) . We could find no precedent or chemical explanation for why PCR or next-generation sequencing would preferentially amplify chimeras , since there are no sequence features distinguishing chimeras from true insertions in small DNA fragments that would cause preferential overamplification of the former in single-cell RC-seq . In fact , prior single-neuron sequencing studies and chimera rates of Illumina libraries and MALBAC show directly that chimeras are not preferentially amplified relative to true genomic sequence fragments and true insertions ( Appendix 3; Figure 2—figure supplement 1B–C; Figure 3—figure supplement 1 ) . Indeed , the use of read counts for mutation analysis is integral to one of the prime purposes of single-cell sequencing , a technology whose development was motivated by two goals: ( a ) tracking which somatic mutations are present together in the same cells to enable lineage tracing; and ( b ) achieving higher signal to noise ratios for somatic mutations , i . e . true mutation to false-positive read count ratios . In single-cell sequencing , somatic mutations appear on average at the same signal level as germline heterozygous mutations ( i . e . 50% of reads at the locus ) , while the fraction of false variant reads at a locus ( e . g . sequencing errors , library PCR mutations , chimeras ) is the same on average regardless of the number of cells sequenced . Accordingly , decreasing the number of cells pooled for sequencing increases the signal to noise ratio of somatic mutations ( see Figure 5 for a simplified mathematical framework for single-cell sequencing ) . Therefore , calling mutations supported by only a single sequencing read is counter to a key feature and objective of single-cell sequencing . Furthermore , although Upton et al . present qPCR experiments as additional evidence for their findings , it is important that the originators of that qPCR method consider single-cell analysis as definitive ( Erwin et al . , 2014 ) , and qPCR results are affected by target L1 specificity ( Appendix 4 ) . 10 . 7554/eLife . 12966 . 012Figure 5 . A mathematical framework for single-cell sequencing . ( A ) In bulk sequencing , a somatic mutation present in k out of n cells pooled together for sequencing ( i . e . mosaicism of k/n ) , with read coverage D at the mutation locus , will be detected on average in k/n·D/2 reads with a variance depending on sampling error; i . e . the number of reads detecting the mutation correlates linearly with the percent mosaicism . In contrast , germline heterozygous and homozygous variants are present in D/2 and D reads , respectively . Due to sequencing artifacts and sequencing errors , a mutation must be detected above a threshold number of reads , T , which also depends on the sequencing depth , D , since errors occur at rates , e , that are a constant fraction on average of the total number of reads ( T=z·e·D; z is a constant chosen based on desired detection sensitivity and specificity ) . The fraction of error reads , e , is a constant on average that is independent of total sequencing depth , D , because library artifacts and sequencing errors occur at rates that are independent of total sequencing depth . The threshold , T , can be reduced with methods reducing sequencing error , but errors are still present in any current sequencing technology . Combining equations simplifies to k/n ≥ 2·z·e . This means that the mosaicism of a somatic mutation must be at least twice the sequencing error rate ( or more , depending on the confidence factor ) for detection to be possible in bulk DNA sequencing , regardless of sequencing depth . Below a certain level of mosaicism that depends on the sequencing error rate , detection is unlikely . Note: for simplicity , the height of the histograms ( # of mutations ) is scaled to the same height , and the equations do not include variance terms . ( B ) In single-cell sequencing , somatic mutations are present at the same signal level on average as germline heterozygous variants ( i . e . D/2 , since k/n = 1 ) , enabling detection of low mosaicism mutations that would otherwise be below detection thresholds of bulk sequencing due to sequencing error . Due to whole genome amplification , single-cell sequencing also leads to greater variance in mutation and error signal level distributions ( non-uniform amplification and dropout ) and entails additional artifacts not present in bulk sequencing , which increases the noise level , e' , but still a lower level on average than true heterozygous mutations . However , the signal distribution of artifacts may still overlap that of true mutations , necessitating careful bioinformatics and modeling of error and true mutation signals along with rigorous validation . Note , for simplicity , the equations here do not include variance terms and bioinformatic modeling usually includes additional parameters other than read count illustrated here . Single-cell sequencing does not achieve increased sensitivity for somatic mutations without cost , because to detect a given mutation with k/n mosaicism , more than n/k single cells may need to be sequenced . The benefit of single-cell sequencing is not to reduce sequencing costs , but rather its ability to overcome limitations due to sequencing error rates on the minimum mosaicism detectable and maintaining information as to which somatic mutations are found within the same cell , which enables lineage tracing . DOI: http://dx . doi . org/10 . 7554/eLife . 12966 . 012 Finally , we emphasize that the bioinformatic and validation approach led to the inflated somatic insertion rate , but not the RC-seq L1 hybridization capture method itself . Our analysis suggests that RC-seq capture , if used with an appropriate single-cell amplification method , careful signal modeling based on true insertions , and rigorous PCR validation , would likely enable cost-effective , high-throughput retrotransposon profiling comparing favorably with other methods such as L1-IP . The corrected RC-seq retrotransposition rate is significant as it aligns to a wholly different regime of potential functional roles for retrotransposition in the brain ( rare normal variation and rare disease ) rather than a "ubiquitous" role . This corrected rate is consistent with rates measured in vitro in neuronal progenitors ( Coufal et al . , 2009 ) and is consistent with the absence of significant somatic L1 insertions in brain tumors ( Helman et al . , 2014; Iskow et al . , 2010; Lee et al . , 2012 ) . These rates do not rule out that there may be rare individuals in whom a somatic L1 insertion affects a gene in enough cells to cause a sub-clinical or overt phenotype , or that elevated L1 rates may occur in particular individuals or disease states . Future single-neuron genomic studies will resolve the rates and mosaicism frequencies of all classes of somatic mutation across the diversity of cell types , regions , and developmental timepoints in the brain . Single-cell genomic analysis has enabled the first systematic measurement of somatic mutation rates in the body but entails additional challenges spanning molecular biology to bioinformatics . Our findings suggest the following elements may aid future single-cell genomics studies: a ) choosing a single-cell amplification method suitable for the studied mutation type; b ) objective metrics evaluating genome amplification coverage , uniformity , dropout , and chimera rates at spatial scales and genomic elements relevant to the mutation type; c ) use of gold-standard germline mutations and chimera rates to build a signal model for calling mutations; and d ) stringent validation experiments . Retrotransposons offer unique advantages as a starting point for developing single-cell genomics methods due to their characteristic sequence signatures allowing definitive validation even when present in only one cell . The lessons learned from the study of somatic retrotransposition are therefore broadly applicable for the nascent field of single-cell genomics .
Sequencing data of single-cell whole-genome sequencing ( WGS ) experiments from Upton et al . were obtained from the European Nucleotide Archive with accession PRJEB5239 . Single-cell RC-seq somatic candidate data ( including sequences ) and bulk RC-seq KNR insertion data were obtained from Upton et al . supplemental tables ( 2015 ) . Single-cell RC-seq KNR ( germline polymorphic ) insertion data ( read counts and junction detection rates ) and bulk RC-seq somatic L1 candidate data were provided by Geoffrey Faulkner upon request . Sequencing data of single cells from Evrony et al . ( 2015 ) and Zong et al . ( 2012 ) used in MDA and MALBAC performance analyses were obtained from those studies as described in Evrony et al . ( 2015 ) . High-coverage bulk DNA sequencing of individual N12877 shown in power spectral analysis ( Figure 4C ) was obtained from the NCBI Sequence Read Archive with accession ERX069504 . RC-seq insertion candidate sequences were analyzed with the aid of standard tools , including the UCSC genome browser ( Kent et al . , 2002 ) , Blat ( Kent , 2002 ) , NCBI BLAST ( http://blast . ncbi . nlm . nih . gov/Blast . cgi ) , RepeatMasker ( Smit et al . , 2010 ) , RepBase ( Jurka et al . , 2005 ) , ClustalW2 ( Larkin et al . , 2007 ) , and L1Xplorer ( Penzkofer et al . , 2005 ) . MALBAC and Illumina sequencing adaptors were trimmed from sequencing reads of Upton et al . MALBAC single-cell WGS samples using the 'phacro' tookit ( http://sourceforge . net/projects/phacro ) ( Hou et al . , 2013 ) with default settings and the MALBAC adaptor: GTGAGTGATGGTTGAGGTCTTGTGGAG . The phacro toolkit was created by the team that developed MALBAC specifically for trimming MALBAC adaptors from MALBAC samples , including the 8bp degenerate 'N' sequence following the adaptor . After adaptor trimming , Upton et al . WGS data was aligned to the hs37d5 human genome reference ( 1000 Genomes Project human genome reference based on the GRCh37 primary assembly ) with bwa ( Li and Durbin , 2009 ) as in Evrony et al . ( 2015 ) . PCR duplicates were removed as in Evrony et al . ( 2015 ) . High resolution genome coverage plots ( Figure 4A ) , low resolution ( ~500 kb bin ) genome coverage plots ( Figure 4B and Figure 4—figure supplement 1 ) , power spectral density analysis ( Figure 4C ) , subsampling genome coverage analysis ( Figure 4—figure supplement 2A ) , and Lorenz curves ( Figure 4—figure supplement 2A ) were calculated and plotted as in Evrony et al . ( 2015 ) ( results summarized in Appendix 2 ) . Plots for samples from individual 1465 and SW480 MALBAC samples in power spectral density analysis ( Figure 4C ) , subsampling analyses ( Figure 4—figure supplement 2A ) , and Lorenz curves ( Figure 4—figure supplement 2B ) were already calculated in Evrony et al . ( 2015 ) and are presented again in this paper to allow comparison to Upton et al . single-cell samples . Median absolute pairwise deviation ( MAPD ) and median absolute deviation from the median ( MDAD ) scores of single-cell quality were calculated in ~500 kb equal-read bins as in Evrony et al . ( 2015 ) . High-resolution genome coverage plots ( Figure 4A ) , power spectral density analysis ( Figure 4C ) , subsampling genome coverage analysis ( Figure 4—figure supplement 2A ) , and Lorenz curves ( Figure 4—figure supplement 2B ) were calculated after pooling all single neurons from normal individual hippocampi ( n = 92 cells ) to create a high-coverage dataset ( 48x ) , since the WGS sequencing depth of individual cells in Upton et al . are not sufficient for high-resolution analyses . A high-coverage ( 5x ) pooled sample of all single neurons from normal individual cerebral cortex ( n = 35 cells ) was also created for the high-resolution genome coverage plot ( Figure 4A ) and power spectral density analysis ( Figure 4C ) . Low-resolution genome coverage plots and analyses ( Figures 4B and Figure 4—figure supplement 1 ) were performed for individual hippocampus single neurons and also separately for the pooled hippocampus single-neuron sample . Low-resolution genome coverage plots of Upton et al . single cells used the pooled cerebral cortex single-neuron sample as a copy number reference . Note that pooling to achieve higher coverage datasets would only improve genome coverage statistics since as samples are pooled , stochastic noise present in individual cells cancels out , leaving systematic noise due to MALBAC and providing a view of MALBAC amplification performance . Low-resolution genome coverage plots and analyses of MDA single neurons ( Evrony et al . , 2015 ) , SW480 MALBAC single cells ( Zong et al . , 2012 ) , and the pooled hippocampus MALBAC single neuron sample ( Upton et al . , 2015 ) were also calculated after subsampling these high coverage samples to lower read depths ( Figures 4B and Figure 4—figure supplement 1 ) , confirming as in Evrony et al . ( 2015 ) that low-resolution genome coverage plots and statistics are minimally affected by increasing read depth > 0 . 1x . Therefore the results and conclusions of low-resolution genome coverage analyses are not due to lower sequencing depth for Upton et al . single cells relative to MALBAC and MDA samples from other studies , as the conclusions are the same after subsampling MALBAC and MDA samples from other studies to lower read depth than Upton et al . single cells . Chromosome X bins in low-resolution genome coverage plots of single cells from individual CTRL-36 ( female ) and the pooled hippocampus single-neuron sample ( which includes CTRL-36 female neurons ) ( Figure 4—figure supplement 1A ) were corrected in each sample to the median of all bins in chromosome X of the sample , since the pooled cortex single neurons used as a copy number reference derived from male samples so chromosome X bins would have inflated copy number without correction . Chromosome Y bins of each CTRL-36 ( female ) hippocampus single neuron were set to a log2 relative copy number of 0 so that they do not affect genome coverage statistics , since CTRL-36 female neurons do not have a Y chromosome and complete dropout of Y-chromosome bins would skew ( i . e . make worse ) genome coverage statistics . Chromosome Y bins of the pooled hippocampus single-neuron sample were also normalized to the median of chromosome Y bins in the sample , since this pooled sample includes CTRL-36 female neurons that do not have a Y chromosome . Discordant and clipped read statistics for Upton et al . single-cell WGS samples ( Appendix 3 ) were calculated as in Evrony et al . ( 2015 ) . Discordant and clipped read statistics for MALBAC single-cell samples from Zong et al . ( 2012 ) and MDA single-cell samples from Evrony et al . ( 2015 ) were already calculated in Evrony et al . ( 2015 ) . Read count histograms of somatic insertion candidates and germline known non-reference ( KNR ) insertions ( Figure 2A and Figure 2—figure supplement 1 ) , which are insertions detected in prior L1 profiling studies that are absent from the human genome reference , were constructed as described below for each L1 profiling method . Upton et al . acknowledge the importance of KNR gold-standard insertions by using them to estimate the sensitivity of their method , but they did not present the distribution of KNR insertion read counts in single cells , which is essential data for calling somatic insertion candidates and evaluating candidate veracity . Read count histograms plot the per sample read counts of candidates and insertions , not their total read count across all samples , which controls for the number of samples profiled per individual and for candidates/insertions present in multiple samples ( necessary for comparing germline KNR insertions that are present in many samples to somatic candidates ) . RC-seq KNR read count histograms ( Figure 2A and Figure 2—figure supplement 1A ) : Single-cell RC-seq KNR read counts were obtained from data provided by Geoffrey Faulkner upon request . Bulk RC-seq KNR read counts were obtained from the 'Polymorphic L1' sheet of Table S2 in Upton et al . ( 2015 ) . The gold-standard set of germline KNR insertions plotted for single cells in Figure 2A and Figure 2—figure supplement 1A consists of insertions identified in prior non RC-seq L1 profiling studies ( i . e . insertions with a prior study annotated in the 'Database ? ' column of Upton et al . tables ) that were detected with ≥ 40 reads in both bulk samples of the individual ( considering detection only in bulk samples corresponding to the individual from whom the single cell derived ) . Insertions that were detected only in a prior RC-seq study ( "Published RC-seq ? ' column ) but not in a prior non RC-seq study ( empty 'Database ? ' column ) were not included in Figure 2A and Figure 2—figure supplement 1A since it is preferable to define a gold standard set of true mutations detected by independent methods . Nevertheless , read count histograms that also include KNR insertions that were identified only in prior RC-seq studies produced nearly identical histograms ( data not shown ) . Therefore , whether or not KNR insertions found only in prior RC-seq studies are included has negligible effect . Bulk KNR insertion read count histograms in Figure 2A and Figure 2—figure supplement 1A show KNR insertions detected at any read count ( i . e . ≥ 1 read ) , since there is no independent gold-standard reference as to which KNR insertions are present in bulk samples of the profiled individuals , and using a ≥ 40 read cutoff would mask the underlying read count distribution by showing only insertions appearing at high read counts . In any case , the key comparison for evaluating RC-seq somatic candidate veracity is between single-cell KNR insertions and single-cell somatic candidates , not between single-cell KNR insertions and bulk KNR insertions . The latter comparison is useful for assessing the quality of single cells versus bulk samples and the effect of MALBAC amplification . Note that germline KNR insertion dropouts in single cells ( read counts of 0 for germline KNR insertions in single cells of an individual known to have the KNR insertion based on bulk samples ) are not included in the read count histograms since single-cell dropout rates affect both KNR insertions and somatic insertions . While for KNR insertions the true state ( presence/absence ) in each cell is known , the true state is unknown for somatic insertions . Therefore , in order to compare germline KNR insertion and somatic candidate read count distributions , KNR dropout calls must be excluded . Also , note that the read count distribution of gold-standard KNR insertions in single-cell RC-seq is bimodal ( Figure 2A ) , with a population of high read count calls and a population of low read count calls . Although KNR insertions appear at lower read depth in single cell RC-seq relative to bulk RC-seq samples and show a bimodal distribution with ~1/3 of calls detected by only one read ( Figure 2A ) , this does not affect the conclusion that the vast majority of single-cell RC-seq somatic insertion candidates are false-positives: only 20 of the 4759 somatic candidates were detected with > 2 reads across all 170 single cells and half of true somatic insertions are expected to be detected at this level based on KNR insertion read counts . However , it does predict that ~1/3 of true somatic insertions would be detected with 1 read . This bimodal distribution of KNR read counts in single-cell RC-seq is due to , both: a ) high variability ( non-uniformity ) in single-cell MALBAC genome amplification at the length scale of L1 insertions ( data not shown; and see Evrony et al . ( 2015 ) : Note S1 , 'Coverage variability analyses' section , Figure S6 , and Figure S7 , as well as Zhang et al . ( 2015 ) for details of non-uniformity at small length scales < 30 kb inherent to MALBAC ) ; and b ) allelic dropout stemming from low-quality of Upton et al . single neurons . The MAPD ( median absolute pairwise deviation ) metric reflects uniformity of genome coverage at large genomic scales ( ~500 kb bins ) , with lower MAPD scores indicating better uniformity . Upton et al . single neurons have mean MAPD scores of 0 . 53 ± 0 . 16 ( SD ) , compared to MAPD 0 . 18 ± 0 . 06 for MALBAC-amplified single cells from Zong et al . ( 2012 ) and MAPD 0 . 33 ± 0 . 06 for MDA-amplified single neurons from Evrony et al . ( 2015 ) . Furthermore , the lower overall read counts of KNR insertions in single cells relative to bulk samples is also partly due to ~3fold lower total reads per sample on average for single-cell samples versus bulk samples . This highlights a further issue when read count filters are not used , in that there is no normalization for different total reads per sample . Somatic insertions are present in a single copy in the genome ( i . e . heterozygous or hemizygous ) in cells harboring the mutation . Most germline KNR insertions ( ~75% ) are present in a single copy per cell as well , since most are in the heterozygous state in individuals of the population . This supports the use of KNR insertions as a reference for the expected read count distribution ( and signal distribution of other parameters ) of somatic insertions . The evidence that most KNR insertions that are present in an individual are heterozygous is based on measured allele frequencies and genotypes of KNR insertions in prior population studies of L1 polymorphism: a ) In the 1000 Genomes project studying mobile element polymorphism ( Stewart et al . , 2011 ) , genotyping of a large number of L1 KNR insertions ( see Table S4 in that study ) found an average heterozygosity of 0 . 85 in individuals harboring the insertions ( i . e . number of individuals heterozygous/ ( number heterozygous + number homozygous ) for each KNR insertion , averaged across all KNR insertions ) . The average allele frequency of these insertions was 0 . 26; b ) In Iskow et al . ( 2010 ) , the average allele frequencies of KNR insertions found by dideoxy sequencing was 0 . 22 ( table S1 in that study ) and < 0 . 2 for insertions found by 454 sequencing ( Figure 2F in that study ) , corresponding to a heterozygosity rate for KNR insertions of at least 0 . 88 in individuals harboring each insertion ( i . e . , allele frequency p = 0 . 22; heterozygosity in individuals with the insertion = 2pq/ ( p2+2pq ) ) assuming insertions are in Hardy-Weinberg equilibrium . Prior studies have shown L1 insertion genotypes are almost always consistent with Hardy-Weinberg equilibrium ( Badge et al . , 2003; Myers et al . , 2002; Seleme et al . , 2006 ) ; c ) Ewing and Kazazian ( 2011 ) also analyzed the 1000 genomes data and found a KNR insertion allele frequency < 0 . 2 ( Figure 1B in that study ) , corresponding to an average heterozygosity >0 . 89 for KNR insertions present in an individual; d ) Huang et al . ( 2010 ) estimate an allele frequency of chromosome X KNR insertions of 0 . 58 and an allele frequency of 0 . 38 for a set of KNR insertions identified by whole-genome profiling , corresponding to an average heterozygosity of 0 . 59 and 0 . 77 , respectively in individuals harboring the insertions; d ) 75% ( 105/140 ) of the KNR insertions detected in individual 1465 in Evrony et al . ( 2015 ) ( gold-standard KNR insertions detected in both bulk samples of the individual ) are heterozygous or hemizygous ( Evrony et al . , 2015 ) ; e ) in dbRIP ( Wang et al . , 2006 ) , the average heterozygosity of polymorphic insertions among individuals with the insertion is 0 . 46 ( with an average allele frequency of 0 . 59 ) . This shows that most KNR insertions in an individual are heterozygous and present in a single copy per cell . We also plotted the RC-seq read count histograms of a pure set of single-copy KNR insertions– those found on chromosome X in male samples– and found a similar distribution of read counts as the full KNR set , with most insertions still detected by multiple reads in single cells: 65% , 12% , and 11% were detected with ≥3 , ≥20 and ≥40 reads per sample ( Figure 2—figure supplement 1A ) . RC-seq somatic candidate read count histogram ( Figure 2A ) : Single-cell RC-seq somatic candidate read counts were obtained from the 'Somatic L1' sheet of Table S2 in Upton et al . ( 2015 ) . Bulk RC-seq somatic candidate read counts were provided by Geoffrey Faulkner upon request . WGS KNR read count histogram ( Figure 2—figure supplement 1B ) : The gold-standard KNR insertion set for the WGS read count histogram is defined as insertions detected in both bulk samples ( cortex and heart ) of the individual with the following parameters ( see Evrony et al . ( 2015 ) for details of parameters ) : a ) ≥ 2 RAM reads on each side of the breakpoint; b ) ≥ 4 clipped reads supporting the insertion call; c ) estimated target-site duplication or deletion ≤ 50 bp in size in the absence of a poly-A tail , or ≤ 250 bp in size if a poly-A tail was detected; d ) at least half of clipped reads at the insertion site aligned to ± 2 bp of the insertion breakpoint; e ) the insertion was detected in prior independent L1 profiling studies from other groups ( see Evrony et al . ( 2015 ) for list of prior L1 profiling studies used ) . L1-IP KNR read count histograms ( Figure 2—figure supplement 1C ) : The gold-standard KNR insertion set for the L1-IP read count histograms was defined as insertions detected with a confidence score ≥ 0 . 5 in at least half of the bulk samples of the individual and detected in prior independent L1 profiling studies of other groups ( see Evrony et al . ( 2012 ) for list of prior L1 profiling studies used ) . The percentage of RC-seq insertions and candidates detected at only the 5' , only the 3' , or both 5' and 3' L1 junctions ( Figure 2B ) were obtained as follows: Germline KNR junction detection data for bulk and single-cell RC-seq samples were provided by Geoffrey Faulkner; these data annotated for each individual sample and each KNR insertion which L1 junctions were detected ( 5' , 3' , or both ) . Junction detection rates of both bulk and single-cell germline KNR insertions shown in Figure 2B are for the same high-confidence KNR insertion set defined for the single-cell KNR read count histogram in Figure 2A ( see 'Read count histograms' in the prior section of the 'Materials and methods' ) . The numerator and denominator units of bulk and single-cell RC-seq KNR junction detection rates are KNR insertion calls , not KNR insertions; i . e . for a hypothetical KNR insertion detected in samples A , B , and C , each of these 3 calls is counted separately because the detection of a KNR insertion in each sample is independent of other samples . Single-cell RC-seq somatic candidate junction detection data were obtained from the 'Somatic L1' sheet of Table S2 in Upton et al . ( 2015 ) . 5'-only detected candidates are those with a negative alignment in the 'Sense L1' column but no antisense read or a negative alignment in the 'Antisense L1' column but no sense read . 3'-only detected candidates are those with a positive alignment in the 'Sense L1' column but no antisense read or a positive alignment in the 'Antisense L1' column but no sense read . Candidates detected at both 5' and 3' junctions are those with both sense and antisense reads . Note that the 'single-cell somatic candidate' junction data available in Table S2 of Upton et al . annotates junction detection per candidate ( regardless of the number of cells in which the candidate was detected ) , in contrast to the 'single-cell KNR insertion' junction data that annotates junction detection for each individual sample in which the insertion was detected . Since 'single-cell somatic candidate' junction detection data is only available annotated per candidate rather than per cell , somatic candidates detected in multiple cells may skew the true junction detection rates and preclude comparison to 'single-cell KNR insertion' rates . Therefore , to allow comparison between 'single-cell somatic candidate' and 'single-cell KNR insertion' junction detection rates , the 'single-cell somatic candidate' junction detection rates in Figure 2B are for candidates detected in only one cell and excludes those detected in multiple cells . Nevertheless , even when including candidates found in more than one cell ( i . e . considering both junctions as detected even when each was detected in a different single cell ) , only 0 . 4% ( 21/4728 ) of single-cell somatic candidates were detected at both junctions– still >25-fold less than the rate for single-cell KNR insertions ( 11% ) and similar to the single-cell somatic candidate rate of 0 . 04% ( 2/4682 ) calculated when excluding candidates found in multiple cells . Briefly , somatic insertion rates were calculated by first counting the number of somatic candidates detected with > 2 reads . Sequences of candidates were then manually examined and definite chimeras were excluded ( Supplementary file 1 ) . In each cell , the remaining number of candidates was adjusted for that cell's sensitivity for gold-standard KNR insertions . Insertion rates per cell type ( Figure 3B ) are an average of the rates across all single cells of that type . Below is a full explanation of the insertion rate calculations: RC-seq somatic retrotransposon insertion rates were calculated using RC-seq read counts of the gold-standard germline KNR insertion set to guide read count filtering . A read count threshold was chosen that would optimize the number of true ( germline KNR and somatic ) insertions above the threshold ( sensitivity ) while minimizing the number of false-positive calls ( specificity ) . Sensitivity for true insertions at any given read count threshold was estimated per single cell using the single-cell RC-seq KNR insertion read count data provided by Geoffrey Faulkner . Sensitivity was calculated as the fraction of high-confidence germline KNR insertions present in the individual ( i . e . insertions detected with ≥ 40 reads in both bulk samples of the individual , and identified in prior non RC-seq L1 profiling studies with a prior study annotated in the 'Database ? ' column of Upton et al . tables ) , that were detected in the single cell above the read count threshold . Specificity at any given read count was estimated using the read count distribution of all single-cell somatic candidate calls ( 'Somatic L1' sheet of Table S2 in Upton et al . ( 2015 ) ) since nearly all are false-positives . As discussed in the main text and in the following paragraph , the latter assumption is valid because of the discrepancy between the read count distributions of KNR insertions versus somatic candidates . As discussed above in the 'Read count histograms' section , the single-cell RC-seq KNR insertion read count distribution is bimodal due to non-uniformity of MALBAC amplification , with high and low read-count sub-populations ( Figure 2A ) . Finite mixture modeling can estimate the proportion of the read count distribution that belongs to each sub-population . Finite mixture modeling estimates the high and low read count sub-populations comprise 1/3 and 2/3 of the single-cell KNR insertion distribution , respectively . In contrast , the read count distribution of single-cell somatic candidates is unimodal , concentrated at low signal with nearly all ( 99 . 6% ) candidates having ≤ 2 reads ( Figure 2A ) . Intuitively , the absence of a high-signal component in the somatic candidate read count distribution indicates nearly all somatic candidates are false-positives . Therefore , the somatic candidate read count distribution can be treated essentially as a false-positive distribution for the purposes of deciding on an optimal read count threshold . More formally , any single-cell somatic candidate distribution is a mixture of two subpopulations: false-positive candidates ( e . g . chimeras ) and true somatic insertions . A finite mixture model can estimate the proportion of somatic candidates that derives from a true somatic insertion subpopulation , using a model of the high read-count component of the true-positive KNR insertion distribution as a guide . This analysis estimates a negligible fraction ( < 0 . 5% ) of single-cell somatic candidates are true somatic insertions . Consequently , we can consider the read count distributions of KNR insertions and somatic candidates as reflecting true and false-positives , respectively . This then allows calculation of estimated sensitivity loss and specificity gain at increasing read count thresholds . Increasing the read count threshold from >0 to >1 read reduces the per-cell sensitivity for true ( KNR ) insertion calls from an average of 45% to 31% ( a 32% reduction ) while reducing the estimated number of false-positive calls by ~97% . Further increasing the threshold to > 2 reads reduces the sensitivity for true insertion calls to 24% ( a further 23% reduction ) and reduces false-positive calls by an estimated additional ~84% relative to the > 1 read threshold– still a large improvement in specificity with a relatively modest reduction in sensitivity . Increasing the read count threshold further to > 3 reads leads to diminishing returns in terms of improved specificity– 18% reduction in sensitivity with 35% reduction in false-positive calls relative to the > 2 read threshold– reflecting the fact that nearly all somatic candidate ( mostly false-positive ) calls are at read counts of 1 and 2 . Therefore a read count threshold of > 2 reads was chosen , which maintains detection of 53% of KNR insertion calls across all single cells and a per-cell KNR detection sensitivity of 24% , while excluding an estimated 99 . 6% of false-positive calls . Once the > 2 reads count threshold was chosen , for each single cell the number of somatic insertion candidates detected with > 2 reads was counted . Candidate sequences were then manually examined and candidates that were definite chimeras were excluded ( see 'RC-seq | Somatic L1 > 2 reads' sheet in Supplementary file 1 for sequence analyses of all candidates ) . For each cell , the remaining number of somatic candidates in the cell was then corrected for the sensitivity for gold-standard KNR insertions achieved in that same cell , i . e . dividing the number of somatic candidates by the fraction of KNR insertions detected in the cell above the chosen threshold , using the gold-standard KNR insertion reference of the individual from whom the single-cell derived , as described above . This final number was the estimated pre-PCR validation somatic insertion rate for the cell , since Upton et al . did not attempt PCR validation for these somatic candidates . Insertion rates per cell type ( Figure 3B ) are an average of the rates across all single cells of that type . Further justification for the > 2 reads threshold is shown by estimates of the pre-PCR validation somatic insertion rate at read thresholds of >1 , > 3 , and > 4 reads . At a read threshold of >1 read , the estimated pre-PCR validation rate across all cells is 2 . 4 ± 3 . 3 ( SD ) per cell prior to manual examination of candidates for chimeras . Adjusting for the chimera rate of 12/20 seen at the > 2 read threshold ( since the chimera rate at a >1 read threshold could only be greater ) , gives a rate of 1 . 1 ± 1 . 4 insertions per cell . At read thresholds of > 3 and > 4 reads , the estimated pre-PCR validation rates across all cells are 0 . 38 ± 1 . 35 and 0 . 44 ± 1 . 69 ( SD ) , respectively , per cell prior to manual examination of candidates for chimeras . Adjusting for chimera rates of 8/13 and 7/12 , respectively , seen in manual examination of the candidates ( Supplementary file 1 ) yields pre-PCR validation rate estimates of 0 . 15 ± 0 . 52 and 0 . 18 ± 0 . 70 , respectively , for the > 3 and > 4 read thresholds . These are similar to the estimate of 0 . 19 ± 0 . 97 at a > 2 read threshold . In summary , the pre-PCR validation rates across all single cells at >1 , > 2 , > 3 , and > 4 read thresholds ( after excluding chimeras ) are 1 . 1 , 0 . 19 , 0 . 15 , and 0 . 18 , respectively . This shows stability of the rate estimate at thresholds of > 2 reads or more , a result of the fact that the vast majority of chimeras appear with 1 or 2 reads , while most true insertions appear at higher read counts . The stability of the rate estimate above thresholds of > 2 reads supports the use of the > 2 read threshold , which optimizes sensitivity and specificity . In contrast , the somatic rate calculated at the > 1 read threshold is higher than the rates calculated at > 2 , > 3 , and > 4 reads and a significant overestimate of the true rate for two reasons: a ) the > 1 read threshold begins to overlap the false-positive chimera distribution , so most candidates at the > 1 read threshold are chimeras . This is confirmed by the read count histogram analyses of KNR insertions and somatic candidates discussed above– namely that there is no discernible population of high read count candidates in the read count distribution of somatic candidates as there is in the KNR insertion read count distribution ( Figure 2A ) , so the population of somatic candidates at read counts of 1 and 2 are nearly all false-positives; b ) this is a pre-PCR validation rate . The somatic insertion rate estimate obtained at a > 1 read threshold is therefore an overestimate that would be confirmed as such after proper PCR validation , while the rate obtained at a > 2 read threshold is a more accurate pre-PCR validation rate estimate . The somatic L1 retrotransposition rate for single neurons from Evrony et al . ( 2015 ) was calculated for comparison to the RC-seq rate . 16 single neurons were sequenced in Evrony et al . ( 2015 ) , but the rate was estimated from the 14 single neurons that were selected randomly for sequencing . The 2 remaining cells in which L1 #1 was detected ( neurons 2 and 77 ) ( Evrony et al . , 2015 ) were excluded from the rate estimate , because they were a priori chosen for whole-genome sequencing as positive controls known to harbor somatic L1 insertions previously detected by the L1-IP method in Evrony et al . ( 2012 ) . Therefore , the calculated rate reflects the 2 of the 14 single neurons ( neurons 6 and 18 ) that harbor the same L1 #2 clonal insertion ( Evrony et al . , 2015 ) . The L1 somatic insertion rate estimate of each neuron was corrected for the neuron's sensitivity for KNR insertions . 3' junction PCR validation of 48 L1-IP candidates with low read counts ( Supplementary file 1 , sheet 'L1-IP | low-read-count' ) was performed as described in Evrony et al . ( 2012 ) . The L1-IP computational pipeline was rerun on raw data from Evrony et al . ( 2012 ) after removing any read count filter . 24 candidates were randomly selected from all candidates detected by only 1 read , and another 24 candidates were randomly selected from all candidates . | The human brain harbors perhaps the most diverse collection of cells among any organ in the body , consisting of neurons and other cells with many different shapes and behaviors . The mechanisms that create this diversity have been a long-standing area of investigation . While neurons can become more diverse through changes in the activity of genes during development and in response to experiences , it has been speculated that genetic differences among neurons may also play a role . The complete set of genes found in an individual is known as its genome . It is often assumed that each cell in an individual's brain has an identical genome . However , mutations accumulate in cells during the lifetime of an individual such that every brain cell may in fact contain a unique set of genetic mutations . The extent and types of such genetic mutations have only recently become accessible using techniques that can examine the genomes of single-cells . Some of these genetic differences may result from the activity of short sections of DNA called retrotransposons , which can copy themselves and move to a different place in the genome . This can introduce genetic mutations that alter how the cell works . Multiple studies have shown that retrotransposon-related mutations are present in human brain cells . Indeed , in 2015 a group of researchers suggested that every neuron in two brain regions called the cortex and the hippocampus contains as many as 16 retrotransposon-related mutations on average , which suggests that retrotransposons may play an essential role in the healthy brain . However , these findings contrasted with previous studies that had shown much fewer mutations . Now , Evrony , Lee et al . have analyzed the data from the 2015 study that led the previous researchers to interpret some artifacts as retrotransposon mutations . Reanalysing the data confirmed that these mutations do indeed occur; however , they are around 50 times less common than had been suggested by the earlier study . This suggests that retrotranspons are more likely to be occasional sources of rare variation or disease , rather than essential contributors to normal brain activity in humans . Further work is needed to examine the rate of these and other types of mutations in different cell types and brain regions , and at different developmental stages . However , to ensure that these studies are robust and reliable , Evrony , Lee et al . also outline a framework to aid the design and analysis of future studies . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Resolving rates of mutation in the brain using single-neuron genomics |
How the very first step in nucleosome assembly , deposition of histone H3-H4 as tetramers or dimers on DNA , is accomplished remains largely unclear . Here , we report that yeast chromatin assembly factor 1 ( CAF1 ) , a conserved histone chaperone complex that deposits H3-H4 during DNA replication , binds a single H3-H4 heterodimer in solution . We identify a new DNA-binding domain in the large Cac1 subunit of CAF1 , which is required for high-affinity DNA binding by the CAF1 three-subunit complex , and which is distinct from the previously described C-terminal winged-helix domain . CAF1 binds preferentially to DNA molecules longer than 40 bp , and two CAF1-H3-H4 complexes concertedly associate with DNA molecules of this size , resulting in deposition of H3-H4 tetramers . While DNA binding is not essential for H3–H4 tetrasome deposition in vitro , it is required for efficient DNA synthesis-coupled nucleosome assembly . Mutant histones with impaired H3-H4 tetramerization interactions fail to release from CAF1 , indicating that DNA deposition of H3-H4 tetramers by CAF1 requires a hierarchical cooperation between DNA binding , H3-H4 deposition and histone tetramerization .
Nucleosomes in eukaryotic cells enable packaging of the DNA within the cell nucleus and provide an important layer in genome regulation . They are composed of an octameric core of histones , around which 147 bp of DNA are wrapped ( Luger et al . , 1997 ) . The majority of nucleosomes in chromatin contain two copies of each of the four major histones H2A , H2B , H3 and H4 that are assembled in a step-wise manner ( Smith and Stillman , 1991 ) . Following the initial H3-H4 tetramer , two H2A-H2B dimers are deposited to complete the octameric core particle . Nucleosome assembly is promoted via the action of histone chaperones , ( De Koning et al . , 2007; Gurard-Levin et al . , 2014 ) . DNA replication requires doubling of the amount of chromatin which is accomplished through recycling of parental and incorporation of newly synthesized histones ( Groth , 2009 ) . It has been suggested that the parental histone modifications are propagated to newly incorporated nucleosomes upon cell division ( reviewed in Margueron and Reinberg , 2010; Probst et al . , 2009 ) , however whether a purely histone-based inheritance mechanism exists remains a matter of debate ( Ptashne , 2013 ) , and the mechanism of how histone modifications are maintained following DNA replication remains unclear . Cumulative evidence supports the model that upon replication , parental H3-H4 are conservatively propagated as tetramers in proliferating cultured cells ( Jackson , 1988; Prior et al . , 1980 ) . Newly deposited H3-H4 tetramers are assembled entirely of new histones thus precluding the acquisition of new histone modifications based on preexisting parental modifications within the same nucleosome ( Katan-Khaykovich and Struhl , 2011; Xu et al . , 2010 ) . Thus , one mechanism to ensure the maintenance of epigenetic modification could involve a read-write mechanism in which parental histone modifications are copied over to newly incorporated nucleosomes ( Probst et al . , 2009; Ragunathan et al . , 2015 ) . A central , unanswered question is how parental H3-H4 tetramers are propagated and how new H3-H4 tetramers are assembled behind the replication fork . The conservative nature of tetramer propagation and de novo assembly could be due to the biochemical properties of the histone chaperone machinery that operates at the replication fork . A number of histone chaperones including antisilencing function 1 ( Asf1 ) and DAXX bind H3-H4 dimers but not tetramers , indicating that chaperon-mediated tetramer assembly requires a two-step mechanism ( Elsässer et al . , 2012; English et al . , 2006; Liu et al . , 2012a; Natsume et al . , 2007 ) . Asf1 associates with H3-H4 and Mcm2 , a subunit of the replicative helicase ( Groth et al . , 2007 ) . A crystal structure of the complex and binding studies reveal a 1:1:1:1 stoichiometry indicating that dimeric H3-H4 is propagated at replication forks ( Huang et al . , 2015; Richet et al . , 2015 ) . Thus , a current model suggests that the Mcm2-Asf1 complex mediates the passing of parental H3-H4 through transient tetramer disruption and conservative reassembly onto DNA behind the replication fork ( Clément and Almouzni , 2015; Huang et al . , 2015; Richet et al . , 2015 ) . This model underlines the importance of H3-H4 dimers as intermediates even for the recycling of parental histones and raises the question of how such dimers are reassembled into tetramers behind the replication fork . Chromatin assembly factor 1 ( CAF1 ) is a histone chaperone complex that deposits new H3-H4 de novo in a DNA-synthesis-dependent manner ( Smith and Stillman , 1989 ) and is functionally conserved throughout eukaryotes . CAF1 contains three subunits: p150 , p60 and p48 ( Cac1 , Cac2 and Cac3 in Saccharomyces cerevisiae ) ( Kaufman et al . , 1997; Verreault et al . , 1996 ) . CAF1 is recruited to the replication fork by interaction of the p150 subunit with proliferating cell nuclear antigen ( PCNA ) , the DNA polymerase processivity clamp ( Gérard et al . , 2006; Shibahara and Stillman , 1999 ) . Human CAF1 preferentially interacts with the replication-dependent histones H3 . 1/2 but not with the replication-independent histone variant H3 . 3 ( Benson et al . , 2006; Tagami et al . , 2004 ) . Budding yeast contains only a H3 . 3 ortholog which is used in both replication-dependent and independent H3 deposition pathways . The C-terminal region of p150 contains a conserved Winged-helix DNA-binding domain ( WHD ) that is thought to contribute to stabilization of CAF1 at the replication fork ( Zhang et al . , 2016 ) . The p60 subunit preferentially associates with Asf1b in vivo ( Abascal et al . , 2013; Gurard-Levin et al . , 2014; Tagami et al . , 2004 ) , and Asf1 is thought to deliver H3-H4 dimers to CAF1 for deposition onto DNA ( Tyler et al . , 1999 ) . The p48 subunit ( Rb-associated protein RbAp48 ) interacts with a single H3-H4 dimer through the N-terminal tail of H3 and the N-terminal helix of H4 ( Nowak et al . , 2011; Schmitges et al . , 2011; Song et al . , 2008; Zhang et al . , 2013 ) . However , in the context of CAF1 , the N-terminal H3 and H4 tails are not essential for histone deposition activity , indicating that additional contacts are made between p150 and/or p60 and the H3-H4 core histone fold ( Shibahara et al . , 2000; Winkler et al . , 2012 ) , a model supported by recent hydrogen/deuterium exchange ( HX ) data and cross-linking MS ( Kim et al . , 2016; Liu et al . , 2016 ) . Whether CAF1 binds a H3-H4 dimer or tetramer has remained controversial . Human CAF1 has been found to bind a single H3-H4 dimer ( Benson et al . , 2006; Tagami et al . , 2004 ) , whereas yeast CAF1 ( yCAF1 ) has been reported to bind H3-H4 tetramers in a non-canonical conformation in vitro and in vivo and prior to deposition onto DNA ( Liu et al . , 2012b; Winkler et al . , 2012 ) . Further , it has been proposed that the Cac1 subunit in isolation is sufficient to enable H3-H4 tetramerization ( Liu et al . , 2016 ) . Together , the model is emerging that H3-H4 are maintained as dimers from synthesis up until tetramer assembly by CAF1 . Whether the tetramerization of H3-H4 occurs on CAF1 prior to deposition onto DNA or via a sequential CAF1-mediated deposition of two H3-H4 dimers onto DNA , remains unclear . To explore the molecular mechanism underlying H3-H4 deposition by CAF1 , we performed structure-function analysis of yCAF1 and of the yCAF1-H3-H4 complex and analyzed the histone deposition reaction onto DNA . We report that yCAF1 binds a single H3-H4 heterodimer and prevents H3-H4 tetramer formation . Our data imply that two yCAF1-H3-H4 complexes cooperate for assembly and deposition of H3-H4 tetramers . Biochemical studies show that the Cac1 subunit contacts DNA through a DNA-binding domain that is located in the region comprising the predicted coiled-coil segment of Cac1 and that high affinity DNA-binding also requires the WHD domain . High affinity DNA-binding by yCAF1 requires a B-form DNA substrate in the range of ~40–80 bp , due to cooperative binding of two yCAF1 complexes . Such extended DNA substrates allow deposition of H3-H4 tetramers by yCAF1 . While DNA binding does not prove necessary for tetramer deposition using purified components , it is required for DNA synthesis-coupled nucleosome assembly in an in vitro assembly system . DNA-binding deficient mutants retained the ability to bind H3-H4 heterodimers yet histone binding per se was not sufficient for the chaperone activity of yCAF1 . In addition , H3-H4 binding to yCAF1 and yAsf1 or Mcm2 was mutually exclusive suggesting a possible hand-over mechanism for final H3-H4 tetramer deposition onto DNA by yCAF1 . Finally , we report that H3-H4 tetramerization is required for release of H3-H4 from yCAF1 during DNA deposition . We thus propose a model in which two yCAF1-H3-H4 complexes cooperatively bind to an extended DNA element enabling deposition of two copies of H3-H4 , the first step in nucleosome formation .
Budding yeast CAF1 ( yCAF1 ) is a heterotrimeric complex containing the Cac1 , Cac2 and Cac3 subunits ( Kaufman et al . , 1997 ) . Cac1 contains a predicted coiled-coil region rich in amino acid residues K/E/R , an acidic E/D domain and a C-terminal WHD , which was shown to interact with DNA ( Zhang et al . , 2016 ) . Cac2 and Cac3 are predicted WD40 domain proteins ( Figure 1A ) . While Cac2 contains a C-terminal Asf1 interaction motif , Cac3 is predicted , based on sequence similarity to the mammalian ortholog RbAp48 , to interact with histones H3-H4 . To determine the molecular architecture of the yCAF1 complex , we established a co-expression system for Cac1 , Cac2 and Cac3 in insect cells ( Figure 1A ) . Expression and purification of this complex yielded a stable , monodisperse heterotrimeric yCAF1 complex with 1:1:1 stoichiometry , as judged from size-exclusion chromatography coupled to multi-angle laser light scattering ( SEC-MALLS; Figure 1B , Table 1 ) . This complex was able to bind histones H3-H4 to yield a homogeneous complex that , despite the additional mass , had a slightly smaller hydrodynamic radius as indicated by mobility on the size-exclusion column ( see also below ) . To derive further insights into the architecture of the yCAF1 complex and its interaction with H3-H4 , we used limited proteolysis to identify stable yCAF1 subcomplexes . In agreement with disorder predictions ( Figure 1—figure supplement 1A ) , we found that while Cac2 and Cac3 were mostly protease cleavage resistant , Cac1 was readily cleaved into smaller fragments ( Figure 1—figure supplement 1B ) . Mass spectrometry analysis revealed a C-terminal fragment of Cac1 ( Cac1T; amino acid residues 230–606 ) . Using this information , we designed a series of N- and C-terminal Cac1 truncations ( Figure 1A ) and co-expressed these with Cac2 and Cac3 in insect cells . We found that all the Cac1 truncation variants were able to bind to Cac2 and Cac3 allowing purification of stable and monodisperse heterotrimeric complexes ( Figure 1C , D ) . These data show that the Cac1 region required for interaction is located in the segment spanning amino acid residues 230–494 . All the yCAF1 variants described here retained the ability to bind H3-H4 , indicating that the deleted regions of Cac1 are not essential for histone interaction . Additional limited proteolysis revealed yCAF1 complex dissociation during SEC analysis ( Figure 1—figure supplement 1C ) . LC-MS analysis showed that peak two contained a complex of Cac1 comprising amino acid residues 234–442 and close to full-length Cac3 spanning amino acid residues 5–422 . This complex did not retain Cac2 binding activity , which migrated as separate peak on the SEC column ( Figure 1—figure supplement 1C ) . This Cac2 fragment , spanning residues 1–434 , was only missing the B domain , a known binding site for yAsf1 ( Malay et al . , 2008 ) . We therefore tentatively assign the Cac2 and Cac3 binding regions on Cac1 to amino acid residues 443–489 and 234–442 , respectively . The assignment of the Cac2 and Cac3 binding regions on Cac1 is in agreement with earlier data from yeast and human CAF1 and emphasizes the conserved nature of the CAF1 complex ( Kaufman et al . , 1995; Krawitz et al . , 2002 ) . 10 . 7554/eLife . 23474 . 003Table 1 . Overall biophysical parameters of yCAF1 . Column labeling: SEC-MALLS ( Size-exclusion chromatography - multi-angle laser light scattering ) ; EQ-AUC ( equilibrium analytical ultracentrifugation ) ; SV-AUC ( sedimentation velocity analytical ultracentrifugation ) ; SAXS ( small angle X-ray scattering ) ; Native MS ( native mass spectrometry ) ; MMSLS ( Molar masses determined by SEC-MALLS ) ; MMAUC ( Molar masses determined by equilibrium analytical ultracentrifugation ) ; s020 , w ( Sedimentation coefficient determined by velocity analytical ultracentrifugation ) ; sth ( computed sedimentation coefficient derived from SAXS envelopes ) ; Rg ( radius of gyration ) ; MMSAXS ( molar masses determined by SAXS ) ; Dmax ( maximum dimension ) ; Vp ( excluded particle Volume ) ; MMMS ( molar masses determined by native MS; MMth ( theoretical molar mass calculated ) . The errors reported for SEC-MALLS are the residual standard deviations of the observed data from the fitted values calculated using Astra . The errors of the AUC experiments are derived from the standard deviations of linear fits of the obtained data points to extrapolate the respective values ( MMAUC and s020 , w ) to zero protein concentration . The errors reported for the parameters derived from SAXS are based on the observed range of results is it possible to obtain , adjusting ( within acceptable theoretical limits ) the data points used for the calculation and as such represent the confidence range of the parameter . Resolution of the ab inito SAXS models was calculated according to ( Tuukkanen et al . , 2016 ) . Errors in native MS were determined according to ( McKay et al . , 2006 ) . CSLS , CSAXS , CMS are the concentrations of samples used for SEC-MALLS , SAXS and native MS respectively . N . D . ( not determined ) DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 003Sec-mallsEq-aucSv-aucSAXSSAXSSAXSSAXSSAXSNative MSSampleMMSLS kDa ( CSLS , µM ) MMAUC kDasth Svedbergs020 , w SvedbergRg nm ( CSAXS mg·ml−1 ) MMSAXS kDaDmax nmVp nm3Resolution ÅMMMS kDa ( CMS µM ) MMth kDayCAF1172 . 4 ± 1% ( 67 ) 180 ± 106 . 26 . 41 ± 0 . 036 . 39 ± 0 . 22 ( 11 . 4 ) 175 ± 326 ± 2307 ± 557 ± 4174 . 49 ± 0 . 30 ( 2 . 5 ) 174 . 0yCAF1-H3-H4198 . 9 ± 1 . 1% ( 50 ) 200 ± 117 . 16 . 84 ± 0 . 066 . 02 ± 0 . 35 ( 10 ) 203 ± 1325 ± 2355 ± 2354 ± 4201 . 00 ± 0 . 01 ( 7 ) 200 . 7yCAF1T142 ± 1% ( 15 ) N . D . N . D . N . D . 5 . 66 ± 0 . 03 ( 30 . 0 ) 127 ± 120 ± 1 . 2222 ± 249 ± 4N . D . 146 . 8yCAF1T-H3-H4153 ± 1% ( 15 ) N . D . N . D . N . D . 5 . 10 ± 0 . 03 ( 13 . 2 ) 163 ± 117 . 3 ± 1 . 3285 ± 252 ± 4N . D . 173 . 610 . 7554/eLife . 23474 . 004Figure 1 . Domain architecture and purification of yCAF1 . ( A ) Domain architecture of Cac1 , Cac2 and Cac3 with K/E/R domain containing a predicted coiled-coil segment; PIP , PCNA interacting peptide; E/D , acidic domain; WHD , winged helix domain; WD40-repeat β-propeller domain; B , Asf1 interaction domain . Constructs used are shown below . yCAF1T and yCAF1V are missing the first two residues of the PIP motif ( 227-Q-x-x-I-x-x-F-F-234 ) in Cac1 ( B ) Determination of the apparent molecular mass of yCAF1 ± H3-H4 using SEC-MALLS . Lines correspond to the UV280nm traces of yCAF1 ( red ) or yCAF1–H3-H4 ( blue ) eluting from the SEC column . Dots correspond to the molar mass determined for yCAF1 or yCAF1 –H3-H4 . ( C ) SDS-PAGE analysis of purified yCAF1 constructs ± H3-H4 . ( D ) Determination of the apparent molecular mass of yCAF1 mutants ± H3-H4 using SEC-MALLS . The UV280nm traces are shown as lines and the molar mass measurements as dots . ( E ) Native mass spectra of yCAF1 in ( E ) the absence or ( F ) the presence of H3-H4 . The yCAF1 ( ★ ) , yCAF1-H3-H4 ( ★ ) and the subcomplexes are labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 00410 . 7554/eLife . 23474 . 005Figure 1—figure supplement 1 . Structural and biochemical characterization of yCAF1 complexes . ( A ) Coiled-coil and disorder probability as determined by COILS and Disopred3 , respectively . Domain arrangements of Cac1 , Cac2 and Cac3 are shown below . ( B ) Limited proteolysis using chymotrypsin digestion revealed stable yCAF1 fragments . Left half of gel: yCAF1 control without protease; I , input . Identity of these fragments was determined by LC/MS . ( C ) Limited proteolysis of yCAF1X followed by SEC analysis . The resulting peaks were analyzed by LC/MS . Peak two contained a complex of Cac1234-442 and Cac35-422 indicating that these fragments interact directly . Peak three contained Cac21-434 . ( D ) Tandem MS spectrum of the 32+ ion of yCAF1-H3-H4 . The 32+ ion population of yCAF1-H3-H4 was subjected to collision-induced dissociation ( CID ) . This experiment broke the complex into subcomplexes ( yCAF1-H3 , yCAF1-H4 , Cac1-Cac3-H3-H4 , Cac1-Cac2-H3-H4 ) and monomers ( H3 , H4 , Cac2 and Cac3 ) and confirms the 1:1:1:1:1 stoichiometry of the yCAF1-H3-H4 complex . ( E ) Native MS analysis of yCAF1-H3-H4 complexes with an excess of H3-H4 . The yCAF1 ( at a concentration of 4 . 6 μM ) was incubated with 1 . 5 time molar excess of H3-H4 for 30 min at 4°C and then analyzed by native MS . The peaks at high m/z range showed that the yCAF1 was bound to a single copy of H3-H4 ( 201 kDa ) . At low m/z range , signal for the tetrameric ( H3–H4 ) 2 was detected ( 53 kDa ) in close proximity to the peaks belonging to Cac2 ( 53 kDa ) . ( F ) Native MS analysis of H3-H4 preparations . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 00510 . 7554/eLife . 23474 . 006Figure 1—figure supplement 2 . MS analysis of yCAF1-H3-H4 and H3-H4 . ( A ) Tandem MS spectrum of the 32+ ion of yCAF1-H3-H4 . The 32+ ion population of yCAF1-H3-H4 was subjected to collision-induced dissociation ( CID ) . This experiment broke the complex into subcomplexes ( yCAF1-H3 , yCAF1-H4 , Cac1-Cac3-H3-H4 , Cac1-Cac2-H3-H4 ) and monomers ( H3 , H4 , Cac2 and Cac3 ) and confirms the 1:1:1:1:1 stoichiometry of the yCAF1-H3-H4 complex . ( B ) Native MS analysis of yCAF1-H3-H4 complexes with an excess of H3-H4 . The yCAF1 ( at a concentration of 4 . 6 μM ) was incubated with 1 . 5 time molar excess of H3-H4 for 30 min at 4°C and then analyzed by native MS . The peaks at high m/z range showed that the yCAF1 was bound to a single copy of H3-H4 ( 201 kDa ) . At low m/z range , signal for the tetrameric ( H3–H4 ) 2 was detected ( 53 kDa ) in close proximity to the peaks belonging to Cac2 ( 53 kDa ) . ( C ) Native MS analysis of H3-H4 preparations . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 006 The yCAF1 complex is thought to enable H3-H4 tetramerization before final deposition onto DNA ( Liu et al . , 2012b , 2016; Winkler et al . , 2012 ) . However , previous immunoprecipitation experiments show that human CAF1 binds a single copy of H3-H4 ( Tagami et al . , 2004 ) . To resolve this apparent controversy , we performed a series of biophysical measurements including SEC- MALLS , analytical ultracentrifugation ( AUC ) and native mass spectrometry ( native MS ) , a technique with high mass accuracy . SEC-MALLS showed a mass of 172 kDa for yCAF1 , consistent with the expected mass of a heterotrimer ( Figure 1B , Table 1 ) . In the presence of H3-H4 we found single species with a mass of 199 kDa , in agreement with binding of one yCAF1 trimer bound to one H3-H4 dimer . We obtained the same mass even when H3-H4 was added in twofold molar excess . All four mutant yCAF1 complexes analyzed also formed monodisperse heterotrimers that bound a single copy of H3-H4 ( Figure 1C , D and Table 2 ) . 10 . 7554/eLife . 23474 . 007Table 2 . Summary of SEC-MALLS data for yCAF1 variants . Column labeling: SEC-MALLS ( Size-exclusion chromatography - multi-angle laser light scattering ) ; MMSLS ( Molar masses determined by SEC-MALLS ) ; MMth ( theoretical molar mass calculated ) . CSLS is the concentration used for SEC-MALLS . Errors reported are the residual standard deviations of the observed data from the fitted values calculated using Astra . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 007SampleMMSLS kDa ( CSLS , µM ) MMth kDayCAF1U166 ± 1 . 1% ( 20 ) 161 . 4yCAF1U-H3-H4184 ± 1% ( 20 ) 188 . 2yCAF1V133 ± 1 . 2% ( 20 ) 134 . 4yCAF1V-H3-H4156 ± 1 . 1% ( 20 ) 161 . 2yCAF1X143 ± 1% ( 15 ) 147 . 2yCAF1X-H3-H4160 ± 1% ( 15 ) 174 . 0 Native MS of yCAF1 showed an ion population at high m/z corresponding to a mass of 174 kDa , in agreement with the mass of intact yCAF1 ( Figure 1E; Table 3 ) . At low m/z we observed peaks of free Cac2 ( 53 kDa ) and Cac3 , ( 50 kDa ) while the intermediate m/z range showed masses corresponding to Cac1-Cac3 and Cac2-Cac3 subcomplexes . In presence of histones , yCAF1 was bound to a single copy of H3-H4 ( 201 kDa ) ( Figure 1F ) . The 32+ ion population of yCAF1-H3-H4 was subjected to tandem MS experiments and confirmed a 1:1:1:1:1 stoichiometry of the yCAF1-H3-H4 complex ( Figure 1—figure supplement 2A ) . Thus , consistent with previous results , we found that isolated yCAF1 forms a heterotrimer containing a single copy of each subunit ( Liu et al . , 2012b; Winkler et al . , 2012 ) . However , none of our data support the model that yCAF1 binds two H3-H4 heterodimers . To further test this , we added excess H3-H4 to yCAF1 and analyzed the sample by native MS . The largest species obtained corresponded to yCAF1 bound to a single H3-H4 heterodimer ( Figure 1—figure supplement 2B ) . Of note , the masses observed for excess H3-H4 correspond to those of tetramers ( Figure 1—figure supplement 2B ) . Native MS analysis of the H3-H4 sample also showed masses corresponding to tetramers indicating that H3-H4 remain associated as tetramers in this experiment ( Figure 1—figure supplement 2C , Table 3 ) . Considering that isolated H3-H4 forms tetramers in solution at equivalent concentrations ( Winkler et al . , 2012 ) and in our native MS experiments ( Figure 1—figure supplement 2C , Table 3 ) , our data suggest that yCAF1 prevents H3-H4 tetramerization . 10 . 7554/eLife . 23474 . 008Table 3 . Summary of native mass spectrometry . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 008Protein sampleConcentration ( μM ) Oligomerization stateMeasured mass ± error ( Da ) *Calculated mass ( Da ) yCAF12 . 5Cac1:Cac2:Cac3174 492 ± 3173965 . 1yCAF12 . 5Cac1:Cac3123 927 ± 5120735yCAF12 . 5Cac2:Cac3103 840 ± 4103755 . 1yCAF12 . 5Cac253 273 ± 653230 . 1yCAF12 . 5Cac350 568 ± 750525yCAF1-H3-H47Cac1:Cac2:Cac3:H3:H4201 002 ± 5200 720 . 3yCAF1-H3-H47Cac2:Cac3:H3:H4130 343 ± 7130510 . 4H3-H410† ( H3-H4 ) 253 015 ± 453510 . 6H3-H410†H3-H426 508 ± 226755 . 3H3-H410†H315 271 ± 215388H3-H410†H411236 ± 311367 . 3*Values reported represent the mean value ± standard deviation according to ( McKay et al . , 2006 ) . Combinations of neighboring m/z values were used to determine distinct M values of a macromolecule . Using these values , a mean value of M and its standard deviation were calculated . †Values reported assume that H3-H4 are tetrameric in solution . In addition , we carried out AUC sedimentation velocity experiments on yCAF1 and yCAF1-H3-H4 ( Figure 2A , B ) . A g ( s* ) analysis showed that there was a minor shift in the peak position with an increase in the concentration yCAF1 or yCAF1-H3-H4 , indicating that predominantly a single species was present at all examined concentrations . The sedimentation coefficients are in agreement with a monomeric complex in solution and also with theoretical Svedberg values calculated from SAXS envelopes ( see below ) . Sedimentation equilibrium experiments showed that the samples contained a single monodisperse species and the residuals showed mostly random distribution ( Figure 2—figure supplement 1 ) . For yCAF1 , the data yielded a molecular mass of 179 kDa , and for yCAF1-H3-H4 of 199 kDa , values close to the calculated masses of these complexes considering equal stoichiometry of the polypeptide chains ( Table 1 ) . 10 . 7554/eLife . 23474 . 009Figure 2 . yCAF1 binds a single H3-H4 heterodimer . ( A ) Sedimentation velocity analytical ultracentrifugation of the yCAF1 complex . ( B ) yCAF1-H3-H4 . Shown is the plot of the sedimentation coefficient distribution at different protein concentrations . ( C ) The experimental SAXS profile ( log intensity ( I ) as a function of the momentum transfer ( q ) ) . Dots with error bars are the experimental scattering data . yCAF1 ± H3-H4 ( blue or orange respectively ) . The normalized fit to the experimental data is superimposed as a black line . Inset: Guinier plot ( log I vs . ( q2 ) of the low q region of the X-ray scattering data ( D ) yCAF1T ± H3-H4 ( blue or orange , respectively ) . Bottom panels: Normalized interatomic distance distribution functions . ( E ) The p ( r ) distribution plot for yCAF1 alone ( blue ) and bound to H3-H4 ( orange ) . ( F ) The p ( r ) distribution plot for yCAF1T alone ( blue ) and bound to H3-H4 ( orange ) . ( G ) Average DAMMIN bead models . Left: yCAF1 ( orange ) and yCAF1-H3-H4 ( blue ) . Middle: yCAF1T ( orange ) and yCAF1T-H3-H4 ( blue ) . Right: Superposition of yCAF1 ( grey ) onto yCAF1T ( orange ) and yCAF1-H3-H4 ( grey ) onto yCAF1T-H3-H4 ( blue ) . Arrows indicate the N-terminal extension of Cac1 and the possible positioning of H3-H4 . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 00910 . 7554/eLife . 23474 . 010Figure 2—figure supplement 1 . Analytical ultracentrifugation of yCAF1 complexes . ( A ) Sedimentation equilibrium analytical ultracentrifugation of yCAF1 . Shown is the plot of the concentration of 3 . 2 μM yCAF1 as a function of the radial distance after reaching equilibrium at 7 . 000 ( blue ) , 10 . 000 ( red ) and 40 . 000 ( orange ) rpm . Solid black lines are derived from a global fit of all datasets to a model describing an ideal non-interacting single-component system . Bottom panel: Random scatter in the residuals indicates that this model describes the data well . ( B ) Plot as described above for yCAF1-H3-H4 at a concentration of 3 . 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 010 To further analyze the structure of yCAF1 , we performed small angle X-ray scattering measurements of yCAF1 and yCAF1T complexes unbound or bound to H3-H4 . yCAF1T lacks the N-terminal region of the Cac1 subunit ( Figure 1A ) . To reduce potentially confounding interparticle effects , we measured scattering data using an in-line size exclusion chromatography system ( SEC-SAXS ) ( Pernot et al . , 2013 ) . The scattering curves showed no sample aggregation , and the linear Guinier range indicated monodisperse protein complexes for all four complexes analyzed ( Figure 2C , D and insets ) . From these data , we obtained an Rg of 6 . 39 nm for yCAF1 while the distance distribution function p ( r ) showed skewed shape characteristic of elongated , multi-domain particles , with a maximum diameter ( Dmax ) of 26 nm ( Figure 2E ) . For yCAF1-H3-H4 , we obtained an Rg of 6 . 02 and a Dmax of 25 nm ( Figure 2E , Table 1 ) . The molecular masses determined from SAXS were 175 kDa ( yCAF1 ) and 203 kDa ( yCAF1-H3-H4 ) , in close agreement with those determined from orthogonal methods ( Table 1 ) . Ab initio reconstructions showed an elongated particle for yCAF1 ( Figure 2G ) . In the H3-H4-bound form , a similar elongated shape was obtained with additional mass towards the center of the particle ( Figure 2G ) . Ab initio calculation of the Svedberg coefficients of yCAF1 and yCAF1-H3-H4 using the molecular envelopes obtained from the SAXS experiments also match the measured Svedberg values from sedimentation velocity AUC experiments ( Table 1 ) , also emphasizing the elongated shape of the particle . For yCAF1T we obtained an Rg of 5 . 66 nm and a Dmax of 20 nm and for the H3-H4-bound complex an Rg of 5 . 10 nm and a Dmax of 17 . 3 nm ( Figure 1D , F and Table 1 ) . Ab initio reconstructions also showed an elongated particle for yCAF1T ( Figure 2G ) . As for yCAF1 , addition of H3-H4 to yCAF1T , resulted in additional mass towards the center of the particle . Based on the overall reduction of the particle diameter of yCAF1T , which matches the expected volume of the flexible N-terminal region of Cac1 , we suggest that the bulk of this segment is located there . Comparison of H3-H4-bound and unbound yCAF1 variants indicates that the histone heterodimer is positioned towards the center of the particle ( Figure 2G ) . We consistently observed , as also shown in SEC-MALLS ( Figure 1B ) , a smaller radius of gyration for the H3-H4-bound yCAF1 complexes , indicative of conformational compaction upon histone binding . Together , a series of rigorous and complementary biophysical techniques led us to conclude that yCAF1 forms an elongated trimer with 1:1:1 stoichiometry , which binds a single H3-H4 heterodimer . Previous data show that Cac1 contains a C-terminal WHD DNA-binding domain ( Zhang et al . , 2016 ) . The WHD binds to a 10–16 bp DNA element in a non-specific fashion with a dissociation constant Kd of ~2 μM ( Zhang et al . , 2016 ) . As previous studies were limited to the isolated WHD , we investigated the DNA-binding requirements of full-length yCAF1 . Purification of the yCAF1 complex from insect cells revealed the presence of contaminating nucleic acids and insect cell histones H2A , H2B , H3 and H4 suggesting that yCAF1 potentially directly interacts with nucleosomes ( Figure 3—figure supplement 1A ) . To test this model , we analyzed binding of yCAF1 to reconstituted nucleosome core particles using recombinantly produced Xenopus histones , which were assembled on a 147 bp DNA fragment containing the 601 nucleosome positioning sequence ( Lowary and Widom , 1998 ) . Native PAGE analysis showed that these preparations contained nucleosomes and a fraction of unbound 147 bp DNA ( Figure 3—figure supplement 1B ) . Titration of the yCAF1 complex showed that in this direct competition assay , yCAF1 preferentially interacted with the naked DNA but not with the nucleosome core particles ( Figure 3—figure supplement 1B , lanes 2–8 ) . Thus , surfaces of H3-H4 that are required for yCAF1 binding are buried once these histones are assembled into nucleosomes . We found that yCAF1 bound to the 147 bp DNA fragment with a dissociation constant KD of ~2 . 1 μM ( Figure 3A , Table 4 ) . Quantitative analysis of the binding reaction showed a Hill coefficient of 2 . 0 , indicative of cooperative binding . yCAF1U , a construct lacking the WHD domain had a KD of ~4 . 1 μM , ( Figure 3A , Table 4 ) . The non-conserved N-terminal region ( amino acids 1–129 ) did not contribute significantly to DNA binding as a construct lacking this segment ( CAF1X ) had similar DNA-binding affinity ( 5 . 7 μM ) . yCAF1U and yCAF1X showed less steep binding isotherms but still largely retained their ability to assemble cooperatively as judged by a positive Hill coefficient ( Table 4 ) . yCAF1T , a variant lacking the N-terminal 229 amino acids including the K/E/R-rich coiled-coil , showed no detectable DNA binding , despite the fact that it contained the WHD domain indicating that the DNA-binding activity of the WHD domain is masked . yCAF1V , which contained a deletion of the WHD , showed also no detectable DNA binding ( Figure 3A ) . Taken together , these data indicate that the region spanning the K/E/R-rich coiled-coil segment of Cac1 contains a DNA-binding domain which appears to hierarchically cooperate with the C-terminal WHD for high-affinity DNA binding as constructs lacking this domain are devoid of DNA binding . The WHD contributes to overall DNA-binding avidity but is not sufficient to enable high affinity binding in the absence of the coiled-coil domain . 10 . 7554/eLife . 23474 . 011Table 4 . DNA binding by yCAF1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 011Protein sampleDNA substrateKD [μM] *Hill coefficientyCAF1147 bp2 . 1 ± 0 . 12 . 0 ± 0 . 484 bp2 . 3 ± 0 . 32 . 2 ± 0 . 342 bp2 . 5 ± 0 . 52 . 2 ± 0 . 417 bp5 . 1 ± 1 . 01 . 3 ± 0 . 2yCAF1U147 bp4 . 1 ± 1 . 94 . 3 ± 5 . 8yCAF1V147 bp>10N . D . yCAF1X147 bp5 . 7 ± 1 . 72 . 9 ± 0 . 6yCAF1T147 bp>10N . D . *Values determined from experiments using the 147 , 84 , 42 or 17 bp DNA fragment . Errors , where reported , correspond to the SEM value of three technical replicates . 10 . 7554/eLife . 23474 . 012Figure 3 . yCAF1 binding to DNA . ( A ) Left panel: Binding curves of yCAF1 variants to 147 bp DNA . Right panels: EMSA showing binding of yCAF1 variants to 147 bp DNA . Free DNA and yCAF1-bound ( B ) DNA are indicated . Wells ( W ) are indicated additionally with red horizontal bars . Increasing amounts of yCAF1 ( 0 . 15 , 0 . 3 , 0 . 61 , 1 . 25 , 2 . 5 , 5 or 10 μM ) were mixed with 1 μM DNA . Error bars represent SEM values of three technical replicates ( B ) Left panel: binding curves of yCAF1 binding to DNA fragments of 20–100 bp ( Sequence information in Table 5 ) . Right panel: EMSA showing yCAF1 binding to free DNA fragments of 20–100 bp . Concentration of the DNA 10–100 bp ladder was 275 nM overall nucleotide base pairs present in the binding reaction . yCAF1- DNA binding was quantified by measuring DNA substrate depletion . ( C ) Left panel: Binding curves of yCAF1 to 17 bp , 42 bp and 84 bp DNA . Right panels: EMSA showing binding of yCAF1 to 17 bp , 42 bp and 84 bp DNA . Wells ( W ) , free DNA and yCAF1-bound ( B ) DNA are indicated . Increasing amounts of yCAF1 ( 0 . 15 , 0 . 3 , 0 . 61 , 1 . 25 , 2 . 5 , 5 , 10 , 20 , 40 or 80 μM ) were mixed with 1 μM DNA . Error bars represent SEM values of three technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 01210 . 7554/eLife . 23474 . 013Figure 3—figure supplement 1 . Analysis of nucleosome binding by yCAF1 . ( A ) Anion exchange chromatography of yCAF1 ( first peak ) revealing a histone-DNA contaminant in the yCAF1 preparation from Trichoplusia ni insect cells ( second peak ) . The identity of all four Trichoplusia ni histones was confirmed by SDS-PAGE and LC-MS . ( B ) EMSA showing that yCAF1 does not bind to nucleosomes but to free DNA . Increasing amounts of yCAF1 ( 0 . 3 μM to 20 μM in two fold steps ) were mixed with 2 μM nucleosome . Free DNA , nucleosomes ( Nuc ) , and DNA-bound yCAF1 ( B ) are indicated . Lane one contained no yCAF1 ( - ) and lane nine contained yCAF1 but no nucleosome . ( C ) Uncropped gel image of Figure 3B - EMSA showing yCAF1 binding to free DNA fragments of 20–100 bp . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 013 Analysis of the DNA sequence length preferences , using a DNA ladder ranging from 20 to 100 bp DNA size ( Table 5 ) , showed that yCAF1 bound DNA fragments longer than >40 bp more efficiently than shorter ones ( Figure 3B , C ) . While this would not be unusual for a DNA-binding factor , the fact that yCAF1 discriminated against DNA shorter than 40 bp was unexpected as the WHD binds DNA of 10–16 bp length ( Zhang et al . , 2016 ) . To further test the interplay between DNA substrate length and yCAF1 binding efficacy we investigated whether yCAF1 can also cooperatively bind to DNA molecules that are shorter than 40 bp in length . We performed DNA-binding assays with 17 bp , 42 bp or 84 bp DNA substrates and analyzed the binding isotherms ( Figure 3C ) . In agreement with our previous results we observed that while 42 bp and 84 bp DNA were clearly bound cooperatively by yCAF1 ( Hill coefficient of 2 . 2 for both substrates ) , the 17 bp DNA substrate displayed standard Michaelis-Menten kinetics ( Hill coefficient of 1 . 3 ) indicative of a single binding site ( Table 4 ) . Of note , while yCAF1 efficiently bound to free B-DNA , the DNA geometry of the nucleosome core particle , a reaction product of yCAF1-mediated assembly , is not compatible with yCAF1 interaction ( Figure 3—figure supplement 1B ) . Together , our results suggest that yCAF1 uses a two-pronged DNA-binding mode involving the WHD and the coiled-coil segment of Cac1 . Optimal yCAF1 DNA binding requires regular B-DNA geometry and a minimum of ~40 bp length . 10 . 7554/eLife . 23474 . 014Table 5 . Sequence information on 10 bp DNA ladder ( Promega ) . AT content ( % ) for all DNA fragments is 60% . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 014Length ( bp ) Sequence10GGACTATACT20GGACTATACTAGACATTGAC30GGACTATACTAGACATTGACGTGGTTGTAA40GGACTATACTAGACATTGACGTGGTTGTAAGATGATCATG50GGACTATACTAGACATTGACGTGGTTGTAAGATGATCATGTGTTAATGGC60GGACTATACTAGACATTGACGTGGTTGTAAGATGATCATGTGTTAATGGCAAGGTGAGTT70CATGATCATCTTACAACCACGTCAATGTCTAGTATAGTCCTACTCTGTGATATGGTTCTCTGTCGATGTA80GCCATTAACACATGATCATCTTACAACCACGTCAATGTCTAGTATAGTCCTACTCTGTGATATGGTTCTCTGTCGATGTA90AACTCACCTTGCCATTAACACATGATCATCTTACAACCACGTCAATGTCTAGTATAGTCCTACTCTGTGATATGGTTCTCTGTCGATGTA100ATGATCATCTAACTCACCTTGCCATTAACACATGATCATCTTACAACCACGTCAATGTCTAGTATAGTCCTACTCTGTGATATGGTTCTCTGTCGATGTA Considering that two yCAF1-H3-H4 complexes need to come together for assembly of two copies of H3-H4 on DNA to form so-called ‘tetrasomes’ , we hypothesized that the DNA length requirements are due to binding of two complexes to an extended DNA substrate for histone deposition . Parenthetically , a DNA substrate of similar length ( ~60–80 bp ) is also required for tetramer binding in the nucleosome ( Luger et al . , 1997 ) . To assess the ability of yCAF1 to assemble tetrasomes , we incubated yCAF1-H3-H4 with a 84 bp DNA fragment derived from the H3-H4 binding region of the 601 nucleosome positioning sequence ( Lowary and Widom , 1998 ) and of sufficient length for salt deposition of a single tetramer ( Figure 4—figure supplement 2A ) . The reactions were analyzed on native PAGE and stained for DNA ( SYBR Safe , left panel ) or protein ( Coomassie , right panel ) . As expected , salt-deposition showed H3-H4 tetrasomes ( Figure 4A , lane 2 and Figure 4—figure supplement 2A ) . A histone H3 L126R/I130R mutant ( H3M ) , which disrupts tetramer formation ( Winkler et al . , 2012 ) , showed a band with higher mobility , interpreted to represent disomes , a dimer of H3M-H4 bound to DNA ( Figure 4A , lane 1 and Figure 4—figure supplement 2B ) . Titration of the yCAF1-H3-H4 complex onto DNA showed the appearance of tetrasomes but no apparent disome assembly intermediates ( Figure 4A , left panel , lanes 3–10 ) . The top part of the gel also showed yCAF1 bound to DNA , apparently in the absence of H3-H4 ( see below ) . 10 . 7554/eLife . 23474 . 015Figure 4 . yCAF1 deposition of H3-H4 . ( A ) EMSA showing tetrasome deposition on 84 bp DNA . Increasing amounts of yCAF1-H3-H4 ( 0 . 15 , 0 . 3 , 0 . 61 , 1 . 25 , 2 . 5 , 5 or 10 μM ) were mixed with 1 μM 84 bp DNA and the bands resolved by native PAGE . Gels were stained for DNA with SYBR Safe ( left panel ) and for protein with Coomassie ( right panel ) . ( B ) As above but for yCAF1V-H3-H4 . ( C ) As above but for yCAF1-H3M-H4 ( H3M contains the L126R/I130R mutation ) . * indicates extracted gel bands that we analyzed by SDS-PAGE ( Figure 4—figure supplement 2D ) . All EMSA experiments were repeated at least two times with consistency . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 01510 . 7554/eLife . 23474 . 016Figure 4—figure supplement 1 . EMSA analysis of H3-H4 deposition . ( A ) EMSA showing H3-H4 deposition by yCAF1T . ( B ) yCAF1U . ( C ) yCAF1X . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 01610 . 7554/eLife . 23474 . 017Figure 4—figure supplement 2 . EMSA analysis of H3-H4 deposition . ( A ) EMSA showing tetrasome and ( B ) disome assembly controls . The position of migration of the dimsome , tetrasome and free DNA are indicated . The gels were stained with SYBR Safe or Coomassie stain as indicated . Increasing amounts of yCAF1-H3-H4 ( 0 . 2 μM to 10 μM in two fold steps ) were mixed with 1 μM 84 bp DNA . ( C ) EMSA showing that yCAF1 binding to the 84 base pair DNA substrate migrates at the same position as yCAF1 that has released its H3-H4 cargo and subsequently bound to excess free DNA . All EMSA experiments were repeated at least two times with consistency . ( D ) Bands indicated by * in Figure 4A , C were extracted from the native PAGE gel and analyzed by SDS-PAGE followed by Coomassie blue staining . The bottom panel shows a high contrast rendering of the bottom part of the SDS-PAGE gel . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 017 The yCAF1T , yCAF1U , yCAF1V and yCAF1X complexes bound to H3-H4 also retained to varying degrees tetrasome deposition activity ( Figure 4B , Figure 4—figure supplement 1A–C ) . yCAF1U , which lacks the WHD , showed normal tetrasome deposition ( Figure 4—figure supplement 1B ) . yCAF1X , which has an additional deletion of amino acids 1–128 of Cac1 , also showed wild-type levels of tetrasome deposition ( Figure 4—figure supplement 1C ) . The DNA-binding deficient yCAF1T deposited tetrasomes and a complex with slightly higher mobility , presumably a tetrasome in a non-canonical position on the DNA substrate ( Figure 4—figure supplement 1A ) . yCAF1V also showed tetrasome assembly activity and in addition a disome assembly intermediate ( Figure 4B , lanes 6–8 ) . In all cases , addition of DNA resulted in release of H3-H4 from yCAF1 as seen by the appearance of tetrasomes and free yCAF1 . This was most clearly seen with the DNA-binding deficient variants yCAF1T and yCAF1V ( Figure 4B and Figure 4—figure supplement 1A ) . With both mutant complexes , addition of yCAF1-H3-H4 to DNA resulted in the appearance of tetrasomes and free yCAF1T or yCAF1V . Only when the free DNA was exhausted from the reaction , the excess yCAF1T-H3-H4 or yCAF1V-H3-H4 complexes were seen ( Figure 4B and Figure 4—figure supplement 1A , right panels , lanes 8–10 ) . As expected for these DNA-binding deficient variants , free yCAF1T or yCAF1V , liberated during the histone deposition reaction , did not interact with the DNA ( Figure 4B , Figure 4—figure supplement 1A , left panels , lanes 3–10 ) . For the DNA-binding competent yCAF1U and yCAF1X , as for wild-type yCAF1 , we observed in addition to tetrasome deposition , that the released yCAF1U or yCAF1X further partitioned into a DNA-bound complex at lower concentrations ( Figure 4—figure supplement 1B–C , left panel , lanes 3–10 ) or free yCAF1U or yCAF1X when in excess over DNA . Together , we conclude that DNA binding by yCAF1 is not necessary to mediate H3-H4 deposition in vitro . Addition of DNA consistently resulted in tetrasome deposition and release of H3-H4 from yCAF1 . In no instance could we detect a yCAF1-H3-H4-DNA co-complex . Supporting this model , a yCAF1-DNA complex formed in the absence of H3-H4 migrated at a similar position on the native PAGE gel as the yCAF1-DNA complexes seen in the H3-H4 deposition reactions ( Figure 4—figure supplement 2C ) . To assess the contribution of the H3-H3’ tetramerization interface , we produced a yCAF1 complex bound to the H3M-H4 mutant and analyzed DNA deposition . This complex was deficient in histone deposition on DNA as seen by a lack of disome or tetrasome deposition . Instead , mostly non-specific binding is seen ( Figure 4C , left panel , lanes 4–10 ) . Apparently this defect arose due to a failure to release histones from yCAF1 ( Figure 4C , right panel , lanes 3–9 ) . Released yCAF1 would be expected to migrate as the free yCAF1 complex ( Figure 4C , lane 12 ) . While some free yCAF1 is seen at the highest concentration used ( 10 μM; Figure 4C , lane 10 ) , the amounts of free yCAF1 were clearly lower than that seen in other DNA deposition assays . We also confirmed the deposition defect directly by extracting the yCAF1-DNA bands from the native PAGE gel followed by analysis by SDS-PAGE . As expected , in the absence of DNA , the extracted yCAF1-H3-H4 band , which migrated close to top of the gel ( * in Figure 4A , lane 11 ) , showed yCAF1 and bound H3-H4 ( Figure 4—figure supplement 2D , lane 1 ) . With wild-type H3-H4 , the DNA-bound complex ( * in Figure 4A , lane 7 ) , showed yCAF1 but no H3-H4 ( Figure 4—figure supplement 2D , lane 2 ) , showing that yCAF1 had released H3-H4 . However with the H3M-H4 mutant , the DNA-bound yCAF1 complex ( * in Figure 4C , lane 8 ) contained H3M-H4 ( Figure 4—figure supplement 2D , lane 3 ) , demonstrating that the deposition defect arose due to a failure to release H3-H4 from yCAF1 . Taken together , these results provide compelling evidence that the ability of H3-H4 to tetramerize upon DNA deposition contributes to histone release from yCAF1 . Our biochemical studies suggest a mechanism in which two yCAF1-H3-H4 complexes cooperatively bind to an extended DNA sequence element to assemble and deposit H3-H4 tetrasomes . We used a Xenopus egg high-speed egg extract ( HSE ) system to study the ability of yCAF1 mutant complexes to perform DNA synthesis-dependent chromatin assembly . To compare DNA-synthesis dependent and independent chromatin assembly , we used a plasmid with ( pBSuv ) or without UV-damage ( pBSo ) , as described previously ( Ray-Gallet et al . , 2007 ) . Indeed , Xenopus HSEs enabled us to follow chromatin assembly independently of the requirement for DNA synthesis when using pBSo ( Figure 5A , upper panel , lanes 3 , 6 ) and in a DNA synthesis-dependent manner when using UV-treated DNA pBSuv ( Figure 5A , lower panel , lane 6 ) . 10 . 7554/eLife . 23474 . 018Figure 5 . DNA-binding of yCAF1 is required for DNA synthesis-coupled nucleosome assembly . ( A ) Nucleosome assembly reactions with either non-UV-treated plasmid ( pBS0 ) or plasmid irradiated with UV ( pBSuv ) in presence of [α−32P] . After an incubation time of 5 or 180 min , DNA was extracted , resolved on an agarose gel and visualized by ethidium bromide staining ( EtBr ) or by autoradiography ( Autradiogram ) . ( B ) 150 ng of pBSuv plasmid was incubated with p150-depleted HSE extracts and complemented with the indicated amounts of yCAF1 . After 3 hr incubation , DNA was extracted , resolved on and agarose gel and visualized as above . ( C ) As in ( B ) but reactions were complemented with 125 ng of the different yCAF1 variants and the amount of pBSuv plasmid was increased to 300 ng ( D ) As in ( B ) but reactions were complemented with 100 ng of the different yCAF1 variants . ( I ) Supercoiled and ( II ) relaxed plasmid . All reactions were repeated two times with consistency . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 01810 . 7554/eLife . 23474 . 019Figure 5—figure supplement 1 . Depletion of Xenopus p150 from HSE . ( A ) p150- or mock-depleted HSEs were analyzed by Western blotting using the indicated antibodies . ( B ) Nucleosome assembly reactions with 300 ng of pBSuv plasmid and 50 or 100 ng of yCAF1 variants showing that none of the yCAF1 mutants reach wild-type level activity . ( C ) Western blot of a FLAG pulldown of yeast PCNA with FLAG tagged yCAF1 WT or yCAF1V to investigate the effect of the N-terminal mutant on PCNA binding . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 019 To assess whether recombinant yCAF1 complexes are able to promote nucleosome assembly in a physiological context , we first immunodepleted the endogenous p150 from HSEs ( Figure 5—figure supplement 1 ) and used these extracts for complementation assays . Such p150-depleted extracts were unable to assemble chromatin in a DNA synthesis-dependent fashion ( Figure 5B , lane 4 ) . Previous data show that these HSEs can be complemented with human p150 ( Quivy et al . , 2001 ) . Here , we report that addition of increasing amounts of yCAF1 rescued DNA synthesis-dependent chromatin assembly ( Figure 5B , lane 4–7 ) . This experiment allowed us to establish limiting concentrations of yCAF1 that allow nucleosome assembly , as described previously ( Mello et al . , 2002 ) . The different CAF1 mutants showed , to varying degrees , defects in DNA synthesis-coupled nucleosome assembly ( Figure 5—figure supplement 1B ) and specific conditions allowed us to discriminate for their efficiency ( Figure 5C , D ) . The construct lacking the coiled-coil K/E/R domain ( yCAF1T ) showed residual nucleosome assembly activity as judged by the relative intensity of supercoiled product ( Figure 5C ) . Constructs lacking the WHD ( yCAF1U ) or both the N-terminal region and WHD ( yCAF1X ) , showed more deficient nucleosome assembly activity ( Figure 5C ) . yCAF1V , a variant containing a deletion of both the coiled-coil and WHD of Cac1 showed the greatest assembly defect , as judged from the amount of supercoiled plasmid product ( labeled I in Figure 5C ) . This defect of yCAF1V was even observable when increasing the yCAF1-DNA ratio such that all other mutants were able to almost fully compensate for the nucleosome assembly defect ( Figure 5D ) . The mutants yCAF1T and yCAF1V contain a partially truncated PIP motif in which the first two residues of the conserved Q-x-x-I-x-x-F-F motif are absent ( Figure 1A ) . As this motif is a PCNA binding site , we asked whether the nucleosome assembly defect of these mutants is due interference with PCNA binding . A FLAG-pulldown experiment showed that yCAF1V was able to interact with PCNA , albeit to a lesser extent than yCAF1 wild-type ( Figure 5-supplemental figure 1C ) . We therefore cannot fully exclude that the defective PIP motif contributes to the phenotype that we describe here . However it is clear that DNA binding is required for yCAF1 activity . The two DNA-binding domains were essential for CAF1 activity as the absence of either domain interfered to different degrees with DNA synthesis-dependent chromatin assembly . A current model suggests that upon DNA replication , parental H3-H4 are transiently maintained as dimers by Mcm2-Asf1 and reassembled into tetramers by CAF1 behind the replication fork ( Clément and Almouzni , 2015; Huang et al . , 2015; Richet et al . , 2015 ) . Asf1 is known to interact with the Cac2/p60 subunit of CAF1 possibly to enable histone transfer between the two chaperones ( Kim et al . , 2016; Malay et al . , 2008; Mello et al . , 2002 ) . To assess a possible histone hand-over mechanism from Asf1-Mcm2 towards yCAF1 , we systematically analyzed various combinations of these histone chaperones in the absence or presence of H3-H4 substrate by SEC-MALLS ( Figure 6 , Table 6 ) . yAsf1 and yCAF1 did not interact under these conditions , presumably due to the transient nature of this interaction ( Malay et al . , 2008 ) . In the presence of H3-H4 , we observed yCAF1-H3-H4 and Asf1-H3-H4 but no higher-order complex ( Figure 6A ) . Addition of a fivefold molar excess of preassembled Asf1-H3-H4 to yCAF1 resulted in appearance of a yCAF1-H3-H4 complex , due to transfer of H3-H4 from Asf1 to yCAF1 ( Figure 6A ) . These data show that yCAF1 is able to receive H3-H4 from Asf1 , but that there is no stable higher-order complex of these histone chaperones . An N-terminal fragment of Mcm2 , spanning amino acid residues 1–160 and sufficient for H3-H4 binding , also did not interact directly with yCAF1 ( Figure 6B , Table 3 ) . In a direct competition assay , H3-H4 interacted preferentially with yCAF1 and Mcm2 migrated in its unbound form ( Figure 6B ) . Incubation of a fivefold molar excess of Mcm2-H3-H4 with yCAF1 resulted in transfer of H3-H4 to yCAF1 ( Figure 6B ) . Finally , addition of equimolar ratios of the three chaperones Mcm2 ( 1-160 ) , yAsf1 and yCAF1 followed by addition of H3-H4 led to the formation of the yCAF1-H3-H4 and yAsf1-H3-H4-Mcm2 complexes . Of note , the molecular masses obtained are compatible with the model that a single H3-H4 dimer is bound to the complexes ( Table 6 ) . 10 . 7554/eLife . 23474 . 020Table 6 . Summary of SEC-MALLS data . Column labeling: Ve ( elution Volume ) ; MMSLS ( Molar masses determined by SEC-MALLS ) ; MMth ( theoretical molar mass calculated ) . When there are more than two proteins in the injected sample , ‘+' indicates the mixing order . In sample 4 , a five-fold molar excess of a preformed yAsf1-H3-H4 complex was incubated with yCAF1 before injection . In sample 9 , a five-fold molar excess of a preformed MCM2-H3-H4 complex was incubated with yCAF1 . The errors reported are the residual standard deviations of the observed data from the fitted values calculated using Astra . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 020Peak 1Peak 2Peak 3SampleVe ( ml ) MMsls ( kDa ) MMth ( kDa ) Ve ( ml ) MMsls ( kDa ) MMth ( kDa ) Ve ( ml ) MMsls ( kDa ) MMth ( kDa ) yCAF19 . 96172 . 1 ± 0 . 1174 . 0––––––yCAF1 + yAsf19 . 96171 . 1 ± 0 . 6174 . 012 . 9138 . 6 ± 0 . 231 . 6–––yCAF1 + yAsf1 + H3 H410 . 00185 . 3 ± 0 . 1200 . 712 . 8564 . 3 ± 0 . 258 . 3–––yCAF1 + yAsf1-H3-H4 1 :59 . 98192 . 9 ± 0 . 2200 . 712 . 8860 . 1 ± 0 . 158 . 3–––yAsf1–––12 . 9636 . 3 ± 0 . 131 . 6–––yAsf1 + H3 H4–––12 . 8658 . 7 ± 0 . 258 . 3–––yCAF1 + MCM29 . 95174 . 1 ± 0 . 7174 . 0–––14 . 3219 . 1 ± 0 . 517 . 6yCAF1 + MCM2 + H3 H49 . 97185 . 9 ± 0 . 1200 . 7––––––yCAF1 + MCM2-H3-H4 1 :59 . 98190 . 5 ± 0 . 7200 . 712 . 0990 . 8 ± 0 . 471 . 0–––MCM2––––––14 . 3218 . 8 ± 0 . 117 . 6MCM2 + H3 H4–––12 . 1682 . 6 ± 0 . 471 . 0–-–yCAF1 + H3 H49 . 97191 . 3 ± 0 . 4200 . 7––––––yCAF1 + yAsf1+Mmc2 + H3 H410 . 05188 . 8 ± 0 . 9200 . 712 . 7967 . 5 ± 0 . 175 . 914 . 3827 . 4 ± 0 . 117 . 610 . 7554/eLife . 23474 . 021Figure 6 . Competition of yCAF1 with yAsf1 or Mcm2 for H3-H4 binding . SEC-MALLS analysis of complexes formed upon mixing of up to three histone chaperones with H3-H4 . In all experiments , lines correspond to the UV280nm traces of the eluting complex . Dots correspond to the molar mass measurements . Eluting fractions were analyzed by SDS-PAGE and the relevant areas of the gels are displayed below each chromatogram using the corresponding color code . When there are two or more proteins in the mixture , the ‘+' in the labeling indicates the order in which the samples were mixed together ( eg . yCAF1 + yAsf1 +H3-H4 indicates that yAsf1 was added to yCAF1 followed by addition of H3-H4 ) . The final protein concentration used was 20 µM for all proteins except where preformed yAsf1-H3-H4 and Mcm2-H3-H4 were supplied in 5-fold molar excess to yCAF1 ( blue lines in A and B ) . ( A ) H3-H4 competition experiments of yCAF1 with yAsf1 . ( B ) H3-H4 competition experiments of yCAF1 with Mcm2 . ( C ) H3-H4 competition experiments of yCAF1 in the presence of yAsf1 and Mcm2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 021
In agreement with recent negative stain electron microscopy data ( Kim et al . , 2016 ) , we found that yCAF1 forms an elongated heterotrimer containing a single copy of each subunit and that H3-H4 binds in a central position . Our deletion mapping is also in agreement with recent cross-linking MS experiments that indicate that the Cac1 subunit scaffolds interactions with Cac2 , Cac3 and histones H3-H4 ( Kim et al . , 2016; Liu et al . , 2016 ) . As the isolated Cac1 subunit is able to interact with H3-H4 ( Liu et al . , 2016 ) , our data are in agreement with the model that Cac1 contributes substantially to H3-H4 binding while Cac2 and Cac3 provide accessory interactions . However , in contrast to previous findings ( Liu et al . , 2012b , 2016; Winkler et al . , 2012 ) , we could not detect the presence of a H3-H4 tetramer bound to yCAF1 even when H3-H4 were supplied in excess and under conditions that favor tetramerization ( Figure 1—figure supplement 2C ) . Previous data are based on fluorescence titration or FRET experiments which show that yCAF1 binds two copies of H3-H4 or a covalently crosslinked H3-H4 tetramer with high ( KDapp ~5 nM ) binding affinity ( Liu et al . , 2012b , 2016; Winkler et al . , 2012 ) . Our data suggest that high-affinity histone binding by yCAF1 is driven , at least in part , by the highly polar DNA-binding surface of H3-H4 . Considering that the DNA-binding surface is maintained in the crosslinked H3-H4 tetramer , it is not too surprising that this substrate binds with high affinity if it can be accommodated sterically . We found that yCAF1 can receive H3-H4 from Asf1 or Mcm2 without forming a stable higher-order complex among these histone chaperones ( Figure 6 ) . Thus , we propose that the mechanism of H3-H4 tetrasome assembly by yCAF1 requires yAsf1- or Mcm2-dependent transfer of H3-H4 towards yCAF1 ( Figure 7 ) . Although the Cac1 subunit contains a reported DNA-binding WHD that binds to 10–16 bp DNA ( Zhang et al . , 2016 ) , this domain is not sufficient for high-affinity DNA binding . In agreement , point mutations that abolish DNA binding of the isolated WHD do not abolish DNA binding by yCAF1 ( Zhang et al . , 2016 ) . Nevertheless , the WHD clearly contributes to overall yCAF1 activity and might play a role in transcriptional silencing activity and the DNA damage response presumably by aiding recruitment of yCAF1 to replication forks ( Liu et al . , 2016; Zhang et al . , 2016 ) . 10 . 7554/eLife . 23474 . 022Figure 7 . Model for yCAF1 recruitment and H3-H4 deposition . Free monomeric yCAF1 ( step 1 ) is loaded with dimeric H3-H4 through association of yAsf1 with the Cac2 subunit . Alternatively , loading can occur through hand over of H3-H4 from Mcm2 ( step 2 ) . yCAF1 binds the histones via their DNA binding and oligomerization surfaces ( step 3 ) . During DNA synthesis , two yCAF1-H3-H4 complexes bind cooperatively to an extended DNA element >50 bp ( step 4 ) to deposit H3-H4 dimers and form tetrasomes . The WHD ( orange ) and coiled-coil ( red ) DNA-binding domains of yCAF1 are required for deposition of H3-H4 tetramers . The requirement of an extended free DNA region together with PCNA interaction may direct yCAF1 activity to replication forks . H2A-H2B chaperones like NAP1 or FACT recognize the tetrasome intermediate and deposit two copies of H2A-H2B ( step 5 ) to form a complete nucleosome ( step 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23474 . 022 High affinity DNA binding by yCAF1 requires a second DNA-binding element that we locate to the amino acid residues 135–230 of Cac1 , a region containing the K/E/R domain and is predicted to form a coiled-coil . This region is also critical in human p150 , as a deletion mutant spanning the equivalent region ( amino acids 311–445 ) abolishes chromatin assembly activity ( Kaufman et al . , 1995 ) . We identified two mutants , yCAF1T and yCAF1V , that were deficient in DNA interaction but that retained H3-H4-binding activity . Counterintuitively , the WHD appears to be inert for DNA binding in yCAF1T ( Figure 3A ) . Previous hydrogen/deuterium exchange ( HX ) data indicate that H3-H4 binding results in structural rearrangements in Cac1 as evidenced by increased HX for amino acids 550–591 ( Liu et al . , 2016 ) . Thus , H3-H4 binding to yCAF1 could result in unmasking of the DNA-binding activity of the WHD . However DNA synthesis-dependent chromatin assembly requires the presence of both the coiled-coil K/E/R and WHD DNA-binding domains . yCAF1 binds preferentially to DNA of 40–80 bp length and we find that binding to such extended DNA elements is cooperative . Assembly of H3-H4 tetrasomes requires a DNA substrate of similar length ~60–80 bp ( Luger et al . , 1997 ) . Cooperative binding usually requires some form of interaction between binding partners but we do not find evidence for yCAF1 or yCAF1-H3-H4 dimerization in solution . Previous studies have indicated , that the isolated Cac1 subunit and the p150 ortholog , have a tendency to dimerize in the absence of their binding partners ( Gérard et al . , 2006; Liu et al . , 2016; Quivy et al . , 2001; Winkler et al . , 2012 ) . A C-terminal fragment of Cac1 , spanning amino acid residues 385–606 , is sufficient to bind H3-H4 and to promote deposition of a H3-H4 tetramer onto DNA ( Liu et al . , 2016 ) , but the physiological relevance of dimerization of isolated Cac1/p150 remains unclear . yCAF1 binds cooperatively with DNA and when loaded with H3-H4 heterodimers , this results in tetrasome deposition . Our model is that histone tetramerization gives directionality to the deposition reaction as it is required for histone release from yCAF1 . H3-H4 tetramers interact with higher binding affinity ( ~1 nM ) with DNA as compared to yCAF1 ( ~5 nM ) ( Andrews et al . , 2010; Winkler et al . , 2012 ) . A H3M-H4 mutant is expected to show decreased DNA-binding affinity as tetramerization stabilizes H3-H4 on DNA . However the H3M-H4 mutant retains high affinity ( ~5 . 5 nM ) yCAF1 binding ( Winkler et al . , 2012 ) , likely explaining the failure to release from yCAF1 in a DNA deposition assay ( Figure 4C and Figure 4—figure supplement 2D ) . yCAF1 has a preference for regular B-DNA as we failed to detect binding to nucleosome core particles that are known to contain DNA with unusual conformational parameters ( Luger et al . , 1997 ) . An extended nucleosome-free DNA region of ~250 bp is present behind and ~300 bp ahead of the replication fork ( Gasser et al . , 1996; Sogo et al . , 1986 ) . Nucleosome assembly occurs as soon as DNA of sufficient length to wrap around the histone octamer core has passed through the replication machinery ( Sogo et al . , 1986 ) . Thus it is conceivable that the DNA-binding preferences of yCAF1 described here allow yCAF1 to ‘sense’ regions of extended nucleosome-free DNA and promote nucleosome assembly preferentially onto the nucleosome-free region of replicating DNA . In contrast , assembled chromatin , which contains nucleosomes separated by small linker DNA ( ~15–20 bp on average in yeast ) , would be expected not to allow the yCAF1 binding thereby preventing its action ( Radman-Livaja and Rando , 2010 ) . This could explain why yCAF1 does not promote significant nucleosome assembly in the absence of ongoing DNA replication . Not unlike what we propose here for yCAF1 , direct DNA binding by HIRA and Rtt106 , histone chaperone complexes involved in deposition of different H3 variants , has been proposed ( Liu et al . , 2010; Ray-Gallet et al . , 2011; Schneiderman et al . , 2012 ) . PCNA has long been reported to act as the key recruitment vehicle for CAF1 at replication forks ( Moggs et al . , 2000; Shibahara and Stillman , 1999; Zhang et al . , 2000 ) . Yeast Cac1 contains one PCNA binding site ( PIP ) that is conserved from yeast to human p150 and is required for interaction with PCNA ( Krawitz et al . , 2002 ) . We suggest that yCAF1 recruitment to sites of DNA replication requires cooperation between DNA binding and PCNA interaction . Such dependencies are frequently seen in chromatin regulators that utilize high-affinity nucleic acid-binding domains coupled to low-affinity binding modules that recognize histone modifications ( Ptashne , 2009 ) . The high-affinity interactions thereby reduce the entropic penalty of bringing the weakly binding modules and their substrate together and expand the dynamic range for local recruitment by increasing affinity up to several-fold ( Ptashne , 2009 ) . Our data are consistent with a model where two yCAF1-H3-H4 complexes cooperatively bind to an extended DNA sequence element to deposit H3-H4 tetrasomes ( Figure 7 ) . In solution , H3-H4 tetramerization in the yCAF1 complex is not permitted , possibly because residues in the H3-H3’ dimerization interface are occluded . Furthermore , considering that H3-H4 binding to yCAF1 or DNA is mutually exclusive , we propose that yCAF1 interacts with the DNA-binding surface of H3-H4 . This model is also consistent with our competition experiments with Mcm2 ( Figure 6 ) . All the yCAF1 variants analyzed allowed tetrasome assembly using purified components , but the presence of the DNA-binding domains is required for DNA synthesis-coupled nucleosome assembly ( Figure 5 ) . H3-H4 tetramerization stabilizes the H3-H4 dimers on DNA and triggers the release from yCAF1 ( Figure 4C ) . This model is in agreement with data showing that H3-H4 dimers do not assemble as tetramers prior to deposition onto DNA ( Benson et al . , 2006; Tagami et al . , 2004 ) and puts forward the importance of a series of coordinated transitions that require the unique DNA-binding properties of yCAF1 . Together , we propose a model for DNA synthesis-coupled chromatin assembly in which ( i ) yCAF1 binds to a single H3-H4 dimer , ( ii ) two yCAF1-H3-H4 complexes can bind to nascent DNA in a concerted manner enabling H3-H4 tetramer formation on DNA , which ( iii ) results in release of histones from yCAF1 . The DNA substrate requirements of the reaction could fit nicely with the length of Okazaki fragments during replication on the lagging strand , the periodic sizing of which also depends on yCAF1-dependent histone deposition ( Smith and Whitehouse , 2012 ) , and would also be relevant during nucleotide excision repair . Future work , in vivo and in other organisms , should address how the mechanistic insights provided here are linked to inheritance of chromatin-based information .
The three yCAF1 subunits Cac1 , Cac2 and Cac3 from Saccharomyces cerevisiae were amplified by PCR from pPK133 , pPK160 and pPK134 and cloned into the pIDC , pIDK and pFL vectors for the Multibac system using InFusion cloning ( Trowitzsch et al . , 2010 ) . Cac3 contained a N-terminal decahistidine tag , Cac2 was FLAG-tagged , both containing TEV protease recognition sites for removal of the tags . N-terminal Cac1 deletions were generated by PCR of the pIDC-Cac1 plasmid and by using forward primers carrying a NdeI site , amplifying the shortened version of Cac1 . The corresponding , NdeI containing reverse primer amplified the plasmid backbone . PCR products were cut with NdeI , purified and re-ligated . Forward primers: pIDC_Cac1_NdeI_129_fwd 5’ ATATCATATGAAGAGAGAACTTTCCTCATCG 3’ pIDC_Cac1_NdeI_230_fwd 5’ ATATCATATGATTGGTAACTTCTTTAAAAAACTAAGCG 3’ Reverse primer: pIDC_BB_NdeI_START_rev 5’ ATATCATATGCGGACCGGGATCCGC 3’ The WHD deletion was generated using a similar approach by using the following primers carrying XhoI sites: pIDC_Cac1_494_XhoI_rev 5’ ATATCTCGAGTTACGATGTTTTGGGTTC 3’ pIDC_BB_XhoI_fwd 5’ ATATCTCGAGGGCCTACGTCGACGAG 3’ Cac1 truncations were designed to remove regions containing predicted disorder and considering limited proteolysis experiments ( Figure 1—figure supplement 1 ) . Baculovirus generation and expression in Hi5 cells was carried out as described before ( Trowitzsch et al . , 2010 ) . 1–2 l Hi5 cell culture expressing yCAF1 constructs were harvested by centrifugation ( 800 g , 20 min , 4°C ) and resuspended in 100 ml lysis buffer ( 20 mM TRIS pH 7 . 8 , 500 mM NaCl , 5 mM Imidazole , 0 . 1% NP-40 , 0 . 5 mM TRIS ( 2-carboxyethyl ) phosphine hydrochloride ( TCEP ) , containing protease inhibitor tablets ( Roche , Switzerland ) and Benzonase . Cells were lysed by sonication for 30 s on ice followed by centrifugation at 30000 g for 45 min at 4°C . The soluble lysate was loaded onto a 5 ml HisTrap FF column ( GE Healthcare , UK ) , pre-charged with Co2+ ions and equilibrated in lysis buffer using a peristaltic pump at 4°C . After loading , the column was washed in with 10 column volumes ( CV ) of lysis buffer , followed by 40 CV of wash buffer ( 20 mM TRIS pH 7 . 8 , 500 mM NaCl , 5 mM Imidazole , 0 . 5 mM TCEP ) . Bound protein was eluted with buffer containing 20 mM TRIS pH 7 . 8 , 300 mM NaCl , 500 mM Imidazole 0 . 5 mM TCEP and subsequently diluted with 20 mM TRIS pH 7 . 8 , 0 . 5 mM TCEP to give a final NaCl concentration of 150 mM NaCl . The sample was loaded onto a 5 ml HiTrap Q column ( GE Healthcare ) and subsequently connected to an AKTA Purifier FPLC system for washing in Q-150 buffer ( 20 mM TRIS pH 7 . 8 , 150 mM NaCl , 0 . 5 mM TCEP ) and elution using a 20 CV gradient of buffer Q-1000 ( 20 mM TRIS pH 7 . 8 , 1 M NaCl , 0 . 5 mM TCEP ) . yCAF1-containing fractions were identified by SDS-PAGE , pooled and concentrated before injection onto a Superdex 200 16/60 size exclusion column ( GE Healthcare ) equilibrated in SEC buffer ( 20 mM BisTRIS pH 6 . 5 , 500 mM NaCl , 0 . 5 mM TCEP ) . Protein purity was assessed and the protein-containing fractions were pooled and concentrated to 8–23 mg ml−1 using Amicon centrifuge filter units ( 100 kDa cutoff ) . Concentrated protein was maintained at 4°C or flash frozen in liquid nitrogen and stored at −80°C . Recombinant Xenopus laevis histones H3 and H4 were expressed , purified and refolded according to standard procedures ( Luger et al . , 1999 ) . The H3 tetramerization mutant ( H3M ) , containing the point mutations L126R and I130R which disrupt tetramer formation ( Winkler et al . , 2012 ) , was created by directed mutagenesis from the wild type plasmid using primers 5’ GCTGGCCCGCAGAAGGCGAGGCGAGAGG 3’ and 5’ CCTCTCGCCTCGCCTTCTGCGGGCCAGC 3’ for I130R and 5’ CAAGGACATCCAGCGGGCCCGCAGAATCC 3’ and 5’ GGATTCTGCGGGCCCGCTGGATGTCCTTG 3’ for L126R and verified by DNA sequencing . H3M was expressed , purified from inclusion bodies following the same procedure as used for the wild type and the presence of the mutation verified by LC-MS . H3-H4 were assembled by dissolving equimolar amounts of each lyophylized histone in unfolding buffer ( 20 mM TRIS pH 7 . 5 , 7 M Guanidinium chloride , 5 mM β-mercaptoethanol ) . After mixing H3-H4 , the sample was incubated for one hour , followed by dialysis for 16–18 hr at 4°C against refolding buffer ( 10 mM TRIS , pH 7 . 5 , 2 M NaCl , 1 mM Na-EDTA , 5 mM β-mercaptoethanol ) . The sample was then run in refolding buffer on an equilibrated HiLoad 16/60 Superdex 75 column ( GE Healthcare ) . Aliquots were stored in 50% glycerol at −20°C . To prepare yCAF1-H3-H4 complexes , reconstituted H3-H4 tetramers were added at twofold molar excess to yCAF1 . The samples were incubated for up to 120 min on ice before loading on a Superdex 200 10/300 GL column ( GE Healthcare ) in SEC buffer . Protein-containing fractions were pooled and concentrated to 20–30 mg ml−1 using Amicon centrifuge filter units ( 100 kDa cutoff ) . Concentrated protein was maintained at 4°C or flash frozen in liquid nitrogen and stored at −80°C . For salt-deposition of histones , we used standard protocols ( Dyer et al . , 2004 ) . Briefly , H3-H4 tetramers or H3M-H4 dimers were mixed with 84 bp or 147 bp DNA derived from the canonical Widom sequence in 20 mM TRIS , pH 8 . 0 , 2 M NaCl , 1 mM EDTA , 1 mM DTT and dialyzed against 1 . 5 M NaCl buffer for 2–3 hr at 4°C . The samples were then transferred into consecutively lower ( first 1 M , 0 . 5 M and then 0 . 25 M ) NaCl concentration buffer for 2 hr each with the second-last dialysis being an overnight step . Samples were then incubated at 37°C for 15 min and then maintained on ice prior to analysis . For analysis of DNA-binding by the EMSA , yCAF1 or yCAF1-H3-H4 were incubated with the 84 bp or 147 bp DNA in EMSA buffer ( 500 mM NaCl , 20 mM BisTRIS , pH 7 . 8 , 0 . 5 mM TCEP , 5% glycerol ) . The samples were maintained on ice for 30 min and then heat shifted at 37°C for 5 min prior to analysis . The binding reactions were analyzed on a 6% native 1x TRIS-Glycine ( 250 mM TRIS , 1 . 92 M glycine , pH 8 . 3 ) Mini-PROTEAN ( Bio-rad , Hercules , CA ) polyacrylamide gel using 1x TRIS-Glycine running buffer . The gel was stained with SYBR Safe ( Thermo Fisher Scientific , Waltham , MA ) to visualize DNA-bound complexes or Coomassie Blue for protein staining . Band intensities were quantified by ImageJ ( Version 1 . 51 ) and the data analyzed by using the Origin software ( Version 9 . 3 ) using a Hill equation binding model . For extraction of protein/DNA bands from native PAGE gels , the bands indicated with an * in Figure 4 were cut out and mechanically homogenized using a syringe with a 1 . 2 mm ⌀ needle . After addition of 1x SDS loading buffer the samples were boiled , spun down and 40 μl loaded on SDS-PAGE for analysis . Size-exclusion chromatography was performed at a flow rate of 0 . 5 ml min−1 on a Superdex 200 Increase 10/300 GL column equilibrated in SEC-MALLS buffer ( 20 mM Bis-TRIS , pH 6 . 5 , 500 mM NaCl , 1 mM DTT or TCEP ) at 21°C . A 30 µl sample of yCAF1 ± H3-H4 at 2–10 mg ml−1 , previously incubated on ice for 10 min to 3 hr , was injected onto the column and multi angle laser light scattering was recorded with a laser emitting at 690 nm using a DAWN-EOS detector ( Wyatt TechnologyCorp . Santa Barbara , CA ) . The refractive index was measured using a RI2000 detector ( Schambeck SFD , Germany ) . The molecular weight was calculated form differential refractive index measurements across the center of the elution peaks using the Debye model for protein using ASTRA software version 6 . 0 . 5 . 3 . To verify the stoichiometry of the yCAF1-H3-H4 complex , 1:1 and 1:2 ratios of yCAF1 to H3-H4 dimers were tested . X-ray scattering data were collected using an inline HPLC setup , at the Bio-SAXS beamline ( BM29 ) of the European Synchrotron Radiation Facility ( Pernot et al . , 2013 ) . Inline size-exclusion chromatography was performed at a temperature of 10°C using a Superdex 200 Increase 10/300 GL column equilibrated in SEC-MALLS buffer . Data were collected with a photon-counting Pilatus 1M detector at a sample-detector distance of 2 . 86 m , a wavelength of λ = 0 . 991 Å and an exposure time of 1 s/frame . A momentum transfer range of 0 . 008 to 0 . 47 Å−1 was covered ( q = 4π sinθ/λ , where θ is the scattering angle and λ the X-ray wavelength ) . Data collected across the peak were subtracted from buffer scattering and the frames showing a constant radius of gyration ( Rg ) were merged for further analysis . Rg values were obtained from the Guinier approximation sRg <1 . 3 using Primus ( Petoukhov et al . , 2012 ) . Distance distribution functions p ( r ) and the Porod volumes Vp were computed from the entire scattering curve using GNOM ( Petoukhov et al . , 2012 ) . The program DAMMIN ( Svergun , 1998 ) , was used to generate 40 low-resolution ab initio shape reconstructions . To select the most typical ab initio model of the complex and estimate its possible conformational space , these reconstructions were pairwise aligned and averaged using DAMAVER ( Volkov and Svergun , 2003 ) . The model with the lowest mean value of normalized spatial discrepancy ( NSD ) was selected as the most typical reconstruction . To assess the resolution and reliability of the reconstructions we used the Fourier Shell Correlation ( FSC ) approach as implemented in SASRES ( Tuukkanen et al . , 2016 ) . Prior to non–denaturing MS analyses , 50 μl of yCAF1 or yCAF1-H3-H4 were buffer exchanged into 250 mM ammonium acetate , pH 6 . 8 using a Superdex 200 3 . 2/300 column mounted on an ÄKTAmicro system ( GE Healthcare Life Sciences ) . The buffer of H3-H4 tetramers was exchanged into 1 M ammonium acetate pH 7 . 0 . The exchange did not affect the complex integrity , as judged from the SEC elution profile . For all the measurements , 2–4 μl of sample were loaded into nanoflow platinum-coated borosilicate electrospray capillaries ( Thermo Electron SAS , France ) . Protein ions were generated using a nanoflow electrospray ( nano-ESI ) source and MS analyses carried out on a quadrupole time-of-flight mass spectrometer ( Q-TOF Ultima , Waters Corporation , U . K . ) . The instrument was modified for the detection of high masses ( Sobott et al . , 2002; van den Heuvel et al . , 2006 ) . The following instrumental parameters were used: capillary voltage = 1 . 2–1 . 3 kV , cone potential = 40 V , RF lens-1 potential = 40 V , RF lens-2 potential = 1 V , aperture-1 potential = 0 V , collision energy = 30–140 V , and microchannel plate ( MCP ) = 1900 V , ToF pressure ≈8 10–6 mbar . For collision induced dissociation experiments , the collision voltage was increased up to 210 V , and collision cell pressure was ≈2 10–4 mbar . All mass spectra were calibrated externally using a solution of cesium iodide ( 6 mg/mL in 50% isopropanol ) and were processed with the Masslynx 4 . 0 software ( Waters Corporation ) and with Massign software package ( Morgner and Robinson , 2012 ) with minimal smoothing and no background subtraction . To calculate the molecular mass ( M ) and to estimate the standard deviation of the measurement , we followed a procedure described previously ( McKay et al . , 2006 ) . Briefly , two neighboring m/z values ( M/z1 and M/z2 ) are determined experimentally ( x and y ) and two equations are written ( M/z1 = x and M/z2 = y ) . Since z1 = z2–1 , the equations are solved to determine M , z1 and z2 using the MassLynx software ( Waters ) . The program takes several combinations of neighboring m/z values to determine distinct M values of a macromolecule . Using these values , a mean value of M and its standard deviation are calculated . The M values were determined from m/z values corresponding to the left edge of the peaks . These values provide the ‘least-adducted’ M of the noncovalent complexes ( McKay et al . , 2006 ) and are reported in Table 1 and Table 3 . Analytical centrifugation was performed using an Optima XL-A analytical ultracentrifuge ( Beckman Coulter , Brea , CA ) with an AN-60 Ti rotor . For sedimentation equilibrium experiments , six-channel cells were used , and data were acquired at a resolution of 0 . 001 cm with twenty replicates at a temperature of 4°C . Reference cells were loaded with buffer 20 mM Bis-TRIS ( pH 6 . 5 ) , 0 . 2 M NaCl , and 0 . 5 mM TCEP . Absorbance at 280 nm was used to monitor concentration gradients . yCAF1 concentrations were 0 . 5 , 1 and 3 . 2 μM , and the speeds were 7 . 000 , 10 . 000 , and 40 . 000 rpm . yCAF1-H3-H4 concentrations were 0 . 5 , 1 and 3 . 2 μM , and the concentration distribution was measured at identical rotor speeds as yCAF1 . Samples were determined to have reached equilibrium when scans taken 4 hr apart showed no systematic differences . The data were analyzed with the program WinNonlin . For sedimentation velocity experiments the purified yCAF1 complexes were loaded into two- sector centerpieces , and buffer 20 mM Bis-TRIS ( pH 6 . 5 ) , 0 . 2 M NaCl , and 0 . 5 mM TCEP was used for the reference chamber . Experiments were performed at 42 , 000 rpm and 4°C . Data were collected at a wavelength of 280 nm , using a spacing of 0 . 003 cm , with one average in the continuous scan mode . No time delay was used , allowing traces to be collected every ~1 min . Sedimentation coefficients were corrected to standard conditions ( 20°C , in water ) using DCDT+ ( version 2 . 4 . 3 ) ( Philo , 2006; Stafford , 1992 ) . The sedimentation velocity data were analyzed to obtain the g ( s* ) distribution of the sample using DCDT+ . The g ( s ) distributions were further analyzed to obtain the apparent molecular weight of the sample using DCDT+ . The protein partial specific volumes were calculated from the amino acid composition to 0 . 722 ml g−1 ( yCAF1 ) and 0 . 733 ml g−1 ( yCAF1-H3-H4 ) and solvent density was calculated through summation of the contribution of buffer components to 1 . 009 g cm−3 at 4°C using the program SEDNTERP . Molecular mass was determined using Beckman software provided as an add-on to Origin version 3 . 8 . All nine data sets were analyzed using a global fit procedure based on a model describing an ideal non-interacting single component system , with local parameters for reference concentrations and base-line offsets and global parameters for the molecular weight . Best-fits were determined through visual inspection of the residuals ( Figure 2—figure supplement 1 ) . To compare the consistency of the hydrodynamic parameters determined from SAXS and AUC , we determined a theoretical sedimation coefficient ( Sth ) from the SAXS beads model by using the program WinHydroPro++ ( Ortega et al . , 2011 ) . Input parameters including solvent density , solvent viscosity and partial specific volume were determined using SEDNTERP as described above . The temperature was 4°C and the theoretical molecular mass was calculated from the primary sequence . The radii of atoms were set to the same values as that obtained from the SAXS bead models . Recombinant Drosophila melanogaster histones H3-H4 ( identical to human histones H3 . 2 ) and Homo sapiens Mcm2 ( 1–160 ) were purified as described previously ( Richet et al . , 2015 ) . Recombinant full length Saccharomyces cerevisiae Asf1 was produced and purified using the same protocol as for Mcm2 except that the HisTrap column was replaced by a nickel-nitrilotriacetic acid ( Ni-NTA ) column ( Qiagen , Germany ) . The flow through was then loaded on an anion exchange column Resource Q ( GE Healthcare ) and Asf1 eluted using a buffer with 50 mM TRIS-HCl and 1 M NaCl . The elution buffer was replaced by 50 mM TRIS-HCl pH 7 . 5 storage buffer using an Amicon device ( Millipore , Billerica , MA ) and an YM10 regenerated cellulose membrane ( Millipore ) . For molar mass determination , purified proteins were analyzed using SEC-MALLS as described previously ( Richet et al . , 2015 ) . yAsf1 and yCAF1 were mixed in equimolar ratios ( final concentration of both chaperones 20 µM ) in 10 mM TRIS pH 7 . 5 , 0 . 5 M NaCl , 0 . 5 mM TCEP ( final volume 110 μL ) . To prevent H3-H4 aggregation a specific order of addition was maintained during sample setup ( histones added last ) . Samples were incubated at 4°C overnight prior to injection of 100 μl of into a Superdex 200 Increase 10/300 GL column ( GE Healthcare ) equilibrated in 10 mM TRIS pH 7 . 5 , 0 . 5 M NaCl , 0 . 5 mM TCEP at a flow rate of 0 . 5 ml . min−1 . Multi angle laser light scattering was recorded with a laser emitting at 690 nm using a DAWN-TREOS detector ( Wyatt TechnologyCorp . Santa Barbara , CA ) . The refractive index was measured using a T-rEX detector ( Wyatt technology . Santa Barbara , CA ) . The molecular weight was calculated from differential refractive index measurements across the center of the elution peaks using the Debye model for protein using ASTRA software version 6 . 1 . 7 . 13 . Xenopus High-Speed Egg extract ( HSE ) preparation and chromatin assembly assays were prepared as described previously ( Ray-Gallet and Almouzni , 2004 ) . After removal of the jelly coat by cysteine treatment , Xenopus laevis eggs were rinsed in extraction buffer ( 10 mM KOH-HEPES pH 7 . 8 , 70 mM KCl , 5% sucrose , 0 . 5 mM dithiothreitol ( DTT ) and protease inhibitors ) and centrifuged at 150 , 000 g for 1 hr at 4°C . The clear ooplasmic fraction was collected , aliquoted and stored at −80°C . Depletions were done by the addition of p150 antibody ( Quivy et al . , 2001 ) coupled to protein A-Sepharose slurry ( CL-4B; Amersham Biosciences , UK ) to HSE for 1 hr at 4°C on a rotating wheel . A pBS plasmid ( Stratagene , La Jolla , CA ) was used to perform the chromatin assembly reaction , damaged by UV-C ( 500 J/m2 ) ( named pBSUV ) or not ( named pBS0 ) . 10 μL of HSE ( depleted or mock depleted ) was added to 150 ng or 300 ng of pBSUV or pBS0 in a buffer containing 5 mM MgCl2 , 40 mM KOH-HEPES pH 7 . 8 , 0 . 5 mM DTT , 4 mM ATP , 40 mM phosphocreatine , 2 . 5 μg of creatine phosphokinase and 5 μCi of [α-32P]dCTP in a final volume of 25 μl . The reaction was incubated at 23°C for 3 hr . After 5 min , yCAF1 complexes or buffer were added to the reaction . Chromatin assembly was stopped by the addition of 25 μl of a mix containing 30 mM EDTA and 0 . 7% SDS . Brief treatments by RNAse A and Proteinase K were followed by phenol-chloroform-isoamyl alcohol DNA extraction . The pellets were resuspended in 16 μl of TE and 4 μl of 5x loading buffer and only half of this solution was loaded on a 1% agarose gel in TAE 1x . The gels migrated at 55 V for 15 hr at 4°C and were then stained with ethidium bromide to visualize total DNA . Finally , gels were dried out and analyzed by Phosphorimager to visualize newly-synthetized DNA . Purified full-length yCAF1 , or yCAF1V complexes were mixed with equal amounts of pure trimeric PCNA ( 10 µM each ) in pulldown buffer ( 50 mM Tris pH 7 . 5 , 500 mM NaCl ) and left for 10 min on ice before incubation with 20 µL equilibrated FLAG beads for 1 hr at 4°C ( shaking ) . Unbound material was removed and the beads were washed three times with 100 µL pulldown buffer . Bound protein was eluted with 2 × 30 µL FLAG peptide ( 0 . 4 mg/ml ) in 20 mM Tris pH 7 . 5 , 300 mM NaCl and 10 min incubation at room temperature each time . Eluted proteins were analyzed by SDS-PAGE and western blot using standard procedures . For western blot analysis , the samples were separated by SDS-PAGE and transferred to a nitrocellulose membrane , confirmed by Ponceau S red staining . The membrane was blocked with 5% defatted milk in TBST and incubated with monoclonal mouse anti-PCNA antibody ( 1:4000 , Abcam , UK ) in TBST over night at 4°C . After washing with TBST , the blot was incubated with HRP-conjugated anti-mouse secondary antibody ( 1:10 000 , Sigma , St . Louis , MO ) in TBST for 1 hr at room temperature . Finally , PCNA was detected by chemiluminescent signal from the ECL Prime Western Blotting Detection Reagent ( Amersham ) . Subsequently , the membrane was stripped using 0 . 2 M glycine , pH 2 . 2 , 0 . 1% SDS and 1% Tween using standard procedures . After incubation with anti-FLAG antibody ( 1:1000 , Sigma ) in TBST for 1 hr at room temperature the membrane was further treated as described for anti-PCNA . | Animal and plant cells contain very long DNA molecules that are tightly packaged by being wrapped around proteins called histones to form structures known as nucleosomes . While this is a useful way to store DNA , it also makes it inaccessible to many proteins and other molecules that activate genes , copy DNA or perform other important cell processes . To enable these processes to take place , the cell can selectively disassemble particular nucleosomes and remove the histone proteins . Afterwards , the nucleosomes must reassemble to repackage the DNA . A single nucleosome contains four pairs of histones , with two pairs consisting of a H3 and a H4 histone . Histone chaperones assemble nucleosomes in a two-step process . First , two of these histone H3-H4 pairs ( collectively known as a tetramer ) interact with DNA to form a group or “complex” . Then , two more pairs of different histones bind to complete the nucleosome . An enzyme called CAF1 is known to attach H3-H4 tetramers onto DNA as the DNA is being copied , which allows nucleosomes to form on the newly made DNA . However , it is not known how CAF1 deposits H3-H4 tetramers onto the DNA . Sauer et al . explored how yeast CAF1 works by carrying out a series of experiments in a cell-free system . The experiments showed that each CAF1 enzyme binds to a single H3-H4 pair . When attached to their histone cargo , two CAF1 enzymes bind to DNA and attach a H3-H4 tetramer onto it . The tetramer has to form in this way for histones to be correctly delivered to DNA after the DNA has been copied . Sauer et al . also identified a new region of the CAF1 enzyme that binds to DNA . Together with another region , this enables CAF1 to bind to an extended stretch of DNA that accommodates the H3-H4 tetramer . Together , the findings explain the sequence of events that take place when CAF1 attaches H3-H4 tetramers onto DNA in the first step of nucleosome formation . Future work will be required to understand the structure of CAF1 in different situations and to find out how the cell targets this enzyme to stretches of DNA that have just been copied . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Insights into the molecular architecture and histone H3-H4 deposition mechanism of yeast Chromatin assembly factor 1 |
Many cellular constituents travel along microtubules in association with multiple copies of motor proteins . How the activity of these motors is regulated during cargo sorting is poorly understood . In this study , we address this issue using a novel in vitro assay for the motility of localising Drosophila mRNAs bound to native dynein-dynactin complexes . High precision tracking reveals that individual RNPs within a population undergo either diffusive , or highly processive , minus end-directed movements along microtubules . RNA localisation signals stimulate the processive movements , with regulation of dynein-dynactin’s activity rather than its total copy number per RNP , responsible for this effect . Our data support a novel mechanism for multi-motor translocation based on the regulation of dynein processivity by discrete cargo-associated features . Studying the in vitro responses of RNPs to microtubule-associated proteins ( MAPs ) and microtubule ends provides insights into how an RNA population could navigate the cytoskeletal network and become anchored at its destination in cells .
Microtubule-based motility plays a major role in the distribution and sorting of organelles , vesicles , and macromolecules within cells . The significance of this process is underscored by the association of mutations in microtubule motor proteins and their co-factors with several neurological disorders ( Hirokawa et al . , 2010; Schiavo et al . , 2013 ) . Furthermore , many pathogens exploit cellular microtubule motors during infection ( Grieshaber et al . , 2003; Greber and Way , 2006; Ramsden et al . , 2007 ) . Despite the fundamental importance of the process , the mechanisms by which cargos are trafficked along microtubules are poorly understood . There is substantial evidence that individual cargos simultaneously associate with multiple microtubule motors ( Gross et al . , 2007 ) . Not only can a single cargo associate with several copies of the same kind of motor ( e . g . , Leopold et al . , 1992; Welte et al . , 1998; Hendricks et al . , 2010; Encalada et al . , 2011; Rai et al . , 2013 ) , but opposite polarity motors are often stably bound ( Ling et al . , 2004; Pilling et al . , 2006; Shubeita et al . , 2008; Soppina et al . , 2009; Hendricks et al . , 2010; Encalada et al . , 2011 ) . Thus , in order to understand cargo trafficking in vivo it is essential to learn how the activity of multiple motors is orchestrated . The mechanisms governing translocation of cargos by teams of microtubule motors have predominantly been tackled in two ways . One set of studies has analysed motility of cargo populations within cells ( e . g . , Kural et al . , 2005; Shubeita et al . , 2008; Reis et al . , 2012; Rai et al . , 2013 ) . Although physiologically relevant , the mechanistic insights that can be derived from this approach are limited by the complex in vivo environment . The second set of studies have analysed the in vitro behaviours of artificial cargos , such as beads or DNA origami , coupled to isolated motor proteins or motor domains ( e . g . , Vale et al . , 1985; Block et al . , 1990; Mallik et al . , 2005; Diehl et al . , 2006; Ross et al . , 2006; Vershinin et al . , 2007; Derr et al . , 2012; Furuta et al . , 2013 ) . These in vitro studies have provided evidence that small increases in the number of purified motors of the same polarity strongly augment the average travel distance in that direction ( Block et al . , 1990; Mallik et al . , 2005; Vershinin et al . , 2007; Derr et al . , 2012; Furuta et al . , 2013 ) . However , these experiments did not include potential regulatory co-factors that associate with motors in vivo . Thus , substantial debate persists over whether net movement of physiological cargo-motor complexes is dominated by motor copy number or by higher order mechanisms that regulate motor activity ( Gross , 2004; Kural et al . , 2005; Shubeita et al . , 2008; Elting and Spudich , 2012; Reis et al . , 2012 ) . Another important unresolved issue is how , when navigating to their destination in vivo , cargo-motor complexes cope with other proteins that decorate microtubules . In vitro experiments have shown that the plus end-directed kinesin-1 motor frequently detaches when encountering microtubule-associated proteins ( MAPs ) ( Vershinin et al . , 2007; Dixit et al . , 2008; Telley et al . , 2009; McVicker et al . , 2011 ) . In contrast , upon such an obstacle encounter , individual bidirectional dynein-dynactins often remain attached to a microtubule and undergo a reversal in travel direction ( Dixit et al . , 2008 ) . It remains to be tested in a defined system how intact transport complexes , containing cargo , multiple motors , and potential regulatory proteins , react to obstacles on their tracks . It is also not known how these complexes respond when they reach the ends of the microtubules . Each of these problems is exemplified by the trafficking of developmentally important mRNAs in the syncytial Drosophila embryo . Cytoplasmic injection of in vitro synthesised fluorescent transcripts has shed light on the mechanisms governing RNA sorting in this system . These experiments have provided evidence that apical mRNA localisation is achieved by a bidirectional translocation process in which , on average , minus end-directed transport by the multi-subunit dynein motor and its large accessory complex dynactin predominates ( Wilkie and Davis , 2001; Bullock et al . , 2006; Vendra et al . , 2007 ) . Upon reaching the apical cytoplasm , the ribonucleoprotein complexes ( RNPs ) are statically anchored by an unknown , dynein-dependent mechanism ( Delanoue and Davis , 2005 ) . mRNAs that are uniformly distributed also move bidirectionally , but with little net directional bias ( Bullock et al . , 2006; Amrute-Nayak and Bullock , 2012 ) . Intriguingly , dynein-dynactin is required for both plus end- and minus end-directed motion of the localising and uniformly distributed RNPs formed upon injection ( Bullock et al . , 2006; Vendra et al . , 2007 ) . Dynein is also needed for efficient spreading of uniformly distributed endogenous RNAs from the perinuclear region , supporting a physiological requirement for the motor complex in bidirectional RNA motion ( Bullock et al . , 2006 ) . These findings , together with the failure to detect functional evidence for the involvement of a kinesin motor ( Vendra et al . , 2007 ) , suggest that plus end movements of RNPs are driven by dynein moving in this direction , a property that has been documented in several in vitro studies of the motor ( Schliwa et al . , 1991; Wang et al . , 1995; Wang and Sheetz , 2000; Mallik et al . , 2005; Ross et al . , 2006; Miura et al . , 2010; Walter et al . , 2012 ) . Net minus end transport of apical transcripts is dependent on RNA localisation signals , which are comprised of specialised stem-loops that recruit additional dynein-dynactin complexes to RNPs through the Egalitarian ( Egl ) and Bicaudal-D ( BicD ) adaptor proteins ( Bullock et al . , 2006; Dienstbier et al . , 2009; Amrute-Nayak and Bullock , 2012 ) . Egl binds directly to the localisation signals ( Dienstbier et al . , 2009 ) and the light chain subunit of dynein ( Navarro et al . , 2004 ) , whereas BicD interacts simultaneously with Egl ( Navarro et al . , 2004; Dienstbier et al . , 2009 ) and multiple sites in the dynein-dynactin complex ( Hoogenraad et al . , 2001; Splinter et al . , 2012 ) . Egl and BicD do not appear to contribute to the binding of the dynein-dynactin complex to RNA at sites other than localisation signals ( Bullock et al . , 2006; Dix et al . , 2013 ) , and the RNA features and protein factors that fulfil this task have not been identified . Recent proteomic work by our group ( Dix et al . , 2013 ) has shown that Lissencephaly-1 ( Lis1 ) is also a component of dynein-dynactin complexes associated with localising and uniformly distributed RNAs . Lis1 promotes the recruitment of dynein-dynactin to RNAs ( Dix et al . , 2013 ) and may also regulate mechanochemistry of the cargo-associated motor ( McKenney et al . , 2010; Huang et al . , 2012; Vallee et al . , 2012 ) . The study of Dix et al . supported the existence of a core functional complex recruited to localisation signals , consisting of Egl , BicD , dynein-dynactin , and Lis1 ( Dix et al . , 2013 ) . However , it is not known whether the dynein-dynactin recruited in this manner is more likely to engage in minus end-directed motion than that recruited elsewhere in the RNA . Alternatively , the localisation signals could drive net minus end motion simply by recruiting more copies of functionally equivalent dynein-dynactin complexes per RNP . In order to begin to address these mechanistic issues , we have developed a novel in vitro RNA motility assay that combines the manipulability of a cell-free system with the physiological relevance of cargo-motor complexes assembled from Drosophila embryo extract . We have used the unique advantages of this system to examine the mechanism of directionally biased motility by multi-motor assemblies , the response to potential obstacles , and the consequences of reaching microtubule ends .
Optical limitations of the Drosophila embryo preclude direct visualisation of RNPs moving along individual microtubules . We therefore previously established an in vitro assay for mRNA motility by capturing RNA-motor complexes from embryo extract through the affinity of motors for microtubules ( Amrute-Nayak and Bullock , 2012 ) . However , detailed characterisation of RNA motility was challenging due to the limited number of motile RNPs in each imaging chamber . We therefore developed RAT-TRAP ( RNA Transport After Tethered RNA Affinity Purification ) , a method to efficiently study RNP motility in an in vitro setting . RNA-motor complexes were assembled by incubating embryo extract with in vitro transcribed RNAs immobilised on streptavidin-coated beads via a streptavidin-binding RNA aptamer ( Figure 1A ) . The RNAs were labelled by the stochastic incorporation of fluorophore-coupled UTP during the transcription reaction , permitting downstream visualisation by fluorescence microscopy . The assembled RNA-motor complexes were washed briefly and eluted from the beads with biotin , which competes for the interaction of the aptamer with streptavidin . This methodology leads to the recruitment of dynein-dynactin complexes to RNA localisation signals in association with Egl , BicD , and Lis1 ( Dix et al . , 2013 ) . To assay in vitro RNP motility , the eluate was supplemented with a saturating level of ATP and added to a chamber containing polarity-marked , GmpCpp-stabilised microtubules bound to a coverslip ( Figure 1A ) . RNAs and microtubules were then visualised with total internal reflection ( TIR ) microscopy . 10 . 7554/eLife . 01596 . 003Figure 1 . A novel in vitro assay to study mRNA motility . ( A ) Schematic of assay; see text for details . SA , streptavidin; Aptamer , streptavidin-binding RNA aptamer . ( B and C ) Stills generated from time-lapse series of motile unidirectional ( B ) and bidirectional ( C ) Cy3-h wild-type ( hWT ) RNPs ( magenta ) . Double-headed arrow indicates orientation of microtubule ( − and + , minus and plus end ) ; plus end ( black arrowhead above top still ) is marked by greater incorporation of HiLyte 647-tubulin ( green ) . ( D and E ) Kymographs ( time-distance plots ) of examples of unidirectional ( D ) and bidirectional ( E ) RNPs; t , time . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 00310 . 7554/eLife . 01596 . 004Figure 1—figure supplement 1 . Assessing RNA copy number in hWT RNPs and tracking accuracy in the RAT-TRAP assay . ( A and B ) Motile RNPs assembled in the presence of equimolar Alexa488 ( A488 ) -labelled h and Cy3-labelled h contained only A488 or Cy3 dyes , but never both . Thus , single fluorescent h molecules were present in each RNP . Kymographs of two examples of RNPs ( A ) and overall quantification from the experimental series ( B ) are shown . d , distance; − , minus end; + , plus end; N , number of motile RNPs analysed . For the experiments in A and B the Alexa488-labelled RNAs and Cy3-labelled RNAs were mixed and incubated with the streptavidin bead matrix before the addition of extract . ( C ) Representative kymograph of h RNPs in the presence of 20 U·ml−1 apyrase ( which is used to deplete ATP and ADP from the chamber [Higuchi et al . , 1997; Ma and Taylor , 1997] ) . Note that RNPs appear static . ( D ) Distribution of instantaneous frame-to-frame displacements generated from automatic tracking of h RNPs on microtubules in the presence of 20 U·ml−1 apyrase . In accordance with previous studies ( Hendricks et al . , 2010 ) , our tracking accuracy was defined as one standard deviation of these displacements . Dashed line shows Gaussian fit from which standard deviation was calculated . R2 , goodness of fit; N , total number of individual displacements analysed ( from 15 RNPs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 004 Our initial experiments focused on the well-characterised hairy ( h ) mRNA , which requires a 124 nt localisation signal ( the h localisation element [HLE] ) in its ∼800 nt 3′-untranslated region ( UTR ) to localise apically in the Drosophila embryo ( Bullock et al . , 2003 ) . Embryo extracts were incubated with immobilised h wild-type 3′ UTR RNAs ( hWT ) , which contained an average of ∼8 Cy3 dyes per molecule . 50–80 microtubule-associated RNPs were typically observed following injection of the eluate into an imaging chamber , with ∼60–70% of these displaying motility during the ∼100 s of data acquisition . There was a strong overall minus end bias to motility of the hWT RNP population , as is the case in vivo ( Bullock et al . , 2003 , 2006 ) . 24 ± 1% of motile RNPs analysed per chamber moved exclusively unidirectionally towards the minus ends of microtubules ( mean ± SEM , 10 chambers; e . g . , Figure 1B , D; Video 1 ) . These movements were highly processive , often ending when RNPs paused or detached upon reaching the microtubule minus end . The remainder of motile hWT RNPs moved bidirectionally , with frequent switches in direction ( e . g . , Figure 1C , E; Video 2 ) and no overt directional bias at the population level ( see below for quantification ) . 10 . 7554/eLife . 01596 . 005Video 1 . Example of unidirectional motion of a Cy3-hWT RNP towards the minus end of a polarity-marked microtubule . RNA , pseudocoloured magenta; microtubule , pseudocoloured green . Microtubule plus end is labelled with a highly fluorescent segment . This RNP pauses upon reaching the minus end of the microtubule; we also observed instances of dissociation of unidirectional RNPs upon reaching the minus end ( Figure 6D , E ) . Video corresponds to 12 . 4 s; width of the image is 16 . 7 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 00510 . 7554/eLife . 01596 . 006Video 2 . Example of bidirectional motion of a Cy3-hWT RNP on an individual microtubule . RNA , pseudocoloured magenta; microtubule , pseudocoloured green . Microtubule plus end located outside of field-of-view ( towards the right ) . Video corresponds to 62 . 5 s; width of the image is 9 . 6 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 006 We have recently shown that , unlike the large RNPs formed after injection of fluorescent RNA into the embryo , endogenous h RNPs contain a single copy of the RNA molecule ( Amrute-Nayak and Bullock , 2012 ) . To assess the copy number of h RNA per RNP in the RAT-TRAP assay , exclusively Alexa488-labelled hWT RNAs and exclusively Cy3-labelled hWT RNAs were incubated simultaneously with the beads before the addition of extract . All motile RNPs assembled in such experiments contained only one type of fluorophore ( Figure 1—figure supplement 1A , B ) . These data indicate that , as was the case for our previous in vitro assay for RNA motility ( Amrute-Nayak and Bullock , 2012 ) , the RAT-TRAP assay reports on the motility of RNPs containing a single fluorescent RNA molecule . The number of motile hWT RNPs observed in RAT-TRAP assays was ∼10-fold greater than that observed using our previous method ( [Amrute-Nayak and Bullock , 2012] and data not shown ) . Furthermore , functionalisation of glass surfaces with polyethylene glycol greatly reduced non-specific binding of fluorescent RNPs to the coverslip in the new assay and thereby allowed the implementation of automatic , sub-pixel tracking of microtubule-associated complexes ( Figure 1—figure supplement 1C , D ) with high temporal precision ( 15 frames per second [fps] ) . Our previous assay ( Amrute-Nayak and Bullock , 2012 ) used manual analysis of kymographs generated from images captured at a much lower frame rate ( ∼3 fps ) . Thus , the RAT-TRAP assay constitutes a substantial step forward in the ability to analyse the movement of RNPs in vitro . To investigate the global properties of unidirectional and bidirectional motion , we analysed the relationship of mean square displacement ( MSD ) of tracked RNP trajectories with time . As expected , unidirectional hWT RNP trajectories showed a quadratic dependence of mean square displacement ( MSD ) over time ( average slope of ∼2 . 0 in a log–log plot [Figure 2A]; see Figure 2—figure supplement 1A , C for additional MSD analysis ) . This observation indicates an underlying active transport process ( Saxton , 1997; Sanchez et al . , 2012 ) . In contrast , the relationship of MSD of bidirectional RNPs over time was linear ( average slope of ∼1 . 0 in the log–log plot ( Figure 2A; Figure 2—figure supplement 1B , C ) . Thus , there appears to be a strong diffusive component to the movement of bidirectional hWT RNPs ( Saxton , 1997; Sanchez et al . , 2012 ) . 10 . 7554/eLife . 01596 . 007Figure 2 . Characterisation of the motile properties of hWT RNPs . ( A ) Mean square displacement ( MSD ) of hWT RNP trajectories as a function of time plotted in a log–log format . Mean slopes ( ± SEM ) were calculated from a linear fit to the data . Plot of MSD vs time on non-logarithmic axes is shown in Figure 2—figure supplement 1A , B . ATP , 2 . 5 mM ATP; ATP-vanadate , 2 . 5 mM ATP plus 100 μM vanadate . N , number of RNPs analysed . See Figure 2—figure supplement 1C for slopes of log–log MSD ( t ) for individual RNPs in each population . ( B–E ) Correlation analysis of mean run length and run velocity for the bidirectional subset of hWT RNPs . Strong correlations exist for individual RNPs between ( B ) mean minus end and mean plus end run length , ( C ) mean minus end and mean plus end velocity , ( D ) mean minus end run length and mean minus end velocity , and ( E ) mean plus end run length and mean plus end velocity . Only bidirectional RNPs with ≥20 runs in total were used for these analyses ( note that no such cut-off was applied for the analysis in H and I ) . R2 , correlation coefficient; N , number of RNPs analysed; m , slope . There is no significant bias in minus end vs plus end motile properties in B and C ( red line represents slope expected for no bias ) . ( F ) Mean number of hWT RNPs bound per μm of microtubule per movie in the presence of 2 . 5 mM ATP , 20 U·ml−1 apyrase and 2 . 5 mM ATP plus 100 μM vanadate ( ATP-vanadate ) . Means were calculate from values for 12 microtubules selected at random in at least three imaging chambers . ( G ) Mean percentage of microtubule-associated hWT RNPs that were unidirectional , bidirectional , or stationary . Means were calculated from 12 microtubules as in F . ( H and I ) Mean run lengths ( H ) and velocities ( I ) of individual runs of bidirectional hWT RNPs . N , number of individual runs of RNPs ( number of RNPs from which the individual runs were extracted was 40 for ATP and 20 for ATP-vanadate ) . See Figure 2—figure supplement 2C , D for distribution of run lengths and velocities . Errors represent SEM in all panels . In F–I , ***p<0 . 001; **p<0 . 01; *p<0 . 05 , compared to the ATP condition for the same parameter ( Mann–Whitney non-parametric t test ) . Images for all analyses in the figure were acquired at 15 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 00710 . 7554/eLife . 01596 . 008Figure 2—figure supplement 1 . Supplementary MSD analysis of hWT RNP trajectories . ( A and B ) Plots of average mean square displacement ( MSD ) vs time for hWT RNPs . See Figure 2A for data plotted on logarithmic axes . ( A ) For unidirectional RNPs there is a quadratic relationship of MSD with time , indicating a deterministic transport process . ( B ) For bidirectional RNPs there is a linear relationship of MSD with time , indicative of a diffusive process . The linear relationship held over all time scales analysed; for example , MSD ( t ) also appeared linear between 0 and 0 . 3 s ( data not shown ) . MSD ( t ) is similar in the presence of both ATP and ATP-vanadate . D , diffusion coefficients of bidirectional RNPs on the microtubule ( calculated from the slopes of the linear fits in B using MSD = 2Dt ) . In A and B , MSD traces were obtained by internal averaging , with the time interval corresponding to one quarter of the total duration of the shortest-lived RNP trajectory analysed ( Saxton , 1997 ) . N , number of RNPs analysed . ( C ) Scatter plot showing the slope of the log–log plot of MSD vs time for individual RNPs obtained by internal averaging , with the time interval for each RNP corresponding to one quarter of the total duration of its trajectory ( Saxton , 1997 ) . Long and short horizontal red lines demarcate the mean and SEM , respectively . Slopes of ∼1 . 0 or ∼2 . 0 are expected for diffusive motion and active transport , respectively . We do not find a significant subset of bidirectional RNPs with a slope of ∼1 . 5 ( such a value has been interpreted as indicative of diffusive motion interspersed with bouts of processive movement [Hendricks et al . , 2010; Sanchez et al . , 2012] ) . Images for all MSD analyses were acquired at 15 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 00810 . 7554/eLife . 01596 . 009Figure 2—figure supplement 2 . Distributions of run lengths and velocities of unidirectional and bidirectional hWT RNPs . ( A and B ) Distribution of lengths ( A ) and velocities ( B ) of individual runs of unidirectional minus end-directed hWT RNPs . Note that no plus end-directed unidirectional runs were observed . N , number of runs ( from 25 RNPs [many RNPs have more than one run due to interruptions of bouts of minus end-directed motility by short-lived pauses] ) . ( C and D ) Distribution of lengths ( C ) and velocities ( D ) of individual runs of bidirectional hWT RNPs in the minus end or plus end direction in the presence of 2 . 5 mM ATP or 2 . 5 mM ATP plus 100 μM vanadate ( ATP-vanadate ) . N , number of runs ( from 40 RNPs for ATP and 25 RNPs for ATP-vanadate ) . Note that due to the relative paucity of long or fast runs , those runs >500 nm or >2000 nm·s−1 were binned together in these plots . The maximum run lengths were: minus end , 2190 nm; plus end , 2008 nm . The maximum run velocities were: minus end , 5270 nm·s−1; plus end , 5800 nm·s−1 . There was a general tendency for individual runs that were long to also be fast ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 009 We next extracted runs from RNP trajectories and analysed their length and velocity . A run was operationally defined as a bout of uninterrupted motion in one direction before a reversal , pause , detachment from the microtubule or curtailment by the end of image acquisition . The mean length and velocity of individual runs of unidirectional hWT RNPs were ∼4 μm and ∼1 . 2 μm . s−1 , respectively ( Figure 2—figure supplement 2A , B ) . The mean velocity of individual runs of bidirectional hWT RNPs in both directions ( ∼0 . 8 μm . s−1 ) was similar to that of the unidirectional counterparts , whereas the mean run lengths were much shorter ( Figure 2—figure supplement 2C , D ) . In both directions the mean length of individual runs of bidirectional RNPs was ∼120 nm; only a small fraction of runs were longer than 500 nm , with the longest ∼2 μm ( Figure 2—figure supplement 2C and accompanying legend ) . These run lengths are much shorter than those previously documented for force-generating dynein and kinesin motors bound to other bidirectional cargos ( Shubeita et al . , 2008; Soppina et al . , 2009; Schuster et al . , 2011; Reis et al . , 2012; Rai et al . , 2013 ) . Analysis of the mean run length and velocities of individual bidirectional hWT RNPs revealed a great deal of variation for each of these values per RNP ( Figure 2B–E ) . This presumably reflects compositional heterogeneity , a feature that has been observed in populations of cargo-motor complexes in vivo ( Encalada et al . , 2011; Schuster et al . , 2011 ) . Despite their variation across the population , the quantitative parameters of motion were highly correlated to each other for individual RNPs ( Figure 2B–E ) . The mean minus end and plus end run lengths were very similar for any given RNP ( Figure 2B ) . Individual RNPs also exhibited mean velocities that were highly similar in both the minus end and plus end direction ( Figure 2C ) . Mean run lengths and run velocities for individual RNPs were also highly correlated ( Figure 2D , E ) . Thus , complexes that move faster on average tended to move further before a run ends . Overall , our correlation analysis reveals a high degree of coupling between motile properties in each direction for individual RNPs . We next studied the effects of alternate nucleotide states on the motility of RNPs . Depletion of ATP and ADP with apyrase ( Higuchi et al . , 1997; Ma and Taylor , 1997 ) led to a significant increase in the number of microtubule-associated RNPs compared to the ATP condition ( Figure 2F ) . In the presence of apyrase there was also a large increase in the proportion of microtubule-bound RNPs that were stationary , with unidirectional behaviour absent and a strong reduction in the proportion of bidirectional complexes ( Figure 2G ) . The affect of apyrase on unidirectional motion is entirely expected as processive movement of dynein is dependent on ATP hydrolysis ( Roberts et al . , 2013 ) . The inhibition of bidirectional motion by apyrase could conceivably be explained by dynein normally stepping towards both the minus and plus end of microtubules through ATP hydrolysis—as has been documented in two previous studies of purified motor complexes ( Ross et al . , 2006; Walter et al . , 2012 ) —with frequent reversals between bouts of motion in each direction . Alternatively , RNPs could undergo passive diffusion along microtubules in the presence of ATP , with the presence of the motor on RNPs stimulating long-term arrest in the absence of ATP due to its tight microtubule binding in the no nucleotide state ( Roberts et al . , 2013 ) . Both these scenarios , or a combination of the two , could account for the relatively short mean run lengths , diffusive MSD properties , and the tight coupling of minus and plus end motile properties for individual bidirectional RNPs in the presence of ATP . To discriminate between these possibilities we performed motility assays in the presence of both ATP and vanadate . Vanadate inhibits the ATPase activity of dynein and mimics the ADP·Pi state , which is associated with weak affinity for microtubules ( Shimizu and Johnson , 1983; Miura et al . , 2010 ) . There was a partial reduction in the number of RNPs bound to microtubules in the presence of vanadate ( Figure 2F ) . As expected , vanadate abolished unidirectional motion of the microtubule-associated RNPs ( Figure 2G ) . Tellingly , the percentage of microtubule-associated RNPs that underwent bidirectional motion was not decreased in the presence of ATP-vanadate compared to ATP alone ( Figure 2G ) . In fact , in the presence of vanadate there was an increase in the proportion of microtubule-associated RNPs that were bidirectional , possibly due to a contribution of RNPs that would otherwise be unidirectional undergoing switching to a bidirectional state . The relationship of MSD vs time was very similar for the bidirectional RNPs in the presence of ATP-vanadate and ATP ( Figure 2A , Figure 2—figure supplement 1B , C ) . Similar distributions of lengths and velocities or individual runs in both the minus and plus end direction were also observed for bidirectional RNPs in both conditions ( Figure 2—figure supplement 2C , D ) . Thus , even the occurrence of relatively long runs of hWT RNPs was not prevented by vanadate ( Figure 2—figure supplement 2C , D ) . Vanadate actually caused a subtle but statistically significant increase in the mean run length and velocity of bidirectional RNPs in the minus and plus end directions ( Figure 2H , I ) , implying a modulation of the diffusive properties of RNPs when dynein is in the ADP . Pi state . Overall , our results demonstrate that an active energy transduction property of dynein is required for unidirectional , minus end-directed RNP motion but not for motion of bidirectional RNPs in either the minus or plus end direction . Thus , individual hWT RNPs can adopt two discrete behaviours in vitro: unidirectional motion , driven by processive movement of dynein that is associated with ATP hydrolysis and bidirectional motion that appears to be dominated by passive diffusion . As described in the introduction , there is substantial debate on the influence of total motor copy number on the translocation of cargos . To assess the consequences of copy number of native motor complexes on the motion of a physiological cargo in vitro we increased or decreased the number of native dynein-dynactin complexes per h RNP . This was achieved by manipulating the HLE RNA sequences that are naturally responsible for their recruitment . The key feature of the HLE is stem-loop 1 ( SL1 ) ( Bullock et al . , 2003 ) , which recruits dynein-dynactin through the adaptor proteins Egl and BicD ( Dienstbier et al . , 2009; Dix et al . , 2013 ) . We therefore generated Cy3-labelled RNAs that contained fusions of the streptavidin-binding aptamer to a h 3′UTR in which the entire HLE is replaced by three copies of the SL1 element ( hSL1x3 ) or a heterologous sequence from the glutathione-S-transferase ( GST ) RNA ( hΔLE ) ( Figure 3A ) . 10 . 7554/eLife . 01596 . 010Figure 3 . Manipulating the copy number of native dynein-dynactin complexes on individual RNPs . ( A ) Schematic of h RNA variants fused to streptavidin aptamers and used in this study . hΔLE and hSL1x3 have replacements of the 124 nt HLE region with , respectively , a heterologous sequence from the Glutathione-S-transferase ( GST ) gene and three copies of stem-loop 1 ( SL1 ) separated by short spacers . ( B ) Example of an RNP in vitro containing Cy3-h RNA ( magenta ) and GFP::Dlic ( green ) , which is immobilised on a polarity-marked microtubule ( cyan ) by the omission of ATP . ( C ) Stepwise GFP photobleaching analysis of RNPs assembled on h RNA variants . Note significant relative changes in the copy number of Dlic and Dmn upon increasing or decreasing copy number of SL1 . N , number of photobleaching traces analysed . See Figure 3—figure supplement 1B for distribution of values . ( D ) Mean run length per RNP for each RNA species . N , number of RNPs analysed . ( E ) Mean proportion of motile RNPs per imaging chamber that are unidirectional or bidirectional . No unidirectional RNPs were observed for hΔLE . N , number of chambers analysed; n , total number of RNPs . ( F and G ) Mean length ( F ) and velocity ( G ) of individual runs of unidirectional RNPs . N , number of runs ( from 25 and 20 RNPs for hWT and hSL1x3 , respectively [many RNPs have more than one run due to interruptions of bouts of minus end-directed motility by short-lived pauses] ) . ( H and I ) Mean length ( H ) and velocity ( I ) of individual runs of bidirectional RNPs . N , number of runs ( from 40 RNPs each for hWT , hΔLE , and hSL1x3 ) . In D–I , all experiments were performed in 2 . 5 mM ATP; error bars represent SEM . ***p<0 . 001; **p<0 . 01; *p<0 . 05 , compared to hWT values for the same parameter ( Mann–Whitney non-parametric t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 01010 . 7554/eLife . 01596 . 011Figure 3—figure supplement 1 . Supplemental data on GFP photobleaching analysis of relative dynein and dynactin copy number per hWT , hΔLE , or hSL1x3 RNP . ( A ) Examples of traces of GFP fluorescent signals ( after background subtraction ) from individual hWT RNPs assembled in the presence of tagged Dlic ( i , ii ) or Dmn ( iii , iv ) . Traces show stepwise decreases in fluorescence , indicative of photobleaching of single GFP molecules . ( i ) and ( iii ) are the raw traces , with ( ii ) and ( iv ) showing fitting with the Stepfinder algorithm ( Kerssemakers et al . , 2006 ) . ( B ) Distributions and means ( ± SEM ) of number of GFP decay steps for GFP::Dlic ( top ) and GFP::Dmn ( bottom ) for RNPs associated with hΔLE , hWT , or hSL1x3 RNAs . ( C ) Distributions and means ( ± SEM ) of total GFP fluorescence at the beginning of imaging ( after background subtraction ) for GFP::Dlic ( top ) and GFP::Dmn ( bottom ) for RNPs associated with hΔLE , hWT or hSL1x3 RNAs . In B and C , means ± SEM were determined from raw data and N represents the number of RNPs analysed . ***p<0 . 001 , compared to hWT values ( Mann–Whitney non-parametric t test ) . Due to differences in fluorescence illumination in the TIR field in x , y , and z ( Axelrod , 1989; Stout and Axelrod , 1989 ) , measurements of total GFP fluorescence are likely to be less accurate than quantification of the number of stepwise decay events . However , the total GFP signal measurements provide further evidence for alterations in Dlic and Dmn copy number following manipulation of the number of SL1 elements . Note that the maximum number of dynein motors recruited by one localisation element and the proportion of recruited motors that are actively engaged during RNP transport is unknown . ( D ) Fluorescent Western Blot with α-Dlic antibodies revealing the levels of GFP-labelled Dlic compared to unlabelled , endogenous Dlic in the GFP::Dlic extracts used for photobleaching analysis . Predicted molecular weights of Dlic and GFP::Dlic are 54 . 5 kDa and 81 . 4 kDa , respectively . The mean percentage of total Dlic that was GFP-labelled was 0 . 49 ± 0 . 01 ( mean ± SEM ) . Note that we previously showed that GFP::Dlic is incorporated in microtubule-associated motor complexes in accordance with its abundance in extracts relative to unlabelled , endogenously expressed Dlic ( Amrute-Nayak and Bullock , 2012 ) . Estimation of the ratio of GFP::Dmn to endogenous Dmn could not be made due to the unavailability of previously published α-Drosophila Dmn antibodies . ( E ) Table illustrating the calculations used to estimate dynein copy number per RNP based on stepwise photobleaching . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 01110 . 7554/eLife . 01596 . 012Figure 3—figure supplement 2 . Distributions of run lengths and velocities of hSL1x3 and hΔLE RNPs . ( A and B ) Distribution of lengths ( A ) and velocities ( B ) of individual runs of unidirectional minus end-directed hSL1x3 RNPs . Note that no plus end-directed unidirectional runs were observed . N , number of runs ( from 20 RNPs [individual RNPs often have more than one run due to interruptions by short-lived pauses] ) . See Figure 3F , G for comparison of means with those of hWT RNPs . ( C and D ) Distribution of lengths ( C ) and velocities ( D ) of individual runs of bidirectional hSL1x3 and hΔLE RNPs in the minus end or plus end direction compared to hWT RNPs . N , number of runs ( from 40 RNPs ) . Note that due to the relative paucity of long or fast runs , those >500 nm or >2000 nm·s−1 were binned together in this plot . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 012 We used stepwise GFP photobleaching to confirm that the number of dynein and dynactin subunits associated with individual RNPs increased with increasing numbers of SL1 elements per RNA molecule . Photobleaching analysis was performed on motor complexes that were assembled on the h RNA variants using embryo extracts expressing GFP-tagged versions of Dynein light intermediate chain ( GFP::Dlic ) or the dynactin subunit p50-Dynamitin ( GFP::Dmn ) and stably bound to microtubules in the absence of nucleotide ( Amrute-Nayak and Bullock , 2012 ) ( Figure 3A–C , Figure 3—figure supplement 1A–C ) . Previous experiments had shown that GFP::Dlic is incorporated into dynein-dynactin complexes in accordance with its abundance in embryo extract relative to endogenous Dlic ( Amrute-Nayak and Bullock , 2012 ) . Thus , by using fluorescent immunoblotting to determine the proportion of total Dlic in the extract that was labelled with GFP ( Figure 3—figure supplement 1D ) , we could approximate the mean number of Dlic copies on each of the h RNA variants ( Figure 3—figure supplement 1E ) . The estimated means ± SEM of Dlic copies for hWT , hΔLE , and hSL1x3 RNPs were , respectively , 6 . 73 ± 0 . 44 , 4 . 36 ± 0 . 33 , and 13 . 85 ± 0 . 67 . It has been reported that there are two Dlic molecules per dynein motor ( King et al . , 1996; Trokter et al . , 2012 ) and hence our estimates of Dlic copy number on h RNA variants are consistent with 3 . 36 ± 0 . 21 , 2 . 18 ± 0 . 16 , and 6 . 92 ± 0 . 33 dynein motors associated with hWT , hΔLE , or hSL1x3 RNPs , respectively . We wish to stress that these values are only estimates of the average dynein copy number on each RNA due to limitations of the photobleaching technique in determining precise copy number of fluorescent molecules ( Das et al . , 2007; Shu et al . , 2007 ) . Nonetheless , our analysis clearly reveals that manipulating SL1 copy number in the different h RNA variants causes significant changes , in relative terms , in the average copy number of dynein and dynactin complexes per RNP . We next analysed the motile behaviour of the different h RNA variants in the RAT-TRAP assay in the presence of ATP . The population of hSL1x3 RNPs had a significantly greater minus end-directed bias than the hWT RNPs ( Figure 3D ) . This was associated with a doubling in the proportion of motile hSL1x3 RNPs undergoing exclusively unidirectional movement towards the minus end of microtubules compared to hWT RNPs ( Figure 3E ) , without substantially altering the mean run lengths and velocities of these events ( Figure 3F , G , Figure 3—figure supplement 2A , B ) . Reducing the copy number of dynein-dynactin per RNP by using the hΔLE RNA resulted in exclusively bidirectional motion ( Figure 3E ) with no significant net directional bias ( Figure 3D ) . Thus , there was a correlation between the probability of unidirectional minus end-directed motion of RNPs and the average number of associated dynein-dynactin complexes . The distribution of run lengths and velocities of the bidirectional RNPs in the minus and plus end directions were similar for hWT , hSL1x3 , and hΔLE RNAs ( Figure 3H , I , Figure 3—figure supplement 2C , D ) . There was , however , a subtle increase in both minus and plus end mean run lengths and velocities with increasing numbers of SL1 elements ( Figure 3H , I , Figure 3—figure supplement 2C , D ) . These data suggest that increasing dynein-dynactin copy number through localisation signals does not have a major influence on the motile properties of bidirectional RNPs . However , the correlation between increased dynein-dynactin number and increased lengths and velocities of runs in both the minus and plus end direction is consistent with this complex playing a direct role in mediating bidirectional movement of RNPs along microtubules . The experiments described above with the hΔLE , hWT , and hSL1x3 RNAs show that the major mechanism through which increasing the number of RNA localisation signals promotes net minus end-directed motion is by increasing the probability of long , unidirectional transport in this direction . Consistent with previous observations ( Bullock et al . , 2006; Amrute-Nayak and Bullock , 2012; Dix et al . , 2013 ) , we find that RNAs lacking localisation signals still recruit dynein-dynactin but move exclusively bidirectionally ( Figure 3C , E ) . Thus , localisation signals may promote unidirectional motion simply by increasing the average copy number of dynein-dynactin complexes bound per RNP . Such a model is consistent with previous work showing that individual purified dynein-dynactin complexes move bidirectionally in vitro , with additional dynein-dynactin complexes bound to a bead driving unidirectional motion in the minus end direction ( Ross et al . , 2006 ) . Alternatively , dynein-dynactins recruited to localisation signals may be more likely to undergo unidirectional motion than those bound elsewhere in the RNA . To discriminate between these possibilities we performed RAT-TRAP assays with Cy3-labelled HLE RNA , which lacks the dynein-dynactin binding sites provided by the rest of the h 3′UTR sequences . Importantly , we first confirmed using stepwise photobleaching with GFP-Dlic extracts that there is a significant reduction in the relative copy number of the motor complex on HLE RNPs compared to hWT RNPs ( Figure 4A , Figure 4—figure supplement 1A , B ) . The average number of GFP::Dlic photobleaching steps on the HLE was in fact statistically indistinguishable from that observed for hΔLE ( 2 . 21 ± 0 . 15 and 2 . 14 ± 0 . 16 , respectively; Figure 4A and Figure 3C ) . 10 . 7554/eLife . 01596 . 013Figure 4 . The HLE alone promotes unidirectional motion . ( A ) Stepwise GFP photobleaching analysis of RNPs assembled on hWT or HLE RNAs; inset , schematic of HLE RNA that was fused to the aptamer sequence . Note the significant decrease in the relative copy number of GFP::Dlic on the HLE compared to hWT . N , number of photobleaching traces analysed . See Figure 4—figure supplement 1A for distribution of values . ( B ) Mean proportion of motile hWT or HLE RNPs that are unidirectional per imaging chamber . N , number of chambers analysed; n , total number of RNPs . ( C ) Mean run length per RNP of motile hWT or HLE RNPs . N , number of RNPs analysed . ( D and E ) Mean length ( D ) and velocity ( E ) of individual runs of unidirectional RNPs . N , number of individual runs of RNPs ( from 25 RNPs each for hWT and HLE ) . ( F and G ) Mean length ( F ) and velocity ( G ) of individual runs of bidirectional RNPs ( from 40 and 20 RNPs for hWT and HLE , respectively ) . Images for all analyses in the figure were acquired at a lower frame rate of 4 . 2 fps; this is because the small number of Cy3 dyes that could be incorporated into the short HLE RNA necessitated imaging with a higher exposure time . Note that the measured mean run lengths of unidirectional and bidirectional RNPs of the same species are higher at 4 . 2 fps than at 15 fps ( Figure 3F , H ) , presumably due to short pauses and frequent reversals ( in the case of bidirectional RNPs ) being missed at the lower frame rate . Velocity of measured runs of bidirectional RNPs of the same species is lower at 4 . 2 fps than at 15 fps , presumably for the same reason , whereas velocity of unidirectional runs is not significantly affected by the different frame rate as short pauses have little affect on the velocity measured for such long runs ( Figure 3G , I ) . ***p<0 . 001; **p<0 . 01; *p<0 . 05 ( Mann Whitney non-parametric t test ) , compared to hWT values for the same parameter; error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 01310 . 7554/eLife . 01596 . 014Figure 4—figure supplement 1 . Supplemental data on the HLE’s recruitment of dynein and motile properties . ( A ) Distributions and means ( ± SEM ) of number of GFP decay steps for GFP::Dlic associated with HLE RNPs , compared to data for hWT RNPs that were acquired in experiments performed in parallel . ***p<0 . 001 , compared to hWT value ( Mann–Whitney non-parametric t test ) . ( B ) Table illustrating the calculations used to estimate dynein copy number per RNP based on stepwise photobleaching . ( C and D ) Distribution of lengths ( C ) and velocities ( D ) of individual runs of unidirectional minus end-directed HLE RNPs , compared to those of hWT RNPs imaged at the same frame rate . Note that no plus end-directed unidirectional runs were observed . N , number of runs ( from 25 RNPs [many RNPs have more than one run due to interruptions of bouts of minus end-directed motility by short-lived pauses] ) . ( E and F ) Distribution of lengths ( E ) and velocities ( F ) of individual runs of bidirectional HLE RNPs in the minus end or plus end direction compared to hWT RNPs imaged at the same frame rate . N , number of runs ( from 40 and 20 RNPs for hWT and HLE , respectively ) . Note that due to the relative paucity of long or fast runs , those >800 nm or >1000 nm·s−1 were binned together in these plots . Data were acquired at 4 . 2 fps for these and other experiments involving the HLE , which alters the measured run lengths and velocities for hWT RNPs compared to those obtained from 15 fps imaging ( see Figure 4 legends for further details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 014 If the total copy number of the dynein complex is the sole determinant of unidirectional transport , one would expect HLE RNPs to be less likely than hWT RNPs to exhibit this behaviour , possibly exhibiting exclusively bidirectional motion as is the case for hΔLE . However , this was not the case . Approximately 65% of RNPs containing the HLE RNA exhibited unidirectional motion towards the minus end of microtubules , compared to 24% of those containing the hWT RNA ( Figure 4B ) . Consequently , the HLE RNP population had a particularly strong minus end directional bias ( Figure 4C ) . These data indicate that total dynein-dynactin copy number is not the primary determinant of unidirectional RNP movement . Instead the HLE promotes the ability of the associated dynein-dynactin to undergo unidirectional transport , with features in the rest of the 3′UTR seemingly antagonising this effect . The mean run lengths and velocities of unidirectional runs of HLE RNPs were indistinguishable from those of hWT RNPs ( ( Figure 4D , E , Figure 4—figure supplement 1C , D ) ; note that in this series of experiments , and others involving the HLE , the image exposure time was increased ( acquisition rate , 4 . 2 fps ) to compensate for the reduced number of Cy3 dyes incorporated into the shorter RNA . Because more pauses and reversals are missed at the lower frame rate , there are differences in the absolute values of measured lengths and velocities of runs for the hWT RNPs compared to analysis at 15 fps , Figure 4 legends ) . The comparison of unidirectional runs of HLE and hWT RNPs , together with the earlier evaluation of unidirectional motility of hSL1x3 and hWT RNPs ( Figure 3F , G ) , demonstrates that once initiated the unidirectional mode of transport is not strongly influenced by the average number of cargo-associated dynein-dynactins . The bidirectional HLE RNPs exhibited similar mean lengths and velocities of individual runs in both the minus and plus end directions and these were no greater than those of the bidirectional subset of hWT RNPs ( Figure 4F , G , Figure 4—figure supplement 1E , F ) . Collectively , our results suggest that the dynein-dynactin recruited specifically by RNA localisation signals determines the likelihood of RNPs initiating an exclusively unidirectional , minus end-directed mode of transport but is not sufficient to cause an increase in the length or velocity of excursions of bidirectional RNPs in either direction . The results documented above shed light on how intrinsic properties of RNPs influence motility along the microtubule . Cargos moving along microtubules in vivo also have to contend with extrinsic factors , including MAPs that can act as obstacles . We were therefore interested in how RNPs assembled in the RAT-TRAP assay would react to the presence of MAPs and how this behaviour is influenced by manipulating dynein-dynactin copy number using the different h RNA variants . We decorated microtubules with two different recombinant , fluorescent MAPs and analysed the consequences on the motility of Cy3-labelled RNPs . The first MAP was a GFP-labelled Drosophila kinesin-1 with a rigor mutation ( Kin401T99N:mGFP ) that prevents its movement along the microtubule ( Telley et al . , 2009 ) . The second was a human Alexa488-labelled tau ( isoform 23 , also known as 0N3R ) ( Dixit et al . , 2008; McVicker et al . , 2011 ) , most of which undergoes diffusive movement on microtubules ( Hinrichs et al . , 2012 ) ( Figure 5—figure supplement 1A–C ) . For both MAPs , stepwise photobleaching analysis of static , microtubule-associated puncta estimated an average of 4–5 monomers per diffraction-limited spot ( Figure 5—figure supplement 1D ) . The presence of kin401T99N:mGFP on microtubules led to a concentration-dependent decrease in the net minus end directional bias of the hWT RNP population ( Figure 5A ) . The presence of tau23 on the microtubules also reduced the overall minus end bias to hWT RNP motion ( Figure 5A ) . This effect of the MAPs was not associated with a significant reduction in the time that hWT RNPs were detected in association with microtubules ( Figure 5B ) . Thus , unlike the case for the wild-type kinesin-1 motor under similar conditions ( Dixit et al . , 2008; Telley et al . , 2009; McVicker et al . , 2011 ) , hWT RNPs did not frequently detach from microtubules bound by kin401T99N:mGFP or tau23 . Thus , the presence of these MAPs appears to alter the motile properties of the RNPs without displacing them from the microtubule . 10 . 7554/eLife . 01596 . 015Figure 5 . Effects of microtubule-associated proteins on the motile properties of h RNPs . ( A ) Mean run length per RNP for hWT RNPs when solutions of Drosophila GFP-tagged kinesin-1 ( aa 1–401 ) rigor mutant ( kin401T99N:mGFP ) and human Alexa488-tau23 ( tau23 ) were previously added into the imaging chamber at the indicated concentrations . ( B ) Mean dwell time of bidirectional RNPs on individual microtubules with or without MAPs . In A and B , N is number of RNPs analysed . ( C ) Proportion of motile hWT RNPs per imaging chamber that are unidirectional in the presence and absence of MAPs . N , number of chambers analysed; n , total number of RNPs . ( D ) Kymographs exemplifying encounters of RNPs ( magenta ) with regions of microtubules containing kin401T99N:mGFP or static tau23 ( green ) . t , time; d , distance . Arrowheads and asterisks mark examples of encounters that are associated with RNP reversal or a passing event , respectively . A large fraction of tau23 underwent diffusive movement along microtubules ( Figure 5—figure supplement 1A–C ) , consistent with recent observations ( Hinrichs et al . , 2012 ) . A microtubule with a single , static patch of tau23 is therefore shown here for clarity . ( E ) Outcomes of individual encounters of hWT RNPs with regions of microtubules associated with kin401T99N:mGFP or the static fraction of tau23 . ‘Minus end’ and ‘Plus end’ indicate the direction of RNP movement prior to the encounter . ‘Simulation’ refers to a dataset in which the positions of microtubule-associated kin401T99N:mGFP puncta from an independent experiment were artificially superimposed on kymographs of hWT RNPs moving in the absence of added MAPs . N , total number of encounters analysed . ( F ) MAPs cause a reduction in the correlation between mean minus and plus end run length for individual bidirectional hWT RNPs ( RNPs with ≥20 runs in total were used for this analysis ) . Red solid line represents the best linear fit and dashed lines represent the 95% confidence interval for the data . R2 , correlation coefficient; N , total number of individual RNPs analysed . ( G ) Outcomes of individual encounters of HLE and hSL1x3 RNPs with regions of microtubules associated with kin401T99N:mGFP . ‘Minus end’ and ‘Plus end’ indicate the direction of RNP movement prior to the encounter . N , total number of encounters analysed . In A–C , error bars represent SEM . ***p<0 . 001; **p<0 . 01 ( Mann Whitney non-parametric t test compared to chambers without MAPs ( A–C ) or Fisher’s exact test compared to observed outcomes in the presence of MAPs ( E ) ) . In all cases , images were acquired at 4 . 2 fps to enable comparison with HLE data . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 01510 . 7554/eLife . 01596 . 016Figure 5—figure supplement 1 . Further characterisation of kin401T99N:mGFP and tau23 puncta on individual microtubules . ( A ) Stills showing appearance of microtubule-associated , recombinant GFP-labelled rigor kinesin ( kin401T99N:mGFP ) , and Alexa488-labelled tau23 added to the imaging chamber at the indicated concentrations . Asterisks mark tau23 spots that were static during the ∼160 s imaging period . ( B ) Kymographs showing examples of diffusive and static Alexa488-labelled tau23 puncta on individual microtubules . ( C ) Quantification of the number of stationary and non-stationary fluorescent puncta per μm per microtubule following addition of 10 nM kin401T99N:mGFP or 10 nM Alexa488-labelled tau23 into the chamber . N , number of microtubules . It was not possible to resolve individual fluorescent spots of kin401T99N:mGFP on microtubules in the 100 nM condition due to the high density . ( D ) Distribution of number of GFP decay steps for individual GFP-labelled rigor kinesin ( i ) or Alexa488-labelled tau23 puncta ( ii ) added to the chamber at 10 nM concentration and subjected to stepwise photobleaching analysis . The assembly of multiple tau proteins in a diffraction-limited spot on the microtubule is consistent with previous observations using this protein concentration ( Dixit et al . , 2008 ) . That this is also the case for GFP-labelled rigor kinesin is surprising , given strong evidence that kinesin-1 is dimeric in solution ( Vale et al . , 1996; Coy et al . , 1999 ) . The dispersed pattern of individual rigor kin401T99N:mGFP puncta on microtubules ( see panel A ) indicates that it is unlikely that multiple proteins per puncta are due to independent association of multiple dimers in a single diffraction-limited spot . Instead multiple kinesins may associate together in solution before microtubule binding . Note that we confirmed that purified kin401T99N:mGFP was predominantly dimeric before concentration using gel filtration . We also observed a maximum of two photobleaching steps of microtubule-associated kin401T99N:mGFP when a dilute solution of the protein ( 1 pM ) was added to imaging chambers , consistent with previous observations ( Telley et al . , 2009 ) . ( E ) Example of simulated kymograph of hWT RNP with 10 nM rigor kinesin obstacles . White arrowheads , events scored as reversals coincident with a GFP signal; asterisks , events scored as the GFP signal being passed . The simulated dataset was generated by overlaying the positions of kin401T99N:mGFP puncta from the ‘observed’ dataset on kymographs of RNAs moving in the absence of obstacles . In this particular case , the position of the rigor kinesin puncta are derived from the kymograph from the ‘observed’ dataset shown in Figure 5D . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 01610 . 7554/eLife . 01596 . 017Figure 5—figure supplement 2 . MAPs curtail the motility of HLE and hSL1x3 RNPs on microtubules . ( A ) Mean run lengths per RNP for HLE and hSL1x3 RNPs with and without kin401T99N:mGFP . N , number of RNPs analysed . ( B ) Mean dwell time of bidirectional RNPs on individual microtubules with and without kin401T99N:mGFP . N , number of RNPs analysed . ( C ) Mean proportion of motile HLE and hSL1x3 RNPs per imaging chamber that are unidirectional with and without kin401T99N:mGFP . N , number of chambers; n , total number of RNPs . ***p<0 . 001; **p<0 . 01 ( Mann Whitney non-parametric t test ) , compared to no obstacle scenario for the same parameter and same RNA species; error bars represent SEM . In all cases , images for these analyses were acquired at 4 . 2 fps . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 017 The reduction in net minus end motion of the hWT population in the presence of MAPs was associated with a significant decrease in the proportion of RNPs that moved exclusively unidirectionally towards the minus end of the microtubule ( Figure 5C ) . Indeed , no unidirectional hWT RNPs were observed in chambers with the highest concentration of kin401T99N:mGFP ( Figure 5C ) . These observations suggest that unidirectional RNPs are capable of switching to a bidirectional mode in response to MAPs . To directly observe the effects of MAPs on RNP motion we visualised individual encounters of hWT complexes with those regions of microtubules containing discrete puncta of kin401T99N:mGFP or static tau23 ( Figure 5D , E ) . Encounters of hWT RNPs with a region of microtubule bound by a kin401T99N:mGFP punctum were rarely associated with RNP detachment or pausing ( <5% of encounters in total ) . The frequency of such outcomes was in fact no different than expected for a segment of undecorated microtubule of similar length , as revealed by simulating the positions of MAPs on kymographs of hWT RNPs moving in the absence of obstacles ( Figure 5E , Figure 5—figure supplement 1E ) . In approximately 25% of cases , RNPs passed the region bound by a rigor kinesin punctum ( Figure 5E ) . Such outcomes presumably include a contribution of events in which the RNP and MAP do not meet each other due to their presence on different microtubule protofilaments . The vast majority ( ∼70% ) of encounters of hWT RNPs with a region bound by a kin401T99N:mGFP punctum coincided with a reversal of the RNP on the microtubule , regardless of the direction of initial motion ( Figure 5E ) . A much lower incidence of reversals ( ∼10–15% ) was expected by random chance , as revealed by analysing the dataset with simulated MAP positions ( Figure 5E ) . Encounters of RNPs with regions of microtubules decorated with static tau23 also led to a reversal in the vast majority of cases , with very few detachments or pauses ( Figure 5E ) . Collectively , these data demonstrate that encounters with MAPs frequently induce reversals of hWT RNPs . Consistent with reversals in these experiments being influenced not just by intrinsic properties of the RNPs but also by the stochastic positioning of MAPs on the microtubule , the correlation between mean minus end and mean plus end travel distance per RNP was reduced in the presence of kin401T99N:mGFP and tau23 ( Figure 5F ) . The influence of MAPs may therefore explain why tightly correlated motion of cargo-motor complexes in each direction has not been previously observed in vivo . We next addressed the influence of dynein-dynactin copy number on the behaviour of RNPs upon encounters with MAPs . We hypothesised that increasing dynein-dynactin copy number on RNPs could facilitate the passing of MAPs by increasing the probability of initiating motility on an obstacle-free protofilament . To test this notion , Cy3-labelled hSL1x3 RNPs , containing significantly more dynein-dynactin copies on average than hWT RNPs , were observed in the presence of kin401T99N:mGFP obstacles . Consistent with what was seen for hWT ( Figure 5A–C ) , the proportion of hSL1x3 RNPs that were unidirectional , and hence the overall minus end bias of the population , was reduced in the presence of the MAP without a significant reduction in the dwell time of RNPs on microtubules ( Figure 5—figure supplement 2A–C ) . Thus , the motile properties of hSL1x3 RNPs along microtubules were sensitive to the rigor kinesin . Analysis of individual encounters of hSL1x3 RNPs with regions of microtubules bound by kin401T99N:mGFP puncta revealed a quantitatively similar behaviour to that observed for hWT RNPs; ∼70% of encounters preceded by either minus or plus end-directed motion were associated with a reversal , while detachments and pauses were rare ( Figure 5G ) . The movement of RNPs containing the short HLE RNA was also curtailed by kin401T99N:mGFP , with the proportion of motile RNPs that were unidirectional reduced significantly ( Figure 5—figure supplement 2A–C ) . Furthermore , despite containing fewer associated dynein-dynactins , HLE RNPs reversed at obstacles with similar probability to hWT and hSL1x3 RNPs ( Figure 5G ) . Collectively , this series of experiments indicates that the total number of associated dynein-dynactin complexes does not influence the ability of RNPs to navigate around microtubule-associated obstacles . The results described above demonstrate that RNPs can respond to microtubule-associated obstacles by reversing direction . To determine whether RNPs can react to other cytoskeletal features , we studied their response to encounters with microtubule ends . Interestingly , encounters of bidirectional hWT RNPs with plus ends always led to a reversal in travel direction ( Figure 6A , B ) . Thus , these RNPs do not detach at the plus end , a behaviour that might facilitate their long-term translocation on the microtubule cytoskeleton in vivo ( ‘Discussion’ ) . 10 . 7554/eLife . 01596 . 018Figure 6 . The responses of RNPs to microtubule ends . ( A ) Kymographs showing examples of encounters of bidirectional hWT RNPs with microtubule plus ends ( i ) and minus ends ( ii and iii ) . RNPs are shown in magenta and microtubules in green ( plus end labelled by greater incorporation of HiLyte 647-tubulin ) . ‘*’ in ( i ) represents a different microtubule with plus end lying along the tracked RNP path . R indicates a number of examples of reversals . Dashed line indicates position of microtubule minus end . ( B ) Outcome of encounters of bidirectional RNPs with minus and plus ends of microtubules . Pauses were defined as events in which RNPs were stationary for longer than 1 frame ( 0 . 236 s ) . N , total number of encounters analysed . ( C ) Duration of minus end pausing events for different bidirectional h RNP variants . Pausing events usually ended with disappearance of the Cy3 signal ( presumably due to detachment from the microtubule or Cy3 photobleaching ) . Open circles indicate the minority of events that ended abruptly due to completion of image acquisition . ( D ) Kymographs showing examples of encounters of unidirectional hWT RNPs with microtubule minus ends leading to pausing ( i ) or detachment ( ii ) . RNPs , microtubules , and minus ends are depicted as in A . t , time; d , distance . ( E ) Quantification of outcomes of encounters of unidirectional RNPs with minus ends of microtubules ( no unidirectional hΔLE RNPs were observed [Figure 3E] ) . N , total number of encounters analysed . *p<0 . 05 ( Fisher’s exact test ) , compared to hWT values for the same parameter . ( F ) Duration of pausing events for different unidirectional h RNP variants . *p<0 . 05 ( Mann Whitney non-parametric t test ) , compared to hWT values for the same parameter . In C and F , long and short horizontal red lines demarcate the mean and SEM , respectively . Images for all of these analyses were acquired at 4 . 2 fps to enable comparison with HLE data . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 018 Reversals of bidirectional hWT RNPs were also common upon reaching the minus end , occurring ∼70% of the time ( Figure 6A , B ) . Detachments of bidirectional hWT RNPs upon immediately reaching a minus end were rare ( ∼5% of total encounters ) ( Figure 6B ) . Approximately 25% of encounters of these RNPs with minus ends resulted in long-term pausing at this site , with a mean dwell time of ∼1 min ( Figure 6B , C ) . By definition , unidirectional hWT RNPs did not reverse at the minus end . Nonetheless , we never observed an example of a reversal at the minus end that was preceded by a long minus end-directed run ( >3 μm ) . Thus , unidirectional motion does not appear to be followed by a reversal at the minus end . Approximately 70% of encounters of unidirectional hWT RNPs with a minus end resulted in pausing , with detachment from the microtubule occurring in all other cases ( Figure 6D , E ) . Collectively , these data demonstrate that hWT RNPs are capable of recognising the minus end of microtubules and being retained there . This process may contribute to the dynein-dependent anchorage of RNAs in the apical cytoplasm that has been observed in vivo ( Delanoue and Davis , 2005 ) ( ‘Discussion’ ) . To study the effect of dynein-dynactin copy number on the response of RNPs to microtubule ends we turned our attention to the behaviour of hΔLE and hSL1x3 RNPs assembled in the RAT-TRAP assay . Bidirectional RNPs containing these RNA species behaved indistinguishably to hWT RNPs at minus and plus ends ( Figure 6B , C ) . Unidirectional hSL1x3 RNPs also responded to minus ends of microtubules in the same manner as unidirectional hWT RNPs ( Figure 6E , F ) . Thus , average dynein-dynactin copy number does not influence the reaction to microtubule ends of RNPs that harbour h 3′UTRs . Interestingly , the mean durations of pauses of unidirectional hWT and hSL1x3 RNPs at minus ends were ∼ fivefold shorter than those exhibited by bidirectional RNPs containing the same RNA species ( Figure 6C , F; e . g . , hWT unidirectional , 10 . 7 ± 1 . 7 s; hWT bidirectional , 58 . 9 ± 8 . 0 s ( p<0 . 0001 , Mann–Whitney non-parametric t test ) ) . This finding led us to consider the possibility that features that promote bidirectional motion aid in the retention of RNPs at the minus end . To explore this hypothesis further , we examined the behaviour at minus ends of the HLE element , which lacks sequences within the rest of the 3′UTR that can recruit dynein-dynactin and drive bidirectional motion ( as revealed by analysis of the hΔLE RNA [Figure 3] ) . Upon reaching the minus end , unidirectional HLE RNPs were more likely to detach than unidirectional hWT and hSL1x3 RNPs ( Figure 6E ) , and the duration of pauses was significantly reduced ( Figure 6F ) . This observation is compatible with the notion that anchorage of RNPs at minus ends is facilitated by features that promote bidirectional movement .
Several previous studies have manipulated the copy numbers of cargo-associated motors in order to elucidate how multiple motors orchestrate sorting . One experimental approach has been to alter the numbers of isolated motors or motor domains attached to artificial cargos such as beads or DNA origami ( e . g . , Block et al . , 1990; Mallik et al . , 2005; Diehl et al . , 2006; Vershinin et al . , 2007; Derr et al . , 2012; Furuta et al . , 2013 ) . Although very informative , these studies did not include physiological cargo , cargo adaptors , and motor co-factors that could potentially modulate motor behaviour . Other studies have used genetic manipulations to alter the numbers of motor complexes available to cargos in vivo ( Shubeita et al . , 2008; Reis et al . , 2012 ) . These approaches are physiologically relevant but cannot rule out influences from the cellular environment , including possible indirect effects of altered motor concentration on other processes that impinge on cargo motility . In this study , we have manipulated the copy number of native motor complexes on a physiological type of cargo , by incubating cellular extracts with RNA variants , and studied the consequences on motility in a defined in vitro setting using high spatiotemporal resolution imaging . We find that h wild-type RNPs associate with dynein-dynactin and can undergo either unidirectional motion in the minus end direction that is highly processive or bidirectional motion that has characteristics of a diffusive process . Our experiments indicate that unidirectional RNP movement is driven by active , ATP hydrolysis-mediated translocation of dynein along the microtubule . The most parsimonious explanation for the bidirectional motion is that it is also due to dynein undergoing back-and-forth movements along the microtubule , a behaviour that has been observed in several studies of the purified motor in vitro ( e . g . , Wang et al . , 1995; Wang and Sheetz , 1999 , 2000; Mallik et al . , 2005; Ross et al . , 2006; Miura et al . , 2010 ) and also appears to occur in vivo ( Ananthanarayanan et al . , 2013 ) . Indeed , we observed bidirectional movement of a significant subset of RNPs containing only the HLE ( ∼30% in the absence of MAPs and ∼60% in their presence [Figure 5—figure supplement 2C] ) , on which we failed to detect binding of a kinesin family member under conditions in which dynein and dynactin were readily detected ( Dix et al . , 2013 ) . Interestingly , bidirectional motion of RNPs is not overtly sensitive to inhibition of dynein’s ability to hydrolyse ATP . Thus , our findings are compatible with those of Miura et al . ( 2010 ) who reported passive diffusion along microtubules of dynein in complex with a dynactin component in the presence of ATP or ATP-vanadate . We found no evidence that bidirectional RNPs can undergo long , ATP hydrolysis-dependent runs in both directions akin to those documented for individual , GFP-labelled dynein-dynactin complexes purified from mouse brain ( Ross et al . , 2006 ) . Varying the number of SL1 elements within the context of the h 3′UTR revealed a correlation between the total copy number of dyneins per RNP and the probability of entering into the unidirectional , minus end-directed state . Previous studies using purified dynein bound to artificial cargos have demonstrated that increasing motor copy number is sufficient to stimulate processive movement towards minus ends ( Mallik et al . , 2005; Ross et al . , 2006 ) . However , experiments with the isolated HLE signal indicated that total motor number is not the key determinant of the unidirectional mode of RNP movement . The HLE alone has a statistically indistinguishable copy number of dynein components to the hΔLE RNA , in which the localisation signal has been replaced by a heterologous sequence in the context of the h 3′UTR , yet only the former RNA is capable of unidirectional motion . These data suggest that features associated with the RNA signal are sufficient to increase the probability of processive movement of the associated dynein . It has recently been shown in Schizosaccharomyces pombe that microtubule-associated dynein can switch from diffusive to processive behaviour upon contacting cortical anchors , an event that regulates the generation of pulling forces on the microtubule ( Ananthanarayanan et al . , 2013 ) . Our data suggest that regulation of dynein processivity by associated factors may be a widespread phenomenon . Collectively , our results support a novel model in which the same cargo species can interact with processive or non-processive dynein , with discrete cargo-associated features regulating the probability of switching between the two behaviours ( Figure 7 ) . Thus , the regulatory mechanisms underpinning sorting of RNPs in this system appear distinct from those of other well-studied bidirectional cargos , which involve the interplay of opposite polarity force-generating motors , such as dyneins and kinesins ( Shubeita et al . , 2008; Ally et al . , 2009; Soppina et al . , 2009; Jolly and Gelfand , 2011; Reis et al . , 2012 ) . How might localisation signals regulate dynein processivity ? One possibility is that proteins recruited by the localisation signal directly regulate the activity of the motor . Two candidates to serve such a role are the adaptor proteins Egl and BicD , which are associated only with the dynein-dynactin bound to the localisation signals ( Bullock et al . , 2006; Dix et al . , 2013 ) . It is also conceivable that the structure or rigidity of the RNA signal plays an architectural role in presenting dynein-dynactin to the microtubule in a manner that favours processive movement . 10 . 7554/eLife . 01596 . 019Figure 7 . Schematic model for the role of RNA localisation signals in controlling net RNP motion . ( Top ) In the absence of RNA localisation signals , RNPs undergo passive diffusion on the microtubule lattice . Non-processive dyneins are recruited to sites in the RNA other than localisation signals through an unknown factor ( X ) . The most parsimonious explanation for the diffusive movement is that it is driven by dynein-dynactin . ( Middle ) The presence of a localisation signal leads to the recruitment of additional dynein-dynactin and a subset of RNPs in the population undergoing processive , minus end-directed transport . This introduces a minus end bias to motion of the RNA population . It is not the increased total copy number of the motor complex that drives unidirectional movement , but rather the ability of localisation signals to increase the probability of associated dynein-dynactin entering into a processive state . ( Bottom ) Addition of more SL1 elements increases further the likelihood of RNPs entering into the unidirectional state , leading to a greater minus end bias to motility of the RNA population . It is not known whether unidirectional motion involves stepping of a single dynein-dynactin or of multiple motor complexes . Resolving this issue will require long-term nanoscale analysis . Note that for simplicity the cartoon does not attempt to depict the absolute copy number of dynein-dynactin complexes on each RNP . DOI: http://dx . doi . org/10 . 7554/eLife . 01596 . 019 Our study also sheds light on how physiological cargo-motor complexes respond to extrinsic factors within the cytoskeletal environment . We show that RNPs frequently reverse when encountering the regions of microtubules bound by puncta of both MAPs studied , rigor kinesin and tau23 . The behaviour of RNPs at MAPs is therefore highly reminiscent of that seen for individual , purified dynein-dynactin complexes in vitro ( Dixit et al . , 2008 ) . Interestingly , comparison of the behaviour of the h RNA variants reveals that the number of dynein-dynactins associated with an RNP does not increase the probability of passing microtubule-associated obstacles . The ability of RNPs to reverse at MAPs may help them navigate to their destination in vivo . For example , reversals upon meeting an obstacle may facilitate encounters of RNPs with intersecting microtubules . Switching of RNPs between microtubules , a behaviour we have observed when intersections occur in our in vitro assays ( unpublished observations ) , could allow these complexes to explore alternative routes to their destination . Reversals of RNPs at a MAP may also give dynein space to switch to a different lateral position on the same microtubule and thereby provide another opportunity to pass the obstacle following resumption of movement in the previous direction . Compatible with this notion , we see RNPs moving on single microtubules that can pass obstacles after multiple attempts ( e . g . , Figure 5D ) . Our analysis of the behaviour of RNPs at microtubule ends indicates that plus end encounters always result in a reversal . The behaviour may also be advantageous in a cellular environment by preventing detachment of RNPs at this point , an outcome that would necessitate a rebinding event before motion on the microtubule network can resume . Interestingly , a subset of RNPs undergoes pausing at the minus ends of microtubules , with a mean dwell time of ∼1 min for bidirectional complexes . These findings demonstrate that the probability of changing directions is different at the minus end and the plus end of the microtubule and that additional in vivo features , such as the γ-tubulin ring complex or other centrosome-associated factors , are not obligatory for long-term retention of RNPs at the minus end . Intrinsic behaviours of RNPs upon reaching the minus end of the α/β-tubulin polymer may therefore contribute to the dynein-dependent anchorage of RNAs in the vicinity of minus ends in vivo ( Delanoue and Davis , 2005 ) . Analysis of hΔLE , hWT , and hSL1x3 RNAs demonstrates that the probability of an RNP undergoing minus end pausing in vitro , as well as the duration of such events , is not influenced by the addition of more dynein-dynactins through localisation elements . This finding offers an explanation for why inhibition of Egl and BicD following translocation of localising RNAs to the apical cytoplasm of the embryo does not affect dynein-dependent anchorage ( Delanoue and Davis , 2005 ) . It is intriguing that the average dwell time of pausing events of unidirectional hWT and hSL1x3 RNPs at minus ends is ∼ fivefold less than that of bidirectional RNPs harbouring the same RNA species . We also find that the unidirectional HLE RNPs , which lack features within the h 3′UTR that can recruit non-processive dynein-dynactin , dwell at the minus end for significantly less time than the unidirectional hWT and hSL1x3 RNPs . One explanation for these findings is that the ability of dynein bound to localisation signals to walk processively off the minus end is antagonised by interactions with the microtubule mediated by non-processive dynein bound at other sites in the RNA . Additional , long-term experiments will be required to test this hypothesis . Nonetheless , our data suggest more generally that features that promote bidirectional motion could assist in the retention of RNPs at minus ends . Collectively , our analysis of encounters of RNPs with MAPs and microtubule ends raises the possibility that the co-existence of unidirectional or bidirectional modes of movement facilitates efficient navigation of an RNA population to its destination in vivo . Processive , unidirectional movement in the minus end direction could be beneficial for rapid , directional movement along regions of the microtubule that are not rich in MAPs . Diffusive motion along microtubules may be valuable for movement of RNPs through an obstacle rich environment and could still contribute to asymmetric sorting as it is associated with long-term retention of complexes at microtubule minus ends . This strategy appears analogous to that used by DNA enzymes and kinesins that depolymerise microtubules , which can employ one-dimensional diffusion to search for their specific target sites ( Helenius et al . , 2006; Gorman and Greene , 2008 ) . Our in vitro work on RNA motility has provided several new insights into how cargo-motor complexes operate and how their behaviour is modulated by encounters with the environment . Our data lead to a model in which discrete cargo-associated features regulate motor processivity , a phenomenon that could not have been recapitulated using minimal motor elements coupled to artificial cargos . Our results also illustrate that intrinsic motile properties of cargo populations in vivo are likely to be obscured by the influence of extrinsic factors including MAPs and microtubule ends . Further exploitation of the RAT-TRAP assay is likely to be an effective strategy for shedding light on molecular mechanisms that underpin intrinsic and extrinsic regulation of cargo motility , particularly when combined with powerful Drosophila gene perturbation techniques . In the longer term , it will be important to understand how the behaviours defined in vitro are integrated during sorting of single RNA molecules in vivo , a goal that necessitates the development of new methods to visualise movement of transcripts in the optically challenging embryo system .
Wild-type embryos were of the OR-R strain . P ( Ubi-GFP::Dlic ) ( Pandey et al . , 2007 ) and P ( mat-tub-α4-GFP::Dmn ) ( Januschke et al . , 2002 ) lead to ubiquitous expression of the GFP fusion proteins during early development and were gifts from J Raff ( Oxford University , UK ) and A Guichet ( Institut Jacques Monod , France ) , respectively . The hairy RNA ( hWT ) used for in vitro motility assays is 822 nt long and represents the majority of the transcript’s 3′UTR . hSL1x3 is a h 3′UTR variant in which the 124 nt h localisation signal ( HLE ) is replaced by a cassette containing three copies of the 46 nt h stem-loop 1 ( SL1; Bullock et al . , 2003 ) separated by 12–17 nt single stranded spacers . The presence of three SL1 elements increases the net minus end bias to h 3′UTR RNP motility in vivo compared to the wild-type HLE ( Bullock et al . , 2006 ) . hΔLE is a h 3′UTR variant in which the HLE is replaced with the same-sized piece of RNA derived from the glutathione-S-transferase ( GST ) gene from Schistosoma japonicum ( generated by PCR from the pGEX6P-1 vector [GE Healthcare] ) . HLE has the isolated 124 nt h localisation signal , together with 25 nt of the 3′ sequence of the h 3′UTR ( the inclusion of the additional sequence was designed to assist folding of the SL2 element of the HLE ) . Sequences encoding hWT , hSL1x3 , hΔLE , and HLE RNAs were introduced into the pTRAPv3 . 0 vector ( Cytostore ) , allowing a fusion RNA to be synthesised from the T7 promoter that contains two copies of the S1 streptavidin-binding aptamer ( Srisawat and Engelke , 2001 ) 5′ to the RNA of interest ( the transcription product from the T7 promoter until the start of h RNA variants is 264 nt long ) . We also introduced 15 nt of RNA sequence at the 5′ and 3′ of the aptamer-HLE ( GCATACCGGATACGC and CCATAGGCATAGCGC , respectively ) as part of the splint ligation procedure for end-labelling the RNA with Cy3 dyes ( see below ) . Cy3-labelled , aptamer-linked hWT , hSL1x3 , and hΔLE RNAs were synthesised with the MEGAscript T7 kit ( Ambion , Foster City , CA ) as per the manufacturer’s instructions with 1 . 875 mM Cy3–UTP ( PerkinElmer , Waltham , MA ) and 5 . 625 mM unlabelled-UTP ( Roche , Switzerland ) . Alexa488-labelled , aptamer-linked hWT RNA was synthesised as described previously ( Bullock et al . , 2006 ) . Unincorporated nucleotides were removed using mini Quick Spin RNA Columns ( Roche ) . The degree of body labelling of RNAs with these dyes was determined with a Nanodrop spectrophotometer ( Thermo Scientific , Waltham , MA ) . Typically , there was an average of ∼8 dyes per aptamer-h 3′UTR RNA molecule . Aptamer-HLE RNA was labelled at each end with Cy3 using the splint ligation technique , as previously described ( Moore and Query , 2000 ) . Briefly , this method uses bridging DNA oligonucleotides to facilitate T4 DNA ligase-mediated ligation of Cy3-labelled 15 nt RNA oligonucleotides ( Integrated DNA Technologies , Coralville , IO ) to the 5′ and 3′ ends of the aptamer-HLE RNA , followed by the removal of the DNA using DNaseI ( Agilent , Santa Clara , CA ) . Biotin-PEG-functionalised glass and passivated counter glass surfaces were prepared as described ( Bieling et al . , 2010; Roostalu et al . , 2011 ) . Imaging was performed at 24 ± 1°C with a total internal reflection fluorescence microscope ( Nikon , Netherlands ) equipped with a 100× objective ( Nikon , 1 . 49 NA Oil , APO TIRF ) , using the following lasers: 150 mW 488 nm , 150 mW 561 nm laser ( both Coherent ( Santa Clara , CA ) Sapphire ) , and 100 mW 641 nm ( Coherent Cube ) . Images were acquired with a back illuminated EMCCD camera ( iXonEM+ DU-897E , Andor , UK ) controlled with µManager software ( http://micro-manager . org/wiki/Micro-Manager ) . The size of each pixel was 105 × 105 nm . Porcine tubulins and polymerisation buffers were purchased from Cytoskeleton , Inc . ( Denver , CO ) . Biotinylated , GmpCpp-stabilised microtubules with plus ends marked by greater incorporation of HiLyte 647 were polymerised as previously described ( Roostalu et al . , 2011 ) . Briefly , long , biotinylated microtubules that were dimly labelled with fluorophore were first produced by polymerisation for 2 hr at 37°C from 1 . 66 µM unlabelled tubulin , 0 . 4 µM biotinylated tubulin , 0 . 15 µM HiLyte 647 tubulin in the presence of 0 . 5 mM GmpCpp ( Jena Bioscience , Germany ) in BRB80 ( 80 mM PIPES , 4 mM MgCl2 , 1 mM EGTA , pH 6 . 8 ) . Polymerised microtubules were sedimented by centrifugation at 20 , 800×g for 8 min at room temperature . In a second step , a short microtubule plus end segment was generated that was brightly labelled with fluorophore . This was achieved by incubation of the dimly labelled microtubules for 30–45 min at 37°C in ‘bright elongation mix’ consisting of 1 . 5 µM NEM-tubulin ( n-ethylmaleimide-tubulin prepared as previous described ( Phelps and Walker , 2000 ) ) , 1 . 33 µM HiLyte 647 tubulin and 0 . 5 mM Gmp-Cpp in BRB80 . Polarity-marked microtubules were then sedimented at 20 , 800×g for 8 min at RT and resuspended in BRB80 containing 40 µM taxol . Microtubules were immobilised on a biotin-PEG-coated glass surface via streptavidin as described ( Bieling et al . , 2010 ) . In control experiments , microtubule length was determined to be 15 ± 1 . 6 µm ( mean ± SEM , N = 30 ) . 50 µl of M-280 streptavidin-coupled Dynabeads ( Invitrogen , Carlsbad , CA ) were washed twice for 10 min in 1 mg·ml−1 bovine serum albumin ( BSA ) on a roller mixer at room temperature . Magnetic beads were then incubated with 1 mg·ml−1 BSA on ice for 60 min . This procedure was designed to block unspecific binding sites on the beads . Subsequently , 1 pmol of fluorescently labelled RNA ( containing the streptavidin aptamers and RNA of interest ) in DXB buffer ( 30 mM HEPES at pH 7 . 3 , 50 mM KCl , 2 . 5 mM MgCl2 , 250 mM sucrose , 5 mM dithiothreitol , 10 μM MgATP , and Complete [EDTA-free] protease inhibitor [Roche] ) with 40 U RNasin RNase inhibitor ( Promega , Madison , WI ) was incubated with the blocked Dynabeads for 2 hr at 4°C . Drosophila embryo extracts were produced from dechorionated 0–6 hr wild-type embryos by homogenising in DXB buffer as described ( Bullock et al . , 2006 ) using 100 μl of DXB buffer per 50 mg of embryos . Embryo extract was then centrifuged for 8 min at 9000 rpm at 4°C to remove debris . Typically , 100 µl embryo extract was added to the Dynabead–RNA complex , supplemented with 40 U of RNasin and incubated for 1 hr at 4°C on a roller mixer . The magnetic beads were washed once with DXB buffer at 4°C and RNA-motor complexes eluted by incubating the Dynabeads with 200 µl of 10 mM biotin ( Invitrogen ) in motility buffer ( 30 mM HEPES/KOH , 5 mM MgSO4 , 1 mM DTT , 1 mM EGTA , 0 . 5 mg/ml bovine serum albumin , pH 7 . 0 ) at 15°C for 20 min in a Thermomixer comfort ( Eppendorf , UK; 950 rpm ) . Biotin competes for the interaction of the aptamer with streptavidin . In control experiments , we confirmed by immunoblotting that components of RNA-motor complexes are present in the eluate from hWT RNA affinity purification experiments by probing for Egalitarian , Bicaudal-D and Dynein heavy chain . These factors were not present when RNA was omitted from the procedure . For motility assays the eluate was transferred immediately to ice and subsequently introduced along with 2 . 5 mM ATP and an oxygen scavenging system ( 1250 nM glucose oxidase , 140 nM of catalase , 71 mM 2-mercaptoethanol , and 24 . 9 mM glucose ) into a flow chamber with polarity-marked , HiLyte 647-labelled microtubules pre-adsorbed to the coverslip . In a subset of experiments sodium orthovanadate or apyrase was added to the eluate ( to a final concentration of 100 μM or 20 U·ml−1 , respectively ) before addition to the flow cell . In the apyrase experiment the eluate was not supplemented with ATP . Microtubules and Cy3-labelled RNA molecules were visualised with a TIR microscope ( 15 fps , that is 66 ms capture time ( 63 ms exposure plus 3 ms image acquisition ) or 4 . 2 fps , that is 236 ms capture time ( 200 ms exposure plus 36 ms image acquisition ) ) ; unless stated otherwise , ∼ 1500 frames or ∼700 frames were collected for 15 fps or 4 . 2 fps imaging , respectively . HLE RNPs were analysed at the lower frame rate ( 4 . 2 fps ) . This was because the reduced number of Cy3 dye molecules per RNA molecule compared to h 3′UTR RNAs ( see above ) necessitated a longer exposure time . Experiments in which the motility of h 3′UTR species was compared to that of HLE were also performed at 4 . 2 fps . Hence , all motility experiments in Figures 4–6 ( involving HLE ) were performed at 4 . 2 fps , with all other experiments performed at 15 fps . For each condition/RNA species , data were collected from three independent days of experiments ( 8–10 imaging chambers ) . Random microtubules in the region-of-interest were selected , followed by analysing the RNPs that moved along them . The movement of RNPs with reference to the polarity of microtubules were analysed using TrackMate in Fiji ( http://fiji . sc/TrackMate ) and a custom Matlab code ( available on request ) . Sub-pixel XY coordinates of motile RNPs were acquired from TrackMate as described ( http://fiji . sc/TrackMate ) using two-dimensional Gaussian fitting . The RNP coordinates were imported into Matlab where they were projected onto a vector along the coordinates of the microtubule ( Hendricks et al . , 2010; Rai et al . , 2013 ) , thereby producing on-axis positions of the RNP with respect to the track . The difference in on-axis position between consecutive frames ( termed d ) was then calculated . The tracking accuracy was determined to be 11 . 3 nm ( i . e . , the standard deviation of instantaneous displacements of tracked immobile RNPs on microtubules [Hendricks et al . , 2010; Figure 1—figure supplement 1C , D] ) . Runs were defined as the sum of consecutive displacements >22 nm in one direction ( i . e . , before termination by pausing , reversal , RNP detachment , or the end of imaging ) . 22 nm was chosen as a cut-off as it is approximately twice the standard deviation of instantaneous displacement measurements for immobile RNPs ( Figure 1—figure supplement 1D ) . A similar calculation was previously used to produce a cut-off for the analysis of the motility of neuronal vesicles in vitro ( Hendricks et al . , 2010 ) . Pauses were defined as ≥1 frame with an instantaneous displacement ≤22 nm . Reversals within RNP tracks were defined as instances when d of consecutive frames i and i+1 were >22 nm and the absolute value of the sign of di minus di+1 equalled 2 ( i . e . , |sign ( di ) -sign ( di+1 ) | = 2 ) , where sign ( d ) = +1 for d >0 and sign ( d ) = −1 for d <0 . The trajectories were then analysed for run length and persistence time and these values used to calculate run velocity . In control experiments , analysis of 20 s bins of RNP motion over the duration of image acquisition ( 100–120 s ) revealed that mean run length and velocity does not decrease over time . This indicates that RNA-motor complexes are relatively stable over the total period of image acquisition . The proportion of motile RNPs that were unidirectional vs bidirectional for each RNA variant or condition was determined by manual analysis of 6–10 imaging chambers from three independent days of experiments . Automatic tracking of a subset of RNPs defined as unidirectional by manual analysis confirmed that they do not contain plus end-directed events . For the analysis in Figure 2G , RNPs were deemed as unidirectional , bidirectional , or stationary ( no motion beyond 1 pixel ) by manual analysis of kymographs . 12 randomly selected microtubules from at least three different imaging chambers were analysed for each condition . Mean squared displacement ( MSD ) analysis was performed with sub-pixel resolution using randomly selected bidirectional and unidirectional RNPs filmed at 15 fps . MSD was calculated in Matlab using internal averaging ( averaging over all pairs ) , to ensure each data point was weighted evenly , for no greater than one quarter of the total travel time , as previously described ( Saxton , 1997 ) . Diffusion coefficients ( D ) of bidirectional RNPs were calculated from a linear fit to the MSD ( t ) data using the equation MSD = 2Dt . The plots in Figures 2B–E and 5F were produced by obtaining the mean of the individual minus and plus end run lengths or velocities for each RNP that had a total number of runs ≥20 . For the experiments documented in Figure 1—figure supplement 1A , B , 1 pmol each of Alexa-488-labelled h and Cy3-labelled h were mixed and captured on streptavidin beads before incubation with embryo extract . The assembled RNA-motor complexes were eluted using biotin and injected into a flow cell containing HiLyte 647-labelled microtubules as describe above . Images of microtubules were captured at the beginning and end of a series of alternating images of the Alexa-488 and Cy3 signals ( 236 ms capture [200 ms exposure +36 ms image acquisition] per channel , 500 frames ) . DNA coding for the first 401 amino acids of conventional kinesin ( kinesin-1 ) from Drosophila melanogaster with monomeric GFP fused to the C-terminal end and His6-z- to the N-terminal end was a gift from T Surrey ( Telley et al . , 2009 ) . We used Quikchange mutagenesis ( Stratagene ) to introduce a T99N substitution in the kinesin motor domain to obtain a non-motile rigor mutant that constitutively associates with the microtubule in a strongly bound state ( Telley et al . , 2009 ) . The recombinant fusion protein was expressed in bacteria ( BL21 ( DE3 ) strain ) and the dimeric fraction purified using gel filtration as described previously ( Telley et al . , 2009 ) . Human tau23 ( 352 AA , 0 inserts and 3 repeats–0N3R , shortest isoform ) was expressed in bacteria ( BL21 ( DE3 ) strain ) , purified , and fluorescently labelled with Alexa-488 maleimide as previously described ( Goedert and Jakes , 1990; McVicker et al . , 2011 ) . Tau23 contains one reactive cysteine , at position 322 within the 3rd repeat . The percentage of protein that was labelled , determined using a NanoDrop ND-1000 spectrophotometer ( Thermo Scientific ) , was ∼80% . Kin401T99N:mGFP was introduced into chambers containing HiLyte 647-microtubules preadsorbed to the coverslip . After washing with 5x chamber volume of motility buffer , RNA-motor complexes were introduced with ATP and oxygen scavenging agent as described above . A still image of the microtubule ( 635 nm excitation , 236 ms capture time [200 ms exposure +36 ms image acquisition] ) was acquired , followed by a still image of rigor kin401T99N:mGFP ( 488 nm excitation , 1 . 036 s capture time [1 s exposure +36 ms image acquisition] ) . This was followed by continuous imaging of the Cy3-RNA channel ( 561 nm excitation , 236 ms capture time [200 ms exposure +36 ms image acquisition] ) for 700 frames to monitor the behaviour of RNA-motor complexes . At the end of this series another still image of kin401T99N:mGFP was acquired as above . Encounter statistics in Figure 5E , G were calculated manually using kin401T99N:mGFP present at the same position during the entire time interval of the movie . We had previously confirmed using continuous imaging of the GFP channel that kin401T99N:mGFP rarely moved during several minutes of filming . Outcomes of encounters were scored when the RNP reached the pixels containing kin401T99N:mGFP fluorescent signal on the kymograph . The same experimental set up was used for tau23 experiments , with the following modifications: a movie was recorded ( 488 nm excitation , 236 ms capture time [200 ms exposure +36 ms image acquisition] , 40 frames ) before and after imaging the RNA-motor complexes . This allowed assessment of which tau23 patches were stationary and only these were used for encounter analysis . Analysis of run lengths and velocities of RNPs in the absence and presence of MAPs was performed using automated tracking and analysis as described above . To quantify simulated encounters of RNPs with kin401T99N:mGFP ( Figure 5E , Figure 5—figure supplement 1E ) , kymographs of static , microtubule-associated kin401T99N:mGFP puncta generated in independent experiments were superimposed on kymographs of RNPs moving on microtubules without any MAPs added to the chamber . Scoring of encounters of RNPs with the simulated obstacles was performed as described above . Outcomes of RNP encounters with microtubule ends were scored manually from kymographs . Pauses were defined as events in which RNPs were stationary for longer than 1 frame ( 0 . 236 s ) . Relative copy numbers of GFP-tagged Dlic and Dmn on RNA-motor complexes were estimated using RNPs assembled and eluted using the same protocol as for RAT-TRAP assays , except no ATP was added to the eluate before introduction into the imaging chamber . The absence of nucleotide was designed to promote stable binding of RNA-motor complexes to microtubules . Note that quantification of the number of GFP molecules in RNPs undergoing motion was precluded by fluctuations in GFP fluorescence intensity . GFP and Cy3 fluorophores were illuminated sequentially for 136 ms and images captured as described above . Positions of HiLyte 647-labelled microtubules were recorded by image capture at the beginning and end of filming . Kalaimoscope Motion Tracker ( Transinsight , Germany ) was used to plot the change over time in fluorescence intensities of GFP signals co-localised with Cy3-RNA signal on microtubules ( after background subtraction ) . The number of photobleaching steps ( discrete decay steps of fluorescent signal intensity ) in each trace that reached baseline levels of fluorescence was determined manually , following published procedures , after a Chung-Kennedy filter was applied on the traces as described ( Chung and Kennedy , 1991; Ulbrich and Isacoff , 2008; Jain et al . , 2011; Badrinarayanan et al . , 2012 ) . We confirmed the accuracy of manual scoring by analysing a subset of the traces using the step-fitting algorithm StepFinder ( Kerssemakers et al . , 2006 ) . Fluorescence traces showing a sudden , large drop in intensity to basal levels were not used . These events , which happened very rarely , could indicate dissociation of the RNA-motor complex from the microtubule . Traces showing excessive fluorescence fluctuation were also excluded from analysis . Background fluorescence values were determined by averaging the fluorescence value from 8–10 randomly chosen pixels in the field-of-view . To estimate the copy numbers of kin401T99N:mGFP or tau23 per diffraction-limited spot , proteins were introduced into an imaging chamber ( in the absence of ATP ) containing HiLyte 647-labelled microtubules bound to the glass , together with oxygen scavenging agent ( at the same concentration as above ) . Images were captured ( 488 nm excitation , 136 ms capture [100 ms exposure +36 ms image acquisition] ) until the fluorescent signal reached basal levels ( indicating photobleaching or detachment ) . Estimation of the number of GFP photobleaching steps was performed manually as describe above , with the same criteria used for exclusion of a subset of traces from the analysis . To evaluate the ratio of GFP-labelled to unlabelled Dlic in extracts of GFP::Dlic embryos we performed immunoblotting with a mouse anti-Dlic P5F5 antibody ( Mische et al . , 2008 ) ( a gift from T Hays , Minnesota University , USA; used at a 1:1500 dilution ) . Signal was detected , used Alexa488-conjugated anti-mouse secondary antibodies ( 1:1000; Invitrogen ) , and a Typhoon imaging system ( GE Healthcare , UK ) . Quantification of the ratio of unlabelled Dlic to GFP-labelled Dlic was performed using ImageJ ( http://rsb . info . nih . gov/ij/ ) following background subtraction . Data plotting and curve fitting was performed with GraphPad ( La Jolla , CA ) Prism 6 and Matlab R2012b ( Mathworks , Natick , MA ) . Evaluations of statistical significance are described in the appropriate legend . | For a cell to do its job , the different components inside it need to be moved to different locations . This is achieved by an elaborate cellular transport system . To move a component to where it needs to be , motor proteins bind to it , often with the assistance of other ‘accessory’ proteins . This cargo-motor complex then moves along a network of tracks within the cell . Viruses also exploit this transport system in order to be trafficked to specific parts of the cell during their life cycles . Many cargos are moved along microtubule tracks . Multiple microtubule motor proteins often attach to the same cargo , but it is unclear how they work together during transport . Previous studies have attempted to address this issue by attaching motor proteins to artificial cargoes , such as synthetic beads . However , these experiments did not include some of the accessory proteins that are thought to play a role during transport within the living cell . Soundararajan and Bullock have now examined how complexes containing multiple motors bound to accessory proteins move molecules of messenger RNA to specific sites within cells . By visualising fruit fly mRNA moving along microtubules attached to a glass surface , the transport process can be studied in detail . It appears that the complexes travel using one of two methods: they either diffuse along the microtubules , which they can do in either direction , or they power themselves along the microtubules , which they can only do in one direction . Although previous experiments with artificial cargos suggested that the number of motors in the complex determines the likelihood of one-way traffic , it appears that one or more accessory proteins are actually in control during mRNA transport . Soundararajan and Bullock also documented how the mRNA-motor complexes react to roadblocks and dead-ends on the microtubule highway . Rather than letting go of the microtubule upon such an encounter , the complexes can reverse back down the track . This behaviour may help them to find a new route to their destination . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2014 | The influence of dynein processivity control, MAPs, and microtubule ends on directional movement of a localising mRNA |
Piezo ion channels are activated by various types of mechanical stimuli and function as biological pressure sensors in both vertebrates and invertebrates . To date , mechanical stimuli are the only means to activate Piezo ion channels and whether other modes of activation exist is not known . In this study , we screened ∼3 . 25 million compounds using a cell-based fluorescence assay and identified a synthetic small molecule we termed Yoda1 that acts as an agonist for both human and mouse Piezo1 . Functional studies in cells revealed that Yoda1 affects the sensitivity and the inactivation kinetics of mechanically induced responses . Characterization of Yoda1 in artificial droplet lipid bilayers showed that Yoda1 activates purified Piezo1 channels in the absence of other cellular components . Our studies demonstrate that Piezo1 is amenable to chemical activation and raise the possibility that endogenous Piezo1 agonists might exist . Yoda1 will serve as a key tool compound to study Piezo1 regulation and function .
Mechanotransduction describes processes by which mechanical forces are converted into biological responses . Mechanotransduction is essential for physiological functions including the sense of touch , hearing , and blood pressure regulation . The molecular mechanisms involved in mechanotransduction have been largely unknown , but mechanically activated cation channels are postulated to play important roles ( Delmas et al . , 2011 ) . Piezo1 and Piezo2 are necessary and sufficient for mechanically activated cation channel activity ( Coste et al . , 2010 ) . These proteins are expressed in various mechanically sensitive cell types , and Piezo1 and Piezo2 have recently been shown to be required for vascular development and touch sensing , respectively ( Li et al . , 2014; Maksimovic et al . , 2014; Ranade et al . , 2014a; Ranade et al . , 2014b; Woo et al . , 2014 ) . In humans , Piezo1 gain-of-function mutations are associated with a hereditary red blood cell condition termed dehydrated hereditary stomatocytosis , while Piezo2 gain-of-function mutations are associated with three phenotypically overlapping conditions termed distal arthrogryposis type 5 , Gordon Syndrome , and Marden–Walker Syndrome ( Piezo2 ) ( Albuisson et al . , 2013; Andolfo et al . , 2013; Bae et al . , 2013; Coste et al . , 2013; McMillin et al . , 2014 ) . Piezo proteins form a distinct class of proteins with no apparent sequence similarity to other proteins and channels ( Coste et al . , 2010; Bae et al . , 2011; Nilius and Honore , 2012 ) . They typically consist of >2000 amino acids with ∼30–40 putative transmembrane segments , and Piezo1 has been shown to assemble as a homotetramer of ∼1 . 2 million daltons . Purified Piezo1 can be reconstituted in lipid bilayers resulting in spontaneous cation currents . This indicates that Piezos constitute channel-forming proteins , as opposed to accessory subunits ( Coste et al . , 2012 ) . Both vertebrate and invertebrate Piezo channels can be activated by mechanical stimuli suggesting an evolutionarily conserved gating mechanism geared to transduce mechanical force ( Kim et al . , 2012 ) . Indeed , to date , mechanical stimuli are the only means to activate Piezo ion channels . In comparison , temperature-activated transient receptor potential ( TRP ) ion channels are polymodal and are the sensors of various chemicals that cause a burning sensation such as capsaicin and mustard oil , as well as endogenous compounds that cause inflammation ( Julius , 2013 ) . Studies on the chemical activation of TRP channels have been crucial to understand the physiological role of these channels and have contributed to mechanistic appreciation of how these ion channels are gated ( Jordt and Julius , 2002; Bandell et al . , 2006; Macpherson et al . , 2007; Cao et al . , 2013; Julius , 2013 ) . The discovery of a chemical agonist of Piezo channels could thus benefit the study of mechanotransduction .
We set out to probe Piezos for chemical-mediated activation . As Piezos are calcium-permeable channels , we hypothesized that Piezo activity could be monitored using calcium-sensitive fluorophores . We tested this by overexpressing Piezo1 in human embryonic kidney ( HEK ) cells and monitoring intracellular calcium in response to pressure exerted on the cell via a blunt glass probe . In Piezo1 expressing cells , a sequence of mechanical stimulations using a piezo-electric driven probe caused reversible calcium responses that increased with increasing probe displacement until ultimately an irreversible Ca2+ signal ensued as a consequence of membrane rupture . In contrast , control cells showed no reversible responses and only irreversible calcium flux upon cell perforation was observed . ( Figure 1A ) . This suggested that Piezo1-mediated calcium flux could be measured via calcium fluorophores . With the objective to identify either a Piezo1 or Piezo2 agonist , we co-transfected HEK cells with mPiezo1 and mPiezo2 cDNAs and screened a collection of ∼3 . 25 million low molecular weight ( LMW ) compounds for their ability to induce calcium influx in these cells . This led us to identify a synthetic compound that elicits Ca2+ flux in Piezo1- but not vector-transfected cells; we named this compound Yoda1 ( see ‘Materials and methods’ and Figure 1D for screen and Yoda1 details ) . Yoda1-induced calcium responses depended largely on calcium influx as the chelation of extracellular calcium dramatically reduced the responses , while the depletion of intracellular calcium stores using thapsigargin did not ( Figure 1B ) . Still , calcium-chelating conditions did not completely abolish the responses , raising the possibility of some functional Piezo1 in intracellular membranes upon overexpression . Concentration-response experiments showed that Yoda1 at micromolar concentrations induced robust Ca2+ responses in cells transfected with either human or mouse Piezo1 , but not Piezo2-transfected cells , indicating its selectivity for Piezo1 ( Figure 1C ) . At higher Yoda1 concentrations ( >∼20 μM ) , the solutions became increasingly opaque . Therefore , the apparent EC50 is likely affected by compound insolubility and may not allow meaningful interpretation . We further tested six distinct hPiezo1 mutants that we previously identified in xerocytosis patients and which exhibited increased mechanically induced currents ( Albuisson et al . , 2013 ) . Invariably , the Yoda1-induced calcium responses were bigger in cells transfected with these mutant channels than cells transfected with wild-type Piezo1 , consistent with their gain-of-function phenotype ( Figure 1—figure supplement 1 ) . The effect of Yoda1 appears to critically depend on the dichloro substitution ( Figure 1D ) , as similar compounds present in the collection lacking the chlorines were not identified in the screen ( data not shown ) . Furthermore , the oxidation state of the thioether group appears critical as no activity could be observed with the sulfone analog ( tested at ≤30 μM , data not shown ) . 10 . 7554/eLife . 07369 . 003Figure 1 . A high-throughput screen identifies a Piezo1 activating chemical , Yoda1 . ( A ) mPiezo1 mediates Ca2+ influx upon mechanical activation . Ratiometric Ca2+ imaging ( Fura-2 ) of human embryonic kidney ( HEK ) 293T cells transiently transfected with Piezo1 or untransfected . Cells were subjected to a series of mechanical stimuli , by pressing a glass probe briefly onto the cell surface for 150 ms ( arrows ) . For each consecutive stimulus , the travel distance of probe was increased by 1 μm ( B ) Yoda1 ( 25 μM ) mediates Ca2+ responses ( 384-well FLIPR ) in HEK cells transiently transfected with mPiezo1 . When indicated , extracellular calcium was chelated by addition of EGTA , or cells were pretreated with thapsigargin to deplete intracellular calcium stores . Traces represent average ± SEM fluorescence of four wells . ( C ) Concentration-response profiles of mouse and human Piezo1 and Piezo2 , transfected HEK293T cells assayed using FLIPR suggesting apparent EC50 of 17 . 1 and 26 . 6 μM for mouse and human Piezo1 , respectively ( 95% confidence interval: 13 . 4 to 21 . 9 , and 20 . 6 to 34 . 4 ) , however , compound ( in ) solubility precludes meaningful conclusions with respect to EC50 ( see text ) . ( D ) Chemical structure of Yoda1 . The functional groups tested chlorines and thioether are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 07369 . 00310 . 7554/eLife . 07369 . 004Figure 1—figure supplement 1 . Piezo1 gain-of-function mutations show increased Yoda1 responses . Dose-response curves of HEK293T cells transiently transfected with hPiezo1 and indicated mutants ( Albuisson et al . , 2013 ) . Calcium responses were determined in response to increasing concentration of Yoda1 by means of 384-well FLIPR . Average fluorescence change ± SEM in four wells is plotted and fitted to a sigmoidal dose-response curve . Please note that , compared to Figure 1C , HEK cells here exhibited a more pronounced background response to Yoda1 ( see vector control ) at higher concentrations . We do not know the cause of such variation in responses . DOI: http://dx . doi . org/10 . 7554/eLife . 07369 . 004 We next sought to assess the effect of Yoda1 on mPiezo1 channel function directly by recording mPiezo1-mediated currents , both in the presence and absence of mechanical force ( Figure 2A–E ) . In HEK cells transiently transfected with mPiezo1 , using a cell-attached patch configuration , currents were measured before and during a series of increasing negative pressures applied via the recording pipette . The presence of Yoda1 caused multiple distinct effects . Firstly , Yoda1 caused a dramatic change in the kinetics of the mechanical responses , as it notably slowed the inactivation phase of the transient currents ( Figure 2A ) . Secondly , Yoda1 sensitized mPiezo1 to activation by pressure as indicated by a leftward shift in the current–pressure relationship , reducing the half maximal activation pressure ( P50 ) by about 15 mm Hg ( Figure 2B , C ) . Lastly , in the absence of negative pressure , we observed small currents in Yoda1-exposed patches . The Yoda1-dependent currents were a fraction of those attained by stretch: 9 . 0 ± 2 . 2% of the maximal attainable current compared to 1 . 5 ± 0 . 4% in control patches without Yoda1 ( Figure 2B , D , E ) . These results suggest that Yoda1 both modifies Piezo1 mechanotransduction currents and partially activates Piezo1 in the absence of externally applied pressure ( note , however , that some membrane tension exists in cell-attached patches even prior to the application of negative pressure ) The partial activation of Piezo1 by Yoda1 might be due to a variety of reasons , including ( 1 ) an indirect mechanism of action ( but see below ) , ( 2 ) an inefficacious gating mechanism ( i . e . , acting as a gating modifier instead of a full agonist ) , or ( 3 ) a very high actual EC50 ( see comment on insolubility above ) . We also tested the effect of Yoda1 ( 10 μM ) in whole-cell configuration , where mechanical pressure can be applied using a piezoelectric-driven glass probe . This concentration did not lead to measurable mPiezo1 currents in the absence of pressure but did cause a clear slowing of the inactivation phase of the mechanically activated currents , similar to the cell-attached patch experiments ( Figure 2F , G ) . No such change in kinetics could be observed for Piezo2 consistent with Yoda1 having a Piezo1 selective effect ( Figure 2H ) . 10 . 7554/eLife . 07369 . 005Figure 2 . Yoda1 functions as a gating modifier of Piezo1 . ( A–E ) mPiezo1-transfected HEK293T cells , cell-attached patch configuration . ( A ) Typical recordings of stretch-activated currents at −80 mV in two mPiezo1-transfected cells with or without 30 μM Yoda1 in the patch pipette . Negative pressure pulses from 0 to −80 mm Hg are applied for 500 ms every 15 s . ( B ) Average normalized current–pressure relationships from mPiezo1-transfected cell recordings with or without 30 μM Yoda1 in the patch pipette ( n = 8 and 12 , respectively ) . ( C ) Average P50 values from individual cells used for panel B ( p < 0 . 05 , Mann–Whitney t-test ) . ( D ) High magnification of recording traces shown in panel A in the absence of stretch stimulation . Left panels are full-trace histograms . ( E ) Average current without stretch stimulation normalized to maximal stretch-activated current from mPiezo1-transfected cells recorded at −80 mV with or without 30 μM Yoda1 in the patch pipette ( n = 8 and 12 , respectively; p < 0 . 05 , Mann–Whitney t-test ) . ( F–H ) mPiezo1- and mPiezo2-transfected HEK293T cells , whole-cell configuration . ( F ) Stimulus displacement in 0 . 5-μm increments every 10 s before ( black trace ) and 1–2 min after bath application of 10 μM Yoda1 ( red trace ) . A 20-mV step was applied in the beginning of each sweep ( sweeps are concatenated and hack marks indicate ∼10 s ) to monitor membrane ( Rm ) and access ( Ra ) resistance . ( G ) The fold change in the inactivation time constant indicates a significant slowing of inactivation during Yoda1 exposure . The effect was completely reversible ( not shown ) . The baseline tau prior to Yoda1 exposure was 16 . 5 ± 1 . 5 ms ( n = 5 ) ( H ) . No effect was observed upon Yoda1 exposure ( up to 5 min ) to the mechanically activated currents elicited in a cell expressing mPiezo2 . Fold change in inactivation time constant was 0 . 89- , 1 . 19- , and 1 . 25-fold ( n = 3 ) . Dotted lines indicated 0 current level ( current traces ) and displacement at which cell was visibly touched ( top ) . *p < 0 . 005 , Mann–Whitney t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 07369 . 005 As discussed above , Yoda1 might act directly on Piezo1 or indirectly via other membrane or even intracellular mediators . To address this , we set out to test its effect on purified mPiezo1 reconstituted in droplet interface lipid bilayers ( DIBs ) . Previously , we reconstituted purified mPiezo1 channels into asymmetric DIBs containing DiPhytanoyl-sn-glycero-3-PhosphoCholine ( DPhPC ) and 1 , 2-dioleoyl-sn-glycero-3-phosphate ( DOPA ) and observed spontaneous cation-selective channel activity ( Coste et al . , 2012 ) . Here , to study mPiezo1 activation , we utilized symmetric DIBs made up of only DPhPC in which reconstituted mPiezo1 does not show constitutive activity ( Figure 3A ) . Application of 1 μM Yoda1-induced discernable single-channel currents with a conductance similar to what we previously observed for spontaneously active mPiezo1 in asymmetric bilayers ( Figure 3B ) . Higher Yoda1 concentrations yielded robust currents with a staircase-like appearance , indicating the presence of multiple ( 30–40 ) functional channels ( Figure 3C , D ) . 10 . 7554/eLife . 07369 . 006Figure 3 . Yoda1 activates mPiezo1 in a membrane-delimited fashion . ( A ) Electrical recordings of reconstituted mPiezo1 in the symmetric DiPhytanoyl-sn-glycero-3-PhosphoCholine ( DPhPC ) bilayers and corresponding all point current histograms without the application of Yoda1 . ( B ) Single-channel electrical recordings of reconstituted mPiezo1 in the symmetric DPhPC bilayers in the presence of 1 μM Yoda1 . The calculated single-channel conductance of outward currents from the corresponding all point current histograms is 98 ± 9 pS in 0 . 5 M KCl , 20 mM HEPES , pH 7 . 4 at V = 100 mV . ( C ) Macroscopic currents of mPiezo1 in the presence of 10 μM Yoda1 ( upper left panel ) followed by the injection of 30 μM blocker RR ( upper right panel ) . The lower left panel is an expansion of the record ( red line ) to highlight multiple-channel openings . The lower right panel shows a complete block of channel activity after 6 s of RR injection . ( D ) Maximum current obtained at the indicated concentrations of Yoda1 ( red bars ) and the subsequent block by RR ( black bars ) . Each concentration point is plotted ( red bars ) as the function of maximum currents obtained in an ‘n’ number of experiments at V = 100 mV . Error bars indicate standard deviation . Note the lack of Piezo activity either without Yoda1 ( n = 9 ) or without mPiezo1 ( n = 10 ) in the bilayers . When indicated , Yoda1 is reconstituted in the DPhPC liposomes prior to the bilayer formation . ( E ) Representative histograms of closed ( left graph ) and open ( right graph ) dwell times extracted from single-channel analysis of mPiezo1 in the presence of 1 μM Yoda1; τ1 closed = 3 ± 1 ms , τ2 closed = 57 ± 15 ms , and τ open = 55 ± 9 ms . ( F ) Representative histograms of closed ( left graph ) and open ( right graph ) dwell times extracted from single-channel analysis of mPiezo1 reconstituted in an asymmetric bilayers ( without Yoda1 ) ; τ1 closed = 5 ± 1 ms , τ2 closed = 47 ± 9 ms , and τ open = 13 ± 4 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 07369 . 006 Next , we assessed single-channel parameters of mPiezo1 reconstituted in symmetric DPhPC bilayers in the presence of 1 μM Yoda1 ( Figure 3E ) . The channel exhibited closed dwell-time distributions that fitted well to a two-component probability density function; τ1 closed = 3 ± 1 ms and τ2 closed = 57 ± 15 ms . The two time constants differ in duration by > 10-fold , a property of a bursting channel , where τ1 closed is the closed time within a burst , and τ2 closed is the closed time between bursts . The open dwell time is fitted to a single-component probability density function with characteristic mean open time τ open = 55 ± 9 ms . Although , no direct comparison can be made for mPiezo1 parameters with and without Yoda1 ( due to lack of mPiezo1 activity in symmetric bilayers without Yoda1 ) , we analyzed single-channel properties of mPiezo1 acquired in asymmetric bilayers without Yoda1 ( Figure 3F ) . In an asymmetric bilayer , the channel exhibited characteristic bursting pattern with two closed time distributions , τ1 closed = 5 ± 1 ms , τ2 closed = 47 ± 9 ms , and an open time distribution τ open = 13 ± 4 ms . The significant difference in τ open ( 13 ± 4 ms vs 55 ± 9 ms ) suggests that mPiezo1 remains open for longer times in the presence of Yoda1 . No significant change was observed in τ closed , suggesting that Yoda1 mainly stabilizes the open state rather than destabilizing the closed state . A detailed biophysical analysis will be required to understand the mechanism of action of Yoda1 . Our present analysis provides the first steps towards this goal . For instance , our lipid bilayer experiments suggest that Yoda1 does not require other proteins or specific lipid domains to exert an effect on Piezo1 . This suggests its effect is either directly on the channel or via long-range membrane-delimited effects , for instance through a change in membrane tension or curvature of the membrane . Indeed , compounds that modify membrane curvature are known to affect mechanically sensitive ion channels ( Patel et al . , 1998 ) . However , such compounds are typically amphipaths which Yoda1 is not ( Sheetz and Singer , 1974 ) . More importantly , the effect of Yoda1 appears governed by stringent structural requirements both on the side of the chemical and on the side of the channel ( as no effect on Piezo2 was observed ) , fitting a model in which Yoda1 directly interacts with Piezo1 . Our electrophysiological experiments in cells suggest that Yoda1 prominently affects the sensitivity and the inactivation kinetics of mechanically induced responses but at best causes a slight mPiezo1 activation in the absence of mechanical stimuli . In the bilayer system , we observed that Yoda1 stabilizes the open channel , potentially explaining the slowing of mPiezo1 inactivation kinetics observed in cells . However , we also observe prominent Yoda1-dependent calcium responses in cell culture and currents in artificial bilayers in the absence of externally applied forces . Therefore , the discrepancy in various levels of channel activity observed with different assays used here remains unexplained , and future in-depth understanding of mechanism of Yoda1 action on Piezo1 might shed light on these apparently disparate observations . Regardless , we show in an accompanying paper that Yoda1 causes Piezo1-dependent red blood cell dehydration , arguing for sufficient activation of the ion channel in the absence of external forces to cause a robust physiological impact . Irrespective of whether Yoda1 acts as a full activator or as a positive modulator , our results suggest that we have identified the first Piezo1 agonist . This finding is important from two perspectives . Firstly , our studies provide the first evidence of non-mechanical activation of a Piezo channel , suggesting that Piezo1 gating does not exclusively depend on changes in mechanical force . This is important , since it raises the possibility that endogenous agonists of Piezo1 exist which may play an important role in modulating mechanotransduction . Secondly , Yoda1 will provide a valuable tool to facilitate studies aimed at elucidating Piezo1 gating mechanism and exploring its functional significance in various biological processes ( see for instance accompanying paper ) .
Mechanically activated whole-cell currents at a holding potential of −80 mV were elicited by indentation by a blunt glass probe as described ( Coste et al . , 2010 ) . Application of vehicle or compounds was achieved by puffer pipette as described ( Dubin et al . , 1999 ) or bath application; results were similar and combined . Voltage ramp-induced currents were recorded as described ( Dubin et al . , 1999 ) . Single-channel acquisition and analysis were performed as described previously ( Coste et al . , 2012 ) . Segments of continuous recordings in the range of 50 s ≤ t ≤ 500 s were used for analysis . The currents were sampled at 20 kHz and filtered at 2 kHz . Additional offline filtering of 1 kHz was applied to the recordings for display . Conductance was determined by fitting a Gaussian curve to the single channel all point current amplitude histograms . Event detection was performed by time-course fitting with the segmental K means ( SKM ) implemented in QuB software . To avoid the detection of erroneous events , the receiver dead time ( td ) was set at 300 μs for all records . Therefore , transitions shorter than td were ignored; transitions longer than td were accepted as events . Open dwell-time distributions are fitted with a single-component probability density function , whereas closed dwell-time distributions are fitted with a two-component probability density function implemented in QuB . The calculated values are reported as mean ± standard deviation , n denote number of experiments . | Within our bodies , cells and tissues are constantly being pushed and pulled by their surrounding environment . These mechanical forces are then transformed into electrical or chemical signals by cells . This process is crucial for many biological structures , such as blood vessels , to develop correctly , and is also a key part of our senses of touch and hearing . In 2010 , researchers discovered a group of ion channels—proteins embedded in the membrane that surrounds a cell—that open up when a force is applied and allow ions such as calcium , potassium , and sodium to flow . This movement of ions generates the electrical response of the cell to the applied force . However , not much is known about how these ‘Piezo’ ion channels work . To investigate this , it is important to be able to precisely control how and when the Piezo channels open . Many other ion channels are studied by using small chemical compounds to activate them , but there were none that were known to act on Piezo proteins . Syeda et al . —including some of the researchers involved in the 2010 work—screened over three million compounds for their ability to cause calcium ions to flow into human cells to try to identify chemicals that activate the Piezo channels . This revealed one promising candidate named Yoda1 , which specifically activated Piezo1: a Piezo protein that had previously been linked to a role in blood vessel development in embryos . To investigate how Yoda1 activates Piezo1 , Syeda et al . placed Piezo1 in an artificial cell membrane that did not contain any other cellular components . When Yoda1 was added to this set up , the Piezo1 channels opened up . This suggests that Piezo1 and Yoda1 interact in a manner that does not require additional cellular components other than a cell membrane . Separate work by Cahalan , Lukacs et al . uses Yoda1 to reveal that Piezo1 helps to control the volume of red blood cells , showing that in the future , Yoda1 could be valuable in research that investigates the roles of Piezo1 . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"structural",
"biology",
"and",
"molecular",
"biophysics",
"neuroscience"
] | 2015 | Chemical activation of the mechanotransduction channel Piezo1 |
The immediate evolutionary space accessible to HIV is largely determined by how single amino acid mutations affect fitness . These mutational effects can shift as the virus evolves . However , the prevalence of such shifts in mutational effects remains unclear . Here , we quantify the effects on viral growth of all amino acid mutations to two HIV envelope ( Env ) proteins that differ at >100 residues . Most mutations similarly affect both Envs , but the amino acid preferences of a minority of sites have clearly shifted . These shifted sites usually prefer a specific amino acid in one Env , but tolerate many amino acids in the other . Surprisingly , shifts are only slightly enriched at sites that have substituted between the Envs—and many occur at residues that do not even contact substitutions . Therefore , long-range epistasis can unpredictably shift Env’s mutational tolerance during HIV evolution , although the amino acid preferences of most sites are conserved between moderately diverged viral strains .
HIV’s envelope ( Env ) protein evolves very rapidly . The major group of HIV-1 that is responsible for the current pandemic originated from a virus that entered the human population ∼100 years ago ( Sharp and Hahn , 2011; Worobey et al . , 2008; Faria et al . , 2014 ) . The descendants of this virus have evolved so rapidly that their Envs now have as little as 65% protein identity ( Lynch et al . , 2009 ) . For comparison , protein orthologs shared between humans and mice have only diverged to a median identity of 78% over 90 million years ( Waterston et al . , 2002; Hedges et al . , 2006 ) . Env’s rapid evolution has dire consequences for anti-HIV immunity , since it erodes the efficacy of most neutralizing antibodies ( Albert et al . , 1990; Wei et al . , 2003; Richman et al . , 2003; Burton et al . , 2005 ) . Because of this public-health importance , numerous studies have experimentally characterized aspects of the ‘evolutionary landscape’ that Env traverses . The immediate evolutionary space accessible to any given Env is largely defined by the effects on viral fitness of all single amino acid mutations to Env . Most mutational studies have measured how just a small number of these mutations affect viral growth in cell culture , although it has recently become possible to use deep mutational scanning to measure the effects of many ( Al-Mawsawi et al . , 2014; Duenas-Decamp et al . , 2016 ) or even all ( Haddox et al . , 2016 ) single amino acid mutations to an Env variant . But interpreting these studies in the context of Env evolution requires addressing a fundamental question: How informative are mutational studies of a single protein variant about constraints on long-term evolution ? During protein evolution , substitutions at one site can change the effect of mutations at other sites ( Natarajan et al . , 2013; Gong et al . , 2013; Harms and Thornton , 2014; Podgornaia and Laub , 2015; Starr and Thornton , 2016; Klink and Bazykin , 2017 ) . We will follow the nomenclature of ( Pollock et al . , 2012 ) to refer to these changes in mutational effects as shifts in a site’s amino acid preferences . Such shifts can accumulate as substitutions become entrenched via epistatic interactions with subsequent changes ( Starr et al . , 2017; Pollock et al . , 2012; Shah et al . , 2015; Bazykin , 2015 ) —although the magnitude of these shifts is usually limited ( Doud et al . , 2015; Chan et al . , 2017; Ashenberg et al . , 2013; Risso et al . , 2015 ) . Given that the Envs of circulating HIV strains represent a vast collection of homologs that often differ at >100 residues , shifts in amino acid preferences could make the outcome of any study highly dependent on the Env used . Indeed , epistasis among a few combinations of Env mutations has been experimentally demonstrated ( da Silva et al . , 2010 ) , and epistatic fitness landscapes have been computationally inferred for a variety of HIV proteins ( Kouyos et al . , 2012; Ferguson et al . , 2013; Mann et al . , 2014; Barton et al . , 2015 ) including Env ( Louie et al . , 2018 ) . However , the only protein-wide experimental studies of how amino acid preferences shift during evolution have examined proteins that are structurally far simpler than Env , which forms a large heavily glycosylated heterotrimeric complex that transitions through multiple conformational states ( Munro et al . , 2014; Ozorowski et al . , 2017 ) . Here , we use an improved version of a previously described deep mutational scanning strategy ( Haddox et al . , 2016 ) to measure the effects on viral growth of all single amino acid mutations to two transmitted-founder virus Envs that differ by >100 mutations . We compare these complete maps of mutational effects to identify sites that have shifted in their amino acid preferences between the Envs . Most sites show no detectable shifts , but 30 sites have clearly shifted preferences . These shifted sites usually prefer a specific amino acid in one Env but have shifted to tolerate many amino acids in the other Env . The shifted sites cluster in structure but are often distant from any amino acid substitutions that distinguish the two Envs , demonstrating the action of long-range epistasis . By aggregating our measurements for both Envs , we identify sites that evolve faster or slower in nature than expected given the functional constraints measured in the lab , probably due to pressure for immune evasion . Overall , our work provides complete across-strain maps of mutational effects that inform analyses of Env’s evolution and function .
The viruses most relevant to HIV’s long-term evolution are those which are transmitted from human-to-human . However , the only prior work that has measured how all Env amino acid mutations affect HIV growth is a study by some of us ( Haddox et al . , 2016 ) that used a late-stage lab-passaged CXCR4-tropic virus ( Peden et al . , 1991 ) . The properties of Env can vary substantially between such late-stage viruses and the transmitted-founder viruses relevant to HIV’s long-term evolution ( Sagar et al . , 2006; Wilen et al . , 2011; Parrish et al . , 2013; Ronen et al . , 2015 ) . We therefore selected Envs from two transmitted-founder viruses , BG505 . W6M . C2 . T332N and BF520 . W14M . C2 ( hereafter referred to as BG505 and BF520 ) , that were isolated from HIV-infected infants shortly after mother-to-child transmission ( Nduati et al . , 2000; Wu et al . , 2006; Goo et al . , 2014 ) . The BG505 Env has been extensively studied from a structural standpoint ( Julien et al . , 2013; Lyumkis et al . , 2013; Pancera et al . , 2014; Huang et al . , 2014; Sanders et al . , 2015; Stewart-Jones et al . , 2016; Gristick et al . , 2016 ) , and variants of this Env are being tested as vaccine immunogens ( Sanders et al . , 2013 , Sanders et al . , 2015; de Taeye et al . , 2015 ) . We used the T332N variant of BG505 Env because it has a common glycosylation site that is targeted by many anti-HIV antibodies ( Sanders et al . , 2013 ) . The BF520 Env was isolated from an infant who developed an early broad anti-HIV antibody response ( Goo et al . , 2014; Simonich et al . , 2016 ) . We have previously created comprehensive codon-mutant libraries of the BF520 Env and used them to map HIV antibody escape ( Dingens et al . , 2017 ) , but these BF520 libraries have not been characterized with respect to how mutations affect viral growth . Both BG505 and BF520 are from clade A of the major ( M ) group of HIV-1 . Figure 1 shows the phylogenetic relationship among these two Envs and other clade A sequences . BG505 and BF520 are identical at 721 of the 836 pairwise-alignable protein sites ( 86 . 2% identity ) . However , in our experiments we mutagenized only the ectodomain and transmembrane domain of Env , and excluded the signal peptide and cytoplasmic tail . The reason is that we measure how Env mutations affect viral growth , which is influenced both by the functionality of Env protein molecules and their expression level . Mutations in the signal peptide and cytoplasmic tail commonly affect Env expression level ( Chakrabarti et al . , 1989; Yuste et al . , 2004; Li et al . , 1994 ) , so we excluded these regions with the goal of reducing the degree to which we simply identified mutations that affected Env expression . In the ectodomain and transmembrane domains of Env , BG505 and BF520 are identical at 549 of the 616 sites ( 89 . 1% identity ) that are alignable across clade A Envs ( Figure 1—source data 1 , Figure 1—source data 2 ) . The divergence between BG505 and BF520 therefore offers ample opportunity to investigate mutational shifts during Env evolution . We have previously described a deep mutational scanning strategy for measuring how all amino acid mutations to Env affect HIV growth in cell culture , and applied this strategy to the late-stage lab-adapted LAI strain ( Haddox et al . , 2016 ) . Here , we made several modifications to this earlier strategy to apply it to transmitted-founder Envs and to reduce the experimental noise . This last consideration is especially important when comparing Envs , since it is only possible to reliably detect differences that exceed the magnitude of the experimental noise . Our modified deep mutational scanning strategy is in Figure 2A . This approach had the following substantive changes: instead of SupT1 cells , we used SupT1 . CCR5 cells ( SupT1 cells that express CCR5 in addition to CXCR4 [Boyd et al . , 2015] ) to support growth of viruses with transmitted-founder , CCR5-tropic Envs; we used more virions for the first passage ( ≥3×106 versus 5×105 infectious units per library ) to avoid bottlenecking library diversity; and rather than performing a full second passage we just did a short high-MOI infection to enable recovery of env genes from infectious virions without bottlenecking ( Figure 2A ) . We performed this deep mutational scanning in full biological triplicate for both BG505 and BF520 ( Figure 2B ) . Our libraries encompassed all codon mutations to all sites in Env except for the signal peptide and cytoplasmic tail . The deep mutational scanning effectively selected for functional Envs as evidenced by strong purifying selection against stop codons . Figure 3A shows the average frequency of mutations across Env in the plasmid mutant libraries , the mutant viruses , and wildtype controls as determined from the deep sequencing . The mutant viruses show clear selection against stop codons and many nonsynonymous mutations ( Figure 3A ) . This selection is more apparent if we correct for the background error rates estimated from the wild-type controls ( Figure 3—source data 1 ) . The error-corrected frequencies of stop codons drop to 3–16% of their original values ( Figure 3—source data 1 ) , with the residual stop codons probably due to some non-functional virions surviving due to complementation by other co-infecting virions . The error-corrected frequencies of nonsynonymous mutations also drop substantially ( 43%–49% of their original values ) , whereas the frequencies of synonymous mutations drop only slightly ( 85%–95% of their original values ) . These trends are consistent with the fact that nonsynonymous mutations are often deleterious , whereas synonymous mutations often ( Zanini and Neher , 2013 ) have only mild effects on viral growth . Figure 3A only summarizes one aspect of the deep mutational scanning data , but Supplementary file 1 and 2 contain detailed plots showing all aspects of the data ( read depth , per-site mutation rate , etc ) as generated by the dms_tools2 software ( Bloom , 2015 , https://jbloomlab . github . io/dms_tools2/ ) . We used the deep mutational scanning data to estimate the preference of each site in Env for each amino acid via the analysis method described in Bloom , 2015 ) . As graphically illustrated in Figure 2A , the preferences for each site are normalized to sum to one . Our libraries were mutagenized at 670 sites in BG505 and 662 sites in BF520 , so 670×20=13 , 400 and 662×20=13 , 240 preferences were estimated for each Env , respectively . The correlations between the preferences from different experimental replicates are in Figure 3B , and the preferences themselves are in Figure 3—source data 2 . These replicate-to-replicate correlations are substantially higher than those for the deep mutational scanning of LAI Env by ( Haddox et al . , 2016 ) , which had replicate-to-replicate Pearson correlations of only R=0 . 45 to 0 . 50 . While the replicates are well correlated across all replicates for both BG505 and BF520 , the replicates for BG505 are more correlated with each other than with replicates for BF520 , and vice versa ( Figure 3B , compare red and blue versus gray plots ) . This fact hints that there are some shifts in amino-acid preferences between the two Envs—something that is investigated with more statistical rigor later in this paper . Note also that there is a trend for highly preferred amino acids to be more strongly preferred in BG505 than BF520 ( most high-preference points in the gray plots in Figure 3B fall above the diagonal ) ; however , this trend does not necessarily reflect differences between the Envs . Rather , there were modest differences in the stringency of selection between our deep mutational scans of BG505 and BF520 ( Figure 3—source data 1 shows that purifying selection better purged stop codons in BG505 ) . In the next section , we correct for these experimental differences by calibrating each dataset to match the stringency of selection in nature . The most immediate question is how authentically the experimental measurements describe the actual selection on Env function in nature . Direct comparisons between experimentally measured amino acid preferences and amino acid frequencies in natural sequences are confounded by the fact that the natural sequences are evolutionarily related . This problem can be overcome by making the comparison in a phylogenetic context to account for the evolutionary relationships among sequences . Specifically , we used our deep mutational scanning data to construct experimentally informed codon models ( ExpCM’s ) for Env’s evolution . An ExpCM is a phylogenetic substitution model that incorporates the functional constraints measured in a deep mutational scanning experiment ( Hilton et al . , 2017 ) . If the experiment captures much of the actual evolutionary constraint on a gene , then an ExpCM will describe the gene’s natural evolution better than a standard phylogenetic codon substitution model . The reason is that standard codon substitution models ( Yang et al . , 2000 ) only model functional constraint via a single parameter that represents the rate of fixation of nonsynonymous protein-altering mutations relative to synonymous ones; this parameter is called dN/dS or ω . In contrast , an ExpCM accounts for the preference of each site for each of the 20 amino acids under the functional selection in the deep mutational scan , and then additionally adds an ω parameter that represents the relative rate of nonsynonymous to synonymous substitutions after accounting for these functional constraints ( Bloom , 2017; Hilton et al . , 2017 ) . Importantly , since we expect some sites in Env to be under diversifying selection from immunity , we extended the ExpCM’s described in Hilton et al . , 2017 ) to draw ω from a gamma distribution as is commonly done for codon-substitution models ( Yang et al . , 2000 ) . Table 1 shows that ExpCM’s informed by the deep mutational scanning of either BG505 or BF520 describe the natural evolution of Env vastly better than a standard codon substitution model . In addition to the improved fit of the ExpCM’s , we can also interpret the ω parameter . Recall that for standard codon substitution models , ω is simply the rate of fixation of nonsynonymous mutations relative to synonymous ones . For such models , the gene-wide average ω is almost always <1 , since purifying selection purges many functionally deleterious amino acid mutations even for adaptively evolving proteins ( Murrell et al . , 2015 ) . Indeed , Table 1 shows that Env’s gene-wide average ω is <one for a standard model . But for ExpCM’s , ω is the relative rate of nonsynonymous to synonymous substitutions after accounting for functional constraints measured in the deep mutational scanning ( Bloom , 2017 ) . For the ExpCM’s , the gene-wide average ω is >1 ( Table 1 ) , indicating that external selection ( e . g . from immunity ) drives Env to fix amino acid mutations faster than expected under a null model that only accounts for functional constraints on the protein . ExpCM’s also have a stringency parameter that relates selection in the experiments to that in nature . Essentially , this parameter indicates how strongly natural selection prefers the amino acids that are preferred in the deep mutational scanning ( Hilton et al . , 2017 ) . A stringency parameter >1 indicates that natural selection prefers the same amino acids as the experiments , but with greater stringency . Both ExpCM’s have stringency parameters >1 ( Table 1 ) —a finding that makes sense , since the stop-codon analysis in the previous section suggests that the experimental selections are more lax than natural selection on HIV . For the entire rest of the paper , we use the experimentally measured preferences re-scaled by the stringency parameters in Table 1 . The reason we do this is to distinguish genuine differences between the two Envs from mere variation in the strength of selection between the two sets of experiments . Re-scaling both sets of preferences to optimally describe Env evolution in nature is a principled way to standardize the measurements; see ( Hilton et al . , 2017 ) and the Materials and methods section entitled ‘Re-scaling the preferences’ for a more detailed explanation . A qualitative way to assess if the deep mutational scanning authentically describes selection on Env function is to visually compare the measurements with existing knowledge . Figure 4 and Figure 5 show the re-scaled across-replicate average of the amino acid preferences for each Env . At sites of known functional importance , these preferences are usually consistent with prior knowledge . For instance , residues T257 , D368 , E370 , W427 , and D457 are important for Env binding to CD4 ( Olshevsky et al . , 1990 ) , and all these amino acids are highly preferred in our deep mutational scanning ( Figure 4 and Figure 5 ) . Likewise , Env has 10 disulfide bonds ( linking sites 54–74 , 119–205 , 126–196 , 131–157 , 218–247 , 228–239 , 296–331 , 378–445 , 385–418 , and 598–604 ) , most of which are important for function ( van Anken et al . , 2008 ) —and the cysteines at these sites are highly preferred in our deep mutational scanning . The deep mutational scanning is also consistent with prior knowledge about sites that are tolerant of mutations . For instance , Env has five variable loops that mostly evolve under weak constraint in nature ( Starcich et al . , 1986; Zolla-Pazner and Cardozo , 2010 ) —and most sites in these loops are mutationally tolerant in our deep mutational scanning ( see sites indicated by gray overlay bars in Figure 4 and Figure 5 , such as 132 to 195 ) . It is beyond the scope of this paper to catalog associations between our measurements and all other prior mutational studies of Env , but the concordance of our findings with the above mutational studies , and the fact that our data improve phylogenetic models of Env’s natural evolution , suggest that our experiments do a reasonable job of authentically measuring functional selection on Env . The most fundamental question that we seek to address is how similar the amino acid preferences are between the two Envs . We have already noted that Figure 3B shows that the preferences are more correlated for replicate measurements on the same Env than for replicate measurements on different Envs . However , simply comparing correlation coefficients does not identify specific sites where mutational effects have shifted , nor does it quantify the magnitude of any shifts . We therefore used a more rigorous approach to identify sites where the amino acid preferences differ between BG505 and BF520 by an amount that exceeds the noise in our experiments . We first re-scaled the preferences from each experimental replicate by the stringency parameter for that Env from Table 1 to calibrate all measurements to the stringency of natural selection . We then identified the 659 sites in the mutagenized regions of Env that are pairwise alignable between BG505 and BF520 ( Figure 6—source data 1 ) . For each site , we calculated the shift in amino acid preferences between Envs using an approach similar to that of ( Doud et al . , 2015 ) as illustrated in Figure 6A . This approach calculates the magnitude of the shift after correcting for experimental noise by comparing the differences in preferences between replicates for BG505 and BF520 to the differences between replicates for the same Env . Figure 6A shows this calculation for a site that has not shifted ( site 598 , which strongly prefers cysteine in both Envs ) , the most shifted site ( 512 , which shifts from being mutationally tolerant in BG505 to strongly preferring alanine in BF520 ) , and two other sites with more intermediate behaviors . The overall distribution of shifts between BG505 and BF520 is shown in Figure 6B . Most sites have relatively small shifts ( close to zero ) , although there is a long tail of sites with large shifts . This tail reaches its upper value with site 512 , which has a shift of 0 . 52 out of a maximum possible of 1 . 0 . How should we interpret this distribution—have mutational effects shifted a lot , or not very much ? We can establish an upper-bound for how much sites might shift by comparing Env to a non-homologous protein . Figure 6B shows the distribution of shifts when comparing Env to influenza’s hemagglutinin protein , which has previously had its amino acid preferences measured by deep mutational scanning ( Doud and Bloom , 2016 ) . Most sites have large shifts between Env and hemagglutinin , with the typical shift being ∼0 . 4 and some approaching the maximum value of 1 . 0 . We can also establish a lower-bound by creating a null distribution for the expected shifts if all differences are simply due to experimental noise . This null distribution is created by randomizing the experimental replicates among Envs . Figure 6B shows that the null distribution is more peaked at zero than the real distribution , and does not have the same prominent tail of sites with large shifts . The answer to the question of how much mutational effects have shifted is therefore nuanced: they have substantially shifted at some sites , but remain vastly more similar between the two Envs than between two unrelated proteins . We can use the null distribution to identify sites where the shifts between BG505 and BF520 are significantly larger than the noise in our experiments ( Figure 6B ) . There are 30 such sites at a false discovery rate of 0 . 1 . Figure 6C shows the amino acid preferences of these significantly shifted sites for each Env . For the majority of shifted sites , one Env prefers a specific amino acid whereas the other Env tolerates many amino acids; for instance , see sites 512 , 516 , 599 , 165 , 605 and 505 in Figure 6C . Such broadening and narrowing of a site’s mutational tolerance is frequently linked to changes in protein stability , with a more stable protein typically being more mutationally tolerant ( Wang et al . , 2002; Bloom et al . , 2006; Gong et al . , 2013; Kumar et al . , 2017 ) . Work with engineered Env protein in the form of ‘SOSIP’ trimer ( Binley et al . , 2000; Sanders et al . , 2002 ) has shown that BG505 SOSIP is more thermostable than BF520 SOSIP ( Verkerke et al . , 2016 ) . Consistent with this fact , sites with altered mutational tolerance are often ( although not always , see sites 165 and 520 in Figure 6C ) more mutationally tolerant in BG505 . Differences in Env’s expression level might also contribute to a general broadening or narrowing of tolerance to subsequent mutations . The reason is that our experiments select for viral growth ( which is affected by both Env function and expression ) , so it is possible that some of the shifts are due to epistatic mutational effects on expression rather than function . However , not all of the significantly shifted sites show a simple pattern of broadening or narrowing of mutational tolerance . For instance , site 288 does not alter its mutational tolerance but rather flips its rather narrow amino acid preference from phenylalanine in BG505 to leucine in BF520 ( Figure 6C ) . Thus , there is variation in both the extent and types of shifts observed . What distinguishes the sites that have undergone significant shifts ? First , we analyzed the distribution of shifted sites in context of Env’s three-dimensional structure . Env’s structure is highly conformationally dynamic and undergoes large changes upon receptor binding and membrane fusion . In an effort to account for these dynamics , we examined multiple conformational states of Env: the closed pre-fusion state ( Stewart-Jones et al . , 2016 ) , the open CD4-bound state ( Ozorowski et al . , 2017 ) , and the post-fusion six-helix bundle ( Weissenhorn et al . , 1997 ) . Figure 7A shows the locations of the shifted sites on the crystal structure of Env in the closed pre-fusion state . There is no visually obvious tendency for shifted sites to preferentially be on Env’s surface or in its core , and statistical analysis of both the closed and open states of Env ( Figure 7B ) finds no association between a site’s relative solvent accessibility and whether its amino acid preferences have shifted . We did not attempt to analyze the association between solvent accessibility and shift for the post-fusion six-helix bundle because crystal structures of this conformation only contain ∼80 Env residues ( Weissenhorn et al . , 1997; Chan et al . , 1997; Tan et al . , 1997 ) . However , Figure 7A does suggest that the sites of significant shifts tend to cluster in Env’s structure . A statistical analysis confirms that there is clustering of shifted sites for the closed and open conformations , with the effect being strongest when we define contacts based on the closest intra-residue distance across these two conformations ( Figure 7C ) . Therefore , the factors that drive shifts in Env’s mutational tolerance often affect physically interacting clusters of residues in a coordinated fashion . We also investigated clustering of shifted sites in the post-fusion six-helix bundle . Because structures of this conformation only resolve the coordinates of ∼80 residues , we did not perform a statistical analysis . However , a qualitative analysis revealed that three of the four shifted sites that are resolved in the post-fusion conformation cluster at one end of the helical bundle ( Figure 7—figure supplement 1 ) . An obvious hypothesis is that strongly shifted sites have substituted between BG505 and BF520 , or physically contact such substitutions . According to this hypothesis , substitutions would alter the local physicochemical environment of the substituted site and its neighbors , thereby shifting the amino acid preferences of sites in the physical cluster . But surprisingly , for both the closed and open conformations , the typical magnitude of shifts is not significantly larger at sites that have substituted , or at sites that contact sites that have experienced substitutions ( Figure 7C ) . For the six-helix bundle , there are five structurally resolved substituted sites , one of which is adjacent to the cluster of shifted sites ( Figure 7—figure supplement 1 ) . The number of resolved shifted and substituted sites in this structure is too small for a meaningful statistical analysis of the type in Figure 7D . However , the cluster of shifted and substituted sites in the six-helix bundle is also present in the closed and open states ( Figure 7—figure supplement 1 ) , and so is included in the statistical analyses in Figure 7D . There is a borderline trend for the significantly shifted sites to be more likely to have substituted between BG505 and BF520 ( Figure 7—source data 1 ) , but most shifted sites have not substituted ( only 8 of the 30 shifted sites differ in amino acid identity between the two Envs ) . The lack of strong enrichment in shifts at substituted sites contrasts with previous protein-wide experimental ( Doud et al . , 2015 ) and simulation-based ( Pollock et al . , 2012; Shah et al . , 2015 ) studies of shifting amino acid preferences , which found that shifts were dramatically more pronounced at sites of substitutions . The difference may arise because these earlier studies examined proteins that are fairly conformationally static ( absolutely so in the case of the simulations ) . The fact that Env is extremely complex and conformationally dynamic ( Munro et al . , 2014; Ozorowski et al . , 2017 ) may increase the opportunities for long-range epistasis to enable substitutions at one site to shift the amino acid preferences of distant sites . Indeed , many of the shifted sites cluster within regions of Env that are highly conformationally dynamic . Figure 7—figure supplement 2 shows the structural context of these clusters in finer detail . One cluster is at the trimer apex where two of Env’s variable loops pack against one another and against an adjacent protomer . These interactions are likely involved in regulating the transition between conformational states , and upon CD4 binding , these loops become highly disordered ( Guttman et al . , 2014; Ozorowski et al . , 2017 ) . Mutations at two of the shifted sites in this cluster ( 165 and 307 ) have been shown to cause Env to assume aberrant conformations , suggesting that these sites can strongly modulate Env’s dynamics ( Lee et al . , 2017 ) . Strikingly , this cluster of shifted sites may reflect previously observed differences in the conformational dynamics of this regions between these two Envs; the V2 region of BF520 SOSIP trimer is more accessible to deuterium exchange than the BG505 SOSIP trimer ( Verkerke et al . , 2016 ) . The other cluster of shifted sites is near a network of hydrophobic amino acids that has been proposed to help transmit the large-scale conformational change that takes place upon CD4 binding ( Ozorowski et al . , 2017 ) . One of the shifted sites ( site 69 ) overlaps with this network , and mutations at another ( site 64 ) have been shown to strongly modulate the relative stability of the open and closed conformations ( de Taeye et al . , 2015 ) . In total , these two clusters consist of nearly half of the shifted sites ( 13 out of 30 ) . One hypothesis why so many shifted sites cluster in these regions is that their dynamic nature allows long-range epistatic interactions to be readily propagated between substituted sites and distant shifted sites . It is difficult to discern exactly how these interactions might occur , but there is certainly a trend for sites that are conformationally dynamic to also be sites that show shifts in their amino acid preferences during evolution . One idea that has recently gained support in the protein-evolution field is that substitutions become ‘entrenched’ by subsequent evolution ( Pollock et al . , 2012; Shah et al . , 2015; Starr et al . , 2017 ) . Entrenchment is the tendency of a mutational reversion to become increasingly unfavorable as a sequence evolves . Given two homologs , if there is no entrenchment then the effect of mutating a site in the first homolog to its identity in the second will simply be the opposite of mutating the site in the second homolog to its identity in the first . But if there is entrenchment , then both mutations will be unfavorable , since the site is entrenched at its preferred identity in each homolog . Figure 8 shows the distribution of effects for mutating all sites that differ between BG505 and BF520 to the identity in the other Env . As expected under entrenchment , the average effect of these mutations is deleterious—although there are a substantial number of sites where the mutational flips are not deleterious . We can get some sense of the magnitude of the entrenchment by comparing the effects of the BG505↔BF520 mutations to the distribution of effects of all possible amino acid mutations ( Figure 8 ) . This comparison shows that even unfavorable inter-Env mutational flips are generally more favorable than random amino acid mutations . Therefore , entrenchment occurs for some but not all substitutions that distinguish BG505 and BF520 , and the magnitude of entrenchment is less than the effect of a typical random mutation . Entrenchment of substitutions therefore contributes to some of the mutational shifts . But given that many of these shifts occur at sites that do not even differ between the Envs ( Figure 7D ) , entrenchment of substitutions is clearly not the only cause of the shifting amino acid preferences . Our experiments measure the effects of mutations on viral growth in a T-cell line in the lab . But HIV actually evolves in humans , where additional selection pressures on Env are undoubtedly present . For instance , antibody pressure might increase the rate of evolution at some sites ( Albert et al . , 1990; Wei et al . , 2003; Richman et al . , 2003 ) , whereas pressure to mask certain epitopes ( Kwong et al . , 2002 ) might add constraint at other sites . Comparing selection in our experiments to natural selection can identify sites that are under such additional pressures during HIV’s actual evolution in humans . We determined whether each site in Env evolves faster or slower in nature than expected given three models: that evolution is purely neutral ( all nonsynonymous and synonymous mutations have equivalent effects ) , that sites are under the protein-level constraint measured in our experiments with BG505 , or that sites are under the constraint measured with BF520 . The first model used a standard dN/dS test ( Kosakovsky Pond and Frost , 2005 ) , whereas the other two models are conceptually similar but account for the experimentally measured amino acid preferences as described by ( Bloom , 2017 ) . All three models test if individual sites evolve faster or slower than expected , but they ‘expect’ different things: the dN/dS model expects nonsynonymous and synonymous mutations fix at the same rate , while the ExpCM expects the rate at a site to depend on the experimentally measured functional constraints . In all cases , the evidence that a site r evolves differently than expected is statistically summarized by the p-value that ωr is > or < 1 . The standard dN/dS model finds hundreds of sites that evolve slower than expected under neutral evolution ( Table 1 , ωr<1 ) , and only a handful of sites that evolve faster than expected under neutral evolution ( Table 1 , ωr>1 ) . This finding is unsurprising , since it is well known that Env is under functional constraint . In contrast , ExpCM’s that test the rates of evolution relative to the experimentally measured constraints find far fewer sites that evolve slower than expected , but many more sites that evolve faster ( Table 1 ) . The sites that evolve slower or faster than expected from the experiments are shown in Figure 9A , B , and overlaid on the logoplots in Figure 4 and Figure 5 as the ωr values . The identified sites are similar regardless of whether we use the experiments with BG505 or BF520 ( Figure 9C ) . The reason the results are similar for both experimental datasets is that ( as discussed above ) the amino acid preferences of most sites are similar in both Envs , suggesting that either dataset provides a reasonable approximation of the site-specific functional constraints across the clade A Envs in Figure 1 . What causes some sites to evolve faster or slower in nature than expected from the experiments ? The answer in both cases is likely to be immune selection . Most of the sites of faster-than-expected evolution are on the surface of Env ( Figure 9A , B and Figure 9—figure supplement 1 ) . Env’s escape from autologous neutralizing antibodies often involves amino acid substitutions in surface-exposed regions ( Moore et al . , 2009 ) , including at many of the sites that evolve faster than expected . Since our deep mutational scanning did not impose antibody pressure , sites where substitutions are antibody-driven will evolve faster in nature than expected from the experiments . Interestingly , immune selection also offers a plausible explanation for the sites that evolve slower than expected . In addition to escaping immunity via substitutions at antibody-binding footprints , Env is notorious for employing a range of more general strategies to reduce its susceptibility to antibodies . These strategies include shielding immunogenic regions with glycans ( Wei et al . , 2003; Stewart-Jones et al . , 2016; Gristick et al . , 2016 ) or hiding them by adopting a closed protein conformation ( Kwong et al . , 2002; Guttman et al . , 2015; Ozorowski et al . , 2017 ) . Sites that contribute to such general immune-evasion strategies will be under a constraint in nature that is not present in our experiments—and indeed , such sites evolve more slowly than expected from our experiments . For instance , we find very little selection to maintain most glycans in our cell-culture experiments . Of the 21 N-linked glycosylation sites shared between BG505 and BF520 , only four are under strong selection to maintain the glycan in our experiments—despite the fact that most are conserved in nature ( Figure 9C and Figure 9—figure supplement 2 ) . This finding concords with prior literature suggesting that these glycans are selected primarily for their role in immune evasion ( Pugach et al . , 2004; Wang et al . , 2013; Rathore et al . , 2017 ) . Similarly , a network of sites that help regulate Env’s transition between open and closed conformations that have different antibody susceptibilities ( Figure 9D ) also evolve slower in nature than expected from our experiments . Therefore , we can distinguish evolutionary patterns that are shaped by simple selection for Env function from those that are due to the additional complex pressures imposed during human infections .
We have experimentally measured the preference for each amino acid at each site in the ectodomain and transmembrane domain of two Envs under selection for viral growth in cell culture . These amino acid preference maps are generally consistent with prior knowledge about sites that are important for protein properties such as receptor binding or disulfide-mediated stability . However , the main value of these maps comes not from comparing them with prior knowledge , but from the fact that such prior knowledge encompasses just a small fraction of the vast mutational space available to Env . Because Env evolves so rapidly , every study of this protein must be placed in an evolutionary context , and our comprehensive amino acid preference maps potentially enable this in ways that prior piecemeal studies of mutations cannot . But these maps come with a potentially serious caveat: each one is measured for just a single Env variant . The major question that our study aimed to answer is whether the maps are still useful for evolutionary questions , or whether Env’s amino acid preferences shift so rapidly that each map only applies to the specific HIV strain for which it was measured . This question is reminiscent of one that was grappled with in the early days of protein crystallography , when it first became possible to build maps of a protein’s structure . Because it was not ( and is still not ) possible to crystallize every variant of a protein , it was necessary to determine whether protein structures could be usefully generalized among homologs . Fortunately for the utility of structural biology , it soon became apparent that closely homologous proteins have similar structures ( Chothia and Lesk , 1986; Sander and Schneider , 1991 ) . This rough generalizability of protein structures holds even for a protein as conformationally complex as Env—for although there are many examples of mutations that alter aspects of Env’s conformation and dynamics ( Kwong et al . , 2000; White et al . , 2010; Almond et al . , 2010; Davenport et al . , 2013 ) , SOSIP trimer structures from diverse Env strains remain highly similar in most respects ( Julien et al . , 2015; Pugach et al . , 2015; Stewart-Jones et al . , 2016; Verkerke et al . , 2016; Gristick et al . , 2016 ) . Our results show that amino acid preference maps of Env also have a useful level of conservation for many purposes . From a qualitative perspective , the amino acid preferences look mostly similar between BG505 and BF520 , and so provide a valuable reference for estimating which mutations are likely to be tolerated at each site in diverse HIV strains . Indeed , we anticipate that the complete maps of mutational effects in Figure 4 and Figure 5 will be useful for future sequence-structure-function studies . From an analytical perspective , a powerful use of our maps is to identify sites that evolve differently in nature than is required by the simple selection for viral growth imposed in our experiments—and the identified sites are largely the same regardless of whether the analysis uses an amino-acid preference map from BG505 or BF520 . Of course , from the perspective of protein evolution , the most interesting sites are the exceptions to the general conservation of amino-acid preferences . Consistent with studies of other proteins ( Natarajan et al . , 2013; Harms and Thornton , 2014; Doud et al . , 2015; Starr et al . , 2017 ) , we find a subset of sites that change markedly in which mutations they tolerate . Some shifted sites simply accommodate more amino acids in the more stable BG505 Env—a type of shift that has been well-documented for other proteins ( Wang et al . , 2002; Bloom et al . , 2006; Gong et al . , 2013; Kumar et al . , 2017 ) . But interestingly , there is no strong trend for shifts to be enhanced at sites that differ between BG505 and BF520 . Recent studies of protein evolution have focused on the idea that substitutions become ‘entrenched’ as sites shift to accommodate new amino acids ( Pollock et al . , 2012; Shah et al . , 2015; Bazykin , 2015; Starr et al . , 2017 ) . Indeed , a prior protein-wide comparison of amino acid preferences across homologs of influenza nucleoprotein found a significant enrichment of shifts at sites of substitutions ( Doud et al . , 2015 ) . But although there is some entrenchment of differences between BG505 and BF520 , this is not the major factor behind the shifts in amino acid preferences: the majority of sites that have shifted between BG505 and BF520 actually have the same wild-type amino acid in both Envs even though the preferences have shifted . This rather surprising result might be due to Env’s exceptional conformational complexity—mutations can cause long-range alterations in Env’s conformation ( Kwong et al . , 2000; White et al . , 2010; Almond et al . , 2010; Davenport et al . , 2013 ) , so it seems plausible that they might also shift mutational tolerance at distant sites . Regardless of the exact mechanism , our large-scale datasets of mutational effects in multiple viral strains should be useful for efforts to computationally parameterize ‘fitness landscapes’ of Env ( Kouyos et al . , 2012; Ferguson et al . , 2013; Mann et al . , 2014; Barton et al . , 2015; Louie et al . , 2018 ) . Our experiments provide highly quantitative data on the mutational tolerance of Env under selection for viral growth in cell culture . These data are amenable to rigorous functional and evolutionary analyses . Here , we have shown how these data can be compared between Envs to identify sites where mutational tolerance shifts with viral genotype , or between experiments and nature to identify sites under different pressure in the lab and in humans . Future experiments that modulate selection pressures in other relevant ways should provide further insight into the forces that drive and constrain HIV’s evolution .
Our codon mutant libraries mutagenized all sites in env to all 64 codons , except that the signal peptide and cytoplasmic tail were not mutagenized . The rationale for excluding these regions is that they are not part of Env’s ectodomain and are prone to mutations that strongly modulate Env’s expression level ( Chakrabarti et al . , 1989; Yuste et al . , 2004; Li et al . , 1994 ) . The codon-mutant libraries were generated using the approach originally described in ( Bloom , 2014a ) , with the modification of ( Dingens et al . , 2017 ) to ensure more uniform primer melting temperatures . The computer script used to design the mutagenesis primers ( along with some detailed implementation notes ) is at https://github . com/jbloomlab/CodonTilingPrimers . For BF520 , the three libraries are the same ones described by ( Dingens et al . , 2017 ) . For BG505 , we created three libraries for this study . The wild-type BG505 sequence used for these libraries is in Supplemental file 3 . The BG505 mutagenesis primers are in Supplemental file 4 . The end primers for the BG505 mutagenesis were: 5’-tgaaggcaaaactactggtccgtctcgagcagaagacagtggcaatgaga-3’ and 5’-gctacaaatgcatataacagcgtctcattctttccctaacctcaggcca-3’ . As with BF520 , we cloned the BG505 env libraries into the env locus of the full-length proviral genome of HIV strain Q23 ( Poss and Overbaugh , 1999 ) using the high-efficiency cloning vector described in ( Dingens et al . , 2017 ) . For this cloning , we digested the cloning vector with BsmBI , and then used PCR to elongate the amplicons to include 30 nucleotides at each end that were identical in sequence to the ends of the BsmBI-digested vector . The primers for this PCR were: 5’-agataggttaattgagagaataagagaaagagcagaagacagtggcaatgagagtgatgg-3’ and 5’-ctcctggtgctgctggaggggcacgtctcattctttccctaacctcaggccatcc-3’ . Next , we used NEBuilder HiFi DNA Assembly ( NEB , E2621S ) to clone the env amplicons into the BsmBI-digested plasmids . We purified the assembled products using Agencourt AMPure XP beads ( Beckman Coulter , A63880 ) using a bead-to-sample ratio of 1 . 5 , and then transformed the purified products into Stellar electrocompetent cells ( Takara , 636765 ) . The transformations yielded between 1 . 5 and 3 . 6 million unique clones for each of the three replicate libraries , as estimated by plating 1:2000 dilutions of the transformations . We scraped the plated colonies and maxiprepped the plasmid DNA; unlike in ( Dingens et al . , 2017 ) , we did not include a 4 hr outgrowth step after the scraping step . For the wild-type controls , we maxiprepped three independent cultures of wildtype BG505 env cloned into the same Q23 proviral plasmid . See Figure 2—figure supplement 1 and Figure 3A for information on the average mutation rate in these libraries as estimated by Sanger sequencing and deep sequencing , respectively . For BG505 , we generated mutant virus libraries from the proviral plasmid libraries by transfecting 293 T cells in three 6-well plates ( so 18 wells total per library ) with a per-well mixture of 2 μg plasmid DNA , 6 μl FuGENE 6 Transfection Reagent ( Promega , E269A ) , and 100 μl DMEM . The 293 T cells were seeded at 5×105 cells/well in D10 media ( DMEM supplemented with 10% FBS , 1% 200 mM L-glutamine , and 1% of a solution of 10 , 000 units/mL penicillin and 10 , 000 μg/mL streptomycin ) the day before transfection , such that they were approximately 50% confluent at the time of transfection . In parallel , we generated wildtype viruses by transfecting one six-well plate of 293 T cells with each wildtype plasmid replicate . At 2 days post-transfection , we harvested the transfection supernatant , passed it through a 0 . 2 μm filter to remove cells , treated the supernatant with DNAse to digest residual plasmid DNA as in ( Haddox et al . , 2016 ) , and froze aliquots at −80∘C . We thawed and titered aliquots using the TZM-bl assay in the presence of 10 μg/mL DEAE-dextran as described in ( Dingens et al . , 2017 ) . We conducted the low MOI viral passage illustrated in Figure 2A in SupT1 . CCR5 cells ( obtained from Dr . James Hoxie; Boyd et al . , 2015 ) . The SupT1 . CCR5 cells tested negative for mycoplasma . The SupT1 . CCR5 cell line was previously created by engineering the parental SupT1 cell line to express CCR5 ( Boyd et al . , 2015 ) . We used antibody staining followed by flow cytometry to validate that our stock of SupT1 . CCR5 cells expressed CCR5 , CXCR4 , and CD4 . There is no validated STR profile for SupT1 . CCR5 cells . However , we performed STR profiling on our stock of cells and compared the results to the ATCC SupT1 ( ATCC #CRL-1942 ) reference profile . We found that 11 of 14 alleles plus both amelogenin alleles matched the reference , with no additional mismatched alleles ins the SupT1 . CCR5 profile . Given the known instability of lymphoma cell lines ( Inoue et al . , 2000 ) , this level of identity suggests that the SupT1 . CCR5 cells are indeed related to the parental SupT1 cells ( Capes-Davis et al . , 2013 ) During this passage , cells were maintained in R10 media , which has the same composition as the D10 described above , except RPMI-1640 ( GE Healthcare Life Sciences , SH30255 . 01 ) is used in the place of DMEM . In addition , the media contained 10 μg/mL DEAE-dextran to enhance viral infection . We infected cells with 4 million ( for replicate 1 ) or 5 million ( for replicates 2 and 3 ) TZM-bl infectious units of mutant virus at an MOI of 0 . 01 , with cells at a starting concentration of 1 million cells/mL in vented tissue-culture flasks ( Fisher Scientific , 14-826-80 ) . At day one post-infection , we pelleted cells , aspirated the supernatant , and resuspended cell pellets in the same volume of fresh media still including the DEAE-dextran . At 2 days post-infection , we doubled the volume of each culture with fresh media still including DEAE-dextran . At 4 days post-infection , we pelleted cells , passed the viral supernatant through a 0 . 2 μm filter , concentrated the virus ∼30 fold using ultracentrifugation as described in ( Dingens et al . , 2017 ) , and froze aliquots at −80∘C . In parallel , for each replicate , we also passaged 2×105 ( for replicate 1 ) or 5×105 ( for replicates 2 and 3 ) TZM-bl infectious units of wildtype virus using the same procedure . To obtain final titers for our concentrated virus , we thawed one of the aliquots stored at −80∘C and titered using the TZM-bl assay in the presence of 10 μg/mL DEAE-dextran . For the final short-duration infection illustrated in Figure 2A , for each replicate we infected 106 TZM-bl infectious units into 106 SupT1 . CCR5 cells in the presence of 100 μg/mL DEAE-dextran ( note that this is a 10-fold higher concentration of DEAE-dextran than for the other steps , meaning that the effective MOI of infection is higher if DEAE-dextran has the expected effect of enhancing viral infection ) . Three hours post-infection , we pelleted the cells and resuspended them in fresh media without any DEAE-dextran . At 12 hr post-infection , we pelleted cells , washed them once with PBS , and then used a miniprep kit to harvest reverse-transcribed unintegrated viral DNA ( Haddox et al . , 2016 ) . The generation , passaging and deep sequencing of BF520 was done in a highly similar fashion , except that we only had a single replicate of the wild-type control . Note that the final passaged BF520 mutant libraries analyzed here actually correspond to the ‘no-antibody’ controls described in ( Dingens et al . , 2017 ) , but that study did not analyze the initial plasmid mutant libraries relative to these passaged viruses , and so was not able to provide measurements of the amino acid preferences . We deep sequenced all of the samples shown in Figure 3A: the plasmid mutant libraries and wildtype plasmid controls , and the cDNA from the final mutant viruses and wildtype virus controls . In order to increase the sequence accuracy , we used a barcoded-subamplicon sequencing strategy . This general strategy was originally applied in the context of deep mutational scanning by Wu et al . ( 2014 ) , and the specific protocol used in our work is described in Doud and Bloom , 2016 ) ( see also https://jbloomlab . github . io/dms-tools2/bcsubamp . html ) . The primers used for BG505 are in Supplementary file 5 . The primers used for BF520 are in ( Dingens et al . , 2017 ) . The data generated by the Illumina deep sequencing are on the Sequence Read Archive under the accession numbers provided at the beginning of the Jupyter notebook in Supplementary file 1 and 2 . We analyzed the deep-sequencing data using the dms_tools2 software package ( Bloom , 2015 , https://jbloomlab . github . io/dms_tools2/ , version 2 . 2 . 4 ) . The algorithm that goes from the deep-sequencing counts to the amino acid preferences is that described in ( Bloom , 2015 ) ( see also https://jbloomlab . github . io/dms-tools2/prefs . html ) . A Jupyter notebook that performs the entire analysis including generation of most of the figures in this paper is in Supplementary file 1 . An HTML rendering of this notebook is in Supplementary file 2 . A repository containing all of this code is also available at https://github . com/jbloomlab/EnvMutationalShiftsPaper ( Haddox et al . , 2018 , copy archived at https://github . com/elifesciences-publications/EnvMutationalShiftsPaper ) . The Jupyter notebooks in Supplementary file 1 and 2 also contain numerous plots that summarize relevant aspects of the deep sequencing such as read depth , per-codon mutation frequency , mutation types , etc . Supplementary file 1 also contains text files and CSV files with the numerical values shown in these plots . Citations are also owed to weblogo ( Crooks et al . , 2004 ) and ggseqlogo ( Wagih , 2017 ) , which were used in the generation of the logoplots . A basic description of the process used to generate the clade A sequence alignment in tree-source data 1 , the alignment mask in tree-source data 1 , and the phylogenetic tree in tree are provided in the legend to that figure . An algorithmic description of how the alignment and tree were generated are in Supplementary file 1 and 2 . For fitting of the phylogenetic substitution models , we used Table 1 ( Hilton et al . , 2017 , http://jbloomlab . github . io/phydms/ , version 2 . 2 . 1 ) to optimize the substitution model parameters and branch lengths on the fixed tree topology intree . The Goldman-Yang ( or YNGKP ) model used in Table 1 is the M5 variant described by Yang et al . , 2000 ) , with the equilibrium codon frequencies determined empirically using the CF3 × 4 method ( Kosakovsky Pond et al . , 2010 ) . For the ExpCM shown in Table 1 , we extended the models with empirical nucleotide frequencies described in Hilton et al . , 2017 ) to also allow ω to be drawn from discrete gamma-distributed categories exactly as for the M5 model . These ExpCM with gamma-distributed ω were implemented in Table 1 using the equations provided by ( Yang , 1994 ) ( see also http://jbloomlab . github . io/phydms/implementation . html#models-with-a-gamma-distributed-model-parameter ) . The preferences were re-scaled by the stringency parameters in Table 1 as described in Hilton et al . , 2017 ) . For both the M5 model and the ExpCM with a gamma-distributed ω , we used four categories for the discretized gamma distribution . Table 1—source data 1 shows the results for a wider set of models than those used in Table 1 . These include the M0 model of ( Yang et al . , 2000 ) , ExpCM without a gamma-distributed ω , and ExpCM in which the amino acid preferences are averaged across sites as a control to ensure that the improved performance of these models is due to their site-specificity . Note how for these Env alignments , using a gamma-distributed ω is very important in order for the ExpCMs to outperform the M5 model—we suspect this is because there are many sites of strong diversifying selection . For detection of sites with faster or slower than expected evolution , we used the approach in ( Bloom , 2017 ) , which is exactly modeled on the FEL approach of ( Kosakovsky Pond and Frost , 2005 ) but extended to ExpCM . This approach estimates a p-value that ωr is not equal to one for each site r using a likelihood-ratio test . The actual point estimates of ωr are unreliable for individual sites due to the limited number of observations , so we report the p-value that ωr is not equal to one , which is a better indication of the strength of the statistical evidence for faster or slower than expected evolution ( Kosakovsky Pond and Frost , 2005; Murrell et al . , 2012 ) . For the Q-values and false discovery rate testing , we considered the tests for ωr>1 and ωr<1 separately . Supplementary file 1 and 2 contains the code that runs Table 1 to reproduce all of these analyses . The amino acid preferences that are directly extracted from the deep sequencing data essentially give the enrichment/depletion of each mutation , normalized to sum to one at each site ( Doud et al . , 2015 , https://jbloomlab . github . io/dms_tools2/prefs . html ) . However , the extent that any mutation is enriched or depleted is a combination of two factors: the inherent effect of that mutation , and the ‘stringency’ of the experimental selection . For instance , if the selection is weak , then deleterious mutations will only be slightly depleted; conversely , if selection is strong , then deleterious mutations will be greatly depleted . The fact that the preferences depend on the stringency of the experimental selection is important if we want to compare results between Envs . The reason is that our goal is to identify differences in the inherent effects of mutations between Envs , not simply find differences due to variation in experimental stringency . Of course , we have done our best to perform the experiments for BG505 and BF520 equivalently , but because these are different viruses with different growth rates , it is impossible to exactly match the experimental stringencies . This can be seen in Figure 3—source data 1 , which shows that stop codons were more depleted for BG505 than BF520 , indicating that selection in our experiments was more stringent for BG505 . How should we best re-scale the preferences ? Raising them to a power is a sensible approach . To see why , imagine a mutation that is depleted 3-fold after 2 rounds of viral growth . If our experiment instead allowed 22=4 rounds of viral growth , then the mutation would be depleted 32=9 -fold . More generally , if a mutation is enriched in frequency by ϕ-fold after n rounds of viral growth , then it will be enriched in frequency by ϕβ-fold after β×n rounds of viral growth . Since the amino acid preferences are conceptually equivalent to the re-normalized enrichments of mutations ( Bloom , 2015 ) https://jbloomlab . github . io/dms-tools2/prefs . html , it therefore makes sense that the re-scaled preference πr , a for amino acid a at site r should be related to the directly measured preference π^r , a by πr , a∝ ( π^r , a ) β . And indeed , this is exactly the re-scaling scheme described in ( Hilton et al . , 2017 ) that we use to re-scale our preferences for BG505 and BF520 . The last point is how to choose the re-scaling parameter β for each Env . It turns out that the features that we have described above for our experiments are also a feature of natural evolution: the expected frequency of a substitution during evolution depends not only on the inherent fitness effect of that mutation , but also on the effective population size , which is conceptually somewhat similar to the stringency of selection . It turns out that in a mutation-selection phylogenetic model of evolution , if the amino acid preferences are taken to represent the ‘fitness effects’ of mutations , then the exponential scaling parameter β is proportional to the effective population size ( Halpern and Bruno , 1998; McCandlish and Stoltzfus , 2014; Bloom , 2014b ) . Therefore , fitting the β parameter using a phylogenetic approach enables standardization of the preferences for the two Envs , and re-scales the preferences so that they best match with the actual stringency of selection observed in nature ( Hilton et al . , 2017 ) . Note that in practice this re-scaling scheme is roughly equivalent to a more heuristic approach that has been used by ( Gray et al . , 2017 ) and others . In this heuristic approach , the log-transformed enrichment ratios from different experiments are adjusted so that the distributions have equal spreads . Since multiplying log-transformed enrichment ratios is equivalent to exponentiating amino acid preferences , these two re-scaling procedures apply the same mathematical transformation . When identifying shifts in amino acid preferences between the two Envs , we needed a way to quantify differences between the Envs while accounting for the fact that our measurements are noisy . The approach we use is based closely on that of Doud et al . , 2015 ) and is illustrated graphically in Figure 6A . The RMSDcorrected value is our measure of the magnitude of the shift . Figure 6A , its legend , and the associated text completely explains these calculations with the following exception: they do not detail how the ‘distance’ between any two preference measurements was calculated . The distance between preferences at each site was simply defined as half of the sum of absolute value of the difference between preferences for each amino acid . Specifically , for a given site r , let πr , ai be the re-scaled preference for amino acid a in homolog i ( e . g . BG505 ) and let πr , aj be the re-scaled preference for that same amino acid in homolog j ( e . g . BF520 ) . Then the distance between the homologs at this site is simply Dri , j=12∑a |πr , ai−πr , aj| . The factor of 12 is used so that the maximum distance will always fall between zero and one . For the analysis in Figuer 8 , the results are presented in terms of the mutational effects rather than the amino acid preferences . If πr , a is the preference of site r for amino acid a and πr , a′ is the preference for amino acid a′ ( both re-scaled by the stringency parameters in Table 1 ) , then the estimated effect of the mutation from a to a′ is simply log ( πr , a′πr , a ) . All code and input data required to reproduce all analyses in this paper are in Supplementary file 1 ( see also Supplementary file 2 ) . A repository containing all of this code is also available at https://github . com/jbloomlab/EnvMutationalShiftsPaper ( Haddox et al . , 2018; copy archived at https://github . com/elifesciences-publications/EnvMutationalShiftsPaper ) . The deep sequencing data are on the Sequence Read Archive with the accession numbers listed in Supplementary file 1 and 2 . | The virus that causes AIDS , or HIV , has a protein called Env on its surface , which is essential for the virus to infect cells . Env can also be recognized by the immune system , which then targets the virus for destruction or blocks it from infecting cells . Unfortunately , Env evolves very quickly , which means that HIV can evade our defenses . However , there are limits to how much this protein can change , since it still needs to perform its essential role in helping viruses enter cells . In the century since HIV first appeared in human populations , the virus has evolved considerably . There are now many HIV strains that infect people , and they bear Env proteins with substantially different sequences . However , it is not clear if these changes in sequence have resulted in Envs from distinct strains being able to tolerate different mutations . To examine this question , Haddox et al . compared how the Envs from two strains of HIV react to modifications in their sequences . They created all possible individual mutations in the proteins , and the resulting collections of mutated viruses were then tested for their ability to infect cells in the laboratory . Most mutations had similar effects in both Env proteins . This allowed Haddox et al . to identify portions of the protein that easily accommodate changes , and portions that must remain unchanged for viruses to remain infectious—at least in the laboratory . Some of these mutations are under different types of pressures when the virus faces the immune system , and those were identified using computational approaches . However , some mutations were tolerated differently by the two Env proteins . Therefore , viral strains differ in how their Env proteins can evolve . The parts of Env that showed differences in mutational tolerance between the strains were not necessarily the parts that differ in sequence . This shows that changes in sequence in one part of the protein can modify how other portions evolve . It remains to be determined whether changes in tolerance to mutations translate into differences in how the virus can escape immunity . This is an important question given that the rapid evolution of Env is a major obstacle to creating a vaccine for HIV . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2018 | Mapping mutational effects along the evolutionary landscape of HIV envelope |
Many postnatal onset neurological disorders such as autism spectrum disorders ( ASDs ) and intellectual disability are thought to arise largely from disruption of excitatory/inhibitory homeostasis . Although mouse models of Rett syndrome ( RTT ) , a postnatal neurological disorder caused by loss-of-function mutations in MECP2 , display impaired excitatory neurotransmission , the RTT phenotype can be largely reproduced in mice simply by removing MeCP2 from inhibitory GABAergic neurons . To determine what role excitatory signaling impairment might play in RTT pathogenesis , we generated conditional mouse models with Mecp2 either removed from or expressed solely in glutamatergic neurons . MeCP2 deficiency in glutamatergic neurons leads to early lethality , obesity , tremor , altered anxiety-like behaviors , and impaired acoustic startle response , which is distinct from the phenotype of mice lacking MeCP2 only in inhibitory neurons . These findings reveal a role for excitatory signaling impairment in specific neurobehavioral abnormalities shared by RTT and other postnatal neurological disorders .
A number of postnatal neurological disorders such as autism spectrum disorders ( ASDs ) and certain types of intellectual disability are thought to involve a disturbance in excitatory/inhibitory homeostasis ( Nelson and Valakh , 2015; Rubenstein and Merzenich , 2003 ) . For example , mouse models of Rett syndrome ( RTT ) , a severe postnatal onset neurological disorder , show diminished excitatory drive and overall reduced cortical microcircuit activity ( Dani et al . , 2005; Wood et al . , 2009 ) . Conversely , the Met conditional deletion mouse ( a model of nonsyndromic , sporadic autism ) shows increased excitatory synaptic connectivity from cortical layer 2/3 to layer 5B corticostriatal neurons ( Qiu et al . , 2011 ) . Certainly the prevalence of epilepsy as a comorbidity with ASDs and intellectual disability disorders would suggest a tendency toward hyperexcitation , but both excitatory and inhibitory neuronal populations are affected in these disorders . It is difficult to discern the contributions of different neuronal groups to pathogenesis , however , in part because genes driving disorders such as RTT , Fragile X syndrome , Angelman syndrome , and tuberous sclerosis ( MECP2 , FMRP , UBE3A , and TSC1 and TSC2 , respectively ) participate in fundamental cellular processes in multiple cell types ( Crino , 2013; Darnell et al . , 2011; Lyst and Bird , 2015; Mabb et al . , 2011 ) , and in part because network alterations always induce compensatory changes in the circuit ( Turrigiano , 2011 ) . One way to circumvent this challenge is to examine the effects of such genes on specific neuronal types . This is the approach we and others have taken to understand the contribution of various neuronal types to the complex , wide-ranging phenotype of RTT , which is caused by loss-of-function mutations in the ubiquitously expressed MECP2 ( methyl CpG-binding protein 2 ) ( Amir et al . , 1999; Trappe et al . , 2001 ) . Removing Mecp2 from SIM-1 expressing hypothalamic neurons , for example , results in hyperphagia , obesity and aggression ( Fyffe et al . , 2008 ) , whereas deficiency of MeCP2 in somatostatin-positive neurons causes seizures and stereotypies ( Ito-Ishida et al . , 2015 ) . The most surprising result , however , given early studies suggesting RTT involves impairment in excitatory signaling ( Chao et al . , 2007; Chapleau et al . , 2009; Dani et al . , 2005; Marchetto et al . , 2010 ) , came from conditional knockout of Mecp2 in GABAergic neurons . These GABAergic conditional knockout mice reproduce most of the features observed in the Mecp2-null mice: premature lethality , stereotyped forepaw motions and other repetitive behaviors , abnormal social interaction , learning and memory deficits , hypoactivity , ataxia , and hindlimb clasping ( Chao et al . , 2010 ) . The only features of the Mecp2-null mice that did not develop in the GABAergic knockouts were tremor and anxiety-like behaviors . What role , then , do the excitatory signaling impairments observed in Rett patients and previous mouse models play in RTT pathogenesis ? To answer this question , we characterized mice with Mecp2 deleted solely in excitatory glutamatergic neurons . We further examined null mice with Mecp2 re-expressed only in the glutamatergic neurons to determine which behaviors healthy excitatory neurons were sufficient to rescue . We found that loss of MeCP2 in glutamatergic neurons results in neurological deficits that are quite distinct from those revealed upon loss of MeCP2 in GABAergic neurons .
To target excitatory neurons specifically , we utilized an Slc17a6-Cre ( also known as vesicular glutamate transporter 2-Cre , referred to as Vglut2-Cre henceforth ) mouse line to express Cre recombinase in the majority of excitatory neurons ( Vong et al . , 2011 ) . Vglut2 is the dominant glutamate vesicular transporter at embryonic and early postnatal stages: by P14 it is mostly replaced by Vglut1 in the hippocampus , cortex , and cerebellum ( Boulland et al . , 2004; Fremeau et al . , 2004; Miyazaki et al . , 2003 ) , which ensures that any Cre activity driven by Vglut2 will have a broad effect on excitatory neurons . Indeed , characterization of Vglut2-Cre expression using Rosa-tdTomato reporter mice revealed that the majority of CamKII-positive glutamatergic neurons in the cortex and hippocampal CA1 region expressed Cre ( 88% and 98% , respectively; Figure 1A and B ) . 99% of Cre-expressing cells were CamKII-positive ( Figure 1A and B ) , indicating that Cre is specifically expressed in excitatory neurons . To conditionally delete Mecp2 in glutamatergic neurons we mated female mice carrying a conditional Mecp2 allele flanked by loxP sites ( Guy et al . , 2001 ) with male Vglut2-Cre+/- mice and generated Mecp2 conditional knockout mice ( Mecp2flox+/y;Vglut2-Cre+/-: CKO ) along with their littermate controls ( Mecp2flox+/y: Flox , wild type: WT , Vglut2-Cre+/-: Cre ) . Immunofluorescence staining of MeCP2 in the CKO mice demonstrated a clear loss of the protein in areas primarily composed of excitatory neurons , such as the cerebral cortex , thalamus , and hippocampal CA1 through CA3 regions ( Figure 1C ) . 10 . 7554/eLife . 14199 . 003Figure 1 . Mecp2 was either deleted or restored specifically in glutamatergic neurons . ( A ) Vglut2-Cre expression was assayed by colocalization of the reporter tdTomato and CamKII in 6-week-old Vglut2-Cre+/-;Rosa26tdTomato male mice . Scale bar , 30 µm . ( B ) Quantification of images in ( A ) showing the percentage of tdTomato positive cells in total CamKII-immunostaining cells ( black ) and the percentage of CamKII-immunostaining cells in total tdTomato positive cells ( red , n = 3 mice , 14 sections ) . ( C , D ) Representative images showing MeCP2 expression in the brain of Flox and CKO mice ( C ) , as well as stop-null and male C-rescue mice ( D ) . Scale bar , 2 mm ( n = 3 mice per genotype ) . ( E ) Fluorescence images of male C-rescue cortex stained for nucleus ( 4′ , 6-diamidino-2-phenylindole , DAPI ) , MeCP2 and CamKII . Scale bar , 30 µm . ( F ) Quantification of images in ( E ) showing the percentage of MeCP2 positive cells in total CamKII-immunostaining cells ( black ) , and the percentage of CamKII-immunostaining cells in total MeCP2 positive cells ( red , n = 3 mice , 18 sections ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 003 To restore MeCP2 function in glutamatergic neurons , we bred Vglut2-Cre+/- male mice to females carrying a Mecp2 conditional rescue allele with a floxed STOP cassette ( Mecp2LSL/+ ) ( Guy et al . , 2007 ) . Male F1 Mecp2LSL/y mice ( referred to as 'stop-null' ) express a truncated , non-functioning version of MeCP2 and phenotypically resemble the constitutive null mutants ( Guy et al . , 2007 ) . Male Mecp2LSL/y;Vglut2-Cre+/- mice ( referred to as 'conditional rescue' or 'C-rescue' ) re-express MeCP2 only in glutamatergic neurons , where Cre excises the STOP cassette . Consistent with published data , there was no MeCP2 expression in stop-null mice ( Guy et al . , 2007 ) , but its expression was restored in regions rich in excitatory neurons in the male C-rescue mice ( Figure 1D ) . Specifically , 95% of CamKII-positive glutamatergic neurons in the cortex of male C-rescue mice expressed MeCP2 while all MeCP2-positive neurons were labeled by CamKII ( Figure 1E and F ) , indicating successful restoration of MeCP2 specifically in glutamatergic neurons . To determine the functional importance of MeCP2 in glutamatergic neurons , we studied the effect of loss of MeCP2 on cortical layer V pyramidal neurons in the somatosensory cortex . Whole-cell patch-clamp recordings from these neurons in 6- to 8-week old mice revealed reduced spontaneous action potential firing in CKO mice ( Figure 2A ) similar to that observed in the Mecp2-null mutants ( Dani et al . , 2005 ) . In the presence of synaptic transmission blockers , neurons in CKO mice showed similar firing rates in response to injected currents as WT neurons , indicating normal intrinsic excitabilities ( Figure 2—figure supplement 1C ) . To explore the cause of reduced spontaneous activity , we focused on the total excitatory and inhibitory drive onto layer V pyramidal neurons . In the presence of ongoing network activity within the slices , we recorded spontaneous excitatory postsynaptic currents ( sEPSCs ) by voltage-clamping the cell membrane at the measured chloride-reversal potential ( WT: −70 . 8 ± 0 . 4 mV; Cre: −69 . 1 ± 0 . 3 mV; Flox: −70 . 7 ± 0 . 5 mV; CKO: −69 . 9 ± 0 . 6 mV ) , while spontaneous inhibitory postsynaptic currents ( sIPSCs ) were recorded by holding the cell at the measured cation-reversal potential ( WT: 14 . 3 ± 0 . 4 mV; Cre: 12 . 9 ± 0 . 4 mV; Flox: 14 . 3 ± 0 . 8 mV; CKO: 12 . 7 ± 0 . 2 mV ) . Interestingly , both excitatory and inhibitory synaptic charges were reduced in CKO mouse neurons compared to controls ( Figure 2—figure supplement 1A and B ) , indicating that the observed reduction in cortical activity is most likely caused by reduced excitatory input to layer V neurons . To determine if the change in sEPSCs is caused by an alteration in quantal responses ( miniature EPSCs , mEPSCs ) or spike-driven EPSCs , we recorded α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor ( AMPAR ) -dependent mEPSCs . The frequency of mEPSCs recorded from CKO mice was significantly decreased when compared to WT mice and showed a trend of reduction although not significant when compared with Cre or Flox mice ( Figure 2—figure supplement 2A ) . The amplitude of mEPSCs and both the amplitude and frequency of miniature IPSCs ( mIPSCs ) did not differ significantly across genotypes ( Figure 2—figure supplement 2A and B ) . These data suggest that the weakened spike-driven excitatory input to layer V pyramidal neurons largely accounts for the reduced firing rate in CKO mice . 10 . 7554/eLife . 14199 . 004Figure 2 . Reduced cortical activity was induced by conditional deletion of Mecp2 in glutamatergic neurons but rescued by conditional restoration of MeCP2 . ( A ) Top: Sample traces of spontaneous firing of Layer V neurons from CKO and control mice . Bottom: Average firing rate of four genotypes of mice . ( B ) Top: Sample traces of spontaneous firing of Layer V neurons from male C-rescue mice and its counterparts . Bottom: Quantification of spontaneous firing rate . n = 3–4 mice and 12–15 cells per genotype . Data are presented as mean ± SEM . *p<0 . 05; **p<0 . 01; ***p<0 . 001; ****p<0 . 0001; by one-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 00410 . 7554/eLife . 14199 . 005Figure 2—figure supplement 1 . Layer V pyramidal neurons in the CKO mice received less excitatory and inhibitory input . ( A ) Top: Sample traces of sEPSCs from Layer V pyramidal neurons . Bottom: Summary of sEPSCs strength calculated as total charge of responses for 120 s . ( B ) Top: Sample traces of sIPSCs from Layer V pyramidal neurons . Bottom: Summary of sIPSCs strength calculated as total charge of responses for 120 s . ( C ) In the presence of excitatory and inhibitory synaptic transmission blockers , firing rate was plotted as a function of current amplitude to determine intrinsic properties . Scale: 100 mV , 200 ms . n = 3–4 mice , >10 cells per genotype . Data are presented as mean ± SEM . *p<0 . 05; **p<0 . 01; ****p<0 . 0001; by one-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 00510 . 7554/eLife . 14199 . 006Figure 2—figure supplement 2 . CKO mice displayed reduced mEPSC frequency and normal mIPSCs . ( A ) Top: Sample traces of mEPSCs . Bottom: Quantification of amplitude and frequency of mEPSCs . ( B ) Top: Sample traces of mIPSCs . Bottom: Average of amplitude and frequency of mIPSCs . n = 4–5 mice , >15 cells per genotype . Data are presented as mean ± SEM . *p<0 . 05; n . s . , not significant; by one-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 00610 . 7554/eLife . 14199 . 007Figure 2—figure supplement 3 . Restoration of MeCP2 in glutamatergic neurons normalized reduced mEPSC frequency in stop-null mice . ( A ) Top: Sample traces of mEPSCs . Bottom: Quantification of amplitude and frequency of mEPSCs . ( B ) Top: Sample traces of mIPSCs . Bottom: Average of amplitude and frequency of mIPSCs . n = 4–5 mice , >15 cells per genotype . Data are presented as mean ± SEM . *p<0 . 05; n . s . , not significant; by one-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 007 To determine if restoration of MeCP2 function in glutamatergic neurons can improve the reduced cortical activity in Mecp2-null mice , we measured spontaneous action potential firing in the male C-rescue , stop-null , and control mice . As expected , layer V pyramidal neurons in the somatosensory cortex of the stop-null mice showed a significantly reduced firing rate ( Figure 2B ) , whereas restoring MeCP2 in excitatory neurons rescued the firing rate to a level similar to that observed in controls . The stop-null mice also showed reduced mEPSC frequency ( Figure 2—figure supplement 3A ) as reported in Mecp2-null mice ( Chao et al . , 2007 ) , while this defect was reversed in C-rescue mice . Frequency and amplitude of mIPSCs in stop-null mice and C-rescue mice were similar to controls ( Figure 2—figure supplement 3B ) . Thus , restoration of MeCP2 only in excitatory neurons is sufficient to maintain layer V pyramidal neuron activity . To understand what features of the RTT phenotype might derive from excitatory signaling impairment , we characterized the CKO and male C-rescue mice . CKO mice died by 10 weeks of age ( Figure 3A ) , similar to Mecp2-null mice ( Guy et al . , 2001 ) . Interestingly , restoration of MeCP2 only in glutamatergic neurons significantly lengthened lifespan , as half of the male C-rescue mice survived for more than 46 weeks , in contrast to the ~12-week median lifespan of stop-null mice ( Figure 3B ) . 10 . 7554/eLife . 14199 . 008Figure 3 . Removal of Mecp2 from glutamatergic neurons led to early death and obesity , which were improved in C-rescue mice . ( A–B ) Survival curves plotted with percentage of mice alive as a function of age ( A , n = 25; B , n = 28–34 ) . ( C–D ) Plot of weight as a function of age ( C , n = 20; D , n = 15–22 ) . '**' indicates a statistical significant difference between CKO or stop-null and controls , while '##' indicates a statistical significant difference between C-rescue and controls . Data are presented as mean ± SEM . ** , ## , p<0 . 01; by two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 00810 . 7554/eLife . 14199 . 009Figure 3—figure supplement 1 . CKO mice gained more weight associated with increased daily food intake . ( A ) Average daily food intake from experimental day 3 to 5 ( 72 hr , P27-30 ) . Values are estimated marginal means adjusted for the difference in initial body weight . ( B–C ) Average daily weight gain ( B ) and fat gain ( C ) from day 1 to 6 while mice were housed in the calorimetry chambers . Values are estimated marginal means adjusted for the difference in initial body weight . ( D ) Average daily energy expenditure during experimental day 3 to 5 from indirect calorimetry . ( E ) Resting metabolic rate calculated based on energy expenditure during fasting . Values for ( D–E ) are estimated marginal means adjusted for the difference in average fat and lean mass . ( F–G ) Average daily activity ( F ) and daily RER ( G ) during experimental day 3 to 5 . ( H ) Average RER during fasting . Data are from 5–7 mice per genotype and are presented as mean ± SEM . **p<0 . 01; ****p<0 . 0001; n . s . , not significant; by one-way ANOVA or ANCOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 00910 . 7554/eLife . 14199 . 010Figure 3—figure supplement 2 . Male C-rescue mice gained less weight associated with reduced daily food intake . ( A ) Average daily food intake from experimental day 3 to 5 ( 72 hr , P27-30 ) . Values are estimated marginal means adjusted for the difference in initial body weight . ( B–C ) Average daily weight gain ( B ) and fat gain ( C ) from day 1 to 6 , while mice were housed in the calorimetry chambers . Values are estimated marginal means adjusted for the difference in initial body weight . ( D ) Average daily energy expenditure from day 3 to 5 in the colorimetry chambers . ( E ) Quantification of resting metabolic rate based on energy expenditure during fasting . Values for ( D–E ) are estimated marginal means adjusted for the difference in average fat and lean mass . ( F–G ) Quantification of average daily activity ( F ) and RER ( G ) during experimental day 3 to 5 . ( H ) Average RER during fasting . Data are from 6–7 mice per genotype and are presented as mean ± SEM . *p<0 . 05; **p<0 . 01; ***p<0 . 001; n . s . , not significant; by one-way ANOVA or ANCOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 010 The CKO mice , being an F1 hybrid of FVB and 129SvEvTac , gained significantly more weight than controls as early as 6 weeks of age and became severely obese with age ( Figure 3C ) , phenocopying the Mecp2-null mice on a 129SvEvTac background ( Heckman et al . , 2014 ) . The male C-rescue mice gained some weight with age , but less than littermates , and they remained underweight ( Figure 3D ) . To identify the cause of weight abnormalities in both CKO and C-rescue mice , we housed mice individually in calorimetry chambers for 6 days from P25-30 to simultaneously monitor food intake , activity , and components of energy balance . After a two-day acclimation period , we assessed average food intake per day over the subsequent 3 days , accounting for initial differences in individual weights . The CKO mice consumed about 25–40% more food than controls ( Figure 3—figure supplement 1A ) , and this hyperphagia was associated with significantly greater weight and fat gain ( Figure 3—figure supplement 1B and C ) . Evaluation of total energy expenditure , resting metabolic rate , and total activity of the mice revealed no significant difference between CKO and controls ( Figure 3—figure supplement 1D–F ) . The CKO mice had a higher respiratory exchange ratio ( RER ) when they had free access to food ( Figure 3—figure supplement 1G ) , but normal RER during fasting ( Figure 3—figure supplement 1H ) , suggesting greater lipogenesis but no impediment to fat utilization . In contrast , P25-30 male C-rescue mice consumed less food than stop-null and control mice ( Figure 3—figure supplement 2A ) , which correlated with less gain of fat and weight ( Figure 3—figure supplement 2B and C ) . They displayed no significant difference in total daily energy expenditure , resting metabolic rate , RER , or total activity compared with controls ( Figure 3—figure supplement 2D–H ) . The obesity of CKO mice and the underweight phenotype of male C-rescue mice therefore appear to be caused by abnormal energy intake rather than altered energy expenditure . Epilepsy occurs in 67%–90% of individuals with RTT ( Cardoza et al . , 2011; Cooper et al . , 1998; Steffenburg et al . , 2001 ) . Seizures and abnormal electroencephalography ( EEG ) have also been reported in Mecp2 mutant mice ( D'Cruz et al . , 2010; Shahbazian et al . , 2002 ) . We found focal seizure-like spike-and-wave discharges ( Figure 4A ) in three out of eight 10-week-old CKO mice that underwent video-EEG recordings . These discharges occurred at 3 . 3 ± 2 . 1 episodes/hour ( n = 3 mice ) and the average duration of episodes was 4 . 1 ± 0 . 4 s ( n = 27 ) . We did not find seizure-like discharges in control mice ( 9 Flox , 6 Cre , and 5 WT mice ) . Interestingly , we observed behavioral seizures on male C-rescue mice older than 25-week-old during routine husbandary . We then conducted EEG recording on four 25- to 30-week-old male C-rescue mice , as well as two WT and two Cre mice that were at similar age . One out of those four male C-rescue mice , but no WT or Cre mice , showed spike-and-wave discharges ( Figure 4B ) . They occurred at 3 . 3 episodes/hour and the average duration of episodes was 4 . 5 ± 0 . 8 s ( n = 10 ) . Thus , deleting MeCP2 only from glutamatergic neurons induced seizure-like abnormal EEGs , while restoring MeCP2 only in this group of neurons is not sufficient to rescue seizures . 10 . 7554/eLife . 14199 . 011Figure 4 . Both CKO and male C-rescue mice developed abnormal EEGs . ( A ) Representative EEG traces from the somatosensory and frontal cortices of 10-week-old WT , Cre , Flox , and CKO mice . Note the spike-and-wave discharge in the CKO mouse . ( B ) Representative EEG traces from the somatosensory and frontal cortices of 25- to 30-week-old WT , Cre , and male C-rescue mice . Note the spike-and-wave discharge in the male C-rescue mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 011 Depletion of Mecp2 in inhibitory GABAergic neurons reproduced most of the phenotype of the Mecp2-null mice except for tremor and anxiety-like behaviors ( Chao et al . , 2010 ) . We therefore asked whether these features depend on impairment of excitatory neurons . Our first set of behavioral tests examined anxiety-like behaviors . At 5 weeks of age , CKO mice showed similar locomotor function as controls when they traveled in the open field ( Figure 5—figure supplement 1A ) . While , in the elevated plus maze test , they stayed longer in the open arms , which can indicate decreased anxiety ( Figure 5A ) ; on the other hand , they spent less time in the light chamber during the light/dark assay and made fewer transitions between the two chambers ( Figure 5C ) , suggesting increased anxiety . These seemingly paradoxical results replicate the altered anxiety-like behaviors of Mecp2-null mice ( Heckman et al . , 2014; Pelka et al . , 2006; Stearns et al . , 2007 ) . Stop-null mice displayed similarly altered anxiety-related behaviors , with increased time spent in the open arms of the elevated plus maze ( Figure 5B ) and less time in , and fewer transitions into , the light chamber ( Figure 5D ) . Interestingly , restoration of MeCP2 solely in glutamatergic neurons was able to fully normalize this behavior , as the male C-rescue mice performed similarly as WT and Cre mice in those assays ( Figure 5B and D ) . This reversal of altered anxiety in the male C-rescue mice was maintained to 20 weeks of age , when most stop-null mice were dead and could not be included in the experiment ( Figure 5—figure supplement 2A and B ) . 10 . 7554/eLife . 14199 . 012Figure 5 . Loss of Mecp2 in glutamatergic neurons led to altered anxiety-like behaviors , tremor , and impaired acoustic startle response , which were normalized in male C-rescue mice . ( A–B ) Time spent in the open arms of the elevated plus maze ( A , n = 16–19; B , n = 9–14 ) . ( C–D ) Average time spent in , and number of transitions into , the light chamber during 10min test in the light/dark box ( C , n = 16–19; D , n = 13–16 ) . ( E–F ) Percentage of mice displayed tremor at 8 weeks of age ( E , n = 17–22; F , n = 18 ) . ( G–H ) Mean of response to the 120 dB acoustic stimulus ( G , n = 15; H , n = 14–18 ) . Data are presented as mean ± SEM . *p<0 . 05; **p<0 . 01; ***p<0 . 001; ****p<0 . 0001; by one-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 01210 . 7554/eLife . 14199 . 013Figure 5—figure supplement 1 . CKO mice have normal locomotor and hearing function . ( A ) Total distance mice traveled during the 30 min open field assay ( n = 16–18 ) . ( B ) Mean hearing thresholds ( dB SPL ) were plotted as a function of stimulus frequency ( n = 8–9 ) . Data are presented as mean ± SEM . n . s . , not significant; by one-way or two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 01310 . 7554/eLife . 14199 . 014Figure 5—figure supplement 2 . Reversal of altered anxiety-like behaviors and impaired acoustic startle response were maintained in 20-week-old mice . ( A ) Behavior of mice in the elevated plus maze , showing time spent in the open arms ( n = 12–21 ) . ( B ) Average time spent in , and number of transitions into , the light side during the 10 min light/dark test ( n = 12–21 ) . ( C ) Average response to a 120 dB stimulus ( n = 12–21 ) . Data are presented as mean ± SEM . *p<0 . 05; ***p<0 . 001; n . s . , not significant; by one-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 014 At 8 weeks of age 80% of CKO mice displayed obvious tremor , which was rarely observed in control mice ( Figure 5E ) . All examined stop-null mice showed tremor , while no male C-rescue mice developed a detectable tremor ( Figure 5F ) . Impaired acoustic startle response has been described in the Mecp2-null mice . CKO mice at 6 weeks of age also showed a diminished response to a 120 dB stimulus ( Figure 5G ) . This reduction in the CKO mice is likely caused by a disturbance in the sensorimotor gating circuit in the lower brainstem ( Koch and Schnitzler , 1997 ) , since auditory brainstem response ( ABR ) measurements showed no obvious differences in hearing of the CKO mice compared to controls ( Figure 5—figure supplement 1B ) . The male C-rescue mice , however , responded similarly to WT and Cre mice at 6 weeks of age , in contrast to the barely responsive stop-null mice ( Figure 5H ) . At 20 weeks of age , the male C-rescue mice displayed an elevated acoustic startle response compared with controls ( Figure 5—figure supplement 2C ) . Taken together , these data imply that deficiency of MeCP2 in glutamatergic neurons underlies altered anxiety-like behaviors , tremor , and impaired acoustic startle response . This hypothesis receives further support from the failure of restoring MeCP2 expression solely in GABA-expressing neurons to rescue most of these features on a null background ( see companion paper Ure et al . , 2016 ) . Despite the tight correspondence between the CKO and male C-rescue mice in anxiety , tremor , and acoustic startle response , not all behaviors in the two lines were complementary . For instance , the 6-week-old CKO mice , C-rescue , and stop-null mice all exhibited ataxia , as indicated by a shorter latency to fall from the accelerating rotarod ( Figure 6A and B ) . It is understandable that glutamatergic neurons would be necessary but not sufficient to ensure motor coordination , given that multiple neural populations and glia cells are necessary to keep the circuit intact . Strikingly , restoration of MeCP2 only in glutamatergic neurons affected behaviors that did not appear as deficits in the CKO mice but did develop in the stop-null mice , such as social interaction and repetitive behaviors . The CKO mice showed normal interest in familiar and non-familiar partners in the partition test of social behavior ( Figure 6C ) , based on the time they spent interacting with partners , and they displayed no repetitive behaviors on the hole-board assay ( Figure 6E ) . The stop-null mice , however , showed increased interest in partners and repetitive behavior , both of which were normalized to WT level in the male C-rescue mice ( Figure 6D and F ) . It is possible that the rescue of social behavior and stereotypies results from compensatory mechanisms elicited by re-expression of MeCP2 in excitatory neurons . 10 . 7554/eLife . 14199 . 015Figure 6 . Both CKO and male C-rescue mice suffered from ataxia but maintained normal social interaction and were free of repetitive behaviors . ( A–B ) Latency to fall from the accelerating rotarod plotted as a function of trial ( A , n = 15–17; B , n = 9–14 ) . ( C–D ) Average time that mice spent interacting with partners when they were housed with a familiar mouse ( Fam1 ) , a novel mouse ( Novel ) , and the same familiar mouse ( Fam2 ) again in the partition test ( C , n = 10–16; D , n = 13–15 ) . ( E–F ) Number of holes explored with ≥2 sequential nose-pokes during the 10 min hole-board assay ( E , n = 10–14; F , n = 11–13 ) . Data are presented as mean ± SEM . *p<0 . 05; **p<0 . 01; ***p<0 . 001; n . s . , not significant; by one-way or two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 015 Our behavioral characterization of male C-rescue mice suggests that there is a benefit from restoring glutamatergic function in RTT mouse models , especially on longevity , obesity , anxiety-like behaviors , tremor , repetitive behaviors , and social interaction . The male mice are not the best model to study rescue effects , however , because most RTT patients are females with mosaic expression of MeCP2 due to random X chromosome inactivation . Female Mecp2-heterozygous mice are thus a more clinically relevant model for studying the effects of restoring MeCP2 only in glutamatergic neurons . We characterized the female F1 generation from the Mecp2LSL/+ and Vglut2-Cre+/- mating . In these F1 females , Mecp2LSL/+ ( referred as 'stop-het' ) functionally corresponds to Mecp2+/- and Mecp2LSL/+;Vglut2-Cre+/- is the female C-rescue . Stop-het mice were significantly overweight by 9 weeks of age , while female C-rescue mice , like the male C-rescue , gained much less weight by 10 weeks of age than controls ( Figure 7A ) . Similar to Mecp2-null mice , the stop-het mice at 10 weeks of age displayed reduced anxiety in the elevated plus maze ( Figure 7B ) , and impaired acoustic startle response ( Figure 7C ) . Unlike the stop-null mice , the stop-het mice were responsive to acoustic stimulus , so we were able to further analyze the pre-pulse inhibition ( PPI ) . The stop-het mice showed increased inhibition at 78 and 82 dB pre-pulses ( Figure 7D ) . Female C-rescue mice performed similarly to controls in all these assays ( Figure 7B–D ) . 10 . 7554/eLife . 14199 . 016Figure 7 . Restoration of Mecp2 in glutamatergic neurons in stop-het mice rescued RTT-like features . ( A ) Plot of weight as a function of age ( n = 16 ) . '*' indicates a statistically significant difference between stop-het and all other genotypes , while '#' indicates a statistically significant difference between female C-rescue and controls . ( B ) Average time spent in the open arms of the elevated plus maze ( n = 16–18 ) . ( C ) Mean of response to the 120 dB stimulus ( n = 16–18 ) . ( D ) Pre-pulse inhibition at 74 dB , 78 dB and 82 dB pre-pulses ( n = 16–18 ) . ( E ) Latency to fall from the accelerating rotarod is plotted as a function of trial ( n = 15–16 ) . ( F ) Average number of footfalls normalized by distance traveled during the 10 min foot-slip assay ( n = 15–16 ) . ( G ) Percentage of mice displayed tremor at 30 weeks of age ( n = 20–28 ) . Data are presented as mean ± SEM . * , # p<0 . 05; **p<0 . 01; ***p<0 . 001; ****p<0 . 0001; by one-way or two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 01610 . 7554/eLife . 14199 . 017Figure 7—figure supplement 1 . Reversal of impaired acoustic startle and increased PPI were maintained at 30-week-old female C-rescue mice . ( A ) Total distance traveled in the open field assay ( n = 15–16 ) . ( B ) Mean time spent in the open arms of the elevated plus maze ( n = 15–16 ) . ( C ) Average response to the 120 dB stimulus ( n = 15–16 ) . ( D ) Percentage of inhibition in response to three pre-pulses ( n = 15–16 ) . Data are presented as mean ± SEM . *p<0 . 05; **p<0 . 01; ***p<0 . 001; ****p<0 . 0001; n . s . , not significant; by one-way or two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 017 We then characterized 30-week-old animals to see if the improvements in certain features in female C-rescue mice persisted . All animals still displayed similar locomotor activity at 30 weeks of age ( Figure 7—figure supplement 1A ) . Anxiety-like behaviors in stop-het mice at this age were not obvious ( Figure 7—figure supplement 1B ) . Correction of impaired acoustic startle response and increased PPI were maintained in female C-rescue mice at 30 weeks of age ( Figure 7—figure supplement 1C and D ) . Unexpectedly , in contrast to the male C-rescue mice , female C-rescue mice showed alleviation of ataxia in their significantly longer latency to fall compared to stop-het mice in the rotarod assay ( Figure 7E ) . This improvement was further confirmed by fewer footfalls in the footslip assay ( Figure 7F ) , which assesses motor coordination independent of body weight . In addition , at 30 weeks of age only one out of 28 ( 3 . 6% ) female C-rescue mice displayed tremor , in contrast to the ~50% stop-het mice that showed tremor ( Figure 7G ) . Taken together , our data indicate that restoration of MeCP2 function in glutamatergic neurons in a female heterozygous background rescues many RTT-like features .
As RTT affects virtually every part of the brain and various MeCP2 models have been so thoroughly characterized , MeCP2 mice are an ideal model to interrogate the consequences of partial impairment of a specific neuronal population . This is particularly necessary to postnatal neurological disorders such as ASDs and intellectual disability in which the neurons are present and only partially impaired . Here we have delineated the contributions of MeCP2 loss in glutamatergic neurons to several neurological deficits shared by RTT and ASDs . Generally excitatory neurons are thought to carry information flow while inhibitory neurons are required for tuning the circuits . Given the entangled nature of excitatory and inhibitory signaling , it is surprising that deleting or restoring MeCP2 only in glutamatergic neurons leads to complementary phenotypes of tremor , anxiety , and acoustic startle response ( Table 1 ) . In the case of acoustic startle response , this is not entirely unexpected: the caudal pontine reticular nucleus ( PnC ) is the major sensorimotor interface , and proper glutamatergic signaling is necessary for the transmission of auditory information onto PnC neurons through action on AMPAR ( Koch and Schnitzler , 1997 ) . As for anxiety and tremor , our findings suggest that the circuits involved are more resistant to inhibitory but not excitatory disturbance , possibly due to their specific connectivity . Similarly , the cerebellar circuits may be more vulnerable to inhibitory alteration since inhibitory Purkinje cells are the sole output of the cerebellar cortex . It is therefore understandable that re-expression of MeCP2 in inhibitory neurons ( see companion paper Ure et al . , 2016 ) but not excitatory neurons was sufficient to rescue motor coordination deficits . It is also not surprising that several other Mecp2-null features , including ataxia , premature lethality , and obesity , can be reproduced in both excitatory and inhibitory Mecp2 CKO mice , since those circuits are disturbed with alteration in either excitation or inhibition . Thus , restoring MeCP2 in neither excitatory nor inhibitory neurons alone is sufficient to fully rescue premature lethality . Similarly , re-expressing MeCP2 in excitatory neurons failed to maintain body weight , as the C-rescue mice were underweight when compared with WT animals , indicating the necessity of MeCP2 in inhibitory neurons to balance the circuit . 10 . 7554/eLife . 14199 . 018Table 1 . Comparison of stop-null , Mecp2 CKO , and male and female C-rescue mice . DOI: http://dx . doi . org/10 . 7554/eLife . 14199 . 018Neurological featuresStop-nullMecp2 CKOMale C-rescueFemale C-rescueTremor++--Altered Anxiety++--Impaired Acoustic Startle Response++--Premature Lethality++Significantly delayed-*Obesity++--Ataxia+++-Repetitive Behavior+--NEAbnormal Social Interaction+--NE+: Presence of feature . -: Absence of feature*: Deficits were absent in stop-het mice . NE: Mice were not examined with this particular assay . Interestingly , the increased social interaction and repetitive behaviors observed in null mice were not induced in the CKO mice but were improved in the male C-rescue mice . It is worth noting that the GABAergic knockout also shows these defects ( Chao et al . , 2010 ) ; it may be that restoring glutamatergic neuron function in the stop-null mice improved GABAergic neuron function through reciprocal interactions . One intriguing finding of this study is that restoring MeCP2 in glutamatergic neurons benefits stop-het females more than stop-null male mice: ataxia was normalized in the female , but not in the male , C-rescue mice . This is reasonable considering that in addition to the MeCP2 expressed in glutamatergic neurons , female C-rescue mice also retain MeCP2 in over 50% of non-glutamatergic neurons ( Young and Zoghbi , 2004 ) , none of which express MeCP2 at all in male C-rescue mice . In this regard , it is interesting to note that re-expressing MeCP2 only in GABAergic neurons is less effective in a heterozygous background than in the Mecp2-null background ( see companion paper Ure et al . , 2016 ) . The opposite outcomes of these rescue strategies suggest that various types of neurons may be differentially sensitive to MeCP2 dosage , and it seems easier for the brain to compensate for the loss of MeCP2 in 50% of inhibitory neurons ( glutamatergic female C-rescue ) than in 50% of excitatory neurons ( inhibitory female C-rescue ) . The possible reasons for this difference might include the distinct roles that these two types of neurons play in the brain circuit and the differential changes in gene expression due to loss of MeCP2 in these two types of neurons ( Kodama et al . , 2012; Telfeian et al . , 2003 ) . Nevertheless , it strongly argues for the use of Mecp2-heterozygous mice as the primary RTT disease model when developing and testing therapeutic strategies . At the circuit level , we demonstrated that deleting MeCP2 only in glutamatergic neurons resulted in reduced cortical activity in layer V pyramidal neurons of the somatosensory cortex , a defect also seen in the Mecp2-null mice . Together with the complete rescue of this defect by restoring MeCP2 only in glutamatergic neurons , this indicates an essential role of MeCP2 in modulating the firing activity of excitatory neurons . The dramatically reduced sEPSC charge reflects weakened excitatory drive onto the layer V neurons of CKO mice . Given that neurons usually need to receive above-threshold excitation to generate action potential ( Hodgkin and Huxley , 1990 ) , the layer V neurons of CKO mice thus fired less often than controls even though they also received less inhibition . It is worth noting that this reduced firing activity in CKO mice may be region specific , as we also observed seizure-like spike-and-wave discharges in the cortex of these mice . Behavioral or electrographic seizures have also been reported in Mecp2 whole brain ( Chao et al . , 2010 ) and forebrain ( Goffin et al . , 2014 ) inhibitory CKO mice , as well as in somatostatin CKO mice ( Ito-Ishida et al . , 2015 ) , although one group reported absence seizure ( Zhang et al . , 2014 ) but another group found no behavioral or electrographic seizures at all ( Goffin et al . , 2014 ) with the forebrain excitatory neuron KO mice . These data suggest that the seizures we observe in Mecp2-null mice are caused by multi-level disruption of circuits that can arise from dysfunction of several different neuronal cell types . This is further supported by the C-rescue mice , as restoring MeCP2 only in excitatory neurons is insufficient to prevent seizure . Previous work has shown that postnatal loss of MeCP2 in forebrain excitatory neurons leads to motor coordination deficits , altered anxiety , and impaired social interaction ( Gemelli et al . , 2006 ) . Through comparing and contrasting the excitatory conditional deletion and conditional rescue models , we confirmed the importance of MeCP2 in glutamatergic neurons to prevent anxiety-like behaviors . Importantly , we also highlighted its contribution to the pathogenesis of tremor , which is a common and highly penetrant symptom in individuals with RTT ( Klauck et al . , 2002; Roze et al . , 2007 ) , and has not been reported in any other cell type-specific Mecp2 conditional deletion mouse . Expanding on the observations from Gemelli et al . and on previous studies on the autonomous dysfunction of RTT ( Julu et al . , 2001; Ward et al . , 2011; Weese-Mayer et al . , 2006 ) , our work suggests that the autonomic dysfunction and premature death of Mecp2-null mice may be mainly driven by impaired excitatory signaling in the brainstem and spinal cord . Our results precisely complement previous knowledge and work from Ure et al . , which shows that deficiency of MeCP2 in GABAergic neurons leads to ataxia , increased social interaction , repetitive behaviors , premature lethality , and obesity , but is less involved in tremor , altered anxiety , and impaired acoustic startle response . This work also provides an example of how primary deficits in excitatory signaling lead to neurological deficits , which has ramifications for other postnatal neurological disorders , with disturbances in excitatory and inhibitory homeostasis , such as ASDs and intellectual disability .
The Vglut2-Cre+/- knock-in line was a gift from Dr . Brad Lowell and was backcrossed to the FVB strain for four generations to generate progeny with more than 99% FVB strain polymorphic markers . Mecp2 CKO mice were obtained by breeding Mecp2flox/+ ( Guy et al . , 2001 ) female mice on the 129S6SvEv strain to Vglut2-Cre+/- male mice . Mecp2 conditional rescue mice were obtained by breeding Mecp2LSL/+ ( Guy et al . , 2007 ) female mice on the 129S6SvEv strain to Vglut2-Cre+/- male mice . CKO mice were compared to wild type , Mecp2flox , and Vglut2-Cre littermate controls . C-rescue mice were compared to wild type , Vglut2-Cre , and Mecp2LSL littermate controls . Mice were housed in an AAALAS-certified animal facility on a 14 hr/10 hr light/dark cycle . All procedures to maintain and use these mice were approved by the Institutional Animal Care and Use committee for Baylor College of Medicine . All the behavioral assays were carried out blinded to the genotype . Mice were habituated in the testing room for at least 30 min before the test . After habituation in the testing room ( 200-lux , 60 dB white noise ) mice were individually placed in the center of an open Plexiglas chamber ( 40 × 40 ×30 cm ) with photo beams ( Accuscan ) to measure their activity for 30 min . Data are shown as mean ± standard error of mean and analyzed by one-way ANOVA with Tukey’s post hoc analysis . The light/dark box was a chamber with a lit side ( 36 × 20 × 26 cm ) and a dark side ( 15 . 5 × 20 × 26 cm ) with a 10 . 5 × 5 cm opening ( OmniTech Electronics ) . After habituation to the testing room ( 200-lux , 60 dB white noise ) , the mouse was placed in the lit side . The amount of time animals spent in each side and the number of transitions between the two sides were recorded by photo beams ( Fusion ) for a 10-minute period . Data are shown as mean ± standard error of mean and analyzed by one-way ANOVA with Tukey’s post hoc analysis . After habituation to the testing room ( 200-lux , 60 dB white noise ) , the mouse was placed in the center of a four-arm maze ( each arm 25 × 7 . 5 cm ) , with two opposing arms enclosed by 15 cm high walls and the other two open . The maze was 50 cm above the ground level . The amount of time animals spent in , and their entries to , each arm were recorded for 10 min with a camera and ANY-maze ( Stoelting Co . ) video tracking software . Data are shown as mean ± standard error of mean and analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were habituated in a room next to the testing room for 30 min . Test mice were brought into the testing room and were placed in a Plexiglas tube inside of a sound-insulated lighted box ( SR-Lab , San Diego Instruments ) . The procedure was carried out as described previously ( Chao et al . , 2010 ) . Startle stimulus is 120 dB and three pre-pulses used are 74 , 78 , and 82 dB . Pre-pulse inhibition was calculated as 1-[averaged startle response to startle stimulus with pre-pulse/averaged response to startle stimulus] x 100 . Data are shown as mean ± standard error of mean . ASR data are analyzed by one-way ANOVA with Tukey’s post hoc analysis . PPI data are analyzed by two-way ANOVA with Tukey’s post hoc analysis . Mice were put on an accelerating rotarod ( Ugo Basile ) whose speed increased from 4 to 40 rpm for the first 5 min and was then maintained at 40 rpm for another 5 min ( or was stopped when all the mice fell ) . Each animal was tested in 4 trials per day for 2 consecutive days , with a 30-minute interval between two trials in the same day . Latency to fall was recorded when the mouse fell from the rod or when the mouse had ridden the rotating rod for two revolutions without regaining control . Data are shown as mean ± standard error of mean and analyzed by two-way ANOVA with Tukey’s post hoc analysis . The apparatus was a floor of parallel-positioned rods within a Plexiglas chamber ( 21 × 21 cm ) . Mice were individually placed in the center of the grid . The number of foot slips and the distance traveled were recorded for 10 min with a camera and ANY-maze ( Stoelting Co . ) video tracking software . The number of footslips was normalized to the total distance traveled in the chamber . Data are shown as mean ± standard error of mean and analyzed by one-way ANOVA with Tukey’s post hoc analysis . This assay was performed as previously described ( Chao et al . , 2010 ) with a few modifications . Experimental mice were singly housed for 24 hr in standard cages with transparent perforated Plexiglas barrier that separates the cage into two compartments . An age- and gender-matched C57BL/6J partner was placed in the opposite side of the partition cage on the second day . Social interaction scoring was carried out on the third day allowing at least 18 hr co-housing of the experimental and partner mice . The amount of time the experimental mouse exhibited direct interest into the partner in a 5-minute period was recorded using a handheld computer ( Psion ) , and analyzed using the Observer program ( Noldus ) . Three different interaction paradigms were assessed sequentially: experimental mouse versus familiar partner ( Familiar 1 ) , experimental mouse versus novel partner ( Novel ) , and experimental mouse versus familiar partner again ( Familiar2 ) . Data are shown as mean ± standard error of mean and analyzed by two-way ANOVA with Tukey’s post hoc analysis . This assay was performed as previously described ( Chao et al . , 2010 ) . Briefly , mice were placed on a Plexiglas frame ( 40 × 40 cm ) with 16 equally spaced holes . Mice were allowed to explore freely for 10 min while the experimenter recorded holes that had been nose-poked by the mouse in order . Increased tendency to continuously poke the same hole no less than twice is used as an indicator of repetitive behaviors . Data are shown as mean ± standard error of mean and analyzed by one-way ANOVA with Tukey’s post hoc analysis . Mice were put on the flat palm of the hand for 30 s for a person who is blind to the genotype to decide if the mouse displayed tremor . The methods were modified from our previous publication ( Hao et al . , 2015 ) . Adult mice were anesthetized with isoflurane . Under aseptic condition , each mouse was surgically implanted with 4 recordings electrodes . Two silver wire electrodes ( 127 µm diameter ) were implanted in the right frontal cortex and somatosensory cortex , respectively , with the reference electrode positioned in the occipital region of the skull . Two tungsten electrodes ( 50 µm diameter ) were aimed at the left hippocampal CA1 ( P2 . 0R1 . 2H1 . 3 ) and dentate ( P2 . 0R1 . 8H1 . 8 ) regions , respectively , with the reference electrode in the corpus callosum . All electrode wires were attached to a miniature connector ( Harwin Connector ) secured on the skull by dental cement . After 2 weeks of post-surgical recovery , EEG activities ( filtered between 0 . 1 Hz and 1 kHz , sampled at 2 kHz ) and the mouse behavior were recorded for 2 hr per day over 3–4 days . Seizure candidates were identified using custom-written algorithms . Briefly , EEG signals were divided into 10-minute segments . A third order Butterworth bandpass filter of 0 . 5 and 400 Hz cutoffs was applied to each segment . The filtered data was divided into 500 ms non-overlapping epochs . Signal changes occurring in the time domain were captured by amplitude correlation ( autocorrelation value between successive epochs ) , root mean square ( average amplitude of the epoch ) , and spike density ( number of spikes normalized to the epoch ) . Signal changes occurring in the frequency domain were captured by frequency band ratio , where the power of the upper band ( 20–50 Hz ) was contrasted with that of the lower band ( 0 . 5–20 Hz ) . EEG signals that exceeded the threshold for all of the above quantitative features were identified as seizure candidates . Each candidate and the corresponding video were visually inspected to identify electrographic seizures . All computation was performed in MATLAB R2015b . This assay was performed as previously described with a few modifications ( Ma et al . , 2011 ) . Mice were group housed before the experiment started . Then they were placed individually in calorimetry chambers for 6 days , with the first 2 days considered as an acclimation phase . Mice had free access to food and water in the first 5 days . Resting metabolic rate ( RMR ) was measured on day 6 in the Comprehensive Laboratory Animal Monitoring System ( CLAMS , Columbus Instruments ) . Feeders were closed at 6 a . m . on day 6 to prevent any further food consumption . The RMR for each mouse was calculated from the two lowest values of energy expenditure from 4 to 8 hr after the start of the fast . Fat and lean mass were measured by quantitative magnetic resonance spectroscopy ( EchoMRI , Houston , TX ) . For food intake , weight and fat gain , energy expenditure , and resting metabolic rate data were analyzed by ANCOVA with Tukey’s post hoc analysis using body weight ( for food intake ) or fat and lean mass ( for energy expenditure ) as covariance ( Tschop et al . , 2012 ) . Values are shown as estimated marginal means ( least square means ) adjusting for covariance ± standard error of mean . For total activity and RER , data are shown as mean ± standard error of mean and analyzed by one-way ANOVA with Tukey’s post hoc analysis . Auditory brainstem responses ( ABRs ) were measured as previously described ( Cai et al . , 2013 ) . Briefly , 6-week-old mice were anesthetized with an intraperitoneal injection of ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) . Normal body temperature was maintained throughout the procedure by placing the mice on a heating pad . Pure tone stimuli from 4 kHz to 48 kHz were generated using Tucker Davis Technologies System 3 digital signal processing hardware and software ( Tucker Davis Technologies , Alachua , FL , USA ) , and the intensity of the tone stimuli was calibrated using a type 4938 1/4″ pressure-field calibration microphone ( Brüel and Kjær , Nærum , Denmark ) . Response signals were recorded with subcutaneous needle electrodes inserted at the vertex of the scalp , the postauricular region ( reference ) and the back leg ( ground ) . Auditory thresholds were determined by decreasing the sound intensity of each stimulus from 90 dB to 10 dB in 5 dB steps until the lowest sound intensity with reproducible and recognizable waves in the response was reached . Data are shown as mean ± standard error of mean and analyzed by two-way ANOVA . Coronal slices ( 350 μm thick ) containing primary somatosensory cortex ( S1 ) were prepared from 6- to 8-week-old mice with a vibratome slicer ( Leica Microsystems Inc . , Buffalo Grove , IL ) . Acute slices were collected in chilled ( 2–5°C ) cutting solution containing ( in mM ) 110 choline-chloride , 25 NaHCO3 , 25 D-glucose , 11 . 6 sodium ascorbate , 7 MgSO4 , 3 . 1 sodium pyruvate , 2 . 5 KCl , 1 . 25 NaH2PO4 , and 0 . 5 CaCl2 . Then , slices were incubated in standard artificial cerebrospinal fluid ( ACSF , in mM ) containing 119 NaCl , 26 . 2 NaHCO3 , 11 D-glucose , 3 KCl , 2 CaCl2 , 1 MgSO4 , and 1 . 25 NaH2PO4 at 37°C for 50 min before being stored at room temperature for recording . All solutions were saturated with 95% O2 and 5% CO2 . Whole-cell recordings were made from visually identified pyramidal neurons in layer 5 region of S1 by using a patch clamp amplifier ( MultiClamp 700 B , Molecular Devices , Union City , CA ) . Microelectrodes with resistance of 2–3 MΩ were pulled from borosilicate glass capillaries ( Sutter Instruments , Novato , CA ) . The intrapipette solution for recording spontaneous firing ( holding at -60 mV ) and mEPSC ( holding at -70 mV ) contained ( in mM ) 140 potassium gluconate , 5 KCl , 10 HEPES , 0 . 2 EGTA , 2 MgCl2 , 4 MgATP , 0 . 3 Na2GTP , and 10 Na2-phosphocreatine ( pH 7 . 2 ) . Picrotoxin ( 100 µM ) , D-2-amino-5-phosphonopentanoic acid ( AP5 , 50 µM ) , and Tetrodotoxin ( TTX , 0 . 1 µM ) were present in the ACSF for mEPSC recording . For measuring mIPSCs , the pipettes were filled with ( in mM ) 145 KCl , 10 HEPES , 2 MgCl2 , 4 MgATP , 0 . 3 Na2GTP , and 10 Na2-phosphocreatine ( pH 7 . 2 ) . AP5 ( 50 µM ) , 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX , 20 µM ) , and TTX ( 0 . 1 µM ) were also present in the ACSF . The recordings of spontaneous firing and spontaneous synaptic activity were performed in modified ACSF ( in mM: 126 NaCl , 25 NaHCO3 , 14 D-glucose , 3 . 5 KCl , 1 CaCl2 , 0 . 5 MgCl2 , 1 NaH2PO4 ) . For measuring sEPSC/sIPSC , the intrapipette solution contained ( in mM ) 120 CsCH3SO3 , 20 HEPES , 0 . 4 EGTA , 5 TEA-Cl ( tetraethylammonium chloride ) , 2 MgCl2 , 2 . 5 MgATP , 0 . 3 GTP , 10 Na2-phosphocreatine , and 1 QX-314 [N- ( 2 , 6-dimethylphenylcarbamoylmethyl ) triethylammonium bromide] ( pH 7 . 2 with CsOH ) . To find the reversal potential , the postsynaptic neurons were voltage-clamped at different potentials ( with 5 mV interval ) between-75 and -55 mV for sEPSC and between -5 and +20 mVs for sIPSC . To test the intrinsic property , synaptic blockers ( APV 50 µM , CNQX 20 µM , and bicuculine 20 µM ) were added into the standard ACSF . The input ( current pulse ) /output ( spikes ) curves were generated using incremental depolarizing current pulses ( 10pA/800 ms ) . The whole-cell recording was performed at ( 30 ± 1°C ) by using an automatic temperature controller ( Warner Instrument , Hamden , CT ) . Data acquisition was performed by using a digitizer ( DigiData 1440A , Molecular Devices ) . Minianalysis 6 . 0 . 3 ( Synaptosoft Inc ) and pClamp 10 ( Molecular Devices ) were used for data analysis . Data were discarded when the rest membrane potential was above −60 mV or the resistance change was over 20% during experiment . Data are shown as mean ± standard error of mean and analyzed by one-way ANOVA with Tukey’s post hoc analysis . Experiments was performed as previously described ( Heckman et al . , 2014 ) . Free-floating 45 μm sagittal sections were cut on a cryostat ( Leica CM3050 S ) , and stained overnight with primary antibody: rabbit anti-MeCP2 ( Cell Signaling 3456S ) ; mouse anti-MeCP2 ( Sigma M6818 ) , rabbit anti-CamKII ( Abcam ab52476 ) , and 2 hr with secondary antibody: goat anti-rabbit Alexa488 ( Thermo Fisher Scientific A11034 ) ; goat anti-mouse Alexa555 ( Thermo Fisher Scientific A-21137 ) . DAPI ( Thermo Fisher Scientific D1306 ) was used to stain nuclei . Sections were mounted with ProLong Gold Antifade Mountant ( Thermo Fisher Scientific P36934 ) . Images were acquired on Zeiss 710 laser-scanning confocal microscope and Leica SP8 microscope . Sample sizes for all analyses were determined by previous experience ( Chao et al . , 2010; Samaco et al . , 2013 ) . Data were analyzed using commercially available statistical software ( Prism6 , and SPSS 20 . 2 ) . | Rett syndrome is a childhood brain disorder that mainly affects girls and causes symptoms including anxiety , tremors , uncoordinated movements and breathing difficulties . Rett syndrome is caused by mutations in a gene called MECP2 , which is found on the X chromosome . Males with MECP2 mutations are rare but have more severe symptoms and die young . Many researchers who study Rett syndrome use mice as a model of the disorder . In particular , male mice with the mouse equivalent of the human MECP2 gene switched off in every cell in the body ( also known as Mecp2-null mice ) show many of the features of Rett syndrome and die at a young age . The MECP2 gene is important for healthy brain activity . The brain contains two major types of neurons: excitatory neurons , which encourage other neurons to be active; and inhibitory neurons , which stop or dampen the activity of other neurons . In 2010 , researchers reported that mice lacking Mecp2 in only their inhibitory neurons develop most of the same problems as those mice with no Mecp2 at all . Now , Meng et al . – who include two researchers involved in the 2010 study – have asked how deleting or activating Mecp2 only in excitatory neurons of mice affects Rett-syndrome-like symptoms . The experiments showed that male mice without Mecp2 in their excitatory neurons develop tremors , anxiety-like behaviors , abnormal seizure-like brain activity and severe obesity; these mice also die earlier than normal mice . Female mice lacking Mecp2 in half of their excitatory neurons ( because the gene is on the X chromosome ) were less affected than the males , and had normal life spans . These symptoms are different from those seen in mice missing Mecp2 only in inhibitory neurons . Meng et al . also found that if Mecp2 was switched on only in excitatory neurons of female mice ( which are a model of human Rett syndrome patients ) the mice were almost completely normal . In male mice ( which show more severe symptoms ) , activating Mecp2 in only the excitatory neurons reduced the anxiety and tremors . These findings suggest that impaired excitatory neurons may be responsible for specific symptoms such as anxiety and tremors amongst other Rett-syndrome-like features . The next challenge is to explore how the loss of Mecp2 changes the activity of excitatory neurons in different brain regions . Further studies could also investigate if drugs that improve the activity of excitatory neurons can be used to treat Rett syndrome patients . Finally , in a related study , Ure et al . asked if activating Mecp2 in inhibitory neurons in otherwise Mecp2-null mice was enough to prevent some of their Rett syndrome-like symptoms . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Manipulations of MeCP2 in glutamatergic neurons highlight their contributions to Rett and other neurological disorders |
Heritable variation in gene expression forms a crucial bridge between genomic variation and the biology of many traits . However , most expression quantitative trait loci ( eQTLs ) remain unidentified . We mapped eQTLs by transcriptome sequencing in 1012 yeast segregants . The resulting eQTLs accounted for over 70% of the heritability of mRNA levels , allowing comprehensive dissection of regulatory variation . Most genes had multiple eQTLs . Most expression variation arose from trans-acting eQTLs distant from their target genes . Nearly all trans-eQTLs clustered at 102 hotspot locations , some of which influenced the expression of thousands of genes . Fine-mapped hotspot regions were enriched for transcription factor genes . While most genes had a local eQTL , most of these had no detectable effects on the expression of other genes in trans . Hundreds of non-additive genetic interactions accounted for small fractions of expression variation . These results reveal the complexity of genetic influences on transcriptome variation in unprecedented depth and detail .
Differences in gene expression among individuals arise in part from DNA sequence differences in regulatory elements and in regulatory genes . Regions of the genome that contain regulatory variants can be identified by tests of genetic linkage or association between mRNA levels and DNA polymorphisms in large collections of individuals . Regions for which such tests show statistical significance are known as eQTLs ( Albert and Kruglyak , 2015 ) . Regulatory variation is widespread in the species for which it has been studied; indeed , in humans , the expression of nearly every gene appears to be influenced by one or more eQTL ( Aguet et al . , 2017; Battle et al . , 2014 ) . In humans , eQTLs are typically mapped by genome-wide association studies ( GWAS ) in unrelated individuals . To cover the genome , human GWAS must test a very large number of variants , resulting in a high multiple-testing burden and low statistical power . As a result , most human eQTL GWAS have been limited to searches for ‘local’ eQTLs that are located close to the genes they influence ( GTEx Consortium , 2015; Lappalainen et al . , 2013 ) . The power to detect local eQTLs is higher because focused local tests reduce the multiple-testing burden , and because local eQTLs tend to have larger effect sizes . However , genome-wide estimates show that most regulatory variation does not arise from local eQTLs . Instead , it arises from ‘distant’ eQTLs , which are located far from the genes they influence , typically on different chromosomes , and which exert their effects through trans-acting factors ( Grundberg et al . , 2012; Wright et al . , 2014 ) . Although trans-acting human eQTLs have been discovered ( Aguet et al . , 2017; Battle et al . , 2014; Brynedal et al . , 2017; Fehrmann et al . , 2011; Grundberg et al . , 2012; Heinig et al . , 2010; Lee et al . , 2014; Small et al . , 2011; Wright et al . , 2014; Yao et al . , 2017 ) , the vast majority remains unknown . As a consequence , we know relatively little about this crucial source of regulatory genetic variation . In model organisms , eQTLs can be identified by linkage analysis in panels of offspring obtained from crosses of genetically different individuals ( Brem et al . , 2002 ) . Whereas GWAS studies are powered to test only genetic variants found at high frequency in the population ( e . g . [Kita et al . , 2017] ) , linkage studies can assay both common and rare variants that differ between the parental strains . In addition , longer blocks of linkage reduce the number of statistical tests required to cover the genome . As a result , many local and distant eQTLs have been discovered in such studies . However , even in linkage studies , sample size limitations have to date resulted in insufficient statistical power to detect most eQTLs . This limitation has manifested itself as ‘missing heritability’: detected eQTLs tend to account for only a fraction of the measured heritable component of gene expression variation . Here , we addressed this limitation by carrying out an eQTL study in a large panel of segregants from a cross between two yeast strains . The high power of our study allowed us to identify eQTLs that account for the great majority of heritable expression variation in this cross , and to characterize the distant component of regulatory variation in unprecedented depth and detail .
We developed an experimental pipeline for high-throughput generation of RNA-seq data in yeast and obtained high-quality expression measurements ( Source data 1 and Source data 2 ) for 5720 genes in 1012 segregants from a cross between a laboratory and a wine strain ( hereafter , BY and RM , respectively ) . We obtained high-confidence genotypes at 11 , 530 variant sites from low-coverage whole-genome sequences of the segregants ( Bloom et al . , 2013 ) ( Source data 3 ) . We used the genotype and RNA-seq data for eQTL mapping and identified 36 , 498 eQTLs for 5643 genes at a false discovery rate ( FDR ) of 5% ( Source data 4 ) . Only 77 genes had no detected eQTL . Among the genes with at least one detected eQTL , the median number was 6 , with a maximum of 21 ( Figure 1A; Supplementary Discussion 1 describes the five genes with 21 eQTLs ) . Previous eQTL mapping in 112 segregants from this cross detected an average of less than one eQTL per gene as a consequence of much lower statistical power ( Brem et al . , 2002; Smith and Kruglyak , 2008 ) . That data set was used to obtain indirect estimates of the distribution of the number of eQTLs per gene ( Brem and Kruglyak , 2005 ) , and these agree closely with the distribution of directly detected eQTLs observed in the current study . For example , Brem and Kruglyak ( Brem and Kruglyak , 2005 ) estimated that at most 3% of genes would be influenced by a single eQTL ( we observed 2 . 6% of such genes ) , and half of genes would require >5 eQTLs ( we observed a median of 6 eQTLs per gene ) . While Brem and Kruglyak estimated that one third of genes would require more than 8 eQTLs , we observed only 23% such genes . Additional eQTLs of very small effect missed by our study likely account for this discrepancy . The observed distribution of the number of loci also closely matched the distribution we reported for loci influencing 160 protein levels studied with the highly powered X-pQTL approach ( Albert et al . , 2014b ) . Our results provide direct demonstration that variation in expression levels of nearly all genes has a complex genetic basis . We used our data to estimate the additive heritability of the expression level of each gene ( i . e . the fraction of expression variance attributable to genetic factors; Figure 1—figure supplement 1; Source data 5 ) . We observed a median heritability of 26% , with a maximum of 95% ( Figure 1B ) . Our estimates are similar to those from population-based studies of gene expression in humans ( Grundberg et al . , 2012; Lloyd-Jones et al . , 2017; Wheeler et al . , 2016; Wright et al . , 2014 ) . The estimates are lower than heritabilities typically seen for organismal traits in this yeast cross ( Bloom et al . , 2013 , 2015 ) , suggesting a greater contribution of environmental and stochastic factors to gene expression variation . Across genes , heritability was positively correlated with mean expression and with expression variance , and negatively correlated with the number of protein-protein and synthetic genetic interaction partners , as well as with gene essentiality ( p≤0 . 005 ) ( Figure 1—figure supplement 2; Supplementary Discussion 2 and 3; Supplementary file 1; Source data 6 ) . In contrast to previous eQTL studies , the detected eQTLs explained most of the estimated additive gene expression heritability ( a median across genes of 71 . 5% ) ( Figure 1B and D 10-fold cross-validation ) . Low missing heritability in our data is explained by the high power of our experiment . We had greater than 90% power to detect eQTLs that explain at least 2 . 5% of expression variance ( Figure 1C ) . The distribution of effect sizes of detected eQTLs is strongly weighted toward small effects ( median 1 . 9% of variance explained; Figure 1C ) , suggesting that the remaining missing heritability is explained by undetected eQTLs with even smaller effects . These results are similar to those observed for organismal traits in this cross ( Bloom et al . , 2013 , 2015 ) . Thus , we have discovered most eQTLs with substantial effects that segregate in this cross , and these jointly account for the great majority of the observed genetic variation in the transcriptome . We found that 2884 genes ( 50% of 5720 expressed genes ) had a local eQTL ( defined as an eQTL whose confidence interval includes the gene it influences ) at genome-wide significance ( Figure 2A ) . This number rose to 4241 genes ( 74% of expressed genes ) when we performed eQTL analysis with only one nearby marker per gene in order to reduce the multiple testing burden ( FDR < 5% ) . Thus , the single pair of yeast isolates used here harbors sufficient local regulatory variation to alter the expression of more than half the genes in the genome . Comparisons with allele-specific expression data ( Albert et al . , 2014a ) support previous results ( Doss et al . , 2005; Ronald et al . , 2005 ) that most but not all local eQTLs act in cis ( Figure 2—figure supplement 1 , Figure 2—figure supplement 2 , Supplementary Discussion 4; Supplementary files 2 , 3 and 4; Source data 7 ) . The vast majority of the genome-wide significant eQTLs did not overlap the genes they influenced ( 92%; 33 , 529 of 36 , 498 ) ; indeed , 86% were located on a different chromosome . Nearly every expressed gene ( 98%; 5606 ) had at least one such distant , trans-acting eQTL ( Figure 2A ) . The individual effect sizes of the trans eQTLs were smaller than those of local eQTLs ( median variance explained 2 . 8-fold less , T-test p<2 . 2e-16; Figure 1C ) . However , for the 2846 genes that had both a local eQTL and at least one distant eQTL , the aggregate effect of the distant eQTLs per gene was larger than that of the local eQTL ( median 2 . 6-fold more variance explained; paired T-test p<2 . 2e-16; Figure 2B ) . Distant eQTLs accounted for the majority of eQTL variance for 85% of genes . Our results directly demonstrate the importance of trans acting variation . The trans eQTLs were not uniformly distributed across the genome ( Figure 3A ) . Instead , they clustered at 102 hotspot loci , each of which affected the expression of many genes ( Brem et al . , 2002; Ghazalpour et al . , 2008; Orozco et al . , 2012 ) ( Figure 3B ) . These hotspots contained over 90% of all trans eQTLs . The eQTLs that mapped outside of the hotspots also clustered more than expected by chance ( randomization p<0 . 001 ) , suggesting the existence of additional hotspots that affect the expression of too few genes to pass the stringent criteria used to define the set of 102 . Isolated trans-acting loci that affect the expression of one or a few genes appear to be uncommon . The 102 hotspots affected a median of 425 genes , ranging from 26 ( a newly discovered hotspot at position 166 , 390 bp on chromosome III ) to 4594 at the previously reported MKT1 hotspot ( Zhu et al . , 2008 ) ( 82% of 5629 genes with any signal at a hotspot; Figure 3B ) . Three additional hotspots each affected more than half of all genes . They include a previously described hotspot at the HAP1 gene ( Brem et al . , 2002 ) ( 3640 genes affected ) , as well as two newly detected hotspots on chromosome XIV . A hotspot at 372 , 376 bp affected 4172 genes and is likely caused by a variant that recently arose in the KRE33 gene in the RM parent used in our cross ( Jerison et al . , 2017 ) . A hotspot at position 267 , 161 bp affected 3169 genes and spans the genes GCR2 , YNL198C , WHI3 and SLZ1 . These results indicate that hotspots can have extraordinarily wide-reaching effects on the transcriptome , with some influencing the expression of the majority of all genes . Widespread effects caused by single loci likely arise from a cascade of effects in which strong primary effects spread through the cellular regulatory network . For example , the BY allele of the transcriptional activator HAP1 carries a transposon insertion that reduces HAP1 function . As expected , the BY allele strongly reduced the expression of known transcriptional targets of HAP1: 26 out of the 69 HAP1 targets present in our data were among the 50 genes with the largest reduction in expression in segregants carrying the BY allele of HAP1 ( p<2 . 2e-16 , odds ratio = 138 ) . In total , only 75 direct transcriptional HAP1 targets are known . Unless previous work missed thousands of HAP1 targets , the vast majority of the 3640 trans eQTLs at HAP1 must reflect indirect , secondary consequences of the direct transcriptional effects . HAP1 is an activator of genes involved in cellular respiration . Thus , the many secondary effects of the BY HAP1 allele on gene expression may be mediated by cellular responses to altered metabolism arising from reduced respiration . Our Supplementary Files and Datasets provide detailed information about each hotspot . Source data 8 contains a table that gives an overview of the hotspots , including their location , genes affected ( details in Source data 9 ) , and analyses of function ( details in Source data 10 ) and transcriptional regulation ( details in Source data 11 ) of the target genes of each hotspot . Supplementary file 5 visually represents each hotspot region , and Supplementary file 6 displays gene networks formed by the strongest target genes of each hotspot . Functional analysis of eQTL hotspots requires identification of the underlying causal genes , which has been challenging to do systematically . We developed a multivariate fine-mapping algorithm that narrows hotspot positions by leveraging information across the genes that map to each hotspot ( Materials and methods ) . Briefly , we used all genes with an eQTL on a given chromosome and regressed out genetic factors on all other chromosomes and additional non-genetic factors . We reduced the dimensionality of the residual expression levels by singular value decomposition to capture linear combinations of traits that account for most of the variance in residual expression . We scanned the given chromosome for genetic influences on the multivariate distribution of these linear combinations , while controlling FDR via permutations . Finally , we used bootstraps to compute confidence intervals for hotspot locations ( Figure 4 and 5 ) . With this approach , we resolved the locations of 26 hotspots to regions containing three or fewer genes ( a total of 58 genes; Figure 4A ) . Three hotspots contained exactly one gene ( GIS1 , STB5 , and MOT3 ) . We previously identified and experimentally confirmed the causal genes at several major hotspots ( MKT1 ( Zhu et al . , 2008 ) , HAP1 ( Brem et al . , 2002 ) , IRA2 ( Smith and Kruglyak , 2008 ) , GPA1 ( Yvert et al . , 2003 ) , and the mating-type locus [Brem et al . , 2002] ) . These were all correctly localized by the algorithm , validating this fine-mapping strategy . The 58 genes at 26 high-resolution hotspots are highly enriched for the causal genes underlying the hotspots , making it possible to systematically study the functions of hotspot regulators . These genes were less likely to be essential or to have a human homolog than other yeast genes but did not differ from other genes in their expression level or the number of physical or genetic interaction partners ( Supplementary file 7 ) . The hotspot genes had highly significant enrichments for gene ontology ( GO ) terms related to transcriptional regulation ( e . g . GO:0006357 ‘regulation of transcription from RNA polymerase II promoter’: 19 genes among the 58; 4 expected; p=4e-9; Figure 4—figure supplement 1; Source data 12 ) , as well as weaker enrichments for terms related to response to nutrient levels ( GO:0031669; ‘cellular response to nutrient levels’: 8 genes , 1 expected , p=6e-6 ) These analyses indicate that causal hotspot genes are disproportionately involved in transcriptional regulation , a signal that was not picked up in an earlier study with fewer , less-well-resolved hotspots ( Yvert et al . , 2003 ) . For example , we fine-mapped a new hotspot that affected 382 genes to a single gene , the transcription factor STB5 ( Figure 4B ) . STB5 is a transcriptional activator of multidrug resistance genes ( Kasten and Stillman , 1997 ) . A previous analysis suggested reduced activity of the STB5 BY allele compared to the RM allele ( Lee and Bussemaker , 2010 ) . Consistent with this observation , we found that the promoters of genes whose expression was lower in the presence of the STB5 BY allele were strongly enriched for STB5 binding sites ( Figure 4C ) . To further examine the role of sequence variation in transcription factor genes , we focused on transcription factors with variants predicted to be damaging to protein function such as premature stop codons or frameshifts . There are eight such genes in our data . Remarkably , six of these ( GAT1 , HMS1 , PUT3 , RFX1 , SRD1 , TBS1 ) were located in a hotspot , often very close to the estimated peak location ( Figure 4—figure supplement 2 ) . None of these expression hotspots have been reported previously , although variation at GAT1 has been reported to influence traits relevant for wine production ( Salinas et al . , 2012 ) , and a premature stop codon in the RM allele of RFX1 has been linked to reduced activity of this transcriptional repressor ( Lee and Bussemaker , 2010 ) . The remaining two transcription factors with predicted damaging mutations did not overlap a hotspot . Of these , a predicted frameshift in YRM1 is very close to the end of the coding region , while a predicted frameshift in STB4 appears to be an annotation artifact: it resides in a region that is annotated as coding but that does not in fact appear to be transcribed ( Figure 4—figure supplement 3 , [Albert et al . , 2014a] ) . Hotspot genes can also influence mRNA levels more indirectly – for instance , by shaping the cellular response to external stimuli such as nutrient availability . For example , we fine-mapped a hotspot , which influenced 645 genes , to an interval on chromosome VIII containing six genes ( Figure 4—figure supplement 4A ) . One of these is ERC1 , which encodes a transmembrane transporter . BY but not RM carries a frameshift in this gene , which removes the last two out of 12 predicted transmembrane helices of the protein ( Fehrmann et al . , 2013 ) . This variant is known to reduce cell-to-cell variability ( or ‘noise’ ) in the expression of a MET17 gene tagged with green fluorescent protein ( Fehrmann et al . , 2013 ) . We found that the BY allele at this hotspot reduced the expression of genes that are highly enriched for the GO category ‘methionine biosynthetic process’ ( GO:0009086 , p=2e-22; Source data 8 and Source data 10 ) . Thus , in addition to reducing MET17 expression noise , the ERC1 frameshift variant is linked to reduced mean expression levels of multiple genes in the methionine biosynthesis pathway ( the MET regulon; Figure 4—figure supplement 4B ) . While the precise compounds that are imported or exported by Erc1p are not known , the ERC1 BY allele reduces cellular levels of S-Adenosylmethionine ( SAM ) ( Breunig et al . , 2014 ) , a key component of methionine and cysteine amino acid metabolism ( Sadhu et al . , 2014 ) . The ERC1 BY allele may down-regulate the MET regulon via its effects on SAM , triggering further transcriptional changes in hundreds of genes . Most known causal variants underlying yeast eQTL hotspots are coding ( HAP1 ( Brem et al . , 2002 ) , MKT1 ( Zhu et al . , 2008 ) , GPA1 , AMN1 ( Yvert et al . , 2003 ) , SSY1 ( Brown et al . , 2008 ) ; [Fay , 2013] ) ; however , change in the expression of a trans-acting factor by a local eQTL is another plausible causal mechanism ( Sudarsanam and Cohen , 2014; Yao et al . , 2017 ) . We found that a higher proportion of hotspots contained genes with a local eQTL than expected by chance ( p=0 . 007; Figure 5A ) . The median effect size of the strongest local eQTL in these hotspots was larger than expected ( p=0 . 003 ) . These enrichments are consistent with some hotspots being caused by local eQTLs that alter the expression of a gene located at the hotspot position , which in turn leads to changes in the other transcript levels that map to the hotspot . On the other hand , the majority of local eQTLs ( 60% ) did not overlap any of the hotspots . Evidently , the expression changes caused by these local eQTLs did not in turn lead to detectable trans effects on many unlinked genes , within the limits of our statistical power . To quantify this observation further , we focused on ‘non-hotspot’ local eQTLs that did not overlap a hotspot and asked how they related to the 10% of non-hotspot trans eQTLs that did not overlap a hotspot . We created non-overlapping genomic bins , each centered on a non-hotspot local eQTL . We then counted how many non-hotspot trans eQTL peaks fell into these bins ( Figure 5B ) . The resulting distribution roughly matched the distribution expected if non-hotspot trans peaks occurred at random locations . To the extent that the distribution differed from random , we found an excess of bins with six or more trans eQTLs ( p<0 . 001 ) . There was also an excess of bins with zero trans peaks ( p<0 . 001 ) . This class comprised the great majority of the distribution ( Figure 5B ) . The genetic architecture that is most consistent with these observations is one in which some local eQTLs are accompanied by multiple trans eQTLs , while most local eQTLs have no detectable trans consequences on the expression of other genes . Even when the causal gene in a hotspot has a local eQTL , it does not automatically follow that this is the causal mechanism . For example , STB5 and ERC1 each had a local eQTL . However , STB5 did not show allele-specific expression , and while there was weak allele-specific expression for ERC1 , it was in the opposite direction of the local ERC1 eQTL ( Source data 7 ) . Therefore , these local eQTLs are unlikely to be caused by cis-acting variants . STB5 and ERC1 carry protein-altering variants between BY and RM , including the known causal ERC1 frameshift in BY . Altered protein activity due to these coding variants may be responsible for the many distant linkages to these hotspots and may also cause the observed local eQTLs in trans , as previously shown for AMN1 ( Ronald et al . , 2005 ) . The Stb5p transcription factor is predicted to target its own promoter ( MacIsaac et al . , 2006 ) , such that its altered activity could influence its own expression . For the transmembrane transporter encoded by ERC1 , the local eQTL might reflect a more indirect mechanism . For each of these hotspots , it seems plausible that a change in protein function , rather than change in gene expression , underlies the hotspot . The degree to which mRNA-based eQTLs also affect the protein levels of their target genes is a fundamental open question ( Battle et al . , 2015; Chick et al . , 2016; Foss et al . , 2007; Ghazalpour et al . , 2011; Picotti et al . , 2013 ) that has been difficult to resolve as a consequence of low statistical power in eQTL and protein QTL ( pQTL ) studies . Low power is expected to lead to poor overlap between eQTLs and pQTLs solely as a result of high false-negative rates . We compared our eQTLs to pQTLs that we had identified earlier for 160 proteins using a powerful bulk segregant approach ( Albert et al . , 2014b ) ( Source data 13 ) . Here , we present results comparing the distant QTLs in both datasets because – by design of our earlier study – the set of distant pQTLs is much larger than the set of local pQTLs . Results for local QTLs are broadly consistent with those for distant QTLs ( Supplementary Discussion 5 ) . Distant pQTLs clustered at hotspots , which broadly mirrored the mRNA hotspots identified here ( Figure 6A ) . However , differences in hotspot architecture exist . For example , a hotspot on chromosome II showed strong pQTL effects ( Albert et al . , 2014b ) but only weak effects on mRNA levels for the same genes , none of which rose to genome-wide significance . In order to avoid downward bias in the overlap between eQTLs and pQTLs caused by false negatives , we focused on strong QTLs in each dataset and asked if they overlapped a significant QTL in the other dataset ( Materials and methods; see Supplementary Note 5 for results based on all distant QTLs ) . Of the 236 strongest eQTLs , 47% ( 111 ) overlapped a pQTL for the same gene . Of the 218 strongest pQTLs , 50% ( 108 ) overlapped an eQTL for the same gene . As a more sensitive alternative to QTL overlap analyses , we computed estimates of agreement based on the π1 statistic ( Storey and Tibshirani , 2003 ) . Of the strong eQTLs , 92% were estimated to match a pQTL . Of strong pQTLs , 63% were estimated to match an eQTL . Thus , while nearly all eQTLs with strong effects on mRNA levels also affect protein levels for the same gene , a larger fraction of strong pQTLs appear to be specific to protein levels . Strong eQTLs without a pQTL clustered primarily at the HAP1 and MKT1 hotspots ( Supplementary file 8; Figure 6B ) . These two hotspots also showed the clearest examples of overlapping eQTLs and pQTLs with opposite direction of effect on the same genes ( Supplementary file 9; Figure 6C ) . Thus , while these hotspots influence both mRNA and protein levels of many genes , their effects on mRNA vs . protein levels of a given gene can be quite different . Strong pQTLs without an eQTL were more widely distributed across the genome ( Supplementary file 10; Figure 6D ) . The contribution of non-additive or ‘epistatic’ genetic interactions to trait variation is a topic of ongoing debate ( Hill et al . , 2008; Mackay , 2014; Mäki-Tanila and Hill , 2014 ) . In particular , demonstration of non-additive effects on human gene expression has been challenging ( Becker et al . , 2012; Brown et al . , 2014; Buil et al . , 2015; Fish et al . , 2016; Hemani et al . , 2014; Wood et al . , 2014 ) . Although clear examples of epistasis have been revealed for yeast gene expression ( Brem and Kruglyak , 2005; Storey et al . , 2005 ) , the limited power of earlier studies had necessitated targeted search strategies rather than a full genome-by-genome scan . We reasoned that the high power of our current dataset should permit a more unbiased view of the contribution of epistasis to mRNA expression variation . We carried out a genome-by-genome scan for non-additive interaction effects on the expression levels of all genes and detected 387 eQTL-eQTL interactions influencing 306 genes ( FDR = 10%; Source data 14 ) . To our knowledge , this is the first unequivocal identification of eQTL interactions from an unbiased genome-by-genome scan . Targeted scans with a reduced multiple testing burden identified larger numbers of interacting pairs of loci: a total of 784 from a scan for interactions between genome-wide significant additive eQTLs and the genome , and a total of 1464 interactions between significant additive eQTLs . We examined the 387 eQTL-eQTL interactions detected in the genome-wide scan in more detail . The locations of interacting eQTLs clustered at certain positions in the genome , generally overlapping the hotspots described above ( Figure 7A ) . In particular , many epistatic interactions involved the HAP1 hotspot ( 79 interactions ) , the KRE33 hotspot ( 66 interactions ) , as well as hotspots containing MKT1 , GAP1 , the mating type locus , and IRA2 . Many interactions connected these hotspots with each other ( Figure 7B ) . For example , 30 genes shared an eQTL interaction between HAP1 and KRE33 , 14 genes shared an eQTL interaction between HAP1 and MKT1 , 13 genes shared an eQTL interaction between KRE33 with IRA2 , and 14 genes shared eQTL interactions between GPA1 and the mating-type locus ( Brem et al . , 2005 ) . The fact that interacting eQTLs colocalize with additive hotspots suggests that epistatic interactions often involve eQTLs that also have additive effects . Indeed , of the 774 markers in the 387 epistatic pairs , 558 ( 72% ) were within 10 kb of a genome-wide significant additive eQTL influencing the same gene . An estimate based on the π1 statistic ( Storey and Tibshirani , 2003 ) showed that at least 84% of epistatic markers have additive effects . An example is SAG1 , which encodes the Alpha-agglutinin of cells with the alpha mating type . In our cross , the alpha mating type was carried by the RM parent . Consequently , expression of SAG1 was higher in segregants that carried the RM allele at the mating type locus ( Figure 7C ) . In addition to this additive effect , the mating type locus was also involved in an interaction with the GPA1 locus , replicating the finding from a targeted search that a BY-specific S469I variant in GPA1 modulates SAG1 expression in alpha but not a cells ( Brem et al . , 2005 ) . Of the 387 interactions , 158 ( 41% ) involved ‘cis by trans’ interactions between distant and local loci ( Figure 7—figure supplement 1A ) . For example , the HAP1 hotspot interacted with local eQTLs for 15 genes ( Figure 7—figure supplement 1A ) . Among these , the RM alleles of HAP1 and SCM4 both increased SCM4 expression , and segregants with the RM genotype at both loci had higher expression levels than expected from additive effects ( Figure 7—figure supplement 1B ) . The HAP1 BY allele is less active due to a transposon insertion ( Brem et al . , 2002 ) . The local SCM4 eQTL may arise from variants that disrupt a HAP1 binding site ( Ter Linde and Steensma , 2002 ) in BY , and this allelic difference may result in a stronger expression difference in the presence of the more active RM Hap1p . In agreement with this model , an epistatic interaction between inferred Hap1p activity and SCM4 expression has been reported ( Parts et al . , 2011 ) . Interestingly , SCM4 is the only annotated direct HAP1 target among the local eQTLs that interact with HAP1 ( Harbison et al . , 2004 ) . Further , the effect sizes of local eQTLs that interacted with HAP1 were not statistically different in segregants that carried the more active RM vs . the less active BY allele at HAP1 ( T-test , p=0 . 9 ) . In all the cases we examined ( KRE33 , MKT1 , GPA1 , and IRA2 ) , the local eQTLs that interacted with these hotspots did not show consistently larger or smaller effect sizes depending on the hotspot allele ( all p≥0 . 3 ) . This argues against a scenario in which most cis by trans interactions are due to variants in transcription factor binding sites whose effect increases with a more active trans regulator . Instead , most cis by trans interaction effects may be mediated through more indirect mechanisms . We quantified the fraction of variation in gene expression that is contributed by epistatic interactions . Pairwise interactions typically explained about 1/10th as much expression variance as did additive loci ( Figure 7—figure supplement 2 ) , with a median of 2 . 2% of expression variance per pair . Thus , genetic interactions contributed only a small minority of trait variance for gene expression levels , which is consistent with what we previously reported for organism-level traits ( Bloom et al . , 2015 ) . We asked if any of the epistatic pairs identified from our unbiased search corresponded to ‘purely’ epistatic interactions without any additive effects . Of the 387 pairs , 111 ( 29% ) were not found in the search between additive loci . Given the small effects of most interactions , this group is likely enriched for false positives from the full search as well as false negatives from the search between additive loci . We searched for specific epistatic pairs in which neither locus had an additive effect ( p>0 . 05 ) , overlapped an additive hotspot ( to avoid any potential undetected additive hotspot effects ) , and where the affected gene was on a different chromosome from both interacting markers ( to avoid any potential undetected additive local eQTLs ) . Only three such pairs existed in our data ( Figure 7—figure supplement 3 ) . In each case , the two genotype classes with non-parental genotype combinations ( BY-RM and RM-BY ) had similar expression levels that differed from those in the two parental genotype combinations ( BY-BY and RM-RM ) , such that the marginal genotype effects canceled out . Even in these three cases , the variance resulting from the interaction was only 2 . 6–3 . 6% . We conclude that any contribution of ‘purely’ epistatic pairs to transcriptome variation must be small .
The high power of our study allowed us to identify genome-wide significant eQTLs that jointly explain over 70% of gene expression heritability . Thus , our eQTL map shows a high degree of completeness , capturing most genetic sources of transcriptome variation in this cross . In particular , our results allowed us to examine the contribution of trans-acting regulatory variation in much greater detail than previously possible in any species . We showed that trans-acting eQTLs are the predominant source of expression variation in this cross . Specifically , trans-eQTLs contribute 2 . 6-fold more to gene expression variance than local eQTLs , remarkably similar to genome-wide estimates in humans of 1 . 8-fold ( Grundberg et al . , 2012 ) and 3 . 4-fold ( Wright et al . , 2014 ) . While based on different experimental designs and study populations , these results from yeast and humans clearly demonstrate the importance of trans-acting variation in eukaryotic species . The vast majority of trans-eQTLs are concentrated at a limited number of hotspot regions that harbor variants with widespread effects on the expression of other genes . The strongest of these hotspots affected the expression of most genes in the genome . The minority of trans eQTLs that fell outside the statistically defined hotspots also clustered more than expected by chance . We saw little evidence for isolated trans-acting loci that affect the expression of one or a few genes . The high density of sequence differences in this cross might have been expected to produce a more diffuse trans-regulatory architecture , with many loci each influencing one or a few genes . Instead , trans-acting regulatory variation arises primarily from several dozen specific loci . The large number and fine resolution of our hotspots allowed us to conduct unbiased analyses of the genes located in these regions . Unlike earlier work examining fewer , less well localized hotspots ( Yvert et al . , 2003 ) , we found a strong enrichment for transcription factors , a class of genes with obvious relevance to gene regulation . Guided by this enrichment , we noticed that almost all transcription factor genes with predicted damaging variants in this cross appear to produce a trans-acting hotspot , with the exception of two genes in which the damaging variants may affect gene function only weakly if at all . This result clearly illustrates the importance of variation in transcription factors in shaping gene expression in trans . At the same time , many trans-acting hotspots are caused by variation in genes other than transcription factors . These hotspots illustrate the importance of indirect trans effects that alter cellular networks and physiology , which in turn results in gene expression change . In an unbiased search , we revealed hundreds of non-additive genetic interactions that influence mRNA levels . Many of these eQTL-eQTL interactions were between eQTLs at two different hotspot regions , or interactions between an eQTL at a hotspot and a local eQTL . Hotspots may play such prominent roles in epistatic interactions because their wide-reaching effects effectively alter the global state of the cell . The effects of many other variants may be larger in some cellular states than others , similar to what is seen when yeast are grown in different environments ( Lewis et al . , 2014; Smith and Kruglyak , 2008 ) . Previous searches for epistatic interactions on gene expression levels in this cross were targeted to genes and loci chosen based on prior additive scans ( Brem et al . , 2005; Storey et al . , 2005 ) . Our unbiased search confirmed the validity of this strategy . The vast majority of epistatic markers we detected also had additive effects , and there was little evidence for interactions without any additive effects . Conversely , many additive eQTLs , including several major additive hotspots , were involved in few if any epistatic interactions ( Figure 7A ) . Thus , epistatic interactions cannot be simply assumed to exist for all additive loci , and the most fruitful strategy to detect epistasis is to test loci with additive effects for interactions with other loci . In our cross between two yeast strains , we identified over one hundred hotspot regions , and found evidence for additional , weaker hotspots . Studies in many multicellular organisms , including plants ( Cubillos et al . , 2012; Fu et al . , 2009; West et al . , 2007; Zhang et al . , 2011 ) , nematodes ( Li et al . , 2006; Rockman et al . , 2010 ) , mice ( Hasin-Brumshtein et al . , 2016; Kelly et al . , 2012; Orozco et al . , 2012 ) and rats ( Heinig et al . , 2010; Hubner et al . , 2005; Kaisaki et al . , 2016 ) have also observed trans-eQTL hotspots , showing that this phenomenon is not restricted to yeast . Compared to our yeast cross , human populations harbor orders of magnitudes more variants , and the human genome has many more genes and more extensive and complex regulatory regions that offer a large target space for regulatory variation . We would expect these characteristics to result in human trans-eQTL architectures even more complex than that we have uncovered here in yeast . Recent studies have made progress in identifying trans-eQTLs in humans . Many of the human trans-eQTLs discovered to date tend to influence the expression of multiple genes ( Battle et al . , 2014; Brynedal et al . , 2017; Fehrmann et al . , 2011; Grundberg et al . , 2012; Heinig et al . , 2010; Lee et al . , 2014; Small et al . , 2011; Wright et al . , 2014; Yao et al . , 2017 ) . Specifically , work in blood and blood-derived cells from up to ~5000 individuals ( Battle et al . , 2014; Fairfax et al . , 2012; Fehrmann et al . , 2011; Pierce et al . , 2014; Westra et al . , 2013; Wright et al . , 2014; Yao et al . , 2017 ) revealed SNPs that were associated with multiple genes in trans , including the HLA locus ( Fehrmann et al . , 2011 ) , LZY in monocytes , and KLF4 in B-cells ( Fairfax et al . , 2012 ) . Multiple genes linking to single SNPs were also found in human adipose tissue ( e . g . in the KLF14 transcription factor gene ( Aguet et al . , 2017; Hore et al . , 2016; Small et al . , 2011 , 2018 ) , skin and lymphoblastoid cell lines ( Grundberg et al . , 2012 ) , and other tissues ( Aguet et al . , 2017 ) . In immune cells , condition-specific hotspots at known regulators of the immune response were seen upon stimulation , and these altered expression of specific immune response pathways ( Fairfax et al . , 2014; Lee et al . , 2014 ) . Shared , weak effects across transcripts ( Brynedal et al . , 2017; Hore et al . , 2016 ) indicated that ‘hundreds of trans-eQTLs each affect hundreds of transcripts’ ( Brynedal et al . , 2017 ) . Together , these observations show that trans-eQTL hotspots exist in humans . Whether trans-eQTL hotspots are as predominant in humans as they are in yeast ( where 90% of our trans-eQTLs mapped to a hotspot ) remains an open question . Human trans-eQTLs could be distributed more uniformly , such that hotspots make up a smaller fraction of all trans-eQTLs , leaving room for more gene-specific trans-eQTLs . There could also be more hotspots , each affecting fewer genes than the strongest hotspots in our cross . While it is difficult to extrapolate from the current , still underpowered human studies , it may be informative to consider that the initial eQTL searches in our yeast cross only found eight hotspots , each affecting at most dozens of genes ( Brem et al . , 2002 ) . The observation that hotspots are the main source of trans variation required the higher statistical power of the present dataset . It remains to be seen if human trans-eQTL discovery in larger samples will follow a similar trajectory . The recently proposed omnigenic model ( Boyle et al . , 2017 ) for the genetic basis of complex trait variation posits that gene regulatory networks are sufficiently densely connected that the change in expression of any one gene , caused by a local eQTL , will ‘percolate’ through the network and alter the expression of all other genes that are expressed in a given cell type . The hotspot loci we described here offer evidence that some regulatory variants can indeed have widespread effects on the transcriptome , in some cases altering the expression of the majority of genes in the genome through precisely the combination of strong direct effects on ‘core’ genes in specific pathways and weak indirect effects on other ‘peripheral’ genes envisioned in the omnigenic model . Hundreds of hotspots may exist in the human genome ( Brynedal et al . , 2017 ) , providing a rich substrate through which regulatory variation may influence complex traits . On the other hand , although we detected local eQTLs for most genes in our cross , the majority of these had no detectable trans effects on the expression of other genes , within the limits of our statistical power . Given that our study had sufficient power to detect weak indirect effects of trans-eQTL hotspots , we believe that most local eQTLs indeed have no meaningful downstream consequences for gene expression . By extension , such local eQTLs may be unlikely to contribute to variation in complex traits . Consistent with this conclusion , modest expression changes for dozens of yeast genes have been found to result in minimal fitness effects ( Duveau et al . , 2017; Keren et al . , 2016 ) . These results argue against the simplest form of the omnigenic model , in which a variant that changes the expression of any one gene has meaningful effects on every other gene . Instead , we observed that trans-eQTLs effects preferentially arise from variation in certain classes of genes . Given the crucial importance of regulatory variation for many complex traits ( Albert and Kruglyak , 2015 ) , the organismal consequences of expression changes caused by different types of eQTLs remain a key area for further research .
We used 1012 meiotic segregants previously generated ( Bloom et al . , 2013 ) from a cross between the prototrophic yeast laboratory strain BY ( MATa; derived from a cross between BY4716 and BY4700 ) and the prototrophic vineyard strain RM ( MATα hoΔ::hphMX4 flo8Δ::natMX4 AMN1-BY; derived from RM11-1a ) . The segregants were grouped according to their previously measured ( Bloom et al . , 2013 ) endpoint colony radius on YNB agar plates into groups of 96 . The strains in each group were rearranged from existing stock plates into a total of 13 96-well plates in YNB medium , grown to saturation , and frozen as glycerol stocks for later growth . Within each group of 96 , strain locations in the 96-well plate were selected at random . Culture and liquid handling was performed on a BioMek FXP instrument or with multichannel pipettes in 96-well format . Our strategy of batching segregants according to their growth on YNB ensures that each 96-well plate contains segregants that grow at comparable rates . This facilitates growing all segregants on a plate such that they reach a similar optical density at 600 nm ( OD ) at the same time . Our batching strategy produces experimental batches that are correlated with growth rates . Because we statistically removed variation among experimental batches prior to eQTL mapping ( see below ) , this design reduces our ability to compare variation in growth rates with variation in gene expression . We deemed this an acceptable trade-off because it considerably simplified handling >1000 samples in a systematic fashion . We processed the batches in a randomized order with respect to their growth rate to avoid confounding processing date with faster or slower growth . We used the rearranged stock plates to inoculate growth cultures in 1 ml YNB medium ( recipe for 1 L: 6 . 7 g yeast nitrogen base with ammonium sulfate and without amino acids; 900 ml H2O; autoclave; add 100 mL of separately autoclaved 20% glucose solution ) in 2 mL deep well plates sealed with Breathe-Easy membranes ( Sigma Aldrich ) , and grew the cultures to saturation on Eppendorf MixMate instruments situated in a 30°C incubator and set to 1100 rounds per minute ( rpm ) . We set the saturated cultures back to OD = 0 . 05 in 1 mL YNB in a fresh deep well plate and continued growth at 30°C . We monitored OD during growth by splitting out 100 µL of culture every other hour , measuring OD on a Synergy two plate reader ( BioTek ) and returning the 100 µL used for measuring OD to the deep-well culture plate . We increased the frequency of measurements as cultures approached OD = 0 . 4 . Once average OD in the plate reached 0 . 4 , we transferred the cultures to sterile Norgen nylon filter plates ( #40008 ) situated on a vacuum manifold . We applied vacuum to remove all growth medium , sealed with aluminum foil seals , and flash froze the entire plate in liquid N2 . The frozen plates were placed on a standard 96-well plate to protect their bottom , wrapped with parafilm , and stored at −80°C until RNA extraction . Note that this procedure provided us with OD measurements up to the exact time point at which cells were harvested . We used Dynabeads mRNA DIRECT kits ( Ambion/Thermo Fisher ) to directly isolate mRNA from cell lysates . To perform the RNA extractions on the BioMek robot , we prepared excess lysis/binding and Wash buffers that permitted the use liquid reservoirs with volumes that exceed that provided in the kits . These buffers were prepared as specified in the Dynabeads kit protocol: Lysis/Binding Buffer: 100 mM Tris-HCl , pH 7 . 5 500 mM LiCl 10 mM EDTA , pH 8 1% LiDS 5 mM dithiothreitol ( DTT ) Washing Buffer A: 10 mM Tris-HCl , pH 7 . 5 0 . 15 M LiCl 1 mM EDTA 0 . 1% LiDS Washing Buffer B: 10 mM Tris-HCl , pH 7 . 5 0 . 15 M LiCl 1 mM EDTA We filled the wells of an Axygen 1 . 1 mL plate ( P-DW-11-C-S ) with about 250 µl acid washed 425–600 µm beads ( Sigma G8722 ) . We added 700 µL lysis buffer to our frozen cell plates , pipetted up and down to resuspend the cells , and applied them to the glass beads in the Axygen plate . The Axygen plate was tightly sealed with an Axymat rubber plate seal ( AM-2ML-RD-S ) , and ground for 10 cycles on a plate-based mini bead beater ( Biospec ) . Each cycle consisted of 1 min beating followed by 1 min on ice . We centrifuged the plate for 4 min at 3000 rpm to separate glass beads and cell debris from the lysate . We pipetted two aliquots of 200 µL of lysate supernatant into two 96-well PCR plates for a total of 400 µL lysate . These plates were sealed , and the RNA melted for 2 min at 65°C in a thermocycler . We implemented a BioMek-assisted procedure to perform the Dynabead protocol with two mRNA enrichment steps . We did not quantify the resulting 11 µL of mRNA and simply used the entire mRNA for reverse transcription and sequencing library preparation . While piloting this procedure , we obtained typical yields of ~30 ng / µL and excellent RNA quality as judged by visualization on 1 . 1% agarose gels stained with ethidium bromide . Ribosomal RNA bands were clearly visible in crude lysate , less visible after the first mRNA enrichment , and absent after the second mRNA enrichment step . After the second mRNA enrichment , mRNA was clearly visible on the gel , with no visible RNA degradation . We performed reverse transcription and sequencing library preparation using the Kapa Stranded mRNA-Seq Kit ( KK8420/21 ) . This kit usually begins by enriching mRNA from total RNA . Because we had already performed mRNA enrichment , we used our entire mRNA as input and began at the RNA fragmentation step by adding 11 µL of ‘KAPA fragment , prime and elute buffer’ to our 11 µL of mRNA . RNA fragmentation was performed on a thermocycler for 6 min at 94°C . The remaining procedure was performed as specified in the Kapa kit manual . Briefly , the fragmented RNA is randomly primed and used for first strand cDNA synthesis , second strand synthesis and marking with dUTP , A-tailing of the double-stranded cDNA , adapter ligation , and PCR for 12 cycles . The dUTP marked second strand is not amplified in PCR , resulting in strand-specific libraries . We used custom designed Truseq-compatible indexing adapters ( IDT ) to allow multiplexing all 96 samples per batch . Prior to use , the two types of Truseq adapters were annealed ( 2 min at 97°C; 72 steps of 1 min at 1°C decreasing temperature; 5 min at 25°C ) to generate forked adapters that can be ligated to the A-tailed cDNA . We did not pool samples between batches . Sequencing libraries were quantified by combining 1 µL of library with 100 µL of Qubit High Sensitivity dsDNA reagent in 96-well plates with black bottom and wells , and reading fluorescence ( excitation 485 nm , emission 528 nm ) on the Synergy two plate reader . We calculated library concentrations by comparing to a standard series obtained by diluting the standard solutions included in the Qubit quantification kit . Standards were measured in triplicate on each library plate . We pooled the libraries in each group to equal molarity and used qPCR ( KAPA Biosystems #KK4854 ) on the pool to obtain the molarity for loading on the sequencer . Gel extraction was not necessary because the RNA fragmentation and bead clean-up that are part of the Kapa protocol resulted in library fragments of the desired size of 200–400 bp . Sequencing was performed for 100 bp single end on Illumina HiSeq 2500 instruments at the UCLA BSRC sequencing core for two lanes per batch , for 26 total lanes . On average , we obtained approximately 3 million reads per sample . Sequencing reads are available in SRA under the accession codes listed in the data availability statement . Adapter sequences were trimmed using trimmomatic ( Bolger et al . , 2014 ) . Reads were pseudoaligned to the 6713 annotated yeast ORF coding sequences from Ensembl build R64-1-1 using kallisto v . 43 . 0 ( Bray et al . , 2016 ) . Kallisto was run in strand-specific mode with parameters –l 150 and –s 8 . For each transcript , we computed transcripts per million reads ( TPM ) as a measure of expression and used log2 ( TPM + 0 . 5 ) for downstream analysis . Segregants with fewer than one million reads were removed from downstream analysis , and 1012 segregants passed this filter . We removed 993 invariant transcripts with identical expression across all segregants or with log2 ( TPM + 0 . 5 ) less than 1 in 50% or more of the segregants . Our final dataset included 5720 transcripts , which were used for downstream analyses ( Source data 1 ) . These transcripts cover 5506 of 5971 open reading frames annotated as ‘verified’ or ‘uncharacterized’ in the yeast genome ( Cherry et al . , 2012 ) . Unless otherwise specified , all remaining analyses were conducted in R ( www . r-project . org ) . Based on the OD measurements collected during growth prior to harvesting , growth rates were calculated for each segregant using the R package grofit and the function gcFitSpline ( Kahm et al . , 2010 ) . The difference between the maximum and minimum OD was recorded for each culture and used as a covariate for downstream analysis ( Source data 2 ) . Our BY and RM parent strains had earlier been sequenced to very high depth ( >200 fold coverage of the genome ) , and GATK ( McKenna et al . , 2010 ) used to identify 48 , 254 sequence variants between them . These variants ( irrespective of whether or not they are part of our marker map ) were screened for potential functional impact using the Ensembl Variant Effect Predictor ( McLaren et al . , 2016 ) . The segregant genotyping is described in ( Bloom et al . , 2013 ) and ( Bloom et al . , 2015 ) . The 1012 segregants used for this study were genotyped at 42 , 052 highly reliable markers , which are a subset of the total 48 , 254 sequence differences between BY and RM . Sets of markers that were in perfect linkage disequilibrium ( i . e . markers never separated by recombination ) among the 1012 segregants were collapsed to one marker . Our final linkage map comprised 11 , 530 unique markers ( Source data 3 ) . A variance component model was used to estimate additive heritability . First , gene expression measurements were corrected for batch covariates and the growth measurement covariate described above using a linear model for each gene P=DG + R where P is a vector of log2 ( TPM + 0 . 5 ) measurements for n segregants for that gene . D is a vector of estimated fixed effect coefficients for technical covariates . G is a matrix of n total segregants by m technical covariates . Technical covariates included experimental batch and the growth rate covariate described above . The vector of residuals is denoted as R . R contains expression phenotypes corrected for batch effect and growth covariate . We fit the variance component model R = a + e where a is the vector of additive genetic effects , and the residual error is denoted by e . The distributions of these effects are assumed to be normal with mean zero and variance–covariance as follows: a ~ N ( 0 , σ2AA ) ; e ~ N ( 0 , σ2EVI ) Here , A is the additive relatedness matrix – the fraction of the genome shared between pairs of segregants . A was calculated using the ‘A . mat’ function in the rrBLUP R package ( Endelman , 2011 ) . σ2A is the additive genetic variance captured by markers . σ2EV is the error variance and I is the identity matrix . Additive heritability was estimated using custom code adapted from Kang et al ( Kang et al . , 2008 ) . Although our heritability estimates are lower bounds due to counting noise in the number of sequencing reads per gene , downsampling of reads suggested that additional sequencing would increase heritability for most genes by at most a few percent ( Figure 1—figure supplement 1 ) . Additionally , we fit a model to estimate the relative contribution of pairwise interactions with R = a + i+e where i ~ N ( 0 , σ2AA ( A°A ) ) and A°A is the Hadamard ( entry-wise ) product of A , which can be interpreted as the fraction of pairs of markers shared between pairs of segregants . σ2AA is the interaction genetic variance captured by all pairwise combinations of markers . The other terms are the same as in the additive-only model . The result of ~1/10 as much variance arising from interactions relative to additive loci is based on the ratio of the average of the A°A term across genes to the average of the A term across genes . When we instead calculated this variance ratio for each gene , we found that the mean across genes was greatly inflated by a few extreme outliers , while the median was very low ( less than 1/100 ) because almost half of the genes had an estimate of zero for the A°A term . Gene positions were extracted from Ensembl ( Yates et al . , 2016 ) ( www . ensembl . org ) build 83 . Various analyses throughout the paper made use of a range of gene-specific features , factors and covariates: ( 1 ) Total variance in expression was calculated as the sum of the additive and residual variance components obtained in our heritability estimates . ( 2 ) Expression level was calculated as the mean log2 ( TPM ) across segregants , ( 3 ) Gene essentiality was coded as a binary factor and obtained from SGD ( Cherry et al . , 2012 ) ( www . yeastgenome . org ) by searching for genes whose SGD deletion phenotype contained the term ‘inviable’ . ( 4 ) dN/dS values were obtained from Supplementary Table S4 in ( Wall et al . , 2005 ) . ( 5 ) The number of protein-protein interactions was obtained from SGD by downloading all ‘physical’ interactions between genes and counting their number per gene . ( 6 ) Synthetic genetic interactions were extracted from data from Costanzo et al . ( 2016 ) which provides genetic interaction data for pairwise gene deletions or disruptions between nearly all essential ( E ) and nonessential ( N ) genes ( Costanzo et al . , 2016 ) . Specifically , we downloaded the ‘NxN’ , ‘NxE’ , and ‘ExE’ raw genetic interaction datasets from http://thecellmap . org/costanzo2016/ , combined them into one table , and extracted the lowest interaction p-value for each gene pair . We restricted this set using the ‘strict’ definition from ( Costanzo et al . , 2016 ) and kept only pairs with interaction p-value<0 . 05 and interaction strength ( epsilon ) >0 . 16 or<−0 . 12 . For each gene , we counted how many genes showed a genetic interaction at these thresholds and used this as our measure of synthetic genetic interactions . Using the ‘lenient’ or ‘intermediate’ definitions did not alter our conclusions . ( 7 ) We defined whether or not a gene is a transcription factor by downloading from SGD all genes annotated to the GO term GO:0003700 ‘transcription_factor_activity_sequencespecific_DNA_binding’ and its child GO terms . ( 8 ) As a proxy for deep evolutionary conservation , we extracted from Ensembl biomart whether or not a gene has a human homolog . Gene ontology ( GO ) associations for each gene were downloaded from the Gene Ontology Consortium ( geneontology . org ) on February 16 , 2016 . We used paralogy information downloaded from the yeast gene order browser ( Byrne and Wolfe , 2005 ) ( http://ygob . ucd . ie/ ) . We tested for gene features associated with the degree of heritability by multiple linear regression . This regression modeled heritability as the dependent variable and the various gene features as predictor variables . We used the ‘summary’ and ‘lm’ functions , and the ‘car’ package in R to perform Type III sum-of-squares ANOVA . This analysis tests for the influence of each feature by dropping it from a full model that includes all other terms , and asking whether this results in a significantly worse fit as judged by F-statistics . The analysis controls for correlations among predictor variables and reports marginal associations only if they are significant over all other terms . We did not include interaction terms among predictor variables . We tested for GO enrichments using the R package topGO ( Alexa et al . , 2006 ) . For analyses in which genes were classified as ‘interesting’ or not ( e . g . whether a gene has heritability ≥90% , or whether it is located in a hotspot ) , we used the Fisher test for enrichment . When using a quantitative gene score as the measure of interest ( e . g . the heritability ) , we used the one-sided t . test implemented in topGO . We used the ‘classic’ scoring method ( Alexa et al . , 2006 ) , that is we did not adjust the enrichments for significance of child GO terms . We ( Bloom et al . , 2015 ) and others ( Yang et al . , 2014; Zeng , 1994 ) have previously noticed that power and precision of QTL mapping on a given chromosome can be increased by controlling for genetic contributions that arise from the other chromosomes in the genome . Our eQTL mapping strategy controls for genomic background in two ways . For each gene , we identified large genetic effects segregating on other chromosomes and included them as covariates while mapping on a given chromosome . We also corrected for any additional polygenic additive background signal on other chromosomes . Then , for each gene we used a forward stepwise procedure to map eQTLs with a false discovery rate procedure . Below we describe our algorithm in greater detail . Throughout , we use the terms ‘eQTL’ and ‘linkage’ interchangeably . For each chromosome of interest and for each gene expression trait we calculated P=DG + CLZL+aL + R CL and ZL are the background eQTL effects identified from the procedure above that are not located on the chromosome of interest . aL ~N ( 0 , σ2aLAL ) σ2aL is the additive genetic variance from all chromosomes excluding the chromosome of interest . AL was calculated using the ‘A . mat’ function in the rrBLUP package using a genetic relatedness matrix that excludes markers from the chromosome of interest . The goal of this step was to obtain expression phenotypes R that can be used to scan for eQTLs on a given chromosome by correcting for sources of variation that do not arise from that chromosome: batch and growth effects , large effects on other chromosomes , and a polygenic term accounting for any additional genetic contributions arising from other chromosomes . The amount of additive variance explained by detected eQTLs was estimated using cross-validation . Segregants were grouped based on the batches used for RNA and library preparation . Each batch of segregants was left out of the procedure one at a time . The eQTL mapping procedure was performed for all the other batches . For the QTL markers detected in this training set and with effects estimated in the training set , the amount of variance explained by the joint model of the set of significant QTL markers was estimated in the held out batch . eQTL confidence intervals were calculated as 1 . 5 LOD drops . We extended the eQTL location confidence intervals to include all markers in perfect LD with the markers used in eQTL detection ( marker correlation = 1 ) . We devised an algorithm with the goal of identifying a set of eQTL hotspots by combining information across genetically correlated transcripts and , most importantly , using co-localizing trans-eQTLs to better narrow hotspot confidence intervals . The algorithm has three major steps . First , we control for unmodeled factors affecting gene expression that may obscure hotspot detection and localization . Second , we use a multivariate statistic to identify eQTL hotspots . Finally , we use a bootstrap procedure to delineate confidence intervals for hotspot location . We describe the steps in greater detail below . To classify an eQTL as local , we required its location confidence interval to overlap the position of the gene . We used gene locations expanded by 1000 bp upstream and 200 bp downstream to account for regulatory variants that may be located in the promoter or the 3’UTR . We initially classified 2969 eQTLs as local . These 2969 eQTLs affected 2884 genes . Closer inspection revealed that these multiple ‘local’ eQTLs per gene often involved one eQTL with a peak very close to the gene and other , more distant eQTLs . These more distant eQTLs probably reflect trans-eQTLs on the same chromosome as their target gene with location confidence intervals broad enough that they happened to overlap the target gene . For our ASE comparisons below , we only used the local eQTLs that were located closest to a given gene . We used ASE data from two sources . The first source is the mRNA data from Supplementary Data S2 from ( Albert et al . , 2014a ) . The second source is previously unpublished data generated by Dr . Noorossadat Torabi in the Kruglyak laboratory ( available in SRA under accession code SRP149494 ) . Both datasets performed mRNA sequencing on a BY/RM diploid hybrid strain . Reads from Albert et al . were ~30 bp in length to match ribosome profiling data presented in that paper , while reads from Torabi et al . were 100 bp . In contrast to ( Albert et al . , 2014a ) , the Torabi data was not strand-specific . We processed the Torabi data exactly as described in ( Albert et al . , 2014a ) . Briefly , reads were aligned to both the BY reference genome and the RM reference genome provided by the Broad Institute ( https://www . broadinstitute . org/fungal-genome-initiative/saccharomyces-cerevisiae-rm11-1a-genome-project ) . We retained reads that mapped uniquely and without mismatch . We considered reads mapping to a set of coding SNPs carefully curated to only contain SNPs with good mapping characteristics ( Albert et al . , 2014a ) and counted the number of reads arising from each allele . Counts for multiple SNPs per gene were summed . We performed hypergeometric downsampling to account for a small difference ( 0 . 5% ) in total read counts mapping to BY vs . RM alleles . Genes with fewer than 20 reads were discarded . Significance of ASE was gauged using a binomial test , and p-values Bonferroni-corrected for multiple testing . Effect sizes are expressed as log2-transformed fold changes , which in turn are calculated as the RM allele count divided by the BY allele count for each gene . The Torabi and Albert dataset are independent replicates of a BY/RM hybrid . We can use this fact to gain confidence in significance calls ( we keep track of whether a gene was determined to have significant ASE in one or both of the datasets ) and the magnitude of ASE ( for each gene , we use the mean log2 fold change of the two datasets ) . Across the two datasets , we had ASE data available for 3340 genes . To compare effect sizes between eQTLs and ASE , we used log2-transformed eQTL fold changes re-computed on un-scaled expression data . ASE fold changes were computed as the log2 of the ratio of RM to BY allele counts . Comparisons of effect size across genes were performed using standardized major axis ( SMA ) analysis ( Warton et al . , 2012 ) . To correlate ASE and eQTL effect sizes to the number of sequence variants upstream of each gene , we defined the upstream interval as the sequence upstream of the start codon up to the neighboring gene for a maximum of 1 , 000 bp . We acknowledge that the length of sequence considered is therefore different for different genes but believe that this is a reasonable approximation of the regulatory upstream regions in yeast ( Lin et al . , 2010 ) . Sequence variants had been obtained from short-read sequencing of the BY and RM strains used in this study ( Bloom et al . , 2013 ) . All data necessary to reproduce the ASE analyses and eQTL comparisons is available in Source data 7 . To gauge the statistical power to detect ASE in the two available ASE datasets , we performed simulations . We focused on two variables: the effect size ( i . e . fold change ) and the total read coverage available for the given gene . ASE data is overdispersed compared to a binomial distribution ( Castel et al . , 2015 ) . To properly account for this overdispersion in our simulations , we used the available ASE data to estimate the overdispersion parameter ρ in a beta-binomial distribution using the R function ‘optim’ . We provided the function with the total read count and the observed allele count for each gene and used 0 . 5 as the ‘true’ probability of success . We estimated ρ separately for the Torabi and the Albert data and found that the latter was somewhat more overdispersed ( ρ = 0 . 0054 ) than the former ( ρ = 0 . 0041 ) . We confirmed that these overdispersion estimates fit the data reasonably well by generating simulated datasets using the observed read counts from the Albert and Torabi datasets for each gene , along with the estimated ρ’s . For each gene , we used the mean of the Albert and Torabi ASE fold changes to compute the target probability of success . These simulated data are meant to represent a random instance of allelic counts across genes , using distributions that closely mimic the real data . We generated 50 such datasets using the Albert and Torabi parameters , respectively . We computed all pairwise correlations between these simulated Albert and Torabi datasets . The median correlation coefficient across genes was r = 0 . 42 in the simulated data , compared to r = 0 . 39 for the observed Albert and Torabi data . Thus , while the simulations underestimate the overdispersion in the real data by a small amount , they provide a reasonable approximation . We used these estimates of ρ to simulate 1000 instances of allele counts for several combinations of total read count and fold change ( Figure 2—figure supplement 2 ) . We computed power as the fraction of simulations in which a binomial test yielded p<0 . 05 or p<1 . 5e-5 , the Bonferroni-corrected threshold for 3340 tests . We also computed the fraction of simulations in which the direction of fold change agreed with the true change . These simulations show that power increases with increasing read counts and effect sizes . Crucially , even with very high sequencing depth of 5000 allele-specific reads per gene , the power to detect ASE of a magnitude typical for the majority of local eQTLs detected here is less than 60% . Most genes in the ASE data have substantially lower coverage: the median coverage is ~1000 in the Torabi data and ~200 in the Albert data . Thus , we expect to miss the ASE effects of the majority of local eQTLs . To explore this relationship more precisely , we conducted gene-matched power simulations . For each gene , we conducted 100 simulations using the estimated ρ and sequencing coverage for the given ASE dataset . We used the observed eQTL fold changes to compute expected probabilities of success . We conducted these simulations separately for the Albert and Torabi data . To estimate which genes are influenced by a given hotspot irrespective of whether these associations reached genome-wide significance for the given gene , we performed a targeted forward scan for linkage at the hotspot locations . For each chromosome and for each gene , we regressed out the effects of significant eQTLs detected on other chromosomes as well as the effects of technical covariates . Then , for each gene , we repeated the procedure described above under ‘Mapping additive eQTL’ . However , here X was defined to be the set of detected hotspot markers for that chromosome , as well as either the closest marker to each gene or , if a gene had a genome-wide significant local eQTL , the local eQTL peak marker for that gene . The same FDR threshold ( 5% ) was used to identify the best model for each gene . For each gene , coefficients for the effects of all significant markers identified by this procedure were determined by multiple regression . The regression coefficients from this model were used to count the number of genes affected by a given hotspot ( i . e . all nonzero entries ) and to rank genes for inclusion in gene ontology ( GO ) and transcription factor ( TF ) binding site ( TFBS ) enrichments , and for plotting in networks . Source data 8 also presents the number of genes with a genome-wide significant eQTL overlapping each hotspot . For each hotspot , TFBS and GO enrichments were calculated on the 100 genes with the strongest increases or decreases in expression from the RM allele at the hotspot , respectively . When fewer than 100 genes were influenced by the hotspot in the given direction , we used all genes influenced in this direction . In Source data 8 , we report GO enrichments for the genes affected by each hotspot . This table shows the top five GO categories for each hotspot that exceeded an enrichment p-value of p≤0 . 05 divided by the number of GO categories tested . We adjusted p-values separately for the Biological Process , Molecular Function , and Cellular Compartment GO subtrees . We did not adjust for the fact that multiple hotspots were tested , nor for the fact that two directions of effect ( higher expression linked to the BY or the RM allele ) were tested . Source data 10 lists all GO enrichment results . TFBS enrichments for genes affected by hotspots were computed using Fisher’s exact test . The underlying relationships between TFs and target genes were based on regulatory relationships between genes downloaded from SGD ( Cherry et al . , 2012 ) on May 1st , 2016 . In Source data 8 , we show all TFs with significant enrichments at a threshold of p≤1 . 3e-6 , corresponding to a Bonferroni-corrected p-value controlling for 38 , 556 tests ( 189 TFs , 102 hotspots , and two directions of effect per hotspot ) . Source data 8 also shows the most significant TFBS enrichment , irrespective of whether this enrichment was significant after multiple testing or not . Source data 11 lists all TFBS enrichment results . We displayed the genes that are affected by each hotspot in a manner that reflects significant correlations between the genes’ expression levels in order to highlight groups of genes that may be functionally related . Because expression levels will become correlated when they are linked to the same hotspot , we wanted to estimate the underlying correlation matrix in a manner that is as free as possible from correlations induced by the hotspots . We fit a sparse conditional Gaussian graphical model ( sCGGM [Zhang and Kim , 2014] ) to the expression data and the genotypes at the hotspot markers . sCGGM decomposes the total correlation matrix among gene expression levels into ‘direct’ effects of the genetic markers on each gene expression level as well as a matrix θyy of ‘indirect’ effects that genes exert on each other . We fit the model on a multi-core processor using an R installation compiled using the Intel Math Kernel Library to speed up linear algebra operations , as provided by the University of Minnesota Supercomputing Institute . Following preliminary cross-validation on a subset of 2000 genes , we set the sCGGM regularization parameters λ1 and λ2 to 0 . 1 . We fit log2 ( TPM ) values for all 5720 expressed genes . Prior to fitting , effects of experimental batch and yeast culture optical density were removed using a linear model . We also removed the effect of the marker closest to each gene . This latter correction was not performed for genes that reside in the window tested by bootstrapping around each hotspot , and for genes with a local eQTL that overlaps the given hotspot . We chose to keep the local effects for these genes because local eQTLs at these genes may underlie the hotspot , and we were interested in preserving such potential ‘direct’ local effects for our visualizations . We used the entries of θyy to generate the network plots in Figure 4—figure supplement 4 and Supplementary file 6 . In spite of the shrinkage imposed by the sCGGM algorithm , very few entries were estimated to be zero . As a practical threshold for plotting , we excluded entries of θyy with absolute values less than 1e-5 . 5 . This threshold was set based on visual inspection of a histogram of all entries in θyy , which showed a bimodal distribution with a clear peak of values exceeding this threshold separated from a peak centered on much smaller values . Network plots were generated using the R igraph package ( Csárdi and Nepusz , 2006 ) . Supplementary file 6 shows the resulting network plots for all hotspots . Several hotspots were located close to chromosome ends . Yeast chromosome ends contain complex structural variation that segregates among isolates and influences traits ( Cubillos et al . , 2011 ) . In some cases , BY and RM differ for the presence of entire subtelomeric blocks of genes ( Bergström et al . , 2014 ) . When a hotspot arises from these regions , the identity of the causal gene cannot be determined using our present segregant panel because each segregant either carries all or none of the genes in these regions . Further , the marker map we used for mapping stops at the borders of these regions . Therefore , these hotspots often have very sharp bootstrap distributions on the first or last marker of the linkage map on the given chromosome . We excluded subtelomeric hotspots from the analyses of genes located in hotspots because the position of the final marker on a chromosome is unlikely to reflect the position of the causal gene , which may well be located distally to the marker . We excluded 13 hotspots whose peak marker is within 5 kb of the end of our linkage map . We focused the remaining analyses of hotspot genes on 26 non-subtelomeric hotspots with confidence regions that contain three or fewer genes , for a total of 58 genes . To analyze features of genes located in hotspots , we performed multiple logistic regression . The dependent variable indicates whether or not a gene resides in one of these hotspots , and the set of gene features described above served as potential predictor variables . Significance tests were performed using likelihood ratio tests for dropping each term from a full additive model without interactions , as implemented in the R car package . Gene ontology analysis on these genes was conducted using Fisher’s exact test in topGO ( Alexa et al . , 2006 ) . Genes in the yeast genome tend to be clustered in co-localized groups with similar functions ( Hurst et al . , 2004 ) . To test if this clustering influences our GO analyses of genes located in hotspots , we performed a randomization analysis . We sampled 1000 sets of 58 neighboring genes that mirror the distribution of the number of genes across the 26 hotspots . For each set , we performed the same GO enrichment analysis as for the actual 58 hotspot genes . Within each set , we counted the number of significant ( at p<0 . 05 ) GO terms and used the fraction of this distribution that matched or exceeded the number observed in the real set as an empirical p-value for whether the enrichment was globally significant . This test showed significantly more enriched terms than randomly expected for the GO Biological Process category ( p=0 . 001 ) , but only a marginal excess for Molecular Function ( p=0 . 060 ) and no significant excess for Cellular Compartment ( p=0 . 695 ) . This analysis is conservative because in the real data we considered GO terms at much more stringent p-value cutoffs than ≤0 . 05 . We further explored the FDR for GO term enrichment at more stringent p-values by dividing the mean number of terms significant at a given threshold across the permutations by the number of significant terms observed in real data . We found that for Biological Process , FDR was <0 . 05 for GO enrichments with p<0 . 005 , which includes all terms described in the paper . Finally , we computed an empirical p-value for each GO term by asking how often its observed p-value is matched or exceeded in the permutations . This analysis controls for different sizes and compositions of the different GO terms . All terms reported to be significant in the text had p<0 . 001 in these analyses; enrichments as strong or exceeding the observed ones were never seen in 1000 random gene sets . We conclude that our GO analysis of genes in hotspots is unlikely to reflect random sampling of genomic regions . Plots of hotspot location and gene content were generated using the R package Gviz ( Hahne and Ivanek , 2016 ) . Supplementary file 5 shows plots of gene content for all hotspots . We used pQTLs for 160 proteins identified in the BY/RM cross ( Albert et al . , 2014b ) . The pQTL coordinates were mapped from the sacCer2 to the sacCer3 genome using the UCSC liftover tool ( https://genome . ucsc . edu/cgi-bin/hgLiftOver ) . The pQTLs were mapped using bulk-segregant analysis ( BSA ) in large pools of segregants where each gene was tagged with green fluorescent protein ( Albert et al . , 2014b ) . BSA does not produce effect sizes in units of gene expression levels or variance . We instead used the allele frequency difference between high and low GFP pools at the pQTL peak position as a measure of pQTL effect size . For eQTLs , we used the coefficient of the correlation between scaled expression levels and marker genotype . We chose this effect size measure because it , like the BSA allele frequency estimates , is bounded by −1 and 1 , resulting in more easily interpretable scatterplots . Using other measures of eQTL effect such as multiple regression coefficients did not change our conclusions about pQTL overlap . For the comparison of strong eQTLs to significant pQTLs , we defined ‘strong’ eQTLs as those eQTLs that explained ≥3 . 5% of phenotypic variance . Our eQTL data had ≥99% power to detect such eQTLs . Under the assumption that the bulk-segregant based pQTL data had similar statistical power as the current eQTL data , these eQTLs should be easily detectable as pQTL if their effects on proteins are similarly strong . For the comparison of strong pQTLs to significant eQTLs , we defined ‘strong’ pQTLs as pQTLs with LOD ≥15 . This threshold was chosen to pick a set of strong pQTLs that had similar size ( 218 pQTLs ) to the set of strong eQTLs ( 238 eQTLs ) . For eQTLs that did not overlap a significant pQTL or vice versa , we used the effect size point estimate at the respective peak position in the non-significant dataset for plotting in Figure 6B and D and for results presented in Supplementary files 8 and 10 . Data underlying these analyses is available as Source data 13 . To compute π1 at eQTL positions in X-pQTL data , we summed counts for 21 SNPs centered on the given eQTL marker for the BY and RM allele in the high and low GFP populations ( Albert et al . , 2014b ) , performed a G-test on these summed counts , and analyzed the resulting p-values using the qvalue package in R ( Storey and Tibshirani , 2003 ) . We chose to sum counts from multiple SNPs because in X-QTL data with relatively low sequencing coverage , the counts at any one SNP are subject to counting noise . For reference , π1 computed based only at the nearest SNP to each eQTL was 0 . 14 for the comparison using all eQTLs , compared to 0 . 66 using the 21 SNP window . We computed π1 on 5 sets of randomly selected genome positions . These random π1 values were all ≤0 . 52 . These high random estimates reflect the high density of X-pQTLs in the genome as well as a potential upward bias due to our choice of SNP window , but are all clearly lower than the observed value . To compute π1 at pQTL positions in eQTL data , we performed t-tests comparing the normalized expression levels for segregants with the BY allele at pQTL markers to those with the RM allele . The resulting p-values were analyzed using the qvalue package ( Storey and Tibshirani , 2003 ) . For each transcript with at least one significant additive eQTL , we fit a model that included the batch and growth covariates , the significant additive eQTLs , and a random effect for polygenic background . The residuals from this model were used for the detection of eQTL-eQTL interactions . The set of markers used for the detection of eQTL-eQTL interactions was reduced to 3106 using the findCorrelation function in the R package caret ( https://cran . r-project . org/web/packages/caret/index . html ) , using a cutoff of 0 . 99 . All unique combinations of markers were tested for each transcript , with the exclusion of the 20 closest markers on the same chromosome . eQTL-eQTL peaks , including the rare case of closely linked eQTL-eQTL interactions occurring on the same chromosome pair , were identified using custom code provided in our code repository . The same procedure was repeated for five random permutations of segregant identities . A false discovery rate was calculated as the ratio of expected to observed peaks at different LOD thresholds . A false discovery rate for the marginal scan was calculated as the ratio of expected to observed peaks at different LOD thresholds for the subset of marker pairs where one of the pairs had a significant additive effect . False discovery rate was controlled at 10% for both scans . Additionally , we tested a model of eQTL-eQTL interactions between significant additive eQTLs only . For each gene , we regressed out the effects of significant additive eQTLs , the effect of polygenic background , and the effects of technical covariates . Then , for each gene we tested only the eQTL-eQTL interaction effect between significant additive eQTL markers . This procedure involved the peak markers detected in the section on eQTL mapping without any marker downsampling . An F-statistic was calculated for each test . The same procedure was repeated 10 times with permutations of segregant identities . From the permutations , the expected number of significant eQTL-eQTL linkages was calculated at various thresholds . False discovery rate was controlled at 10% for this procedure . Source data 14 lists all identified genetic interactions . Circle plots were generated using the ggBio package in R ( Yin et al . , 2012 ) . Pair markers were deemed to overlap an additive eQTL hotspot if they were within 10 kb of the given hotspot interval . | Every individual’s genome is unique , with variations in the DNA sequence at many thousands of points . Each difference is a change in one or more ‘letters’ of the DNA code . Some of these DNA letter variations have consequences for the way the individual looks or behaves . They can influence these traits either by changing the sequence of a protein encoded by a gene; or by changing when , where or how much a gene is active . Studying how individual differences in the DNA influence gene activity requires a very large amount of data on many individuals within a species . Only recently have such large datasets become available . These have made it possible to study these regulatory differences in unprecedented detail . Albert , Bloom et al . set out to map as many regulatory genetic variants as possible in budding yeast – a popular model organism used in many branches of science . The approach involved measuring how active every gene in the genome was , and which genetic variants influenced whether each gene’s activity was turned up or down , in more than 1 , 000 different strains of yeast . Thousands of regions of the DNA turned out to influence regulation of genes . The analysis revealed that almost every gene is influenced by a complex set of regulatory regions all over the genome . Some hotspot regions were found to influence thousands of genes at once . The findings provide the most complete set of data for studying the effects of variation in DNA sequence on genetic regulation in any species , and can act as a model for researchers to carry out similar experiments in other species . Ultimately , these results could help understand exactly how differences in genome sequence help to make individuals unique . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"genetics",
"and",
"genomics"
] | 2018 | Genetics of trans-regulatory variation in gene expression |
During cell locomotion and endocytosis , membrane-tethered WASP proteins stimulate actin filament nucleation by the Arp2/3 complex . This process generates highly branched arrays of filaments that grow toward the membrane to which they are tethered , a conflict that seemingly would restrict filament growth . Using three-color single-molecule imaging in vitro we revealed how the dynamic associations of Arp2/3 complex with mother filament and WASP are temporally coordinated with initiation of daughter filament growth . We found that WASP proteins dissociated from filament-bound Arp2/3 complex prior to new filament growth . Further , mutations that accelerated release of WASP from filament-bound Arp2/3 complex proportionally accelerated branch formation . These data suggest that while WASP promotes formation of pre-nucleation complexes , filament growth cannot occur until it is triggered by WASP release . This provides a mechanism by which membrane-bound WASP proteins can stimulate network growth without restraining it .
Control of actin dynamics is essential to many cellular processes , including motility , vesicle trafficking , and cell division ( Pollard and Cooper , 2009 ) . The Actin related protein 2/Actin related protein 3 ( Arp2/3 ) complex nucleates new ( daughter ) filaments from the sides of existing ( mother ) filaments in response to activating stimulus from the Wiskott-Aldrich Syndrome Protein ( WASP ) family ( Pollard , 2007; Padrick and Rosen , 2010; Campellone and Welch , 2010 ) . Membrane-associated WASP proteins integrate upstream signals and activate Arp2/3 complex in the correct place and time to produce actin structures that perform a variety of cellular functions including motility . The verprolin homology-central-acidic ( VCA ) domain of WASP family proteins binds to monomeric actin and the Arp2/3 complex , and is both necessary and sufficient for the WASP proteins to stimulate nucleation ( Machesky and Insall , 1998; Miki and Takenawa , 1998; Machesky et al . , 1999; Rohatgi et al . , 1999; Pollard , 2007 ) . VCA acts to promote daughter nucleation by Arp2/3 complex in several ways . VCA engagement promotes a conformational change in Arp2/3 complex that repositions Arp2 and Arp3 ( Robinson et al . , 2001; Goley et al . , 2004; Rodal et al . , 2005; Rouiller et al . , 2008; Xu et al . , 2012; Hetrick et al . , 2013 ) . This conformational change is thought to be required to initiate daughter filament growth . Further , the initial monomers of the nucleated filament are delivered by the V region , or WASP homology 2 ( WH2 ) domain ( Rohatgi et al . , 1999; Hertzog et al . , 2004; Irobi et al . , 2004; Chereau et al . , 2005; Padrick et al . , 2011 ) . However , the Arp2/3-VCA-actin complex is not active on its own; an additional stimulus must be provided by the mother filament ( Mullins et al . , 1998; Achard et al . , 2010 ) . Thus , when stimulated by VCA the Arp2/3 complex only nucleates filaments from the sides of existing filaments . This produces branched filament arrays in vitro ( Mullins et al . , 1998; Machesky et al . , 1999; Blanchoin et al . , 2000; Amann and Pollard , 2001; Achard et al . , 2010 ) , which resemble the branched networks found in cells ( Svitkina and Borisy , 1999; Vinzenz et al . , 2012 ) . An additional feature of the system is that simultaneous binding of two VCA peptides to Arp2/3 complex greatly potentiates daughter nucleation ( Padrick et al . , 2011 ) . It is likely that WASP oligomerization is a broadly used mechanism of activation ( Padrick et al . , 2008 ) , and a number of cellular factors that dimerize or multimerize WASP have been identified ( Padrick and Rosen , 2010 ) . The molecular interactions and structural rearrangements outlined above contribute to VCA acting at more than one step in the nucleation pathway , although the pathway is not fully defined . It is established that VCA stimulates branch formation by accelerating the association of Arp2/3 complex with the mother filament ( Smith et al . , 2013 ) . In addition , there is evidence from kinetic analyses that a VCA-dependent ‘activation step’ follows filament binding during nucleation ( Marchand et al . , 2001; Zalevsky et al . , 2001; Beltzner and Pollard , 2008; Smith et al . , 2013 ) . The nature of this step remains unknown but has been hypothesized to arise from conformational changes in the Arp2/3 complex ( Marchand et al . , 2001; Beltzner and Pollard , 2008 ) . The activation step substantially limits the efficiency of nucleation ( Smith et al . , 2013 ) . An interesting feature of WASP activation of daughter nucleation is that Arp2/3 complex must associate with membrane bound activators at an early stage in the process and yet be separated from those activators at a subsequent stage . In cells , branched filament networks have their barbed ends directed toward membranes ( Small et al . , 1978; Svitkina and Borisy , 1999; Pollard and Borisy , 2003; Vinzenz et al . , 2012 ) . The characteristic geometry of the branches nucleated by the Arp2/3 complex dictates that both mother and daughter filaments grow toward the membrane . However , WASP proteins are linked with activators on the membrane ( Padrick and Rosen , 2010 ) , so that VCA-bound Arp2/3 complex should be tethered to the membrane . This tethering creates a steric problem , in that the growing ends of the filaments are held against , and possibly have their growth limited by , the membrane . However , this problem is eventually resolved ( Figure 1 , right ) . In lamellipodia , VCA-containing WAVE proteins stay associated with the leading edge , while the Arp2/3 complex is distributed throughout the actin mesh ( Lai et al . , 2008 ) . In both budding and fission yeasts , Arp2/3 complex is separated from membrane-bound activators during endocytosis ( Kaksonen et al . , 2003; Sirotkin et al . , 2005 ) . Analogously , in propulsive actin ‘comet tails’ Arp2/3 complex is found throughout the tail while its activators stay ( largely ) associated with the motile bacterium , virus , or vesicle ( Welch et al . , 1997; Egile et al . , 1999; Loisel et al . , 1999; Taunton et al . , 2000; Weisswange et al . , 2009 ) . In vitro , branches are released from the budding yeast WASP family member Las17 ( Martin et al . , 2006 ) . In all of these systems Arp2/3 complex can disengage from the surface attached activator within a short time of the onset of daughter filament growth . 10 . 7554/eLife . 01008 . 003Figure 1 . Pathway of Arp2/3 complex mediated actin branch formation activated by WASP protein dimers on the inside surface of a cell membrane , as deduced from previous studies . Within the white arrow Arp2/3 complex is activated by VCA , detaches from the membrane and initiates daughter filament elongation . The order of these steps and how they are coordinated remains unclear . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 003 These observations raise a fundamental question: how is the binding of Arp2/3 complex to filament sides coordinated with binding and release of VCA and the initiation and growth of the daughter filament ( Figure 1 ) ? Previous studies have proposed that VCA may dissociate from Arp2/3 complex prior to initiation of daughter filament growth . The observation that ATP hydrolysis by Arp2 decreases affinity for VCA suggested a possible trigger for dissociation ( Dayel et al . , 2001; Dayel and Mullins , 2004 ) , although hydrolysis was later shown to be dispensable for filament nucleation and to instead control disassembly of the branch ( Martin et al . , 2006; Ingerman et al . , 2013 ) . Consideration of biochemical and structural data on WH2-actin interactions ( Egile et al . , 1999; Higgs et al . , 1999; Hertzog et al . , 2004; Irobi et al . , 2004; Chereau et al . , 2005; Boczkowska et al . , 2008 ) led to speculation that during Arp2/3-mediated nucleation , the WH2 region of VCA may initially block the barbed end of actin delivered to the Arp2 or Arp3 subunit and thus must move or dissociate prior to daughter filament elongation ( Padrick et al . , 2011; Ti et al . , 2011; Hetrick et al . , 2013 ) . However , direct observation of the sequence of molecular events associated with the initiation of daughter filament growth and the release of VCA has been lacking . Multi-wavelength single-molecule fluorescence colocalization methods ( Hoskins et al . , 2011; Friedman and Gelles , 2012 ) are a powerful approach to elucidating the reaction pathways and identifying key regulated steps in processes that involve multiple macromolecular components . Here we used three-color single molecule fluorescence experiments to directly visualize the sequence and rates of the key steps in the pathway through which VCA dimers , Arp2/3 complexes , and actin filaments associate with one another and generate a new actin branch . The work reveals that the activation step in nucleation is likely the release of VCA dimers from the nascent branch , such that VCA dissociation is the trigger for daughter filament growth . The problem of filament growth against a membrane to which it is tethered is therefore solved by a mechanism in which release from the membrane tether is required for filament initiation .
To follow the coordination of VCA association with Arp2/3 complex during actin branch formation we labeled each protein with a fluorescent probe and visualized their colocalization dynamics using colocalization single molecule spectroscopy ( CoSMoS ) ( Friedman et al . , 2006; Hoskins et al . , 2011; Friedman and Gelles , 2012 ) . Actin was labeled with a blue-excited dye ( on 10% of monomers ) and tagged with biotin ( on 1% of monomers ) to enable tethering to microscope slides . Saccharomyces cerevisiae Arp2/3 complex was labeled with a red-excited dye targeted to a SNAP tag fused to the C-terminus of the Arc18 ( ArpC3 ) subunit ( Smith et al . , 2013 ) . We used a green-excited , Cy3 dye bis-maleimide derivative to label and covalently dimerize the VCA from N-WASP , hereafter called diVCA , which includes the second WH2 motif ( V ) through the C-terminus of the protein . Like other dimeric VCAs ( Padrick et al . , 2008 ) ( Figure 2—figure supplement 1 ) , this diVCA construct was able to stimulate the activity of the Arp2/3 complex ( Figure 2—figure supplement 2; ‘Materials and methods’ ) at low-nanomolar concentrations suitable for single molecule imaging , whereas labeled monomeric VCA constructs did not . Using this combination of tagged proteins ( Figure 2A ) we directly observed individual Arp2/3 complex and diVCA molecules binding to immobilized actin filaments and nucleating new branches . During this process , single molecules of Arp2/3 complex and diVCA were observed to bind together to locations on filament sides ( e . g . , Figure 2B–C at t = 0; Figure 2—figure supplement 3 ) . In nearly all cases ( 83 ± 9% S . E . , based on 877 Arp2/3 complex observations ) , Arp2/3 complex and diVCA arrived simultaneously within the experimental time resolution ( 0 . 15 s ) , indicating that diVCA was bound to Arp2/3 complex prior to filament engagement , and that both proteins bound to filaments as a unit ( ‘Materials and methods’ ) . The sparse labeling of actin and high fluorescence intensity of mother filaments did not permit us to detect the arrival of actin monomers with VCA and Arp2/3 complex . However , under the reaction conditions ( 5 nM diVCA and 1 μM actin ) , with a KD of diVCA for actin of ∼300 nM ( which is only slightly altered by the presence of Arp2/3 complex , Figure 2—figure supplement 4 ) , we expect ∼77% of diVCAs to have at least one actin bound . Thus , most of the diVCA-Arp2/3 complexes observed to bind mother filament should contain actin and thus have all the molecular factors needed for nucleation . We hereafter refer to this filament-bound complex as the ‘nascent branch’ , an intermediate in the pathway to daughter filament assembly . 10 . 7554/eLife . 01008 . 004Figure 2 . Rapid release of dimeric VCA from the nascent branch precedes nucleation . ( A ) Design of an experiment to observe diVCA-activated branch nucleation by Arp2/3 complex on the sides of surface-immobilized actin filaments . Blue , green and red stars denote fluorescent dye labels AlexaFluor 488 ( AF488 ) , Cyanine 3 ( Cy3 ) , and Dy649 that are excited with blue , green , and red lasers , respectively . ( B ) Image sequence of the same microscope field of view taken at each of the three dye wavelengths ( rows ) at five selected time points ( t; columns ) . Images record the colocalization of an individual Arp2/3 complex and diVCA molecule at t = 0 ( yellow arrowhead ) followed by nucleation and growth of a daughter filament at that location ( red arrowhead ) . Solution contained 5 nM Cy3-diVCA , 5 nM SNAP-tagged Arp2/3 complex labeled with Dy649 ( Arp2/3-SNAP649 ) , and 1 µM actin , 10% AF488-labeled . Bar: 1 μm . See Video 1 . ( C ) Recordings of daughter filament length and branch site fluorescence intensities from the nucleation event in B . Arrow marks the time of daughter filament nucleation estimated by extrapolating the daughter length fit line to zero length ( Smith et al . , 2013 ) . Plot at bottom is a magnified view showing that Arp2/3 complex and diVCA labels appear simultaneously ( t = 0 ) followed by rapid release of diVCA ( t = 0 . 2 s ) . ( D ) Cumulative lifetime distributions of Arp2/3 complex and diVCA on filament sides after binding of an Arp2/3-diVCA complex to the filament ( N = 752 ) . Smooth lines indicate two- ( diVCA ) or three-exponential ( Arp2/3 complex ) fits yielding the indicated fit parameters ( ‘Materials and methods’ ) . Main plot shows the data for time <10 s; inset shows the full distribution with the exception of one outlier . ( E ) Comparison of the time ( ±S . E . ) of daughter filament initiation with the time of diVCA release from the nascent branch in individual branch nucleation events by diVCA-Arp2/3 complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 00410 . 7554/eLife . 01008 . 005Figure 2—figure supplement 1 . N-WASP VCA dimers bind tightly to Arp2/3 complex and stimulate its actin nucleation activity . ( A ) Schematic of Glutathione-S-transferase ( GST ) fusions with N-WASP VCA-containing domains . N-terminal GST-fusions , which are naturally tight dimers , were made to VVCA ( N-WASP residues 393–505 ) , VCA ( residues 432–505 ) , and NL-VCA ( NL [gray] is a sequence , residues 420-–430 , found N-terminal to the human WASP VCA ) . ( B ) GST-VCA stimulates actin polymerization by Arp2/3 complex . Records indicate pyrene actin fluorescence increase due to filament polymerization 2 µM actin , 10% pyrene-labeled and bovine Arp2/3 complex , with or without GST-VCA . ( C ) GST-VVCA is more active at saturation than is GST-VCA . Bars indicate the number ( ±S . E . ) of actin filament barbed ends present at the midpoint of polymerization assays like those shown in B . The concentration series show that stimulation by all constructs saturates at or below 25 nM and that dimer with two V domains per subunit ( GST-VVCA ) is more active that the single V domain construct ( GST-VCA ) even when an extended linker is incorporated into the latter ( GST-NL-VCA ) . ( D ) A dye-labeled , single cysteine mutant N-WASP VVCA-A462C-AF594 binds to bovine Arp2/3 complex , as determined by fluorescence anisotropy ( points ) . Fit to a direct binding isotherm yields KD = 150 ± 5 nM ( 68% C . I . ) . ( E ) Binding of GST-VVCA and GST-VCA to bovine Arp2/3 complex ( 200 nM ) . Points indicate interference with binding of VVCA-A462C-AF594 in a fluorescence anisotropy assay like that in D caused by addition of the indicated amount of unlabeled competitor GST-VCA or GST-VVCA . Fits to a competition-binding isotherm ( lines ) yield the dissociation equilibrium constants ( ±S . E . ) shown in the inset . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 00510 . 7554/eLife . 01008 . 006Figure 2—figure supplement 2 . VCA dimers crosslinked through Cy3 stimulate Arp2/3 to a similar extent as GST-VCA dimers . ( A ) Schematics of N-WASP VVCA and VCA dimers made via fusion to GST ( as in Figure 2—figure supplement 1 ) or by covalent crosslinking of single-cysteine mutant subunits with a bis-maleimide Cy3 . The diXVCA material is similar to diVVCA but has the first V domain mutated at several sites to reduce binding to actin . ( B ) Sequence alignment of the diXVCA construct with wild-type N-WASP VVCA ( residues 393–505 of human N-WASP ) . White rectangle , C ( GGS ) 4 linker; X , V , C , and A domains are indicated in the same colors as in A . Cys residues are highlighted; the C431A mutation was made so that the XVCA peptide has only a single cys to react with the Cy3 bis-maleimide crosslinker . ( C and D ) Pyrene-actin fluorescence records ( as in Figure 2—figure supplement 1B ) indicating rates of actin filament polymerization in the presence of 10 nM yeast Arp2/3 complex and 2 μM rabbit muscle actin supplemented with 25 nM ( C ) or 200 nM ( D ) of the indicated VCA dimers . The records show that diXVCA is of somewhat higher nucleation activity ( maximum slope ) than diVCA , but the insertion of the X domain sequence does not increase the activity to the levels seen for diVVCA . Thus , the length and nature of the linkage between the two VCA segments in the dimer may have a subtle effect on the activity , but is not the origin of large difference in activity between diVCA and diVVCA . This reduced activity of the diVCA construct , together with the significant basal branch nucleation activity of S . cerevisiae Arp2/3 complex in the absence of WASP proteins ( Wen and Rubenstein , 2005; Smith et al . , 2013 ) likely contribute to the modest ( ∼twofold; see text ) activation of branch formation by di-VCA we observe . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 00610 . 7554/eLife . 01008 . 007Figure 2—figure supplement 3 . Arp2/3 complex and diVCA usually bind to and release from filaments as a unit when no daughter filament is formed . Left: Merged images ( Figure 2B ) and fluorescence intensity records ( Figure 2C ) showing an example of simultaneous binding ( t = 0 ) and release ( dashed vertical line ) of diVCA and Arp2/3 complex on an actin filament . Right: Three more examples . Experimental conditions were the same as in Figure 2 . Bar: 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 00710 . 7554/eLife . 01008 . 008Figure 2—figure supplement 4 . Association with Arp2/3 complex does not affect binding of VCA to actin . ( A ) Binding , measured using fluorescence anisotropy ( points ) , during titration of rabbit muscle actin into 20 nM VCA-AF488 . Fit to a binding isotherm ( line ) yielded KD 220 ± 10 ( 68% C . I . ) nM . ( B ) Competition binding experiment to determine the KD of diVCA for actin . Fluorescence anisotropy of 20 nM VCA-AF488 was monitored in the presence of 150 nM rabbit muscle actin , and the indicated concentration of diVCA . Fit to a competition binding isotherm ( line ) yielded KD 340 ± 60 nM ( see text ) . ( C ) Binding affinity of non-polymerizable actin ( ‘Materials and methods’ ) for VCA is minimally perturbed by the presence of yeast Arp2/3 complex . Fluorescence anisotropy of 10 nM of VCA-AF594 was measured in the presence of increasing concentrations of non-polymerizable actin ( red circles ) , yeast Arp2/3 complex ( blue squares ) , or non-polymerizable actin in the presence of 300 nM yeast Arp2/3 complex ( green inverted triangles ) . Fits to single site binding isotherms ( lines ) yielded KD 70 ± 10 nM , 65 ± 4 nM , and 105 nM ± 15 nM , respectively . Changes in anisotropy are expected to be dominated by the binding of actin , as the fluorophore location is proximal to the N-terminus of the V domain , and away from the Arp2/3 complex binding motifs . Increased fluorescence anisotropy in the presence of both Arp2/3 complex and actin indicates that both species can bind simultaneously . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 00810 . 7554/eLife . 01008 . 009Video 1 . The diVCA-stimulated actin branch formation event shown in Figure 2B–C . Red: Arp2/3-SNAP649 ( 5 nM in solution ) ; green: Cy3-diVCA ( 5 nM ) ; blue: actin-AF488 ( 1 µM , 10% labeled ) . Arp2/3 complex and diVCA images were recorded every 0 . 05 s; actin images were recorded every ∼12 s . Playback rate: real time . Bar: 1 μm . Many diVCA-Arp2/3 complexes are observed to transiently associate with actin filaments . One such nascent branch complex ( yellow arrowhead ) releases diVCA shortly after it appears , leaving Arp2/3 complex stably associated with the mother filament ( red arrowhead ) , where it subsequently initiates daughter filament elongation . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 009 Most nascent branches , although containing the necessary components for nucleation , dissociated quickly ( typically in <1 s ) without producing a daughter filament . This is consistent with our previous observations that the vast majority of Arp2/3 complex binding events are non-productive ( Smith et al . , 2013 ) . In nearly all cases ( 97 . 3 ± 0 . 4% , based on 758 nascent branch observations ) , diVCA and Arp2/3 complex dissociated simultaneously as a unit ( Figure 2—figure supplement 3 ) . The on-filament lifetime distributions of diVCA and Arp2/3 complex have identical short ( τ1 and τ2 ) components ( Figure 2D at time <1 s ) , consistent with the conclusion that the two molecules are released from the filament as an Arp2/3-diVCA complex . In contrast to the large majority of non-productive filament encounters , a small fraction of Arp2/3-diVCA filament binding events led to formation of a daughter filament . A merit of the single-molecule approach is that we could characterize these rare productive events ( e . g . , Figure 2B–C ) independently of the excess of non-productive events . In the productive events , Arp2/3 complex and diVCA release were not simultaneous . There was no evidence that the Arp2/3 complexes which formed branches ever dissociated; the value of τ3 is set by the photobleaching lifetime of the dye-labeled Arp2/3 complex ( Smith et al . , 2013 ) . In contrast , diVCA dissociated rapidly ( typically in <1 s ) from the productive Arp2/3-diVCA-filament complexes . Consistent with this observation , the on-filament lifetime distribution of diVCA lacks a long component ( τ3 ) that is present in the on-filament lifetime distribution of Arp2/3 complex ( Figure 2D at time > approximately 4 s ) . Thus , the data demonstrate that daughter nucleation is essentially always accompanied by Arp2/3 complex retention and diVCA release . To determine whether diVCA release occurs before or after the onset of daughter filament growth , we measured the time at which each daughter filament initiated elongation by extrapolating daughter length records ( as in Figure 2C , top ) . These filament initiation times were then compared to the times of diVCA release from the same nascent branch . Filament initiation time measurements were imprecise because of the uncertainties inherent in measuring daughter filaments of sub-micrometer lengths . Nevertheless , within this experimental uncertainty we observed that the initiation of daughter filament growth always occurred at or after the time of diVCA release ( 41 observations; Figure 2E ) . This was true even in the comparatively rare cases in which diVCA persisted on the nascent branch for times >1 s before dissociating . Taken together , these data suggest that the daughter filament cannot initiate unless and until VCA is released from the nascent branch . Thus , diVCA release may serve as the trigger for daughter growth . Next we asked whether diVCA can bind to Arp2/3 complex after branches have formed , in order to better understand how WASP recruits and activates free Arp2/3 complex yet does not stay bound to Arp2/3 complex in branch junctions and restrict network growth ( ‘Introduction’ ) . To address this question , we tethered individual dye- and biotin-labeled Arp2/3 complexes to the microscope slide and visualized the binding of freely diffusing diVCA and ( non-biotinylated ) actin filaments ( Figure 3A ) . Most of the tethered Arp2/3 complexes ( >80% ) were observed to bind diVCA . Binding lasted for tens or hundreds of seconds when no filaments were nearby ( e . g . , Figure 3B at time <0 ) . With 2 nM diVCA in solution , individual tethered Arp2/3 complexes were nearly continuously occupied , suggesting a dissociation equilibrium constant KD < 2 nM similar to bulk affinity measurements on other diVCA and Arp2/3 complex species ( Figure 2—figure supplement 1E ) . Further , we never observed the nucleation of a new actin filament from an isolated surface-tethered diVCA-Arp2/3 complex , consistent with previous conclusions that Arp2/3 complex cannot nucleate a daughter unless it is bound to a pre-existing mother filament ( Machesky et al . , 1999; Blanchoin et al . , 2000; Achard et al . , 2010 ) . 10 . 7554/eLife . 01008 . 010Figure 3 . VCA dimers form long-lived complexes with Arp2/3 complex before filament binding but not after branch formation . ( A ) Experimental design to observe diVCA binding and nucleation of actin filaments on immobilized Arp2/3 complexes . Arp2/3-SNAP was tethered to the slide surface via a bi-functional SNAP substrate that incorporated both a Dy649 dye and a biotin-terminated PEG chain; we monitored binding of fluorescently labeled diVCA and actin filaments from solution . ( B ) Example record showing the length of a nucleated daughter filament and the fluorescence intensity from actin ( blue ) and diVCA ( green ) at an individual tethered Arp2/3 complex molecule . The solution contained 1 μM actin ( 10% AF488 labeled ) and Cy3-diVCA ( 2 nM ) . Fluorescence from the tethered Arp2/3 complex ( red trace ) remained steady and above background ( red dash ) throughout . Schematics show the inferred complexes present at the indicated times . Time zero is the time of diVCA release . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 010 In some cases , we did observe association of a mother filament ( formed in solution ) with some tethered Arp2/3-diVCA complexes , followed by growth of a daughter filament . This yielded a branch junction that remained stably colocalized with the tethered Arp2/3 complex . In those events ( e . g . , Figure 3B at time >0 ) , branch formation essentially abolished binding of diVCA; any appearance of diVCA on branch junctions was transient ( <0 . 1 s ) and occurred at a low frequency ( ∼0 . 7 × 106 M−1 s−1; N = 2 ) comparable to non-specific binding at randomly chosen points on the microscope slide ( 1 . 0 ± 0 . 6 × 106 M−1 s−1 ) . The data demonstrate that the affinity of diVCA for isolated Arp2/3 complex is high , whereas the diVCA affinity for Arp2/3 complex in the branch junction is comparatively low . The low affinity of diVCA for the branch is consistent with data from previous studies ( Egile et al . , 2005; Martin et al . , 2006 ) . In addition , our measurements suggest that in the cell , once WASP proteins dissociate and the branch forms the Arp2/3 complex incorporated in the branch junction is unlikely to reassociate with membrane-linked WASP proteins and thus will not restrict filament network growth . To challenge the model that diVCA release from the nascent branch is required to initiate daughter filament growth , we next engineered a series of mutations in diVCA ( Figure 4A; Figure 4—figure supplement 1 ) . The goal was to modestly perturb VCA interactions with Arp2/3 complex or actin without altering the reaction pathway by which diVCA stimulates branch formation . Guided by previous biochemical data ( Zalevsky et al . , 2001; Panchal et al . , 2003; Chereau et al . , 2005 ) , we mutated each of the three regions of N-WASP VCA . The D435S/A436D mutation in the V-region ( diVCA-V* ) was designed to perturb actin affinity ( Chereau et al . , 2005 ) , whereas mutations in the C-region ( I467A , diVCA-C* ) and A-region ( Δ486–488 , diVCA-A* ) were designed to perturb interactions with Arp2/3 complex ( Zalevsky et al . , 2001; Panchal et al . , 2003 ) . 10 . 7554/eLife . 01008 . 011Figure 4 . diVCA mutations alter the stability of Arp2/3 complex-diVCA-actin monomer assemblies . ( A ) Arrangement of V , C , and A domains in native N-WASP and in the diVCA constructs used in this study ( w . t . is wild-type ) . Asterisks mark the domains bearing targeted mutations ( substitution of one or two residues , or a three-residue deletion; Figure 4—figure supplement 1A ) in the three mutant constructs . ( B ) Fluorescence anisotropy detected binding of AF488-labeled N-WASP VCA with rabbit muscle actin , in the presence of competitor wild-type ( same data as in Figure 2—figure supplement 4B ) or mutant Cy3-diVCA constructs ( symbols ) . Data were fit ( lines ) with competition binding isotherms incorporating the coupled equilibria ( ‘Materials and methods’ ) yielding KD values 340 ± 60 ( S . E . ) nM for wild-type diVCA , 660 ± 80 nM for diVCA-V* , 260 ± 40 nM for diVCA-C* , and 250 ± 40 nM for diVCA-A* . ( C ) Example Cy3-diVCA fluorescence intensity records recorded on individual tethered Arp2/3 complexes ( Figure 3 ) : Cy3-diVCA wild-type or C* mutant ( 0 . 5 nM ) molecules binding and dissociating in the presence of 1 μM actin monomers but no filament . ( D ) Cumulative lifetime distributions of diVCA-Arp2/3 complexes in the presence of monomeric actin observed in records like those in B . Smooth lines are biexponential fits ( Table 1 ) . Inset is a magnified view of the indicated data range . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 01110 . 7554/eLife . 01008 . 012Figure 4—figure supplement 1 . Design of the diVCA mutant constructs and characterization of Arp2/3 complex binding by the diVCA-C* mutant . ( A ) Sequence alignments showing the substitution ( colored residues ) and deletion ( dash ) mutations in the diVCA mutant constructs . Lines mark the regions of VCA that interact with G-actin and with Arp2/3 complex . The C ( GGS ) 4 linker and the cys residue used to react with Cy3 bis-maleimide are marked . ( B ) Binding of GST-VCA and GST-VCA-C* to bovine Arp2/3 complex assessed by a competition assay . Conditions and analysis as in Figure 2—figure supplement 1E; blue curve is repeated from that panel . The fits reveal KD values ( 68% C . I . ) indicating that GST-VCA-C* has a ∼fourfold weaker affinity for Arp2/3 complex than does GST-VCA ( inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 01210 . 7554/eLife . 01008 . 013Figure 4—figure supplement 2 . Single molecule analysis of wild-type and mutant diVCA binding to and dissociating from tethered Arp2/3 complex . ( A ) – ( C ) Binding rate measurements . ( A ) Example curves from single experiments showing the cumulative distribution of the lifetimes of Arp2/3 complexes unoccupied by diVCA constructs . diVCA constructs were present at 0 . 1 nM; other conditions as in Figure 4D . Fitting to a single exponential function ( smooth curves ) yielded binding rates . ( B ) Measured binding rates ( ±S . E . ) were proportional to diVCA-C* concentration . ( C ) Second order binding rate constants , kV+ ( ±S . E . ) , calculated from experiments like those in A ( Table 1 ) at diVCA construct concentrations in the range 0 . 1–1 . 0 nM . ( D ) The effect of photobleaching on the slow component ( τV , 2; Table 1 ) of the lifetime distributions for association of diVCA constructs with tethered Arp2/3 complex ( Figure 4D ) . Experiments were performed over a range of excitation green laser powers and the dependence of the observed dissociation rate ( 1/τV , 2; ±S . E . ) on power was globally fit to quantify the photobleaching rate ( slope ) and the photobleaching-corrected dissociation rates for each diVCA construct ( intercepts ) . The τV , 1 and τV , 2 values reported in Table 1 were taken from experiments at the weakest laser powers and were equal within experimental error to the values obtained after photobleaching correction . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 013 We measured the affinity of the mutants for monomeric actin using a fluorescence anisotropy competition assay . Here , labeled VCA reported on binding to monomeric actin by an increase in fluorescence anisotropy ( as in Figure 2—figure supplement 4A ) . As expected from the design , wild-type diVCA , diVCA-C* , and diVCA-A* had similar affinities for actin , while diVCA-V* bound more weakly ( ∼650 nM vs ∼300 nM for wild-type; Figure 4B ) . We next compared the rates of wild-type and mutant diVCAs binding to and dissociating from Arp2/3 complex using surface-tethered Arp2/3 complexes in the presence of actin monomers but not bound to actin filament . For wild-type diVCA and all three mutants , we observed repeated association and dissociation of the diVCA molecules with surface-tethered Arp2/3 complex ( e . g . , Figure 4C ) . These events allowed measurement of the lifetime distributions for the diVCA bound and dissociated states of Arp2/3 complex , and thus determination of the binding and dissociation rate constants . The mutations had only modest effects , at most 2 . 7-fold , on the rate of binding of diVCA to Arp2/3 complex ( Figure 4—figure supplement 2A–C; Table 1 ) . The dissociation rates varied over a wide range , with the A* mutant dissociating from Arp2/3 complex more slowly than wild-type , and the V* and C* mutants dissociating more rapidly ( Figure 4D ) . Both wild-type and mutant complexes displayed lifetime distributions that were fit well with two exponential components ( Figure 4D , Figure 4—figure supplement 2D , Table 1 ) , indicating that at least two distinct diVCA-Arp2/3 complex assemblies or conformations were present . In a more detailed analysis with the V* mutant , we saw no evidence that the short and long lifetime components segregated into different subpopulations of individual Arp2/3 complexes . These observations suggest even individual diVCA-Arp2/3 complexes participated in multiple states . The presence of multiple different complexes is consistent with previous observations including that Arp2/3 complex has two ( or more ) binding sites for VCA ( Padrick et al . , 2011; Ti et al . , 2011; Xu et al . , 2012 ) , and that alternative conformations of Arp2/3 complex ( Goley et al . , 2004; Rodal et al . , 2005 ) may have different affinities for VCA . 10 . 7554/eLife . 01008 . 014Table 1 . Colocalization kinetics and activities of diVCA and Arp2/3 complexDOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 014no VCAdiVCA w . t . diVCA V*diVCA C*diVCA A*Arp2/3 complex off filament ( Figure 4D , Figure 4—figure supplement 2 ) N ( groups ) 5331 kV+ ( 107 M−1 s−1 ) 16±5 ( SEM ) 7±3 ( SEM ) 6±1 ( SEM ) 16±1 ( SE ) NV152895713391262 τV1 ( s ) 8±13 . 6±0 . 42 . 4±0 . 140±10 AV124±3%42±2%98±2%28±9% τV2 ( s ) 61±356±314±6250±30Arp2/3 complex on filament ( Figure 6B ) NA7158774071089597 kA+ ( 104 M−1 s−1 ) 2 . 0±0 . 32 . 1±0 . 31 . 3±0 . 22±10 . 8±0 . 2 fAV0 . 83±0 . 090 . 7±0 . 10 . 59±0 . 080 . 7±0 . 1 fV−0 . 026±0 . 0040 . 041±0 . 0090 . 018±0 . 0030 . 015±0 . 004 fB0 . 006±0 . 0020 . 008±0 . 0020 . 029±0 . 0080 . 015±0 . 0030 . 013±0 . 003Arp2/3 complex at branch sites ( Figure 5A , Figure 6C–E ) NB57696940 τV* ( s ) 0 . 7±0 . 10 . 54±0 . 080 . 37±0 . 040 . 7±0 . 2 kV* ( s−1 ) 0 . 04±0 . 010 . 08±0 . 020 . 05±0 . 010 . 022±0 . 008 kB ( M−1 s−1 ) 120±40160±50320±90200±100100±30Parameter descriptions:N = number of groups of observations used to calculate binding rate of diVCA to isolated Arp2/3 complexes . kV+ = second order rate constant for diVCA binding to Arp2/3 complexes . NV = number of observations of diVCA on isolated Arp2/3 complexes . τV1 = first characteristic lifetime of diVCA on isolated Arp2/3 complexes . AV1 = percent of diVCA that dissociate from Arp2/3 complexes with time constant τV1 . τV2 = second characteristic lifetime of diVCA on isolated Arp2/3 complexes . NA = number of observations of Arp2/3 complexes on the sides of select filaments . kA+ = second order rate constant for Arp2/3 complex binding filament sides ( per filament subunit ) . fAV = fraction of Arp2/3 complexes that bind filament sides coincident with diVCA . fV- = fraction of diVCA-Arp2/3-filament complexes that release diVCA . fB = fraction of diVCA-Arp2/3-filament complexes that nucleate a daughter filament . NB = number of observations of branch formation from diVCA-Arp2/3-filament complexes . τV* = mean lifetime of diVCA on nascent branches . kV* = rate constant for diVCA release from the nascent branch . kB = second order rate constant for branch formation ( per mother filament subunit ) . From these experiments , we conclude that we have created a panel of mutants that modestly alter the association of diVCA with its binding partners . The V* and C* mutants , by disrupting interactions with Arp2/3 complex and monomeric actin , respectively , produce ternary diVCA-actin-Arp2/3 complexes that are less kinetically stable than wild-type . Conversely , the A* mutant produces a more stable ternary complex than wild-type . The diVCA mutants have distinct activities in stimulating branch nucleation by Arp2/3 complex . In bulk solution ( Figure 5—figure supplement 1 ) wild-type diVCA boosted Arp2/3 complex-dependent nucleation by 2 . 0 ± 0 . 5-fold ( p=0 . 017; measured by concentration of filament barbed ends at the midpoint of the reaction , Figure 5—figure supplement 1B ) . The V* and C* mutants were more potent than wild-type ( p=0 . 018 and 0 . 053 , respectively ) , stimulating nucleation 10 ± 3-fold and 6 ± 3-fold . Conversely , the A* mutant was less potent than wild-type ( p=0 . 040 ) , stimulating only 1 . 4 ± 0 . 4 fold . In all of these experiments , the diVCA construct was at 25 nM , a concentration that produces near-maximal stimulation ( Figure 5—figure supplement 2A ) . Similar effects of the mutant diVCA constructs were seen in real-time observations of individual filaments being nucleated ( Figure 5A ) . As expected , all three diVCAs were able to promote nucleation from the sides of existing filaments . More importantly , the mutant diVCA constructs shared with wild-type the key molecular behaviors discussed previously . Like wild-type , mutant diVCA constructs released from the nascent branch prior to initiation of the daughter filament ( Figure 5B; Figure 5—figure supplement 3 ) , and did not bind to Arp2/3 complex after the branch formed ( Figure 5C , D; this later behavior could not be verified for the A* mutant because of its low activity in the experiments using tethered Arp2/3 complexes ) . Overall , these results suggest that the mutant constructs stimulate Arp2/3 complex by the same mechanism as wild-type . Moreover , the rank order of the diVCA construct nucleation activities was identical in the single-molecule measurements of the rate of branch formation observed on existing filaments ( Figure 5A ) and in bulk measurements of concentration of filaments generated in solution ( Figure 5—figure supplement 1B ) . Thus , the mutants provide a range of activities both above and below wild-type that can be seen in both experimental modes , and the mechanism by which the mutants stimulate branch formation appears to be identical to wild-type . 10 . 7554/eLife . 01008 . 015Figure 5 . diVCA constructs differ in the rate but not the pathway of activity in stimulating branch formation . ( A ) Rate ( ±S . E . ) of initiation of daughter filament growth by Arp2/3 complex in the absence or presence of diVCA wild-type and mutant constructs . kB , the second order rate constant for the appearance of branches on existing filaments , per subunit , was calculated from observations of branch formation on existing filaments , as in Figure 2 ( ‘Materials and methods’ ) . ( B ) Comparison of the time ( ±S . E . ) of daughter filament initiation with the time of diVCA release from the nascent branch for wild-type ( data replotted from Figure 2E ) and mutant constructs ( Figure 5—figure supplement 3 ) . ( C and D ) Example records showing the length of a nucleated daughter filament and the fluorescence intensity from actin , diVCA , and individual tethered Arp2/3 complex molecules , as in Figure 3 . Mutant Cy3-diVCA was 0 . 5 nM V* in C , or 1 . 0 nM C* in D . The merged fluorescence images in C were recorded at the indicated times and the white squares mark the area from which the fluorescence was integrated to produce the intensity records . Scale bar , 1 µm . Both mutants bound readily to tethered Arp2/3 prior to but not after branch formation . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 01510 . 7554/eLife . 01008 . 016Figure 5—figure supplement 1 . Analysis of diVCA mutant activities in bulk actin polymerization assays . ( A ) Stimulation of actin filament assembly by 10 nM Arp2/3-SNAP complex activated by 25 nM wild-type or mutant diVCA constructs . Plots show records of pyrene fluorescence in assays containing 2 µM actin , 5% pyrene-labeled . ( B ) Actin nucleation activities from the data in A and a replicate experiment ( mean ± S . D . ; ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 01610 . 7554/eLife . 01008 . 017Figure 5—figure supplement 2 . Saturation of stimulation of Arp2/3 complex actin nucleation activity by diVCA constructs . ( A ) Actin nucleation activity measured as in Figure 5—figure supplement 1 with 10 nM yeast Arp2/3 complex and specified diVCA construct concentrations . Activity saturates above 50 nM for three constructs; for diVCA-A* activity is too low to detect saturation in this assay . ( B ) Saturation of diVCA and diVCA-A* in competing with stimulation of actin nucleation by VVCA . Mean values from 3–4 replicate measurements . Error bars indicate S . E . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 01710 . 7554/eLife . 01008 . 018Figure 5—figure supplement 3 . Relationship relation between diVCA release from the nascent branch and daughter filament initiation for each of the three mutant diVCA constructs . These panels show the data presented in Figure 5B replotted separately to make the error bars more visible . V* , C* , and A* data sets contain 41 , 49 , and 27 observations , respectively , where the measured daughter initiation time was >0 . 1 s . Asterisks mark two observations in which daughter filament initiation occurred significantly before disappearance of the diVCA fluorescent spot . We speculate that in these rare events the diVCA molecule may have become irreversibly crosslinked at or near the branch junction and that spot disappearance is caused by dye photobleaching , not by diVCA release . Consistent with this explanation , both of these outliers show uncharacteristically long diVCA release times . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 018 Since high diVCA activity appears to correspond with weak binding to Arp2/3 complex , we next tested the hypotheses that the rate of diVCA release from the nascent branch quantitatively explains the nucleation activities of different mutants . While we already measured the dissociation rates of the diVCA constructs from Arp2/3 complex when the latter is not bound to the side of a mother filament ( Figure 4D; Table 1 ) , the rates of dissociation from the filament-bound Arp2/3 complex ( the nascent branch ) might be different . Our observations ( Figure 2 ) showed that once a nascent branch forms , it can have multiple fates ( Figure 6A ) . Most often , the Arp2/3-diVCA complex simply dissociates from the filament ( Figure 6A , thick red arrow ) . In a small fraction of the nascent branches , fV− , diVCA departs first , leaving behind Arp2/3 complex bound to mother filament ( Figure 6A , activated complex ) where it may subsequently initiate a daughter filament . Furthermore , only a fraction of activated complexes subsequently formed a branch , so that the fraction of nascent branches that successfully produced a daughter filament , fB , is less than fV− . For some of the mutant constructs fV− or fB values ( Figure 6B ) could not be unambiguously distinguished from wild type in pairwise comparisons ( p=0 . 05–0 . 15 ) ; others differed significantly ( p=0 . 005–0 . 05 ) from wild-type ( asterisks in Figure 6B ) . 10 . 7554/eLife . 01008 . 019Figure 6 . Release of diVCA from nascent branches is rare and limits the rate of daughter nucleation . ( A ) Schematic mechanism of diVCA stimulated branch formation ( see text ) . The key activation step , release of diVCA from the nascent branch , is highlighted . ( B ) Classification of nascent branch fates observed in single molecule experiments ( e . g . , Figure 2B and C; Figure 2—figure supplement 3 ) . Overall bar height indicates the fraction ( ±S . E . ) of nascent branches that release diVCA leaving behind a filament-bound Arp2/3 complex . Filled bar height shows the fraction ( ±S . E . ) of nascent branches that nucleate a daughter filament . ( C ) Cumulative lifetime distributions of diVCA molecules on the subset of filament-bound Arp2/3 complexes observed to produce branches in single-molecule experiments . Inset: mean lifetimes ( ±S . E . ) . ( D ) Rate constants ( ±S . E . ) for diVCA dissociation from the nascent branch , calculated from the mean lifetimes in C and release efficiencies in B . ( E ) Correlation between the rate constant of diVCA-stimulated Arp2/3 complex branch nucleation ( from Figure 5A ) and the rate constant of diVCA release from the nascent branch ( from D ) . Correlation coefficient r = 0 . 9928 is unlikely to arise by coincidence ( p=0 . 0045 ) . Dotted line is a linear fit constrained to pass through the origin . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 01910 . 7554/eLife . 01008 . 020Figure 6—figure supplement 1 . Correlation between the actin nucleation activity of Arp2/3 activated by wild-type and mutant diVCA constructs ( from Figure 5—figure supplement 1B ) and the rate constant of diVCA release from the nascent branch ( from Figure 4F ) . Dotted line is a linear fit constrained to pass through the origin . The correlation coefficient r = 0 . 9598 is unlikely to be coincidental ( p=0 . 027; ‘Materials and methods’ ) . Error bars indicate S . E . s . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 020 We also measured the lifetimes of diVCA constructs on the nascent branches that ultimately produced daughter filaments . The mean lifetime ( τV* , equal to the reciprocal of the sum of the rate constants for diVCA departure and Arp2/3 complex dissociation; ‘Materials and methods’ ) differed little between the different diVCA constructs ( Figure 6C ) . Only the C* mutant had a significantly shorter mean lifetime on the nascent branch than wild-type diVCA ( 2 ± 1-fold , p=0 . 015 ) , whereas the V* and A* mutants did not ( p=0 . 17 and 0 . 48 , respectively ) . This observation suggests ( ‘Materials and methods’ ) that the mutations do not substantially alter the rate of the main pathway of nascent complex breakdown , which is the dissociation of the intact Arp2/3-diVCA complex from filament sides ( Figure 6A , thick red arrow ) . Instead , the mutations principally affect the fraction of nascent complexes that release diVCA , and are productive in forming branches . Based on the measurements of Figure 6B , C , we can calculate the rate constant for dissociation of wild-type and mutant diVCAs from the nascent branch ( Figure 6A , kV−* ) as fV−/τV* ( Figure 6D ) . Strikingly , this rate of release of diVCA from the nascent branch is proportional , within experimental uncertainty ( Figure 6E ) , to the branch formation rate constant kB measured above ( Figure 5A; Figure 6—figure supplement 1 ) . Despite the relatively subtle effects of the mutants on the measured rate constants , the high correlation coefficient ( r = 0 . 993 ) is unlikely to be coincidental ( p=0 . 0045; ‘Materials and methods’ ) . This relationship provides strong support for the hypotheses that diVCA release is a prerequisite for daughter filament formation , and that the rate of release of diVCA limits the rate at which nascent branches initiate daughter filament growth .
By placing three different colors of fluorescent labels on Arp2/3 complex , diVCA , and actin , we directly observed the sequence and kinetics of key steps in the branch nucleation pathway . The observations confirm our previous results ( Smith et al . , 2013 ) that branch nucleation is inefficient even at near-saturating VCA protein concentrations , with only a small fraction of Arp2/3 complex-mother filament associations yielding branches . In contrast to the tight binding of diVCA to Arp2/3 complex in solution , we show that diVCA binding is undetectable when Arp2/3 complex is incorporated into a branch junction . Significantly , we observed that the branch formation process is strictly dependent on release of diVCA from the filament-bound Arp2/3-diVCA complex . Taken together with earlier work , the data support a specific mechanism for diVCA-stimulated actin nucleation by Arp2/3 complex in which diVCA plays a dual role ( Figure 6A ) . Initially , diVCA stimulates assembly of the nascent branch by associating tightly with Arp2/3 complex and actin monomers , which promotes the binding of Arp2/3 complex to the sides of filaments . However , once the nascent branch forms , diVCA plays an inhibitory role: diVCA must dissociate before daughter filament can grow . In mutants that alter the interactions of diVCA with Arp2/3 complex or monomeric actin , alterations of the rate at which diVCA leaves the nascent branch exactly parallel changes in the efficiency of branch formation . These observations strongly suggest that diVCA dissociation is the key , rate-limiting step in daughter filament nucleation . The key mechanistic conclusions reached here with dimerized VCA constructs in vitro are likely to apply to the mechanism of activation of branch nucleation by native WASP oligomers in vivo . Data on dimeric VCA showing a minimum crosslinker length for high activity ( Padrick et al . , 2011 ) strongly suggests that the system passes through a form with two VCAs bound at some point during nucleation . The crosslinkers and spacer sequences used here allow a maximum VCA separation of 122 Å ( Padrick et al . , 2011 ) . During Arp2/3 complex activation in diverse contexts in living cells , WASP family proteins are separated by similar or even smaller distances . For example , when dimerized by activators such as EspFu , the WASP protein VCA regions would be ∼80 Å apart if the proline rich region is considered as a random coil polymer ( Padrick et al . , 2008 ) . Similarly , ActA density on the surface of Listeria separates its VCA-like sequences by 19 nm , close enough to function as VCA dimers ( Footer et al . , 2008 ) . N-WASP proteins recruited to moving PIP2-rich vesicles have a high local density , such that their average spacing is <50 Å ( Co et al . , 2007 ) , and in rocketing vesicles , N-WASP proteins are recruited by Nck onto the vesicle surface such that their average separation is ∼100 Å ( Ditlev et al . , 2012 ) . Bzz1 and Cdc15 are the activators of the fission yeast WASP family protein Wsp1 during endocytosis , and they array their SH3 domains ( which engage Wsp1 ) at a density sufficient to bring VCA regions from adjacent Wsp1 proteins within 50–80 Å ( Arasada and Pollard , 2011 ) . The reaction scheme in Figure 6A describes the essential features of the Arp2/3 complex-dependent filament nucleation process . This mechanism , derived from direct observation of single molecules as opposed to fitting of bulk data , is broadly consistent with proposed kinetic schemes for Arp2/3 complex nucleation of filaments ( Zalevsky et al . , 2001; Beltzner and Pollard , 2008; Smith et al . , 2013 ) . The dominant pathway is that VCA and actin monomer associate with Arp2/3 complex in solution , and this complex then binds to an existing filament , after which an activation step occurs , which allows the daughter filament to elongate . Here we add two informative points . First , we ascribe a distinct mechanism to the activation step , the release of diVCA . Previously this step had been ascribed to structural rearrangements within the Arp2/3 complex ( Dayel et al . , 2001; Zalevsky et al . , 2001; Beltzner and Pollard , 2008 ) . Second , our measurements indicate that most engagements of diVCA-Arp2/3 complexes with the mother filament are resolved by dissociation of the complex from filament without releasing VCA , an idea distinct from previous models ( Beltzner and Pollard , 2008 ) . It is possible that the observed inefficiency in branch formation is to allow for positive regulation of branch formation by factors not present in our experiment ( Smith et al . , 2013 ) . Consistent with this idea , cortactin has been recently demonstrated to accelerate release of WASP proteins from Arp2/3 complex ( Helgeson and Nolen , 2013 ) . Since WASP proteins are tethered to the cell membrane , it is also possible that mechanical tension between the filament network and the membrane plays a similar role in promoting WASP release and consequent daughter growth , which might allow alteration of cell motility in response to mechanical stimuli . While our scheme encompasses key features of the mechanism of diVCA stimulation of Arp2/3 complex-mediated branch formation , it should be noted that the scheme shown in Figure 6A is not complete . The clearest indication of this is that the diVCA lifetime distribution observed in Figure 2D is multi-exponential , whereas the scheme of Figure 6A predicts only a simple single-exponential distribution . The data can be explained if there are two or more conformations of filament-bound diVCA-Arp2/3 complexes that differ in kinetic stability . This proposal is consistent with our previous kinetic analysis with monomeric VCA ( Smith et al . , 2013 ) and with demonstrations that Arp2/3 complex exists in multiple conformations in solution ( Goley et al . , 2004; Rodal et al . , 2005; Xu et al . , 2012 ) . Formulating a more complete kinetic mechanism of diVCA stimulation that accounts for the multiple diVCA-Arp2/3 complex conformations will require additional data . Our model in which association of VCA with the nascent branch inhibits initiation of daughter filament growth is consistent with the known interaction of the WH2 motif ( V-region ) with actin in a conserved cleft involved in longitudinal filament contacts ( Hertzog et al . , 2004; Irobi et al . , 2004; Chereau et al . , 2005 ) . We suspect that inhibition of daughter filament elongation from nascent branches results from the WH2 motif staying engaged with the cleft , occluding addition of the next actin subunit in the daughter filament ( Egile et al . , 1999; Higgs et al . , 1999 ) . Further support for this model is realized by recent reports showing that VCA peptides covalently crosslinked to actin monomers are inactive in stimulating Arp2/3 complex-mediated actin nucleation ( Boczkowska et al . , 2008; Ti et al . , 2011 ) . Moreover , the C-region of VCA likely occupies a similar location on Arp2 and Arp3 that the V-region occupies on actin , preventing insertion of subdomain 2 of the actin recruited by the V-region into the cleft on the bound Arp subunit ( Hertzog et al . , 2004; Irobi et al . , 2004; Chereau et al . , 2005; Boczkowska et al . , 2008; Padrick et al . , 2011; Ti et al . , 2011 ) . Thus , the molecular structures are consistent with the observed dual function of VCA in both stimulating daughter nucleation ( by recruiting actin monomers to Arp2/3 complex ) and in suppressing daughter nucleation ( by blocking assembly of the daughter filament ) . These structural models are also consistent with the low affinity of diVCA for branch junctions . If neither the V-region nor the C-region can bind the branch ( when the actin and Arp clefts are occupied by D-loops of daughter filament actins ) , then the affinity of VCA for the branch may be of the order of that observed for isolated A-region . This affinity has been reported to be approximately 9 μM ( Marchand et al . , 2001 ) , consistent with the lack of observed binding under the conditions necessary for single molecule observations . In cells , active WASP proteins are predominantly tethered to membranes . Taken together , our data suggests a straightforward mechanism by which WASP regulation of Arp2/3 complex can cause branches to form preferentially at membranes without having membrane attachment restrict network growth ( Figure 7 ) . Binding of Arp2/3 complex to VCA dimers in the absence of mother filaments is of high affinity and of long lifetime , allowing dimerization of WASP by upstream activators to promote recruitment of Arp2/3 complex to the membrane surface ( Figure 7 , ‘WASP dimer association with Arp2/3 complex’ ) . Association with VCA promotes filament binding by Arp2/3 complex ( ‘Arp2/3 complex engagement of filament’ ) , but that process is readily reversible and the nascent branch most frequently is simply lost through dissociation ( ‘Filament release’ ) . More rarely , WASP detaches ( ‘WASP release’ ) leaving Arp2/3 complex associated with the mother filament . This severs the direct linkage to the membrane , and only then allows the daughter filament to nucleate and grow ( ‘Initiation of daughter filament elongation’ ) . Our data show that once the branch forms , WASP does not rebind . The dissociation of WASP prior to nucleation and the lack of rebinding provide an appealing explanation for how WASP dimers stimulate branched network formation at the membrane without interfering with network growth . 10 . 7554/eLife . 01008 . 021Figure 7 . Model of WASP-Arp2/3 complex stimulated actin branch formation at cell membranes ( see text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01008 . 021
The bis-maleimide crosslinked N-WASP VCA dimers were prepared from human N-WASP VCA ( amino acid residues 432–505 ) , with the sequence CGGSGGSGGSGGS appended at the N-terminus . This sequence was generated by PCR using overlapping primers encoding the N-terminal extension . The resulting PCR product was cloned into a plasmid derived from pGEX2T with a TEV protease site between the sequence for GST and the multiple cloning site ( which was modified to include a 5′ NdeI site and 3′ BamHI site ) . Mutations were introduced by designing overlapping primers containing the desired mutations , which were used in PCR amplification of the region 5′ to the modification and 3′ to the modification . These products were used as template in a second round of PCR that fused them into a single cassette containing the desired mutation . The resulting product was digested with NdeI and BamHI and ligated into the above described expression vector . Cloned products were verified by DNA sequencing . Crosslinked VCA dimers were prepared using a common strategy . Proteins were expressed in Escherichia coli BL21 ( DE3 ) T1R cells using IPTG induction at 37°C for 3 hr . Cells were harvested , resuspended in 25 ml of buffer ( 20 mM Tris pH 8 . 0 , 50 mM NaCl , 2 mM EDTA , 1 mM DTT and 1 mM PMSF ) per l of culture , and frozen at −80°C until needed . Cells were lysed using a cell disruptor and clarified by centrifugation at 19 , 500 rpm in a JA25 . 50 rotor ( Beckman Coulter Inc . , Brea , CA ) . Clarified lysate was purified using DEAE Sepharose FF ( GE Healthcare Biosciences , Pittsburgh , PA ) ion exchange chromatography , followed by Glutathione Sepharose ( GE Healthcare Biosciences ) chromatography , followed by cleavage with TEV protease . Cleaved VCA peptides were purified by SOURCE 15Q ion exchange chromatography . Next , the pooled VCA was concentrated and the DTT removed , by dilution and passing over a 0 . 5 ml SOURCE 15Q column in buffer lacking reducing agent . BMCy3 ( #C959070 , Toronto Research Chemicals , Toronto , Canada ) was prepared as a 20 mM stock in anhydrous DMSO . VCA materials were quantified by absorption at 280 nm , and BMCy3 was added to a final concentration of one equivalent ( about 80 μM depending on the preparation ) , split over three additions separated by 10 min incubations at room temperature . After a final 20 min incubation at room temperature , the reaction was quenched with 2 mM DTT , and the reaction was purified using SOURCE 15 Q ion exchange chromatography , followed by Superdex 200 gel filtration chromatography ( GE Healthcare Biosciences ) . Purification was tracked using SDS-PAGE analysis , and concentrations were determined from absorption at 552 nm , using an extinction coefficient of 150 , 000 M−1 cm−1 . GST-VCA and GST-VVCA were produced from vectors lacking the CGGSGGSGGSGGS N-terminal extension . Expression and purification followed a similar protocol as for the crosslinked VCA dimers , with the modification that GST was not cleaved from the product following Glutathione Sepharose purification . Purification continued with SOURCE 15Q ion exchange chromatography and Superdex 200 gel filtration chromatography . Concentrations were measured using absorbance at 280 nm . The fluorescence anisotropy probe for actin binding , VCA-AF488 , was produced from a similar vector encoding N-WASP VCA amino acid residues 430–505 ( with the introduced mutations S430C and C431A , and lacking the CGGSGGSGGSGGS extension ) . Expression and purification was the same as for the VCA dimers , through the SOURCE 15Q step in buffer without reducing agent . At this point , 1 ml of 40 μM VCA was labeled with 150 μM AlexaFluor488 maleimide ( #A-10254 , Molecular Probes/Life Technologies , Grand Island , NY , 40 mM stock in anhydrous DMSO ) . After 2 hr reaction at room temperature , the reaction was quenched with 2 mM DTT and VCA-AF488 was purified by SOURCE15Q and Superdex 75 chromatography . Labeling efficiency was judged to be nearly 100% from a shift in mobility by SDS-PAGE ( small shift judged using a reference of flanking unlabeled material ) . Concentration was measured as the absorbance at 492 nm , using an extinction coefficient of 71 , 000 M−1 cm−1 . The fluorescence anisotropy probe for Arp2/3 complex binding , VVCA-A462C-A594 , was produced by using a similar method to VCA-AF488 with two modifications . First , the vector used encoded N-WASP VCA amino acid residues 393–505 , with two mutations , C431A and A462C . Second , the VCA was labeled with AlexaFluor 594 maleimide ( #A-10256 , Molecular Probes/Life Technologies , 40 mM stock in anhydrous DMSO ) , instead of AlexFluor 488 maleimide . Quantification was performed using absorbance at 588 nm and an extinction coefficient of 96 , 000 M−1 cm−1 . The fluorescence anisotropy probe used for examining simultaneous binding of yeast Arp2/3 complex and Drosophila 5C actin , VCA-AF594 , was produced by using a similar method to VCA-AF488 , but substituting the AlexaFluor 594 maleimide ( mentioned above ) for AlexaFluor 488 maleimide . S . cerevisiae Arp2/3 complex ( used in Figure 2—figure supplements 2 and 4 , Figure 5—figure supplement 2 ) was purified from commercial baker’s yeast ( #05020 , Red Star Yeast Company , Milwaukee , WI ) using a method adapted from published protocols ( Egile et al . , 1999; Lechler et al . , 2001 ) with added SOURCE 15Q and Superdex 200 chromatography steps ( Doolittle et al . , 2013b ) . Endogenous bovine Arp2/3 complex ( used in Figure 2—figure supplement 1 , Figure 4—figure supplement 1B ) was purified from calf thymus using previously described methods ( Higgs et al . , 1999; Doolittle et al . , 2013c ) . SNAP-tagged Arp2/3 complex was purified from recombinant S . cerevisiae and labeled as previously described ( Smith et al . , 2013 ) , except that labeling used BG-649 and BG-649-PEG-biotin ( ‘Synthesis of SNAP-tag substrates BG-649 and BG-649-PEG-biotin’ below ) to yield Arp2/3-SNAP649 and Arp2/3-SNAP649-biotin . Rabbit muscle actin , pyrene-labeled actin , AlexaFluor488-labeled actin , and biotinylated actin were purified as described ( Spudich and Watt , 1971; Smith et al . , 2013 ) . Non-polymerizable Drosophila melanogaster 5C actin was prepared according to established methods ( Joel et al . , 2004 ) , but with the mutation D287A/V288A/D289A instead of A204E/P243K . Characterization of this mutation will be described elsewhere ( Zahm et al . , 2013 ) . In the syntheses , commercially available compounds were used without further purification and reaction yields are not optimized . Reversed-phase high-performance liquid chromatography ( HPLC ) was performed on Agilent LC/MS Single Quad System 1200 Series ( analytical ) and Agilent 1100 Preparative-scale Purification System ( semi-preparative ) . Analytical HPLC was performed on Waters Atlantis T3 C18 column ( 2 . 1 × 150 mm , 5 µm particle size ) at a flow rate of 0 . 5 ml/min with a binary gradient from Phase A ( 0 . 1 M triethyl ammonium bicarbonate [TEAB] or 0 . 1% trifluoroacetic acid [TFA] in water ) to Phase B ( acetonitrile ) and monitored by absorbance at 280 nm . Semi-preparative HPLC was performed on VYDAC 218 TP series C18 polymeric reversed-phase column ( 22 × 250 mm , 10 µm particle size ) at a flow rate of 20 ml/min . Mass spectra were recorded by electrospray ionization ( ESI ) on an Agilent 6210 Time-of-Flight ( TOF ) or 6120 Quadrupole LC/MS system . BG-649 was prepared by reacting the building block BG-NH2 ( New England Biolabs , Ipswich , MA ) with the dye N-hydroxysuccinimide ester DY-649 NHS ( Dyomics GmbH , Jena , Germany ) as described previously ( Keppler et al . , 2004 ) . BG-NH2 ( 0 . 54 mg , 2 . 0 µmol ) was dissolved in anhydrous DMF ( 0 . 5 ml ) . DY-649 NHS ( 2 . 0 mg , 2 . 0 µmol ) and triethylamine ( 0 . 4 µl , 3 . 0 µmol ) were added and the reaction mixture stirred overnight at room temperature . The solvent was removed under vacuum and the product purified by reversed-phase HPLC using 0 . 1 M TEAB/acetonitrile gradient . Yield: 74% . BG-649: ESI-MS m/z 1095 . 2 [M-H]− ( calculated for C48H56N8O14S4 , m/z 1095 . 3 ) . The bifunctional BG-649-PEG-Biotin ( which includes both a DY-649 dye and a biotin moiety ) was prepared by successive couplings of commercially available α-N-Fmoc-ε-N-Dde-lysine ( Merck KGaA , Darmstadt , Germany ) with BG-NH2 ( New England Biolabs , Ipswich , MA ) , N- ( + ) -biotin-6-aminocaproic acid N-succinimidyl ester ( Sigma-Aldrich , St . Louis , MO ) and DY-649 NHS ( Dyomics GmbH , Jena , Germany ) according to synthetic route described previously ( Kindermann et al . , 2004 ) . BG-649-PEG-Biotin was synthesized as follows: BG-NH2 ( 250 . 0 mg , 0 . 92 mmol ) was dissolved in anhydrous DMF ( 8 ml ) . HBTU ( N , N , N′ , N′-Tetramethyl-O- ( 1H-benzotriazol-1-yl ) uronium hexafluorophosphate ) ( 368 . 0 mg , 0 . 97 mmol ) , triethylamine ( 135 µl , 0 . 97 mmol ) , and Fmoc-Lys ( Dde ) -OH ( 515 . 5 mg , 0 . 97 mmol ) were added and the reaction mixture stirred overnight at room temperature . The reaction mixture was poured onto water ( 80 ml ) . The white solid was collected by filtration , washed twice with water , and dried in dessicator under vacuum overnight . Yield: 91% . BG-Lys ( Dde ) -Fmoc ( 50 mg , 63 . 7 µmol ) was dissolved in anhydrous in DMF ( 5 ml ) . Et2NH ( 19 . 8 µl , 191 . 1 µmol ) was added and the reaction mixture stirred overnight at room temperature . The solvent was removed under vacuum for 6 hr and the residue dissolved in DMF ( 3 ml ) . Fmoc-12-amino-4 , 7 , 10-trioxadodecanoic acid ( 29 . 7 mg , 66 . 9 µmol ) , triethylamine ( 26 . 6 µl , 191 . 1 µmol ) and HBTU ( 36 . 3 mg , 95 . 6 µmol ) were added and the reaction mixture stirred for 1 hr at room temperature . The reaction completion was monitored by LC/MS . The solvent was removed under vacuum and the product purified by reversed-phase HPLC using 0 . 1 M TEAB/acetonitrile gradient . Yield: 50% . BG-Lys ( Dde ) -PEG-NHFmoc: ESI-MS m/z 988 . 4 [M+H]+ ( calculated for C53H65N9O10 , m/z 988 . 5 ) . BG-Lys ( Dde ) -PEG-NHFmoc ( 31 . 6 mg , 31 . 9 µmol ) was dissolved in anhydrous DMF ( 2 ml ) . Et2NH ( 9 . 9 µl , 95 . 7 µmol ) was added and the reaction mixture stirred overnight at room temperature . The solvent was removed under vacuum for 6 hr and the residue dissolved in DMF ( 2 ml ) . N- ( + ) -biotin-6-aminocaproic acid NHS ( 14 . 5 mg , 31 . 9 µmol ) and triethylamine ( 13 . 3 µl , 95 . 7 mmol ) were added and the reaction mixture stirred overnight at room temperature . The reaction completion was monitored by LC/MS . A 2% solution of hydrazine in DMF ( 0 . 5 ml ) was added and the reaction mixture stirred for 1 hr at room temperature . The solvent was removed under vacuum and the product purified by reversed-phase HPLC using 0 . 1% TFA in water/acetonitrile gradient . Yield: 75% . BG-Lys ( NH2 ) -PEG-Biotin: ESI-TOFMS m/z 939 . 4873 [M-H]− ( calculated for C44H68N12O9S , m/z 939 . 4880 ) . BG-Lys ( NH2 ) -PEG-Biotin ( 2 . 3 mg , 2 . 13 µmol ) was dissolved in anhydrous DMF ( 1 ml ) . DY-649 NHS ( 2 . 1 mg , 2 . 13 µmol ) and triethylamine ( 0 . 45 µl , 3 . 2 µmol ) were added and the reaction mixture stirred overnight at room temperature . The solvent was removed under vacuum and the product purified by reversed-phase HPLC using 0 . 1 M TEAB/acetonitrile gradient . Yield: 73% . BG-649-PEG-Biotin: ESI-TOFMS m/z 882 . 3166 [M-2H]2− ( calculated for C79H110N14O22S5 , m/z 882 . 3188 ) . Actin assembly kinetics measurements were performed in a fashion similar to previously published ( Cooper et al . , 1983; D’Agostino and Goode , 2005; Padrick et al . , 2008 , 2011; Doolittle et al . , 2013a ) . Briefly , a rabbit muscle actin stock ( 5% pyrene labeled in Figure 5—figure supplement 1 or 10% pyrene labeled in Figure 2—figure supplement 1 and 2 , and Figure 5—figure supplement 2 ) in buffer G ( 2 mM Tris pH 8 . 0 , 200 μM CaCl2 , 1 mM NaN3 , 100 μM ATP , 0 . 5 mM DTT ) was combined with 1/10th volume 10 mM EGTA , 1 mM MgCl2 , and then with enough buffer G-Mg ( same as buffer G but substituting MgCl2 for CaCl2 ) to dilute the overall actin concentration to 4 μM . After a 2 min incubation in this buffer , the actin solution was combined with an equal volume of Arp2/3 complex and VCA materials in double strength KMEI buffer ( such that the final solution was 10 mM imidazole pH 7 . 0 , 50 mM KCl , 1 mM EGTA , 1 mM MgCl2 , 0 . 5 mM DTT , with one half concentration of buffer G carrying over from the actin solution ) . For experiments designed to enable quantitative comparison of actin assembly rates ( Figure 5—figure supplement 1 ) to results from single molecule analysis ( Figure 6D , Figure 6—figure supplement 1 ) the buffer was supplemented with additional components ( final concentrations: 10 mM DTT , 0 . 2 mM ATP , 15 mM glucose , 0 . 02 mg/ml catalase , 0 . 1 mg/ml glucose oxidase , 0 . 1% bovine serum albumin , 1 mM 6-hydroxy-2 , 5 , 7 , 8-tetramethylchroman-2-carboxylic acid ( Trolox ) , 1 mM 4-nitrobenzyl alcohol , and 0 . 5 mM propyl gallate ) so as to maintain similar buffer conditions in the two experiments . Reactions were then immediately placed in a cuvette in a PTI Quantamaster spectrofluorometer . Pyrene-actin fluorescence was observed over time by exciting at 365 nm and observing at 407 nm . For some assays ( Figure 2—figure supplement 1 and 2 , Figure 5—figure supplement 2 ) , 10% pyrene labeling was used , and the reactions were placed into 96-well plates and followed using a plate reader ( VarioSkan Flash , Thermo Scientific , Hudson , NH ) . Actin filament barbed end concentrations were evaluated at 50% polymerization ( t50 ) by first scaling the pyrene fluorescence intensity over the full range of filament concentrations ( 0–1 . 9 µM ) , then fitting the slope ( actin assembly rate ) between 42% and 58% polymerization . These rates were then divided by the approximate filament elongation rate ( 10 subunits per second ) to obtain the barbed end concentrations ( Figure 5—figure supplement 1B ) . Fold stimulation of branch formation by diVCA was calculated by subtracting the barbed ends created by actin alone from the total barbed ends formed in the presence of Arp2/3 complex , then dividing the concentrations of these excess barbed ends formed in the presence diVCA by those formed in the absence of diVCA . Binding of VCA to actin was monitored by fluorescence anisotropy ( Figure 2—figure supplement 4 , Figure 4B ) . 20 nM N-WASP VCA-AF488 was mixed with the indicated concentrations of actin and Cy3-diVCA competitor with buffer additions to bring the final mixture to 10 mM imidazole pH 7 . 0 , 50 mM KCl , 1 mM EGTA , 1 mM MgCl2 , 0 . 5 mM DTT , 0 . 1 mM ATP , and 1/10th residual concentration of buffer G . The mixtures were incubated for 3 min at room temperature prior to placing in a 3 mm by 3 mm cuvette in a T-form PTI Quantamaster Spectrafluorometer , equipped with Glan-Thompson polarizers . Emission intensity was averaged for 3 min and converted to anisotropy values , after correcting for background signal intensity and G-factor . Competition binding experiments ( including N-WASP diVCA competition ) were performed with 20 nM N-WASP VCA-AF488 , 200 nM rabbit muscle actin , and the indicated concentrations of VCA dimers . Binding isotherms ( both direct and competition binding ) were fit to a complete competition binding solution for a single site receptor , using Levenberg-Marquardt nonlinear least squares methods , with bound and free fluorescence anisotropy as fit parameters . Direct binding isotherms were fit with the concentration of competitor ligand set to zero . Fitting of the competition-binding isotherm used the direct binding KD that was obtained from fitting the direct binding isotherm . Fit values for free fluorescence anisotropy values were similar to the actual free anisotropy , and the values of free and bound fluorescence anisotropy determined from competition binding experiments was similar to that of the direct binding system . Binding of VCA to bovine Arp2/3 complex in solution ( Figure 2—figure supplement 1D–E , Figure 4—figure supplement 1B ) was monitored by fluorescence anisotropy using the same basic protocol as for actin binding , with a few modifications . First , the reporter was 20 nM VVCA-A462C-AF594 . Next , as there was no actin present , there was no residual buffer G in the mixture . Finally , the mixture was incubated for 10 min prior to measurement , and the fluorescence intensity data averaged for 5 min . Solution binding of VCA to non-polymerizable Drosophila 5C actin , and to yeast Arp2/3 complex , was followed by fluorescence anisotropy . Acquisition and processing was similar to the actin binding assay described above , except 10 nM VCA-A594 was used , and either Drosophila 5C actin was added or endogenous budding yeast Arp2/3 complex was added . From fitting the Arp2/3 complex titration curve , 300 nM was judged to have essentially complete binding , and this was added to a separate titration of actin with Arp2/3 complex present . In fitting all three data sets , a single binding site was assumed on VCA . Single molecule imaging was performed on a custom built multi-wavelength total internal reflection fluorescence ( TIRF ) microscope , described previously ( Friedman et al . , 2006; Hoskins et al . , 2011; Friedman and Gelles , 2012; Smith et al . , 2013 ) . Briefly , the microscope design permitted selective fluorescence excitation of molecules immobilized on the surface of a glass flow chamber ( Smith et al . , 2013 ) using three lasers at wavelengths 488 nm , 532 nm , and 633 nm . Emissions were split into short ( <635 nm ) and long ( >635 nm ) wavelengths and focused on different locations on the camera , to allow for simultaneous acquisition of two-color fluorescence ( Friedman et al . , 2006 ) . Experiments were performed using two different modes of operation described below , one for recording Arp2/3 complex and diVCA interacting with tethered filaments , and another for recording diVCA and filaments interacting with tethered Arp2/3 complexes . All experiments were performed in TIRF buffer: 50 mM KCl , 1 mM MgCl2 , 1 mM EGTA , 10 mM imidazole pH 7 . 2 , 10 mM DTT , 0 . 2 mM ATP , 15 mM glucose , 0 . 02 mg/ml catalase , 0 . 1 mg/ml glucose oxidase , 0 . 1% bovine serum albumin ( BSA ) , and 2% dextran . To suppress blinking of the fluorophores , TIRF buffer was supplemented with a mixture of triplet state quenchers: 1 mM 6-hydroxy-2 , 5 , 7 , 8-tetramethylchroman-2-carboxylic acid ( Trolox ) , 1 mM 4-nitrobenzyl alcohol , and 0 . 5 mM propyl gallate , which were dissolved in DMSO at a 200× stock concentration . Prior to flow chamber assembly ( Smith et al . , 2013 ) , surfaces were first cleaned by sonication in detergent ( 2% Micro-90 , 1 hr ) , ethanol ( 1 hr ) , KOH ( 0 . 1 M , 30 min ) , and deionized water ( 10 min ) , then coated with a mixture of methoxy-poly ( ethylene glycol ) -silane and biotin-poly ( ethylene glycol ) -silane ( mPEG-sil-2000 and biotin-PEG-sil-3400; Laysan Bio Inc . , Arab , AL ) in 80% ethanol , pH 2 HCl , baked overnight at 70°C . Immediately before each experiment , the chambers were incubated in 0 . 5 mg/ml BSA and 0 . 03 mg/ml streptavidin in successive washes . The oxygen scavenging activity of the glucose/glucose oxidase/catalase system in our buffers resulted in a gradual decrease in pH over time through the course of our experiments ( Shi et al . , 2010 ) . We measured this pH change in our microscope flow chambers using a pH-sensitive fluorophore ( SNARF-4F , pKa ∼6 . 4; Molecular Probes/Life Technologies ) , by recording the ratio of fluorescence emission intensities in the two color channels of our microscope , exciting at fixed wavelength ( 533 nm ) . The ratiometric response from 0 . 1 µM SNARF-4F was first calibrated using strong buffers ( 0 . 1 M phosphate buffer ) over a pH range of 5 . 8–8 . 0 ( increment 0 . 2 ) . The time dependent pH of our TIRF buffer was then measured and found to decrease from 7 . 0 to 6 . 5 over the typical time course of our experiments ( ∼30 min ) . The effect of pH on Arp2/3 complex mediated actin filament assembly was also tested and showed that decreasing the pH from 7 to 6 increased nucleation twofold without VCA , while stimulation by diVCA and mutants differed by <1 . 5-fold . Thus , we expect that the buffer conditions used for the single molecule experiments had only a minor influence on diVCA stimulation of branch formation by Arp2/3 complex . For tethered-filament experiments , a mixture of 10% AF488 labeled actin , 1% biotinated actin , and unlabeled actin was allowed to polymerize for 2–8 hr at 3 μM in TIRF buffer without dextran . Preassembled filaments were then diluted 40-fold , flowed into the microscope observation chamber coated with a 1:100 mixture of biotin-PEG-silane:PEG-silane , and allowed to adhere to the surface . The chamber was then rinsed with TIRF buffer and the reaction mixture was introduced: 1 μM actin ( 10% AF488 labeled ) , 5 nM Arp2/3-SNAP649 , and 5 nM Cy3-diVCA . Data was recorded for 15–30 min in cycles of a single 50 ms frame of fluorescence emission using 488 nm laser excitation ( to image actin filaments ) followed by continuous acquisition of 200 frames ( 50 ms per frame ) of emissions using dual 532 nm and 633 nm excitation ( to image Arp2/3 complex and diVCA ) . Each cycle was followed by a ∼1 s delay during which the focus was automatically adjusted ( Smith et al . , 2013 ) , bringing the total cycle interval to ∼12 s . For tethered-Arp2/3 experiments , 1 nM Arp2/3-SNAP-649-biotin was introduced into a microscope observation chamber coated with a 1:2000 mixture of biotin-PEG-silane:PEG-silane , and allowed to adhere to the surface . The chamber was then rinsed with TIRF buffer and the reaction mixture was introduced: 1 μM actin ( 10% AF488 labeled ) and 0 . 1–1 nM Cy3-diVCA . Data was recorded for 30–50 min in cycles of a single frame of fluorescence emission using dual 488 nm and 633 nm excitation ( to image actin filaments and tethered-Arp2/3 ) followed by continuous acquisition of emissions using 532 nm excitation ( to image VCA ) . For these experiments the frame duration was varied in the range 0 . 05–1 s and the number of frames per cycle was adjusted such that each cycle interval was 14–28 s . The frame duration was increased to allow for the use of lower power excitation , so as to characterize and correct for the effect of photobleaching on the observed lifetimes of Cy3-diVCA bound to tethered Arp2/3 complexes ( Figure 4—figure supplement 2D ) . For each experiment , we calibrated the transformation matrix for superimposing images from the split emission recordings by imaging fluorescent beads that emitted in both the short- and long-wavelength color channels of the microscope . This transformation corrected for the translocation of the two color images on the camera , as well as differences in the rotation and magnification in the two channels . For records of colocalization on tethered Arp2/3 complexes we also corrected for drift in the microscope stage using automated tracking of the centers of 2D-Gaussian-fit fluorescence emissions from individual Arp2/3-SNAP649-biotin molecules stably tethered to the microscope slide . Image processing and kinetic analysis was performed in ImageJ ( National Institutes of Health , Bethesda , MD ) , Origin ( OriginLab Corp . , Northampton , MA ) and with custom programs developed in Matlab ( Mathworks , Natick , MA ) . | Most cells are neither perfect spheres nor amorphous blobs , but instead have characteristic shapes that enable them to carry out specific roles within tissues or organs . These shapes are established by a type of scaffolding , called the cytoskeleton , that gives structure to the cell , and also forms networks over which other proteins , and even organelles , can travel . The filaments that make up the cytoskeleton are composed of various proteins , one of which is called actin . Cellular actin filaments can grow by adding new actin molecules , and actin filaments can also have ‘branches’ that fork out from the mother filament . Branches grow out of an assembly of seven proteins known as the Arp2/3 complex , which attaches to the side of the mother filament . Branch growth is triggered by binding to the Arp2/3 complex of an additional protein , WASP , but the sequence of events required to initiate a new branch is not well understood . In particular , WASP is bound to cell membranes; at some point it must detach from the Arp2/3 complex so that the nearness of the membrane does not interfere with the growth of branches . Now , Smith et al . uncover how branch formation is triggered , and define a new role played by WASP in this process . It is known that a specific region of the WASP protein called the VCA domain binds to both the Arp2/3 complex and actin . Smith et al . studied how this domain could initiate branch formation , and showed that a pair of VCA domains linked to each other , along with an Arp2/3 complex , could interact jointly with an existing actin filament before a new branch formed . However , new branches did not form unless the VCA-domain pair detached from the actin filament , leaving the Arp2/3 complex behind . Additionally , Smith et al . found that mutant VCA-domain pairs detached from the actin filament at different rates , which then determined the chance that a new branch formed . These findings—and those of Helgeson and Nolen published concurrently in eLife—suggest that , in cells , two WASP proteins first recruit the Arp2/3 complex to the membrane , and that together they interact with an existing actin filament . The WASP proteins then release the filament , and only then does the Arp2/3 complex initiate the formation of an actin branch . Since the Arp2/3 complex is no longer attached to WASP , subsequent growth of the branch is not physically limited by linkage to the membrane . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2013 | Three-color single molecule imaging shows WASP detachment from Arp2/3 complex triggers actin filament branch formation |
The target of rapamycin ( TOR ) protein kinase forms multi-subunit TOR complex 1 ( TORC1 ) and TOR complex 2 ( TORC2 ) , which exhibit distinct substrate specificities . Sin1 is one of the TORC2-specific subunit essential for phosphorylation and activation of certain AGC-family kinases . Here , we show that Sin1 is dispensable for the catalytic activity of TORC2 , but its conserved region in the middle ( Sin1CRIM ) forms a discrete domain that specifically binds the TORC2 substrate kinases . Sin1CRIM fused to a different TORC2 subunit can recruit the TORC2 substrate Gad8 for phosphorylation even in the sin1 null mutant of fission yeast . The solution structure of Sin1CRIM shows a ubiquitin-like fold with a characteristic acidic loop , which is essential for interaction with the TORC2 substrates . The specific substrate-recognition function is conserved in human Sin1CRIM , which may represent a potential target for novel anticancer drugs that prevent activation of the mTORC2 substrates such as AKT .
Sin1 ( SAPK-interacting protein 1 ) was identified in a yeast two-hybrid screen as a protein that interacts with the stress-activated Spc1/Sty1 MAP kinase ( MAPK ) in the fission yeast Schizosaccharomyces pombe ( Wilkinson et al . , 1999 ) . Because of the stress-sensitive phenotype of the sin1 null ( ∆sin1 ) mutant , Sin1 was proposed to regulate the stress MAPK cascade; however , a later scrutiny failed to find a clear functional link between Sin1 and the MAPK cascade ( Ikeda et al . , 2008 ) . Sin1 orthologs have been identified in diverse eukaryotic species through various screens . A partial cDNA clone encoding a human ortholog was isolated in a genetic screen for suppressors of the RAS function in budding yeast ( Colicelli et al . , 1991 ) . Biochemical isolation of proteins binding to mammalian MAPK kinase kinase , MEKK2 , also identified a human Sin1 ortholog ( Cheng et al . , 2005 ) . A Dictyostelium homolog RIP3 was isolated in a yeast two-hybrid screen for proteins interacting with mammalian Ha-Ras ( Lee et al . , 1999 ) . Identification of Sin1 orthologs as a component of the TOR ( target of rapamycin ) protein complex provided an important clue for the cellular function of Sin1 ( Loewith et al . , 2002; Wedaman et al . , 2003; Lee et al . , 2005; Frias et al . , 2006; Jacinto et al . , 2006; Yang et al . , 2006 ) . TOR is a serine/threonine-specific protein kinase conserved from yeast to humans , forming two distinct protein complexes referred to as TOR complex 1 ( TORC1 ) and TOR complex 2 ( TORC2 ) , the latter of which contains Sin1 as a stable subunit ( Wullschleger et al . , 2006 ) . Mammalian TORC2 ( mTORC2 ) functions downstream of the PI3K pathway , activating AGC-family protein kinases , such as AKT , protein kinase Cα ( PKCα ) and SGK1 , by direct phosphorylation of their hydrophobic motif ( Sarbassov et al . , 2004; Hresko and Mueckler , 2005; Sarbassov et al . , 2005; Guertin et al . , 2006; Facchinetti et al . , 2008; García-Martínez and Alessi , 2008; Ikenoue et al . , 2008; Lu et al . , 2010 ) . Thus , mTORC2 appears to serve as a signaling node that controls protein kinases pivotal to the regulation of cellular metabolism and proliferation ( Cybulski and Hall , 2009 ) . In the SIN1 knockout/knockdown cell lines , no functional mTORC2 is formed ( Frias et al . , 2006; Jacinto et al . , 2006; Yang et al . , 2006 ) and therefore , it has been proposed that SIN1 is required for the assembly/integrity of mTORC2 . In addition , immunoprecipitation of SIN1 resulted in co-purification of AKT , raising the possibility that SIN1 might serve as a scaffold for AKT ( Jacinto et al . , 2006; Cameron et al . , 2011 ) . Also in fission yeast , Sin1 is a stable subunit of TORC2 that assembles around Tor1 , one of the two TOR kinases in this organism ( Hayashi et al . , 2007; Matsuo et al . , 2007 ) . Fission yeast TORC2 is regulated by the Rab GTPase Ryh1 ( Tatebe et al . , 2010 ) and phosphorylates S546 within the hydrophobic motif of the Gad8 kinase , which has significant sequence homology to human AKT and SGK1 ( Matsuo et al . , 2003; Ikeda et al . , 2008 ) . As in the tor1 null ( ∆tor1 ) strain , Gad8-S546 is not phosphorylated in the ∆sin1 strain ( Ikeda et al . , 2008; Tatebe et al . , 2010 ) . In addition , ∆sin1 mutant cells show phenotypes indistinguishable from those of ∆tor1 and ∆gad8 cells , including sterility and hypersensitivity to stress ( Wilkinson et al . , 1999; Kawai et al . , 2001; Weisman and Choder , 2001; Matsuo et al . , 2003; Ikeda et al . , 2008 ) . These observations indicate that Sin1 plays a critical role as a TORC2 subunit in phosphorylating and activating Gad8 , although the exact molecular function of Sin1 remains obscure . In this report , we present a set of evidence that the Sin1 subunit of fission yeast TORC2 binds and recruits Gad8 kinase for phosphorylation . The evolutionarily conserved central region of Sin1 , called conserved region in the middle ( CRIM ) ( Schroder et al . , 2004 ) , is sufficient to recognize and bind Gad8 . In addition , the CRIM domain of human SIN1 can bind mTORC2 substrates such as AKT , PKCα and SGK1 , suggesting the conserved role of Sin1 CRIM in substrate recognition by TORC2 . The NMR structure of Sin1 CRIM shows a ubiquitin-like fold with a negatively charged , protruding loop , which plays a key role in recruiting the TORC2 substrates .
In mTORC2 , it has been proposed that the SIN1 subunit is required for the interaction between mTOR and the RICTOR subunit , as their association was not detectable in the SIN1 knockout cell line ( Jacinto et al . , 2006 ) . Although this observation might be due to the reduced level of RICTOR in the absence of SIN1 ( Frias et al . , 2006 ) , we tested whether absence of Sin1 affects the TORC2 integrity in the fission yeast S . pombe , in which TORC2 is not essential for cell viability . Immunoprecipitation of the FLAG epitope-tagged Tor1 detected its association with the RICTOR ortholog Ste20 in wild-type , ∆sin1 , ∆wat1 and ∆bit61 cell lysate , indicating that Tor1 interacts with Ste20 in the absence of Sin1 or other TORC2 subunits ( Figure 1A ) . Reciprocal experiments also confirmed this conclusion ( Figure 1B ) . Similarly , the mLST8 ortholog Wat1 co-precipitated with Tor1 from the lysate of strains lacking other TORC2 subunits , including Sin1 ( Figure 1—figure supplement 1A and B ) . It has also been found that Sin1 is not required for the association of Bit61 , a fission yeast ortholog of PROTOR/PRR5 ( Hayashi et al . , 2007 ) , with TORC2 ( Tatebe and Shiozaki , 2010 ) . Together , these results indicate that the Sin1 subunit is not crucial for the integrity of S . pombe TORC2 and , even in the absence of Sin1 , the other TORC2 subunits assemble into a complex . 10 . 7554/eLife . 19594 . 003Figure 1 . CRIM , but not the RBD and PH domain , is essential for Sin1 function . ( A , B ) The RICTOR ortholog Ste20 associates with Tor1 in the absence of Sin1 . ( A ) Co-purification of Ste20-myc with FLAG-Tor1 was tested by anti-FLAG ( ‘α-FLAG’ ) affinity beads in wild-type ( CA7087 ) , ∆sin1 ( CA7143 ) , ∆bit61 ( CA7150 ) and ∆wat1 ( CA7151 ) cell lysate . A ste20:myc strain expressing untagged Tor1 ( CA6435 ) was used as a negative control ( tor1+ ) . ( B ) Co-purification of FLAG-Tor1 with Ste20-myc was tested by anti-myc ( ‘α-myc’ ) affinity beads as in ( A ) . A FLAG:tor1 strain expressing untagged Ste20 ( CA6530 ) was used as a negative control ( ste20+ ) . ( C ) The CRIM , but not the RBD and C-terminal PH domain , of Sin1 is required for TORC2 to phosphorylate Gad8 . TORC2-dependent phosphorylation of Gad8-S546 and total Gad8 were detected by immunoblotting ( Tatebe et al . , 2010 ) in cell lysate from a ∆sin1 strain ( CA5126 ) carrying plasmids to express full-length ( 1–665 aa ) or various Sin1 fragments fused to the myc epitope . ( D ) Expression of the mutant Sin1 lacking the C-terminus rescues the ∆sin1 phenotype . Their stress sensitivity was evaluated by growth on solid YES medium containing either 1 M KCl or 0 . 1 M CaCl2 for 2 days at 30°C . CRIM , Conserved region in the middle; PH , Pleckstrin homology; RBD , Ras-binding domain . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 00310 . 7554/eLife . 19594 . 004Figure 1—figure supplement 1 . Sin1 is dispensable for the interaction between Tor1 and the LST8 ortholog Wat1 . ( A ) Five fission yeast strains ( CA7183 , CA7213 , CA7286 , CA7317 , CA7329 ) were grown to mid log phase , and the cell lysate was subjected to immunoprecipitation with anti-FLAG-affinity gel , followed by immunoblotting with anti-myc and anti-FLAG antibodies . ( B ) Five fission yeast strains ( CA6530 , CA7213 , CA7286 , CA7317 and CA7329 ) were tested as in ( A ) with anti-myc-affinity gel . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 00410 . 7554/eLife . 19594 . 005Figure 1—figure supplement 2 . The CRIM domain is not required for Sin1 to bind the other TORC2 subunits . ∆sin1 strains expressing FLAG-tagged Tor1 ( CA6870 ) or Ste20 ( CA6275 ) were transformed with an empty vector , or a plasmid that expresses the myc epitope-tagged full-length Sin1 ( 1–665 aa ) or Sin1 fragments lacking CRIM ( ∆291–390 aa , 1–290 aa; Figure 1C ) . Sin1-myc was precipitated with anti-myc-affinity beads and co-purification of FLAG-Tor1 and Ste20-FLAG was examined by anti-FLAG immunoblotting . CRIM , Conserved region in the middle . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 005 Sin1 orthologs from yeast to humans have a Ras-binding domain ( RBD ) and a pleckstrin homology ( PH ) domain at their C terminus ( Figure 1C ) ( Schroder et al . , 2007; Pan and Matsuura , 2012 ) . It was previously reported that the C-terminal 164 amino acid residues of S . pombe Sin1 , including the PH domain and a part of the RBD , are essential for Sin1 function in vivo ( Wilkinson et al . , 1999 ) . Unexpectedly , however , our immunoblotting experiments with the antibodies against Gad8 phosphorylated at S546 within the hydrophobic motif ( Tatebe et al . , 2010 ) found that the TORC2-dependent phosphorylation of Gad8 was not compromised in S . pombe cells expressing Sin1 lacking the RBD and PH domain ( 1–530 aa and 1–390 aa , Figure 1C ) . Consistently , the stress-sensitive phenotypes of the ∆sin1 mutant ( Matsuo et al . , 2007; Ikeda et al . , 2008 ) was complemented by expressing those truncated Sin1 proteins ( Figure 1D ) , confirming that the C-terminal region of Sin1 including its PH domain is dispensable for the Sin1 function in fission yeast . On the other hand , further C-terminal truncation ( 1–290 aa ) or internal deletion ( ∆291–390 aa ) , which removes the CRIM ( Schroder et al . , 2004 ) , completely abolished the Sin1 function ( Figure 1C and D ) , although those Sin1 mutant proteins were still capable of forming a complex with the other TORC2 subunits ( Figure 1—figure supplement 2 ) . These results strongly suggest an essential role of the Sin1 CRIM in the TORC2 function . During a yeast two-hybrid screen to identify fission yeast proteins that interact with the Gad8 kinase ( Materials and methods ) , we isolated cDNA clones encoding Sin1 . Immunoprecipitation of FLAG epitope-tagged Gad8 from fission yeast cell lysate co-purified a small amount of Sin1 ( Figure 2A ) , confirming weak but detectable interaction between the two proteins in fission yeast . Yeast two-hybrid assays using a series of truncated Sin1 fragments indicated that Gad8 binds to the amino acid residues 281–400 of Sin1 ( Figure 2—figure supplement 1A ) , which mostly overlap with the CRIM domain and are thus referred to as SpSin1CRIM hereafter . When SpSin1CRIM fused to GST was expressed in S . pombe , Gad8 was co-purified with this GST fusion ( Figure 2B ) , confirming that this 120-residue region of Sin1 is sufficient for the interaction with Gad8 . Moreover , induced overexpression of the GST-SpSin1CRIM fusion significantly reduced the Gad8-S546 phosphorylation ( Figure 2C ) ; one likely possibility is that the expressed SpSin1CRIM fragment can compete with TORC2 for Gad8 . 10 . 7554/eLife . 19594 . 006Figure 2 . Sin1CRIM binds specifically to the TORC2 target AGC kinases . ( A ) Sin1 interacts with the TORC2 substrate Gad8 in S . pombe . Co-purification of Sin1-myc with Gad8-FLAG ( ‘FL’ ) precipitated by anti-FLAG beads was tested in a sin1:myc gad8:FLAG strain ( lane 2; CA6993 ) . sin1:myc gad8+ ( lane 1; CA6984 ) and gad8:FLAG sin1+ ( lane 3; CA6281 ) strains were used as negative controls . ( B ) SpSin1CRIM can bind Gad8 in S . pombe . SpSin1CRIM fused to GST was expressed in a gad8:FLAG strain ( CA6281 ) , and proteins collected on glutathione ( GSH ) -beads and the cell lysate were analyzed by Coomassie blue staining and anti-FLAG immunoblotting . ( C ) Overexpressed SpSin1CRIM inhibits Gad8 phosphorylation by TORC2 . GST or GST- SpSin1CRIM ( 247–400 ) were induced ( ‘On’ ) from the thiamine-repressible nmt1 promoter and the crude cell lysate was analyzed by immunoblotting . ( D ) Human Sin1CRIM binds mTORC2 substrates , but not the mTORC1 substrate S6K1 . FLAG-tagged AKT , PKCα , Sgk1 and S6K1 were expressed in HEK-293T and the cell lysate was incubated with bacterially produced MBP- or GST-fused HsSin1CRIM , which were immobilized onto amylose- and GSH-beads , respectively . Proteins bound to the beads and the cell lysate were analyzed by anti-FLAG immunoblotting and Coomassie blue staining . ( E ) SpSin1CRIM does not interact with Psk1 , a S . pombe TORC1 substrate . Bacterially produced GST-SpSin1CRIM bound to GSH-beads was incubated with cell lysate from gad8:FLAG ( CA6281 ) and psk1:FLAG ( CA8070 ) strains of S . pombe . Proteins bound to the GSH-beads and the cell lysate were analyzed by anti-FLAG immunoblotting and Coomassie blue staining . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 00610 . 7554/eLife . 19594 . 007Figure 2—figure supplement 1 . Interaction between fission yeast Sin1 – Gad8 and human Sin1 – AKT2 in yeast two-hybrid assays . ( A ) Interaction between various fission yeast Sin1 fragments and Gad8 in yeast two-hybrid assays . Abbreviation: CRIM , Conserved Region In the Middle; RBD , Raf-like Ras-binding domain; PH , pleckstrin homology . CRIM spans the amino acid residues 253–383 of fission yeast Sin1 ( Schroder et al . , 2004 ) ; RBD and PH do the residues 438–514 and 560–658 , respectively ( Schroder et al . , 2007 ) . Interaction was assessed by histidine auxotrophy: ‘+’ , interaction; '–' , no interaction . ( B ) Interaction of Gad8 fragments with Sin1 in yeast two-hybrid assays . Abbreviations: C2 , C2 domain; Pkinase , protein kinase domain; T , protein kinase C-terminal domain . Domain prediction is based on the motifs PF00069 , PF00433 , and PF00168 in the Pfam database . ( C ) Interaction between various human Sin1 central fragments and human AKT2 in yeast two-hybrid assays . CRIM spans the amino acid residues 137–267 of human Sin1i2 ( Schroder et al . , 2004 ) , while PH does the amino acid residues 348–450 ( Schroder et al . , 2007 ) . Note that the Sin1i2 isoform does not contain the intact RBD . ( D ) Interaction of human AKT2 fragments with human SIN1 in yeast two-hybrid assays . Abbreviations: PH , PH domain; Pkinase , protein kinase domain; T , protein kinase C terminal domain . Domain prediction is based on PF00069 , PF00433 , and PF00169 in the Pfam database . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 00710 . 7554/eLife . 19594 . 008Figure 2—figure supplement 2 . Yeast two-hybrid assay for interaction between HsSin1CRIM and PKCα . Interaction was examined by HIS3 reporter gene expression in the histidine auxotrophic HF7c budding yeast strain . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 00810 . 7554/eLife . 19594 . 009Figure 2—figure supplement 3 . Yeast two-hybrid assay for interaction between SpSin1CRIM and fission yeast Sck1 , Sck2 , Psk1 , Pck1 , Pck2 . Interaction was examined by HIS3 reporter gene expression in the histidine auxotrophic HF7c budding yeast strain . The Sck2 and Pka1 baits exhibited auto-activation in HIS3 reporter gene expression independently of the SpSin1CRIM prey ( labeled by asterisks ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 00910 . 7554/eLife . 19594 . 010Figure 2—figure supplement 4 . Yeast two-hybrid assay for cross-species interaction between Sin1 , Gad8 , and AKT2 . Interaction was examined by HIS3 reporter gene expression in the histidine auxotrophic HF7c budding yeast strain . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 010 Because of the evolutionary conservation of the CRIM domain among Sin1 orthologs ( Schroder et al . , 2004 ) , we examined whether human SIN1 CRIM binds to the mTORC2 substrates , such as AKT , PKCα and SGK1 ( Sarbassov et al . , 2005; Jacinto et al . , 2006; García-Martínez and Alessi , 2008; Cameron et al . , 2011 ) . Although the physical interaction between human SIN1 and AKT has been controversial ( Jacinto et al . , 2006; Cameron et al . , 2011; Lu et al . , 2011 ) , our yeast two-hybrid assays found that the residues 122–314 of human SIN1 , referred to as HsSin1CRIM hereafter , interacted with AKT as well as PKCα ( Figure 2—figure supplements 1C , D and 2 ) . Furthermore , in an in vitro binding assay , AKT , PKCα as well as SGK1 were co-precipitated with bacterially produced HsSin1CRIM ( Figure 2D ) . Although the interaction between SIN1 and SGK1 has been reported ( Lu et al . , 2011 ) , our experiments showed for the first time that the CRIM domain of SIN1 is sufficient for the interaction with SGK1 . Another mTORC2 substrate , PKCε , is known to interact with SIN1 fragments that include the CRIM domain ( Cameron et al . , 2011 ) . Together , these results strongly suggest that the ability of the CRIM domain to bind the TORC2 substrate kinases is conserved between S . pombe and humans . Despite the structural similarity among the AGC kinase family members , TORC2 phosphorylates only a subset of them; human S6K1 and its fission yeast ortholog Psk1 are AGC kinases whose hydrophobic motif is phosphorylated by TORC1 rather than TORC2 ( Magnuson et al . , 2012; Nakashima et al . , 2012 ) . Consistently , human S6K1 and S . pombe Psk1 showed no detectable binding to HsSin1CRIM and SpSin1CRIM , respectively ( Figure 2D and E ) . Psk1 and other S . pombe AGC kinases that are not regulated by TORC2 , including Sck1 , Sck2 , Pck1 and Pck2 ( Nakashima et al . , 2012; Madrid et al . , 2015 ) , also failed to interact with SpSin1CRIM in yeast two-hybrid assays ( Figure 2—figure supplement 3 ) . Thus , it appears that , in both fission yeast and humans , the CRIM domain of Sin1 can selectively bind the TORC2 substrates among the members of the AGC kinase family . It should be noted , however , that the CRIM domain does not seem to recognize the TORC2 phosphorylation sites per se , because CRIM can interact with Gad8 and AKT lacking their C-terminal hydrophobic motif ( Figure 2—figure supplement 1B and D ) . For better understanding of the molecular mechanism by which the CRIM domain recognizes the TORC2 substrates , we sought amino acid residues essential for the SpSin1CRIM function . Random mutagenesis followed by yeast two-hybrid screens isolated several SpSin1CRIM missense mutations that abolish the interaction with Gad8 ( Figure 3A ) . Each of the isolated mutations was introduced into GST-SpSin1CRIM , and all the successfully expressed mutant proteins showed significantly compromised interaction with Gad8 in fission yeast cells ( Figure 3B; data not shown ) . Those confirmed mutations were then introduced to the full-length Sin1 protein tagged with the myc epitope and expressed in the ∆sin1 strain for further analyses . Complementation of the ∆sin1 defects in Gad8 phosphorylation ( Figure 3C ) and cellular stress resistance ( Figure 3D ) was not observed even with the mutant Sin1 proteins whose expression was comparable to that of the wild-type protein . On the other hand , with the exception of the marginal effect of L348S , those amino acid substitutions within CRIM had no apparent impact on the ability of Sin1 to interact with Tor1 ( Figure 3E ) . Thus , these Sin1 mutant proteins were incorporated into TORC2 , but the mutations to the CRIM domain compromised the interaction with Gad8 , leading to the inability of TORC2 to phosphorylate Gad8 . 10 . 7554/eLife . 19594 . 011Figure 3 . Sin1CRIM forms a ubiquitin-fold domain that binds TORC2 substrates . ( A ) Yeast two-hybrid screens isolated SpSin1CRIM mutations that abrogate the interaction with Gad8 ( red ) . Pairwise alignment of the fission yeast and human CRIM sequences was performed using the SSEARCH program ( Pearson , 1991 ) . ( B ) Defective interaction of the mutant SpSin1CRIM with Gad8 . The Sin1CRIM ( residues 247–400 or 281–400 ) fragments and those with the indicated mutations were expressed as GST-fusion in a gad8:FLAG strain of S . pombe ( CA6281 ) , and proteins collected on GSH-beads and the cell lysate were detected by Coomassie blue staining and anti-FLAG immunoblotting . ( C ) TORC2 with the Sin1CRIM mutations fails to phosphorylate Gad8-S546 . A ∆sin1 strain ( CA5126 ) was transformed with a plasmid to express myc-tagged wild-type or mutant Sin1 carrying one of the CRIM mutations , and the cell lysate was analyzed by immunoblotting . ( D ) Stress sensitivity of the Sin1 mutants in ( C ) was examined by growth on YES agar medium containing either 1 M KCl or 0 . 1 M CaCl2 . ( E ) CRIM mutations do not compromise the incorporation of Sin1 into TORC2 . A ∆sin1 FLAG:tor1 strain ( CA6870 ) was transformed with the plasmids used in ( C ) , and co-purification of the wild-type and mutant Sin1-myc with FLAG-Tor1 precipitated by anti-FLAG beads was tested . A ∆sin1 strain expressing untagged Tor1 ( CA5126 ) was used as a negative control ( tor1+ ) . ( F ) A ribbon representation of the lowest energy structure of SpSin1CRIM ( 272–397 ) . The protruding acidic loop region is marked by a dashed oval . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 01110 . 7554/eLife . 19594 . 012Figure 3—figure supplement 1 . Mutated residues of SpSin1CRIM in the tertiary structure . The Sin1 residues whose mutation abrogates the interaction with Gad8 ( Figure 3A ) are shown as ball-and-stick models in the NMR structure of the CRIM domain . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 012 The experiment above demonstrated that the CRIM domain of Sin1 is essential for TORC2 to bind and phosphorylate the substrate Gad8 . However , the isolated mutations are scattered throughout the SpSin1CRIM ( Figure 3A ) , providing little mechanistic insight into the substrate recognition . NMR chemical shifts and 1H-15N heteronuclear NOE values ( Kataoka et al . , 2014; Furuita et al . , 2014 ) indicated that SpSin1CRIM is a highly loop-rich but structurally ordered . In general , loop-rich proteins are difficult targets to determine the structures by X-ray and NMR . Therefore , a newly developed method that allows determination of loop-rich protein structures by paramagnetic relaxation enhancement NMR techniques ( Furuita et al . , 2014 ) was applied to SpSin1CRIM , with additional structure refinement by the XPLOR-NIH software . The solution structure of SpSin1CRIM has three α helices and four β strands , which are arranged around the longest helix α2 ( Figure 3F and Table 1; the atomic coordinates of the refined structure have been submitted to the Protein Data Bank ) . The overall structure of SpSin1CRIM belongs to the ubiquitin superfold family ( Kiel and Serrano , 2006 ) , but more than half of the structured region consists of loops . 10 . 7554/eLife . 19594 . 013Table 1 . Structural statistics for SpSin1CRIM . RDC correlation coefficient was caluculated using the program PALES ( Zweckstetter and Bax , 2000 ) . Ramachandran analysis was performed using PROCHECK 3 . 5 . 4 ( Laskowski et al . , 1996 ) . Deviations from ideal geometry was caluculated using PDB Validation Server . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 013NOE upper distance restraintsShort-range ( |i−j|<=1 ) 636Medium-range ( 1<|i−j|<5 ) 132Long-range ( 5<=|i−j| ) 199Total967PRE distance restraints748Dihedral angle restraintsφ99ψ110χ112Hydrogen-bond restraints0RMS Deviations ( 272–397 ) ( Å ) backbone0 . 89 ± 0 . 11 heavy1 . 36 ± 0 . 12 RDC correlation coefficient0 . 94 ± 0 . 02 ViolationsNOE ( >0 . 5 Å ) 0PRE ( >0 . 5 Å ) 0Dihedral ( >5° ) 0Maximum violationNOE ( Å ) 0 . 49PRE ( Å ) 0 . 15Dihedral ( ° ) 3 . 1Ramachandran analysis ( 272–397 ) ( % ) Most favored regions91 . 3Additional allowed regions7 . 7Generously allowed regions0 . 5Disallowed regions0 . 6Deviations from ideal geometryBond lingths ( Å ) 0 . 011Bond angles ( ° ) 0 . 14 The majority of the mutations that abrogate the Gad8 binding ( Figure 3A ) are mapped to the buried residues in the structure of SpSin1CRIM ( Figure 3—figure supplement 1 ) and are likely to significantly perturb the overall folding of the domain , thus disrupting the functional TORC2-Gad8 interaction ( Figure 3C ) . A distinctive characteristic of SpSin1CRIM is its protruding loop structure formed around residues 352–361 ( Figures 3F and 4A ) , which are rich in acidic amino acid residues with solvent-exposed side chains ( Figure 4—figure supplement 1 ) . In addition , the primary sequence of this loop region is highly conserved among Sin1 orthologs ( Figure 4B and Figure 4—figure supplement 2 ) . Substitutions of the seven acidic residues in this loop region with asparagine and glutamine ( ‘poly NQ’ , Figure 4B ) completely eliminated the affinity of SpSin1CRIM for Gad8 ( Figure 4C ) without significantly disturbing the overall folding ( Figure 4—figure supplement 3 ) , suggesting that the negative charge of the protruding loop is essential . Further analysis of individual substitution mutants suggested that the acidic residues located in the second half of the loop ( D358 , E359 and D360 ) play more important roles ( Figure 4—figure supplement 4 ) . Two conserved hydrophobic residues within this loop region , L357 and F361 , also appear to contribute to the interaction with Gad8; alanine substitution of each of these residues partially compromised the Gad8 binding ( Figure 4—figure supplement 5 ) , and the L357A/F361A double mutant ( ‘AA’ , Figure 4B ) showed very little interaction with Gad8 ( Figure 4C ) . Consistently , the TORC2-dependent phosphorylation of Gad8-S546 was dramatically impaired in strains expressing Sin1 with those amino-acid substitutions ( Figure 4D ) , which did not affect the association between Sin1 and the Tor1 kinase within the TORC2 complex ( Figure 4E ) . These results indicate that the protruding acidic loop of SpSin1CRIM is essential for TORC2 to recruit and phosphorylate its substrate , Gad8 . 10 . 7554/eLife . 19594 . 014Figure 4 . The protruding acidic loop of CRIM is essential for Sin1 to interact with TORC2 target kinases . ( A ) The molecular surface of SpSin1CRIM colored according to the electrostatic potential ranging from positive ( blue ) to negative ( red ) charge . The protruding acidic loop is marked by a dashed oval . ( B ) Alignment of the fission yeast and human Sin1CRIM sequences that correspond to the acidic protrusion . The ‘poly NQ’ and ‘AA’ show the amino acid substitutions used in ( C ) ~ ( G ) . ( C ) The acidic and hydrophobic residues in the protruding loop of SpSin1CRIM are essential for Gad8 binding in vitro . The recombinant wild-type and mutant GST-SpSin1CRIM on GSH-beads were incubated with cell lysate from a gad8:FLAG strain ( CA6281 ) . Proteins bound to the beads and the cell lysate were analyzed by anti-FLAG immunoblotting and Coomassie blue staining . ( D ) TORC2 with the mutations to the acidic loop of Sin1CRIM fails to phosphorylate Gad8 . A ∆sin1 strain ( CA5126 ) was transformed with a plasmid to express myc-tagged wild-type or mutant Sin1 carrying the poly-NQ and AA mutations , and the cell lysate was analyzed by immunoblotting . ( E ) The acidic loop mutations do not compromise the incorporation of Sin1 into TORC2 . A ∆sin1 FLAG:tor1 strain ( CA6870 ) was transformed with the plasmids used in ( D ) , and co-purification of the wild-type and mutant Sin1-myc with FLAG-Tor1 precipitated by anti-FLAG beads was tested . A ∆sin1 strain expressing untagged Tor1 ( CA5126 ) was used as a negative control ( tor1+ ) . ( F ) The acidic and hydrophobic residues in the conserved loop region of HsSin1CRIM are important to bind AKT . FLAG-tagged AKT was expressed in HEK-293T , and the cell lysate was incubated with the recombinant wild-type and mutant MBP-HsSin1CRIM on amylose-beads . Proteins bound to the beads and the cell lysate were analyzed as in ( C ) . ( G ) The conserved acidic loop region of HsSin1CRIM is important for mTORC2-dependent phosphorylation of AKT-S473 . The wild-type and mutant SIN1 lacking the entire CRIM ( ‘∆CRIM’ ) or acidic region ( '∆acidic’ ) as well as the poly NQ SIN1mutant were expressed with the myc tag in the SIN1-deficient MCF-7 cells . Insulin-stimulated phosphorylation of AKT-S473 was analyzed by immunoblotting . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 01410 . 7554/eLife . 19594 . 015Figure 4—figure supplement 1 . The acidic side chains of the conserved loop region do not interact with other residues of the CRIM domain . The NMR structure of the acidic loop in SpSin1CRIM ( residues 352–361; Figure 4B ) is shown as a stick model . The side chains of the most residues within the acidic loop region face outward and do not interact with other residues of the CRIM domain . The Ligplot+ ( Laskowski and Swindells , 2011 ) and WHAT IF ( Vriend , 1990 ) programs identified neither hydrogen bond nor salt bridge between the loop region and the rest of the CRIM domain . Arg349 and Ile350 are predicted by LIGPLOT software to have hydrophobic interaction with the loop region and are shown by a surface model . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 01510 . 7554/eLife . 19594 . 016Figure 4—figure supplement 2 . The acidic loop region is highly conserved among Sin1 orthologs . Multiple alignment was performed by MSAprobWS ( Liu et al . , 2010 ) via Jalview ( Waterhouse et al . , 2009 ) . Sequences included in the alignment are YOL078W ( S . cerevisiae ) , ANID_06304 ( A . nidulans ) , SPAPYUG7 . 02c ( S . pombe ) , gi52453 ( T . adhaerens ) , NP_077022 . 1 ( H . sapiens ) and XP_002649140 . 1 ( D . discoidum ) . The secondary structure elements of SpSin1CRIM determined by DSSP ( Kabsch and Sander , 1983; Touw et al . , 2015 ) are shown above with cylinders and arrows as α-helices and β-sheets , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 01610 . 7554/eLife . 19594 . 017Figure 4—figure supplement 3 . An overlay of 1H-15N HSQC spectra of the wild-type ( blue ) and ‘poly NQ’ mutant ( red ) of SpSin1CRIM . The NMR spectrum of the ‘poly NQ’ mutant resembles that of the wild-type protein , suggesting that the mutation does not significantly disturb the overall folding of the CRIM domain . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 01710 . 7554/eLife . 19594 . 018Figure 4—figure supplement 4 . Point mutations to the acidic residues in the conserved SpSin1CRIM loop compromise the interaction with Gad8 . The recombinant wild-type and mutant GST-SpSin1CRIM on GSH-beads were incubated with cell lysate from a gad8:FLAG strain ( CA6281 ) . Proteins bound to the beads and the cell lysate were analyzed by anti-FLAG immunoblotting and Coomassie blue staining . Unfused GST and the Sin1 RBD and PH domains ( residues 401–665 ) fused to GST were used as negative controls . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 01810 . 7554/eLife . 19594 . 019Figure 4—figure supplement 5 . Alanine substitution of the conserved hydrophobic residues within the SpSin1CRIM loop reduces the Gad8 binding . The recombinant wild-type and mutant GST-SpSin1CRIM on GSH-beads were incubated with the cell lysate from a gad8:FLAG ( CA6281 ) strain . Gad8 bound to the beads were analyzed by anti-FLAG immunoblotting , followed by densitometric quantification . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 01910 . 7554/eLife . 19594 . 020Figure 4—figure supplement 6 . Inactivation of SIN1 by CRISPR/Cas9-mediated genome editing was confirmed by immunoblotting and genomic sequencing . ( A ) Expression of the SIN1 protein is abolished in several clones of the MCF-7 human culture cells that are transfected with a SIN1-targeting CRISPR/Cas9 construct . In clones #1 , 3 , and 9 , SIN1 ( indicated by arrowheads ) was not detected at all by immunoblotting in either crude cell lysate or anti-RICTOR immunoprecipitate . Accordingly , AKT phosphorylation at the hydrophobic motif was completely lost in those clones . Clone #3 was used in the subsequent experiment shown in Figure 4G . ( B ) Genomic sequencing of SIN1-deficient clone #3 detected a small insertion and small deletions in Exon 6 of human SIN1 that resulted in frameshift mutations or an internal deletion . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 02010 . 7554/eLife . 19594 . 021Figure 4—figure supplement 7 . The CRIM domain of human Sin1 is not required for its interaction with mTOR and RICTOR to form mTORC2 . ( A ) The truncated SIN1 fragments and the SIN1 mutant proteins lacking the conserved acidic loop ( ∆236–245 ) or the CRIM domain ( ∆156–314 ) . ( B ) The wild-type and mutant SIN1 proteins shown in ( A ) were expressed with the myc epitope tag in the SIN1-deficient MCF-7 cells . Co-purification of SIN1-myc with mTOR and Rictor was tested by anti-myc immunoprecipitation ( IP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 02110 . 7554/eLife . 19594 . 022Figure 4—figure supplement 8 . The human Sin1CRIM and its acidic loop are essential for phosphorylation of the mTORC2 targets . ( A ) FLAG epitope-tagged AKT was co-expressed with the indicated SIN1 mutants ( see Figure 4—figure supplement 7A ) in the SIN1-deficient MCF-7 cells . AKT was collected onto anti-FLAG-affinity beads , and immunoblotting was carried out to detect phosphorylation of Ser-473 and Thr-450 in its hydrophobic and turn motifs , respectively . ( B ) FLAG-tagged PKCα was co-expressed with the indicated SIN1 mutants in the SIN1-deficient MCF-7 cells . PKCα was collected onto anti-FLAG-affinity beads and analyzed by immunoblotting . mTORC2-dependent phosphorylation of PKCα is known to increase its stability ( Ikenoue et al . , 2008 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 022 The corresponding ‘poly NQ’ and ‘AA’ substitutions in the HsSin1CRIM ( Figure 4B ) partially compromised its affinity for AKT ( Figure 4F ) . The importance of the conserved acidic residues was further tested by expressing the ‘poly NQ’ SIN1 mutant protein in a human cell line where the SIN1 gene was inactivated by CRISPR/Cas9-mediated genome editing ( Cong et al . , 2013; Mali et al . , 2013 ) ( Figure 4—figure supplement 6 ) . As reported previously ( Frias et al . , 2006; Jacinto et al . , 2006; Yang et al . , 2006 ) , absence of functional SIN1 resulted in the loss of mTORC2-dependent phosphorylation of AKT-S473 in the hydrophobic motif , a defect rescued by expressing wild-type SIN1 ( Figure 4G ) . The AKT phosphorylation was hardly detectable when re-introduced SIN1 lacks the entire CRIM domain ( ‘∆CRIM’ ) or its acidic loop region of residues 236–245 ( ‘∆acidic’ ) ( Figure 4G ) , although these mutant SIN1 proteins can interact with mTOR and RICTOR ( Figure 4—figure supplement 7 ) . Only a low level of the AKT-S473 phosphorylation was observed when the ‘poly NQ’ mutant was expressed ( Figure 4G ) , and similar results were obtained for the mTORC2-dependent phosphorylation of AKT-T450 in the turn motif and PKCα-S657 in the hydrophobic motif ( Figure 4—figure supplement 8 ) . Together , these observations strongly suggest that the conserved acidic residues within the CRIM domain are important for the SIN1 function as mTORC2 subunit . Although SpSin1CRIM alone cannot replace the Sin1 function in vivo ( Figure 5—figure supplement 1 ) , we found that it complemented the stress-sensitive phenotypes of the ∆sin1 mutant when fused to Ste20 , the RICTOR equivalent in S . pombe TORC2 ( Figure 5A ) . Consistently , expression of the Ste20-SpSin1CRIM fusion was found to induce phosphorylation of Gad8-S546 in the ∆sin1 strain ( Figure 5B ) . With the L348S or L364S substitutions that prevent SpSin1CRIM from binding Gad8 ( Figure 3B ) , the Ste20-SpSin1CRIM failed to bring about the Gad8 phosphorylation ( Figure 5B ) . As expected , these substitutions as well as the ‘poly NQ’ and ‘AA’ mutations to the acidic loop ( Figure 4 ) abrogated the ability of Ste20-SpSin1CRIM to complement the ∆ste20 ∆sin1 phenotype ( Figure 5—figure supplement 2 ) . Therefore , the hybrid subunit enables phosphorylation and activation of Gad8 in the absence of Sin1 most likely by recruiting Gad8 to TORC2 . Thus , the CRIM domain alone appears to be sufficient for the substrate-recruiting function of the Sin1 subunit . 10 . 7554/eLife . 19594 . 023Figure 5 . Substrate recruitment by Sin1CRIM determines the specificity of TORC2 . ( A and B ) SpSin1CRIM fused to the Ste20 subunit can substitute for the Sin1 function in TORC2 . In ( A ) , a ∆sin1 strain ( CA5126 ) was transformed with an empty vector , or a plasmid to express the ste20+ , sin1+ or ste20:sin1CRIM genes . Growth of the transformants was tested at 30°C on YES plates with and without 0 . 1 M CaCl2 or 0 . 6 M KCl . In ( B ) , phosphorylation of Gad8-S546 was examined by immunoblotting in a ∆sin1 strain ( CA5126 ) carrying a plasmid to express FLAG-tagged Ste20 fused to SpSin1CRIM with and without L348S/L364S mutations . ( C ) TORC2 without Sin1 can phosphorylate Gad8 fused to the Ste20 subunit . ∆sin1 strains with tor1+ ( CA7471 ) and tor1-D2137A ( CA7395 ) alleles were transformed with a plasmid to express Ste20 fused to Gad8 . Expression and phosphorylation of the Ste20-Gad8 fusion were detected by antibodies against Gad8 and phosphorylated Gad8-S546 , respectively . ( D ) Gad8 fused to the Mip1 subunit of TORC1 is phosphorylated when TORC1 is active . A ∆sin1 strain ( CA5126 ) carrying a plasmid to express Gad8 fused to Mip1 , the Raptor ortholog in fission yeast , was grown to mid-log phase , followed by nitrogen starvation ( -N ) to inactivate TORC1 . Expression and phosphorylation of the Mip1-Gad8 fusion were analyzed as in ( C ) . ( E ) The Mip1-Gad8 fusion is incorporated into TORC1 . Mip1-Gad8 was expressed in FLAG:tor1 ( CA6530 ) and FLAG:tor2 ( CA6904 ) strains , from which TORC2 and TORC1 were collected onto anti-FLAG beads , respectively , and co-purification of Mip1-Gad8 was examined by anti-Gad8 immunoblotting . Alanine substitutions of Mip1 residues 316–318 ( ‘DLF’ ) or 366–369 ( ‘NWIF’ ) prevented the fusion protein from being incorporated into TORC1 . ( F ) Phosphorylation of Gad8-S546 in the Mip1-Gad8 fusion protein incorporated into TORC1 . A ∆sin1 strain ( CA5126 ) was transformed with an empty vector ( − ) or plasmids to express Mip1-Gad8 with or without the mutations that abrogate its incorporation into TORC1 as shown in ( E ) . Expression and phosphorylation of the Mip1-Gad8 proteins were analyzed as in ( D ) . The endogenous Gad8 was monitored as a loading control ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 02310 . 7554/eLife . 19594 . 024Figure 5—figure supplement 1 . The CRIM domain alone is not sufficient to replace the Sin1 function in vivo . A ∆sin1 strain ( CA5126 ) was transformed with an empty vector , a plasmid to express the full-length Sin1 ( ‘Sin1 WT’ ) , or that to express SpSin1CRIM . Growth was tested at 30°C with and without high osmolarity stress of 1 M KCl . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 02410 . 7554/eLife . 19594 . 025Figure 5—figure supplement 2 . When fused to Ste20 , SpSin1CRIM can replace the Sin1 function in a manner dependent on its ability to bind Gad8 . A ∆ste20 ∆sin1 ( CA7471 ) strain was transformed with an empty vector , a plasmid expressing the ste20+ gene , or a plasmid that expresses either wild-type or mutant SpSin1CRIM fused to the Ste20 subunit . Transformants were incubated at 30°C in the absence ( ‘no stress’ ) or in the presence of 1 M KCl . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 025 To further confirm that Sin1 is not essential for the catalytic activity of TORC2 per se , we expressed Gad8 as a fusion to the Ste20 subunit , so that Gad8 physically associates with TORC2 even in the absence of Sin1 . As shown in Figure 5C , the Gad8 phosphorylation was detectable within the Ste20-Gad8 fusion protein in ∆sin1 cells that express the wild-type Tor1 kinase , but hardly in those expressing Tor1-D2137A with compromised catalytic activity ( Tatebe et al . , 2010 ) . Thus , TORC2 lacking Sin1 can phosphorylate the hydrophobic motif of Gad8 recruited to TORC2 , indicating that Sin1 is not essential for the catalytic activity of TORC2 . We also constructed a fusion gene between gad8+ and mip1+ , which encodes a S . pombe TORC1 subunit homologous to mammalian RAPTOR ( Matsuo et al . , 2007; Alvarez and Moreno , 2006 ) , and expressed it in the ∆sin1 mutant . The hydrophobic motif of Gad8 within the Mip1-Gad8 fusion was phosphorylated ( Figures 5D and 0 min ) , which became negligible immediately after nitrogen starvation that inactivates TORC1 but not TORC2 ( Alvarez and Moreno , 2006; Uritani et al . , 2006; Matsuo et al . , 2007; Nakashima et al . , 2012; Hatano et al . , 2015 ) . This result suggests that TORC1 can phosphorylate Gad8 , a TORC2 substrate , when Gad8 is artificially brought to TORC1 . Indeed , co-precipitation experiments showed that the Mip1-Gad8 fusion protein forms a complex preferentially with Tor2 , the catalytic subunit of TORC1 , rather than with Tor1 ( Figure 5E ) . This association with Tor2 was significantly compromised by introducing alanine substitutions to residues 316–318 or 366–369 of Mip1 in the fusion protein ( ‘DLF->AAA’ and ‘NWIF->AAAA’ , respectively ) ; the equivalent substitutions in human RAPTOR are known to abrogate its interaction with mTOR ( Kim et al . , 2002 ) . As expected , the compromised interaction of Mip1-Gad8 with Tor2 dramatically undermined the phosphorylation of the Gad8 hydrophobic motif ( Figure 5F ) , confirming that the detected phosphorylation is dependent on the association of Gad8 with TORC1 . Although TORC1 and TORC2 in fission yeast have different catalytic subunits , Tor2 and Tor1 , respectively ( Hayashi et al . , 2007; Matsuo et al . , 2007 ) , TORC1 is capable of phosphorylating the TORC2 substrate Gad8 , when Gad8 physically interacts with TORC1 .
Unlike in other model organisms , TORC2 in the fission yeast S . pombe is not essential for viability . In this study , we utilized this experimental organism to clarify the molecular function of Sin1 , an evolutionarily conserved subunit of TORC2 . In the sin1 null mutant of S . pombe , we observed assembly of the Sin1-free TORC2 , which can phosphorylate Gad8 fused to the Ste20 subunit , suggesting that ensuring the functional assembly and integrity of TORC2 is not an intrinsic function of Sin1 . Although the budding yeast Sin1 ortholog was implicated in targeting TORC2 to the plasma membrane ( Berchtold and Walther , 2009 ) , the membrane localization of fission yeast TORC2 is not dependent on Sin1 ( Tatebe et al . , 2010 ) . On the other hand , our data presented here demonstrate that Sin1 is the substrate-recruiting subunit of S . pombe TORC2 . Furthermore , the critical role of the CRIM domain in substrate recognition appears to be conserved in the human SIN1 . We showed for the first time that recombinant HsSin1CRIM can bind the mTORC2 substrates such as AKT , PKCα and SGK1 in vitro , but not the TORC1 substrate S6K1 . Interestingly , the fission yeast CRIM also interacts with AKT in yeast two-hybrid assays ( Figure 2—figure supplement 4 ) , implying the evolutionarily conserved structure and function of CRIM . Budding yeast TORC2 additionally requires a PH domain protein , Slm1 , to recruit the Ypk1 kinase for phosphorylation ( Berchtold et al . , 2012; Niles et al . , 2012 ) . However , the unique Slm ortholog SPAC637 . 13c in fission yeast has no apparent role in Gad8 phosphorylation by TORC2 ( manuscript in preparation ) , and no Slm ortholog has been found in metazoan . CRIM is the signature sequence of Sin1 orthologs in diverse eukaryotes ( Schroder et al . , 2004 ) , and its solution structure shows a ubiquitin-like fold . Ubiquitin superfold domains identified in intracellular signaling proteins often mediate protein-protein interaction; their interaction surfaces are diverse , involving different segments of the ubiquitin fold as well as loop regions ( Kiel and Serrano , 2006; Sumimoto et al . , 2007 ) . We found that SpSin1CRIM has a distinctive acidic loop that projects from the ubiquitin-like fold . Our experiments with human SIN1 also suggested a critical role of the conserved acidic stretch in the SIN1-AKT interaction and phosphorylation of AKT . Eliminating the sequence that spans this acidic stretch of SIN1 also compromises mTORC2-dependent phosphorylation of PKCα ( Cameron et al . , 2011 ) , although such a relatively large deletion may disturb the overall folding of Sin1CRIM . It should be noted that our results do not exclude the possibility of additional SIN1-substrate interaction outside of the CRIM domain . Like AKT , SGK1 binds HsSin1CRIM ( Figure 2D ) , but it was also reported that the Q68H mutation to the N-terminal region of SIN1 impairs its interaction with SGK1 ( Lu et al . , 2011 ) . The Q68H mutation , however , does not affect SIN1-AKT interaction and therefore , additional SIN1-substrate interaction outside of the CRIM domain may be substrate-specific . Yeast two-hybrid analyses showed that HsSin1CRIM interacts with the kinase catalytic domain of AKT ( Figure 2—figure supplement 1D ) . On the other hand , the kinase domain fragment of Gad8 lacking the N-terminal , non-catalytic domain failed to interact with SpSin1CRIM ( Figure 2—figure supplement 1B ) . Interestingly , this N-terminal region of Gad8 and its budding yeast orthologs , Ypk1 and Ypk2 , contain a conserved auto-inhibitory sequence ( Kamada et al . , 2005; our unpublished results ) ; without the N-terminal auto-inhibitory domain , the Gad8 kinase domain is expected to be in its active conformation , which might affect the interaction with the SpSin1CRIM . We have not succeeded in localizing the CRIM interaction to any subfragment of the Gad8/AKT catalytic domains , and further study is required to elucidate how Sin1CRIM recognizes the TORC2 substrate kinases , including Sin1CRIM–kinase co-crystallization . We show that Gad8 , a TORC2 substrate , can be phosphorylated by TORC1 when recruited to TORC1 by gene fusion . Thus , the substrate specificity of the TOR complexes seems to be determined by their ability to recruit specific substrates through the regulatory subunits . In mTORC1 , RAPTOR is such a regulatory subunit that recognizes the TOR Signaling ( TOS ) motif in S6K1 and 4E-BP1 ( Hara et al . , 2002; Schalm and Blenis , 2002; Schalm et al . , 2003 ) , although many TORC1 substrates have no apparent TOS motif . Our study has shown that Sin1 is the substrate-recruiting subunit of TORC2 , and that its CRIM domain is the determinant of the substrate specificity . Hyperactivation of mTORC2 is often a key event in cancer development ( Guertin et al . , 2009; Lee et al . , 2012; Lawrence et al . , 2014; Carr et al . , 2015 ) , but there is no effective drug that specifically inhibits mTORC2 . A possible strategy for development of mTORC2 inhibitors is to find small molecules that target the substrate-binding interface of mTORC2 ( Sparks and Guertin , 2010 ) , and our study is expected to serve as a basis for such an approach .
S . pombe strains used in this study are listed in Table 2 . Growth media and basic techniques for S . pombe have been described previously ( Alfa et al . , 1993 ) . Stress sensitivity of S . pombe strains was assessed by streaking or spotting on YES agar plates at 37°C or those containing 2 M sorbitol , 1 M KCl or 0 . 1 M CaCl2 at 30°C ( Wang et al . , 2005 ) . Gene disruption and gene tagging with the epitope sequences have been described previously ( Bähler et al . , 1998; Shiozaki and Russell , 1997 ) . QuikChange kit ( Stratagene ) and PrimeSTAR Mutagenesis Basal Kit ( Takara Bio Inc ) were used for site-directed mutagenesis . 10 . 7554/eLife . 19594 . 026Table 2 . S . pombe strains used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 19594 . 026Strain ID Genotype Source , Reference CA101h− leu1–32 Lab stockCA4593h− leu1–32 ura4-D18 ∆tor1::ura4+ ( Kawai et al . , 2001 ) CA4767h− leu1–32 ura4-D18 ∆tor1::ura4+ ∆sin1::kanMX6 ( Ikeda et al . , 2008 ) CA4855h− leu1–32 ura4-D18 ∆tor1::ura4+ ste20:5FLAG ( kanR ) This studyCA5021h− leu1–32 ∆ste20::kanMX6 ( Ikeda et al . , 2008 ) CA5126h− leu1–32 ∆sin1::kanMX6 ( Ikeda et al . , 2008 ) CA5142h− leu1–32 ura4-D18 ∆gad8::ura4+ ( Matsuo et al . , 2003 ) CA5827h- leu1–32 ura4-D18 ∆tor1::ura4+ ∆gad8::ura4+ This studyCA6275h− leu1–32 ste20:5FLAG ( kanR ) ∆sin1::kanMX6 This studyCA6281h− leu1–32 gad8:5FLAG ( kanR ) ( Tatebe et al . , 2010 ) CA6323h− leu1–32 tor1D2137A This studyCA6435h− leu1–32 ste20:13myc ( kanR ) ( Tatebe et al . , 2010 ) CA6530h− leu1–32 FLAG:tor1 ( hph ) ( Hayashi et al . , 2007 ) CA6855h− leu1–32 FLAG:tor1 ( hph ) bit61:13myc ( kanMX6 ) ( Tatebe and Shiozaki , 2010 ) CA6859h− leu1–32 bit61:13myc ( kanMX6 ) ( Tatebe and Shiozaki , 2010 ) CA6870h− leu1–32 FLAG:tor1 ( hph ) ∆sin1::kanMX6 This studyCA6904h− leu1–32 FLAG:tor2 ( kanR ) ( Hayashi et al . , 2007 ) CA6984h− leu1–32 sin1:13myc ( kanMX6 ) ( Tatebe et al . , 2010 ) CA6993h− leu1–32 sin1:13myc ( kanMX6 ) gad8:5FLAG ( kanMX6 ) ( Tatebe et al . , 2010 ) CA7087h− leu1–32 FLAG:tor1 ( hph ) ste20:13myc ( kanMX6 ) ( Hatano et al . , 2015 ) CA7092h− leu1–32 FLAG:tor1 ( hph ) sin1:13myc ( kanMX6 ) ( Hatano et al . , 2015 ) CA7143h− leu1–32 FLAG:tor1 ( hph ) ste20:13myc ( kanMX6 ) ∆sin1::kanMX6 This studyCA7147h− leu1–32 FLAG:tor1 ( hph ) sin1:13myc ( kanMX6 ) ∆gad8::ura4+ This studyCA7150h− leu1–32 FLAG:tor1 ( hph ) sin1:13myc ( kanMX6 ) ∆bit61::ura4+ This studyCA7151h− leu1–32 FLAG:tor1 ( hph ) ste20:13myc ( kanMX6 ) ∆wat1::kanMX6 This studyCA7155h− leu1–32 FLAG:tor1 ( hph ) ste20:13myc ( kanMX6 ) ∆bit61::ura4+ This studyCA7172h− leu1–32 FLAG:tor1 ( hph ) ∆sin1::kanMX6 bit61:13myc ( kanMX6 ) This studyCA7183h− leu1–32 wat1:13myc ( kanMX6 ) This studyCA7189h− leu1–32 ura4-D18 FLAG:tor1 ( hph ) ste20:13myc ( kanMX6 ) ∆gad8::ura4+ This studyCA7200h− leu1–32 FLAG:tor1 ( hph ) sin1:13myc ( kanMX6 ) ∆wat1::kanMX6 This studyCA7213h− leu1–32 FLAG:tor1 ( hph ) wat1:13myc ( kanMX6 ) ( Hatano et al . , 2015 ) CA7217h− leu1–32 FLAG:tor1 ( hph ) ∆ste20::kanMX6 bit61:13myc ( kanMX6 ) This studyCA7222h+ leu1–32 ura4-D18 his7–366 FLAG:tor1 ( hph ) ∆ste20::kanMX6 sin1:13myc ( kanMX6 ) This studyCA7286h− leu1–32 FLAG:tor1 ( hph ) ∆sin1::kanMX6 wat1:13myc ( kanMX6 ) This studyCA7307h− leu1–32 ura4-D18 FLAG:tor1 ( hph ) bit61:13myc ( kanMX6 ) ∆gad8::ura4+ This studyCA7317h− leu1–32 ura4-D18 FLAG:tor1 ( hph ) wat1:13myc ( kanMX6 ) ∆bit61::ura4+ This studyCA7318h− leu1–32 ura4-D18 FLAG:tor1 ( hph ) wat1:13myc ( kanMX6 ) ∆gad8::ura4+ This studyCA7319h− leu1–32 FLAG:tor1 ( hph ) ∆wat1::kanMX6 bit61:13myc ( kanMX6 ) This studyCA7329h− leu1–32 ura4-D18 FLAG:tor1 ( hph ) ∆ste20::kanMX6 wat1:13myc ( kanMX6 ) This studyCA7395h− leu1–32 ( hph ) FLAG:tor1D2137A ( kanMX6 ) ∆sin1::kanMX6 This studyCA7471h− leu1–32 ∆ste20::kanMX6 ∆sin1::kanMX6 This studyCA8070h− leu1–32 psk1:5FLAG ( kanMX6 ) This study To isolate Gad8-interacting proteins by yeast two-hybrid assays , the entire open reading frame of gad8+ was cloned in the pGBT9 vector ( Clontech ) to express the GAL4 DNA-binding domain fused with Gad8 . A fission yeast cDNA library constructed in the pGAD GH vector ( Clontech ) and the budding yeast HF7c strain were used ( Tatebe et al . , 2008 ) . Yeast transformants were screened by histidine auxotrophy and β-galactosidase assay . Yeast two-hybrid screens to isolate Sin1CRIM mutants defective in the interaction with Gad8 was conducted as follows . Randomly mutagenized DNA fragments encoding the fission yeast Sin1CRIM were prepared by standard PCR using premix Ex Taq ( Takara Bio Inc . ) . pGAD GH was linearized by restriction enzyme digestion at two sites in the multi cloning site . The budding yeast HF7c strain carrying pGBT9-gad8+ was transformed with both mutagenized SpSin1CRIM DNA fragments and linearized pGAD GH . Resultant transformants were screened by histidine auxotrophy . GAL4 transcriptional activation domain-SpSin1CRIM fusion in each candidate transformant was examined in immunoblotting with antibodies against GAL4 transcriptional activation domain ( RRID:AB_669111 ) to eliminate all the nonsense mutations . Deficiency in the interaction was verified by recovery and re-transformation of pGAD GH-mutant SpSin1CRIM . Mutation sites were determined by DNA sequencing . pGEX-KG or pMAL-c2x expression plasmids carrying various Sin1 fragments were transformed into BL21 E . coli strain . Exponentially growing BL21 cultures were transferred from 37°C to 16°C , followed by IPTG addition at the final concentration of 0 . 1 mM . Cultures were incubated overnight at 16°C for maximum production of recombinant protein . Cells were harvested by centrifugation and stored at −80°C until use . Frozen cells were thawed in ice-cold TBS ( 20 mM Tris-HCl [pH 7 . 5] , 150 mM NaCl ) and disrupted by sonication . After adding Triton X-100 at the final concentration of 1% , lysate was cleared by 15-min centrifugation at 20 , 800 x g . Supernatant was incubated with glutathione sepharose ( GE Healthcare ) or amylose resin ( New England Biolabs ) for 1 hr , followed by extensive wash with ice-cold TBS containing 1% Triton X-100 . Cell breakage and all the following purification steps were performed at 4°C . Human HEK-293T ( RRID:CVCL_0063 ) and MCF-7 ( RRID:CVCL_0031 ) cells were cultured in DMEM ( Wako ) supplemented with 10% ( v/v ) FBS . HEK-293T ( RCB2202 ) and MCF-7 ( RCB1904 ) were obtained from RCB ( Riken Cell Bank , Japan ) , neither of which we further authenticated . Mycoplasma contamination was examined by DAPI staining and fluorescence microscopy . Transfection was performed with Lipofectamine 2000 ( Life Technologies Japan Ltd . ) or polyethylenimine . FLAG-tagged AKT2 , PKC , SGK1 , and S6K1 are transiently expressed under the control of the CMV promoter of p3xFLAG-CMV-7 . 1 ( Sigma-Aldrich ) . The guide sequence ‘GGACACCGATTTCCCCCCGC’ targeting Exon 6 of the human SIN1 gene was cloned into pX330 ( Addgene #42230 ) ( Cong et al . , 2013 ) . pCAG-EGxxFP was used to examine efficiency of the target DNA cleavage by the guide sequence and Cas9 ( Mashiko et al . , 2013 ) . The resultant plasmid DNA was co-transfected with the puromycin-resistant marker DNA into MCF-7 human culture cells . Forty-eight hours after co-transfection , cells were re-plated in growth media containing 0 . 4 µg/ml puromycin , followed by isolation of drug-resistant colonies . Expression of full length SIN1 protein in each isolated clone was tested by immunoblotting with anti-SIN1 antibodies ( RRID:AB_661901 ) ( Figure 4—figure supplement 6A ) . Genomic DNA flanking the guide and PAM sequence in SIN1 was amplified by PCR and cloned for DNA sequencing ( Figure 4—figure supplement 6B ) . Protein-protein interactions were tested by co-precipitation experiments followed by immunoblot detection . Luminescent Image Analyzer LAS-4000 ( Fujifilm ) was used for quantification in immunoblot . Fission yeast cell lysate was prepared in lysis buffer ( 20 mM HEPES-KOH [pH 7 . 5] , 150 mM potassium glutamate or sodium glutamate , 0 . 25% Tween-20 , 50 mM NaF , 10 mM sodium pyrophosphate , 10 mM p-nitrophenyl phosphate , 10 mM β-glycerophosphate , PMSF , aprotinin , leupeptin and the protease inhibitor cocktail for use in purification of Histidine-tagged proteins [Sigma-Aldrich P8849] ) , followed by centrifugation for 15 min at 20 , 800 x g . The total protein levels of supernatant were quantified by Bradford assay . For interaction between myc-tagged protein and FLAG-tagged protein in fission yeast cells , supernatant was prepared and incubated for 2 hr with anti-FLAG M2-affinity gel ( RRID:AB_10063035 ) or EZview Red anti-c-myc-affinity gel ( RRID:AB_10093201 ) , followed by extensive wash . Resultant samples were subjected to immunoblotting with anti-c-myc ( RRID:AB_631274; RRID:AB_10092917 ) and anti-FLAG ( RRID:AB_10596509; RRID:AB_259529 ) antibodies . For interaction of Gad8FLAG with GST-SpSin1CRIM expressed in fission yeast , GST-SpSin1CRIM expression was induced under the control of the thiamine-repressive nmt1 promoter in a fission yeast gad8:FLAG strain . Supernatant was prepared and incubated with glutathione sepharose ( GE Healthcare ) for 2 hr , followed by extensive wash . GST-SpSin1CRIM and Gad8-FLAG were detected in Coomassie Brilliant Blue ( CBB ) staining and immunoblotting with anti-FLAG M2 antibodies ( RRID:AB_10596509; RRID:AB_259529 ) , respectively . For interaction of Gad8-FLAG with bacterially produced GST-SpSin1CRIM , GST-SpSin1CRIM was first purified onto glutathione sepharose . GST-SpSin1CRIM bound to glutathione sepharose was incubated for 1 hr with supernatant prepared from a fission yeast gad8:FLAG strain . After extensive wash , precipitates were subjected to CBB staining for GST-SpSin1CRIM detection and immunoblotting with anti-FLAG antibodies ( RRID:AB_10596509; RRID:AB_259529 ) for Gad8-FLAG detection , respectively . For interaction of human AGC kinases with HsSin1CRIM , human cell lysate was prepared in lysis buffer ( 20 mM HEPES-KOH [pH 7 . 5] , 150 mM sodium glutamate , 10% glycerol , 0 . 25% tween-20 , PMSF , and the protease inhibitor cocktail for use in purification of Histidine-tagged proteins [Sigma P8849] ) . Bacterially expressed GST-HsSin1CRIM and MBP-HsSin1CRIM were purified onto glutathione sepharose ( GE Healthcare ) and amylose resin ( New England Biolabs ) , respectively , followed by incubation with human cell lysate . The experimental procedure for structure determination of SpSin1CRIM with MTSL spin labels has been reported previously ( Furuita et al . , 2014; Kataoka et al . , 2014 ) . In brief , the cDNA encoding SpSin1CRIM was inserted into pCold-GST vector ( Hayashi and Kojima , 2008 ) . Nine single cysteine mutants were generated using the QuikChange site-directed mutagenesis method ( Stratagene ) . The wild-type and mutant proteins were overexpressed in Escherichia coli RosettaTM ( DE3 ) ( Novagen ) . Following purification , the spin-labeled reagent MTSL [ ( 1-oxyl-2 , 2 , 5 , 5-tetramethyl-∆3-pyrroline-3-methyl ) methanethiosulfonate] was attached to the thiol moiety of the introduced cysteine residues of the mutant proteins . Chemical shift assignments were performed using a conventional method . Dihedral angles were estimated using the program TALOS+ ( Shen et al . , 2009 ) and three bond JC' Cγ and JNCγ coupling constants ( Hu and Bax , 1997 ) . In order to obtain NOE distance restraints , 15N- and 13C-edited NOESY spectra of the wild-type protein were recorded . PRE distance restraints were calculated using intensity ratios of 1H-15N HSQC spectra of MTSL-conjugated mutant proteins in the paramagnetic and diamagnetic states ( Battiste and Wagner , 2000 ) . The structure calculations and automated NOE assignments were performed by CYANA 3 . 95 ( Güntert et al . , 1997 ) , in which PRE distance restraints were used in combination . The obtained structures were refined using Xplor-NIH 2 . 31 ( Schwieters et al . , 2006 , 2003 ) with a single MTSL nitroxide label at each mutated position . The 10 MTSL-conjugated structures ( PDB ID: 2RUJ ) were further refined using Xplor-NIH 2 . 31 . First , MTSL-conjugated cysteines in each structure were mutated back to wild-type residues , and each structure was subjected to energy minimization . Then , using these structures as initial models , structure refinements were performed with NOE distance , PRE distance and dihedral angle restraints . The PRE distance restraints were introduced between Cβ of mutated residues and amide protons with the error of ±7 Å . A total of 100 structures ( 10 structures per initial model ) were calculated , and the 10 lowest energy structures were selected and analyzed . The structural restraints and statistics are summarized in Table 1 . The atomic coordinates of the refined SpSin1CRIM structure have been deposited in the Protein Data Bank with accession code 2RVK . | Human cells contain a group of proteins that together make up a molecular machine called the TOR complex 2 , or TORC2 for short . TORC2 contains an enzyme that changes the activities of other proteins by tagging them with phosphate groups , a process known as phosphorylation . TORC2 has control over many proteins and it can significantly affect the overall behaviour of a cell . Yet it is not known how TORC2 specifically targets these proteins over others in the cell . The components of TORC2 are very similar in yeast and humans , and Tatebe , Murayama et al . have now used a species of yeast called Schizosaccharomyces pombe to examine the proteins in TORC2 . The experiments revealed that one protein called Sin1 is responsible for selecting the targets for TORC2 . Analysis of Sin1 from both yeast and humans identified that the ability to select target proteins is due to a specific part in the middle of this protein . Further experiments revealed that the middle portion of Sin1 has a shape known as a ubiquitin-fold , which is a common feature of proteins that selectively interact with other proteins . More detailed examination of the middle part of Sin1 is still needed to understand why it attaches to some proteins and not others . In particular , TORC2 is known to control a protein called AKT , which plays an important role in some cancers . Understanding how to specifically stop TORC2 from activating AKT could result in new and targeted approaches to treating these cancers . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2017 | Substrate specificity of TOR complex 2 is determined by a ubiquitin-fold domain of the Sin1 subunit |
Escherichia coli single-stranded ( ss ) DNA binding ( SSB ) protein mediates genome maintenance processes by regulating access to ssDNA . This homotetrameric protein wraps ssDNA in multiple distinct binding modes that may be used selectively in different DNA processes , and whose detailed wrapping topologies remain speculative . Here , we used single-molecule force and fluorescence spectroscopy to investigate E . coli SSB binding to ssDNA . Stretching a single ssDNA-SSB complex reveals discrete states that correlate with known binding modes , the likely ssDNA conformations and diffusion dynamics in each , and the kinetic pathways by which the protein wraps ssDNA and is dissociated . The data allow us to construct an energy landscape for the ssDNA-SSB complex , revealing that unwrapping energy costs increase the more ssDNA is unraveled . Our findings provide insights into the mechanism by which proteins gain access to ssDNA bound by SSB , as demonstrated by experiments in which SSB is displaced by the E . coli recombinase RecA .
Escherichia coli single-stranded DNA binding protein ( EcoSSB ) is an essential protein involved in most aspects of genome maintenance ( Meyer and Laine , 1990; Lohman and Ferrari , 1994; Shereda et al . , 2008 ) . It binds with high affinity and little sequence specificity ( Lohman and Overman , 1985; Lohman and Ferrari , 1994 ) to single stranded ( ss ) DNA intermediates formed during DNA replication , recombination , and repair , protecting them from both nucleolytic and chemical damage . SSB also interacts directly with more than a dozen proteins involved in genome maintenance , regulating their access to ssDNA and bringing them to their sites of action ( Shereda et al . , 2008 ) . EcoSSB is one of the most extensively studied ssDNA binding proteins . It consists of four identical subunits ( ∼19 kDa each ) that form a functional tetramer ( Raghunathan et al . , 1997 , 2000 ) ( Figure 1A ) that is stable over a wide range of solution conditions and at sub-nanomolar protein concentrations ( Lohman and Overman , 1985; Bujalowski and Lohman , 1991b ) . Each monomer contains an oligonucleotide/oligosaccharide binding ( OB ) fold that contains the ssDNA binding site ( Raghunathan et al . , 2000 ) . Thermodynamic studies have shown that EcoSSB tetramers bind and wrap ssDNA in a variety of binding modes that differ primarily in the number of OB folds that interact with the tetramer ( Lohman and Ferrari , 1994 ) . Three different binding modes have been identified on poly ( dT ) at 25°C , termed ( SSB ) 65 , ( SSB ) 56 and ( SSB ) 35 , which occlude 65 , 56 , and 35 nucleotides ( nt ) per tetramer , respectively , with a fourth mode observed at 37°C that occludes 40 nt ( Bujalowski and Lohman , 1986 ) . These modes can reversibly interconvert , with the transitions influenced primarily by salt concentration and type as well as protein binding density on the DNA ( Bujalowski and Lohman , 1986 ) . The ( SSB ) 35 mode also binds ssDNA with high cooperativity , forming protein clusters ( Sigal et al . , 1972; Ruyechan and Wetmur , 1975; Lohman et al . , 1986; Kozlov et al . , 2015 ) that may be important during DNA replication ( Lohman et al . , 1988 ) . It has been suggested that SSB utilizes all of these binding modes during its different roles in genome maintenance ( Lohman et al . , 1988 ) and that transitions between modes may control access of other proteins to the ssDNA ( Wessel et al . , 2013; Bhattacharyya et al . , 2014 ) . 10 . 7554/eLife . 08193 . 003Figure 1 . Unwrapping of ssDNA from Escherichia coli SSB under mechanical tension . ( A ) Crystal structure ( Protein Data Bank ID number 1EYG ) and schematic representation of an E . coli SSB tetramer wrapped by 70 nt of ssDNA ( blue ) in the ( SSB ) 65 mode . From 5′ to 3′ , ssDNA interacts with the yellow , purple , green and red subunits . ( B ) Schematic of SSB unwrapping experiment . A DNA construct consisting of two long double-stranded DNA ( dsDNA ) handles and a short ( dT ) 70 ssDNA site is tethered between two optically trapped beads in the absence of SSB ( Position 1 , panel C ) . When moved to the stream containing SSB ( Position 2 ) , a single SSB tetramer binds to the ssDNA site at low tension ( ∼0 . 5 pN ) . The tethered DNA is moved back to the blank stream ( Position 1 ) and a ramping force is applied . Stretching the nucleoprotein complex to >20 pN causes the SSB to dissociate . ( C ) Experimental flow chamber . Two separate streams containing experimental buffer only ( red , Position 1 ) and buffer plus 0 . 5 nM SSB ( blue , Position 2 ) form a laminar interface with minimal mixing . ( D ) Representative force-extension curves ( FECs ) . Relaxing curves ( red ) are obtained after SSB dissociation , and are well fit to a polymer model of bare DNA ( black dotted line , ‘Materials and methods’ ) . Stretching curves ( purple ) of the SSB-ssDNA complex deviate from a model assuming the protein adopts the ( SSB ) 65 wrapping mode at all forces ( black dashed line ) . Cartoon illustration of SSB unwrapping shows the SSB behavior at particular forces . ( E ) Change in extension upon SSB wrapping vs applied force . The change in extension is determined from the extension difference between stretching and relaxing curves in ( D ) . Individual traces ( gray ) are binned and averaged to yield a mean change in extension ( black opened circle; error bars are S . D . ) . The data deviates from the model ( dashed line , determined from the difference between the dashed and dotted lines in ( D ) ) at forces >1 pN . Representative traces ( red , green , and blue ) display the differences between the individual and averaged traces . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 00310 . 7554/eLife . 08193 . 004Figure 1—figure supplement 1 . Dissociation of SSB upon DNA stretching . Averaged stretching ( blue ) and relaxing ( red ) FEC from Figure 1D , and bare DNA FEC ( green ) . Both the relaxing and bare DNA stretching curves are fitted to the polymer elasticity model with 3260 bp dsDNA handles and 70 nt ssDNA ( black dashed line , ‘Materials and methods’ ) . The model assumes zero extension at zero force and fits the data . The resulting fits are consistent with each other , indicating that SSB has dissociated during stretching . Error bars are S . D . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 00410 . 7554/eLife . 08193 . 005Figure 1—figure supplement 2 . Single-stranded DNA polymer modeling . Representative FEC of stretching and relaxing a DNA construct containing 3260 bp dsDNA handles and 70 nt ( green ) or 140 nt ( orange ) ssDNA . The total extension of the tether is modeled by the sum of dsDNA and ssDNA extensions . The dsDNA segment is modeled using the extensible worm-liked chain ( XWLC ) , while the ssDNA segment is fitted to the snake-like chain ( SLC; ‘Materials and methods’ ) . Black dashed and dotted lines are fits to the 70 nt and 140 nt ssDNA constructs , respectively . The extension difference ( inset , blue ) between 70 nt and 140 nt ssDNA constructs illustrates the validity of the ssDNA elasticity model over short lengths ( 70 nt ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 00510 . 7554/eLife . 08193 . 006Figure 1—figure supplement 3 . Dissociation force of SSB-ssDNA . Cartoon schematic and representative traces showing combined fluorescence and DNA extension measurements . A DNA construct bound by fluorescently labeled SSB , SSBf , is stretched ( blue ) and relaxed ( red ) under mechanical force . Upon reaching a force ∼10 pN , SSBf dissociates from the DNA as indicated by the decrease in fluorescence . The relaxing curves from the corresponding FECs match the polymer elasticity model of bare DNA ( black dotted line , ‘Materials and methods’ ) indicating that the SSB has dissociated during stretching . The dissociation force from the FECs is consistent with the fluorescence data . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 00610 . 7554/eLife . 08193 . 007Figure 1—figure supplement 4 . Sample chamber . Image and schematic of a laminar flow chamber . Two glass coverslips are used to sandwich patterned parafilm ( Nescofilm ) . For illustration purposes , food dye of different colors is flowed into the chamber via inlet tubing at a rate of 100 µl/hr . Two streams , one containing experimental buffer only ( red , 1 ) , and the other containing buffer plus SSB ( blue , 2 ) , merge into the central channel but do not mix appreciably due to the laminar flow . The chamber design allows rapid exchange of buffer conditions by moving the optical traps across the stream interface . The top channel ( yellow ) is loaded with anti-digoxigenin beads , while the bottom channel ( green ) is loaded with DNA-bound streptavidin beads . Both beads diffuse through glass capillaries into the middle channel where the optical trapping experiment is performed . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 00710 . 7554/eLife . 08193 . 008Figure 1—figure supplement 5 . DNA construct . Schematic of single-stranded DNA construct . The DNA construct consists of three separate fragments ligated together ( ‘Materials and methods’ ) : ‘Right Handle’ ( RH ) , ‘Left Handle’ ( LH ) , and ‘Binding Site’ ( BS ) . The handles served as functionalized linkers that connect to trapped beads through biotin-streptavidin and digoxigenin-anti-digoxigenin linkages and spatially separate the beads from the protein binding site . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 00810 . 7554/eLife . 08193 . 009Figure 1—figure supplement 6 . SSB binds to dT70 in the fully wrapped ( SSB ) 65 mode at a 1:1 molar ratio in 100 mM Tris buffer . Results of an equilibrium titration of Cy5- ( dT ) 70-Cy3-dT-3′ ( 0 . 1 μM ) with SSB ( left panel; 100 mM Tris-HCl , 20 mM NaCl , 0 . 1 mM EDTA , 25°C ) plotted as normalized Cy5 fluorescence ( Fn = ( F − F0 ) /F0 ) vs molar ratio of total SSB protein ( tetramer ) to total DNA concentrations ( where F0 is the fluorescence intensity of DNA alone and F is the fluorescence measured at each point in the titration ) . The biphasic character of the binding isotherm indicates that two types of complexes can form , the first having one and the second having two tetramers bound and characterized by high and intermediate FRET values ( ( SSB ) 65 and ( SSB ) 35 modes , respectively ) . The continuous line represents the best fit to the data based on a two-site model ( Roy et al . , 2009 ) with equilibrium binding constants , k1 = 1 × 1010 M−1 ( minimum estimate ) and k2 = ( 1 . 21 ± 0 . 04 ) × 108 M−1 and two additional parameters F1 = 10 . 1 ± 0 . 1 and F2 = 4 . 8 ± 0 . 1 , reflecting the maximum Cy5 fluorescence observed for one and two tetramers bound , respectively . Species distribution predicted from the best fit parameters listed above ( right panel ) . At low concentration of SSB tetramers the protein binds to dT70 exclusively in the fully wrapped ( SSB ) 65 binding mode , although as the SSB concentration increases ( [SSB]tot/[dT70]tot > 1 ) the ( SSB ) 35 binding mode starts to form in which two SSB tetramers are bound to one molecule of dT70 . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 009 Crystallographic studies of a C-terminal truncation of the SSB tetramer ( SSBc ) with two molecules of ( dC ) 35 bound suggest a model for the ( SSB ) 65 mode in which 65 nt of ssDNA wrap around an SSB tetramer in a topology resembling the seams on a baseball ( Raghunathan et al . , 2000 ) ( Figure 1A ) . Based on this structure , a model for the ( SSB ) 35 mode has also been proposed ( Raghunathan et al . , 2000 ) . Less is known about the wrapping configurations of the other binding modes , especially the ( SSB ) 56 mode that has only been detected on long poly ( dT ) ssDNA ( Bujalowski and Lohman , 1986 ) . However , various techniques such as electron microscopy ( Chrysogelos and Griffith , 1982; Griffith et al . , 1984 ) , SSB fluorescence quenching ( Lohman and Overman , 1985; Bujalowski and Lohman , 1986 , 1989a , 1989b; Lohman et al . , 1986 ) and sedimentation ( Bujalowski et al . , 1988 ) have provided some basic constraints . Recent single-molecule studies have provided new insights on SSB-ssDNA complex dynamics . Single-molecule FRET ( smFRET ) measurements characterized transitions between binding modes ( Roy et al . , 2007 ) and established that EcoSSB tetramers can diffuse along ssDNA ( Roy et al . , 2009 ) by a reptation mechanism ( Zhou et al . , 2011 ) . Force spectroscopy approaches have also proven useful in studying single-stranded DNA binding protein interactions with DNA ( Pant et al . , 2005; Shokri et al . , 2006; Hatch et al . , 2007 , 2008 ) . Force not only adds another variable to perturb protein-DNA interactions but also provides a well-defined reaction coordinate to quantify the energy landscape governing those interactions . Using a combination of optical traps and smFRET , Zhou et al . ( 2011 ) showed that force gradually unravels ssDNA from EcoSSB and proposed that the energy landscape for SSB-ssDNA interactions is smooth , with few barriers to unwrapping . Here , we present direct observations of a single EcoSSB tetramer interacting with ssDNA using force spectroscopy combined with single-molecule fluorescence microscopy . Applying mechanical force to destabilize the SSB-ssDNA complex and facilitate transitions between binding modes , we show that the ssDNA exhibits discrete wrapping states consistent with the known ( SSB ) 65 , ( SSB ) 56 and ( SSB ) 35 binding modes . Our results are compatible with putative models of the ( SSB ) 35 structure ( Raghunathan et al . , 2000 ) and reveal a likely wrapping configuration for the ( SSB ) 56 mode . SSB- ( dT ) 70 complexes exhibit reversible force-induced transitions between modes without dissociation and SSB can diffuse along ssDNA in the different binding modes , indicating a highly dynamic complex . The data also reveal details of the energy landscape for SSB-ssDNA interactions . In contrast to previous suggestions ( Zhou et al . , 2011 ) , the landscape contains multiple barriers between discrete wrapping conformations , suggesting a distinct wrapping pathway for EcoSSB . Moreover , the energy density is unbalanced , such that the energy cost of unwrapping increases as ssDNA is unraveled from its ends . These findings along with studies of the competition between E . coli SSB and the RecA recombinase protein demonstrate how SSB bound in its different modes might regulate accessibility to ssDNA of other genome maintenance proteins .
We used dual trap optical tweezers to stretch a SSB-ssDNA complex mechanically . As shown in Figure 1B , two trapped functionalized micron-sized beads were tethered together by a DNA construct consisting of a 70-nt poly ( dT ) ssDNA segment flanked by two long double-stranded DNA ( dsDNA ) ‘handles’ ( ‘Materials and methods’ ) . The length of the ssDNA was chosen to accommodate one SSB tetramer in its ( SSB ) 65 binding mode . We also worked under salt conditions and protein concentrations known to favor the ( SSB ) 65 mode in the absence of mechanical tension ( Bujalowski and Lohman , 1986; Roy et al . , 2007 ) ( ‘Materials and methods’ ) . Force-extension curves ( FECs ) of this construct in the absence of protein ( Figure 1—figure supplement 1 , green ) were in excellent agreement with theoretical models of DNA elasticity ( ‘Materials and methods’; Figure 1—figure supplement 1 , black dashed line ) . The total extension of the ‘bare’ DNA molecule , xbare , is given by the sum of the extensions of the dsDNA handles and the ssDNA binding site at a tension F: ( 1 ) xbare ( F ) =ξds ( F ) ·Nds+ξss ( F ) ·Nss , where ξds ( F ) and ξss ( F ) are the extension of one dsDNA base pair and one ssDNA nucleotide given by the extensible worm-like chain ( XWLC , Bustamante et al . , 1994 ) and ‘snake-like chain ( SLC ) ’ model ( Saleh et al . , 2009 ) , respectively ( ‘Materials and methods’; Figure 1—figure supplement 2 ) . Nds = 3260 bp is the total length of the dsDNA handles and Nss = 70 nt is that of the ssDNA loading site . To investigate a single SSB tetramer-ssDNA complex , protein in solution was added to the construct ( ‘Materials and methods’; Figure 1B , C ) for a short period of incubation , allowing one SSB to bind the 70-nt ssDNA . The molecule was then stretched in the absence of free proteins in solution ( Figure 1B , C ) . FECs of stretching and relaxing many molecules are shown in Figure 1D . The stretching FECs ( violet ) of the SSB-DNA complex displayed a shorter extension compared to those without protein due to ssDNA compaction by the SSB . Upon stretching to a force >20 pN and relaxing the molecule , the FECs ( Figure 1 , red ) matched those in the absence of protein ( Figure 1—figure supplement 1 , green ) , indicating that the SSB had dissociated during the stretching process . We confirmed that a single SSB was loaded onto the DNA and dissociated at high force through simultaneous fluorescence detection of dye labeled protein . Using an instrument combining optical traps with a single-molecule fluorescence confocal microscope ( Comstock et al . , 2011 ) , we detected SSB site-specifically labeled with an average of one AlexaFluor555 fluorophore ( SSBf ) as we obtained a FEC ( Figure 1—figure supplement 3; ‘Materials and methods’ ) . The average dissociation force was 10 . 3 ± 0 . 9 pN , consistent with previous reports ( Zhou et al . , 2011 ) . Integrating the area between protein-bound and bare FECs to the force at which the complex spends half its time bound and half unbound yielded a value for the SSB-ssDNA wrapping free energy of 22 ± 2 kBT ( ‘Materials and methods’ ) similar to a previously reported value ( Zhou et al . , 2011 ) . The difference in extension between stretching and relaxing FECs provides information on the SSB-ssDNA wrapping conformation as a function of force . For SSB-bound DNA , we first considered that SSB adopted the canonical ( SSB ) 65 structure ( Raghunathan et al . , 2000 ) . We thus expected a FEC given by Equation 1 with Nss = 70 − 65 = 5 nt due to occlusion by the SSB . As shown in Figure 1D , the stretching FECs ( violet ) diverged significantly from this theoretical model ( black dashed line ) . Figure 1E displays the extension difference , Δx , between the stretching and corresponding relaxing curves as a function of tension F , averaged over many molecules ( N = 36; black points ) , and the corresponding theoretical model ( black dashed line ) . The agreement between model and data at tensions <1 pN is consistent with 65 nt being wrapped around SSB at low forces . Beyond this force , however , Δx is consistently below the prediction , indicating that the SSB wraps <65 nt of ssDNA , in agreement with earlier measurements ( Zhou et al . , 2011 ) . Interestingly , neither the data in Figure 1E nor in those previous studies ( Zhou et al . , 2011 ) provide evidence for discrete wrapping morphologies such as ( SSB ) 56 and ( SSB ) 35 as observed in ensemble studies . If different SSB modes are stable and interconvertible , discrete transitions in the extension would have been expected in the stretching-relaxing experiment . However , detecting intermediates would be possible only if the rate at which the force was ramped was slower than the transitions between intermediates . Moreover , averaging over multiple molecules here and in Zhou et al . ( 2011 ) likely conceals transitions between SSB-ssDNA wrapping intermediates . Example individual traces ( Figure 1E , blue , red , and green curves ) support this view by illustrating the variability among FECs and their divergence from the average behavior ( black ) . Rips in some of these traces ( for example , the red traces at 5 pN ) suggest that SSB may undergo transitions between different wrapping states . To investigate the presence of intermediate wrapping states further , we measured binding of individual SSB tetramers to the ssDNA at constant tension by operating the optical trap in a force-clamp mode ( [Neuman and Block , 2004] , ‘Materials and methods’ ) . As shown in Figure 2A , a DNA construct was initially held in the optical tweezers at a desired constant tension ( 2–10 pN ) and protein was added . After a short time , an SSB binds , and the DNA is compacted upon wrapping . At the end of each observation , protein was dissociated by increasing the tension to a force ( ∼25 pN ) at which SSB cannot remain stably bound . This cycle was repeated numerous times to monitor new protein binding to the same DNA construct . 10 . 7554/eLife . 08193 . 010Figure 2 . Intermediate ssDNA wrapping states of SSB under tension . ( A ) Schematic of SSB constant force wrapping experiment . A DNA construct is held between two optical traps under a constant tension between 2–10 pN in the presence of protein . An extension change , Δx , is measured upon SSB binding , wrapping or unwrapping ssDNA . At the end of each observation , SSB is removed by stretching the DNA construct to high force ( >20 pN ) . ( B ) Representative time traces of SSB-ssDNA wrapping at 2 , 5 , 7 , and 9 pN ( red , green , blue , and purple respectively ) . Extension change data were acquired at 66 kHz and boxcar averaged to 10 Hz ( dark color ) . In all traces , SSB first binds and compacts ssDNA as indicated by an extension decrease . Depending on tension , SSB displays several intermediate wrapping states . Black dashed lines represent the mean extension change of each particular wrapping state . ( C ) Extension change distribution from many SSB wrapping traces at constant tensions between 2–10 pN . The color map matches that in ( B ) . Solid lines are multi-Gaussian fits to the distributions . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 01010 . 7554/eLife . 08193 . 011Figure 2—figure supplement 1 . Single SSB binding and wrapping transitions . Schematic and representative traces illustrating a wrapping experiment with fluorescently labeled SSB , SSBf . A DNA construct is held between two optical traps at a constant tension of 2 , 5 , and 9 pN ( left , middle , and right panels ) . An extension change , Δx , is measured upon SSBf wrapping or unwrapping ssDNA . Upon SSBf binding , a decrease in extension ( gray ) and increase in fluorescence ( green ) are observed simultaneously ( all panels ) . A further decrease in extension ( middle panel ) does not result in further increase in fluorescence , indicating that the same SSB wraps additional ssDNA . At high forces ( right panel ) extension increases correspond to SSB dissociation . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 011 Figure 2B shows the change in DNA end-to-end extension , Δx , upon binding of SSB as a function of force . Using bare DNA as a reference ( set to 0 nm ) , negative extension changes correspond to ssDNA wrapping and positive changes to release of wrapped DNA . At low tensions ( <3 pN ) , we observed that individual SSBs bind and compact ssDNA in a single step ( Figure 2B ) . SSBs remained bound to the ssDNA indefinitely at these tensions . In contrast , at higher tensions , ( 3–8 pN ) , we observed multiple steps upon SSB binding , with dynamic transitions among 2 to 3 distinct states ( Figure 2B , dashed lines ) depending on tension , but no dissociation of SSB . We interpret these dynamic changes in extension as wrapping and unwrapping transitions between intermediate conformations of a single ssDNA-SSB complex . Working at low SSB concentrations ( 0 . 5 nM ) favored the likelihood that multiple SSBs do not bind during one cycle . We corroborated this interpretation with measurements of fluorescently labeled SSBf . Figure 2—figure supplement 1 shows that a single SSB tetramer was responsible for the observed wrapping-unwrapping dynamics . Near the dissociation force ( 9–10 pN ) , we observed multiple instances of one-step wrapping followed by complete release of ssDNA . At these forces , SSB is unable to bind the DNA tether stably , and the observed transitions correspond to protein binding and dissociation . This interpretation is also confirmed by measurements using fluorescent SSBf ( Figure 2—figure supplement 1 , right panel ) , in which dissociation events correlate with loss of fluorescence . Figure 2C shows the combined extension change distributions from many individual SSBs at different tensions . Similarly to the force-ramp results , Δx decreases as tension increases , indicating that the amount of ssDNA wrapped by SSB decreases . However , in contrast to the force-ramp experiment , the constant force experiment provides evidence for intermediate wrapping conformations of SSB , since multiple states are observed at many tensions . The areas under the peaks in the distributions indicate that SSB spends different amounts of time in these particular states . As tension is increased , the SSB-ssDNA complex shifts to states with smaller Δx , corresponding to lower extents of ssDNA wrapping . We considered the possibility that these intermediate DNA wrapping states correspond to the different SSB binding modes observed on poly ( dT ) in ensemble measurements ( Bujalowski and Lohman , 1986 ) . Figure 3A displays the mean extension changes from the peaks of the distributions in Figure 2C . Interpreting these changes in extension , Δx , and attributing these to binding modes required a detailed model . As shown in Figure 3B , ssDNA wrapping by SSB contributes in two ways to the extension of the DNA tether: ( i ) it removes Nw ssDNA nucleotides wrapped by the SSB , and ( ii ) it adds length due to the effective physical size of the SSB-ssDNA complex , xSSBeff , as noted in other mechanical unfolding studies ( Gao et al . , 2012 ) . The extension of the wrapped DNA molecule , xwrap , is thus: ( 2 ) xwrap ( F ) =ξds ( F ) ·Nds+ξss ( F ) · ( Nss−Nw ) +xSSBeff ( Nw , F ) . 10 . 7554/eLife . 08193 . 012Figure 3 . SSB wrapping modes . ( A ) Mean change in extension Δx vs tension for each wrapping state , derived from the peaks of the distributions in Figure 2C . Error bars represent S . E . M . and were determined by bootstrapping . The dashed line is the model in Figure 1D . Solid lines represent models of Δx based on Equation 3 for SSB wrapping Nw = 65 , 56 , 35 , and ∼17 nt ( purple , blue , green , and red , respectively; ‘Materials and methods’ ) . Data points are clustered into 4 groups corresponding to those states ( purple , blue , green , and red circles ) . ( B ) Schematic representation of Δx . Top: Bare ssDNA ( with Nss = 70 nt ) and its extension , xbare , based on a polymer elasticity model Equation 1 ( ‘Materials and methods’ ) . Bottom: SSB-wrapped ssDNA showing the number of wrapped nucleotides , Nw ( <70 , red ) and the remaining unwrapped nucleotides ( Nss − Nw , blue ) . The extension of wrapped DNA , xwrap is calculated from an elasticity model and the effective physical size of the SSB-ssDNA complex , xSSBeff , Equation 2 ( ‘Materials and methods’ ) . Δx is the difference between xwrap and xbare , Equation 3 . ( C ) Number of wrapped nucleotides Nw vs tension F . Each data point in ( A ) is mapped to Nw using the model described in the text ( ‘Materials and methods’; Figure 3—figure supplement 1 ) . Dotted lines represent the maximum possible range of Nw for each colored group of points based on xSSBeff being <6 . 5 nm ( Figure 3—figure supplement 1 , left panel ) . Dashed lines represent a tighter range of possible Nw for each group of points derived from the SSB-ssDNA structure ( Figure 3—figure supplement 1 , middle panel ) . Error bars represent this range for each individual data point . The shaded areas represent the tightest range of possible Nw for each group based on the ‘hotspot’ analysis described in the text ( Figure 3—figure supplement 1 , right panel ) . The points are the best estimates of Nw from the model . The shaded areas and solid lines in ( C ) map directly to those in ( A ) . Cartoon schematics depict possible wrapping modes corresponding to the 4 groups . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 01210 . 7554/eLife . 08193 . 013Figure 3—figure supplement 1 . SSB wrapping models . Three-level modeling of SSB wrapping configurations . Schematics of SSB , wrapped ssDNA ( blue ) , and the distance between wrapped ends , xSSB ( black arrow; top panels ) . Each extension change data point Δx ( F ) in Figure 3A corresponds to a curve in the space of possible Nw and xSSB , according to Equation 5 ( colored curves , bottom panels ) . The widths of the curves correspond to the error bars in Figure 3A . Selected data points from Figure 3A are displayed ( purple: F = 0 . 8 pN , Δx = 11 nm , blue: 4 pN , 14 nm , green: 7 pN , 10 nm , and red: 9 pN , 7 nm ) . At the first level of modeling ( left panels ) , xSSB is assumed to be limited only by the size of the protein ( i . e . , xSSB < 6 . 5 nm; dark gray shaded area ) . The range of possible Nw corresponding to each selected data point is shown by the colored dotted lines . At the second level ( middle ) , the range of possible xSSB is refined by utilizing the ( SSB ) 65 crystal structure . The end-to-end distance between every pair of nucleotides ni and nj along the ssDNA in the structural model defines a lower and upper bound of xSSB for each Nw ( gray shaded area ) . This , in turn , narrows down the range of possible Nw for each data point ( colored dashed lines ) . At the third level ( right ) , four ‘hotspots’ , residues on each SSB monomer with which nucleotides interact most strongly ( green molecular surfaces in the schematic and green nucleotides ) , are used to refine the estimates for xSSB further . Three regions near the hotspots ( black contours ) are identified and used to calculate xSSB . The numbering ( 1 , 2 , and 3 ) corresponds to the configurations shown in Figure 3—figure supplement 2 . This analysis provides the narrowest estimate for the range of Nw for each data point Δx ( colored bands ) . The best estimates for Nw are obtained from the center of this range ( black dots ) ; these are plotted in Figure 3C vs force . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 01310 . 7554/eLife . 08193 . 014Figure 3—figure supplement 2 . SSB wrapping pathway . Crystal structures and schematics of SSB wrapping ssDNA ( blue ) in different wrapping modes . Each mode illustrates possible wrapping configurations that correspond to the regions , numbered 1 , 2 , and 3 in Figure 3—figure supplement 1 . As tension increases ( from left to right ) , SSB wraps less ssDNA , and the number of hotspots interacting with ssDNA ( green molecular surfaces in structures , black dots in schematics ) decreases . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 01410 . 7554/eLife . 08193 . 015Figure 3—figure supplement 3 . Wrapping modes of SSB mutant . Schematic of wrapping experiment using SSBm , a SSB mutant in which Trp-54 is replaced by Ser-54 . Comparison of extension change distributions between wild-type SSB ( left panels ) and SSBm ( right ) . At the same tensions ( 3–5 pN ) , SSBm wraps less ssDNA than wild-type SSB , and is more likely to wrap 35 nt . The mean number of wrapped nucleotides vs tension was estimated in the same way as for wt SSB ( Figure 3C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 015 The extension change upon wrapping , Δx , is the difference between xwrap and the extension of the bare molecule xbare , given by Equation 1: ( 3 ) Δx ( F ) =ξss ( F ) ·Nw−xSSBeff ( Nw , F ) . xSSBeff accounts for the distance between the two ends of the wrapped ssDNA on the SSB ( Figure 3B ) . This geometrical term depends on the size of the SSB and the geometry of wrapped ssDNA around the protein , and is thus a function of Nw ( and F ) . For example , based on the proposed model for the ( SSB ) 65 structure ( Raghunathan et al . , 2000 ) xSSBeff ( Nw=65 ) <2 nm since the ends of the wrapped ssDNA exit at nearly the same point on the protein ( Figure 1A ) . In the ( SSB ) 35 structural model , however , the ssDNA strand exits at opposite ends of the protein and xSSBeff ( Nw=35 ) is predicted to be ∼5 . 5 nm . xSSBeff must also account for the rotational degree of freedom of the nucleoprotein complex , and only the projection along the direction of the applied force contributes to the extension of the DNA tether . As force F is exerted , a torque is applied on the complex , orienting it along the direction of tension . This effect is modeled by ( 4 ) xSSBeff ( Nw , F ) =xSSB ( Nw ) ·L ( FxSSB/kBT ) , where xSSB is the distance between wrapped ssDNA ends in the protein's frame of reference ( Figure 3B ) and L ( z ) ≡ coth ( z ) − 1/z is the orientation factor , derived from the alignment of a particle undergoing rotational Brownian motion to an external torque ( ‘Materials and methods’ ) . Substituting Equation 4 into Equation 3 provides an expression for the measured extension change Δx at each force F in terms of the SSB-ssDNA configuration parameters Nw and xSSB . Thus , for each data point Δx ( F ) in Figure 3A there exists a set of possible values for the pair Nw and xSSB ( ‘Materials and methods’ ) . Figure 3—figure supplement 1 displays how selected data points from Figure 3A each project onto a curve of allowed values in the space of Nw and xSSB ( colored lines ) . Structural considerations limit the range of possible Nw and xSSB . The fact that xSSBeff can be no greater than the size of the SSB ( i . e . , 0 < xSSB < 6 . 5 nm ) places a restriction on the range of possible values Nw can have for each Δx ( Figure 3—figure supplement 1 left panel , dotted colored lines; Figure 3C dotted colored lines ) . We limited the range of Nw further by utilizing the ( SSB ) 65 structure ( Raghunathan et al . , 2000 ) to restrict the potential geometries of any intermediate wrapping states . By measuring the end-to-end distance between every pair of nucleotides separated by Nw nt along the ssDNA in the structural model , we imposed a lower and upper bound on xSSB at each force F ( Figure 3—figure supplement 1 middle panel , gray contours and shaded area; ‘Materials and methods’ ) . This refined range of possible Nw restricts our observed wrapping intermediates to four bands centered around Nw = ∼65 , 50–60 , 30–40 , and 10–20 nt ( Figure 3C dashed colored lines ) . The first three correspond well with the ( SSB ) 65 , ( SSB ) 56 , and ( SSB ) 35 wrapping states observed at 25°C on poly ( dT ) . A better estimate for xSSB and Nw at each force F was obtained by recognizing that specific amino acid residues within EcoSSB are known to contact the ssDNA . Trp-40 , Trp-54 , Trp-88 and Phe-60 have been shown to play important roles in maintaining protein-DNA stability ( Casas-Finet et al . , 1987; Khamis et al . , 1987; Ferrari et al . , 1997 ) . Crystal structure analysis also implicates Trp-54 and Arg-56 as important in creating pockets of positive electrostatic potential on the SSB surface for ssDNA to bind ( Raghunathan et al . , 2000 ) . Lastly , a DNA density map generated by all-atom molecular dynamics ( MDs ) simulations of SSB ( Maffeo , 2015 ) in solution with free oligonucleotides showed that DNA interacts most strongly to regions on each monomer near residues 54–56 ( Trp-88 and Phe-60 are also located near this region ) ( Figure 3—figure supplement 1 right schematic , residues highlighted in green; ‘Materials and methods’ ) . Based on these results , we identified the Trp-54/His-55/Arg-56 cluster as a ‘hotspot’ , residues on each SSB monomer that may serve as anchor points along the DNA wrapping path on the SSB . Our best estimates for Nw at each force F , shown in Figure 3C ( colored points ) , were obtained by considering the distances between groups of nucleotides near each hotspot ( Figure 3—figure supplement 1 right panel , black contours; ‘Materials and methods’ ) . Our models consistently show that ssDNA unwraps in discrete steps with tension , instead of gradually as proposed previously ( Zhou et al . , 2011 ) . As tension increases from 0–8 pN , the number of wrapped nucleotides decreases in a stepwise manner from 65 to 56 to ∼35 nt ( Figure 3C , purple , blue , and green points , respectively ) , matching very well to the known binding modes . The best estimates for Nw and xSSB also generate models for the ssDNA wrapping conformations for each intermediate ( Figure 3C; schematics and Figure 3—figure supplement 2 ) . Control experiments using an SSB mutant confirm our analysis . Mutation of Trp-54 to Ser was previously shown to disrupt interactions with ssDNA and favor wrapping in the ( SSB ) 35 mode ( Ferrari et al . , 1997 ) . We similarly found that the number of nucleotides wrapped by this mutant was lower than that of the wild type SSB , with Nw = 35 nt being the most probable wrapping conformation over the range of tensions assayed ( Figure 3—figure supplement 3 ) . We next investigated whether the different wrapping states of SSB affect its dynamics on ssDNA , in particular its ability to diffuse . We monitored simultaneously the wrapping state of SSB and its position on ssDNA using the combined optical tweezers-confocal fluorescence microscope . We measured the latter using smFRET between the DNA construct modified with a single acceptor fluorophore ( Cy5 ) at the 5′ ss-dsDNA junction and fluorescent SSBf labeled with an average of one donor fluorophore ( AlexaFluor555 ) ( Figure 4A ) . 10 . 7554/eLife . 08193 . 016Figure 4 . SSB binding modes and diffusion mechanism . ( A ) Schematic of fluorescently labeled SSB , SSBf , ssDNA wrapping experiment . A Cy5-labeled DNA construct is tethered between two optical traps under a constant tension of 5 pN . Upon binding of an AlexaFluor555-labeled SSB , both DNA extension change , Δx , and single-molecule FRET are measured simultaneously . ( B ) Scatter plot of FRET efficiency and Δx . Data ( circles ) are assigned to 4 states ( red ( i ) , blue ( ii ) , black ( iii ) , and green ( iv ) ) based on the value of FRET and Δx . A density map of the combined FRET-extension states overlaid with the scatter plot confirms that the data can be separated into 4 states . Cartoon illustrations of nucleoprotein complexes demonstrate possible SSB wrapping configurations corresponding to the 4 assigned states . ( C ) Representative traces showing combined fluorescence and DNA extension measurements . Change in extension ( top; boxcar averaged to 50 Hz ) and fluorescence ( middle; boxcar averaged to 0 . 5 Hz ) of donor ( SSBf , green ) and acceptor ( Cy5 , red ) are measured simultaneously . Together , FRET efficiency ( bottom; blue ) and extension change ( top; black ) reveal the SSB wrapping states ( i and ii , iii and iv ) and their dynamics ( ssDNA wrapping/releasing and sliding ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 01610 . 7554/eLife . 08193 . 017Figure 4—figure supplement 1 . Mechanism of SSB diffusion . Cartoon illustrations of nucleoprotein complexes diffusing along ssDNA with different proposed mechanisms . Schematic FRET efficiency and Δx displaying multiple transitions between states ( i , ii , iii , iv ) . In a sliding or reptation mechanism , FRET transitions occur independently of changes in wrapping state ( top panel ) . A rolling mechanism involves SSB displacement by wrapping one end of DNA followed by releasing the other ( bottom panel; i → iii → ii or ii → iii → i ) . No examples ( 0 of N = 82 ) of rolling are observed in our experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 017 Upon SSBf binding to ssDNA held at a constant 5 pN tension , we observed transitions between the two wrapping states with Nw = 35 nt and 56 nt , based on the analysis from the previous section . We also observed transitions between two FRET states with high ( E ∼ 0 . 5 ) and low FRET efficiencies ( E ∼ 0 ) corresponding to SSBf positioned at the 5′ ss-dsDNA junction vs the 3′ end , respectively . As shown in Figure 4B , all four combined extension-FRET states could be detected in our data: ‘i’—35 nt wrapping and low FRET , ‘ii’—35 nt wrapping and high FRET , ‘iii’—56 nt wrapping and high FRET , and ‘iv’—56 nt wrapping and low FRET . Inspection of individual time traces revealed cases in which transitions in extension and FRET were correlated . Figure 4C ( left ) shows an example of such a transition from state i → iii → i , in which an SSB in ( SSB ) 35 mode wraps an additional ∼20 nt of ssDNA from the 5′ end into ( SSB ) 56 mode , then releases the same end of DNA . This confirms our interpretation that these changes in extension represent transitions between binding modes . Alternately ( Figure 4C; middle and right ) we observed cases in which FRET transitions occurred independently of changes in wrapping state . The two-state time traces indicate SSB diffusing across the sensitive distance range of smFRET ( about one Förster radius , ∼6 nm = 18 nt [Forster , 1948] ) and support a reptation mechanism for SSB diffusion ( Figure 4—figure supplement 1 ) , as previously proposed ( Zhou et al . , 2011 ) . Diffusion of SSB occurred in both ( SSB ) 35 ( Figure 4C; middle ) and ( SSB ) 56 ( Figure 4C; right ) wrapping modes . We reasoned that the lifetimes of the high FRET states in these traces correspond approximately to the time the protein takes to diffuse by one Förster radius from the ss-dsDNA junction , and estimated a diffusion constant D ≈ 27 nt2/s for the ( SSB ) 35 mode and 15 nt2/s for the ( SSB ) 56 mode . This range of values is consistent with prior reports ( Roy et al . , 2009 ) when accounting for temperature ( ∼23°C in our measurements ) and the expected reduction in D due to the 5 pN tension ( Roy et al . , 2009; Zhou et al . , 2011 ) . We observed no examples ( 0 of N = 82 ) of transitions from state i → iii → ii—wrapping one end of DNA and releasing the other—providing no support for a ‘rolling’ mechanism of diffusion ( Romer et al . , 1984 ) ( Figure 4—figure supplement 1 ) .
Due to its homotetrameric nature , the EcoSSB protein can bind ssDNA in a number of different modes that differ in the number of nucleotides occluded in complexes with long ssDNA ( Bujalowski and Lohman , 1986; Lohman and Ferrari , 1994; Roy et al . , 2007 ) . SSB-ssDNA complexes can transition between these modes in vitro and their stabilities can be modulated by changes in solution conditions ( salt , pH , temperature ) as well as the SSB to DNA ratio . Our experiments show that force can also be used to control the ssDNA wrapping state of EcoSSB . This has revealed stable intermediate states of ( dT ) 70 ssDNA wrapping around a single SSB tetramer that correlate well with the known [NaCl]-induced poly ( dT ) binding modes , ( SSB ) 65 , ( SSB ) 56 , ( SSB ) 35 that have been observed for SSB binding to longer poly ( dT ) ( Lohman and Overman , 1985; Bujalowski and Lohman , 1986 ) . The observation of stable force-induced SSB- ( dT ) 70 intermediates provides new details about the likely wrapping topologies of the different binding modes . Our results are consistent with the ssDNA wrapping topology proposed for the ( SSB ) 65 mode based on a crystal structure ( Figure 3C; schematic , and Figure 3—figure supplement 2 ) ( Raghunathan et al . , 2000 ) . They also suggest that the ( SSB ) 56 mode has ssDNA bound to all four subunits , but with the 3′ terminal ssDNA end unraveled to the nearest hotspot ( Figure 3C; schematic , and Figure 3—figure supplement 2 ) . This model is consistent with studies ( Bujalowski and Lohman , 1989a , 1989b ) suggesting that all 4 monomers of an SSB tetramer interact with ssDNA upon binding a molecule of ( dT ) 56 . At forces in the range of 5–8 pN , we observe between 1 to 3 separate states wrapping 30–40 nt . Our data and analysis are not sensitive enough to ascribe specific wrapping conformations to each . We believe at least two conformations wrapping ∼35 nt are consistent with the observed extension changes , one of which is nearly identical to the proposed ( SSB ) 35 structure ( Raghunathan et al . , 2000 ) ( Figure 3C schematic , and Figure 3—figure supplement 2 ) . Interestingly , prior studies ( Roy et al . , 2007 ) have suggested the existence of an alternate ‘ ( SSB ) 35b’ mode that occludes 35 nt but is structurally distinct from ( SSB ) 35 , consistent with our observations . At tensions >8 pN , we also observed a stable intermediate reflecting ∼17 nt of bound ssDNA ( Bujalowski and Lohman , 1989a , 1989b , 1991b ) . Here , a multitude of wrapping conformations around two monomers is consistent with the data ( Figure 3C schematic , and Figure 3—figure supplement 2 ) . Although fluorescence quenching studies ( Bujalowski and Lohman , 1991a ) suggest that ( dT ) 16 would bind to one monomer of SSB , partial interactions with two monomers in our structural model may sum to those of a monomer . It is possible that near dissociation , wrapping geometries could be more heterogeneous . Prior studies have shown that EcoSSB can bind to ssDNA as short as ( dT ) 8 ( Krauss et al . , 1981 ) . However , we do not observe long-lived intermediates wrapping less than ∼17 nt before SSB dissociation . Analyzing the transitions between wrapping intermediates ( Figure 2B ) reveals that almost every transition ( N = 373 out of 380 total , 98% ) occurs between adjacent wrapping states , that is , between ( SSB ) 56 and ( SSB ) 35 , but never directly between ( SSB ) 56 and ( SSB ) 17 . This suggests a single , linear kinetic pathway for wrapping ( Figure 3—figure supplement 2 , right to left ) and unwrapping ( left to right ) . This proposed pathway is corroborated by measurements of E . coli SSB in competition with RecA for ssDNA . As shown in Figure 5A , B , we first loaded a single SSB tetramer onto ssDNA at a force of 5 pN , where our analysis shows the protein interconverts between the ( SSB ) 56 and ( SSB ) 35 modes . We then added RecA to the complex under conditions favoring polymerization into ssDNA-RecA filaments ( ‘Materials and methods’ ) . ( To prevent polymerization of RecA onto the dsDNA handles , the construct was synthesized with the 70-nt ssDNA loading site flanked by short non-DNA spacers [‘Materials and methods’] ) . In the absence of SSB , RecA extends the construct by ∼10 nm as it fills the ssDNA ( Figure 5—figure supplement 1 ) , consistent with previous reports that ssDNA-RecA filaments are 50% longer than dsDNA ( Hegner et al . , 1999; Galletto et al . , 2006 ) ( ‘Materials and methods’ ) . When RecA is added to ssDNA wrapped by a single SSB , RecA takes longer to polymerize but eventually removes the SSB in a stepwise fashion ( Figure 5C ) . Analyzing the measured extension changes from many measurements ( Figure 5D; ‘Materials and methods’ ) reveals that the SSB is unraveled in discrete steps , corresponding to the same pathway of intermediates , ( SSB ) 35 → ( SSB ) 17 → unbound , as proposed above ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 08193 . 018Figure 5 . Unwrapping of ssDNA from SSB by RecA filament formation . ( A ) Schematic representation of SSB-RecA experiment . A standard DNA construct consisting of a 70-nt single-stranded DNA ( ( dT ) 70 ) fragment was synthesized to contain two internal 18-atom hexa-ethylene-glycol spacers at both ss-dsDNA junctions ( cyan; ‘Materials and methods’ ) . The spacers prevent RecA filament formation onto the dsDNA . The construct is tethered in the presence of SSB ( Position 1 ) . After the SSB binds , the tethered DNA is moved to the stream containing RecA for observation ( Position 2 ) . ( B ) Experimental flow chamber for SSB-RecA experiment . Two separate streams contain experimental buffer plus 0 . 5 nM SSB ( red , Position 1 ) and buffer plus 125 nM RecA and 125 μM ATP-γS ( blue , Position 2 ) . ( C ) Representative time traces showing competition between RecA and SSB on ssDNA ( green , blue , red ) . Transient wrapping-unwrapping of SSB slows down the nucleation of RecA . Formation of RecA filament extends ssDNA ( blue box ) , displaces the SSB , and stops after reaching the spacers at the ss-dsDNA junctions . The dotted lines correspond to the model in ( D ) . ( D ) Extension change distribution of SSB-RecA intermediates at a constant tension of 5 pN ( pink ) obtained from many RecA filament formation time traces ( N = 25 ) . Five states representing SSB-RecA dissociation intermediates are illustrated ( schematics ) and assigned to peaks of the distribution . Extensions corresponding to these states are predicted using polymer models of elasticity ( black dots and dotted lines , ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 01810 . 7554/eLife . 08193 . 019Figure 5—figure supplement 1 . RecA filament formation on modified single-stranded DNA . Schematics and representative time traces showing RecA filament formation experiment . A DNA construct consisting of two long dsDNA handles , a short 70-nt ssDNA site , and two spacers ( cyan , ‘Materials and methods’ ) is held between two optical traps at a constant tension of 5 pN in the blank buffer . The construct is then moved into the buffer stream containing 125 nM RecA and 125 µM ATP-γS . A change in extension , Δx , is measured while RecA polymerizes , extending the ssDNA . Upon reaching the spacers , RecA filament formation stalls . The extension change distribution from many RecA filament formation time traces ( blue , black , green; N = 22 ) are consistent with the polymer elasticity model of bare DNA and RecA-filled DNA ( black dots; ‘Materials and methods’ ) , indicating that RecA has fully polymerized on ssDNA . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 019 The ability to measure the extension of each wrapping state as a function of force also allows us to construct an energy landscape for the SSB-ssDNA complex . Using the extension histograms in Figure 2C , we determined the probabilities of occupying specific wrapping modes at each force , and from these we calculated the free energy differences between modes ( ‘Materials and methods’; for simplicity , we ascribed intermediates with similar Nw to the same wrapping state ) . We also used the lifetimes of each wrapping state and transition probabilities at each force ( Figure 2B ) to estimate the barrier heights between states ( ‘Materials and methods’ ) . Our analysis ( Figure 6 ) shows that the free energy of wrapping into the ( SSB ) 65 mode is 21 ± 1 kBT , in excellent agreement with the area between protein-bound and bare FECs ( 22 ± 2 kBT; Figure 1D ) . Interestingly , this wrapping free energy is not distributed evenly among the 65 nt . Instead , we find that 73% of the energy is concentrated in the first 35 nt wrapped ( energy density = 0 . 44 ± 0 . 02 kBT/nt ) . In contrast , the ( SSB ) 65 and ( SSB ) 56 states are separated by only ∼0 . 7 kBT ( energy density ∼0 . 07 kBT/nt ) . This finding suggests that the last ∼10 nt wrapped are more susceptible to unraveling and thus might be more accessible to other proteins competing for ssDNA . This unbalanced energy density profile may provide a mechanism by which SSB is displaced by the recombinase RecA , which requires a foothold of 6–17 nt to polymerize into filaments ( Joo et al . , 2006; Bell et al . , 2012 ) . We note that in the RecA/SSB competition experiment ( Figure 5 ) , we observe RecA filaments forming only once the SSB transitions to the ( SSB ) 35 mode , granting access to >14 nt of ssDNA . 10 . 7554/eLife . 08193 . 020Figure 6 . Energy landscape of SSB wrapping . Energy landscapes of a single SSB wrapping ssDNA at representative forces reconstructed from extension change probability distributions vs tension ( Figure 2C ) . The potential wells correspond to the stable SSB-ssDNA intermediates ( cartoon schematics ) : ( SSB ) 65 , ( SSB ) 56 , ( SSB ) 35 , ( SSB ) 17 , and unbound , respectively . The energy associated with each intermediate is determined from the occurrence probabilities for each state ( squares , ‘Materials and methods’ ) . The barrier heights and positions ( circles ) are determined from the state lifetimes ( ‘Materials and methods’ ) . In the absence of tension , SSB wraps ssDNA in the ( SSB ) 65 binding mode . Increasing tension ( brown , orange , cyan , purple lines correspond to 0 , 3 , 7 , 9 pN , respectively ) tilts the energy landscape , changes the free-energy difference between wrapping intermediates , and favors different SSB-ssDNA binding modes . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 02010 . 7554/eLife . 08193 . 021Figure 6—figure supplement 1 . Occurrence probability of SSB wrapping intermediates . Extension change distributions ( left panels ) of many SSB wrapping events obtained from force-ramp experiments ( 1 pN ) and constant force experiments ( 2–10 pN ) . Individual wrapping intermediates are analyzed and assigned to corresponding SSB binding modes based on Figure 3C . At all tensions , the probability of each SSB binding modes ( right panels , color bars ) is derived from the area under the distributions . The model ( black circles , ‘Materials and methods’ ) obtained from the energy landscape in Figure 6 matches well with the experimentally derived probabilities . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 02110 . 7554/eLife . 08193 . 022Figure 6—figure supplement 2 . Modeling of transition rates between SSB wrapping intermediates . Unwrapping ( solid circles ) and wrapping ( open squares ) transition rates between different SSB wrapping intermediates vs force . The rates were determined from dwell times and transition probabilities in Figure 2C ( ‘Materials and methods’ ) . The data were fit globally ( unwrapping , dashed line; wrapping , dotted line ) using expressions of the form Equation 11 and Equation 12 using as parameters the three barriers and distances to the transition state G35/56‡ , G17/35‡ , G0/17‡ , x35/56‡ , x17/35‡ , and x0/17‡ ( ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08193 . 022 Our measurements that SSB can diffuse on ssDNA while in different wrapping modes provide insights into how SSBs could be redistributed along ssDNA by other proteins seeking access to ssDNA . The observation of SSB-ssDNA rearrangements without unwrapping or rewrapping ( Figure 4 ) points to a sliding mechanism of diffusion in which ssDNA reptates along the protein , consistent with prior models ( Zhou et al . , 2011 ) . In Figure 5 , we believe RecA polymerization likely slides the SSB to one ssDNA-dsDNA junction prior to unravelling it ( Roy et al . , 2009; Bell et al . , 2012 ) . Interestingly , the data in Figure 4 suggest that diffusion may be faster in the ( SSB ) 35 mode . The transition rates between FRET states are ∼1 . 8× larger in the ( SSB ) 35 mode than in the ( SSB ) 56 mode . The observation that a smaller site size leads to faster diffusion is consistent with reports that human RPA , which covers 30 nt , has a larger diffusion coefficient than EcoSSB in its ( SSB ) 65 mode ( Nguyen et al . , 2014 ) . Previous work has proposed that different wrapping modes may be used selectively in different DNA metabolic processes ( e . g . , replication vs recombination ) ( Sancar et al . , 1981; Lohman et al . , 1988 ) . How and which of these modes are used for particular processes remains unclear , as experimental proof of this proposition has proven difficult to obtain in vitro . We anticipate that the control of SSB wrapping mode by applied force may be a useful experimental tool to test this hypothesis . | The DNA double helix consists of two strands coiled around each other . However , there are many instances when DNA must be separated into its individual strands—for example , when the DNA sequence needs to be copied . These single-stranded structures are highly prone to damage . For protection , the single-stranded DNA can wrap around single-stranded DNA binding ( SSB ) proteins , which also control how other maintenance proteins interact with the DNA . SSB proteins from the bacteria species Escherichia coli wrap single-stranded DNA into a variety of topologies known as binding modes . By using a technique that uses a laser to exert forces on an individual DNA molecule , Suksombat et al . unraveled DNA from a single SSB protein . This revealed that the unraveling occurs in a series of steps that correspond well to the known binding modes . These steps also provide the energies required to unravel the single-stranded DNA . Further experiments showed that SSBs can slide along DNA without having to change their binding mode . The unraveling and sliding mechanisms are likely to be used by other proteins to gain access to DNA coated with SSBs . The next step is to understand how SSBs interact with these other proteins , and how their various wrapping configurations affect this interaction . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Structural dynamics of E. coli single-stranded DNA binding protein reveal DNA wrapping and unwrapping pathways |
The RIG-I-like receptors ( RLRs ) play a major role in sensing RNA virus infection to initiate and modulate antiviral immunity . They interact with particular viral RNAs , most of them being still unknown . To decipher the viral RNA signature on RLRs during viral infection , we tagged RLRs ( RIG-I , MDA5 , LGP2 ) and applied tagged protein affinity purification followed by next-generation sequencing ( NGS ) of associated RNA molecules . Two viruses with negative- and positive-sense RNA genome were used: measles ( MV ) and chikungunya ( CHIKV ) . NGS analysis revealed that distinct regions of MV genome were specifically recognized by distinct RLRs: RIG-I recognized defective interfering genomes , whereas MDA5 and LGP2 specifically bound MV nucleoprotein-coding region . During CHIKV infection , RIG-I associated specifically to the 3’ untranslated region of viral genome . This study provides the first comparative view of the viral RNA ligands for RIG-I , MDA5 and LGP2 in the presence of infection .
Because of increasing global population and worldwide exchanges , infections by RNA viruses are expanding with a huge impact on public health . New infections are emerging almost every year ( pandemic flu , West Nile encephalitis , Severe Acute Respiratory Syndrome , Middle East Respiratory Syndrome , chikungunya , Ebola ) , tropical fevers are expanding ( dengue , zika ) and even previously controlled diseases are re-emerging ( measles , poliomyelitis ) . Although several efficient vaccines against RNA viruses are used in human medicine , we still lack potent therapeutic treatments for the majority of viral infections as well as rational strategies to create efficient new vaccines . Recent progress in our understanding of cellular pathways controlling viral replication suggests that modulating host cell functions early upon viral infection could inhibit a large panel of RNA viruses ( Cheng et al . , 2011; Shirey et al . , 2011; Es-Saad et al . , 2012; Guo et al . , 2012; Lucas-Hourani et al . , 2013 ) . The RIG-I-like receptors ( RLRs ) appear to be located at the frontline of the evolutionary race between viruses and the host immune system ( Vasseur et al . , 2011 ) . These are cellular proteins that detect invasion of viral nucleic acid species inside the cytoplasm and initiate innate immune responses against viral infections that limit virus replication and trigger an adequate adaptive immune response ( Errett and Gale , 2015 ) . RLR signalling operates ubiquitously . Hence , RLRs represent attractive strategies for antiviral therapy and vaccine development . The RLR family of pattern recognition receptors ( PRRs ) is a group of cytosolic RNA helicases that can identify viral RNA as non-self via binding to pathogen-associated molecular pattern ( PAMP ) motifs . To date three RLR members have been identified: RIG-I ( Retinoic acid-Inducible Gene-I ) , MDA5 ( Melanoma-Differentiation-Associated gene 5 ) , and LGP2 ( Laboratory of Genetics and Physiology 2 ) ( reviewed in [Loo and Gale , 2011; Dixit and Kagan , 2013] ) . They share a number of structural similarities including their organization into three distinct domains ( Figure 1A ) : i ) an N-terminal region consisting of tandem caspase activation and recruitment domains ( CARD ) , ii ) a central DExD/H box RNA helicase domain with the capacity to hydrolyze ATP and to bind RNA , and iii ) a repressor domain ( RD ) embedded within the carboxy-terminal domain ( CTD ) . These RNA helicases interact with particular signatures of viral RNA , most of which are still unknown . Upon ligand recognition , RLRs bearing the CARD domain ( MDA5 and RIG-I ) , undergo a conformational change that permits the CARD domain to be recruited and to oligomerize with MAVS either in the peroxisome or the mitochondrion . This activates signalling pathways leading to translocation of the transcription factors IRF3 , IRF7 and NF-kB into the nucleus to initiate expression and secretion of type I IFNs and other proinflamatory cytokines . Secreted type I IFNs bind to their receptors and activate the JAK/STAT signalling pathway inducing the expression of more than 300 IFN-stimulated genes ( ISGs ) bearing IFN-stimulated response elements ( ISREs ) ( de Veer et al . , 2001 ) . If the virus has no means for subverting the interferon pathway , the infected tissue turns into an antiviral state leading to i ) apoptosis of the infected cells , ii ) limited propagation of the virus by the expression of ISGs in neighbouring cells and iii ) generation of a cytokine storm that triggers the specific adaptive immune response as well as favouring immune cell infiltration from the cardiovascular system . 10 . 7554/eLife . 11275 . 003Figure 1 . Rig-I Like Receptor ( RLR ) gene expression in stable cell lines encompassing ST-RLRs . ( A ) Schematic representation of the protein domains for each RLR . Domain boundaries are indicated for human RIG-I , MDA5 , and LGP2 proteins according to interpro ( www . ebi . ac . uk/interpro ) . ( B ) LGP2 , MDA5 and RIG-I mRNA levels in ST-RLR cells . RLR mRNA expression were calculated by relative RT-qPCR using specific probes for LGP2 , MDA5 or RIG-I ( on 100 ng of total RNA ) . Ct were normalized using a specific probe against GAPDH house keeping gene . Percentage of mRNA expression was done by setting HEK293 cells as 100% of gene expression for each probe . Samples were analyzed in triplicates with standard deviation represented on the figure . ( C–F ) Analysis of RLR protein expression in ST-RLR cells and efficiency of tagged RLR purification by affinity chromatography . ST-RLR cell lysates ( INPUT ) were affinity-purified using STrEP-Tactin beads ( OUTPUT ) . Western blot analysis was performed using ( C ) α-STrEP-Tag , ( D ) anti-LGP2 , ( E ) anti-MDA5 or ( F ) anti-RIG-I antibodies . ( G ) IFNβ promoter activity assay in ST-RLR cells . Cells were transfected with pIFNβ-FLuc , pTK-Rluc and either mock , poly ( I:C ) or 5’3P . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 003 Members of the RLR family have been implicated in the recognition of a variety of viruses ( Dixit and Kagan , 2013; Goubau et al . , 2013; Patel and Garcia-Sastre , 2014 ) . RIG-I confers recognition of hepaciviruses and of members of the Paramyxoviridae , Rhabdoviridae , and Orthomyxoviridae families . For example , 5’copy-back defective-interfering ( DI ) genomes produced by numerous negative-sense RNA viruses specifically associate with RIG-I and activate IFN induction ( Strahle et al . , 2006; Baum et al . , 2010; Komarova et al . , 2013; Runge et al . , 2014 ) . MDA5 detects members of the Picornaviridae , Caliciviridae , and Coronaviridae . Flaviviridae , Reoviridae , and Arenaviridae are detected by both MDA5 and RIG-I . LGP2 can act both positively or negatively upon activation by different viruses ( Moresco and Beutler , 2010; Deddouche et al . , 2014 ) . Several DNA viruses have also been reported to activate the RLR pathway , including Herpes simplex virus-1 , Adenovirus , Epstein-Barr virus , Vaccinia virus and Hepatitis B virus ( Choi et al . , 2009; Sato et al . , 2014 ) . In the case of DNA viruses , poly dA:dT DNA sequences trigger IFN responses after RNA polymerase III transcription and detection by RIG-I ( Ablasser et al . , 2009; Chiu et al . , 2009 ) . Surprisingly , intracellular bacteria , Listeria monocytogenes , Legionella pneumophila , Salmonella typhimurium , Shigella flexneri also activate type I IFN responses through RIG-I signalling and Plasmodium RNAs are sensed by MDA5 during malaria infection ( Chiu et al . , 2009; Monroe et al . , 2009; Liehl et al . , 2013; Patel and Garcia-Sastre , 2014 ) . Although human RLRs have recently received considerable attention , to our knowledge , nobody has yet simultaneously explored viral RNA partners that bind the three known RLRs under the same experimental conditions . In addition , only few studies characterised molecular features of the RLR ligands in the presence of viral infection ( Baum et al . , 2010; Deddouche et al . , 2014; Runge et al . , 2014 ) . Thus , it is difficult from existing observations to get a clear picture of i ) the biological ligands for each of the RLRs and ii ) the functional differences between RLRs . To study virus-host RNA-protein interactions during viral infection , we previously developed and validated a high-throughput riboproteomic approach based on One-STrEP-tagged protein affinity purification and next-generation sequencing ( NGS ) ( Komarova et al . , 2013 ) . This protocol allows exploring biologically active macromolecular complexes within infected cells . Here , we applied this method to study viral RNA signatures sensed by RIG-I , MDA5 and LGP2 cytosolic receptors upon infection with both negative- and positive-sense RNA viruses . Such study provides a better understanding of the mechanisms and roles of RLR associated RNAs . It helps to conceive new therapeutic approaches such as modulators of innate immunity or new vaccine adjuvants .
To identify and compare RLRs viral RNA partners upon infection with different viruses , we generated human HEK293 cell lines stably expressing One-STrEP-tagged LGP2 , or MDA5 , or RIG-I proteins ( Figure 1A ) . One-STrEP tag provides a reliable , rapid and efficient protocol for ribonucleoprotein ( RNP ) and protein complexes purification ( Komarova et al . , 2011; 2013 ) . We tagged the RLRs N-terminally since previous studies have shown that adding N-terminal tags to RLRs had no negative effect on functional interactions with their RNA ligands ( Pichlmair et al . , 2009; Rehwinkel et al . , 2010; Zhang et al . , 2013a; Deddouche et al . , 2014 ) . These cell lines were assigned ST-RLR cells ( ST-LGP2 , ST-MDA5 and ST-RIG-I , respectively ) . In addition , a stable cell line ( assigned ST-CH ) expressing the Cherry protein instead of tagged RLRs was generated as a negative control to allow subtraction of non-specific RNA binding . Using RT-qPCR analysis and affinity purification followed by western blot analysis we validated the expression of tagged RLRs by each stable cell line ( Figure 1B–F ) . We then analysed the functional profiles of the ST-RLR cell lines upon transfection with RIG-I and MDA5 agonists by using a classical IFN-β promoter activity assay ( Figure 1G ) . We observed that transfection of ST-MDA5 and ST-LGP2 cells with MDA5 agonist poly ( I:C ) elicited an increased IFN-I response . In contrast , transfection of ST-LGP2 cells with the RIG-I agonist 5’triphosphate-RNA ( 5’3P ) masked the IFN-I response ( Figure 1G ) . These observations are in accordance with previous studies showing that over-expression of MDA5 or simultaneous co-expression of MDA5 and LGP2 synergize type I IFN response upon activation with MDA-specific agonists or upon infection with encephalomyocarditis virus ( EMCV ) ( Bruns et al . , 2013 ) . LGP2 over-expression has also been previously found to inhibit RIG-I signalling ( Moresco and Beutler , 2010 ) . Altogether this demonstrates that the three stable cell-lines that we established express tagged RLRs that are functional and can be purified by affinity chromatography . To test whether the over-expression of RLRs influences the efficiency of RNA virus replication we infected ST-RLR stable cell lines with a representative of negative- or positive-sense RNA virus: measles virus ( MV ) and chikungunya virus ( CHIKV ) , respectively . MV belongs to the family Paramyxoviridae ( order Mononegavirales ) and is often considered as a prototypical member of negative-sense RNA viruses . MV genome is used as a template by the viral polymerase to replicate viral genome and to synthesize mRNA molecules encoding six structural proteins: N , P , M , F , H , L and two non-structural virulence factors: V and C . To determine the kinetics of MV replication in ST-RLR cells we used recombinant MV expressing the Firefly luciferase ( Fluc ) gene from the viral genome ( rMV2-Fluc [Komarova et al . , 2011] ) . We observed that at low multiplicity of infection ( MOI ) MV replication was slightly less efficient in ST-RIG-I cells than in the two other cell lines ST-MDA5 and ST-LGP2 ( Figure 2A ) . MV replication was also monitored by immunostaining of MV nucleoprotein ( MV-N ) and flow cytometry analysis at 24 and 48 hr post-infection ( Figure 2B ) . Again , we observed that viral replication was slightly reduced in cells expressing an additional copy of RIG-I protein , particularly at low MOIs . 10 . 7554/eLife . 11275 . 004Figure 2 . Efficiency of ST-RLR cells infection by negative-sense RNA virus ( MV ) . ( A ) Efficiency of Fluc ( rMV2/Fluc ) expressing MV replication in ST-RLR cells . ST-RLR cells were infected with the rMV2/Fluc ( MOIs: 0 . 05 , 1 , 2 . 5 ) . Luc activity was analyzed 5 , 19 , 24 and 48 hr post-infection . ( B ) Efficiency of MV replication in ST-RLR analyzed by FACS . ST-RLR cells were infected by MV . After 24 and 48 hr , cells were harvested , fixed and stained using an anti-N antibody to measure percentage of N positive cells . Experiments were performed two times and data represent means ± SD of the technical triplicates of the most representative experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 004 CHIKV is a member of the family Togaviridae with a positive-sense single-stranded RNA genome . Non-structural proteins ( NSP1-4 ) are directly translated from the viral genome ( 49S ) at early stage of viral replication , whereas structural proteins C-E3-E2-6K-E1 ( capsid C , envelope glycoproteins E3 , E2 and E1 and the ion channel protein 6K ) are encoded by subgenomic mRNA ( 26S mRNA ) during the late viral cycle . To assess CHIKV replication kinetics in tagged RLR cell lines we used a molecular clone of CHIKV ( CHIKV-Rluc ) at different MOI: 0 . 05 , 1 , 2 . 5 . In this virus the reporter Renilla luciferase ( Rluc ) gene is expressed as an integrated part of the non-structural polyprotein of CHIKV . The CHIKV-Rluc allows monitoring the early stage of viral replication ( Henrik Gad et al . , 2012 ) . CHIKV-Rluc replicated in the 4 cell lines as detected by strong Rluc expression and classical growth kinetics for this virus ( Figure 3A ) . However , similarly to MV we observed that CHIKV replication was reduced at low MOI in ST-RIG-I cells . This observation was confirmed by measuring the accumulation of E2 glycoprotein , a late viral product , in ST-RLR cells during infection with wt CHIKV . Again , viral replication was delayed in cells expressing an additional copy of RIG-I protein ( Figure 3B ) . 10 . 7554/eLife . 11275 . 005Figure 3 . Efficiency of ST-RLR cells infection by positive-sense RNA viruses ( CHIKV ) . ( A ) Replication efficiency of CHIKV-Rluc in ST-RLR cells . ST-RLR cells were infected with a CHIKV-Rluc ( MOIs: 0 . 05 , 1 , 2 . 5 ) . renLuc activity was analyzed 0 , 5 , 8 , 10 , 13 , 19 , 24 , 32 and 40 hr post-infection . ( B ) Efficiency of CHIKV replication in ST-RLR cells analysed by FACS . ST-RLR cells were infected with wt CHIKV ( MOIs: 0 . 05 , 1 , 2 . 5 ) . Immunostaining of the E2 glycoprotein was performed and percentage of positive cells was determined . ( C ) Analysis of early and late steps of CHIKV replication . ST-RLR cells were infected with CHIKV-Rluc or wt CHIKV at an MOI 1 . Rluc activity was measured for the CHIKV-Rluc infection ( left axis ) 6 , 9 , 15 , 18 , 23 and 28 hr post-infection . wt CHIKV-infected cells were harvested 6 , 9 , 15 , 18 and 23 hr post-infection , stained with an antibody recognizing double stranded RNA or E2 ( right axis ) and percentage of positive stained cells was determined by FACS . Experiments were performed two times and data represent means ± SD of the technical triplicates of the most representative experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 00510 . 7554/eLife . 11275 . 006Figure 3—figure supplement 1 . Experimental approaches used to determine early and late steps of CHIKV replication . ST-RLR cells were infected with CHIKV-Rluc or wt CHIKV at an MOI of 1 . Rluc activity was measured for the CHIKV-Rluc infection . wt CHIKV-infected cells were harvested stained with an antibody recognizing double stranded RNA or E2 and percentage of positive stained cells was determined by FACS . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 006 These experiments demonstrate that both negative- and positive-sense RNA viruses replicate in cells expressing additional copies of tagged RLRs . Thus , ST-RLR cells provide suitable conditions to study RNA ligands of RLRs within infected cells . As showed on Figure 2B all three ST-RLRs were efficiently infected with MV ( MOI 1 ) at 24 hr post-infection ( with more than 60% of MV-N positive cells ) . We have previously observed extremely valuable virus-host interaction networks in the same conditions of infection ( Komarova et al . , 2011 ) . Therefore , we choose these conditions to purify RLR proteins complexed with MV RNA ligands . In the case of CHIKV infection , we distinguished early and late stages of virus replication . Indeed , it is believed that at early stage of the viral cycle dsRNA intermediary products are sensed by RLR receptors to activate type I IFN signalling . CHIKV non-structural proteins translation and viral genome replication are performed at early steps of replication . This stage can be monitored using CHIKV-Rluc . We observed Rluc accumulation starting from 8 hr post-infection in the ST-RLR and in ST-CH cells ( Figure 3A and C left axis ) . In addition , dsRNA presence was assessed in cells infected with wt CHIKV by using intracellular staining with an anti-dsRNA antibody and flow cytometry analysis . We observed dsRNA accumulation starting from 10 hr post-infection ( Figure 3C right axis ) . Further , efficient translation of the structural E2 glycoprotein that corresponded to the late stage of CHIKV replication was observed at 15 hr post-infection in all three ST-RLR cell lines ( Figure 3B and C right axis ) . Schematic representation of the three technical approaches is represented ( Figure 3—figure supplement 1 ) . These experiments showed that purification of CHIKV RNA ligands complexed with RLRs should be performed between 10 and 15 hr post-infection with CHIKV at MOI of 1 ( Figure 3C ) . To purify the RNAs associated with RLRs upon infection with negative- or positive-sense RNA viruses , ST-RLRs cells were infected with MV or with CHIKV at MOI1 . Cells were harvested 24 hr ( MV ) and 13 hr ( CHIKV ) post-infection and RLRs together with associated partners were purified from total cell lysates by affinity chromatography ( see Materials and methods ) . The RNA molecules interacting with RLRs or the negative control CH were then extracted and designated as RLR/RNA or the negative control CH/RNA . We tested the capacity of these RNA molecules to induce an IFN-mediated antiviral response . RNA samples extracted from LGP2 , MDA5 , RIG-I or CH were transfected into STING-37 reporter cells that stably express the Fluc gene under the control of a promoter sequence containing five IFN-Stimulated Response Elements ( ISRE ) ( Lucas-Hourani et al . , 2013 ) . We observed that RNA ligands purified from LGP2 and RIG-I receptors in the presence of MV or CHIKV infections had an increased immunostimulatory activity in comparison with MDA5-specific RNA ligands ( Figure 4A and B ) . As expected , RNAs purified from negative control CH provided poor immunostimulatory activity in the same experimental setup as well as RNA purified from RLR complexes in the absence of a viral infection ( Figure 4 ) . These results demonstrate that RNA ligands bound to ST-RLRs are specifically enriched for immunoactive RNA molecules , and that our experimental strategy is sensitive enough for the isolation of RLR-specific RNA ligands for both negative- and positive-sense RNA viruses within infected cells . 10 . 7554/eLife . 11275 . 007Figure 4 . Immunostimulatory activity of RNA-ligands co-purified with ST-RLRs upon infection with MV and CHIKV . ST-RLR cells were infected with MV for 24 hr ( A ) , CHIKV for 13 hr ( B ) or mock infected ( C ) . Total cell lysate was used for total RNA purification ( INPUT ) and for affinity purification of RLR RNA complexes , followed by RNA extraction ( OUTPUT ) . Immunostimulatory activity was assessed by transfection into STING-37 reporter cell lines ( Lucas-Hourani et al . , 2013 ) . Fluc activity was measured and normalised to mock transfected cells , 5’3P was used as a positive control . Experiments were performed two times and data represent means ± SD of the technical triplicates of the most representative experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 007 To further evaluate the requirement of LGP2 , MDA5 and RIG-I for IFN response to MV and CHIKV , we depleted STING-37 cells of LGP2 , or MDA5 , or RIG-I via transduction with lentiviral vectors expressing shRNA against each of these receptors or with a control shRNA . The level of LGP2 , MDA5 , or RIG-I mRNAs in puromycin-resistant transduced cell populations was assessed by quantitative RT-qPCR . Silencing decreased LGP2 , MDA5 and RIG-I levels to 41% , 48% and 32% , respectively ( Figure 5A ) . These cells were assigned STINGshLGP2 , STINGshMDA5 , STINGshRIG-I and STINGshNeg , respectively ( see Materials and methods ) . 10 . 7554/eLife . 11275 . 008Figure 5 . Innate sensing of MV and CHIKV infections by different RLRs . ( A ) LGP2 , MDA5 and RIG-I mRNA levels in STINGshRLR cells . STING-37 reporter cell line was transduced by lentiviral vectors expressing an shRNA directed against either LGP2 ( shLGP2 ) , or MDA5 ( shMDA5 ) , or RIG-I ( shRIG-I ) , or non-silencing ( shNeg ) . qPCR analyses with specific probes against the mRNA of each of the RLRs were performed . Relative mRNA expression was done using GAPDH as reference gene and shNeg as reference sample . ISRE activation in STINGshRLR cells by MV or MVΔV ( B ) or CHIKV ( C ) infection . 5’3P and poly ( I:C ) were used as controls . Results are represented as fold increase of ISRE expression compared to mock infected cells . ( D ) Analysis of STINGshRLR promoter activity in the presence of different loads of CHIKV . MOIs 0; 0 . 001; 0 . 01; 0 . 1 and 1 were used . Fluc activity was measured 13 hr post-infection and normalized by setting MOI 0 as 100% of ISRE background activity for the corresponding cell line . ( E ) RLR-dependent IRF3 phosphorylation upon CHIKV infection . IRF3 phosphorylation was analysed in STINGshRLR cells after CHIKV infection by Western Blot using a specific antibody recognizing the phosphorylated form of IRF3 . Experiments were performed three times and data represent means ± SD of the technical triplicates of the most representative experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 008 Upon MV infection only the cells that had RIG-I receptor silenced were strongly impaired in ISRE promoter activation , while MDA5 and LGP2 silencing did not affect signalling ( Figure 5B ) . This was independent of the MV-V protein , which is known to subvert the RLR response through interaction with MDA5 and LGP2 ( Andrejeva et al . , 2004; Childs et al . , 2013; Motz et al . , 2013 ) , as MV∆V virus deleted of the V protein behave similarly . Upon CHIKV infection , we failed to detect any activation of ISRE promoter in the conditions previously established at 13 hr post-infection ( Figure 5C ) . Further analysis showed that increasing CHIKV MOI reduced Fluc expression under the control of ISRE promoter in all silenced cell lines , even at early time post-infection ( Figure 5D ) . Old-world alphaviruses are known to shutdown transcription in infected cells ( Garmashova et al . , 2006; White et al . , 2011; Akhrymuk et al . , 2012; Bouraï et al . , 2012 ) . However , CHIKV infection induces IRF3 phosphorylation and nuclear translocation ( White et al . , 2011 ) . Therefore , we used western blot analysis of RLR-dependent type I IFN signalling to evaluate IRF3 phosphorylation in CHIKV-infected STINGshRLR cells . Interestingly , we observed that only in cells deficient in RIG-I , IRF3 was no longer phosphorylated ( Figure 5E ) . These data validated the important role of RIG-I in viral sensing for both MV and CHIKV infections . Using this experimental set up we failed to validate LGP2 and MDA5 implication in type I IFN activation . These results contradict the fact that immunoactive RNA ligands were co-purified with LGP2 from either MV- or CHIKV-infected cells and MDA5/RNAs possessed only slight immunostimulatory activity ( Figure 4A , B ) . To clarify this question we proceeded to the NGS of RLR-associated RNAs . To identify the nature of RNA molecules bound to RLRs in an unbiased manner , strand-specific NGS analysis was performed using the Illumina technology . To distinguish between RLR-associated RNA molecules and non-specific binding to the beads , co-affinity purification experiments were performed in parallel on ST-CH cells . All experiments were performed in triplicates . Approximately , 16 million reads of 51 nucleotides in length were obtained from each sequencing sample and were mapped to the viral genome . This gave MV genome coverage ranging from 274X to 2433X . We also performed NGS analysis of total RNA samples . Total RNA results provided a classical example of the transcription gradient , which is the characteristic of the order Mononegavirales . Indeed , MV transcription is initiated from a single promoter at the 3’ end of the genome and at each gene junction the virus RNA-dependent RNA-polymerase either falloff or continue the transcription of the downstream gene ( Cattaneo et al . , 1987; Plumet et al . , 2005 ) . Due to this gradient of transcription , MV promoter-proximal genes are expressed more efficiently than promoter-distal genes . The normalized read counts ( see bellow ) on the MV genome are shown in Figure 6—figure supplement 1 , where only the first position of the reads is taken into account . We performed statistical analysis of reads for each RLR that were enriched in the RLR/RNA samples compared to the CH/RNA samples , but lacked enrichment in total RNA of the RLR sample test compared to total RNA of the CH control sample . For this we performed differential analyses between RLR/RNA and CH/RNA on one hand , and between total RLR RNA and total CH RNA on the other hand . The normalization and differential analyses were performed with R ( Team , 2013 ) and the DESeq2 package . A p-value adjustment was performed according to the Benjamini and Hochberg ( BH ) procedure ( Benjamini and Hochberg , 1995 ) . Positions were considered significantly enriched when their adjusted p-value was lower than 0 . 05 . Importantly , there were no differences except for the ST-RIG-I cells for the total RNA profiles of the ST-RLR cells upon infection with MV ( Supplementary file 1 ) . These differences were taken into consideration upon statistical analysis . The distributions of normalized read counts matching the MV genome sequences were represented along the viral genome with the X axis corresponding to all possible positions on the viral genome , and the Y axis showing the normalized number of reads that begin at that position ( Figure 6A–C ) . 10 . 7554/eLife . 11275 . 009Figure 6 . NGS analysis of specific RLR viral partners purified upon infection with MV . MDA5/RNA ( A ) , LGP2/RNA ( B ) and RIG-I/RNA ( C ) samples were subjected to Illumina strand-specific NGS analysis . Sequencing reads were mapped to the MV genome and only the first nucleotide was retained . Differential analyses were performed between RLR/RNA and CH/RNA on one hand and total RLR/RNA and total CH/RNA on the other hand . The distributions of normalized read counts matching the MV genome are represented along the viral genome with the X axis corresponding to all possible positions on the MV genome , and the Y axis showing the normalized number of reads that begin at that position on the positive ( + ) or the negative ( - ) strand of the genome . Significantly enriched reads are represented in red and non-significantly enriched reads are in blue . N , P/V/C , M , F , H , L are MV genome regions coding for the corresponding proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 00910 . 7554/eLife . 11275 . 010Figure 6—figure supplement 1 . NGS profile of total RNA aligned on the MV genome from ST-CH cells infected with the MV . First position of raw read counts are plotted in a strand specific manner ( light blue=Watson , dark blue=Crick strand ) per position . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 010 The distribution of the normalized sequencing reads in the RLR samples collected during MV infection differed a lot for each of the three cytosolic receptors . Indeed , NGS data analysis showed specific MDA5 association with RNA sequences of positive polarity , which most likely correspond to the MV-N mRNA or a read-through transcript of the MV leader region containing N coding sequence ( Figure 6A ) . LGP2 associated RNA sequences represented only a short part of the N coding region and were also enriched in L gene-derived RNAs with both positive and negative polarities ( Figure 6B ) . Surprisingly , we failed to detect any RIG-I specific reads aligning to the MV genome ( not shown ) . This was unexpected since in the above-presented silencing experiment , RIG-I was the only important of the three RLRs sensor for innate immunity activation upon MV infection ( Figure 5B ) . Thus , we performed a second RIG-I purification and NGS experiment . For this second RIG-I/RNA sequencing the viral genome coverage ranged from 144X to 1856X . Again , we obtain similar read profile with only few short positions along MV genome statistically enriched in the RIG-I/RNA sample ( Figure 6C ) . The failure to detect any specific MV RNA ligands for RIG-I could be explained by the lack of RIG-I-specific MV RNA ligands , or by a deficiency of tagged RIG-I protein for RNA ligands purification . MV 5’ copy-back DI genomes are known interactors of RIG-I . Therefore , we looked for the presence of DI RNA molecules in ST-RIG-I cells infected with MV by using universal 5’ copy-back DI genome specific primers . RT-PCR analysis failed to detect any 5’ copy-back DI RNA in infected ST-RIG-I cells ( Figure 7A ) . Interestingly , when the same cells were infected with MV△V virus deficient in V protein , two DNA fragments of approximately 1 and 4 kb were amplified from total RNA samples using 5’ copy-back DI genome specific primers ( Figure 7A ) . Furthermore , sequencing analysis revealed that the 1 kb RT-PCR amplicon contained all the 5’ copy-back DI-RNA features . Indeed , this 1236 nucleotide-long DI genome included genomic 'trailer' ( Tr ) and the reverse complement of the Tr sequences at the 3’ and 5’ extremities and the exact complementary of the extremities of 201 nucleotides able to hybridize to form a stem-loop structure ( Figure 7—figure supplement 1 ) . Sequencing analysis of the 4-kb amplicon suggested that it could correspond to a mosaic type of MV DI genome ( Marriott and Dimmock , 2010 ) . Further , the immunostimulatory activity of RIG-I/RNA purified from MV△V infected cells demonstrated a significant increase in ISRE stimulation compared to RIG-I/RNA purified from MV infected cells ( Figure 7B ) . Interestingly , the immunostimulatory activity of total RNA purified from MV and MV△V infected ST-RIG-I cells was similar ( Figure 7—figure supplement 2 ) . 10 . 7554/eLife . 11275 . 011Figure 7 . 5’ copy-back DI-genome is specifically associated with RIG-I upon infection with MVΔV . ( A ) RT-PCR amplification of 5’ copy-back DI-genome from cells infected with MVΔV using specific primers . Primers JM396 and JM403 were used for 5’ copy-back DI-genome amplification and JM402 and JM396 – for MV full-length genome amplification ( Shingai et al . , 2007 ) . ( B ) Comparison of immunostimulatory activities of RIG-I associated RNAs purified from MV and MVΔV infected cells . 10 , 5 , 2 and 1 ng of RNA were tranfected into STING-37cells , 5’3P , poly ( I:C ) and IFNβ ( 200 UI/mL ) were used as controls and Fluc activity was measured after 24 hr post-transfection . Experiments were performed two times and data represents means ± SD of the technical triplicates of the most representative experiment . ( C ) RIG-I recognizes 5’ copy-back DI-genome in MVΔV infected cells . ST-RIG-I cells were infected with MVΔV . RIG-I/RNA samples were subjected to Illumina strand-specific NGS . Reads were mapped to the MV genome and only the first nucleotide was retained in the X axis . Normalization and presentation of NGS results as Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 01110 . 7554/eLife . 11275 . 012Figure 7—figure supplement 1 . The 5’ copy-back DI genome of MVΔV . ( A ) Exact sequence of the 5’ copy-back 1 , 236 nucleotide-long DI genome of MVΔV . Nucleotides at the position where viral polymerase resumes synthesis to transcribe the complementary 'stem' structure are indicated . ( B ) Structure of the 5’copy-back DI-genome: The 1236 DI-genome was submitted to RNAfold . The structure is coloured by base-pairing probability . The free energy of the structure is -608 , 81 kcal/mol . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 01210 . 7554/eLife . 11275 . 013Figure 7—figure supplement 2 . Comparison of immunostimulatory activities of total RNA purified from ST-RIG-I cells infected by either MV or MVΔV recombinant viruses . 100 , 10 , 5 , 2 and 1 ng of total RNA were tranfected in STING-37 cells and Fluc activity was measured . Experiments were performed 2 times and data represents means ± SD of the technical triplicates of the most representative experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 013 Since 5’ copy-back DI genomes are known interactors of RIG-I , we used MVΔV infection to verify the efficiency of specific RIG-I ligands purification from ST-RIG-I cells . As above , RIG-I/RNAs were analysed by NGS . Again as a control , we used CH/RNA samples purified in a parallel experiment . As previously , the experiments were performed in triplicate , sequencing reads were mapped to MV genome and statistical analysis was performed . As expected , when normalized read counts were represented along MV genome , a vast majority of reads aligned to the 5’-end of the genome with sequences of both positive and negative polarities ( Figure 7C ) . This 5’ MV genome region perfectly corresponds to the 1236 nts-long 5’ copy-back DI genome characterized above ( Figure 7—figure supplement 1 ) . These data clearly confirm the role of RIG-I in sensing 5’ copy-back DI RNAs and suggest a role of the MV-V protein in controlling the DI genome’s formation . They also demonstrate that ST-RIG-I cell line is a potent tool to purify specific RNA ligands of RIG-I upon infection . To evaluate AU/GC composition of the RNA sequences found to be specifically interacting with the three RLRs upon infection with MV , we performed additional bioinformatic analysis of the NGS data . AU content was calculated in a sliding window of 200 nts with one nucleotide step size and was compared to the mean count within that 200 nt window . Figure 8A , B , C demonstrates the AU composition of RIG-I/MDA5/LGP2 specific RNA partners . Our results show that RIG-I was preferentially binding AU-rich RNA regions of the MV genome , but only when the DI genomes were generated upon infection with MVΔV . Further , using Mfold algorithm we performed in silico analysis of the potential to form RNA secondary structures by the different MV genome regions . We observed that the 3’-end of the genome up to beginning of the M fragment possessed regions with lower Free Energy ( Figure 8D ) . 10 . 7554/eLife . 11275 . 014Figure 8 . In silico analysis of NGS data . ( A , B , C ) AU content of RLR-specific RNA ligands . Number of sequenced reads ( extended to 200nts ) with a given AU content . Significantly enriched reads/positions are represented in orange and non-significantly enriched reads are coloured in blue . ( A ) MDA5 , ( B ) LGP2 , ( C ) RIG-I NGS data . ( D ) Secondary structure analysis of the MV genome . Either 250 ( red ) or 500 ( blue ) nucleotide long MV genome fragments were analysed . ΔG ( free energy ) vs . position on the MV genome is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 014 Similar protocol of NGS analysis of RLR-specific RNA partners was applied in CHIKV-infected ST-RLR cells . Approximately , 16 million reads of 51 nucleotides in length were obtained from each sequencing sample and were mapped to the CHIKV genome . This gave viral genome coverage ranging from 597X to 15363X . We performed the same statistical analysis as for MV samples . Interestingly , this time total RNAs profiles of ST-RLR cells infected with CHIKV were different with respect to the negative control ST-CH cell line ( Supplementary file 1 ) . Again , these differences were taken into consideration upon statistical analysis . The distributions of normalized read counts matching the CHIKV genome sequences were represented along the viral genome with the X axis corresponding to all possible positions on the viral genome , and the Y axis showing the normalized number of reads that begin at that position ( Figure 9 ) . 10 . 7554/eLife . 11275 . 015Figure 9 . Analysis of purified RIG-I-specific RNA partners by NGS upon CHIKV infection . RIG-I/RNA samples were subjected to Illumina strand-specific NGS . Sequencing reads were mapped to the CHIKV genome and only the first nucleotide was retained in the X axis . Normalization and presentation of NGS results as Figure 6 . The start of subgenomic RNA transcription is shown with the black arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 01510 . 7554/eLife . 11275 . 016Figure 9—figure supplement 1 . NGS profile of total RNA aligned on the CHIKV genome from ST-CH cells infected with the CHIKV . First position of raw read counts are plotted in a strand specific manner ( ligth blue=Watson , dark blue= Crick strand ) per position . The start of subgenomic RNA transcription is shown with the black arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 016 First , we studied NGS data obtained for the total RNA samples . Visual inspection of the aligned reads on the CHIKV genome shows an enrichment of the C-E2-E1-coding region corresponding to the CHIKV subgenomic RNA ( Figure 9—figure supplement 1 ) . Indeed , previous studies have shown that alphaviruses subgenomic RNAs are more abundant than genomes in cytoplasm of infected cells ( Strauss and Strauss , 1994; Pushko et al . , 1997 ) . Although the average number of reads obtained for the CHIKV genome in all RLR samples was higher than for MV RLR/RNAs , after the statistical analysis LGP2/RNA and MDA5/RNA samples failed to show any specific enrichments in reads corresponding to the CHIKV genome ( data not shown ) . In contrast , for the RIG-I/RNA sample we observed a specific enrichment in ( + ) sense RNA reads corresponding to the 3’ untranslated region ( 3’-UTR ) of the CHIKV genome ( Figure 9 ) . Thus , our protocol of RLR-associated RNA ligands purification allowed us to observe that each of these cytosolic sensors has a specific RNA profile upon infection with different RNA viruses . We confirmed the relative amount of RLR-associated viral RNA by using qPCR . For this , numerous primers aligning on MV and CHIKV genomes were designed ( Figure 10A and Supplementary file 2 ) . The obtained enrichment in mRNA coding for MV-N protein in MDA5/RNA sample was compared to the CH/RNA sample . This qPCR analysis used three different probes aligning at the beginning and the end of the MV-N mRNA ( Figure 10A ) and validated that MDA5-associated RNAs most likely represent transcripts coding for the MV-N protein ( Figure 10B ) . As negative control three other MV mRNAs coding for M , P and H proteins were looked for and not found to be enriched in MDA5/RNA samples ( Figure 10B ) . We then validated that LGP2 specifically binds the 5’-end of the coding region of the N gene , as only the RNA fragment located the farthest to the 5’-end of the N mRNA ( N1 ) was significantly enriched using qPCR ( Figure 10C ) . Further , we validated the specific interaction of RIG-I with the 3’-end of CHIKV genome . For this , we used a pair of primers detecting only the 49S genomic RNA ( CHIKV2 ) , and three others detecting both the 49S genomic and 26S subgenomic RNAs ( CHIKV5 , 6 , 11 ) of which only CHIKV11 was located at the 3’-end of the genome ( Figure 10A ) . Indeed , only the primer located at the 3’-end of the genome was significantly enriched compared to the other sets of viral specific primers and to a cellular housekeeping gene ( Figure 10D ) . Furthermore , we analysed immunostimulatory activity of RLR-specific regions on MV genome by transfecting in vitro transcribed RNA fragments in A549 epithelial cells permissive to MV . We identified that only sequences at the 5’-end of the N mRNA had an enhanced capacity to activate the IFN response ( Figure 10E ) . Finally , we validated by classical co-immunoprecipitation ( co-IP ) approach the results obtained with our ST-RLR purification technique . For this we infected a typical virus-permissive immune cell model ( acute monocitic leukemia THP-1 cell line ) to perform MDA5-specific RNAs isolation by co-IP and RNA analysis by qPCR . This experiment validated that MDA5 has a predisposition to bind MV-N coding region ( Figure 10—figure supplement 1 ) . 10 . 7554/eLife . 11275 . 017Figure 10 . qPCR analysis of specific RLR RNA signatures from MV- and CHIKV-infected cells and their immunostimulatory activity . ( A ) Locations of qPCR primers on MV and CHIKV genomes . ( B ) MDA5 specific interaction with the N coding region upon MV infection . MDA5/RNA samples were subjected to RT-qPCR analysis with specific primers within N , P , M and H mRNAs . Relative Normalized Fold Enrichments ( NFE ) against the control CH samples are shown . Small window represents comparison of NFE of the amplicons obtained with set of primers located in the N region ( N2 , N4 , N5 ) to the NFE of primers located elsewhere in the genome ( P1 , M1 , H1 ) . Comparisons were performed by a non-parametric Mann Whitney Test . ( C ) LGP2 specific interaction with the 5’-end of the N coding region upon MV infection . LGP2/RNA samples were subjected to RT-qPCR analysis with specific primers within N , P and M mRNAs and housekeeping gene GusB . Relative NFE against the control CH samples are shown . Fold enrichment for different primers were compared by One-Way-ANOVA and a Tukey Multiple Comparison test . ( D ) RIG-I specific interaction with the 3’-end of the CHIKV genome . RIG-I/RNA samples were subjected to RT-qPCR analysis with specific primers along CHIKV genome . Fold enrichment for different primers were compared by One-Way-ANOVA and a Tukey Multiple Comparison test ( *p<0 . 05 ***p<0 . 001 ) ( E ) Immunostimulatory activity of in vitro transcribed RNA fragments corresponding to the RLR-specific regions on MV genome . RNA fragments were synthesized in vitro and transfected in A549 cells . IFNβ mRNA induction was measured by RT-qPCR analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 01710 . 7554/eLife . 11275 . 018Figure 10—figure supplement 1 . co-IP of MV-N mRNA on MDA5 from human monocytes . THP1 cells were infected with MV at an MOI of 2 for 24 hr . MDA5-associated RNA molecules were obtained by co-IP . As a negative control , antibodies against a missing protein ( anti-STrEP-Tag ) were used . qPCR analysis of specific MDA5 RNA ligands was performed as on Figure 10 . The experiment was performed two times and data represent means ± SD of the technical triplicates of the most representative experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11275 . 018 These results are in good agreement with NGS results , confirming the relevance of our high-throughput protocol to study viral RNA signatures sensed by RIG-I , MDA5 and LGP2 cytosolic receptors upon infection with different RNA viruses .
RLRs are located at the frontline of the evolutionary race between viruses and the host immune system . These cytoplasmic proteins detect invasion of viral RNA in the cytoplasm of infected cells and trigger innate immunity . The inflammatory response is necessary for limiting viral replication and initiating a specific adaptive immune response . Most of the studies on RLR specific viral RNA ligands are performed in the absence of a productive virus infection ( for review see [Yoneyama et al . , 2015] ) . These studies are based on transfection within cells of RNAs coming from infected cells , expression of viral-coding sequences from plasmids , or transfection of in vitro transcribed RNA or synthetic RLR agonists such as poly ( I:C ) . Recently , to characterize RNA ligands of RLRs in the presence of active viral infections ( Sendai and Influenza viruses , then MV and EMCV ) , NGS has been applied on RNAs isolated from infected cells by co-IP . In these studies the methodological approach was based on either Photo Activable-Ribonucleoside-enhanced CrossLinking and co-IP ( PAR-CLIP ) or simple co-IP followed by NGS . Using these approaches for Sendai and Influenza viruses only RIG-I specific RNA partners were analysed ( Baum et al . , 2010 ) , for EMCV only LGP2-specific RNA partners were sequenced ( Deddouche et al . , 2014 ) , and for MV RIG-I and MDA5 specific RNA partners were studied simultaneously ( Runge et al . , 2014 ) . The disadvantages of RNA-protein isolation by co-IP approaches are that ( i ) they rely on the availability of high-affinity antibodies against the protein of interest and ( ii ) they often lack an appropriate negative control . Negative control antibodies with different affinities and directed against proteins that are absent from the cells can easily underestimate the background noise generated by co-IP followed by NGS . Also , for the PAR-CLIP methodology , the incorporation of the nucleoside analogue 4-thiouridine ( 4SU ) within RNA during the active cycle of viral replication could change the molecular properties of viral and cellular RNAs , thus creating an artificial environment . To overcome these problems we applied One-STrEP-tag fusion protein affinity chromatography to purify RLRs . For this , we generated three single-clone stable cell lines expressing either tagged RIG-I , or MDA5 , or LGP2 ( Figure 1 ) . In addition , a stable cell line expressing CH protein fused to the same One-STrEP-tag was used as a control for non-specific binding . These tools allow studying the three RLRs in the same setting and to characterize their specific viral RNA ligands in the course of infection . Indeed , we isolated specific RLR RNA ligands from these cells infected with MV and CHIKV , respectively a negative- and a positive-sense RNA virus . Three biological replicates were performed for each RLR and an appropriate negative control ( CH/RNA ) allowed subtracting non-specific binding and to perform rigorous statistical analyses . Despite the fact that these cells over-expressed one RLR , they were susceptible to infection with different RNA viruses ( Figures 2 and 3 ) . By applying multiple functional tests we determined the best conditions for ST-RLR cells infection ( Figure 3C ) . We validated that they were efficient for RLR-specific RNA ligands purification and allowed extraction of immunoactive RNA molecules ( Figure 4 ) . Consistent with previous studies we observed that RIG-I/RNAs provided a stronger immunostimulatory activity in comparison to MDA5/RNA molecules ( Ikegame et al . , 2010; Runge et al . , 2014 ) . In the same experimental set-up we showed that LGP2-specific RNA had a similar immunostimulatory activity to RIG-I/RNAs ( Figure 4 ) . Thus , for the two infections more immunoactive RNA ligands were purified with LGP2 and RIG-I than with MDA5 . For CHIKV infection , specific RNA ligands of cytosolic RLRs have not been established yet . The inflammatory response in non-hematopoietic cells is known to be dependent on MAVS , the adaptor molecule necessary for both RIG-I and MDA5 , however which of the two receptors is involved in sensing CHIKV is still unknown ( Schilte et al . , 2010 ) . We found that RIG-I specifically interacts with the 3’-UTR of CHIKV genome . Despite the higher coverage of the CHIKV genome compared to MV genome in our NGS data , we failed to observe specific enrichment in CHIKV reads on LGP2 and MDA5 receptors . The observation that RIG-I specifically interacts with CHIKV 3’-UTR is very promising ( Figure 9 ) . Indeed , the 3’-UTR of positive-sense RNA viruses has already been shown to be particularly immunoactive ( Saito et al . , 2008; Schnell et al . , 2012; Kell et al . , 2015 ) . CHIKV 3’-UTR contains multiple 21 nucleotide-long repeated sequence elements forming dsRNA structures , which are possible targets for RIG-I ( Zhang et al . , 2013b ) . Interestingly , the 3’-UTR has recently been described as a region particularly sensitive to the evolutionary pressure exerted by mosquito vectors ( Chen et al . , 2013 ) . Thus , we observed that in the mammalian host RIG-I targets the same viral region . To our knowledge this is the first time that RIG-I affinity for the 3’-end of a positive-sense RNA virus is demonstrated in physiological conditions . For MV infection we compared our NGS data with those previously published for RIG-I and MDA5 using the PAR-CLIP coupled to NGS approach ( Runge et al . , 2014 ) . Interestingly our study and the Runge et al . data showed that i ) RLRs preferentially interact with MV-N and MV-L derived RNA , ii ) RIG-I and not MDA5 interacts with viral RNA of both negative- and positive-sense polarities originating from MV 5’ copy-back DI genomes ( Figure 7C , [Runge et al . , 2014] ) . However , we observed several discrepancies . We found that MDA5 and not RIG-I specifically interacted with MV-N coding sequences . RIG-I specific positive- and negative-sense reads only represented 5’ copy-back DI-genome and were equally distributed along the 5’-end of the MV genome ( Figure 7C ) . Similar NGS profile of the 5’ copy-back DI genome on RIG-I was previously observed in the context of Sendai virus infection ( Baum et al . , 2010 ) . In this study only a single population of 546 nts-long 5’ copy-back DI RNA genome has been specifically enriched on RIG-I . Thus , the concordance of our results for RIG-I with Baum et al . study nicely confirms the relevance of our strategy . In the future , a deeper NGS study should be done to explain why the RIG-I/RNA sample collected during MV infection possesses immunostimulatory activity despite the absence of MV RNA ligands specifically bound to RIG-I ( Figure 4A and 6C ) . We propose that self-RNA molecules processed by the action of the antiviral endoribonuclease RNaseL on cellular RNA could be specific RNA ligands that bind RIG-I and induce IFN-β expression . Indeed , RNaseL induces type I IFN via production of short-length RNA molecules of viral and cellular origins by the mechanism that includes interaction with MDA5 and RIG-I ( Malathi et al . , 2007; 2010 ) . In the current experimental setup we used a sample multiplexing that allowed an appropriate analysis of relatively long viral-derived RNA and not cellular RNA molecules . Alternatively , it is possible that MV RIG-I-specific RNA ligands are short length and thus not compatible with the classical Illumina NGS protocol applied in this study . In the future this protocol will be adapted to study small size RLR partners and to study RLR-specific cellular partners . Recently Deddouche et al . have isolated a pertinent MDA5 agonist by co-IP of LGP2/RNA complexes purified from cells infected with EMCV , a positive-sense RNA virus . Using NGS , they mapped this RNA on the L region of the EMCV genome . Numerous functional assays were performed to validate that the L region is a key determinant of the MDA5 stimulatory activity ( Deddouche et al . , 2014 ) . This work supported previous studies showing that in an infected cell , MDA5 and LGP2 positively cooperate with each other ( Venkataraman et al . , 2007; Satoh et al . , 2010; Bruns and Horvath , 2012; Childs et al . , 2013 ) . In vitro studies of MDA5 suggested that MDA5 hypothetically recognizes long dsRNA molecules like poly ( I:C ) ( Kato et al . , 2008 ) . MDA5 binds dsRNA from the stem of the molecule and starts forming filaments towards its extremities ( Peisley et al . , 2011; Wu et al . , 2013 ) . Recent studies of autoimmune disorders have proposed that MDA5 detects aberrant secondary structures formed by self mRNAs ( Liddicoat et al . , 2015 ) . All these are evidences that most certainly MDA5 natural ligands are RNA stem structures embedded within different RNAs . In concordance with this , our results show that MDA5 recognizes the integrity of the MV-N mRNA , certainly through the recognition of stem loop structures . Further , our in silico ( Mfold ) prediction of RNA secondary structures within the different parts of the MV genome suggested that MV-N mRNA had a potential to form stem loop structures ( Figure 8D ) . It has recently been shown that LGP2 can stabilize shorter filaments of MDA5 around dsRNA of 100 nts in length ( Bruns et al . , 2014 ) . Our data further confirmed this MDA5/LGP2 synergy . We demonstrated that upon MV infection MDA5 and LGP2 were binding similar viral RNA ligands . Indeed , these LGP2 specific RNAs were encompassing the MV-N segment ( Figure 6A and B ) . However , while the whole MV-N mRNA was bound by MDA5 , the LGP2-specific reads were mostly localized in the 5’-end of the N region . Furthermore , when transfecting in vitro transcribed RNAs corresponding to the MDA5- and LGP2-specific regions of the MV genome in human epithelial cells , we identified a higher immunostimulatory activity of the 5’-end of the N mRNA compared to RNA fragments of similar size but coming elsewhere from the genome ( Figure 10E ) . These data suggest that , upon MV infection , LGP2 and MDA5 specifically interact with the same RNA agonist and that LGP2 could be involved in stabilizing short MDA5 filaments formed on secondary structures localized at the 5’-end of the MV-N mRNA . In summary , using One-STrEP-RLRs affinity chromatography purification and NGS we provide the first simultaneous visualisation of specific RNA ligands for RIG-I , MDA5 and LGP2 in living cells and in the presence of different RNA viruses . Our results show that each of these cytosolic sensors has its individual RNA profile upon infection with different RNA viruses ( Figures 6 , 7 and 9 ) . Previous studies demonstrated RIG-I specific binding to the AU- and polyU/UC-rich regions of viral genomes ( Saito et al . , 2008; Schnell et al . , 2012; Runge et al . , 2014; Kell et al . , 2015 ) . In our results RLR-specific binding cannot simply be explained by primary RNA composition or AU/GC content of RNA ligands . Indeed , we found RIG-I preferential binding to the AU-rich sequences ( Figure 8C ) , but this binding was observed only when a dsRNA-containing DI genome was produced by the recombinant MV△V . These results suggest that RIG-I recognizes specific RNA secondary structures . MDA5 and LGP2 specificity to AU-rich RNAs was less pronounced ( Figure 8A and B ) . Additional studies should be performed to obtain a clearer view on the similarities and the differences of RNA structures recognized by each of the three RLRs . For this , we are currently applying our One-STrEP-RLR approach to three other RNA viruses as well as to an intracellular bacterium . As a consequence , this work offers new strategies for immune and antiviral therapies by targeting the RLR pathway for the therapeutic control of viral infections , enhancing the immune response for vaccines , and conceiving strategies for immune suppression to control inflammation or specific autoimmune diseases .
HEK-293 ( human embryonic kidney , ATCC CRL-1573 ) , Hela ( human cervix adenocarcinoma epithelial , ATCC CCL-2 ) and A549 cells ( human lung adenocarcinoma epithelial , ATCC CCL-185 ) were maintained in DMEM-Glutamax ( GIBCO , Thermo Fisher Scientific , Waltham , Massachusetts ) supplemented with 10% heat-inactivated FCS ( Invitrogen , Thermo Fisher Scientific , Waltham , Massachusetts ) and 100 U/ml/100 μg/ml of Penicillin-Streptomycin ( GIBCO ) . THP-1 ( monocyte cells , ATCC TIB-202 ) were cultured at 0 . 5–7 × 105 cells/ml in RPMI 1640 ( GIBCO ) containing 10% heat-inactivated FCS , 100 U/ml/100 μg/ml of Penicillin-Streptomycin . For all cell lines mycoplasma contamination testing status was routinely verified ( #30-1012K , ATCC ) . To establish clonal stable cell lines expressing One-ST-RLR , we used a modified pEXPR-IBA105 plasmid carrying GW cassette ( pEXPR-IBA105GW ) kindly provided by Dr . Yves Jacob ( Unité de Génétique Moléculaire des Virus ARN , Institut Pasteur ) . First , the RLR sequences ( LGP2 , MDA5 or RIG-I ) were amplified by standard PCR from a human spleen cDNA library ( Invitrogen ) using specific primers with AttB1 and AttB2 sequences included ( Supplementary file 2 ) . The corresponding DNA fragments were cloned by in vitro recombination into pDONR207 entry vector ( BP reaction , Gateway , Invitrogen ) . RLR genes ( MDA5: AF095844; LGP2: AAH14949 RIG-I: CCDS6526 . 1 ) were finally recombined from pDONR207 to One-STrEP-tag pEXPR-IBA105GW by in vitro recombination ( LR reaction ) . These new plasmids were transfected into HEK-293 cells using JetPrime reagent ( #114–15 , Polyplus Transfection , Strasbourg , France ) . Two days later , culture medium was removed and replaced by fresh medium containing G418 at 500 µg/ml ( #G8168 , SIGMA , St . Louis , Missouri ) . Transfected cells were amplified and subsequently cloned by serial limit dilution . At least 5 clones were screened for each RLR to detect the tagged protein expression by qPCR and Western Blot . For STING37shRLR cell lines , we generated lentiviral vectors using the canonical triple transfection of HEK293T cells by a VSVg envelope , an encapsidation p8 , 74 plasmid kindly provided by Dr . Pierre Charneau ( Zufferey et al . , 1997 ) and a pGIPZ vector plasmids ( Thermo Scientific , Waltham , Massachusetts ) expressing either an shRNA with no target ( RHS4346 , shNeg ) , or targeting LGP2 ( shLGP2 , RHS4430-99166661-V2LHS_116526 ) , MDA5 ( RHS4430-101128286-V3LHS_300657 , shMDA5 ) or RIG-I ( RHS4430-99619609-V2LHS_197176 , shRIG-I ) , all bearing a puromycin selection marker . Vectors were titrated according to manufacturer’s instructions in HeLa cells . ISRE reporter cell line ( STING37 [Lucas-Hourani et al . , 2013] ) was transduced at an MOI of 0 . 3 and 48 hr later puromycin ( 5 μg/ml ) was added to the media to select properly transduced cells . CHIKV 06–49 strain was isolated from the serum of an adult patient with arthralgia/ myalgia in Saint Louis city , Réunion , France in December 2005 ( Schuffenecker et al . , 2006 ) . CHIKV strain 06–49 was titrated on VERO cell by TCID50 ( 50% Tissue Culture Infective Dose ) . Recombinant CHIKV-Rluc which contains the Rluc reporter gene inserted between the non-structural nsP3 and nsP4 proteins has already been described ( Henrik Gad et al . , 2012 ) . Attenuated MV Schwarz vaccine strain ( MV ) and recombinant MV Schwarz expressing Fluc ( rMV2-Fluc ) from an additional transcription unit derived from MV have been previously described ( Combredet et al . , 2003; Komarova et al . , 2011 ) . To prevent V protein expression from MV , a two-step PCR strategy was used to generate MVΔV virus . Two PCR fragments were amplified using MV2313 ( 5’-ATCTGCTCCCATCTCTATGG ) and MV2504 ( 5’-TCTGTGCCCTTCTTAATGGG ) for the first fragment and MV2485 ( 5’-CCCATTAAGAAGGGCACAGA ) and MV3385 ( 5’-AGGTTGTACTAGTTGGGTCG ) for the second fragment . These PCRs introduced a mutation interfering with RNA editing ( UUAAAAAGGGCACAGA native sequence was mutated to UUAAgAAGGGCACAGA ) . The two PCR products were combined in a second PCR reaction using MV2313 and MV3385 primers . The produced mutated HindIII-SpeI MV fragment was moved into the pTM-MVSchwarz plasmid after digestion with the corresponding restriction enzymes . Recombinant virus was rescued as previously described ( Combredet et al . , 2003 ) and named MV△V . Immunoblot analysis was performed to validate the lack of MV-V expression by MVΔV ( data not shown ) . Virus stocks were produced on VERO cells , and titrated by TCID50 . Protein extracts were resolved by SDS-polyacrylamide gel electrophoresis on 4–12% Criterion gels ( BioRad , Hercules , California ) with MOPS running buffer and transferred to cellulose membranes ( GE Healthcare , Little Chalfont , United Kingdom ) with the Criterion Blotter system ( BioRad ) . The following antibodies were used: an anti-STrEP-Tag ( #34850 , Qiagen , Hilden , Germany ) , anti-LGP2 ( NBP1-85348 , Novus , Littleton , Colorado ) , anti-MDA5 ( #5321 , Cell Signaling , Danvers , Massachusetts and AT113 , EnzoLifescience , New York , NY ) , anti-RIG-I ( D14G6 , Cell Signaling ) or monoclonal anti-β-actin antibody ( A5441 , Sigma ) , IRF3 ( #11904 , Cell Signaling ) or phosphor-IRF3 ( ab76493 , Abcam , Cambridge , UK , ) . HRP-coupled anti-mouse ( NA9310V , GE Healthcare ) or anti-rabbit ( RPN4301 , GE Healthcare ) were used as secondary antibodies . Peroxidase activity was visualized with an ECL Plus Western Blotting Detection System ( #RPN2132 , GE Healthcare ) . MV intracellular staining was performed with mouse anti-N mAb ( clone 25 , [Giraudon and Wild , 1981] ) and FITC coupled Goat Anti-mouse Ab ( BD Biosciences , Franklin Lakes , New Jersey ) . For CHIKV , intracellular staining was performed with either FITC-conjugated anti-CHIK . E2 mAB 3E4 ( Bréhin et al . , 2008 ) or anti-dsRNA mAb ( J2-1201 , Scicons , Szirák , Hungary ) followed by anti-mouse-APC antibody ( A865 , Invitrogen ) . STING37shRLR cells ( 2 x 105 per well ) were plated in a 24-well plate and infected with appropriate MOIs . For rMV2-Fluc or CHIKV-Rluc infection , at each time point cells were lysed by Passive Lysis buffer ( E1941 , Promega , Fitchburg , Wisconsin ) , and luciferase activity was measured with Bright-Glow Luciferase assay system ( E2650 , Promega ) or Renilla Glow Luciferase Assay System ( E2720 , Promega ) , correspondingly . For immunostaining , cells were washed twice with phosphate-buffered saline ( PBS ) and 2% foetal calf serum ( FCS ) and then fixed in PBS containing 4% paraformaldehyde . Cells were permeabilized with PermWash buffer ( #554723 , BD Biosciences ) , incubated with the primary antibody at 4°C for 30 min , washed in PermWash and incubated with the secondary antibody . Cells were washed twice with PBS 2%FCS before analysis by flow cytometry using a MACSQuant cytometer ( Miltenyi Biotec , Bergisch Gladbach , Germany ) and analysis was done with the software FlowJo ( vers 7 . 6 ) . ST-RLR cells ( 5x107 ) were infected at an MOI of 1 for 24 hr ( MV ) or for 13 hr ( CHIKV ) . Cells were washed twice with cold PBS and lysed in 6 ml of lysis buffer ( 20 mM MOPS-KOH pH7 . 4 , 120 mM of KCl , 0 . 5% Igepal , 2 mM β-Mercaptoethanol ) , supplemented with 200 unit/ml RNasin ( #N2515 , Promega ) and Complete Protease Inhibitor Cocktail ( #11873580001 , Roche , Penzberg , Germany ) . Cell lysates were incubated on ice for 20 min with gentle mixing every 5 min , and then clarified by centrifugation at 16 , 000 g for 15 min at 4°C . A 250 μl aliquot of each cell lysate was used to perform total RNA purification using TRI Reagent LS ( #T3934 , SIGMA ) . The remaining of cell lysate was incubated for 2 hr on a spinning wheel at 4°C with 200 μl of StrepTactin Sepharose High Performance beads ( # 28-9355-99 , GE Healthcare ) . Beads were collected by centrifugation ( 1600 g for 5 min at 4°C ) and washed twice for 5 min on a spinning wheel with 5 ml of washing buffer ( 20 mM MOPS-KOH pH7 . 4 , 120 mM of KCl , 2 mM β-Mercaptoethanol ) supplemented with 200 unit/ml RNasin and Complete Protease Inhibitor Cocktail . Precipitates were eluted using biotin elution buffer ( IBA ) . RNA purification was performed using TRI reagent LS ( T3934 , SIGMA ) . RNA was dissolved in 80 μl of DNase-free and RNase-free ultrapure water . Extracted RNAs were analyzed using Nanovue ( GE Healthcare ) and Bioanalyser RNA nano kit ( #5067–1511 , Agilent , Santa Clara , California ) . RNA molecules isolated from ST-RLR/RNA complexes were treated for library preparation using the Truseq Stranded mRNA sample preparation kit ( Illumina , San Diego , California , ) according to manufacturer’s instruction . To analyze all RNA species present , the initial poly ( A ) RNA isolation step was omitted . Briefly , the fragmented RNA samples were randomly primed for reverse transcription followed by second-strand synthesis to create double-stranded cDNA fragments . No end repair step was necessary . An adenine was added to the 3'-end and specific Illumina adapters were ligated . Ligation products were submitted to PCR amplification . Sequencing was performed on the Illumina Hiseq2000 platform to generate single-end 51 bp reads bearing strand specificity . Sequenced reads have been cleaned from adapter sequences and low complexity sequences ( see appendix ) . Reads were aligned to the MV Schwarz strain reference genome ( http://www . ncbi . nlm . nih . gov/nuccore/FJ211590 . 1 ) , and to CHIKV49 strain reference genome ( http://www . ncbi . nlm . nih . gov/nuccore/AM258994 ) using bowtie ( version:0 . 12 . 7 , options: -a --best -q -m50 -e50 --chunkmbs 400 [Langmead et al . , 2009] ) . The first position of each read ( taking into account the strandness ) was used for statistical analysis using the KNIME software ( Jagla et al . , 2011 ) . Counts per position where calculated using pileup from Samtools ( Li et al . , 2009 ) . Reads cleaning was performed using version 0 . 10 of clean_ngs ( https://github . com/PF2-pasteur-fr/clean_ngs ) in single-end mode and the following parameters: minLen=15 , maxLen5 3’ threshold = 5’ threshold = 61 ( ASCII value ) . Adapter sequences: TGGAATTCTCGGGTGCCAAGG; TGGAATTCTCGGGTGCCAAGGAACTCCAGTCACNNNNNNATCTCGTATGCCGTCTTCTGCTTG; parameters for both sequences: Identity 0 . 85 ( min identity between adapter and read ) , QC threshold= 61 ( ASCII value , bases with values inferior are considered as a match to the adapter , remove as many adapter as possible ) min overlap=7; truncate adapter = 0 ( don’t truncate adapter at 5’-end ) ; leader sequence = 0 ( use adapter sequence as a regular adapter at the 3’-end ) . Statistical analyses were performed with R version 3 . 0 . 2 ( Team , 2013 ) and bioconductor packages . Normalization and differential analyses were carried out between RLR/RNA and Ch/RNA on one hand , and between total RLR RNA and total CH RNA on the other hand . In both cases the normalization was performed with DESeq2 version 1 . 2 . 1 with the normalization parameter locfun="shorth" and default parameters otherwise ( Love et al . , 2014 ) . For the differential analysis we used DESeq2 with default settings . Raw p-values were adjusted according to the BH procedure ( Benjamini and Hochberg , 1995 ) . Adjusted p-values were considered significant if they were lower than 0 . 05 . Positions of interest were those enriched in reads in RLR/RNA compared to CH/RNA , and not enriched in the total RNA comparison . AU content was calculated in a sliding window of 200 nts ( step size=1 ) and compared to the mean count within the 200 nt window and to the sum of sequenced reads ( extended to 200 nt ) . Viral RNA sequences were subjected to secondary structure prediction using Mfold software ( version 3 . 6 , MAX=1 ) using standard parameters to all sliding windows ( window size= ( 250 and 500 ) ; step size=1 ) from the reference genome . The copy-back DI-RNAs were amplified from RNA extracted from total RNA extracted from ST-RIG-I using two sets of MV primers ( Shingai et al . , 2007 ) : for DI-RNAs , 396 ( A: 5’-TATAAGCTTACCAGACAAAGCTGGGAATAGAAACTTCG ) and 403 ( C: 5’-CGAAGATATTCTGGTGTAAGTCTAGTA ) and for standard genome , 396 ( A ) and 402 ( B: 5’-TTTATCCAGAATCTCAAGTCCGG ) . Reverse transcription was performed with Super Script III ( 18080–0440–044 , Invitrogen ) and a PCR amplification was achieved with Q5 High-Fidelity 2X Master Mix ( M0494S , NEB , Ipswich , Massachusetts ) . The PCR-amplified product A-C was cloned into pTOPO vector ( Invitrogen ) and sequenced . RLR mRNA expression profile in ST-RLR cells was performed by RT-qPCR on total RNA isolated with the RNeasy Mini Kit ( Qiagen ) . Reactions were performed with 100 ng of RNA using TaqMan RNA-to-Ct 1-Step Kit ( # 4392938 , Applied Biosystems , Waltham , Massachusetts ) and 1 μl of custom TaqMan Gene Expression Assays ( Hs00920075_m1 - for LGP2; Hs01070332_m1 – for MDA5; Hs00204833_m1 – for RIG-I; Hs99999905_m1 for GAPDH; Hs01077958_s1 – for IFN-β ) . We used Applied Biosystem StepOnePlusTM system . Results were normalized using expression levels of the GAPDH housekeeping genes and RLR expression level in HEK293 cells was settled as 100% of gene expression . All the measures were performed in triplicate and analysed by StepOnePlusTM software . For RLR/RNA enrichment profile , RNA was extracted with TRI Reagent LS before or after affinity chromatography purification on StrepTactin Sepharose . Starting from 200 ng of RNA , cDNA synthesis was achieved in 20 µL using the SuperScript VILO cDNA Synthesis Kit ( #1648108 , Life Technologies , Thermo Fisher Scientific , Waltham , Massachusetts ) following manufacturer's recommendations . Reactions were performed on 2 ng of cDNA using Power SYBR® Green PCR Master Mix ( #4367659 , Life Technologies ) . Reactions were performed in a final volume of 20 μl in the presence of 0 , 6 μM of appropriate primers ( Supplementary file 2 ) . MV and CHIKV specific primers were designed using Primer Express Software ( Applied Biosystems ) . RNA Fold Enrichment was calculated according to the following formulas: Fold Enrichment ( FE ) :FE ( Ct ) = 2- ( Ct ( query ) - Ct ( GAPDH ) ) beads - ( Ct ( query ) - Ct ( GAPDH ) ) total Normalized Fold Enrichment ( NFE ) :NFE ( primer ) = RLR . FE ( primer ) mCherry . FE ( primer ) Short 5’3P-bearing RNA molecules were obtained from pCIneo plasmid linearized by XbaI before transcription . T7 transcription reactions were carried out with a T7 RiboMAX Express Large Scale RNA Production System ( # P1320 , Promega ) . RNA was purified using TriLS reagent and analyzed with Bioanalyser RNA nano kit ( Agilent ) . Poly ( I:C ) was from Amersham Biosciences ( # 27-4729-01 , Amersham , UK ) . To determine ST-RLR responsiveness to synthetic ligands , expression of IFN-β was determined by transient transfection of reporter plasmid pIFNβ-Fluc containing the Fluc gene under the control of the IFN-β promoter ( IFN-b-pGL3 [Lin et al . , 2000] ) . For RNA transfection analysis , ST-RLR cells were plated in 24-well plates ( 2 x 105 per well ) . After 24 hr , cells were transfected with JetPrime reagent with pIFNβ-Fluc ( 250 ng/well ) , plasmid harbouring a thymidine kinase ( Tk ) promoter just upstream of the Rluc gene ( pTK-Rluc , 25 ng/well ) and 10 ng of either poly ( I:C ) or 5’3P . After 24 hr , cells were lysed , and the Fluc and Rluc activities were measured in cell lysates using Dual-luciferase Reporter Assay System ( Promega ) according to manufacturer’s instructions . Reporter activity was calculated as a triplicate of the ratio of Fluc activity to the reference Rluc activity . To determine the immunostimulatory activity of RLR RNA partners , STING-37 cells were plated one day before transfection in 24-well ( 2 × 105 cells per well ) . RLR-specific or total RNA ( 10 ng ) was transfected using jetPRIME together with 250 ng of pcINEO empty vector as a carrier . 24 hr post-transfection , cells were lysed with passive buffer and the Fluc activity was measured using the Bright-Glo Luciferase Assay System . Viral sequences were synthesized using specific primers ( Supplementary file 2 ) with the forward primers presenting the T7 promoter sequence ( attgtaatacgactcactataggg ) at the 5’-extremity . Amplicons were purified in 1% agarose gel and used for in vitro transcription . Desired RNA molecules were further separated from the 3’extension by-products ( Triana-Alonso et al . , 1995 ) on 5% TBE-Urea Polyacrylamide Gel ( #345–0086 , Biorad ) . Purified RNAs were analyzed with Bioanalyser RNA pico kit ( Agilent ) . 10 ng of purified RNA was transfected in A549 cells and IFN-β mRNA was quantified by RT-qPCR analysis . Similar to Runge et al . immunoprecipitation protocol was used but with little modifications: without crosslinking and the 20 mM MOPS-KOH pH7 . 4 , 120 mM of KCl , 0 . 5% Igepal , 2 mM β-Mercaptoethanol supplemented with 200 unit/ml RNasin and Complete Protease Inhibitor Cocktail buffer was used for cell lysis and washing steps . | An immune system can protect against disease-causing microbes such as viruses . Human cells contain three different receptors that can recognize and respond when a virus enters and begins to replicate inside . These receptors include RIG-I , MDA5 and LGP2 , and they are collectively known as the RIG-I-like receptors . RIG-I-like receptors specifically recognize viruses that store their genetic material in the form of molecules of RNA . However , the specific viral parts that trigger RIG-I-like receptors to respond remain almost completely unknown . RNA viruses include well-known and re-emerging viruses such as polio and measles , as well as chikungunya – a virus that is spread by mosquitoes and causes illness worldwide . This means that understanding how RIG-I-like receptors identify RNA viruses and then trigger an immune response to eradicate them has the potential to inform the development of vaccines and antiviral therapies for many diseases . Sanchez David et al . now describe and validate a new experimental approach to determine the distinct viral regions that are recognized by human RIG-I-like receptors . The approach involves purifying the RIG-I-like receptors out of infected cells and then working out the sequence of RNA fragments that bind to the receptors . This approach revealed that each of the three human RIG-I-like receptors detected different viral RNA sequences during a measles infection . On the other hand , only RIG-I could recognize a specific part of the chikungunya virus genome . All together , the experiments illustrate how to identify the RNA sequences recognized by any of the three human RIG-I-like receptors during infection by a RNA virus . With the ability to gain this kind of insight , it may soon be possible to develop ways of using the RIG-I-like receptor pathway to control viral infections and enhance the body’s immune response to vaccination . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2016 | Comparative analysis of viral RNA signatures on different RIG-I-like receptors |
The transcription factor STAT5 is fundamental to the mammalian immune system . However , the relationship between its two paralogs , STAT5A and STAT5B , and the extent to which they are functionally distinct , remain uncertain . Using mouse models of paralog deficiency , we demonstrate that they are not equivalent for CD4+ 'helper' T cells , the principal orchestrators of adaptive immunity . Instead , we find that STAT5B is dominant for both effector and regulatory ( Treg ) responses and , therefore , uniquely necessary for immunological tolerance . Comparative analysis of genomic distribution and transcriptomic output confirm that STAT5B has fargreater impact but , surprisingly , the data point towards asymmetric expression ( i . e . paralog dose ) , rather than distinct functional properties , as the key distinguishing feature . Thus , we propose a quantitative model of STAT5 paralog activity whereby relative abundance imposes functional specificity ( or dominance ) in the face of widespread structural homology .
Signal transducers and activators of transcription ( STAT ) family proteins are an evolutionarily conserved set of transcription factors , which operate downstream of cytokine and hormone receptors to convert extracellular stimuli into biochemical signals that instruct gene expression ( Villarino et al . , 2015; Stark and Darnell , 2012 ) . In mammals , STAT5 is unique because it is encoded by two genes , termed Stat5a and Stat5b , derived from a relatively recent duplication event ( Wang and Levy , 2012 ) . In fact , the ancestral STAT5 gene appears to have duplicated on two separate occasions during vertebrate evolution . Once in teleosts , resulting in two paralogs on different chromosomes , and again in eutherians , resulting in the two contiguous mammalian paralogs ( Liongue et al . , 2012 ) . Because of their recent divergence , STAT5A and STAT5B are homologous at the DNA , RNA and protein levels , which has led to persisting questions about whether they are redundant or functionally distinct . Genetically engineered mice lacking Stat5a or Stat5b have provided compelling evidence for both arguments . On one hand , there are phenotypic differences; Stat5a-deficient mice exhibit poor mammary function ( Liu et al . , 1997 ) , reduced hematopoietic stem cell proliferation ( Zhang et al . , 2000 ) and diminished antibody class switching ( Kagami et al . , 2000 ) , while Stat5b-deficient mice exhibit dwarfism ( Udy et al . , 1997 ) , more pronounced lymphopenia , and greater defects in cytokine-driven lymphocyte proliferation ( Moriggl et al . , 1999b; Imada et al . , 1998 ) . On the other hand , deletion of Stat5a and Stat5b has comparable effects on some physiological processes , such as eosinophil recruitment ( Kagami et al . , 2000 ) , and the most dramatic phenotypes , such as infertility , anemia and perinatal lethality , are evident only in mice lacking both paralogs , which implies redundancy and/or cooperativity ( Teglund et al . , 1998; Socolovsky et al . , 1999; Cui et al . , 2004 ) . Genome-wide DNA-binding profiles also support both viewpoints . The target repertoires for STAT5A and STAT5B mostly overlap , which implies redundancy , but there are also a subset of sites that may be differentially bound , which implies specificity ( Liao et al . , 2008; 2011; Yamaji et al . , 2013; Kanai et al . , 2014 ) . Consistent with the latter point , humans with germline mutations in STAT5B exhibit a range of clinical abnormalities , indicating that STAT5A cannot compensate for some vital functions ( Kanai et al . , 2012 ) . Compound STAT5 deficiency manifests striking immunological abnormalities in mice , most notably lymphopenia , splenomegaly and autoimmunity . These are typically attributed to its role downstream of the common gamma chain ( ɣc ) receptor and its dedicated Janus kinase , Jak3 ( Moriggl et al . , 1999b; Snow et al . , 2003; Yao et al . , 2006 ) . The ɣc is shared by 6 different cytokines , IL-2 IL-4 , IL-7 , IL-9 , IL-15 and IL-21 , each of which employs a unique co-receptor subunit that determines which cell types can respond ( Rochman et al . , 2009 ) . ɣc cytokines impact all lymphocytes but have been most extensively studied in CD4+ 'helper' T cells , the key orchestrators of adaptive immunity . Among the many functions ascribed to the ɣc-STAT5 axis in this lineage are the ability to promote Th1- and Th2-type effector responses , to support T cell memory , to promote activation-induced cell death , to suppress Th17-type and T follicular helper cell ( Tfh ) responses , and to promote T regulatory cell ( Treg ) responses ( Moriggl et al . , 1999a; Liao et al . , 2008; 2011; Dooms et al . , 2007; Zhu et al . , 2003; Kagami et al . , 2001; Lenardo , 1991; Laurence et al . , 2007; Ballesteros-Tato et al . , 2012; Johnston et al . , 2012; Mahmud et al . , 2013 ) . To assess redundancy between STAT5 paralogs , we developed a mouse model where STAT5A and/or STAT5B were reduced but not absent , allowing us to compare their respective functions while avoiding the confounding lymphopenia associated with complete STAT5 deficiency . These studies reveal STAT5B as the dominant paralog in helper T cells; exhibiting far greater impact on pathogenic effector and host-protective regulatory responses and , therefore , uniquely required for immunological tolerance . Surprisingly , genome-wide DNA binding and transcriptome surveys did not uncover widespread differences in target gene selection but , instead , point towards relative abundance as the key distinguishing factor . Thus , we propose that asymmetric expression ( i . e . paralog dose ) , rather than differential function , determines the dominant STAT5 paralog in lymphoid cells .
To investigate the relationship between STAT5A and STAT5B , we generated a series of mice with pre-determined combinations of Stat5 alleles , ranging from two alleles each of A and B ( 4 total ) to one allele of either A or B ( Figure 1A ) ( Yamaji et al . , 2013 ) . We refer to each genotype according to the total number of Stat5 alleles that are retained . For example , two-allele Stat5a-deficient mice lack both Stat5a alleles but retain two of Stat5b ( Stat5a-/- Stat5b+/+ ) , while one-allele Stat5a-deficient mice lack both Stat5a alleles and retain just one of Stat5b ( Stat5a-/- Stat5b+/- ) . All 8 genotypes were born at the expected Mendelian ratios and survived beyond 6 months of age , thereby demonstrating that a single allele of either paralog is sufficient to prevent the perinatal lethality seen in STAT5-null mice ( Data not shown ) ( Hoelbl et al . , 2006; Cui et al . , 2004 ) . Red blood cell counts and hematocrits were comparable across all genotypes , indicating that a single allele is also enough to support erythropoiesis , but white blood cell ( WBC ) counts were sharply reduced in one-allele Stat5a- or Stat5b-deficient mice , as well as two-allele Stat5b-deficient mice . By contrast , two-allele Stat5a-deficient mice had relatively normal WBC counts ( Figure 1B ) . 10 . 7554/eLife . 08384 . 003Figure 1 . Stat5b is required for immunological tolerance . ( A ) Cartoon depicts the mutant mice used in this study . Genotypes are grouped according to total Stat5 alleles . ( B ) Bar graphs show averaged RBC , hematocrit and WBC counts . ( C ) Scatter plot shows kidney pathology scores . ( D ) Scatter plot shows urinary albumin/creatinine protein ratios . ( E ) Bar graph shows ELISA measurements ( O . D . ) for anti-double stranded DNA antibodies in serum . ( F ) Micrographs show representative H & E kidney sections ( 40X magnification ) . ( B–E ) Number of Stat5a , Stat5b and total Stat5 alleles ( i . e . genotype ) is explained in the key below each graph . Data are compiled from 3–5 mice per genotype . Error bars indicate standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 003 None of the STAT5 mutants exhibited histological abnormalities in the liver , spleen or intestine , tissues known to be affected in STAT5-null mice ( Data not shown ) ( Snow et al . , 2003; Yao et al . , 2006 ) . However , Stat5b-deficient mice did exhibit kidney pathology with a penetrance of 75% or 25% , depending on whether they harbored one or two Stat5a alleles ( Figure 1C ) . Afflicted individuals presented a loss of glomerular structure , proteinuria and systemic anti-DNA antibodies ( Figure 1D–F ) . Therefore , as in humans , Stat5b is required for immunological tolerance in mice but , given the clear difference between having one or two Stat5a alleles , redundancy and/or cooperativity is also evident . To probe for immunological phenotypes , we first assessed the cellularity and composition of primary lymphoid organs . Although not completely lymphopenic like STAT5-null mice ( Yao et al . , 2006 ) , one- and two-allele Stat5b-deficient mice did have fewer splenocytes than WT controls ( Figure 2A ) . Cell counts were also reduced in one-allele Stat5a-deficient mice , suggesting that , while STAT5B may be dominant , STAT5A does have substantial influence . Lymph node cellularity was similarly affected by the loss of either paralog and , in fact , all genotypes with less than three-alleles had reduced cell counts ( Figure 2A ) . 10 . 7554/eLife . 08384 . 004Figure 2 . Impact of STAT5 paralog deficiency on B and T cells . ( A ) Bar graphs show averaged cell counts for spleens ( top row ) and lymph nodes ( bottom row ) . Error bars indicate standard deviation . ( B ) Percentages of CD4+ T cells ( CD3+ CD4+ CD8α- ) , CD8+ T cells ( CD3+ CD4- CD8α+ ) and B cells ( CD3- B220+ ) were measured in spleens ( top row ) and LNs ( bottom row ) . Box plots show log2 fold changes relative to wild type controls ( WT=0; not shown ) . Dotted red lines indicate two-fold changes . ( A–B ) Number of Stat5a , Stat5b and total Stat5 alleles is explained in the key below each graph . Data are compiled from 5 experiments . ( C ) Contour plots show percentages of GL7+ Fas+ germinal center B cells in lymph nodes . Scatter plot shows percentages of LN resident GC B cells compiled from 3 experiments ( 3–4 mice per group ) . Genotypes are ordered as in Figure 1E ( WT mice = white , one-allele Stat5a-deficient mice = blue , one-allele Stat5b-deficient mice = orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 004 Frequencies of CD4+ T cells were comparable across all genotypes , whereas CD8+ T cells were reduced in one-allele Stat5b-deficient mice and , to a lesser extent , in one-allele Stat5a-deficient mice ( Figure 2B ) . By contrast , B cells were increased in one-allele Stat5b-deficient mice and , consistent with the appearance of auto-antibodies , GL7+ Fashigh IgDlow germinal center ( GC ) B cells were dramatically enriched ( Figure 2B–C and data not shown ) . One-allele Stat5a-deficient mice had a more modest accumulation of GC B cells , again , illustrating both the relevance and redundancy of STAT5A ( Figure 2B–C ) . The ability to promote B cell responses is a defining characteristic of CD4+ ‘helper’ T cells ( Crotty , 2011 ) . Therefore , given the appearance of GC B cells , we next investigated the CD4+ T cell compartment . Not surprisingly , there was a marked accumulation of CD44high IL-7Rαlow effector/memory T cells in Stat5b-deficient mice which , as with the incidence of kidney disease , was more pronounced in those bearing one-allele of Stat5a than in those bearing two ( Figure 3A-B & Figure 3—figure supplement 1A ) . We also measured production of IFN-ɣ and IL-17 , two effector cytokines that are dysregulated in STAT5-null mice ( Laurence et al . , 2007 ) . IFN-ɣ+ cells were highly enriched in the autoimmune-prone Stat5b-deficient mice but not age-matched Stat5a-deficient counterparts , suggesting that STAT5B may be particularly important for limiting Th1-type responses . IL-17A+ Th17-type cells were also increased but this trend did not reach statistical significance ( Figure 3C–D ) . 10 . 7554/eLife . 08384 . 005Figure 3 . Aberrant effector T cell responses in the absence of Stat5b . ( A ) Contour plots show percentages of CD44low IL-7R+naive and CD44high effector/memory CD4+ T cells in the spleens of 8 week-old mice . ( B ) Scatter plots show percentages of naive CD4+ T cells in spleens ( left ) and lymph nodes ( right ) . Genotypes are ordered as in Figure 1C . ( C ) Contour plots show percentages of IFN-ɣ+ or IL-17+ CD4+ T cells in the spleens of 8 week-old mice . ( D ) Box plots show log2 fold changes for IFN-ɣ+ and IL-17+ cells relative to wild type controls ( WT=0; not shown ) . ( E ) Contour plots show percentages of PD1+ CXCR5high Tfh cells in the spleens of 8 week-old mice . ( F ) Box plots show log2 fold changes for Tfh cells in spleens ( left ) and LNs ( right ) relative to wild type controls ( WT=0; not shown ) . ( D and F ) Genotypes are ordered as in Figure 1D . Data are compiled from 4 experiments ( 3–6 mice/group ) . Dotted red lines indicate two-fold changes . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 00510 . 7554/eLife . 08384 . 006Figure 3—figure supplement 1 . Impact of Stat5 allele depletion on effector T cell responses . ( A ) Contour plots show percentages of CD44low IL-7R+ naive and CD44high effector/memory CD4+ T cells . ( B ) Contour plots show percentages of PD1+ CXCR5high T follicular helper cells . ( A and B ) Data are from spleens of 8 week-old mice and are representative of 4–5 experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 006 CD4+ Tfh cells specialize in promoting B cell responses ( Crotty , 2011 ) . Mirroring the abundance of GC B cells , there was dramatic accumulation of PD1high CXCR5high ICOShigh Tfh cells in one-allele Stat5b-deficient mice , and a more modest enrichment in one-allele Stat5a-deficient mice ( Figure 3E–F , Figure 3—figure supplement 1B & data not shown ) . At least two interpretations can be made for this disparity; either the two proteins are not functionally equivalent or the two genes have different outputs . Consistent with the latter view , mice lacking one-allele each of Stat5a and Stat5b ( i . e . double-heterozygotes ) had more Tfh cells than those lacking two-alleles of Stat5a , despite having the same total number of alleles . Moreover , the percentage of Tfh cells was comparable between two-allele Stat5b-deficient mice and one-allele Stat5a-deficient mice , suggesting that two alleles of Stat5a are roughly equal to one allele of Stat5b ( Figure 3F & Figure 3—figure supplement 1B ) . CD4+ T regulatory ( Treg ) cells expressing the forkhead transcription factor , FOXP3 , are essential for immunological tolerance ( Malek and Castro , 2010 ) . Given the importance of STAT5 in Treg cells ( Mahmud et al . , 2013 ) , we next inspected this subset . Unlike STAT5-null mice , which exhibit a profound lack of Treg cells ( Yao et al . , 2007; Burchill et al . , 2006 ) , frequencies of splenic FOXP3+ cells were relatively normal across our STAT5 mutants . However , due to differences in overall cellularity , absolute counts were significantly lower in one- and two-allele Stat5b-deficient mice , as well one-allele Stat5a-deficient mice . A similar trend was observed for LN resident Treg cells; frequencies were comparable to WT controls but total numbers were reduced in all genotypes bearing less than 3 total alleles ( Figure 4—figure supplement 1 ) . To further characterize the Treg compartment , we measured IL-2Rα , a component of the IL-2 receptor that is critical for Treg cell homeostasis and function . It is also a both upstream and downstream of STAT5 signaling and , thus , can be viewed as an indicator of STAT5 activity ( Malek and Castro , 2010 ) . We found that the percentage of IL-2Rα+ Treg cells mirrored the total number of Stat5 alleles; it was slightly reduced in mice with three alleles , lower in those with 2 , and lower still in those with 1 ( Figure 4A–B ) . We also noted that residual IL-2Rα+ Treg cells from one-allele mice had reduced suppressive capacity and were unable to maintain expression of IL-2Rα during in vitro culture ( Figure 4C–E ) . Each of these phenotypes was more pronounced in the absence of Stat5b than Stat5a , again , illustrating both the dominance of the former and the relevance ( and/or redundancy ) of the latter . 10 . 7554/eLife . 08384 . 007Figure 4 . T regulatory cell function is impaired in Stat5b-deficient mice . ( A ) Contour plots show percentages of IL-2Rα+ cells within the FOXP3+ Treg compartment in spleens of 8 week-old mice . ( B ) Box plots show log2 fold changes in the ratio of IL-2Rα+/IL-2Rα- Treg ( WT=0; not shown ) . LN ( top ) and spleen ( bottom ) data are compiled from 5 experiments ( 4–6 mice/group ) and genotypes ordered as in Figure 1D . ( C ) IL-2Rα+ Treg cells from WT and Stat5a- or Stat5b-deficient mice were used for in vitro suppression assays . Histograms show CFSE dilution of responder T cells . ( D ) Line graph shows percent suppression across a range of responder:Treg ratios . Baseline is set according to WT controls at a 1:1 ratio . Data are compiled from 3 experiments . ( E ) Line graph shows the percent Treg cells that remained IL-2Rα+ during in vitro suppression . ( F ) IL-2Rα+ Treg cells were cultured with IL-2 for 72 hr . Contour plots show the percentage of FOXP3+ Treg cells expressing TBX21 ( top ) or IL-2 ( bottom ) . ( G ) Box plots show log2 fold changes for TBX21+ , FOXP3+ and IL-2+ cells relative to wild type controls ( WT=0; not shown ) . Data are compiled from 3 experiments . Dotted red lines indicate two-fold changes . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 00710 . 7554/eLife . 08384 . 008Figure 4—figure supplement 1 . Impact of Stat5 allele depletion on Treg cells . Bar graphs show total numbers of Treg cells in spleens and LNs of 8 week old mice ( 4–6 mice per genotype ) . Number of Stat5a , Stat5b and total Stat5 alleles ( i . e . genotype ) is explained in the key below each graph . Error bars indicate standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 008 Given the appearance of IFN-ɣ+ effector T cells in Stat5b-deficient mice , we next asked whether Stat5b-deficient Treg cells express TBX21 , a transcription factor required for Treg cells to limit Th1-type responses ( Koch et al . , 2009 ) . Similar to conventional T cells ( Liao et al . , 2011 ) , we found that IL-2 was sufficient to induce TBX21 in WT Treg cells ( Figure 4F ) . This effect was slightly reduced in the absence of Stat5a but almost completely abolished in the absence of Stat5b , consistent with the disparity of other STAT5-dependent parameters ( e . g . T cell and Th1 cell frequencies ) . Both Stat5a- and Stat5b-deficient Treg cells maintained FOXP3 expression similar to WT controls and , surprisingly , both gained the ability to produce IL-2 , a cytokine that is typically restricted in this lineage ( Figure 4F–G ) . STAT5 is required for in vitro differentiation of induced regulatory ( iTreg ) T cells ( Yao et al . , 2007 ) . To dissect the contributions of STAT5A and STAT5B , we purified naive CD4+ T cells from our STAT5 mutants , cultured them under iTreg polarizing conditions and compared expression of FOXP3 . We found that , although both paralogs appear to play a role , there were far fewer FOXP3+ cells in one-allele Stat5b-deficient cultures than in ( Figure 5A ) . We also found that deletion of either paralog endowed FOXP3+ iTreg cells with the ability to produce IL-2 , which suggests that , beyond differentiation , STAT5 may limit the inflammatory potential of this subset ( Figure 5A–B ) . 10 . 7554/eLife . 08384 . 009Figure 5 . Defective iTreg differentiation in the absence of Stat5b . ( A ) Naive CD4+ T cells were cultured under iTreg-inducing conditions . Contour plots show percentages of FOXP3+ and IL-2+ cells . ( B ) Box plots show log2 fold changes for FOXP3+ and IL-2+ cells relative to wild type controls ( WT=0; not shown ) . Data are compiled from 3 experiments and genotypes ordered as in Figure 1D . ( C ) Naive T cells from one-allele Stat5a- or Stat5b-deficient mice were cultured as in ( A ) and processed for RNA-seq . GSEA plots show enrichment of Treg signature genes within the Stat5a-deficient ( top ) or Stat5b-deficient ( bottom ) datasets relative to WT controls . ( D ) Heat map shows a selection of STAT5-regulated , Treg signature transcripts . Data are presented as log2 fold changes relative to WT controls ( not shown ) . RNA-seq analyses are compiled from 2 biological replicates per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 009 Next , we compared the transcriptomes of Stat5a- and Stat5b-deficient T cells cultured under iTreg polarizing conditions . Gene set enrichment analysis revealed that the overall Treg gene signature - defined by a combination of FOXP3- and IL-2-dependent transcriptional programs ( Hill et al . , 2007 ) - was similarly affected in both genotypes , meaning that there were no broad qualitative differences ( Figure 5C ) . However , there were obvious quantitative differences; several key genes , including Foxp3 and Il2ra , were more affected by the loss of STAT5B than STAT5A ( Figure 5D ) . Thus , while both paralogs can impact Treg cell biology , deletion of Stat5b is clearly more disruptive , befitting its dominant station within immunological tolerance . To define the molecular basis for phenotypic differences between Stat5a- and Stat5b-deficient T cells , we employed a bioinformatic approach . First , we compared their transcriptomes either directly ex vivo or after acute exposure to STAT5-activating stimuli . The ex vivo set included naive T cells andTreg cells , while the in vitro set included naive T cells cultured with IL-7 and effector T cells cultured with IL-2 ( Figure 6A & Figure 6—figure supplement 1A ) . These pairings were chosen to match the expression patterns of requisite ɣc co-receptors; IL-7R , which is highly expressed on naive T cells , and IL-2Rα , which is highly expressed on effector T cells ( Rochman et al . , 2009 ) . One-allele mice were used because they exhibited the most dramatic T cell phenotypes . 10 . 7554/eLife . 08384 . 010Figure 6 . Redundancy and specificity of STAT5 paralogs for gene transcription . ( A ) Cartoons depict the cell types and experimental conditions used for RNA-seq . ( B ) Histograms show STAT5 paralog preference for all STAT5-regulated transcripts . Those which were more influenced by the loss of Stat5a are positioned to the left ( X<0 ) while those that were more influenced by the loss of Stat5b are positioned to the right ( X>0 ) . Dotted red lines denote equivalence ( X=0 ) and numbers indicate median paralog preference . ( C ) Pie charts depict paralog-specific transcripts . Those impacted only in Stat5a-deficient cells are indicated in blue , those impacted only in Stat5b-deficient cells are indicated in orange and those impacted in both genotypes are indicated in black . ( D ) Heat maps show a selection of STAT5-regulated transcripts . Data are presented as the log2 fold change relative to WT controls ( not shown ) . ( E ) IL-2Rα protein was measured in T cells treated with IL-2 ( left ) or IL-6 ( right ) . Box plots show log2 fold changes for mean fluorescence intensity relative to wild type controls ( WT=0; not shown ) . Genotypes are ordered as in Figure 1D . Dotted red lines indicate a two-fold change . ( A–D ) RNA-seq analyses are compiled from 2–3 biological replicates per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 01010 . 7554/eLife . 08384 . 011Figure 6—figure supplement 1 . Transcriptomic analysis of Stat5a- and Stat5b-deficient T cells . ( A ) Cartoons depict the experimental conditions used for RNA-seq . CD4+ T cells were purified from WT and one-allele Stat5a- or Stat5b-deficient mice , then transcriptomes measured either directly ex vivo or after in vitro treatment with acute STAT5 stimuli . The ex vivo set included naive cells ( first row ) and IL-2Rα+ Treg cells ( second row ) , while the in vitro set included naive cells exposed to IL-7 ( third row ) and effector cells exposed to IL-2 ( fourth row ) . ( B ) Multidimensional scaling ( MDS ) plots show the overall relatedness between experimental groups . Each biological replicate is represented by black ( WT ) , blue ( Stat5a-deficient ) or orange ( Stat5b-deficient ) elements . ( C ) Volcano plots show fold changes and variances for all transcripts relative to WT controls . Those exhibiting >1 . 5 fold change and p<0 . 05 are highlighted . Numbers indicate the sum of transcripts that were down-regulated ( upper left ) or up-regulated ( upper right ) in Stat5a- or Stat5b-deficient cells . Dotted red lines are drawn at 2 fold changes and 0 . 05 p values . ( D ) Venn diagrams indicate the number of transcripts exhibiting >1 . 5 fold change and <0 . 05 p values only in Stat5a-deficient cells ( blue ) , only in Stat5b-deficient cells ( orange ) or in both genotypes ( black ) . ( E ) XY plots show log 2 fold change for STAT5-regulated genes in Stat5a-deficient ( x axis ) versus Stat5b-deficient ( y axis ) cells . Blue and orange elements represent transcripts designated as STAT5A- or STAT5B-specific , respectively . Dotted red lines are drawn at 1 . 5 fold changes . ( A–-E ) Analyses are compiled from 2 to 3 biological replicates per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 01110 . 7554/eLife . 08384 . 012Figure 6—figure supplement 2 . Transcriptomic analysis of Stat5a- and Stat5b-deficient T cells . Genome browser tracks show transcript abundance in WT ( grey ) , Stat5a-deficient ( blue ) or Stat5b-deficient ( orange ) cells . Vertical RPKM scale varies from gene to gene ( but not across experimental conditions ) and is denoted by the numbers at the upper left of each column . Shown is one of 2–3 biological replicates per genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 012 We began our transcriptomic survey by performing multidimensional scaling of the datasets , thereby gaining a broad overview of the experimental groups . Stat5a- and Stat5b-deficient cells typically clustered together and equidistant from WT controls , suggesting that the loss of either paralog has comparable genome-wide effects ( Figure 6—figure supplement 1B ) . Next , we used statistical variance to identify differentially expressed genes . Surprisingly , we found widespread discord between Stat5a- and Stat5b-deficient cells; many transcripts appeared dysregulated in the absence of one paralog or the other ( Figure 6—figure supplement 1C–D ) . However , upon close inspection , we concluded that this disparity was largely due to the arbitrary fold-change cutoff that was used . Most genes that were designated as STAT5B-specific were also affected by the loss of STAT5A ( and vice versa ) , albeit to a lesser degree that did not reach our 2-fold threshold ( Figure 6—figure supplement 1E ) . To avoid this statistical artifact , we devised a ‘paralog preference’ scale whereby all STAT5-sensitive genes were compiled and ranked according to how much they were impacted by the loss of Stat5a or Stat5b . This analysis revealed a binomial distribution for all experimental conditions . The majority of genes were in central bins , affected by both STAT5A and STAT5B , while membership in peripheral bins decreased steadily as paralog preference increased . Importantly , all curves were shifted towards STAT5B , suggesting that STAT5-sensitive genes are generally more impacted by STAT5B than STAT5A ( Figure 6B ) . This latter trend was also evident at the protein level; IL-2-driven ( but not IL-6-driven ) induction of IL-2Rα was clearly more diminished in Stat5b-deficient cells than in Stat5a-deficient counterparts ( Figure 6E ) . Collectively , these data affirm that STAT5B is dominant over STAT5A while , at the same time , demonstrating pervasive redundancy at the level of gene transcription . Beyond quantitative differences , our transcriptomic survey also revealed qualitative differences between Stat5a- and Stat5b-deficient cells . Using strict analysis criteria , we discovered that between 12% and 22% of all STAT5-sensitive genes can be classified as paralog-specific , meaning that they are solely dependent on either STAT5A or STAT5B . The absolute number of paralog-specific genes varied across cell states and stimuli , with the largest allotment found in IL-7-treated naive cells , and was typically skewed towards STAT5B ( Figure 6C & Figure 6—figure supplement 1E ) . Thus , we can create 2 general categories: ‘pan-STAT5' genes that are regulated by both STAT5A and STAT5B ( e . g . Pdk1 , Cish , Lta ) and 'paralog-specific' genes that are regulated by either STAT5A ( e . g . Smc6 ) or STAT5B ( e . g . Cd74 ) ( Figure 6D & Figure 6—figure supplement 2 ) . Given that pan-STAT5 genes are much more numerous , we propose that phenotypic differences between Stat5a and Stat5b deficient T cells are due largely to paralog preference , owing to the fact that deletion of Stat5b has greater quantitative impact , with limited contribution from qualitative , paralog-specific effects . Functional divergence between STAT5A and STAT5B could be due to differences in target gene selection . Previous studies have addressed this issue by comparing genomic distributions by ChIP-seq ( chromatin immunoprecipitation followed by massively parallel sequencing ) using separate , paralog-specific antibodies in WT cells ( Liao et al . , 2008; 2011; Kanai et al . , 2014 ) . We took an alternative approach involving a single antibody that recognizes both paralogs and T cells from Stat5a- or Stat5b-deficient mice , as well as ‘double heterozygotes’ ( hereafter referred to as Stat5a/bhet mice ) . In line with previous studies , 1275 unique regions of STAT5 occupancy were called for WT cells ( Figure 7A ) . By contrast , Stat5a/bhet cells had fewer peaks ( 658 total; Figure 7A ) that tended to be less robust ( i . e . lower signal intensity ) than those found in WT controls ( Figure 7D ) , indicating that changes in STAT5 availability can impact genomic distribution even when both paralogs are present . Total peaks were also reduced in Stat5a-deficient cells ( 609 total ) and almost expunged in Stat5b-deficient cells ( 97 total ) , again , illustrating both the relevance of the former and the dominance of the latter ( Figure 7A ) . STAT5 peaks were similarly localized across all genotypes - they typically congregated near transcriptional start sites but could also be found at distal regions , sometimes >100 kb from annotated genes - and were highly enriched for STAT-binding motifs ( Figure 7B & Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 08384 . 013Figure 7 . Influence of paralog dose on genomic distribution of STAT5 . ( A ) CD4+ T cells from WT , Stat5a/bhet and two-allele Stat5a- or Stat5b-deficient mice were cultured in the presence of IL-2 , then processed for pan-STAT5 ChIP-seq . Bar graph shows the total number of STAT5-bound peaks in each genotype . ( B ) Histogram shows distribution of STAT5-bound peaks relative to transcriptional start sites ( TSS ) . ( C ) Circos plot shows overlap of STAT5-bound beaks across genotypes . Connection width represents the number of overlapping peaks . Only peaks shared with WT cells are shown . Those found only in WT cells are presented as a white semi-circle at the top . ( D ) Bar graph shows the percentage of WT peaks detected in each genotype ( WT=100% ) . Violin plot depicts the total number of sequenced tags ( i . e . peak intensity ) for peak shared with WT controls . ( E ) Genome browser tracks show STAT5 peaks near selected genes . Numbers indicate the maximum peak height within the interval . ( A–E ) Data are representative of two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 01310 . 7554/eLife . 08384 . 014Figure 7—figure supplement 1 . Transcription factor motifs associated with STAT5 binding peaks . CD4+ T cells were isolated from WT , Stat5a/bhet and two-allele Stat5a- or Stat5b-deficient mice , then cultured in the presence of IL-2 and processed for ChIP-seq . Chart shows the top ten transcription factor-binding motifs associated with STAT5 peaks in each genotype . p Values and enrichment ( % STAT5 peaks with indicated motif ÷% random peaks with indicated motif ) are shown for one of two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 01410 . 7554/eLife . 08384 . 015Figure 7—figure supplement 2 . Correlation between STAT5 binding and transcription of Il2ra , Bcl2 and Bcl6 . ( A ) Effector CD4+ T cells from one-allele Stat5a- or Stat5b-deficient mice were cultured in the presence of IL-2 and processed for RNA-seq . ( B ) Effector CD4+ T cells from one-allele Stat5b-deficient mice were transduced with control or STAT5A-expressing retrovirus and processed for RNA-seq . ( C ) Effector CD4+ T cells from WT , Stat5a/bhet and two-allele Stat5a- or Stat5b-deficient mice were cultured in the presence of IL-2 and processed for ChIP-seq . ( A–C ) Genome browser tracks show transcript abundance or STAT5 binding for selected genes . Numbers indicate either RPKM values for the most detected exon ( RNA-seq ) or maximum STAT5-binding peak intensity within the interval ( ChIP-seq ) . Shown is one of two biological replicates for each genotype and/or experimental condition . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 015 Most peaks found in Stat5b-deficient cells could be matched to peaks in Stat5a-deficient , Stat5a/bhet or WT cells ( Figure 7C ) . This implies a hierarchy whereby certain sites are preserved even when STAT5B is absent . STAT5B-independent peaks tended to occur near genes that were highly occupied in WT controls and whose expression was highly dysregulated in Stat5b-deficient cells ( e . g . Cish , Lta ) , suggesting that only the most robust ( i . e . high-affinity ) STAT5-binding sites were preserved ( Figure 8 ) . Peaks detected within Stat5a/bhet and Stat5a-deficient cells also tended to be highly occupied in WT controls and dysregulated in STAT5-deficient cells , but the trend was not as dramatic , indicating that , while a full complement of STAT5 alleles may be necessary to achieve optimal responses , STAT5B has the greater influence on genomic distribution ( Figure 8 ) . 10 . 7554/eLife . 08384 . 016Figure 8 . Preservation of high affinity targets in the absence of STAT5B . Circle plot relates STAT5 occupancy and STAT5-dependent transcription for genes bound in: ( 1 ) only WT cells ( white circle ) , ( 2 ) WT and Stat5a/bhet , cells ( grey circle ) , ( 3 ) WT , Stat5a/bhet , and Stat5a-/- cells ( blue circle ) , ( 4 ) WT , Stat5a/bhet , Stat5a-/- and Stat5b-/- cells ( orange circle ) . STAT5 ChIP-seq peaks were assigned to genes based on proximity to transcriptional start sites ( +/- 10 kb ) . X axis denotes the average height of gene-associated peaks in WT cells . Y axis denotes the average mRNA expression variance ( -log10 p value ) for the corresponding peak-associated genes . Variance is derived from the comparison WT and one-allele Stat5a- or Stat5b-deficient cells cultured in the presence of IL-2 ( from Figure 6 ) . Size of each circle represents the total number of gene-associated peaks in each group ( number is shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 016 Among the genes that were engaged by STAT5 in WT cells and dysregulated in Stat5-deficient cells was Il2ra , which , as discussed , is a known STAT5 target gene that is critical for Treg function and homeostasis ( Figure 7E & Figure 7—figure supplement 2 ) . Another was Bcl6 , considered the master transcription factor for Tfh differentiation ( Crotty , 2011 ) ( Figure 7E & Figure 7—figure supplement 2 ) . In this case , STAT5-binding appears to be a negative regulatory event; multiple studies ( including the present work ) have shown that STAT5 can suppress Bcl6 expression in T cells ( Oestreich et al . , 2012; Liao et al . , 2014 ) . Thus , taken together , our RNA-seq and ChIP-seq data provide a molecular rationale for the Treg and Tfh phenotypes seen in Stat5-deficient mice . Based on our RNA-seq and ChIP-seq studies , we reasoned that asymmetric expression , rather than widespread paralog-specific activity , likely explains the phenotypic differences between Stat5a- and Stat5b-deficient T cells . To explore this possibility , we mined various transcriptome datasets ( including our own ) and confimed that , indeed , Stat5b is more abundant than Stat5a at the mRNA level ( Figure 9—figure supplement 1 ) . Next , we used flow cytometry to measure total STAT5 protein in naive , regulatory ( Treg ) , follicular ( Tfh ) and effector/memory T cells . Regardless of cellular subset , the results were clear: removing one-allele of Stat5b ( Stat5a+/+ Stat5b+/- ) had greater impact than removing one-allele of Stat5a ( Stat5a+/+ Stat5b+/- ) while , at the other end of the spectrum , retaining one-allele of Stat5b ( Stat5a-/- Stat5b+/- ) than retaining one-allele of Stat5a ( Stat5a+/- Stat5b-/- ) ( Figure 9A ) . A similar trend was observed for tyrosine-phosphorylated STAT5 dowsntream of IL-2 or IL-7 ( Figure 9B ) . Thus , we conclude that STATB makes a greater contribution to the total STAT5 protein pool . To determine how STAT5 availability ( i . e . paralog dose ) influences gene expression , we transduced Stat5b-deficient T cells with a STAT5A-expressing retrovirus , thereby increasing the total amount of STAT5 without re-introducing STAT5B . We first validated the system by measuring Il2ra , a well-documented STAT5 target , and found it to be highly induced at both the mRNA and protein levels ( Figure 10A ) . Transcriptomic analysis revealed that , overall , ectopic STAT5A mobilized 320 genes , most of which fall within the pan-STAT5 category ( e . g . Cish , Lta ) ( Figure 10A–B ) . Applied to our paralog preference scale , these genes did not favor STAT5A , meaning that they were similarly affected in Stat5a- and Stat5b-deficient cells ( Figure 10A ) , and GSEA revealed a high degree of enrichment for both STAT5A- or STAT5B-dependent gene sets ( Figure 10C ) . Thus , our data support the idea that differences in STAT5 protein concentrations underlie many ( if not most ) of the transcriptomic divergence between Stat5a- and Stat5b-deficient cells . Having established that ectopic STAT5A can rescue gene expression in Stat5b-deficient T cells , we next asked whether it can rescue cellular differentiation . For these studies , naive T cells from wild type , Stat5a/bhet or one-allele Stat5b-deficient mice were cultured under iTreg polarizing conditions , transduced with either control or STAT5A retrovirus and FOXP3 measured to assess lineage commitment . As expected , FOXP3 was reduced in control-transduced Stat5a/bhet cells and almost completely abolished in control-transduced Stat5b-deficient cells ( Figure 10D ) . However , when ectopic STAT5A was introduced , the percentage of FOXP3+ cells became comparable across all genotypes and , whether endogenous ( top row ) or ectopic ( bottom row ) , there was a clear linear correlation between STAT5 and FOXP3 protein levels ( Figure 10D ) . IL-2Rαwas also diminished in both Stat5a/bhet and Stat5b-deficients cells , and was restored by ectopic STAT5A ( Figure 10E ) . These data argue that a threshold concentration of STAT5 must be reached to institute the Treg program and , given the conspicuous effect of ectopic STAT5A on WT cells ( Figure 10D & Figure 10—figure supplement 1A ) , they imply that STAT5 is a limiting resource for this process . Although they share a common instructive cytokine ( TGF-β ) , Th17 cells and Treg cells have opposing pro- and anti-inflammatory functions . STAT5 is key to this divergence - it promotes Treg responses at the expense of Th17 responses – so we next investigated the effect of paralog dose on Th17 differentiation . We found that the percentage of IL-17+ cells was 4-fold higher in Stat5a/bhet Th17 cultures and >25-fold higher in Stat5b-deficient Th17 cultures than in WT controls , consistent with a high paralog dose threshold , and most importantly , that ectopic STAT5 not only extinguished IL-17 but also induced FOXP3 in all genotypes , thereby demonstrating that changes in STAT5 concentration can tip the balance between effector and regulatory T cells programs ( Figure 10—figure supplement 1B ) .
Although the importance of STAT5 is widely recognized , there is no consensus on whether its closely related paralogs , STAT5A and STAT5B , are redundant or functionally distinct . Assuming the latter , it is also unclear how specificity can be achieved given their extensive structural homology . Both positions are grounded in sound experimental evidence but , until the present studies , there has been no comprehensive inquiry on their relationship in immune cells . We addressed this longstanding question in primary CD4+ helper T cells , the principal orchestrators of adaptive immunity . Using a combination of genetic and genomic approaches , we demonstrate that STAT5B is dominant over STAT5A and , thus , plays a non-redundant role in controlling effector and regulatory T cell responses . This conclusion is based on phenotypic differences between Stat5a- and Stat5b-deficient mice , as well as bioinformatic analyses showing that STAT5B has greater impact on both selection and transcription of STAT5 target genes . The disparity does not appear to be due to differences in genome wide distribution or transcriptional capacity but , instead , relates to differences in relative abundance . Consistent with the latter point , our loss- and gain-of-function studies demonstrate that a threshold concentration of STAT5 must be reached to execute STAT5-dependent gene expression and differentiation programs . Based on these findings , we submit that STAT5A and STAT5B are largely redundant at the molecular level , but not at the cellular or organismal levels , where STAT5B is dominant . It has been proposed that the target repertoires of STAT5A and STAT5B vary due to subtle differences in their DNA-binding domains ( Boucheron et al . , 1998 ) . However , this notion has been disputed because the nature and location of the amino acid substitutions may not alter protein structure enough to impact specificity . In addition , multiple studies have shown that the consensus DNA-binding motifs for STAT5A and STAT5B are identical , although it should be noted that these measured optimal binding to synthetic oligonucleotides in cell free systems , leaving open the possibility that divergent binding properties become apparent only at lower affinity sites or in the context of native chromatin ( Soldaini et al . , 1999; Ehret et al . , 2001 ) . Indeed , differential binding of STAT5A or STAT5B has been detected at several loci in primary immune cells but it remains unclear whether this reflects bonafide differences in specificity or other factors that may influence target gene selection ( Liao et al . , 2008; 2011; Yamaji et al . , 2013; Kanai et al . , 2014 ) . For instance , it is known that STAT5A and STAT5B can exhibit distinct phosphorylation patterns , so preferential binding may reflect cell type- or stimulus-specific differences in activation rather than distinct targeting capabilities ( Caldenhoven et al . , 1998; Hennighausen and Robinson , 2008; Rosen et al . , 1996; Meinke et al . , 1996 ) . Our work supports this latter view by establishing that , even before activation , relative abundance of STAT5A versus STAT5B determines which paralog will dominate a given transcriptional response . 10 . 7554/eLife . 08384 . 017Figure 9 . Relative abundance of STAT5A and STAT5B in helper T cells . ( A ) Total STAT5 protein was measured in naive , Treg , Tfh and effector/memory T cells . Donut charts indicate the percentage of total STAT5 protein accounted for by each paralog . Histograms show representative flow cytometry data from one of three experiments . ( B ) Naive , Treg and effector/memory T cells were treated with IL-2 or IL-7 and phospho-STAT5 measured by flow cytometry . Box plots show log2 fold changes for the percentage of p-STAT5high cells relative to wild type controls ( WT=0; not shown ) . Genotypes are ordered as in Figure 1D . Dotted red lines indicate a two-fold change ( 3–4 replicates/group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 01710 . 7554/eLife . 08384 . 018Figure 9—figure supplement 1 . Relative abundance of STAT5A and STAT5B in helper T cells . ( A ) Donut charts indicate the percentage total STAT5 mRNA ( top ) or protein ( bottom ) accounted for by each paralog in ex vivo naive or Treg cells . mRNA data are compiled from 2–3 RNA-seq replicates and protein data are compiled from 3 flow cytometry replicates ( B ) Donut charts indicate the percentage total STAT5 mRNA accounted for by each paralog in mouse naive or Treg cells . Data were sourced from the Immunological Genome Project ( top ) or the EMBL-EBI Expression Atlas ( bottom ) . ( C ) Donut charts indicate the percentage total STAT5 mRNA accounted for by each paralog in human naive or Treg cells . Data were sourced from the BioGPS Primary Cell Atlas . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 01810 . 7554/eLife . 08384 . 019Figure 9—figure supplement 2 . Stat5a and Stat5b are transcribed from opposite DNA strands . Genome browser tracks show relative abundance and DNA strand origin of Stat5a and Stat5b mRNAs within mouse spleen , thymus and mammary gland . Transcripts originating from the minus ( - ) strand are presented in the ascending orientation while those from the plus ( + ) strand are in the descending orientation . Numbers indicate RNA-seq FPKM values for the most detected exon within the interval . Data are sourced from the ENCODE consortium and one of 2 biological replicates visualized using the UCSC genome browser . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 01910 . 7554/eLife . 08384 . 020Figure 9—figure supplement 3 . Putative lymphocyte-restricted enhancers within the Stat5a/b locus . Genome browser tracks display DNAse I hypersensitivity sites ( DHS ) within the mouse Stat5a/b locus . Shown are data for primary CD4+ T cells ( naive , activated , Treg ) , as well as spleen , thymus and multiple non-lymphoid tissues . 3 putative lymphoid-specific DHS sites are highlighted . Numbers indicate maximum tag count for all DHS sites within the interval . Data are sourced from the ENCODE consortium and one of 2 biological replicates visualized using the UCSC genome browser . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 020 Recent work has shown that small oscillations in transcription factor availability can have genome-wide consequences ( Brewster et al . , 2014 ) . The idea that STAT5 concentration can impact cellular function also has precedent . Of particular interest are studies reporting severe immunological phenotypes in transgenic mice which over-express STAT5A or STAT5B ( Kelly et al . , 2003a; Kelly et al . , 2003b ) , and studies showing that Stat5a/b haplo-insufficiency ameliorates contact hypersensitivity ( Nivarthi et al . , 2014 ) . In addition , we have previously demonstrated that ectopic STAT5A can expand the target repertoire of STAT5 in mouse embryonic fibroblasts ( Zhu et al . , 2012 ) , and have explored the concept of STAT5 gene dosage in the context of mammary development , finding that a high STAT5 threshold must be reached for mammary epithelial cell differentiation ( Yamaji et al . , 2013 ) . Given that STAT5A is the dominant paralog in mammary epithelium , we can infer that asymmetric expression of STAT5 paralogs is not just a feature of immune cells , and that it must be controlled in a tissue-specific manner ( Metser et al . , 2016 ) . Several mechanisms may explain this phenomenon . First , it is known that Stat5a and Stat5b are transcribed from opposite DNA strands and , thus , may be subject to strand-specific modes of regulation ( Figure 9—figure supplement 2 ) . Second , differential transcription could be achieved through paralog-specific enhancer elements whose accessibility is tissue- and/or cell type-restricted . We have recently characterized an intergenic enhancer that drives expression of Stat5a in mammary epithelium and have identified multiple DNase hypersensitivity sites within the Stat5b locus that are present in T cells but not in non-lymphoid tissues , perhaps indicating an analogous mechanism for immune cells ( Metser et al . , 2016 ) ( Figure 9—figure supplement 3 ) . Third , paralog-specific epigenetic modifications , such as histone or DNA methylation , may impose distinct transcriptional outputs , as shown for tumor cells ( Zhang et al . , 2007 ) . Fourth , the 3’ UTRs of Stat5a and Stat5b are highly divergent so it is possible that their mRNAs are subject to post-transcriptional regulation via distinct sets of microRNAs and/or RNA-binding proteins ( Liu et al . , 1995 ) . Beyond asserting the dominance of STAT5B , our work also affirms the importance of STAT5A . Several observations support this latter point: 1 ) deletion of one Stat5a allele exaggerates the gross and cellular phenotypes of Stat5b-deficient mice , 2 ) transcription of STAT5 target genes is typically influenced by both STAT5A or STAT5B , and 3 ) ectopic STAT5A can rescue gene expression in Stat5b-deficient cells . Furthermore , just one-allele of either Stat5a or Stat5b is sufficient to prevent the perinatal lethality and anemia seen in STAT5-null mice , suggesting that molecular redundancy protects the most critical ‘life-and-death’ functions . We also identified a small subset of genes that appear to be regulated by either STAT5A or STAT5B , some of which have known immunological functions . Given that STAT5B is more abundant , it can be argued that all STAT5B-dependency may be due to a high paralog dose threshold , but this cannot explain the appearance of STAT5A-dependent genes . Thus , we propose that phenotypic differences between Stat5a- and Stat5b-deficient T cells result from widespread ‘paralog preference’ and circumscribed 'paralog specificity' . Our ChIP-seq studies indicate that the overall availability of STAT5 , whether STAT5A or STAT5B , has profound influence on target gene selection . Previous studies have compared genomic distribution of STAT5A and STAT5B in primary CD4+ T cells and found that they mostly overlap , thereby supporting the idea of redundancy ( Liao et al . , 2008; 2011; Kanai et al . , 2014 ) . However , they also identified a subset of sites that are occupied by one paralog or the other and , thus , have been taken as evidence for paralog specificity . All such comparisons ( including ours ) should be interpreted with care . Shared sites can be appointed with confidence but , due to technical confounders ( e . g . differences in antibody affinity ) , incongruent sites cannot be definitively classified as STAT5A- or STAT5B-specific . The mechanisms underlying differential binding must also be considered . It is possible that bona fide paralog-specific binding sites do exist , but these are probably only a minor fraction . In most cases , differential binding likely reflects competition; the more abundant paralog is more likely to be detected . Given this nuance , claims that certain genes are uniquely regulated by STAT5A or STAT5B should be tempered . For instance , it has been suggested that Bcl2l1 is regulated only by STAT5A and that Bcl2 , Il2ra and Foxp3 are regulated only by STAT5B ( Kanai et al . , 2014; Jenks et al . , 2013 ) . Our data indicate that these are more accurately described as 'pan-STAT5' genes that are more impacted by deletion of one paralog or the other . 10 . 7554/eLife . 08384 . 021Figure 10 . Paralog dose governs STAT5-driven gene transcription . ( A ) CD4+ T cells from one-allele Stat5b-deficient mice were transduced with STAT5A retrovirus , then processed for RNA-seq . Contour plots ( left ) show correlation between the transduction marker ( hNGFR ) and IL-2Rα protein . Volcano plot ( middle ) shows log2 fold changes and variances for all transcripts relative to control retrovirus . Those exhibiting >1 . 5 fold change and p<0 . 05 are depicted in blue . Dotted red lines indicate 2 fold change and 0 . 05 p value . Histogram ( right ) shows STAT5 paralog preference for transcripts mobilized by ectopic STAT5A . Dotted red line denotes equivalence and number indicates median paralog preference . ( B ) Heat map shows selected transcripts in STAT5A-transduced helper T cells ( top row ) or IL-2 treated Stat5a- or Stat5b-deficient cells ( bottom rows; from Figure 6 ) . Data are presented as the log2 fold change relative to controls ( not shown ) . ( C ) GSEA plots show enrichment of STAT5A-dependent ( left ) or STAT5B-dependent ( right ) genes within the STAT5A-RV dataset . ( A–C ) . RNA-seq analysis is compiled from 2 biological replicates . ( D ) CD4+ T cells from WT , Stat5a/bhet and one-allele Stat5b-deficient mice were transduced with control ( top row ) or STAT5A ( bottom row ) retrovirus under iTreg polarizing conditions . Contour plots show total STAT5 and FOXP3 protein levels in transduced cells . ( E ) Histograms denote IL-2Rα protein levels on transduced cells . ( D–E ) Shown is one of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 02110 . 7554/eLife . 08384 . 022Figure 10—figure supplement 1 . STAT5 paralog dose tips the balance between effector and regulatory T cell programs . CD4+ T cells from WT , Stat5a/bhet and one-allele Stat5b-deficient mice were were transduced with control or STAT5A retrovirus under ( A ) non-polarizing or ( B ) Th17-polarizing conditions . Contour plots denote FOXP3 and IL-17Aprotein levels in transduced cells . Shown is one of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08384 . 022 STAT5 is essential for immunological tolerance ( Mahmud et al . , 2013 ) . This principle is well illustrated in humans with congenital STAT5B defects , who typically manifest a range of autoimmune symptoms ( Kanai et al . , 2012 ) , and is further supported by the present work , which demonstrates that Stat5b deficiency leads to spontaneous kidney disease in mice . The link between STAT5 and autoimmunity is often attributed to its role downstream of IL-2/IL-2Rα in Treg cells ( Malek and Castro , 2010; Mahmud et al . , 2013 ) . Our work clearly endorses this viewpoint and brings to mind the autoimmune phenotype of Treg-deficient Scurfy mice which , like Stat5b-deficient mice , exhibit both autoantibodies and kidney disease {Aschermann:2013gn} . We demonstrate that , similar to STAT5B-deficient humans ( Cohen et al . , 2006 ) , Treg cells are functionally compromised in Stat5b-deficient mice , but , surprisingly , the baseline frequency of FOXP3+ cells was not reduced , likely reflecting immunological and/or environmental differences between the two species . We also present new ideas about why STAT5-deficient Treg cells are impaired . First , they acquire the ability to produce IL-2 , a cytokine that is typically restricted in Treg cells ( Malek and Castro , 2010 ) . This finding is consistent with previous studies demonstrating that STAT5 can suppress IL-2 production by conventional T cells and , given that Treg cells are thought to operate , in part , by consuming IL-2 , it provides one explanation for their ineffectiveness ( Villarino et al . , 2007; Pandiyan et al . , 2007 ) . Second , they fail to express TBX21 , a transcription factor that is required to limit Th1-type T cell responses ( Koch et al . , 2009 ) . Previous studies have shown that STAT1-activating cytokines ( e . g . interferons , IL-27 ) can induce TBX21 in Treg cells but we are the first to show that IL-2 , a STAT5-activating cytokine , can do it ( Hall et al . , 2012; Koch et al . , 2012 ) . Aside from its role in Treg cells , STAT5 promotes immunological tolerance via effector cell-intrinsic mechanisms . Given the dramatic accumulation of Tfh cells in our STAT5 mutants , the capacity of Tfh cells to promote autoimmunity , and recent work showing that STAT5 can suppress Tfh differentiation ( Ballesteros-Tato et al . , 2012; Johnston et al . , 2012 ) , we conclude that exaggerated Tfh responses factor heavily in the autoantibody responses and attendant kidney pathology seen in Stat5b-deficient mice . Our data also suggest an intimate relationship between STAT5 and BCL6 , the ‘master’ transcription factor for Tfh cells . We report that STAT5 directly engages the Bcl6 locus , where it likely acts as a transcriptional repressor , and that STAT5 binding sites are often enriched for BCL6 motifs , consistent with published accounts of co-localization between these two transcription factors ( Y . Zhang et al . , 2012; Liao et al . , 2014 ) . These findings strongly implicate Tfh cells in the pathogenesis of Stat5b-deficient mice but , since these are germline ‘knockouts’ , we must consider the ( likely ) possibility that intrinsic defects in other cell types contribute to the autoimmune phenotype . For instance , multiple dendritic cell subsets are known to be dysregulated in Stat5- or Jak3-deficient mice ( Esashi et al . , 2008; Yamaoka , 2005 ) , and its influence on non-immune cells , particularly downstream of hormone receptors , cannot be ignored ( Kuhrt and Wojchowski , 2015; Hennighausen and Robinson , 2008 ) . Because of its prominent role within the immune system , STAT5 has long been viewed as an attractive target for therapeutic intervention . Clinical use of STAT5-activating cytokines and growth factors ( e . g . IL-2 , erythropoetin ) is now commonplace and the recent approval of Jak3 inhibitors for the treatment of autoimmune disease and malignancy points to sustained interest in this pathway ( Villarino et al . , 2015 ) . Consequently , a detailed understanding of how STAT5 signaling works is imperative not only to inform new drugs , but also to improve existing regimens . The present study yields multiple clinically relevant insights and , in particular , raises two key issues that should be considered . First , partial inhibition of STAT5 expression and or activity may be sufficient to have desired effects on immune cell function . Second , targeting of STAT5A may be safer ( though perhaps less robust ) than targeting of STAT5B . Therefore , taking a broad view , our findings provide a molecular rationale for exploiting STAT5 paralog redundancy in clinical settings .
STAT5 mutants were generated as described ( Yamaji et al . , 2013 ) . Briefly , mice lacking the entire Stat5 locus ( Stat5a/b+/- ) were crossed with mice lacking one-allele of Stat5a ( Stat5a+/- Stat5b+/+ ) or Stat5b ( Stat5a+/+ Stat5b+/- ) to produce 8 combinations of Stat5 alleles ( Figure 1A ) . We refer to each genotype according to the total number of Stat5 alleles that are retained . For example , two-allele Stat5a-deficient mice lack both Stat5a alleles but retain two Stat5b alleles ( Stat5a-/- Stat5b+/+ ) , while one-allele Stat5a-deficient mice lack both Stat5a alleles but retain one Stat5b allele ( Stat5a-/- Stat5b+/- ) . CD45 . 1+ C57BL/6 mice were purchased from Jackson Labs ( Bar Harbor , ME ) . Animals were handled in accordance with NIH guidelines and all experiments approved by the NIAMS Animal Care and Use Committee . Complete blood counts were taken from 8 to 12-week old mice ( NIH Clinical Center , Division of Veterinary Services , Bethesda , MD ) . Anti-double stranded DNA antibodies were measured in serum collected from 4 to 6 month old mice ( Calbiotech , Spring Valley , CA ) . Albumin/creatinine ratio was measured in urine collected from 4 to 6 month old mice ( Exocell , Philadelphia , PA ) . Spleen and lymph node ( cervical , axillary , brachial and inguinal ) cellularity was measured in 8–12 week old mice using a Nexcelom X1 Cellometer ( Lawrence , MA ) . Kidneys were dissected from 4 to 6-month old mice , fixed , embedded in paraffin , sectioned and stained with haematoxylin and eosin ( American Histolabs , Gaithersburg , MD ) . Blinded scoring was performed by a veterinary pathologist ( Diagnostic & Research Services Branch , National Institutes of Health , Bethesda , MD ) . Specimens from at least 3 mice per genotype were inspected . Micrograph images were collected using a BioRevo BZ-9000 digital microscope ( Keyence , Itasca , IL ) . For surface proteins , cells were stained directly ex vivo with fluorochrome labelled anti-mouse CD3ε , CD4 , CD8α , CD25 ( IL-2Rα ) , CD44 , CD45R ( B220 ) , CD95 ( FAS ) , CD127 ( IL-7R ) , CD185 ( CXCR5 ) , CD279 ( PD1 ) , GL-7 , and IgD . For intracellular proteins , cells were fixed and permeabilized using transcription factor staining buffer set ( eBioscience , San Diego , CA ) , then stained with fluorochrome labelled anti-mouse FOXP3 and/or TBX21 . For cytokine production , cells were stimulated with Phorbol 12-myristate 13-acetate and ionomycin for 4 hr ( 50 ng/ml and 500 ng/ml , respectively; Sigma-Aldrich , St . Louis , MO ) , treated with Brefeldin A for 2 hr ( 10 μg/ml; Sigma-Aldrich ) , fixed ( 2% formaldehyde; Sigma-Aldrich ) , permeabilized ( 0 . 25% Saponin; Sigma/Aldrich ) , and stained with fluorochrome-labelled anti-mouse IFN-ɣ , IL-2 and/or IL-17A . For IL-2Rα induction , naive CD4+ CD44low CD25- cells were purified from pooled lymph nodes and spleens using a FACS Aria Cell Sorter ( >98% purity; BD Biosciences , San Diego , CA ) . These were stimulated with plate-bound anti-CD3 ( 10 μg/ml; Clone 17A2 ) and soluble anti-CD28 ( 1 μg/ml; Clone 37 . 51 ) in the presence of soluble anti-mouse IL-2 , IL-4 and IFN-ɣ ( 10 μg/ml each; Clones S4B6 , BVD6-24G2 and XMG1 . 2; BioXcell , West Lebanon , NH ) for 18 hr , then treated with human IL-2 ( 100 units/ml; NIH/NCI BRB Preclinical Repository ) or mouse IL-6 ( 20 ng/ml; eBioscience ) for 18 hr and stained with fluorochrome labelled anti-mouse CD25 . For tyrosine-phosphorylated STAT5 , splenocytes were treated directly ex vivo with human IL-2 ( 100 units/ml ) or mouse IL-7 ( 20 ng/ml; eBioscience ) for 1 hr , or stimulated with anti-CD3 and anti-CD28 in the presence of anti-mouse IL-2 for 18 hr , then pulsed with human IL-2 for 1 hr ( 100 units/ml ) . These were then fixed with 2% formaldehyde , permeabilized with 100% methanol and stained with Alexa Fluor 647-labelled anti-human/mouse pY694 STAT5 ( Clone 47; BD Biosciences ) in conjunction with fluorochrome labelled anti-mouse CD3ε , CD4 , CD25 , CD44 , CD127 and/or FOXP3 . Total STAT5 protein was measured in splenocytes directly ex vivo or following retroviral transduction of purified CD4+ T cells ( described below ) . In both cases , cells were fixed with 2% formaldehyde , permeabilized with 100% methanol , then stained with a rabbit polyclonal IgG that recognizes both STAT5A and STAT5B ( sc-835; Santa Cruz Biotechnology , Santa Cruz , CA ) in conjunction with fluorochrome labelled anti-mouse CD3ε , CD4 , CD25 , CD44 , CD127 , ( IL-7R ) , CD185 ( CXCR5 ) , CD279 ( PD1 ) , IL-17A and/or FOXP3 . Phycoerythrin-labelled goat anti-rabbit IgG was used for detection ( ac-3739; Santa Cruz Biotechnology ) . Normal rabbit IgG was used as a negative control ( ac-2027; Santa Cruz Biotechnology ) . All fluorochrome-labelled antibodies were purchased from eBioscience , BD Biosciences or Biolegend ( San Diego , CA ) , unless noted otherwise . Data were collected on a FACSverse cytometer ( BD Biosciences ) and analyzed using FlowJo software ( FlowJo LLC , Ashland , OR ) . Compiled cytometry data are presented as scatter plots where each element represents a single replicate ( horizontal line indicates the mean ) , or box plots where the the fold change for each replicate was calculated relative to WT controls and log 2 transformed ( horizontal line indicates the mean and whiskers indicate minimum and maximum values ) . Cells were maintained in supplemented tissue culture medium ( RPMI-1640 with 10% fetal calf serum , 1% sodium pyruvate , 1% nonessential amino acids , 0 . 1% β-Mercaptoethanol , 100 U/ml penicillin , 100 mg/ml streptomycin; Life Technologies , Grand Island , NY ) and cultured at a density of 0 . 25–0 . 5 x 106 cells/ml in flat bottomed 96 well plates ( 200 ml/ well; Sigma/Costar , St . Louis , MO ) . For in vitro suppression assays , CD4+ CD25high Neuropilin+ Treg cells were sorted from WT and one-allele Stat5a- or Stat5b-deficient mice . Naive , CD4+ CD44low CD25- responder cells were sorted from congenic CD45 . 1 mice and labelled with Carboxyfluorescein succinimidyl ester ( CFSE; Sigma-Aldrich ) . CD11c+ antigen presenting cells ( APCs ) were purified from WT mice using positive selection beads ( Miltenyi Biotec ) . 5 x 104 CD4+ responder cells were stimulated with soluble anti-mouse CD3ε ( 1 μg/ml ) in round bottom 96-well plates containing 1 x 104 APCs and varying numbers of Treg cells , ranging from 5 x 104 ( 1:1 ratio ) to 1 . 56 x 103 ( 1:32 ratio ) . After 96 hr , cells were stained with fluorochrome-labelled anti-mouse CD4 , CD45 . 1 , and CD25 . Percent suppression was calculated relative to WT controls and reflects the percentage of responder cells exhibiting at least one cell division . For 'Treg only' cultures , cells were stimulated with anti-CD3 and anti-CD28 in the presence human IL-2 ( 100 units/ml ) for 72 hr . For iTreg differentiation , naive CD4+ CD44low CD25- cells were sorted and cultured for 72 hr in the presence anti-CD3 , anti-CD28 , human TGF-β ( 10 ng/ml; R&D Systems , Minneapolis , MN ) , human IL-2 ( 100 units/ml ) and anti-mouse IL-2 , IL-4 and IFN-ɣ . Cell sorting was used to purify cells from pooled lymph nodes and spleens of WT and one-allele Stat5a- or Stat5b-deficient mice ( >99% purity ) . Ex vivo groups included naive T cells ( CD4+ CD44low CD25- ) and Treg cells ( CD4+ CD25high Neuropilin+ ) . In vitro groups included naive T cells that were treated with mouse IL-7 for 18 hr , effector T cells that were stimulated with anti-CD3 and anti-CD28 in the presence of human IL-2 for 72 hr , and induced Treg cells . All cultures included anti-mouse IL-2 , IL-4 and IFN-ɣ ( 10 μg/ml each ) . Equal numbers of cells ( 0 . 5–2 . 5 x 105 ) were collected for each replicate . These were lysed in Trizol reagent and total RNA isolated by phenol-chloroform extraction with GlycoBlue as co-precipitant ( 7-15 μg per sample; Life Technologies ) . Single-end libraries were prepared with 0 . 1–0 . 5 μg of total RNA using the TruSeq RNA Sample Preparation Kit V2 and sequenced for 50 cycles with a HiSeq 2500 instrument ( 4–6 samples multiplexed per lane; Illumina , San Diego , CA ) . 50 bp reads were then mapped onto mouse genome build mm9 using TopHat and further processed using Cufflinks ( Garber et al . , 2011 ) . 2–3 biological replicates were sequenced per genotype for every cell type and culture condition . QC-passing read counts are presented in Supplementary file 1 . Datasets are normalized based on RPKM ( reads per kilobase exon model per million mapped reads ) and purged of micro-RNAs , sno-RNAs and sca-RNAs . To minimize fold-change artifacts caused by low abundance transcripts , a small offset ( 0 . 2–0 . 3; equal to the second quartile of each dataset ) was added to all RPKM values ( Warden , Yuan , and Wu , 2013 ) . When multiple fragments were detected for a single gene , only the most abundant ( i . e . highest average RPKM across all 3 genotypes ) was considered for downstream analyses . Transcripts with RPKM values of less than 1 for all genotypes within a given cell type or condition were excluded . Fold change and variance across genotypes and biological replicates were calculated using EdgeR ( Robinson , McCarthy , and Smyth , 2009 ) . Transcripts were classified as differentially expressed if they exhibited a >1 . 5 fold change and significant pairwise variance ( p<0 . 05 ) relative to WT controls . The 500 transcripts with greatest variance within each cell type or condition were used for multidimensional scaling ( MDS ) using the RobiNA software package ( Lohse et al . , 2012 ) . A 'paralog preference' scale was devised to illustrate the relative impact of Stat5a - or Stat5b- deficiency . First , all transcripts that were differentially expressed in Stat5a- or Stat5b- deficient cells ( relative to WT controls ) were pooled to generate a single list of STAT5-regulated genes for each cell type or condition . Next , the absolute fold change was calculated and multiplied by the higher of the two RPKMs ( WT or KO ) , thereby generating a 'paralog score' . Note that the use of absolute fold change negates the distinction between up- and down-regulated genes , while the multiplication step improves the score for high-abundance transcripts . The paralog score for STAT5B was then divided by the paralog score for STAT5A and the resulting 'preference score' was log 2 transformed so that transcripts which are more impacted by the loss of STAT5B are assigned positive values while those which are more impacted by the loss of STAT5A are assigned negative values . All transcripts were then segregated into 12 bins according to preference scores ( Bin 1 includes values of less than -5 , Bin 2 ranges from -5 to -4 , and so on ) . Data are displayed as histograms and the median preference score is indicated . To identify ‘paralog-specific’ transcripts , we first identified those exhibiting >1 . 5 fold change and significant variance ( p<0 . 05 ) when comparing Stat5a- or Stat5b-deficient cells directly to one another . Next , we refined this list by stipulating that transcripts must be differentially expressed in one KO relative to WT controls ( >1 . 5 fold change ) but not in the other ( <1 . 2 fold change ) . Rare transcripts with opposite expression patterns ( i . e . up-regulated in one genotype but down-regulated in the other ) were excluded . Data are presented as pie charts . All volcano plots , XY plots , histograms and pie charts were generated with the DataGraph software suite ( Visual Data Tools , Inc . ) . Heat maps were generated with Multi Experiment Viewer ( MeV; J . Craig Venter Institute , La Jolla , CA ) . Genome browser files ( BigWig format ) were processed to remove intronic reads using TopHat and are displayed with the Integrative Genomics Viewer ( IGV; Broad Institute , Cambridge , MA ) . GSEA analysis was performed as described ( Subramanian et al . , 2005 ) . Unabridged RNA-seq datasets were used in conjunction with the following user-generated Gene Sets: 1 ) Treg signature genes ( 132 members ) ( Hill et al . , 2007 ) , 2 ) IL-2-regulated , STAT5A-dependent genes ( 258 members ) ( from the comparison of WT and 'one copy' Stat5a-deficient T cells; Figure 6 ) , 3 ) IL-2-regulated , STAT5B-dependent genes ( 329 members ) ( from the comparison of WT and 'one copy' Stat5b-deficient T cells; Figure 6 ) . Enrichment score curves and member ranks were generated by the GSEA software ( Broad Institute ) . Normalized enrichment score ( NES ) , false discovery rate ( FDR ) and nominal p Value is shown on each plot . See Supplementary file 2 for RPKM , fold change and p values for all experimental groups and conditions , Supplementary file 3 for paralog preference calculations and Supplementary file 4 for paralog-specific genes . Cell sorting was used to purify naive CD4+ CD44low CD25- cells from WT , Stat5a/bhet and two-allele Stat5a- or Stat5b-deficient mice ( >99% purity ) . These were stimulated with anti-CD3 and anti-CD28 in the presence of human IL-2 for 48 hr ( 10 U/ml with anti-mouse IL-2 , IL-4 and IFN-ɣ ) , then pulsed with IL-2 ( 100 U/ml ) for one hour before fixing with 1% formaldehyde . They were then lysed ( 1 x 107 cells/sample ) , sonicated and immuno-precipated using a polyclonal rabbit anti-mouse IgG that recognizes both STAT5A and STAT5B ( ab7969; Abcam , Cambridge , MA ) . Recovered STAT5-bound DNA fragments , along with un-precipitated ‘input controls’ , were blunt-end ligated to adaptors and single-end libraries constructed using the NEBNext ChIP-Seq Library Prep for Illumina kit ( New England Biolabs , Ipswich , MA ) . Sequencing was performed on a HiSeq 2500 instrument ( 50 cycles; Illumina ) and short reads ( 50 bps ) aligned using Bowtie ( Langmead et al . , 2009 ) . Non-redundant reads were mapped to the mouse genome ( mm9 ) and aggregated into peaks and using MACS 1 . 4 . 2 ( Feng et al . , 2012 ) . Only peaks with >3 fold enrichment over background and p values <0 . 00005 were called . Positive false discovery rates , or q-values , were calculated empirically for each peak and all were below 0 . 2% ( Storey , 2003 ) . 2 biological replicates were sequenced per genotype . 'bamCorrelate' from deepTools 1 . 5 was used to calculate Spearman's rank correlation coefficients as a measure of inter-replicate variability ( WT=0 . 82 , Stat5a/bhet=0 . 83 , Stat5a-deficient=0 . 82 , Stat5b-deficient=0 . 81; all pairwise p-values <2 . 2 . x 10-16 ) ( Ramirez et al . , 2014 ) . Read depth for all replicates is presented in Supplementary file 1 . Peaks were annotated to the nearest known gene using HOMER ( Zhang et al . , 2008; Heinz et al . , 2010 ) . Localization was calculated as the percentage of peaks found within 10 kb intervals of the nearest transcriptional start sites and plotted as histograms . Direct comparison between experimental groups ( i . e . peak overlap ) was done with PAPST ( Bible et al . , 2015 ) . Circos plot was generated by inputing the number of shared peaks between experimental groups to the Circos Table Viewer ( http://mkweb . bcgsc . ca/tableviewer ) ( Krzywinski et al . , 2009 ) . Violin plot was generated by inputing tag numbers to the online BoxPlotR applet ( http://boxplot . tyerslab . com ) ( Editorial , 2014 ) . Transcription factor motif analysis was done with HOMER using an 'in house' database generated by applying de novo motif discovery to published ChIP-seq datasets . Genome browser files are displayed with IGV . Strand-specific RNA sequencing data were generated by the ENCODE Transcriptome Group from Cold Spring Harbor Laboratories ( U . S . A . ) and the Center for Genomic Regulation ( Spain ) ( https://genome . ucsc . edu/cgi-bin/hgTrackUi ? hgsid=424400999_9OI4vJsT1sakRAPyi9mNSC7V81zc&g=wgEncodeCshlLongRnaSeq ) . DNAseI hyper-sensitivity data was generated by the University of Washington ENCODE group ( https://genome . ucsc . edu/cgi-bin/hgTrackUi ? hgsid=424401115_qgaAWZ6Xs38F1laFE3UuHnvAG7AS&g=wgEncodeUwDnase ) . Data are used in accordance with the ENCODE data release policy ( Yue et al . , 2014 ) and visualized with the UCSC genome browser , focusing on the mouse Stat5a/b locus ( chr11:100642045-100746483 ) . Retroviral vector expressing phosphatase-insensitive STAT5A was generated as described ( Zhu et al . , 2003 ) . Plasmids were transfected into Phoenix packaging cells using Lipofectamine ( Invitrogen ) and the resulting viral supernatants used to transduce CD4+ cells from WT , Stat5a/bhet or one-allele Stat5b-deficient mice . These were stimulated ( anti-CD3/CD28 ) in the presence of anti-mouse IL-2 for 48 hr , exposed to viral supernatant for 1 hr ( at 2200 rpm , 18°C ) , and cultured for an additional 48 hr in the presence of human IL-2 ( 100 U/ml ) . For some experiments , cells were cultured under iTreg ( 10 ng/ml human TGF-β ) or Th17 ( 2 . 5 ng/ml human TGF-β + 20 ng/ml mouse IL-6 ) polarizing conditions before and after transduction ( both in the presence of anti-mouse IL-4 and IFN-ɣ ) . For RNA-seq , 1-2 x 105 cells expressing the bicistronic transduction marker ( human NGFR ) were purified by cell sorting . Transcripts that were significantly impacted ( >1 . 5 fold change , p<0 . 05 ) by ectopic STAT5A relative to empty vector were enumerated using EdgeR . Transcriptome data for CD4+ naive and Treg cells was sourced from: 1 ) Immunological Genome Project ( mouse microarrays: http://www . immgen . org ) , 2 ) EMBL-EBI Expression Atlas ( mouse RNA-seq: https://www . ebi . ac . uk/gxa/experiments/E-MTAB-2582 ) , 3 ) BioGPS Primary Cell Atlas ( human microarrays: http://biogps . org/dataset/BDS_00013/ ) , and 4 ) our RNA-seq catalogue ( described above; Figure 6 ) . Normalized expression values ( microarray signal intensity or FPKM ) for Stat5a and Stat5b were first divided by one another to generate a paralog ratio which was then converted to a percentage ( % total STAT5 mRNA accounted for by each paralog ) and presented as pie charts . Total STAT5 protein was measured by flow cytometry in naive ( CD3ε+ CD4+ CD44low IL-7R+ ) , Treg ( CD3ε + CD4+ FOXP3+ ) , Tfh ( CD3ε + CD4+ PD1+ CXCR5high ) and effector/memory ( CD3ε +CD4+CD44high ) T cells from one- or three-allele Stat5a- or Stat5b-deficient mice , as well as Stat5a/bhet mice and WT controls . Mean fluorescence intensity ( MFI ) was first divided by the baseline ( i . e . WT controls ) to generate 'fold change' values which , in turn , were divided across Stat5a- and Stat5b-deficient genotypes bearing the same total number of alleles . The resulting paralog ratios ( one for one-allele cells and one for three-allele cells ) were then averaged , converted to a percentage ( % total STAT5 protein accounted for by each paralog ) and presented as pie charts . Unpaired ANOVA was used to quantify statistical deviation between experimental groups . In all figures , black asterisks denote significant differences ( p<0 . 05 ) between the indicated group and WT controls . Orange asterisks denote significant differences between Stat5a- and Stat5b-deficient mice bearing the same total number of STAT5 alleles . All sequencing data have been deposited to the Gene Expression Omnibus under the accession number GSE77656 . | The immune system in mammals is one of the most complex networks in the animal kingdom . One way that its many components communicate is via proteins called cytokines , which are released by cells and detected by receptors on the surface of other cells . This leads to the activation of signals inside the responding cells that alter the activity of genes and , ultimately , direct how they behave . STAT5 is a signal protein that is activated when certain cytokines bind to receptors on the cell surface . Consequently , it is an attractive target for drug therapies that seek to alter immune responses and there is keen interest in understanding how it works . It is an unusual protein in that there are two versions – termed STAT5A and STAT5B – that are produced by two separate genes . Together , STAT5A and STAT5B are fundamental to the immune system but there is considerable debate about whether they perform the same job or have distinct roles . Villarino et al . used a combination of genetic and genomic approaches to investigate how both versions of STAT5 work in mice . The experiments show that STAT5B plays a much bigger role in immune cells than STAT5A . Unexpectedly , the experiments indicate that the disparity is not due to differences in protein activity , but is caused by differences in the amount of these proteins in cells . Villarino et al . ’s findings resolve longstanding questions about the relationship between STAT5A and STAT5B within the immune system . A logical next step is to find the molecular mechanisms responsible for causing different amounts of STAT5A and STAT5B to be produced in immune cells . Future work will also compare the roles of STAT5A and STAT5B in non-immune cells and explore whether it might be possible to develop therapies that specifically target one version and not the other . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"&",
"methods"
] | [
"cell",
"biology",
"immunology",
"and",
"inflammation"
] | 2016 | Signal transducer and activator of transcription 5 (STAT5) paralog dose governs T cell effector and regulatory functions |
Lipoprotein RcsF is the OM component of the Rcs envelope stress response . RcsF exists in complexes with β-barrel proteins ( OMPs ) allowing it to adopt a transmembrane orientation with a lipidated N-terminal domain on the cell surface and a periplasmic C-terminal domain . Here we report that mutations that remove BamE or alter a residue in the RcsF trans-lumen domain specifically prevent assembly of the interlocked complexes without inactivating either RcsF or the OMP . Using these mutations we demonstrate that these RcsF/OMP complexes are required for sensing OM outer leaflet stress . Using mutations that alter the positively charged surface-exposed domain , we show that RcsF monitors lateral interactions between lipopolysaccharide ( LPS ) molecules . When these interactions are disrupted by cationic antimicrobial peptides , or by the loss of negatively charged phosphate groups on the LPS molecule , this information is transduced to the RcsF C-terminal signaling domain located in the periplasm to activate the stress response .
The outer membrane ( OM ) of Gram-negative bacteria is an asymmetric bilayer with lipopolysaccharide ( LPS ) and phospholipids in the outer and inner leaflets , respectively ( Silhavy et al . , 2010 ) . LPS is a glycolipid that consists of three domains: lipid A , the core and the O-antigen ( Raetz and Whitfield , 2002 ) . Several sugars in lipid A and the core are phosphorylated conferring negative charge to the LPS molecule . In the OM , these negatively charged groups are bridged by divalent cations , which help to establish strong lateral interactions between LPS molecules . In addition to stabilizing the OM , these lateral interactions contribute to the unique barrier properties of the OM making it impermeable to hydrophobic compounds , detergents and dyes ( Nikaido , 2003 ) . The OM also protects Gram-negatives from the host innate immunity factors and antibiotics limiting their effectiveness . In order to disrupt the OM , many organisms produce cationic antibacterial peptides ( CAMPs ) that bind LPS ( Hancock and Diamond , 2000 ) . As a result of this binding , the OM is permeabilized and this not only facilitates further uptake of the CAMPs but also sensitizes Gram-negatives to antibiotics and host-factors , including lysozyme . For this reason , several CAMPs are “last hope” antibiotics against antibiotic-resistant Gram-negative bacteria ( Li et al . , 2006 ) . Because of the importance of OM integrity and barrier function for survival , Gram-negative bacteria have developed several envelope stress responses to monitor and combat environmental insults . One such envelope response , Rcs ( Regulator of Capsule Synthesis ) , is activated strongly by OM and PG stress ( Majdalani and Gottesman , 2005 ) . Rcs controls the expression of capsule exopolysaccharides that are exported to the cell surface and help to stabilize the OM ( Gottesman et al . , 1985 ) . In addition , Rcs downregulates flagella expression ( Francez-Charlot et al . , 2003 ) , shifting bacteria from planktonic to a biofilm growth mode ( Ferrières and Clarke , 2003; Latasa et al . , 2012 ) , which is often associated with further increased resistance . Rcs is conserved in Enterobacteriaceae and , for many enteric pathogens , it is important for virulence and/or survival in the host ( Erickson and Detweiler , 2006; Hinchliffe et al . , 2008 ) . Rcs is one of the most complex signal transduction pathways in bacteria involving at least seven proteins in four different cellular compartments . RcsF is an OM lipoprotein that acts as a sensory component ( Majdalani and Gottesman , 2005 ) . Unlike most lipoproteins in E . coli , RcsF is anchored to the outer leaflet of the OM by its lipid moiety and contains an N-terminal domain that is surface exposed ( Konovalova et al . , 2014 ) . The short , hydrophilic transmembrane domain of RcsF is threaded through the lumen of β-barrel proteins ( OMPs ) exposing the C-terminal signaling domain in the periplasm ( Konovalova et al . , 2014 ) . The inner membrane ( IM ) protein RcsC is a hybrid histidine kinase , which autophosphorylates and passes the phosphate through the IM phosphotransferase protein RcsD to a cytoplasmic DNA-binding response regulator RcsB ( Stout and Gottesman , 1990; Takeda et al . , 2001; Majdalani and Gottesman , 2007 ) . RcsB , either alone or in combination with other regulators , such as RcsA ( Stout et al . , 1991 ) , BglJ ( Venkatesh et al . , 2010 ) or GadE ( Krin et al . , 2010 ) regulates expression of target genes . In addition , IM protein IgaA negatively regulates the activity of RcsC ( Cano et al . , 2002; Domínguez-Bernal et al . , 2004 ) . Based on genetic analysis as well as some interaction studies it has been proposed that RcsF can interact directly with the periplasmic domain of IgaA to alleviate inhibition of RcsC ( Domínguez-Bernal et al . , 2004; Cho et al . , 2014 ) . Although the signal transduction pathway itself is reasonably well characterized , how RcsF senses the many different envelope defects that induce this system remain poorly understood . The fact that LPS is found exclusively in the outer leaflet of the OM and that RcsF is required to sense LPS structural defects prompted us to search for a surface-exposed domain . Our discovery that RcsF exists in a transmembrane complex with OMPs suggested that this interlocked structure is required for sensing the LPS defects ( Konovalova et al . , 2014 ) . However , another study , which focused primarily on Rcs activated by PG stress caused by A22 or mecillinam treatment , proposed a model in which the RcsF/OMP complex has no function in RcsF sensing and signaling and only serves an inhibitory role during steady state growth . This model was based on the observation that Rcs induction by A22 depends on newly synthesized RcsF ( Cho et al . , 2014 ) . Here , we show that the RcsF/OMP complex is functional and that LPS defects are sensed directly by the surface-exposed domain of RcsF .
Several mutations in the LPS biosynthesis pathway that result in the production of LPS molecules with altered structure are known to activate Rcs ( Parker et al . , 1992 ) . However , the use of chemical inducers avoids phenotypic adaptation and enables kinetic analysis , which can provide important insights . Therefore , we sought a small molecule that could be used to test the function of RcsF/OMP complexes in sensing LPS perturbations . Polymyxin B ( PMB ) is a CAMP with bactericidal activity and its mechanism of action is well established ( Daugelavicius et al . , 2000 ) . When used at the minimal inhibitory concentration ( MIC , 2–8 µg/ml ) , PMB binds to LPS and destabilizes the OM outer leaflet resulting in marked increase in OM permeability . When cells are incubated with much higher concentrations of PMB , it can also integrate into the IM causing a lethal disruption of the membrane potential . PMB is a known RcsF-dependent inducer of the Rcs pathway ( Farris et al . , 2010 ) . In the following experiments , we used 0 . 5 µg /ml PMB to treat cells in mid-log phase ( OD600 of 0 . 5 , 5*108 cells/ml ) . At this cell density , the MIC of PMB was determined to be 8 µg /ml . Therefore , the concentration of PMB we used to induce the Rcs system is more than 10 fold below the MIC value and growth is not affected under these conditions . We confirmed that 0 . 5 µg/ml PMB causes OM , but not IM , stress in our strain background using two routine assays developed to study the effect of CAMPs on the bacterial envelope ( Figure 1A and B ) ( Loh et al . , 1984; Wu et al . , 1999 ) . The NPN ( 1-N-phenylnaphthylamine ) uptake assay is used to quantitatively monitor OM permeability caused by CAMPs ( Loh et al . , 1984 ) . NPN is an environmentally sensitive fluorescent dye that fluoresces in the membrane environment but not in solution . Due to its hydrophobic nature , it cannot penetrate Gram-negative OM due to the presence of LPS . However , when CAMPs compromise the OM , NPN enters the cell and integrates into membranes resulting in an increase in fluorescence . When mid-log cells were incubated with 0 . 5 µg /ml PMB we observed only a small increase in fluorescence , about 1 . 5 fold , compared to a 6-fold increase when cell were treated with 8 µg /ml PMB ( Figure 1A ) . 10 . 7554/eLife . 15276 . 003Figure 1 . PMB causes a specific OM defect . ( A ) PMB at 0 . 5 µg/ml causes a slight OM permeability defect based on increased uptake and fluorescence of NPN dye . Graphs represent mean normalized end-point fluorescence +/-SD , n=3 . ( B ) This concentration of PMB does not cause depolarization of the IM . Unlike gramicidin , PMB is unable to release PMF-dependent DiSC3 ( 5 ) dye . Graphs represent mean normalized end-point fluorescence +/-SD , n=3 . ( C ) Kinetics of Rcs induction upon PMB and Glb treatment at the mRNA ( upper panel ) and protein level ( lower panel ) . Induction was monitored using a chromosomal PrprA-lacZ reporter by qRT-PCR or β-galactosidase assays . For mRNA quantification , graphs represent relative expression values normalized to a no treatment control for each time point +/- SD . β-galactosidase activity represent mean Vmax normalized to OD600 , +/- SEM , n=3 . ( D ) PMB induces Rcs but not the Cpx or SigmaE stress responses . Induction was monitored by following the relative expression of PrprA-lacZ ( Rcs ) , cpxP ( Cpx ) or rpoE ( SigmaE ) by qRT-PCR . Graphs represent mean +/- SEM , n=3 . ( E ) PMB does not cause OMP assembly defects based on immunoblot analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 00310 . 7554/eLife . 15276 . 004Figure 1—figure supplement 1 . Glb induces Rcs but not the SigmaE stress response . Induction was monitored by following the expression of PrprA-lacZ ( Rcs ) and rpoE ( SigmaE ) by qRT-PCR . Graphs represent relative expression values normalized to a no treatment control for each time point , mean +/- SEM , n=3DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 004 The diSC3 ( 5 ) ( Dipropylthiadicarbocyanine iodide ) release assay is based on a membrane potential-sensitive fluorescent dye ( Wu et al . , 1999 ) . DiSC3 ( 5 ) accumulates in the IM in a proton motive force ( PMF ) -dependent manner , where it self-quenches the fluorescence . However , when PMF is compromised , diSC3 ( 5 ) is released from the IM resulting in increased fluorescence . We treated diSC3 ( 5 ) -labeled cells with 0 . 5 µg /ml PMB or an MIC ( 12 . 5 µg /ml ) of Gramicidin , a CAMP known to be PMF uncoupler . Gramicidin A but not PMB caused an increased fluorescence due to diSC3 ( 5 ) release ( Figure 1B ) . Taken together , these assays demonstrate that at low concentrations PMB does not cause IM depolarization and generates only a small increase in OM permeability , consistent with previous results ( Daugelavicius et al . , 2000 ) We monitored the induction kinetics of the Rcs system in response to PMB using a chromosomal PrprA-lacZ reporter fusion ( Majdalani et al . , 2002 ) . rprA encodes a small regulatory RNA that stimulates translation of the mRNA for the stationary phase σ factor RpoS ( Majdalani et al . , 2002 ) . Expression of rprA is regulated exclusive by RcsB and PrprA-lacZ is used as a specific reporter for Rcs stress response activation ( Majdalani and Gottesman , 2007 ) . For this experiment , we grew the MC4100 PrprA-lacZ strain ( from now on WT , the parent for all strains ) to midlog phase . We then added PMB and followed the Rcs induction over time by monitoring expression of PrprA-lacZ reporter using qRT-PCR and β-galactosidase assays . PMB causes a strong and almost immediate induction of Rcs , as quickly as 2 min on the RNA level and 10 min on the protein level ( Figure 1C ) . In order to test whether PMB induces other envelope stress responses or is specific for Rcs , we followed the activity of two other major envelope stress responses , Sigma E and Cpx by monitoring the expression of well-established markers , rpoE ( Mutalik et al . , 2009 ) and cpxP ( Danese and Silhavy , 1998 ) . Figure 1D shows using qRT-PCR that PMB did not induce either of these stress responses , demonstrating that PMB induces only Rcs . We also monitored OMP levels during PMB treatment to see whether PMB causes OMP assembly defects . Levels of the three major OMPs did not change upon PMB treatment ( Figure 1E ) . Because we did not observe OMP assembly defects and because the Sigma E response , a sensitive monitor of OMP assembly , was not induced we conclude that low levels of PMB do not inhibit the Bam complex . Taken together , we conclude that at these sub-MIC levels , PMB causes a specific OM defect . Globomycin ( Glb ) is an inhibitor of LspA , the lipoprotein signal peptidase ( Inukai et al . , 1978; Yamagata et al . , 1983; Dev et al . , 1985 ) . Glb prevents maturation of lipoproteins resulting in the accumulation of OM lipoproteins in the outer leaflet of the IM . It is well established that , when RcsF accumulates in the outer leaflet of the IM , due to Lol-avoidance mutations that alter the signal sequence or defects in lipoprotein maturation and/or export , it strongly induces the Rcs response ( Shiba et al . , 2004; 2012; Tao et al . , 2012 ) . Indeed , we observed strong Rcs induction in response to 5 µM Glb ( 0 . 5 MIC ) . This induction required at least 15 min on the RNA level and 20 min on the protein level ( Figure 1C ) . Clearly , the kinetics of Rcs induction are much slower with Glb than with PMB . Glb will prevent assembly of essential lipoproteins involved in both LPS and OMP assembly; however , depletion of these proteins to levels low enough to interfere with these assembly processes requires generations ( Malinverni et al . , 2006; Wu et al . , 2006 ) . Indeed , Glb does not induce the Sigma E response even after 60 min exposure ( Figure 1—figure supplement 1 ) . Therefore , we conclude that Glb is a chemical inducer of Rcs that is independent of OM damage . There is no reciprocal transport of lipoproteins from the OM to the IM; therefore , only newly synthesized OM lipoproteins accumulate in the IM under Glb treatment . Therefore , the kinetics of the Glb-induced Rcs response can also serve as a temporal marker for induction due to mislocalization of newly synthesized RcsF . The observation that Glb induced Rcs with slower kinetics than with PMB suggested that induction with PMB does not require newly synthesized RcsF . To test this directly , we used the antibiotic Kasugamycin ( Ksg ) , which is an inhibitor of translation initiation ( Okuyama et al . , 1971 ) . After 15 min pre-treatment with Ksg we added PMB or Glb and followed the expression of PrprA-lacZ by qRT-PCR ( Figure 2 ) . As expected Ksg completely abolished Rcs induction in response to Glb ( Figure 2 , lower panel ) , however Ksg did not block the Rcs response to PMB , demonstrating that Rcs induction is independent of de novo protein synthesis and does not require newly synthesized RcsF ( Figure 2 , upper panel ) . Since RcsF is transported to the OM immediately after it is translocated from the cytoplasm , this result suggests that RcsF molecules already located in the OM are required to respond to PMB treatment . 10 . 7554/eLife . 15276 . 005Figure 2 . The PMB-induced Rcs response is independent of de novo protein synthesis . Cell cultures were pretreated with Ksg to inhibit protein synthesis for 15 min prior to addition of PMB or Glb . Rcs induction was then monitored by qRT-PCR . Graphs represent relative expression values normalized to a no treatment control for each time point , mean +/- SEM , n=3DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 005 RcsF forms an interlocked complex with multiple OMPs , such as OmpA , OmpC and OmpF ( Konovalova et al . , 2014 ) . However , it is unclear how much uncomplexed RcsF is present in the OM . Since OmpA is the major RcsF-interacting partner ( Konovalova et al . , 2014 ) , we reasoned that loss of OmpA will lead to significant reduction in RcsF/OMP complexes ( Figure 3A and B ) and , if this complex senses LPS defects , this reduction should affect the response to PMB but not to Glb . 10 . 7554/eLife . 15276 . 006Figure 3 . RcsF/OMP complexes are required for sensing OM stress . ( A ) Topology and assembly pathway of RcsF/OMP complexes ( based on ( 14 ) ) . The lipidated N-terminus of RcsF is anchored in the outer leaflet of the OM exposing residues 16–49 on the cell surface . The transmembrane segment ( residues 50–65 ) of RcsF is threaded through the lumen of the OMP exposing the C-terminal domain in the periplasm . RcsF/OMP complexes are assembled by the Bam machine . Not all OMPs are complexed with RcsF . The effect of different mutants used in this study on the assembly of RcsF/OMP complexes is shown . ( B ) The effect of ompA , bamE and rcsF_A55Y mutations on RcsF crosslinking to BamA and OmpA ( upper panel ) and the total RcsF , BamA and OmpA levels based on immunoblot analysis . ( C ) The ompA , bamE and rcsF_A55Y mutants do not respond to PMB ( upper panel ) but respond to Glb ( lower panel ) treatment based on expression of PrprA-lacZ . Graphs represent mean β-galactosidase activity +/- SEM , n=3 ( D ) The bamE and rcsF_A55Y mutants result in decreased PrprA-lacZ . Graphs represent mean β-galactosidase activity +/- SEM , n=3 . Corresponding OD600 graphs and untreated controls are shown in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 00610 . 7554/eLife . 15276 . 007Figure 3—figure supplement 1 . Growth and kinetcs of Rcs induction in the assembly mutants including untreated cotrols . ( A ) MIC values for the RcsF/OMP assembly defective strains . ( B ) Kinetics of PrprA-lacZ expression in RcsF/OMP assembly defective strains upon treatment with PMB , Glb ( same as Figure 3C ) and untreated controls together with corresponding growth curves . Graphs represent mean β-galactosidase activity or OD600 +/- SEM , n=3DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 00710 . 7554/eLife . 15276 . 008Figure 3—figure supplement 2 . Growth phenotype of the assembly mutants . ( A ) Growth curves of the assembly mutants in waaP background . Strains were grown at 37° C in 1 ml of LB in 24 well plate and OD600 was monitored with a Biotek Synergy 1 plate reader . Note , waaP ompA strains display a synthetic growth phenotype . Graphs represent mean OD600 +/- SEM . ( B ) Plate phenotype of the assembly mutants in the WT and waaP backgrounds . Strains were grown on LB agar at 37⁰ C overnight . Note , that the waaP mutation confers a RcsF-dependent mucoid phenotype . Introduction of a bamE mutation in waaP strain results in a loss of this phenotype . pZS21::rcsF complements waaP rcsF mutant but does not confer a gain-of-function phenotype in the WT background . pZS21::rcsF_A55Y failed to complement the mucoid phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 008 We followed the kinetics of Rcs induction in response to PMB and Glb in the WT and the ompA mutant by monitoring normalized LacZ activity produced by the PrprA-lacZ reporter over time ( Figure 3C ) . PMB treatment did not induce Rcs in the ompA mutant ( Figure . 3C , upper panel ) . In contrast , Rcs was induced in the ompA mutant upon Glb treatment with similar kinetics to that observed for the WT ( Figure 3C , lower panel ) suggesting that ompA is required specifically for PMB-induced OM stress . OmpA is one of the most abundant OMPs in E . coli . To confirm that lack of Rcs response to PMB is the result of loss of RcsF/OMP complexes and not due to damage caused by the loss of the OmpA protein itself , we sought to identify RcsF/OMP assembly defective mutants that would lack RcsF/OmpA complexes but express both the individual proteins . Here we describe two such mutants ( Figure 3A and B ) . RcsF residue A55 is within the region of RcsF predicted to reside in the lumen of OMPs based on the results of site-specific photo-crosslinking ( Figure 3A ) ( Konovalova et al . , 2014 ) . In the course of constructing pBPA variants for site-specific crosslinking , we noticed that the A55-pBPA variant was not functional for LPS sensing and did not crosslink to OMPs . We hypothesized that this residue might be important for RcsF/OMP assembly . Here we report that a mutation rcsF_A55Y results in a significant reduction in the assembly of RcsF/OmpA complexes ( Figure 3B , upper panel ) . Photo-crosslinking also shows that RcsF interacts with BamA and it is thought that this reflects a role for BamA in the assembly of the RcsF/OMP complexes ( Konovalova et al . , 2014 ) . The rcsF_A55Y mutation also results in a significant reduction in the levels of RcsF/BamA complex . bamE encodes one of the non-essential lipoproteins of the Bam complex ( Sklar et al . , 2007 ) . bamE mutants displayed a striking phenotype in RcsF assembly: levels of RcsF/OmpA complexes are severely reduced ( Figure 3B , upper panel ) and increased RcsF/BamA crosslinking is observed ( Figure 3B , upper panel ) . Levels of the individual proteins , RcsF , OmpA or BamA , were not affected in these assembly mutants ( Figure 3B , lower panel ) . Both of the mutants , rscF_A55Y and bamE , failed to activate Rcs in response to PMB induction ( Figure 3C , upper panel ) ; however , both responded to Glb treatment ( Figure 3C , lower panel ) . LacZ activity was lower than the WT , likely due to the lower steady state levels of LacZ activity in the untreated rscF_A55Y and bamE cells . To verify that RcsF/OMP complexes are more generally required for sensing LPS-stress and not specific to PMB , we tested the effect of the ompA , rcsF_A55Y and bamE mutations on Rcs signaling in the LPS biosynthesis mutant waaP ( rfaP ) ( Figure 3D ) . waaP encodes lipopolysaccharide core heptose one ( Hep ( I ) kinase which catalyzes the addition of a negatively charged phosphate group to the LPS core ( Parker et al . , 1992; Yethon et al . , 1998 ) . waaP null mutations induce Rcs approximately 8 fold over the WT , resulting in a mucoid phenotype ( Figure 3—figure supplement 2 ) . Deletion of ompA in waaP background resulted in a strong synthetic interaction and this strain had a growth defect in liquid culture ( Figure 3—figure supplement 2 ) . For this reason , we did not perform experiments with this strain . When the rcsF_A55Y or bamE mutation was introduced in the waaP strain , no growth defect was observed demonstrating that the loss of OmpA not the RcsF/OmpA complex causes this synthetic interaction ( Figure 3—figure supplement 2 ) . Strikingly , both assembly mutations significantly reduced the expression of the PrprA-lacZ reporter in the waaP strain and resulted in loss of the mucoid phenotype ( Figure 3D , Figure 3—figure supplement 2 ) indicating that Rcs was not induced . PrprA-lacZ expression in the rcsF_A55Y strain was still somewhat stimulated by the waaP mutation , likely because the assembly of RcsF/OMP complexes was not completely abolished . Taken together , we conclude that RcsF/OMP complexes are required for sensing LPS defects . Mutations in several LPS biosynthesis genes result in Rcs induction ( Parker et al . , 1992 ) . We systematically analyzed the effect of mutations in non-essential genes in the LPS biosynthesis pathway for the ability to induce Rcs ( Figure 4A , B ) . We found mutations that result in defects in inner core biosynthesis and phosphate modification strongly induce Rcs ( Figure 4B ) . waaCFPG mutations confer a phenotype , known as a deep rough phenotype; these strains are mucoid and unlike the WT are sensitive to detergents , bile salts and hydrophobic antibiotics ( Austin et al . , 1990; Kamio and Nikaido , 1976; Parker et al . , 1992 ) . To differentiate between altered LPS structure and increased permeability as potential inducing signals for Rcs we analyzed several OM biogenesis mutants that also display strong sensitivities to antibiotics without affecting LPS structure . 10 . 7554/eLife . 15276 . 009Figure 4 . RcsF senses alteration in LPS structure . ( A ) Structure and biosynthesis of E . coli K-12 LPS according to ( Raetz and Whitfield , 2002 ) . Non-essential enzymes are shown in bold . ( B ) The effect of mutations in non-essential genes in the LPS biosynthesis pathway on Rcs induction based on PrprA-lacZ expression . ( C and D ) OM permeability is not a physiological inducing signal for Rcs . ( C ) Mutations that cause defects in asymmetry ( pldA mlaA ) or LPS export ( lptD4213 and lptE_R91D K136D ) do not induce Rcs . RcsF is not generally inhibited in these strains because Rcs can still be induced by introducing the waaP mutation . ( D ) Mutations pldA mlaA , lptD4213 and lptE_R91D K136D confer OM permeability defects ( see text for references ) assayed by plating 10-fold serial dilutions of overnight cultures onto LB plates supplemented with antibiotics or detergents . Note , arabinose ( Ara ) is required for growth of lptE_R91D K136D mutant . ( E ) The addition of Mg2+ reduces Rcs signaling in the waaP background in a phoP-independent manner . ( F ) Lipid-truncated PMB derivative , PMBN , induces Rcs in concentration-dependent manner . Graphs B , C , E and F represent mean β-galactosidase activity +/- SEM , n=3DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 009 First , we tested the double mlaA pldA mutant , which knocks out two complementary pathways controlling OM lipid asymmetry ( Dekker , 2000; Malinverni and Silhavy , 2009 ) . This mutant has substantially increased levels of phospholipids in the outer leaflet of the OM ( Malinverni and Silhavy , 2009 ) . Figure 4C shows that Rcs is not induced in this strain . In addition , we tested mutants in the LPS export pathway , lptD4213 and lptE_R91D , K136D ( Ruiz et al . , 2005; Malojčić et al . , 2014 ) . Rcs was also not induced in these strains ( Figure 4C ) . The permeability phenotype of lptE_R91D , K136D is mild but mlaA pldA and lptD4213 are as sensitive to detergents and hydrophobic antibiotics as deep rough mutants , e . g . waaP ( Figure 4D ) . Importantly , RcsF was not generally inhibited in these strains , because it could be activated by introducing the waaP mutation ( Figure 4C ) . These results demonstrate that OM permeability and/or disrupted asymmetry of the OM is not a physiological signal for Rcs . The result above suggested that RcsF senses alterations of LPS structure in waa mutants rather than permeability . Mutations in waaC , waaF and waaG result not only in a truncated core ( Figure 4B ) , but also in the loss of core phosphorylation ( Yethon et al . , 2000; Yethon and Whitfield , 2001 ) because the complete inner core with the first glucose is a substrate for WaaP kinase ( Yethon et al . , 2000; Yethon and Whitfield , 2001 ) . The waaP mutation does not introduce core truncations but results in the loss of a majority of these core phosphates because WaaP activity is a prerequisite for WaaY phosphorylation of Hep ( II ) ( Yethon et al . , 1998 ) . Our results demonstrate that waaP mutation and therefore decreased phosphorylation of LPS is sufficient to fully induce Rcs ( Figure 4B ) . LPS phosphates are essential for establishing cation-mediated LPS cross-bridges in the OM . LB is a Mg2+ -limiting medium , and does not contain a sufficient amount of cations to saturate LPS ( Papp-Wallace and Maguire , 2008; Nikaido , 2009 ) . Addition of Mg2+ is known to stabilize the OM of the deep rough mutants ( Chatterjee et al . , 1976 ) , likely through stabilization of the lipid A phosphates . Interestingly , when 10 mM Mg2+ was added to LB , it decreased Rcs induction in waaP mutant ( Figure 4E ) . The two-component PhoPQ system responds to low Mg2+ concentrations ( Soncini et al . , 1996; Kato et al . , 2003 ) . PhoPQ and Rcs are known to have partially overlapping regulons in several enterobacteria ( Hagiwara et al . , 2003; García-Calderón et al . , 2007 ) . Unlike in Salmonella , PhoPQ in E . coli does not regulate lipid A modifications through the Pmr pathway ( Winfield and Groisman , 2004; Rubin et al . , 2015 ) . Because we observed a Mg2+ dependent effect on Rcs in a waaP mutant , we analyzed Rcs induction in waaP phoP mutant ( Figure 4E ) . Rcs was still induced and responded to Mg2+-addition in the waaP phoP mutant ( Figure 4E ) . Therefore , we concluded that Mg2+ dependent Rcs signaling in waaP mutants was independent of PhoPQ and is likely a result of reinforced cross-bridges between lipid A-phosphates . The effect of LPS charge and presence of cations on Rcs signaling lead us to suggest that RcsF senses the strength of LPS lateral interactions . Binding of a cationic peptide , such as PMB to LPS would also neutralize the charge and weaken LPS lateral interactions . PMB also contains lipid in addition to a cyclic peptide ring , and the lipid integrates into and can disorganize the membrane bilayer ( Vaara , 1992 ) . We therefore tested whether charge-neutralization of LPS by PMB is sufficient to induce Rcs . For this , we analyzed the ability of a lipid-truncated PMB derivative , PMB nonapetide ( PMBN ) to induce Rcs . PMBN also neutralizes LPS but does not disturb the bilayer and is non-toxic . Like PMB , PMBN also induces Rcs , but higher concentrations are required to offset the lower affinity of binding to LPS ( Figure 4F ) ( Vaara and Viljanen , 1985; Thomas and Surolia , 1999 ) . Therefore , we conclude that charge-neutralization of LPS by PMB is sufficient to induce Rcs . If RcsF/OMP complexes sense LPS defects directly , it seems likely that the surface-exposed domain of RcsF acts as the sensory domain . One of the interesting features of surface-exposed region of RcsF is the abundance of positively charged amino acids ( Figure 5A ) . To investigate the role of these positive charges in sensing LPS defects , we substituted the five Arg and Lys residues in the surface-exposed region ( 16–48 ) with Ala to generate a charge substitution mutant gene we refer to as rcsF_A5 ( Figure 5A ) . We verified that RcsF/OmpA complexes are formed with the same efficiency as in RcsF_WT ( Figure 5B ) and that this mutant protein responded normally to Glb ( Figure 5C , lower panel ) . Next , we tested how this mutant protein responded to PMB ( Figure 5C , upper panel ) . Although not completely abolished , the Rcs response to PMB was significantly reduced as compared to WT ( Figure 5C , upper panel ) . Likewise , introduction of rcsF_A5 into the waaP background results in significantly reduced expression of PrprA-lacZ and loss of the mucoid phenotype ( Figure 5D , Figure 5—figure supplement 2 ) . We conclude that the positive charge of surface-exposed region of RcsF is important for sensing alterations in LPS . 10 . 7554/eLife . 15276 . 010Figure 5 . Positive charge of the surface-exposed region of RcsF is required for LPS-sensing . ( A ) Amino acid sequences and total charge of the surface exposed domain of RcsF_WT and the charge substitution mutant RcsF_A5 . Positively charged residues ( Lys and Arg ) which were substituted by Ala are underlined . ( B ) Charge substitution mutations do not affect RcsF crosslinking to BamA and OmpA . ( C ) The rcsF_A5 mutant does not respond to PMB ( upper panel ) but does respond to Glb ( lower panel ) treatment based on expression of a PrprA-lacZ reporter analyzed by beta-galactosidase assay . ( D ) The rcsF_A5 mutant results in decreased PrprA-lacZ expression in the waaP background . Graphs C and D represent mean β-galactosidase activity +/- SEM , n=3 . Corresponding OD600 graphs and untreated controls are shown in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 01010 . 7554/eLife . 15276 . 011Figure 5—figure supplement 1 . Kinetics of PrprA-lacZ expression in RcsF/OMP charge substitution strains upon treatment with PMB , Glb ( same as Figure 5C ) and untreated controls together with corresponding growth curves . Graphs represent mean β-galactosidase activity or OD600 +/- SEM , n=3DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 01110 . 7554/eLife . 15276 . 012Figure 5—figure supplement 2 . Plate phenotype of the charge substitution mutant in the WT and waaP background . Strains were grown on LB agar at 37⁰C overnight . Note , that the waaP mutation confers a RcsF-dependent mucoid phenotype . pZS21::rcsF complements waaP rcsF mutant but does not confer gain-of-function phenotype in the WT background . pZS21::rcsF_A5 failed to complement mucoid phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 15276 . 012
The lipidated amino terminus of lipoprotein RcsF is displayed on the cell surface; the carboxy-terminal signaling domain resides in the periplasm and the short , hydrophilic transmembrane domain is protected from the lipid environment by threading through an OMP , most often OmpA . We show here that this interlocked , heterodimeric structure functions directly to sense known Rcs inducing signals that cause perturbations in the outer leaflet of the OM caused either by treatment with CAMPs or by mutations that cause LPS structural defects . We show that altered OM permeability or asymmetry are not physiological inducers of RcsF . Instead , RcsF senses the state of LPS lateral interactions and it is activated when these interactions are perturbed either by: i ) neutralization of LPS negative charge by CAMPs ( such as PMB ) as a result of direct binding; ii ) decreased LPS phosphorylation as a result of mutations in LPS biosynthesis pathway; iii ) by lack of cations to stabilize LPS cross-bridges . Moreover , we show that the positive charge of surface-exposed domain of RcsF is necessary for sensing these LPS defects . An alternative model for RcsF signaling has been proposed by Cho et al . ( 2014 ) . This model proposes that in unstressed cells , RcsF is constantly assembled into RcsF/OmpA complexes that have no function in sensing and signal transduction . Signals that induce Rcs do so by inhibiting the function of the Bam complex . This prevents RcsF/OMP assembly and allows newly synthesized RcsF to remain free in the periplasm where it can engage the IM components of the signal transduction system . According to this model , activation of Rcs signaling depends on de novo protein synthesis ( Cho et al . , 2014 ) . Here , we have provided multiple lines of evidence demonstrating that this model cannot explain how RcsF senses changes in the OM outer leaflet . First , as noted above , we show that the surface-exposed domain of RcsF/OmpA functions directly to sense these defects . Second , we provide evidence that PMB does not affect β-barrel assembly nor does it induce the SigmaE stress response , which is a sensitive indicator of Bam complex function and the presence of unfolded OMPs in the periplasm . Finally , we show that PMB induction of the Rcs system happens even in the presence of a protein synthesis inhibitor . Two other categories of mutations or chemical agents are also known Rcs inducers: i ) perturbations affecting lipoprotein maturation , such as pgsA ( Shiba et al . , 2004 ) , Glb ( this study ) , or trafficking ( lolA depletion ) ( Tao et al . , 2012 ) ; ii ) perturbations in PG biogenesis , such as those caused by treatment with mecillinam and other β-lactams ( Laubacher and Ades , 2008 ) , lysozyme ( Callewaert et al . , 2009; Ranjit and Young , 2013 ) , A22 ( Cho et al . , 2014 ) , or genetically by introducing multiple pbp knockouts ( pbp4 , 5 , 7 , ampH ) ( Evans et al . , 2013 ) . It is well established that perturbations in lipoprotein assembly lead to the accumulation of RcsF in the inner membrane allowing interaction with downstream components ( Shiba et al . , 2012 ) . Using Ksg , we show that this mechanism clearly depends on RcsF de novo synthesis and is independent of RcsF/OMP assembly in the OM . Activation of RcsF in response to perturbations in PG biogenesis also depends on protein synthesis and is likely independent of RcsF/OMP complexes ( Cho et al . , 2014 ) . However , it is unclear how PG defects induce Rcs . RcsF does not interact with PG , and therefore , activation might not be direct . For example , β-lactams ( including mecillinam ) , A22 and mutations in pbp genes induce not only Rcs but also the Cpx response ( Laubacher and Ades , 2008; Delhaye et al . , 2016; Evans et al . , 2013 ) and , at least in the case of the pbp4 , 5 , 7 , and ampH mutations , Rcs induction relies fully on Cpx ( Evans et al . , 2013 ) . Because A22 and mecillinam treatment decrease RcsF/BamA crosslinking , it was proposed that RcsF monitors the activity of the Bam complex ( Cho et al . , 2014 ) . However , there is no evidence that any of the treatments that affect PG affect the function of the Bam complex . We propose an alternative explanation in which newly synthesized RcsF is engaged in signaling prior its interactions with BamA . Reduction of RcsF/OMP complexes may facilitate this type of signaling by increasing proportion of periplasmic RcsF . Our work also provides important insights into the assembly pathway for the remarkable , interlocked RcsF/OMP complexes . The first of these relates to BamE . bamE encodes one of the non-essential lipoproteins of the Bam complex ( Sklar et al . , 2007 ) , and the function of BamE is not well-understood . Recent studies have uncovered a role for BamE in modulating BamA activity ( Rigel et al . , 2012; 2013 ) . However , bamE null mutations confer modest phenotypes with only slight defects in OMP assembly ( Sklar et al . , 2007; Rigel et al . , 2012 ) . Here , we show that BamE plays an essential role in assembly of RcsF/OMP complexes . In a strain lacking BamE , RcsF/OmpA complexes are almost undetectable . Nonetheless , RcsF is still recognized by BamA . In fact , increased RcsF /BamA crosslinking is observed in a bamE strain . This finding further supports the model that RcsF/BamA crosslinking represents an intermediate in RcsF/OMP assembly pathway and suggests that RcsF binds BamA before it binds the OMP . Secondly , we identified a mutation , rcsF_A55Y that alters a trans-lumen residue and inhibits assembly of RcsF into OMP complexes . We showed that the A55Y substitution prevents RcsF binding to BamA and this in turn prevents assembly of the RcsF/OMP complex . It is important to note that this mutation does not affect the interactions between RcsF and the downstream signaling components , as the rscF_A55Y mutant protein was still able to activate Rcs in response to Glb . We do not yet understand how the inducing signal is transduced by the RcsF/OMP complexes from the cell surface to the IM components of the Rcs system . The RcsF/OMP complexes do not disassemble when OM defects are detected . Therefore , we hypothesize that small conformational changes within complex can facilitate interaction between the RcsF carboxy-terminal signaling domain and the downstream components to activate the signaling pathway and we are currently attempting to detect these changes in the RcsF/OMP complexes in response to LPS defects .
All strains used in this study , including previously published strains ( Majdalani et al . , 2002 , Malinverni et al . , 2006 , Malojčić et al . , 2014 , Silhavy et al . , 1984 ) are listed in Supplementary file 1 . Strains were grown in LB ( 10 g/L tryptone , 5 g/L yeast extract , 10 g/L NaCl ) at 37°C . LB was supplemented with 10 mM MgSO4 when indicated . Arabinose was added at the final concertation of 0 . 2% to support growth of lptE_R91D K136D mutant . Antibiotics were added when appropriate at the following concentration: chloramphenicol 20 µg/ml , kanamycin 25 µg/ml , tetracycline 20 µg/ml . Strains were generated by P1vir transduction ( Silhavy et al . , 1984 ) . The kanamycin resistance cassette was cured from Keio derived mutants with plasmid pCP20 , as previously described ( Datsenko and Wanner , 2000 ) . NPN uptake assay was performed according to ( Loh et al . , 1984 ) . Briefly , AK-265 strain was grown to an OD600 of 0 . 5 , cells were collected by centrifugation and washed twice with 5 mM HEPES , pH 7 . 2 . Cells were resuspened to OD600=0 . 5 and NPN ( 1-N-phenylnaphthylamine , Sigma ) was added to a final concentration of 10 µM . 200 µl of cell suspension was pipetted into wells of black 96 well plates . PMB was added to a final concentration of 0 . 5 or 8 µg/ml . Fluorescence ( excitation 350 nm/emission 420 nm ) was measured for 10 min with 1 min interval using BioTek Synergy 1 plate reader . Endpoint fluorescence was normalized as a fold of untreated sample ( vehicle control ) and values represent mean with SD between three independent measurements . diSC3 ( 5 ) ( Sigma ) release assay was performed following the protocol by ( Zhang et al . , 2000a ) but adopted for a plate-reader format . Briefly , AK-265 strain was grown to an OD600 of 0 . 5 , cells were collected by centrifugation and washed twice with 5 mM HEPES , pH 7 . 8 . Cells were resuspened to OD600=0 . 05 in 5 mM HEPES , pH 7 . 8 with 0 . 2 mM EDTA . diSC3 ( 5 ) was added to the final concentration of 0 . 4 µM . 180 µl of cell suspension was pipetted into wells of black 96 well plates . Uptake of diSC3 ( 5 ) dye was monitored by a decrease in fluorescence excitation 622 nm/emission 670 nm using BioTek Synergy 1 plate reader . After fluorescence signal reached steady state ( appr . 30 min ) , 20 µl of 1 M KCl solution was added to equilibrate the cytoplasmic and external K+ concentrations . PMB or gramicidin ( Sigma ) was added where applicable to the final concentration of 0 . 5 and 12 . 5 µg/ml , respectively , and incubated for 15 min . Fluorescence was measured and normalized as a fold of untreated sample ( vehicle control ) and values represent mean with SD between three independent measurements . For analysis of RNA level induction of Rcs , overnight cultures of AK-265 were back diluted to 2x107 cells/ml and grown to an OD600 of 0 . 5 . Where applicable , cultures were then treated with 500 µg/ml Ksg or a vehicle control for 15 min . Cultures were then treated with 0 . 5 µg/ml PMB or 5 µM Glb . Samples for RNA analysis were harvested at the indicated time points and immediately fixed with a 2X volume of RNAprotect Bacterial Reagent ( Qiagen , Germantown , MD ) as per manufacturer instructions . The fixed cells were lysed and RNA was harvested using the RNeasy kit ( Qiagen ) with on column DNase ( Qiagen ) digestion as per manufacturer instructions . The RNA was quantitated using a Synergy H1 Hybrid Reader ( BioTek , Winooski , VT ) and cDNA was synthesized from 1 µg of RNA using a High Capacity cDNA reverse transcription kit ( Thermo Fisher Scientific , Austin , TX ) in a 20 µl reaction . For qPCR , 10 µl reactions were run in triplicate using PerfecCTa SYBR Green FastMix R0X ( Quanta Biosciences , Gaithersburg , MD ) , 0 . 5 µM primers , and 2 µl of a 1:500 dilution of cDNA samples . The following primers were utilized to quantitate lacZ ( left: 5’-GAAAGCTGGCTACAGGAA-3’; right 5’-GCAGCAACGAGACGTCA-3’ ) , rpoE ( left: 5’-TGGCCTGAGCTATGAAGAGATAG-3’ , right: 5’CCTGATAAGCGGTTGAACTTTG-3’ [Denoncin et al . , 2012] ) , cpxP ( left: 5’-TGCTGAAGTCGGTTCAGGCGATAA-3’ , Right: 5’-TCTGCTGACGCTGATGTTCGGTTA-3’ ) , nrdR ( left: 5’-ATGCATTGCCCATTCTGTTT-3’ , right: 5’-CCGCTACGCAATTTCTCTTC-3’ ) , and ubiJ ( left: 5’-GTTATCGCCTACGCCAGTGT-3’ , 5’-GGCTTTGCTGATTCCTTCAG-3’ ) . RNA expression of rprA was not directly analyzed as high levels of secondary structure prevent accurate quantification by qRT-PCR ( data not shown ) . The qPCR reactions were run on the StepOne Plus RealTime PCR System ( Thermo Fisher Scientific ) using the StepOne Software V2 . 3 ( Thermo Fisher Scientific ) on the following program: 95°C for 10 min , followed by 40 cycles of 95°C for 15 s , 60°C for 1 min ( acquisition ) . Absolute quantification for each primer set was performed based on Ct values calculated by automatic thresholding compared to a standard curve of 101 to 107 copies per reaction of E . coli K12 MG1655 genomic DNA . Relative expression of lacZ , rpoE , and cpxP was calculated using nrdR ( Figure 2A ) and ubiJ ( Figure 2B and C ) as endogenous control genes , as they have been found to be invariant in a wide range of conditions ( Heng et al . , 2011 ) . For calculating fold induction , relative expression values were normalized a no treatment control for each time point . The kinetics of lacZ induction ( Figure 1C upper panel ) are indicated as a representative experiment with error bars representing one standard deviation . Other qRT-PCR experiments ( Figure 1D and Figure 2 ) are indicated as the mean of three independent biological replicates+/-SEM . Strains were grown to an OD600 of 0 . 5–0 . 7 , washed twice in PBS ( Na2HPO4 10mM , KH2PO4 1 . 8 mM , KCl 2 . 7 mM , NaCl 137 mM , pH 7 . 6 ) and concentrated to OD600=10 in PBS . Formaldehyde was added to the final concentration of 0 . 7% and crosslinked for 12 min at room temperature . Crosslinking was stopped by adding of Tris-Cl ( 100 mM final ) ; cells were collected by centrifugation and suspended in SDS loading buffer . Samples were heated at 65° C for 15 min and analyzed by immune blotting with anti-RcsF antibodies . For a time course experiments , overnight cultures of corresponding strains were diluted 1:100 in LB ( supplemented with kanamycin 25 µg/ml when needed ) and grown to an OD600 of 0 . 5 in a shaking water bath at 37°C . Cultures were then treated with 0 . 5 µg/ml PMB or 5 µM Glb for a course of 60 min . Samples were taken every 10 min for a β-galactosidase assay and for OD600 measurement . For a β-galactosidase assay , 100 µl samples were taken and added directly to 900 µl of Z buffer ( 60 mM Na2HPO4 , 40 mM NaH2PO4 , 10 mM KCl , 1 mM MgSO4 , 50 mM β-mercaptoethanol , 0 . 03% SDS ) . 50 µl of chloroform was added to stop growth and mix vigorously by pipetting . After collecting all samples , 100 µl of cell each lysate was mixed with 100 µl of 4 mg/ml O-nitrophenyl-β-D-galactopyranoside ( ONPG ) solution in Z buffer . β-galactosidase activity was analyzed by a kinetic measurement of OD420 in a BioTek Synergy 1 plate reader and Vmax was determined using Gen5 software . Vmax was normalized by OD600 . Experiments were performed in three biological replicates and mean values +/- SEM were plotted . Graphs were built by GraphPad Prism 6 software . For a single point measurement , corresponding strains were grown to an OD600 of 0 . 5–0 . 7 and samples for a β-galactosidase assay were taken and analyzed as described above . | Many disease-causing bacteria have an outer membrane that surrounds and protects the cell , while many hosts of these bacteria produce molecules called antimicrobial peptides that disrupt this outer membrane . In response to this attack , bacteria have evolved a defense system to reinforce their membrane when antimicrobial peptides are present . However , it was not clear how the bacteria sensed these peptides . One clue came from a recent discovery that the bacterial protein required for sensing the peptides is threaded through a barrel-shaped protein to expose a section of it on the bacterial cell’s surface . Now , Konovalova et al . have tested if this surface-exposed domain directly detects damage to the outer membrane caused by the antimicrobial peptides . The investigation revealed several mutants of Escherichia coli that still make the sensor protein but are unable to thread it through the barrel-shaped protein and place a portion on the cell surface . Konovalova et al . showed that these mutants are essentially “blind” to the presence of antimicrobial peptides , and thus prove that it is the surface-exposed domain that works as the sensor . Antimicrobial peptides bind to a major component of the outer membrane and disrupt its normal interactions . Further experiments showed that positively charged sites in surface-exposed domain of the sensor are required to detect these changes and transmit this information inside the cell . Future studies are now needed to understand how the sensor is assembled inside the barrel-shaped protein , and how the danger signal is sent across the membranes that envelope bacterial cells to activate the defense system inside the cell . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | A lipoprotein/β-barrel complex monitors lipopolysaccharide integrity transducing information across the outer membrane |
Understanding the neural basis of consciousness is fundamental to neuroscience research . Disruptions in cortico-cortical connectivity have been suggested as a primary mechanism of unconsciousness . By using a novel combination of positron emission tomography and functional magnetic resonance imaging , we studied anesthesia-induced unconsciousness and recovery using the α2-agonist dexmedetomidine . During unconsciousness , cerebral metabolic rate of glucose and cerebral blood flow were preferentially decreased in the thalamus , the Default Mode Network ( DMN ) , and the bilateral Frontoparietal Networks ( FPNs ) . Cortico-cortical functional connectivity within the DMN and FPNs was preserved . However , DMN thalamo-cortical functional connectivity was disrupted . Recovery from this state was associated with sustained reduction in cerebral blood flow and restored DMN thalamo-cortical functional connectivity . We report that loss of thalamo-cortical functional connectivity is sufficient to produce unconsciousness .
Understanding the neural mechanisms of consciousness is a fundamental challenge of current neuroscience research . We note that precise definitions of the words ‘consciousness’ and ‘unconsciousness’ remain ambiguous . However , loss of voluntary response in a task is an effective and clinically relevant definition of unconsciousness . Over the past decade , important insights have been gained by using functional magnetic resonance imaging ( fMRI ) to characterize the brain during sleep , and disorders of consciousness ( DOC ) such as coma , vegetative states , and minimally conscious states ( Greicius et al . , 2008; Horovitz et al . , 2008; Boly et al . , 2009; Cauda et al . , 2009; Horovitz et al . , 2009; Larson-Prior et al . , 2009; Vanhaudenhuyse et al . , 2010; Samann et al . , 2011; Fernandez-Espejo et al . , 2012; Picchioni et al . , 2014 ) . In addition to being critical for patients to safely and humanely undergo traumatic surgical or diagnostic procedures , anesthesia has long been recognized as a tool for studying the neural mechanisms of loss and recovery of consciousness ( Beecher , 1947 ) . Disruption of cortico-cortical connectivity has been proposed as a central mechanism to explain unconsciousness associated with general anesthesia , sleep , and DOC ( Massimini et al . , 2005; Mashour , 2006; Alkire et al . , 2008; Boly et al . , 2009; Cauda et al . , 2009; Boveroux et al . , 2010; Vanhaudenhuyse et al . , 2010; Langsjo et al . , 2012; Jordan et al . , 2013 ) . However , recent evidence suggests that cortico-cortical functional connectivity may be maintained during unconsciousness ( Greicius et al . , 2008; Horovitz et al . , 2008; Larson-Prior et al . , 2009 ) . To develop more precise neuroanatomic and neurophysiological characterizations of this and other putative mechanisms , studies of anesthetic-induced unconsciousness could benefit from the use of multimodal imaging approaches , combined with neurophysiological understanding of the brain state induced by the anesthetic drug being studied . A challenge in using anesthetics to study unconsciousness is stating precise , possible mechanisms of anesthetic-induced brain states , and using an anesthetic which acts at single rather than multiple brain targets to test those mechanism ( Rudolph and Antkowiak , 2004; Franks , 2008; Brown et al . , 2011 ) . Among the several anesthetics in current use today , dexmedetomidine is an appealing choice for testing a specific mechanism of how an anesthetic alters the level of consciousness . Dexmedetomidine selectively targets pre-synaptic α2-adrenergic receptors on neurons projecting from the locus ceruleus to the pre-optic area ( Correa-Sales et al . , 1992; Chiu et al . , 1995; Mizobe et al . , 1996 ) . This leads to activation of inhibitory outputs to the major arousal centers in the midbrain , pons , and hypothalamus producing a neurophysiological and behavioral state that closely resembles NREM II sleep ( Figure 1 ) ( Nelson et al . , 2003; Huupponen et al . , 2008; Akeju et al . , 2014 ) . Dexmedetomidine also acts at the locus ceruleus projections to the intralaminar nucleus of the thalamus , the basal forebrain and the cortex ( Espana and Berridge , 2006 ) . 10 . 7554/eLife . 04499 . 003Figure 1 . Schematic of dexmedetomidine signaling illustrating similarities to the mechanism proposed for the generation of non-rapid eye movement II sleep . Dexmedetomidine binds to α2 receptors on neurons emanating from the locus ceruleus to inhibit NE release in the POA . The disinhibited POA reduces arousal by means of GABA- and galanin-mediated inhibition of the midbrain , hypothalamic , and pontine arousal nuclei . Dexmedetomidine also acts at the locus ceruleus projections to the intralaminar nucleus of the hypothalamus , the basal forebrain and the cortex and on post-synaptic α2 receptors . 5HT , serotonin; Ach , acetylcholine; BF , basal forebrain; DA , dopamine; Dex; dexmedetomidine; DR , dorsal raphe; GABA , gamma aminobutyric acid receptor subtype A; Gal , galanin; His , histamine; LC , locus ceruleus; LDT , laterodorsal tegmental area; NE , norepinephrine; PAG , periaqueductal gray; POA , preoptic area; PPT , pedunculopontine tegmental area; TMN , tuberomamillary nucleus . DOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 003 Therefore , the aim of this study was to use a novel integrated positron emission tomography and magnetic resonance imaging ( PET/MR ) approach to characterize brain resting-state network activity and metabolism during dexmedetomidine-induced unconsciousness . This strategy offers several appealing features . First , PET scanning using fludeoxyglucose ( 18F-FDG ) provides a sensitive estimate of the brain's metabolic state in terms of the cerebral metabolic rate of glucose ( CMRglc ) . Second , comparison of CMRglc results with regional cerebral blood flow ( rCBF ) estimates obtained from pulsed Arterial Spin Labeling ( pASL ) fMRI signals allows a direct assessment of whether , as previously reported , flow-metabolism coupling is maintained during dexmedetomidine-induced unconsciousness ( Drummond et al . , 2008 ) . Third , we use the CMRglc results to inform the brain functional connectivity analysis derived from the Blood Oxygen Level Dependent ( BOLD ) fMRI signals .
In 10 healthy volunteers , we used EEG recordings , and an auditory task to confirm that dexmedetomidine induced a loss of voluntary responsiveness , and exhibited a neurophysiological profile that was similar to non rapid eye movement ( NREM ) II sleep ( Akeju et al . , 2014 ) . We then studied 18F-FDG uptake during baseline and dexmedetomidine-induced unconsciousness in these 10 healthy volunteers in a separate experiment where we defined loss of consciousness as the onset of sustained eye closure and lack of response to a verbal request to open the eyes ( Figure 2A ) . During both the baseline and dexmedetomidine-induced unconsciousness 18F-FDG study visits ( two per subject ) , fMRI data were also recorded . Consistent with a global decrease in glucose metabolism , we found that during the unconscious state , there was a broad reduction in 18F-FDG standardized uptake values in the brain ( Figure 2—figure supplement 1A ) . To quantify this result , we calculated CMRglc during the two 18F-FDG visits ( Figure 2—figure supplement 1B ) . We found reduced CMRglc in thalamic , frontal , and parietal brain regions during the unconscious state ( Figure 2B , Table 1 ) . The difference in CMRglc exhibited a spatial distribution consistent with the previously described DMN ( Raichle et al . , 2001; Damoiseaux et al . , 2006 ) and both left and right FPNs ( Table 1 ) ( Damoiseaux et al . , 2006; Vincent et al . , 2008 ) . We found these network-specific CMRglc changes interesting because the DMN has been linked to stimulus-independent thought and self-consciousness ( Raichle et al . , 2001; Mason et al . , 2007; Vincent et al . , 2008 ) , while the FPNs have been linked to conscious perception of the external environment and executive control initiation ( Boly et al . , 2008; Vincent et al . , 2008 ) . 10 . 7554/eLife . 04499 . 004Figure 2 . CMRglc and rCBF are decreased in both the default mode and the frontal–parietal network brain regions during unconsciousness . ( A ) Study outline of healthy volunteers recruited to undergo both baseline and dexmedetomidine-induced unconsciousness PET brain imaging in a random order . ( B ) Group wise ( n = 10 ) dexmedetomidine-induced unconsciousness vs baseline changes in CMRglc ( PET ) displayed on the MNI152 standard volume . Significant CMRglc decreases were localized to brain regions that make up the Default Mode and the Frontal-Parietal Networks . ( C ) Schematic of fMRI obtained concurrently in combined PET/MR visits ( n = 10 ) and MR/only visits ( n = 7 ) . ( D ) Group wise ( n = 17 ) dexmedetomidine-induced unconsciousness vs awake changes in rCBF ( fMRI ) displayed on the MNI152 standard volume . Significant rCBF decreases were also localized to brain regions that make up the Default Mode and the Frontal-Parietal Networks . The brain regions that were not included in the rCBF estimation are shaded in the darker hue . ACC , anterior cingulate cortex; CMRglc , cerebral metabolic rate of glucose; Dex , dexmedetomidine; dmPFC , dorsomedial prefrontal cortex; fMRI , functional magnetic resonance; IPL , infraparietal lobule; IPL , infraparietal sulcus; PCC , posterior cingulate cortex; R , right; rCBF , regional cerebral blood flow; RSC , restrosplenial cortex; vmPFC , ventromedial prefrontal cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 00410 . 7554/eLife . 04499 . 005Figure 2—figure supplement 1 . PET SUV changes during unconsciousness . ( A ) Group wise ( n = 10 ) Dexmedetomidine-induced unconsciousness vs baseline changes in SUV ( PET ) displayed on the MNI152 standard volume highlighting that only SUV decreases were notable during unconsciousness . ( B ) Group wise ( n = 10 ) baseline and dexmedetomidine-induced changes in CMRglc ( PET ) displayed on the MNI152 standard volume . MNI , Montreal neurological institute; PET , positron emission tomography; SUV , standardized uptake value . DOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 00510 . 7554/eLife . 04499 . 006Table 1 . CMRglc , awake vs dex-induced unconsciousnessDOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 006MNI coordinate ( mm ) LabelZ-statxyzcluster size ( # voxels ) cluster p-valueawake > dex-induced unconsciousness R middle/inferior frontal gyrus4 . 3552424179192 . 95E-12 R medial orbital gyrus4 . 222438−16 L inferior frontal gyrus4 . 02−523610 L superior frontal gyrus , medial part4−62040 L middle frontal gyrus3 . 94−421838 L middle frontal gyrus3 . 86−34866 R middle frontopolar gyrus3 . 79−30582 L lateral orbital gyrus3 . 58−4628−12 L medial orbital gyrus3 . 56−2632−22 R middle frontal gyrus3 . 55482044 R superior frontal gyrus3 . 4826256 R inferior frontal gyrus , pars opercularis3 . 3754164 L inferior frontal gyrus , pars opercularis3 . 36−46200 L superior frontal gyrus , medial part ( 2 ) 3 . 33−44222 inferior rostral gyrus3 . 15048−10 R anterior insula2 . 8240220 L Posterior Cingulate Gyrus5 . 56−4−243854432 . 85E-05 L precuneus4 . 27−12−6422 R precuneus3 . 8612−6028 L supramarginal gyrus4 . 58−40−464228610 . 0029 L angular gyrus3 . 76−48−6834 R supramarginal gyrus4 . 5944−464822140 . 0112 R angular gyrus434−6244 R thalamus4 . 5212−20416100 . 0444 L thalamus3 . 89−14−264CMRglc , cerebral metabolic rate of glucose; Dex , dexmedetomidine; L , left; MNI , Montreal Neurological Institute; R , right . To further understand the effects of dexmedetomidine-induced unconsciousness on rCBF and brain functional connectivity in these networks , we increased our statistical power for fMRI data analysis by studying an additional seven volunteers using fMRI only ( Figure 2C ) . Analysis of the rCBF difference maps between the unconscious and baseline states showed decreases in rCBF during unconsciousness that overlapped with the same DMN and FPN regions that demonstrated reductions in CMRglc ( Figure 2D , Figure 2—figure supplement 1 , Table 2 ) . This is consistent with maintenance of blood flow and metabolism coupling during dexmedetomidine-induced unconsciousness . 10 . 7554/eLife . 04499 . 007Table 2 . rCBF , awake vs unconsciousDOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 007MNI coordinate ( mm ) LabelZ-statxyzcluster size ( # voxels ) cluster p-valueawake > unconscious L thalamus5 . 36−10−1412108764 . 20E-11 L midbrain5 . 28−32−10 L posterior cingulate cortex5 . 14−2−4628 R thalamus5 . 038−1210 R midbrain4 . 44−6−36−12 L precuneus4 . 44−2−7032 R precuneus4 . 38−7642 R intracalcarine cortex4 . 24−844 L intracalcarine cortex4 . 18−2−822 R hippocampus4 . 1730−34−4 L supramarginal gyrus/intraparietal sulcus4 . 03−34−5044 L hippocampus3 . 09−26−22−12 R frontopolar cortex4 . 71−2464217730 . 00463 L frontopolar cortex4 . 322062−6 L middle frontal gyrus3 . 77−223828 R inferior frontal gyrus , orbital part2 . 885040−14 L dorsal anterior cingulate cortex3 . 9310382212620 . 0227 R dorsal anterior cingulate cortex3 . 75−103820 L subgenual anterior cingulate cortex3 . 3−224−10 R pregenual anterior cingulate cortex3 . 21642−6unconscious > awake n . s . L , left; MNI , Montreal Neurological Institute; n . s , not significant; R , right . Given that we observed reduced flow and metabolism in regions known to correspond to the DMN and FPNs ( Raichle et al . , 2001; Damoiseaux et al . , 2006; Vincent et al . , 2008 ) , we next asked whether intrinsic functional connectivity within these networks was altered with dexmedetomidine-induced unconsciousness . First , we confirmed that during both wakefulness and unconsciousness , the DMN and bilateral FPNs were consistently identified in our volunteers ( Figure 3 , Table 3 ) . Next , we performed a comparison of changes in these networks between the unconscious and awake states . When we compared the DMN between these two states , we found no difference in functional connectivity between the cortical regions of the DMN and FPNs , suggesting that cortico–cortico functional connectivity may be maintained during unconsciousness ( Figure 3A–B , Table 3 ) . However , we observed a reduction in functional connectivity between the thalamus and the DMN in the unconscious state ( Figure 3A–B , Table 3 ) . This loss of thalamic functional connectivity was observed in a region consistent with intralaminar , midline , mediodorsal , and ventral anterior nuclei . However , our ability to precisely resolve the thalamic nuclei implicated in loss of thalamo-cortical functional connectivity is limited because our methods are not sensitive to small focal differences . We also found that functional connectivity of the left cerebellar representation of the DMN ( Buckner et al . , 2011 ) was significantly reduced during unconsciousness ( Figure 3B , Table 3 ) . When we compared the FPNs in these two states , we found that the left and right FPNs showed opposite changes in functional connectivity with a cerebellar cluster , likely indicating a switch in hemispheric dominance for cortico-cerebellar functional connectivity to this region during unconsciousness ( Figure 3C–F , Table 4 , Table 5 ) . Functional connectivity of the cerebellar representations of the FPNs ( Buckner et al . , 2011 ) was also significantly reduced ( Figure 3D , F , Table 4 , Table 5 ) . Since cortico-cerebellar connections are mediated by the thalamus ( Guillery and Sherman , 2002 ) , the changes we observed in cortico-cerebellar functional connectivity are likely related to impaired thalamic functioning . Taken together , these results show that unconsciousness is associated with decreased thalamic CMRglc , rCBF , and functional connectivity to the DMN and support the notion that the thalamus plays a critical role in mediating unconsciousness . 10 . 7554/eLife . 04499 . 008Figure 3 . Changes in the default mode and the bilateral frontal parietal networks during unconsciousness . ( A ) The DMN extracted from BOLD signals during the awake ( n = 16 ) and unconscious states ( n = 16 ) and displayed on the MNI152 standard volume . ( B ) A comparison the DMN network during the unconscious vs the awake state identified the putamen and the vlPFC as regions that exhibited increased functional connectivity during dexmedetomidine-induced unconsciousness . This comparison also identified the thalamus and the cerebellar representation of this network as regions that exhibited decreased functional connectivity during dexmedetomidine-induced unconsciousness . Cortico–cortico functional connectivity within this network was maintained . ( C ) The right FPN extracted from BOLD signals during the awake and unconscious states ( n = 16 ) and displayed on the MNI152 standard volume . ( D ) A comparison of the right FPN network during the unconscious vs the awake state identified the infraparietal lobule and a right cerebellar cluster as regions that exhibited increased functional connectivity during dexmedetomidine-induced unconsciousness . This comparison also identified the cerebellar representation of this network as a region that exhibited decreased functional connectivity during unconsciousness . Cortico–cortico functional connectivity within this network was maintained . ( E ) The left FPN extracted from BOLD signals during the awake and unconscious states ( n = 16 ) and displayed on the MNI152 standard volume . ( F ) A comparison of the left FPN network during the unconscious vs the awake state identified a right cerebellar cluster and the cerebellar representation of the left FPN as regions that exhibited decreased functional connectivity during dexmedetomidine-induced unconsciousness . Cortico–cortico functional connectivity within this network was maintained . BOLD , blood oxygen level dependent; DMN , default mode network; FPN , Frontoparietal Network; R , right; vlPFC , ventrolateral prefrontal cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 00810 . 7554/eLife . 04499 . 009Figure 3—figure supplement 1 . SUV , CMRglc , rCBF changes during unconsciousness and the relationship to the DMN and bilateral FCNs . ( A ) Surface projection of the voxelwise , cluster corrected analysis illustrating overlapping decreases in CMRglc ( n = 10 ) and rCBF ( n = 17 ) in regions corresponding to the DMN , and the bilateral FCNs during unconsciousness . ( B ) Surface projection of the DMN , rFCN and lFCN extracted from study volunteers ( n = 17 ) . These networks were identified from the awake , unconscious , and recovery state BOLD signals . CMRglc , cerebral metabolic rate of glucose; dlPFC , dorsolateral prefrontal cortex; dmPFC , dorsomedial prefrontal cortex; DMN , default mode network; FPN , frontoparietal network; IPL , infraparietal lobule; IPL , infraparietal sulcus; MNI , Montreal neurological institute; PCC , posterior cingulate cortex; rCBF , regional cerebral blood flow; R , right; RSC , retrosplenial cortex; vlPFC , ventrolateral prefrontal cortex , vmPFC , ventromedial prefrontal cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 00910 . 7554/eLife . 04499 . 010Table 3 . DMNDOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 010MNI coordinate ( mm ) LabelZ-statxyzcluster size ( # voxels ) cluster p-valueawake > unconscious R thalamus4 . 042−1484380 . 00193 L thalamus3 . 9−4−148 L cerebellum ( Crus II ) 3 . 93−44−64−442700 . 0429unconscious > awake R inferior frontal gyrus , pars opercularis3 . 835612203990 . 00381 R inferior frontal gyrus , pars triangularis3 . 35543210 L putamen3 . 47−246−83820 . 00517 L insula3 . 43−36−10−4DMN , Default Mode Network; L , left; MNI , Montreal Neurological Institute; R , right . 10 . 7554/eLife . 04499 . 011Table 4 . rFCNDOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 011MNI coordinate ( mm ) LabelZ-statxyzcluster size ( # voxels ) cluster p-valueawake > unconscious L cerebellum ( Crus II ) 3 . 99−36−70−404720 . 000724 L cerebellum ( VIIb ) 3 . 83−34−62−50 L cerebellum ( VI ) 3 . 28−28−62−32unconscious > awake R cerebellum ( VI ) 4 . 0212−68−286931 . 97E-05 R cerebellum ( Crus II ) 3 . 6726−78−42 R cerebellum ( Crus I ) 3 . 2420−74−24 R precuneus2 . 9926−60185190 . 000323 R parietal operculum4 . 158−12125160 . 00034 R insula3 . 2344−40 R cerebellum ( Crus I ) 4 . 2844−56−322920 . 0212 R fusiform gyrus3 . 5930−42−20 L angular gyrus3 . 72−38−86202720 . 0319 L angular gyrus3 . 56−48−7830L , left; MNI , Montreal Neurological Institute; R , right; rFCN , right Frontoparietal Control Network . 10 . 7554/eLife . 04499 . 012Table 5 . lFCNDOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 012MNI coordinate ( mm ) LabelZ-statxyzcluster size ( # voxels ) cluster p-valueawake > unconscious R cerebellum ( Crus II ) 4 . 2438−74−5012235 . 33E-09 R cerebellum ( VIIb ) 4 . 138−56−50 R cerebellum ( VI ) 4 . 0828−62−32 R intracalcarine cortex3 . 584−86−2 L middle frontal gyrus3 . 89−3212544900 . 000341 L superior frontal gyrus , medial part3 . 52−43246unconscious > awake n . s . L , left; lFCN , left Frontoparietal Control Network; MNI , Montreal Neurological Institute; n . s , not significant; R , right . In order to further examine the role of the thalamus in unconsciousness , we next performed a seed-based functional connectivity analysis using the thalamic region with decreased DMN functional connectivity as the seed region . This thalamic seed also overlapped with thalamic regions showing reduced rCBF and CMRglc . Our seed-based analysis allowed us to more thoroughly examine changes in thalamic functional connectivity to all brain regions . We found that the thalamic seed was significantly functionally disconnected from posterior regions of the DMN ( posterior cingulate cortex , precuneus , and inferior parietal lobules ) during unconsciousness ( Figure 4A , Table 6 ) . Notably , the cortical regions showing reduced thalamic functional connectivity also overlapped with those demonstrating reduced flow and metabolism ( Figure 4B ) . 10 . 7554/eLife . 04499 . 013Figure 4 . Seed based functional connectivity analysis of the brain region with overlapping changes in CMRglc , rCBF , and functional connectivity . ( A ) Brain wide representation of regions connected to the thalamic seed during the awake and unconscious states ( n = 16 ) . A comparison of thalamic seed functional connectivity during the unconscious vs the awake state identified the PCC , precuneus , thalamus , visual cortex , IPL and IPS as regions exhibiting decreased functional connectivity during unconsciousness . ( B ) Overlap of the brain regions with changes in CMRglc , rCBF , and functional connectivity . Notably , the posterior thalamus , PCC and the precuneus exhibited overlapping changes . CMRglc , cerebral metabolic rate of glucose; IPL , infraparietal lobule; IPL , infraparietal sulcus; PCC , posterior cingulate cortex; R , right; rCBF , regional cerebral blood flow . DOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 01310 . 7554/eLife . 04499 . 014Table 6 . Thalamic seedDOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 014MNI coordinate ( mm ) LabelZ-statxyzcluster size ( # voxels ) cluster p-valueawake > unconscious R thalamus5 . 3312−12834454 . 05E-16 L posterior cingulate cortex4 . 95−6−3630 R posterior cingulate cortex4 . 736−2834 L thalamus4 . 48−16−1010 R caudate nucleus3 . 44121018 L globus pallidus3 . 44−122−2 L precuneus4 . 03−6−563612301 . 79E-07 R precuneus3 . 7112−6036 R angular gyrus4 . 0736−605010949 . 54E-07 R supramarginal gyrus3 . 6748−4836 R cerebellum ( crus II ) 4 . 078−76−309068 . 76E-06 R cerebellum ( crus I ) 3 . 8422−88−22 R intracalcarine cortex3 . 752−1004 L cerebellum ( crus I ) 3 . 14−20−84−28 L supramarginal gyrus4 . 6−38−52428312 . 21E-05 L angular gyrus3 . 43−32−7648unconscious > awake L inferior frontal gyrus , pars opercularis4 . 21−5614814581 . 80E-08 L superior temporal gyrus4 . 2−62−104 L middle temporal gyrus4 . 04−50−2−22 L insula3 . 55−384−6 R paracentral lobule3 . 774−32589634 . 41E-06 L paracentral lobule3 . 64−10−3858 R superior frontal gyrus , lateral part3 . 4414−656 R precentral gyrus3 . 2314−2066 L postcentral gyrus3 . 2−16−4472 R superior frontal gyrus , medial part2 . 8610250 R postcentral gyrus2 . 7216−3870 R superior temporal gyrus548−42108681 . 39E-05 R middle temporal gyrus4 . 5344−28−6 R precentral gyrus3 . 5154−246 L postcentral gyrus4 . 03−62−8364530 . 00401 L precentral gyrus3 . 77−60−436 L superior temporal gyrus3 . 97−66−44103400 . 0244L , left; MNI , Montreal Neurological Institute; R , right . We next investigated whether the rCBF changes during the unconscious state returned to baseline during the recovery state . We identified 10 volunteers who responded to verbal instructions ( opened eyes and gave a thumbs-up signal , though they were still mildly sedated ) during at least 6 min of the recovery ASL scan . Surprisingly , when we compared the recovery state to the unconscious state in these volunteers , we found that the rCBF did not increase ( cluster corrected , Figure 5A ) . In fact , when explored at a more liberal threshold ( uncorrected , p < 0 . 05 ) , further decreases in rCBF were observed ( Figure 5B ) . When we compared the recovery state to the awake state , we found that the decrease in rCBF showed a spatial distribution similar to that observed during the unconscious state ( Figure 5C ) . Next , we aligned all subjects to unconscious and recovery time points in order to study the dynamics of rCBF changes within the statistically significant clusters . We confirmed that rCBF decreased during loss of consciousness and that this decrease was maintained during recovery from this state in all regions analyzed ( Figure 5—figure supplement 1 ) . These results suggest that cortical rCBF increase was not sufficient for the early stages of recovery from unconsciousness induced by dexmedetomidine . 10 . 7554/eLife . 04499 . 015Figure 5 . Cortical increase in rCBF is not evident during the recovery of consciousness . ( A ) Cluster corrected ( n = 10 ) rCBF comparison of the recovery state vs the unconsciousness state displayed on the MNI152 standard volume highlighting no change in rCBF at recovery . ( B ) Uncorrected ( n = 10 ) rCBF comparison of the recovery state vs the unconsciousness state displayed on the MNI152 standard volume suggesting that a decrease in rCBF occurred during the recovery state . ( C ) Cluster corrected ( n = 10 ) rCBF comparison of the recovery state vs the awake state displayed on the MNI152 standard volume highlighting a spatial distribution of rCBF decrease similar to that observed during unconscious vs the awake state ( n = 17 ) comparison . IPL , infraparietal lobule; IPL , infraparietal sulcus; PCC , posterior cingulate cortex; rCBF , regional cerebral blood flow; R , right; RSC , retrosplenial cortex . The brain coverage of rCBF estimation is shown in Figure 1D . DOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 01510 . 7554/eLife . 04499 . 016Figure 5—figure supplement 1 . Cortical decrease in rCBF persists during the recovery of consciousness . Statistically significant rCBF clusters from Figure 1D were extracted and intersected with the Harvard–Oxford cortical and subcortical map arbitrarily thresholded at 30 . Subjects were aligned to LOC/ROC and then median baseline normalized rCBF values obtained for each time point . For LOC ( n = 17 ) and ROC ( n = 16 ) , an average of baseline-normalized rCBF values corresponding to the 2 min immediately preceding and post LOC/ROC behavioral time points are represented . The decreased rCBF values from baseline observed during the LOC were sustained during the ROC . L , left; LOC , loss of consciousness; Post , posterior; rCBF , regional cerebral blood flow; R , right; ROC , return of consciousness . DOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 016 We found that thalamic functional connectivity to the DMN was restored during the recovery state ( Figure 6A–B , Table 7 ) . We also found that the opposing left and right FPN changes in the cerebellar cluster ( Figure 3B–C ) were also reversed ( Figure 6C–F , Table 7 ) . When we analyzed the functional connectivity of our thalamic seed of interest , we found that thalamic functional connectivity to the posterior cingulate cortex and thalamo-thalamic functional connectivity were restored during the recovery state ( Figure 6G–H , Table 7 ) . Taken together , these results provide further evidence that dexmedetomidine-induced unconsciousness is associated with decreased thalamic functional connectivity to the DMN and supports the notion that the thalamus plays a critical role in mediating unconsciousness and recovery from this state . 10 . 7554/eLife . 04499 . 017Figure 6 . Functional connectivity changes observed during recovery of consciousness . ( A ) The DMN extracted from BOLD signals obtained during the recovery state ( n = 15 ) displayed on the MNI152 standard volume . ( B ) A comparison the DMN network during the recovery vs the awake state identified the thalamus as the only region with partial recovery of functional connectivity . ( C ) The right FPN extracted from BOLD signals during the recovery state ( n = 15 ) and displayed on the MNI152 standard volume . ( D ) A comparison of the right FPN during the recovery vs the awake state showed that right cerebellar cluster from Figure 3B , C now exhibited decreased right FPN functional connectivity . ( E ) The left FPN extracted from BOLD signals during the recovery state ( n = 15 ) and displayed on the MNI152 standard volume . ( F ) A comparison of the left FPN during the recovery vs the awake state showed that right cerebellar cluster from Figure 3B , C now exhibited increased left FPN functional connectivity . ( G ) Brain wide representation of regions connected to the thalamic seed during the recovery state ( n = 15 ) displayed on the MNI152 standard volume . ( H ) A comparison of thalamic seed functional connectivity during the recovery vs the awake state identified the PCC and thalamus as regions with increased functional connectivity at recovery . BOLD , blood oxygen level dependent; DMN , default mode network; FPN , Frontoparietal Network; PCC , posterior cingulate cortex; R , right . DOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 01710 . 7554/eLife . 04499 . 018Table 7 . RecoveryDOI: http://dx . doi . org/10 . 7554/eLife . 04499 . 018MNI coordinate ( mm ) LabelZ-statxyzcluster size ( # voxels ) cluster p-valueDMN , recovery > unconscious R thalamus3 . 32−84780 . 0345 L thalamus3−4−48DMN , unconscious > recovery n . s . rFCN , recovery > unconscious n . s . rFCN , unconscious > recovery R cerebellum ( VI ) 3 . 9214−70−282460 . 00172 R cerebellum ( Crus I ) 3 . 8122−76−28 R cerebellum ( Crus II ) 3 . 0624−80−38lFCN , recovery > unconscious R cerebellum ( VI ) 3 . 5812−66−301180 . 012 R cerebellum ( Crus I ) 3 . 2938−58−32930 . 0232 R cerebellum ( Crus II ) 3 . 0532−64−42lFCN , unconscious > recovery n . s . Thalamic seed , recovery > unconscious posterior cingulate cortex3 . 530−40223670 . 00367 L thalamus3 . 49−10−22123480 . 00471 L putamen2 . 51−1442Thalamic seed , unconscious > recovery n . s . DMN , default mode network; L , left; lFCN , left Frontoparietal Control Network; MNI , Montreal Neurological Institute; n . s , not significant; R , right; rFCN , right Frontoparietal Control Network .
The notion that disruption of thalamo-cortical functional connectivity could constitute a neural-network mechanism responsible for unconsciousness that is distinct from disruption of cortico-cortical functional connectivity is supported by a number of findings . During , NREM I and II sleep in humans , DMN cortico-cortical functional connectivity remains preserved ( Horovitz et al . , 2008; Larson-Prior et al . , 2009 ) while thalamo-cortical functional connectivity is disrupted ( Picchioni et al . , 2014 ) . However , during NREM III , there is a functional disconnection between frontal and parietal nodes of the DMN ( Horovitz et al . , 2009; Samann et al . , 2011 ) , and disrupted thalamo-cortical functional connectivity ( Picchioni et al . , 2014 ) . Likewise , in-depth electrode studies in humans show that changes in thalamic activity during sleep precedes changes in cortical activity , suggesting that disrupted thalamo-cortical functional connectivity may underlie changes in consciousness ( Magnin et al . , 2010 ) . Also , during propofol anesthesia-induced unconsciousness , decreases in thalamo-cortical functional connectivity , alongside disruptions in cortico-cortical functional connectivity in the DMN and FPNs , have been described in healthy volunteers ( Boveroux et al . , 2010; Ni Mhuircheartaigh et al . , 2013 ) Functional reintegration of the thalamus to key regions of the cortex appears to be necessary for the recovery of consciousness from this state ( Langsjo et al . , 2012 ) . Norepinephrinergic neurons in the locus ceruleus project to thalamic mediodorsal , midline , intralaminar nuclei , which in turn have neurons that project to the cingulate cortex ( Jones and Yang , 1985; Vogt et al . , 1987; Buckwalter et al . , 2008; Vogt et al . , 2008 ) . Our results show that these connections may comprise a functional circuit that is disrupted during dexmedetomidine-induced unconsciousness . Moreover , this functional circuit appears specifically to involve the posterior cingulate cortex , distinct from anterior or mid-cingulate cortex . In our thalamic seed-based functional connectivity analysis , we found that during dexmedetomidine-induced unconsciousness , thalamic regions consistent with intralaminar , midline , mediodorsal , and ventral anterior nuclei were functionally disconnected from posterior DMN structures ( precuneus and the posterior cingulate cortex ) . Perhaps more importantly , recovery from dexmedetomidine-induced unconsciousness was associated with restored thalamo-thalamic and thalamic functional connectivity to the posterior cingulate cortex . Structural connectivity between the thalamus and the posterior cingulate cortex has also recently been correlated with the severity of DOC and used to subcategorize patients in minimally conscious states ( Fernandez-Espejo et al . , 2012 ) . The above observations lead us to conclude that a noradrenergic circuit involving the thalamus and posterior cingulate cortex may be important for recovery from both dexmedetomidine-induced unconsciousness and DOC states . An FDG-PET study has recently described a linear relationship between metabolism in the precuneus and central thalamus in severely brain-injured patients with DOC ( Fridman et al . , 2014 ) . We note that our observed reduction in blood flow in the precuneus and thalamus during dexmedetomidine-induced unconsciousness parallels this result . However , during recovery from dexmedetomidine-induced unconsciousness , we did not observe an increase in rCBF to any brain regions . Since our current and previously reported ( Drummond et al . , 2008 ) results suggest that dexmedetomidine does not decouple rCBF and CMRglc , we surmise that metabolism does not return to baseline during the earliest phases of recovery from dexmedetomidine-induced unconsciousness . Thus , the reduced rCBF we report after recovery from dexmedetomidine-induced unconsciousness could reflect a residual drug effect state in which subjects are responsive to external stimuli but remain in a sedated state . This residual effect may be mediated by the broad and diffuse nor-adrenergic projections from the locus ceruleus to the rest of the brain ( Vogt et al . , 2008 ) . This result is interesting because it suggests that recovery from unconsciousness may occur in a graded fashion . For example , performance on more complex cognitive tasks likely differs substantially between the awake and recovery states . This suggests that re-establishment of functional connectivity between vital brain regions precedes increases in rCBF and CMRglc . Return of rCBF and CMRglc to baseline levels may be indicative of the final stages of recovery from dexmedetomidine-induced unconsciousness . Future investigations incorporating cognitive and psychomotor vigilance tasks during recovery from experimental models of unconsciousness would lend further insight on whether this is a drug-specific effect . Nevertheless , this finding is interesting because it suggests that recovery from unconsciousness states may proceed prior to significant increases in rCBF . Drawing from previous work on sleep ( Horovitz et al . , 2008 , 2009; Larson-Prior et al . , 2009; Magnin et al . , 2010; Samann et al . , 2011; Picchioni et al . , 2014 ) and anesthesia ( White and Alkire , 2003; Alkire et al . , 2008; Boveroux et al . , 2010; Liu et al . , 2013; Ni Mhuircheartaigh et al . , 2013 ) , our results suggest a hierarchy of brain network disruptions that can result in different states of altered consciousness . In NREM II ( non-slow wave ) sleep and dexmedetomidine-induced unconsciousness , patients can be easily aroused to respond to verbal commands by sufficiently strong external stimuli . In these states , cortico-cortical functional connectivity is preserved ( Horovitz et al . , 2008; Larson-Prior et al . , 2009 ) , but thalamo-cortical functional connectivity is disrupted ( Picchioni et al . , 2014 ) . In propofol anesthesia-induced unconsciousness and some DOC states , where patients cannot be aroused by external stimuli , both cortico-cortical and thalamo-cortical functional connectivity are disrupted ( Boveroux et al . , 2010 ) . In the case of sleep , dexmedetomidine-induced unconsciousness , and some minimally conscious states , maintained cortico-cortical functional connectivity likely allows for the cortex to be ‘primed’ , ready to recover from the altered consciousness states with restoration of thalamic functional connectivity , which could be triggered by ascending midbrain and brainstem inputs ( Solt et al . , 2014 ) or direct therapeutic intervention at the thalamic level ( Schiff et al . , 2007; Williams et al . , 2013 ) . In other altered consciousness states , such as during general anesthesia or comatose states , markedly disrupted cortico-cortical functional connectivity could further impair the ability to recover consciousness . We note that while it might be theoretically possible to attain unconsciousness by disrupting only cortico-cortical functional connectivity while preserving thalamo-cortical functional connectivity , it appears , from our study and others , that this does not happen during sleep and anesthesia-induced unconsciousness . Compared to the findings supporting a role for disrupted cortico-cortical functional connectivity during unconsciousness , our findings suggest that thalamo-cortical connectivity constitutes a more fundamental disruption required to produce unconsciousness . This insight could lead to a more objective characterization , diagnosis , and prognosis of DOC . Also , the state of cortico-cortical and thalamo-cortical functional connectivity could be used to better select patients who may benefit from therapeutic interventions . For instance , in the event of maintained cortico-cortical functional connectivity but impaired thalamo-cortical functional connectivity with intact thalamo-cortical structural connections ( Fernandez-Espejo et al . , 2012 ) , therapies aimed at restoring thalamo-cortical functional connectivity might be more likely to help patients recover . These therapies could be pharmacological ( Williams et al . , 2013 ) , electrical ( Schiff et al . , 2007 ) , or could involve other forms of stimulation . Since dexmedetomidine appears to disrupt functional network activity in a manner similar to NREM II sleep , our results also suggest that alpha-2-receptor agonists could be further developed as effective sleep therapeutic drugs . At present , there is a debate pertaining to whether anesthetized and unresponsive patients ( without motor impairments ) can remain internally conscious to subjective experiences ( internally generated or provoked by external stimuli ) . This debate has been fueled by clinical studies of anesthetized patients that have reported rare intraoperative episodes of awareness ( resulting from light anesthetic depth ) under anesthesia ( Ghoneim et al . , 2009 ) and dreaming under anesthesia ( Brandner et al . , 1997; Leslie et al . , 2005; Errando et al . , 2008; Sanders et al . , 2012 ) . Noreika et al . studied this phenomenon with dexmedetomidine , sevoflurane , propofol , and xenon ( Noreika et al . , 2011 ) and confirmed that subjective experiences of consciousness occur during clinically defined unconsciousness states . However , the experimental paradigm that was used to test for subjective experiences of consciousness involved a gradual increase in drug concentrations . This empiric gradual increase did not directly target anesthetic drug dosing to a neurophysiologically defined brain state nor does it inform us on the amount of time that was spent in lighter levels of anesthesia . Thus , the reported subjective experiences could have occurred when the drug concentration levels were very low . This line of reasoning is supported by evidence confirming that subjective experiences of dreaming under general anesthesia likely occur when patients are emerging from general anesthesia ( Leslie et al . , 2009 ) . Subjective experiences of consciousness also occur during REM sleep ( Hobson , 2009 ) . To a limited degree , they are also thought to occur during NREM sleep , especially at sleep onset in NREM I and during NREM II sleep periods encountered later at night ( Hobson , 2009 ) . However , the current methods for scoring sleep states remain imprecise and highly subjective , with limited state and temporal resolution , and high intra- and inter-scorer variability , raising the possibility that clinically scored NREM I and II sleep may include periods of arousal . In clinical sleep scoring , trained technicians visually score the sleep time-series data in 30-s epochs according to semantically defined sleep stages . These scoring standards grossly simplify sleep electroencephalographic dynamics . For example , consider the characterization of sleep onset . The American Academy of Sleep Medicine defines sleep onset as the first appearance of any 30-s epoch that contains at least 15 s of sleep ( Iber et al . , 2007 ) . This implies that sleep onset is a binary process . However , recent EEG studies in humans clearly demonstrate that sleep onset , at both a neurophysiological and behavioral level , is a continuous process . As such , methods designed to track moment-to-moment continuous variations in the EEG are significantly better at predicting when subjects lose wakefulness , measured in terms of a behavioral task , than traditional sleep scoring methods ( Prerau et al . , 2014 ) . In particular , subjects can continue to perform awake behavior despite being scored as asleep ( NREM I ) using traditional sleep scoring methods . Thus , it is possible that the subjective experiences of consciousness that have been ascribed to NREM sleep could have occurred during subjectively scored sleep stages that included un-scored periods of arousal or wakefulness . Despite this rapidly evolving understanding of subjective experiences during altered arousal under sleep and anesthesia , we have tried in this study to focus on a particular state of dexmedetomidine-induced unconsciousness . We achieved this state through rapid administration of dexmedetomidine and confirmed that this dose and form of drug administration produces a NREM II-like state ( Akeju et al . , 2014 ) . We then applied a combination of fMRI and FDG-PET imaging that would characterize the overall time-integrated brain activity during this state . Clinical evidence shows that motor-evoked potentials , and thus motor function from cortex down , remain intact in this state , making it unlikely that the loss of responsiveness we observed is only a result of disrupted motor function ( Moore et al . , 2006; Garavaglia et al . , 2014 ) . Therefore , we feel it is appropriate to refer to the state of altered arousal described in this manuscript as dexmedetomdine-induced unconsciousness . A long-standing debate in anesthesiology relates to the fundamental systems-level mechanisms for anesthesia-induced unconsciousness , with some investigators proposing a thalamic switch ( Alkire et al . , 2000 ) and others proposing disrupted cortico-cortical connections as the central mechanism ( Massimini et al . , 2005; Mashour , 2006; Alkire et al . , 2008 ) . Our results , in combination with previous work , resolve this conflict by showing how both mechanisms could co-exist but at different parts of the continuum of altered arousal or unconsciousness . Moreover , it suggests a mechanistic interpretation for varying levels of anesthesia-induced unconsciousness , where disrupted thalamo-cortical functional connectivity , but intact cortico-cortical functional connectivity reflects lighter states of unconsciousness and disruptions of both cortico-cortical and thalamo-cortical functional connectivity reflect deeper states of unconsciousness . These insights could guide development of system-specific anesthetic drugs that possess minimal side effects and improved monitoring of patients in both the operating room and intensive care settings .
This study was conducted at the Athinoula A Martinos Center for Biomedical Imaging at the Massachusetts General Hospital . The Human Research Committee and the Radioactive Drug Research Committee at the Massachusetts General Hospital approved the study protocol . After an initial email/phone screen , potential study subjects were invited to participate in a screening visit . At the screening visit , informed consent including the consent to publish was requested after the nature and possible consequences of the study was explained . All subjects provided informed consent and were American Society of Anesthesiology Physical Status I with Mallampati Class I airway anatomy . After providing informed consent , a standard pre-anesthesia assessment was administered and a blood toxicology screen was performed to ensure that subjects were not taking drugs that might confound the interpretation of study results . A complete metabolic panel and a urine pregnancy test were also obtained to confirm that all values were within normal ranges and to confirm non-pregnant status , respectively . A total of 20 subjects participated in screening visits and 18 subjects were deemed eligible to participate in the study protocol . Two subjects were deemed ineligible after medical evaluation and one subject was lost to follow-up after providing informed consent . During all study visits , subjects were required to take nothing by mouth for at least 8 hr . A urine toxicology screen was performed to ensure that subjects had not taken drugs that might confound interpretation of the results . A pregnancy test was also administered ( serum for PET visits , urine for fMRI only visits ) for each female subject to confirm non-pregnant status . We studied awake , unconscious , and recovery states in humans , using an integrated positron PET/MR approach , in healthy volunteers , 18–36 years of age . Brain imaging was performed with the Biograph mMR scanner ( Siemens Healthcare , Erlangen , Germany ) , which allows simultaneous acquisition of whole-body PET and 3 Tesla MRI data . The fully integrated PET detectors use avalanche photoiodide technology and lutetium oxyorthosilicate crystals ( 8 × 8 arrays of 4 × 4 × 20 mm3 crystals ) . The PET scanner's transaxial and axial fields of view are 594 mm and 25 . 8 cm , respectively ( Drzezga et al . , 2012 ) . Approximately , 5 mCi of FDG was purchased from an outside approved vendor and was administered as a bolus immediately before the initiation of the dexmedetomidine ( described below ) . A PET-compatible 16-channel head and neck array was used to acquire the MR data . At the beginning of the imaging visit , we acquired structural MRI ( MPRAGE volume , TR/TE = 2100/3 . 24 ms , flip angle = 7° , voxel size = 1 mm isotropic ) for the purpose of anatomical localization , spatial normalization , and surface visualization of the imaging data . A 6:08 min pASL scan and a 6:15 min BOLD fMRI scans were performed ( ‘awake’ pASL and BOLD scans; Figure 1C ) , in order to assess rCBF and functional connectivity at baseline . The pASL scans were collected using the ‘PICORE-Q2TIPS’ MRI labeling method ( Luh et al . , 1999 ) ( TR/TE/TI1/TI2 = 3000/13/700/1700 ms , voxel size = 3 . 5*3 . 5*5 mm , number of slices = 16 ) . ‘Tag’ images were acquired by labeling a thick inversion slab ( 110 mm ) , proximal to the imaging slices ( gap = 21 . 1 mm ) . ‘Control’ images were acquired interleaved with the tag images , by applying an off-resonance inversion pulse without any spatial encoding gradient . At the beginning of each pASL scan , an M0 scan ( i . e . , the longitudinal magnetization of fully relaxed tissue ) was acquired for rCBF quantification purposes . The pASL imaging volume common to all subjects scanned covered most of the cerebrum , and extended ventrally to the midbrain and dorsally to the vertex ( Figure 1D ) . BOLD fMRI data were collected using a whole brain T2*-weighted gradient echo BOLD echo planar imaging pulse sequence was used ( TR/TE = 3000/35 ms , flip angle = 90° , voxel size = 2 . 3 × 2 . 3 × 3 . 8 mm , number of slices = 35 ) . After the awake scans , dexmedetomidine was administered as a 1 mcg/kg loading bolus over 10 min , followed by a 0 . 7 mcg/kg/hr infusion to maintain unconsciousness . Another pASL scan was initiated at the onset of the dexmedetomidine infusion , with identical imaging parameters as the awake pASL scan , except for a longer duration ( 20:08 min ) . This longer acquisition was performed in order to ensure that perfusion data were collected for a long enough period to capture the transition from the awake to the unconscious state . During the infusion period , the study anesthesiologists monitored cuff blood pressure , capnography , electrocardiogram , and pulse-oximetry . Volunteers were instructed to keep their eyes open during the course of the study; loss of consciousness was defined as the onset of sustained eye closure and lack of response to a verbal request to open the eyes . After the onset of unconsciousness , we then performed a 6:15 min ‘unconscious’ BOLD fMRI scan ( to evaluate unconsciousness-related changes in functional connectivity ) . After acquisition of all images , the dexmedetomidine infusion was discontinued , and a 20:08 min pASL scan was performed to assess changes in rCBF upon recovery . Spontaneous eye opening and a positive response to give a thumbs-up signal were used to determine recovery of consciousness . All 10 subjects who experienced spontaneous eye opening during the final 6:08 min of the recovery of consciousness pASL successfully executed on the thumbs-up signal request . An additional 6:15 min ‘recovery’ BOLD fMRI scan was performed , in order to assess variations in functional connectivity upon recovery . PET data collected from 10 subjects ( two visits each ) and stored in list mode format were binned into sinograms . In addition to the static frame corresponding to 40–60 min post FDG administration , dynamic frames of progressively longer duration were generated for quantitative analysis . PET images were reconstructed using an ordered subsets expectation maximization algorithm , with 3 iterations and 21 subsets . Corrections were applied to account for variable detector efficiencies and dead time , photon attenuation and scatter and radioactive decay using the software provided by the manufacturer . We employed a method that is similar to the generation of attenuation maps for computed tomography data to generate our head attenuation maps from the MPRAGE data . This head attenuation map was combined with the hardware ( i . e . , RF coil , patient table , etc ) attenuation map provided by the manufacturer . Spatial smoothing was performed post-reconstruction using a 4 mm full-width-at-half-maximum ( FWHM ) Gaussian kernel . Standardized uptake values ( SUV; i . e . , mean radioactivity/injected dose/weight ) were computed voxelwise from the emission data collected 40–60 min post-injection . In order to quantitatively assess metabolic changes , cerebral metabolic rate of glucose ( CMRglc ) were computed from the dynamic PET frames with the non-invasive method proposed by Wu ( 2008 ) , using the whole-brain as our reference region . SUV and CMRglc maps were then coregistered with the high resolution MRI scan using BBREGISTER tool ( Greve and Fischl , 2009 ) from the FreeSurfer suite ( http://surfer . nmr . mgh . harvard . edu/ ) , normalized to the Montreal Neurological Institute ( MNI ) space using nonlinear registration ( FNIRT , from the FSL suite; FMRIB's Software Library , www . fmrib . ox . ac . uk/fsl/ ) ( Smith et al . , 2004 ) and then smoothed with an 8 mm FWHM kernel . A paired , whole-brain , voxel wise , mixed effects analysis was conducted to compare SUV and CMRglc across visits ( n = 10 ) . Statistical parametric maps were thresholded using clusters determined by a voxel-wise threshold ( Z > 2 . 3 ) and a ( corrected ) cluster significance threshold of p = 0 . 05 ( Worsley , 2001 ) . fMRI data were collected from 17 subjects . However , BOLD data ( awake , unconscious , and recovery ) were successfully collected in only 15 subjects . This is because BOLD data was collected with a different set of parameters in one subject and another subject exited the scanner immediately prior to the recovery BOLD scan . ASL data preprocessing was performed using a combination of analysis packages including FSL , and Freesurfer . The ‘tag’ , ‘control’ , and M0 scans were first motion-corrected using MCFLIRT ( Jenkinson et al . , 2002 ) . Then , tag and control scans were surround subtracted ( i . e . , given each tagX , [ ( controlX−1 + control X+1 ) /2 − tagX] ) to achieve perfusion-weighted images . Quantification of rCBF was performed using the ASLtbx toolbox ( Wang et al . , 2008; Wang , 2012 ) from the following data: the whole 6 min awake scan , the last 6 min of the 20 min induction of unconsciousness scan ( i . e . , the portion of the scan during which all subjects were unconscious; ‘unconscious’ scan ) and the last 6 min of the 20-min recovery of consciousness scan ( i . e . , the portion of the scan during which 10 subjects had recovered; ‘recovery’ scan ) . rCBF maps were coregistered with the high resolution MR scan using Freesurfer's BBREGISTER tool ( Greve and Fischl , 2009 ) , normalized to MNI space using nonlinear registration ( FNIRT ) ( Smith et al . , 2004 ) and then smoothed with a 7 mm FWHM kernel . A paired , voxel wise , mixed effects analyses were conducted to compare rCBF between the awake and unconscious states data ( n = 17 ) . This contrast was computed only within voxels imaged in all subjects ( see Figure 2D for coverage common to all subjects ) , as assessed through the computation of a conjunction mask from all the MNI-registered M0 scans . Given that rCBF estimation in the white matter presents methodological challenges ( particularly given the poor signal-to-noise ratio and longer arterial transit time ) ( Liu et al . , 2010 ) , voxels classified as white matter in ( the arbitrary value of ) ≥ 70% of subjects by the Harvard–Oxford Subcortical Structural Atlas ( Center for Morphometric Analyses , http://www . cma . mgh . harvard . edu/fsl_atlas . html ) were excluded from the analyses . Subsequently , another paired analysis was performed to compare rCBF between the unconscious and recovery states . In this analysis , only the 10 subjects who were awake for the full duration of the last 6 min of the 20-min recovery scans were included . This analysis was focused within search areas determined by the statistically significant clusters in the ‘unconscious vs awake’ contrast . Unless otherwise specified , all rCBF statistical parametric maps were thresholded using clusters determined by a voxel-wise threshold ( Z > 2 . 3 ) and a ( corrected ) cluster significance threshold of p = 0 . 05 ( Worsley , 2001 ) . Resting-state BOLD fMRI data were used to perform functional connectivity analyses . Datasets were slice-timing corrected ( SLICETIMER ) , motion-corrected ( MCFLIRT ) , brain-extracted ( BET ) , registered to MPRAGE ( BBREGISTER ) and then to the MNI152 atlas ( FNIRT ) , smoothed with a 5 mm FMWH kernel and high-pass filtered ( cutoff = 0 . 008 Hz ) . As ASL and PET analyses revealed that unconsciousness was associated with a reduction in glucose metabolism and rCBF in regions belonging to distinct resting-state networks ( Default Mode and both Frontoparietal Control networks; DMN , FCN ) , we assessed how intrinsic functional connectivity of these specific networks was affected by unconsciousness , using the dual regression approach ( Filippini et al . , 2009; Zuo et al . , 2010 ) . All the preprocessed datasets were concatenated to create a single 4D dataset . A probabilistic Independent Component Analysis ( pICA ) ( Beckmann et al . , 2005 ) was performed using Multivariate Exploratory Linear Optimized Decomposition into Independent Components ( MELODIC ) on this concatenated 4D dataset , limiting the number of independent components ( ICs ) to 25 , as in previous publications using the dual regression approach ( Filippini et al . , 2009; Napadow et al . , 2010 , 2012; Kim et al . , 2013; Loggia et al . , 2013 ) . From the pool of the 25 ICs , the DMN , right and left FCNs were clearly identified . Group-level spatial maps for each of these RSNs were used as a set of spatial regressors in a General Linear Model ( GLM ) , in order to identify the individual subjects' time course associated with each group-level map . These time courses were then variance normalized and used as a set of temporal regressors in a GLM , to find subject-specific maps associated with the different group-level independent components . In this GLM , explanatory variables also included six motion parameters and the time courses from ventricles and white matter as covariates of no interest . Subject-specific maps were compared across states ( awake , unconscious , recovery ) using paired t-tests . As our dual regression ICA approach identified altered DMN functional connectivity to the thalamus , we used this thalamic region as a seed for seed-based functional connectivity analysis . This follow-up analysis was done with the purpose of ( 1 ) determining whether specific portions of the DMN ( e . g . , more posterior regions within the network ) may be responsible for driving the effect and ( 2 ) assessing changes in functional connectivity between the thalamus and other , non-DMN regions . The extracted fMRI time series , variance normalized , was used as a regressor in a GLM for the awake , unconscious , and recovery data . The same nuisance regressors adopted in the dual regression analyses ( see above ) , with the addition of global signal time course were used in the seed-based analyses . Statistical parametric maps were thresholded using clusters determined by a voxel-wise threshold ( Z > 2 . 3 ) and a ( corrected ) cluster significance threshold of p = 0 . 05 ( Worsley , 2001 ) . From representative clusters identified in the various analyses , data were extracted and displayed for illustrative purposes . Human brain atlases were used for anatomical reference of the cerebrum ( Mai et al . , 2008 ) and the cerebellum ( Diedrichsen et al . , 2009 ) . | Although we are all familiar with the experience of being conscious , explaining precisely what consciousness is and how it arises from activity in the brain remains extremely challenging . Indeed , explaining consciousness is so challenging that it is sometimes referred to as ‘the hard question’ of neuroscience . One way to obtain insights into the neural basis of consciousness is to compare patterns of activity in the brains of conscious subjects with patterns of brain activity in the same subjects under anesthesia . The results of some experiments of this kind suggest that loss of consciousness occurs when the communication between specific regions within the outer layer of the brain , the cortex , is disrupted . However , other studies seem to contradict these findings by showing that this communication can sometimes remain intact in unconscious subjects . Akeju , Loggia et al . have now resolved this issue by using brain imaging to examine the changes that occur as healthy volunteers enter and emerge from a light form of anesthesia roughly equivalent to non-REM sleep . An imaging technique called PET revealed that the loss of consciousness in the subjects was accompanied by reduced activity in a structure deep within the brain called the thalamus . Reduced activity was also seen in areas of cortex at the front and back of the brain . A technique called fMRI showed in turn that communication between the cortex and the thalamus was disrupted as subjects drifted into unconsciousness , whereas communication between cortical regions was spared . As subjects awakened from the anesthesia , communication between the thalamus and the cortex was restored . These results suggest that changes within distinct brain regions give rise to different depths of unconsciousness . Loss of communication between the thalamus and the cortex generates the unconsciousness of sleep or light anesthesia , while the additional loss of communication between cortical regions generates the unconsciousness of general anesthesia or coma . In addition to explaining the mixed results seen in previous experiments , this distinction could lead to advances in the diagnosis of patients with disorders of consciousness , and even to the development of therapies that target the thalamus and its connections with cortex . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2014 | Disruption of thalamic functional connectivity is a neural correlate of dexmedetomidine-induced unconsciousness |
Post-translational modifications of proteins have emerged as a major mechanism for regulating gene expression . However , our understanding of how histone modifications directly affect chromatin function remains limited . In this study , we investigate acetylation of histone H3 at lysine 64 ( H3K64ac ) , a previously uncharacterized acetylation on the lateral surface of the histone octamer . We show that H3K64ac regulates nucleosome stability and facilitates nucleosome eviction and hence gene expression in vivo . In line with this , we demonstrate that H3K64ac is enriched in vivo at the transcriptional start sites of active genes and it defines transcriptionally active chromatin . Moreover , we find that the p300 co-activator acetylates H3K64 , and consistent with a transcriptional activation function , H3K64ac opposes its repressive counterpart H3K64me3 . Our findings reveal an important role for a histone modification within the nucleosome core as a regulator of chromatin function and they demonstrate that lateral surface modifications can define functionally opposing chromatin states .
Histone modifications are central to the regulation of all chromatin-based processes . Four core histones—H3 , H4 , H2A , and H2B—comprise the nucleosomal core particle , and each may be decorated with multiple covalent modifications , including acetylation , methylation , phosphorylation , sumoylation , and ubiquitination ( Kouzarides , 2007 ) . To date , most attention has focused on modifications within the flexible N-terminal tails of histones , which extend from their own nucleosome . Due to their accessibility , ‘reader’ or effector proteins selectively bind to modified sites in the tails to mediate downstream effects . In this way ‘readers’ can provide a relatively simple mechanism enabling cells to decipher the so-called ‘histone-code’ , to facilitate the regulation of biological processes such as transcription , DNA replication , and damage repair . Interestingly , covalent modifications also occur within the globular domain of histones ( Garcia , 2009 ) , especially at positions that are in close contact with the nucleosomal DNA wrapped around each octamer . In particular , modifications on the outer surface of the histone octamer , the so-called lateral surface , have the potential to directly influence chromatin structure by altering histone–histone or histone–DNA interactions ( Cosgrove , 2007; Tropberger and Schneider , 2010; Tropberger et al . , 2013 ) . Because of their structurally important position close to the DNA , one can directly address the mechanism ( s ) by which these lateral surface modifications impact on nucleosome dynamics and chromatin function . This is in contrast to tail modifications that play a mainly indirect role in chromatin regulation through recruitment of effector proteins . Understanding how histone modifications ultimately impact on chromatin function is still an open challenge and this has not been helped by the fact that the repertoire of known modifications is far from complete ( Tan et al . , 2011 ) . Here , we have investigated the functional mechanism of an uncharacterized acetylation site , lysine 64 of histone H3 ( H3K64ac ) , which lies within the H3 globular domain . We demonstrate that this lateral surface modification can directly influence nucleosomal stability and dynamics , which consequently affects transcriptional regulation .
We used mass spectrometry to identify novel histone modifications and found an uncharacterized acetylation site on histone H3 lysine 64 ( H3K64ac ) ( Figure 1A , Figure 1—figure supplement 1A , B ) . H3K64 is the first amino acid of the H3 alpha1 helix , the first of three alpha helices in the histone fold . It is found on the lateral surface of the histone octamer in close proximity to the inner gyre of DNA ( Davey et al . , 2002; Figure 1B ) at a location distinct from other potentially acetylated residues , such as H3K56 . 10 . 7554/eLife . 01632 . 003Figure 1 . Acetylation of K64 in histone H3 is a novel histone modification . ( A ) CID MS/MS spectrum of the tryptic peptide ( K ( ac ) LPFQR; m/z [MH22+] 415 . 74823 ) derived from endogenous histone H3 demonstrating K64 acetylation . The presence of the b2 , the y5 , the y5-NH3 ( y5* ) and the immonium ions derived from ( ɛ ) acetyl-lysine ( enlarged spectrum inlet ) are used for site localization to lysine 64 . ( B ) H3K64 is on the lateral surface of the histone octamer . 3D modelling of a nucleosome . H3 dimer is shown in blue and the proximity of acetylated H3K64 ( red rectangle ) with the DNA is highlighted in the zoomed-in inset . In red are shown the main-chain and the side-chain of lysine 64 . The acetyl group linked to the side-chain terminal nitrogen atom ( in blue ) appears in green , with its oxygen atom in purple . All hydrogen atoms are displayed in grey . ( C ) Peptide competition of H3K64ac immunoblot . H3K64ac antibody was pre-adsorbed with 50 pmoles/ml of indicated peptides . Amido black staining is shown as loading control ( bottom panel ) . ( D ) Immunoblot analysis of H3K64 acetylation state in different cell lines , treated ( + ) or not ( − ) with HDACs inhibitor Na-Butyrate . Untreated MCF7 nuclesosomes were used as control . Ponceau staining is shown as loading control ( bottom panel ) . ( E ) H3K64ac is enriched in euchromatic regions . H3K64 acetylation ( ac , upper panels ) and tri-methylation patterns ( me3 , lower panels ) in MEFs . DAPI dense foci represent pericentric heterochromatin . Single sections are shown . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 00310 . 7554/eLife . 01632 . 004Figure 1—figure supplement 1 . MS/MS spectra of H3K64ac peptides . CID MS/MS spectra of a chymotryptic ( LIRK ( ac ) LPF; m/z [MH22+] 464 . 80363 ) and a tryptic ( K ( ac ) LPFQR; m/z [MH22+] 415 . 74823 ) peptide ( A and B panels , respectively ) derived from endogenous histone H3 demonstrating the acetylation of K64 . The mass difference between the b4 and b3 ion of the chymotryptic peptide LIRK ( ac ) LPF indicates the presence of acetyl-lysine . This is further corroborated by the virtually identical MS/MS spectral overlay of the tryptic peptide K ( ac ) LPFQR ( see also Figure 1 ) either derived from endogenous H3 ( illustrated in black ) or the synthetic peptide STELLIRK ( ac ) LPFQRLVGC ( amide ) ( highlighted in red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 00410 . 7554/eLife . 01632 . 005Figure 1—figure supplement 2 . Anti-H3K64ac antibody validation . ( A ) Western blot analysis with H3K64ac antibody ( left panel ) on nuclear extracts from untreated ( − ) or Na-butyrate-treated ( + ) HeLa cells . Ponceau staining is shown as loading control ( right panel ) . ( B ) , ( C ) , ( D ) , and ( E ) Immuno-dot blots showing specific reactivity of the H3K64ac antibody . Aliquots of indicated pmoles of linear H3 peptides were spotted on the membrane . Of note , detection of peptides in ( E ) has been checked with H3K9ac , H3K18ac , and H3K27ac specific antibodies . ( F ) Peptide competition assay of H3K64ac immunoblot . H3K64ac antibody was pre-adsorbed with 50 pmoles/ml of indicated peptides . Ponceau staining is shown as loading control ( bottom panel ) . ( G ) The H3K64ac antibody detects globular , tailless H3 . Nucleosomes from NIH3T3 cells were incubated with no ( − ) or increasing amounts of trypsin to cleave the histone tails and immunoblotted as indicated for H3K9ac , H3K18ac , H3K27ac , H3K64ac and H3 . The position of full-length and tailless H3 is indicated and loading was controlled by ponceau staining . ( H ) Peptide competition assay in immunofluorescence . MEFs were treated overnight with Na-butyrate ( NaB ) and then stained with H3K64ac antibody , pre-incubated with 100 pmoles/ml of either unmodified or acetylated K64 peptide . ( I ) Immunoblot analysis of H3K64 acetylation state in additional cell lines ( Drosophila S2 , mouse MEFs , human HeLa ) with H3K64ac antibody . Inhibition of HDACs by Na-butyrate treatment increases acetylation levels . Ponceau staining is shown as loading control ( bottom panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 00510 . 7554/eLife . 01632 . 006Figure 1—figure supplement 3 . H3K64ac is excluded from heterochromatic foci . DAPI dense foci ( panel 1 ) represent pericentric heterochromatin where H3K9me3 is enriched and H3K64ac excluded ( compare panels 2 , 3 with 4 and panels 5 with 6 , representing a magnified view of the cell indicated by a white star ) . In panel 4 and 6 H3K9me3 appears in red and H3K64ac in green . Shown are single sections , scale bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 006 To study the biological function of H3K64ac , we first performed an exhaustive characterization of an antibody raised against this modification . This antibody specifically detects endogenous acetylated histone H3 ( Figure 1—figure supplement 2A ) and recognizes a K64-acetylated peptide with a high degree of specificity ( Figure 1—figure supplement 2B–E ) compared to other H3 acetylated lysines . Its recognition of H3 was efficiently competed by the immunizing peptide , but not by other peptides containing acetylated , methylated , or unmodified histone regions ( Figure 1C , Figure 1—figure supplement 2F ) . Furthermore , limited tryptic digestion of native nucleosomes , which removes the H3 tails whilst leaving the DNA-protected H3 core region largely intact , confirmed the antibody’s specificity . Indeed , this digestion treatment resulted in the loss of signal for the tail modifications , such as H3K9ac , H3K18ac , and H3K27ac . In contrast , the truncated H3 core was still recognized by the H3K64ac antibody at levels comparable to the undigested H3 ( Figure 1—figure supplement 2G ) . Using this antibody , we found that H3K64ac is present in a variety of mouse and human cell lines and tissues suggesting a rather ubiquitous function ( Figure 1D , Figure 1—figure supplement 2I ) . Upon HDAC-inhibitor treatment , H3K64 acetylation levels increased ( Figure 1D , Figure 1—figure supplement 2I ) . Immunofluorescence ( IF ) showed a distinct nuclear localization of H3K64ac with a relative depletion from heterochromatin ( Figure 1E , Figure 1—figure supplement 3 , compare panels 5 and 6 , and Figure 1—figure supplement 2H ) . Interestingly , this localization pattern is the opposite to that of H3K64me3 ( Figure 1E ) . We previously established that H3K64me3 is a novel repressive mark enriched in pericentromeric heterochromatin that might help to ‘lock’ the conformation and/or position of the nucleosome , and consequently the surrounding chromatin ( Daujat et al . , 2009; Lange et al . , 2013 ) To obtain a comprehensive picture of the genomic distribution of H3K64ac , we performed ChIP-on-chip assays using chromatin isolated from mouse embryonic stem ( ES ) cells and Nimblegen tiling microarrays ( Figure 2—figure supplement 1A; Lienert et al . , 2011 ) . In line with the euchromatic localization detected in IF , we found strong enrichment of H3K64ac at the transcriptional start site ( TSS ) of active genes ( Figure 2A , Figure 2—figure supplement 1A–C ) . At TSSs , we detected a strong correlation between the enrichment of H3K64ac and RNA Polymerase II occupancy ( Figure 2B ) , as well as with the presence of active histone marks ( Figure 2—figure supplement 1D ) , suggesting a role for H3K64ac in transcriptional activation . Of note , local H3K64ac enrichment is indicative of the steady-state mRNA level of the respective gene ( Figure 2C ) and it is not simply due to increased histone H3 density ( Figure 2—figure supplement 1E ) . Consistent with this , H3K64ac enrichment is anti-correlated with repressive marks such as H3K27me3 and H3K64me3 ( Figure 2—figure supplement 1D ) . Furthermore , we find H3K64ac strongly enriched at enhancers ( Figure 2A ) , with a preference for active enhancers , where it co-localizes with established enhancer marks such as H3K27ac , H3K4me1 , and p300-binding ( Figure 2D; Creyghton et al . , 2010; Rada-Iglesias et al . , 2011 ) . Multiple single gene validations by ChIP-qPCR confirmed the genome-wide data and also revealed that H3K64ac levels are very low at repetitive elements ( Figure 2—figure supplement 2A , B ) , where H3K64me3 is highly enriched ( Daujat et al . , 2009 ) , again indicating opposing genomic localizations for these two marks . 10 . 7554/eLife . 01632 . 007Figure 2 . H3K64ac is enriched genome-wide at active regulatory regions . ( A ) H3K64ac is predominantly localized to active TSS and enhancers in mouse ES cells ( see also D ) . Boxplot showing H3K64ac signal intensity ( log2 ChIP/input ) for all microarray probes at active TSS , inactive TSS , enhancers , gene bodies , and intergenic regions . ( B ) Comparison of H3K64ac and RNA Pol II at TSS . Scatterplot showing signal density distribution and global correlation . The green line is a loess-fitted trend line . ( C ) Meta-gene plot showing H3K64ac enrichment around TSS grouped according to their expression level . ( D ) Boxplot comparing H3K64ac to ‘enhancer-specific’ histone modifications and p300 levels ( Creyghton et al . , 2010 ) at active enhancers , inactive enhancers and control regions ( enhancer regions shifted by 100 kb ) . H3K64ac levels were measured by ChIP-on-chip ( left Y axis ) , whereas the other modifications and p300 by ChIP-seq ( right Y axis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 00710 . 7554/eLife . 01632 . 008Figure 2—figure supplement 1 . H3K64ac distribution within active chromatin . ( A ) Scatterplot showing global correlation between raw data from the two biological H3K64ac ChIP-on-chip replicate experiments . The green line is a loess-fitted trend line . ( B ) UCSC browser screenshots of H3K64ac enrichment at the proximal promoter of active genes ( Eif3a , upper panel and Taf5 , lower panel ) . ( C ) UCSC browser screenshots of H3K64ac enrichment at the proximal promoter of inactive genes ( Cyp2c55 and Cyp2c65 ) . For ( B ) and ( C ) shown are the enrichments for H3K64ac ( as log2 values of ChIP/input ) for one biological replicate . Areas in light blue highlight the region surrounding the TSS of the indicated genes . ( D ) Scatterplots showing global correlation at TSS between H3K64ac and ‘active’ modifications such as H3K9ac and H3K4me2 and ‘repressive’ marks such as H3K27me3 and H3K64me3 . The green line is a loess-fitted trend line . ( E ) Meta-gene plot showing total H3 enrichment around TSS grouped according to their expression level . Of note , H3 density is rather uniform and the slight enrichment proximal to the TSS cannot explain the strong enrichment of H3K64ac detected in the same region ( Figure 2C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 00810 . 7554/eLife . 01632 . 009Figure 2—figure supplement 2 . Genome-wide data validation by ChIP-qPCR . ( A ) Native ChIP analysis of H3K64ac on different genomic regions in mouse ES cells . H3K64ac is enriched at the TSS of active genes , but depleted from inactive genes and repeats . ( B ) Native ChIP analysis performed as in ( A ) . As expected the active H3K4me3 mark is enriched at active genes , but depleted from inactive genes and repeats , in an exact opposite manner to H3K9me3 enriched in repetitive regions . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 009 To study H3K64ac in a dynamic system , we followed the kinetics of H3K64ac during ES cell differentiation ( Figure 3A ) . In pluripotent ES cells , H3K64ac is strongly enriched at active pluripotency-associated genes such as Nanog , Pou5f1 and Dppa3 , whereas after retinoic acid-induced differentiation this enrichment shifts towards active differentiation-associated genes ( e . g . , Hoxb3 and Hoxd3 , Pax6 ) demonstrating that H3K64ac levels reflect transcriptional activity during differentiation . Since we found H3K64ac tightly associated with transcribed regions , we next asked if H3K64ac is enriched on specific H3 variants . In line with previous findings that H3 . 3 has covalent modifications associated with transcriptionally active chromatin ( Hake et al . , 2006 ) , we found the highest enrichment of H3K64ac on the H3 variant H3 . 3 ( Figure 3B , Figure 3—figure supplement 1A ) . Importantly these experiments also show that mutation of K64 to R results in a loss of detection of H3 by the H3K64ac antibody used ( Figure 3—figure supplement 1B ) , suggesting high specificity of the antibody . 10 . 7554/eLife . 01632 . 010Figure 3 . H3K64ac is enriched on active genes . ( A ) ChIP analysis of H3K64ac on pluripotency genes and differentiation-specific genes . Real-time PCR analysis for indicated promoter regions and gene desert in undifferentiated ( yellow bars ) or retinoic acid-induced ( blue bars ) ES cells . ( B ) Distribution of H3K64ac among the H3 variants . Flag-HA-tagged histones H3 . 1 , H3 . 2 , or H3 . 3 were immunoprecipitated and probed with H3K64ac antibody . Average quantification of three biological replicates ( ±SD ) showing H3K64ac signal over HA relative to H3 . 1 . ( C ) The active alleles of ICRs are specifically marked by H3K64ac . Native ChIP performed on primary embryonic fibroblasts of ( C57Bl/6 x JF1 ) F1 genotype followed by radioactive PCR across polymorphic nucleotides between the paternal JF1 ( M . m . molossinus ) and the maternal C57BL/6J ( B6 ) genomes . Single-strand conformation polymorphisms ( SSCP ) were revealed by electrophoresis through a non-denaturing agarose gel . The left panels show PCRs on control genomic DNAs to depict the SSCP polymorphisms used . In the input chromatin ( input ) , the parental were equally represented at the loci analysed . U , unbound fraction; B , bound fraction; M , maternal allele; P , paternal allele . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 01010 . 7554/eLife . 01632 . 011Figure 3—figure supplement 1 . Immunoblots of H3K64ac distribution among H3 variants . ( A ) Representative immunoblot of Flag-immunoprecipitated H3 variants , probed with H3K64ac antibody ( top panel ) and HA antibody as loading control ( bottom panel ) . An antibody control lane with an H3 . 2K64R mutant is also shown . ( B ) Antibody specificity was verified using H3 . 1K64R and H3 . 3K64R point mutants . Flag immunoprecipitated K64R mutants or wt variants were probed with H3K64ac antibody ( top ) , and anti-HA antibody as loading control ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 01110 . 7554/eLife . 01632 . 012Figure 3—figure supplement 2 . The active alleles of ICRs are specifically marked by H3K64ac . UCSC browser screenshots of H3K64ac enrichment at the KvDMR1 ICR region in mouse ES cells validating enrichment at ICRs . Shown are the log2 values of ChIP/input measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 012 Our staining and ChIP data suggested mutually exclusive distribution patterns of H3K64 acetylation and methylation . To corroborate this in a functional model , we made use of imprinted loci; an ideal system to study potentially opposing histone marks , as one allele is transcriptionally silent whilst the other one is active ( Singh et al . , 2010 ) . We analysed five imprinting control regions ( ICRs ) and in each case we found that the transcriptionally active alleles were specifically enriched in H3K64ac , whereas the inactive ones were enriched in H3K64me3 ( Figure 3C , Figure 3—figure supplement 2 ) . These data suggest that H3K64ac and H3K64me3 can define functionally opposing chromatin states . To identify the enzyme ( s ) responsible for H3K64ac , we systematically depleted candidates from different HAT families . In these assays , knockdown of p300 and CBP , but not of other HATs decreased the steady-state levels of H3K64ac ( Figure 4A , Figure 4—figure supplement 1A ) . This decrease was most pronounced at p300/CBP-specific genomic target regions ( Figure 4B ) . In line with this , overexpression of p300 resulted in increased levels of H3K64ac ( Figure 4C , Figure 4—figure supplement 1B ) , and p300 and H3K64ac distributions showed a strong correlation ( Figure 4—figure supplement 1C ) . Moreover , p300 and CBP can acetylate H3K64 in vitro on free H3 ( Figure 4D ) and within chromatin ( Figure 4—figure supplement 1D ) . Altogether these data clearly establish p300/CBP as H3K64 acetyltransferases , not excluding the presence of additional H3K64 acetyltransferases . 10 . 7554/eLife . 01632 . 013Figure 4 . p300 acetylates H3K64 in vivo and in vitro . ( A ) siRNA-mediated depletion of HATs ( as indicated ) in MCF7 cells . Immunoblot analysis of global H3K64ac levels and additional modifications as controls for siRNA efficiency . Anti-H3 blot and Ponceau staining are shown as loading controls . ( B ) ChIP analysis of H3K64ac enrichment on different mouse genomic regions ( as indicated ) upon depletion of p300/CBP ( yellow ) compared to the control knock-down ( scramble , blue ) . ( C ) Overexpression of p300 in HEK293 cells . Control ( empty vector ) or p300 overexpressing cells co-expressed mCherry ( red , top panel ) and were assessed for H3K64ac levels ( green ) in immunoflourescence . ( D ) In vitro HAT assay with p300 or CBP using recombinant H3 ( wt or K64A mutant ) as substrate and probed with the H3K64ac antibody ( top panel ) . Ponceau staining as loading control ( middle panel ) and H3K9ac western blot as activity control ( bottom panel ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 01310 . 7554/eLife . 01632 . 014Figure 4—figure supplement 1 . Validation and supporting experiment to establish p300 activity on H3K64 . ( A ) Validation of HATs knock-down . Upon transfection of targeted siRNA , mRNA levels of indicated HATs are strongly reduced compared to the scramble control . ( B ) p300 overexpression in HEK293 cells ( marked by mCherry expression , red ) leads to an increase in the acetylation level of H3K9 ( left panel ) , a p300 target , but no increase in H4K16 acetylation ( right panel ) . ( C ) Scatterplot showing signal density distribution and global correlation between H3K64ac and p300 at TSS . Green lines are loess-fitted trend lines . ( D ) Immunoblot showing specific p300 activity on H3K64 ( top ) within an immobilised chromatin template . p300 was recruited by the transcriptional factor VP16 . Autoradiography ( 3H-acetyl-CoA incorporation , middle ) as enzyme activity control and ponceau staining as loading control ( bottom ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 014 Given the location of H3K64ac on the nucleosome’s lateral surface , a potential mechanism is that it acts through modulating ATP-dependent chromatin remodelling and/or nucleosome stability . To investigate this , we produced recombinant histone H3 acetylated on K64 in Escherichia coli using site-specific genetically directed incorporation of acetyl-lysine ( Neumann et al . , 2009; Figure 5—figure supplement 1A , B ) . To address effects on chromatin remodelling , we incubated H3K64ac and unmodified nucleosomes with two different chromatin remodelers belonging to two different Snf2 subfamilies , Chd1 and RSC ( Flaus et al . , 2006; Clapier and Cairns , 2009 ) . Our results show that Chd1 repositioned H3K64-acetylated nucleosomes faster than unmodified nucleosomes . This was not the case when we used the RSC enzyme ( Figure 5—figure supplement 2 ) . This suggests that acetylation of H3K64 could differentially affect remodelling enzymes . Next , we sought to investigate whether H3K64ac also impacts on passive fluctuations in nucleosome structure and interrogated whether H3K64ac affects the stability of DNA association with histone octamers within nucleosomes . To do this , we attached fluorescent dyes to specific sites on the DNA , 35 bp from each end of the nucleosomal DNA , and performed FRET measurements ( Neumann et al . , 2009 ) . Using this strategy , we observed that the FRET interaction for H3K64ac nucleosomes was more sensitive to salt-disruption than unmodified nucleosomes at NaCl concentrations ranging between ∼0 . 5 M and 1 . 0 M ( Figure 5A ) . In parallel , we also measured the salt-dependent nucleosome stability by single-molecule FRET , which again resulted in a lower stability of H3K64ac nucleosomes ( Figure 5—figure supplement 3 ) . Together these data demonstrate a decreased stability of H3K64ac nucleosomes , distinguishing this acetylation from H3K56ac that was reported not to significantly affect nucleosome stability under comparable conditions ( Neumann et al . , 2009 ) . 10 . 7554/eLife . 01632 . 015Figure 5 . H3K64 acetylation affects DNA-octamer interactions . ( A ) H3K64ac nucleosomes are more sensitive to salt-induced disruption in in vitro FRET assays . Left , different views of a nucleosome with H3K64 highlighted in blue ( spacefill ) and the internal positions of the dye-labeled bases 35 bp from the end of the nucleosome . Right , the average ( ±s . e . ) of three separate titrations is shown for unmodified ( black squares ) , H3K9ac ( unfilled squares ) , and H3K64ac nucleosomes ( unfilled triangles ) . ( B ) Competitive in vitro nucleosome assembly reactions suggest that H3K64ac weakens histone–DNA interactions . Unmodified , H3K9ac , and H3K64ac nucleosomes were assembled in the presence of an excess of non-specific competitor DNA . The average ( ±s . e . ) of three independent competitive assembly reactions is shown . ( C ) In spermatogenesis H3K64ac levels are high in elongating spermatids during the wave of massive chromatin remodelling . Schematic of mouse spermatogenesis ( top ) . Immunofluorescence stainings ( IF ) show high H3K64ac levels in elongating spermatids ( E ) and below detection limit in spermatocytes ( P ) and round spermatids ( R ) . Scale bar , 5 µm . ( D ) Immunohistochemistry stainings ( IHC ) on sections of mouse testis tubules . H3K64ac ( brown ) is enriched in elongating spermatids ( stages IX-X ) , which are present near the lumen . Scale bar , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 01510 . 7554/eLife . 01632 . 016Figure 5—figure supplement 1 . Validation of the substrates used in the in vitro assays . ( A ) Validation of H3 acetylation on the octamer substrates used in the in vitro assays . The immunoblots show the acetylation levels of the reconstituted octamers specifically acetylated either on H3K64 or H3K9 compared with the unmodified H3 . Ponceau staining is shown as loading control . ( B ) Staining of protein gel used to check that equal amounts of proteins had been used in the experiments and to assess H2A–H2B dimer , as well as unmodified , H3K64ac , and H3K9ac tetramer qualities . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 01610 . 7554/eLife . 01632 . 017Figure 5—figure supplement 2 . Effect of H3K64ac on repositioning by RSC and Chd1 . RSC or Chd1 were incubated with a mixture containing equal amounts of differentially fluorescent-labeled nucleosomes: unmodified ( Cy5 ) , H3K64ac ( Cy3 ) . Bottom panels show representative native PAGEs used to calculate the relative rates of remodeling by RSC and Chd1 . Top panels show the relative rates of remodeling determined by fitting the data with a single exponential function ( solid line for unmodified , dashed line for H3K64ac ) . The rates of nucleosome positioning ( ±s . e . ) calculated from three independent remodeling reactions are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 01710 . 7554/eLife . 01632 . 018Figure 5—figure supplement 3 . Single-molecule analysis of unmodified and H3K64ac salt-dependent nucleosome stability . The fraction of intact nucleosomes at each salt concentration was normalized to the fraction at 0 mM NaCl . Every data point is the average ( ±SD ) of at least three measurements . The data were fit with a sigmoidal function to determine the NaCl concentrations at which the FRET signal is 0 . 5 ( C0 . 5 , see Supplemental methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 01810 . 7554/eLife . 01632 . 019Figure 5—figure supplement 4 . H3 acetylation and H3K64me3 stainings in spermatogenesis . ( A ) Immunohistochemistry ( IHC ) stainings on sections of mouse testis tubules for different acetylation sites on H3 . Note that modification shows distinct patterns of enrichment across the differentiation stages of spermatocytes . Top panels show the specific antibody staining ( brown ) together with the counterstaining ( eosin , red ) . In the middle planels , only the staining from each specific antibody ( brown ) is shown . The area in the dashed-line box is magnified 3X in the bottom panel . Scale bar , 20 μm . ( B ) Immunofluorescence on sections of mouse testis tubules . H3K64me3 signal is strongly detected in the chromocenter of round spermatids as well as in elongating spermatids , in particular at regions where histone replacement occurs late ( arrows ) , scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 01910 . 7554/eLife . 01632 . 020Figure 5—figure supplement 5 . Structural analysis of H3K64 interactions . ( A ) Distances in Å between DNA backbone phosphate oxygen atoms and H3K64 main-chain amide ( red text ) or terminal nitrogen atoms ( black text ) in the high-resolution ( 1 . 9 Å ) crystal structure of the nucleosome ( PDB ID 1KX5 ) . The H3K64 main-chain amide directly interacts with the DNA phosphate backbone , this places the H3K64 side-chain within a range to form direct interactions with the DNA . The measured distances are shown as yellow-dashed lines . ( B ) Two alternate side-chain rotamer conformations of H3K64 ( light blue ) are shown that demonstrate that the H3K64 side-chain has the potential to form direct ionic interactions with the DNA phosphate backbone . ( C ) The side-chain of H3K64 is part of an extensive water-mediated hydrogen bonding network between histones and DNA in the high-resolution crystal structure of the nucleosome ( PDB ID 1KX5 ) . Water atoms are shown as red spheres and hydrogen bonds as dashed black lines . Each of the displayed hydrogen bonds can be traced back to the water molecule bound to H3K64 . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 020 Modifications of histones that affect interactions between the H3–H4 tetramer and DNA also have the potential to influence the efficiency of chromatin assembly . To address whether H3K64ac affects the affinity of histone–DNA binding , we performed competitive nucleosome reconstitution assays ( Thåström et al . , 2004 ) . Figure 5B shows that H3K64ac reduces the level of nucleosome assembly , relative to unmodified H3 , indicating that H3K64 acetylation reduces histone–DNA binding affinity . In contrast , acetylation at H3K9 had no detectable effect ( Figure 5A , B ) , illustrating functional differences between acetylation of lateral surface and histone tail lysines . In concert , these observations demonstrate that H3K64ac directly influences histone–DNA association and nucleosome stability . This could provide a mechanistic explanation as to why H3K64ac is found in vivo at sites where increased nucleosome mobility/instability is required , such as the TSS of active genes ( Segal et al . , 2006 ) . In vivo , one of the most dramatic chromatin-reorganization events that require nucleosome instability occurs during spermatogenesis . In mammalian elongating spermatids , histones are initially replaced by the transition proteins and then by protamines ( Gaucher et al . , 2010 , 2012 ) . In spermatocytes and round spermatids H3K64ac was below the limit of detection . However , the levels of H3K64ac dramatically increased in elongating spermatids ( Figure 5C , D ) . At approximately the same time , we also detected the presence of some specific H3 tail acetylations ( Figure 5—figure supplement 4A ) . Notably , we observed this increase in H3K64ac precisely during the period of nucleosome disassembly and replacement of histones by transition proteins in spermiogenesis . Our present data suggest a potential role for lateral surface modifications , such as H3K64ac , in this replacement process by creating a less stable , more ‘open’ chromatin state . In contrast to this , H3K64me3 is enriched in the part of the spermatid where genomic regions that undergo late histone replacement are located , suggesting that H3K64me3 may ‘stabilize’ particular regions of chromatin thereby protecting them from histone exchange ( Figure 5—figure supplement 4B ) . To test whether H3K64ac can indeed promote histone displacement from chromatin , we assembled chromatin with unmodified H3 , H3K64R , or H3K64ac octamers on a template containing five GAL4-binding sites in front of a MLP promoter ( Figure 6—figure supplement 1 ) and performed a histone eviction assay ( Figure 6A , left panel ) . Using this assay , we detected more H3 in the evicted fraction from H3K64ac chromatin than from H3K64R chromatin ( Figure 6A , right panel , Figure 6—figure supplement 2A ) , suggesting that the destabilization introduced by H3K64ac facilitates histone displacement and eviction . This could explain at least in part how H3K64ac can directly impact on nucleosome stability at active promoters and is in line with our observation that H3K64ac is associated with regions of the genome that are transcriptionally active or require histone exchange/turnover . Of note , we detected equivalent binding of Nap1 to each of the H3 species used ( unmodified , K64R , or K64ac ) indicating that the observed difference in histone eviction is not due to differential Nap1 histone binding ( Figure 6—figure supplement 2B ) . 10 . 7554/eLife . 01632 . 021Figure 6 . H3K64ac can promote histone eviction and facilitate transcription . ( A ) H3K64ac promotes histone eviction . Schematic of eviction assay ( left ) shows chromatin templates assembled on a biotin-labeled pG5-MLP promoter fragment . Recruitment of transcriptional activators and p300 causes displacement of histone octamers , and histones can be trapped onto a supercoiled plasmid ( not shown ) by Nap1 . Immunoblot of the evicted fraction ( right , supernatant ) with anti-H3 antibody shows the total amount of H3 released from unmodified H3 , H3K64R , or H3K64ac chromatin templates in the presence or absence of acetyl-CoA . Ponceau staining as loading control and autoradiography ( 3H-acetyl-CoA incorporation ) on the immobilized fraction as control for p300 activity are shown . ( B ) The acetylation mimic H3K64Q promotes transcription of early-response genes . NIH3T3 cells overexpressing either wild type , K64R , or K64Q FlagHA-H3 . 3 were stimulated with 25 nM TPA for the indicated times . mRNA levels of the indicated genes were analysed by qPCR . Gapdh is shown as control gene . Average of three experiments ( ±s . e . ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 02110 . 7554/eLife . 01632 . 022Figure 6—figure supplement 1 . Validation of the chromatin templates used in the in vitro assay . Representative gels for chromatin assembly onto pG5-MLP promoter DNA fragment ( ≈ 680 bp ) . Ethidium bromide staining of a 0 . 8% agarose gel ( left ) showing chromatin reconstitution efficiency at different octamer:DNA ratios . Comparable chromatin templates were selected and used in the assays ( arrows ) . Ethidium bromide staining of native 5% polyacrylamide gel ( right ) shows that comparable chromatin templates were used in the assays . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 02210 . 7554/eLife . 01632 . 023Figure 6—figure supplement 2 . Control experiments for the eviction assay involving the role of Nap1 . ( A ) Eviction assay performed as described in the Figure 6A , but comparing the effect of Nap1 ( +/− Nap1 ) . Anti-H3 immunoblot of the supernatant fraction shows that the presence of Nap1 is essential for this assay . Ponceau staining as loading control and autoradiography ( 3H-acetyl-CoA incorporation ) on the immobilized fraction as control for p300 activity are shown . ( B ) Coomassie staining of the Nap1 pull-down of each H3 species showing that Nap1 binds equally unmodified , K64R , or K64ac H3s . Top gel shows the pull-down material from Ni2+ beads with or without Nap1; bottom gel shows the input before the pull-down , as control . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 02310 . 7554/eLife . 01632 . 024Figure 6—figure supplement 3 . Controls for H3 . 3 expression and gene induction . ( A ) Anti-HA immunoblot ( top ) shows that the generated NIH3T3 stable cell lines express each of the exogenous FlagHA-H3 . 3 types ( wt , K64R , or K64Q ) . Untransfected control is used for signal specificity; ponceau staining is used for loading control; asterisk indicates the molecular size of the FlagHA-H3 . 3 shifted upwards compared to endogenous H3 . ( B ) Expression of the negative control gene Hsp70 upon TPA induction . Hsp70 is inducible only by heat shock and its expression does not significantly vary among each stable cell line upon TPA stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 024 Finally , to demonstrate a function for H3K64ac in gene expression in vivo , we expressed wildtype H3 . 3 , H3 . 3K64Q ( acetylated lysine mimetic ) and H3 . 3K64R ( retains positive charge but is non-acetylatable ) in NIH3T3 cells ( Figure 6—figure supplement 3A ) and assayed for effects in TPA-dependent early-response gene activation . In this setup , H3K64Q promoted the expression of certain genes , such as c-fos , Egr1 , and c-myc , ( Figure 6B ) , above the levels obtained when assaying wildtype H3 . 3 or the acetylation-deficient H3 . 3K64R . Notably , expression of control genes such as Gapdh and Hsp70 was unaffected ( Figure 6B , Figure 6—figure supplement 3B ) . The observed transcriptional effects due to a single acetylation mimic within the globular domain strongly suggest that H3K64ac has an intrinsic ability to impact upon transcription mechanisms in vivo . Taken together , our results provide novel mechanistic insights into how a single histone modification , H3K64ac , within a nucleosome could specifically regulate nucleosome stability and transcription .
We have identified and functionally characterized a novel histone modification located on the octamer’s lateral surface in close proximity to the DNA backbone of the inner gyre . The main chain atoms of H3K64 are in direct contact with the DNA phosphate backbone and this places the H3K64 side chain close enough to potentially contact the DNA , with the average H3K64 side-chain nitrogen-to-DNA distance approximately 6 . 1 Å ( Figure 5—figure supplement 5A , B ) . Lysine acetylation results in increased bulk and charge neutralization of the lysine side-chain , and it is likely that acetylation of lateral surface lysines ( e . g . , H3K64 ) could interfere with histone–DNA interactions . Interestingly , in the high-resolution crystal structure of the nucleosome ( PDB 1KX5; Davey et al . , 2002 ) H3K64 is found to be involved in an extensive network of water-mediated hydrogen bonds that link numerous water molecules , histone side- and main-chain atoms and the DNA ( Figure 5—figure supplement 5C ) . Acetylation of H3K64 would be predicted to disrupt this hydrogen bond network . Davey et al . estimate from their crystallographic analysis that water-bridged interactions between histones and DNA are as important to nucleosome stability as the direct histone–DNA contacts ( Davey et al . , 2002 ) . These water-bridges provide a plasticity that allows the histone core to adapt to myriad variations in the DNA sequence and conformation . They postulate that disruption of even a single water–bridge interaction ( e . g . , acetylation of H3K64 ) has the potential to significantly affect nucleosome stability . This is indeed fully supported by our findings—we demonstrate that H3K64ac exerts a direct effect on nucleosome stability and facilitates histone eviction . H3K64 acetylation could cause the reduced FRET signal we observed in the salt-dependent dissociation assay in multiple ways . A pre-requisite for the loss of histones from the nucleosome is the dissociation of DNA from the surface of the histone octamer . Given the proximity of H3K64ac to DNA it is likely that it may facilitate this dissociation . This loss of histone–DNA contacts caused by H3K64ac may aid the loss of H2A–H2B dimers , contributing to the observed change in FRET signal . Alternatively , following the loss of H2A–H2B dimers H3K64ac may favour dissociation of DNA from histones H3 and H4 , or H3K64ac may affect water-mediated hydrogen bonds that directly influence the association of H2A–H2B dimers . These modes of action are not mutually exclusive and they are consistent with the observed effects of H3K64ac in chromatin assembly in vitro . They also all have the potential to influence chromatin organization during cycles of assembly/disassembly at gene regulatory elements . We detect an approximately 60 mM decrease in the salt-dependent stability between H3K64ac and unmodified/H3K9ac nucleosomes . This difference in salt stability of H3K64ac appears to be in a range consistent with the observed changes in nucleosome dynamics . Gansen et al . ( 2013 ) showed that simultaneous acetylation of all sites in H3 results in a decrease in nucleosome salt stability of approximately 100–130 mM NaCl . Furthermore , a fundamental change in the histone content of the nucleosome via incorporation of the histone H2A variant H2A . Z , increases salt stability by up to ∼70 mM ( Park et al . , 2004 ) . Moreover , the observed enrichment of H3K64ac at active gene promoters and during chromatin reorganization in spermiogenesis argues for a role of H3K64ac in creating an open , permissive chromatin state . Indeed , this is supported by increased histone eviction and the finding that a mutation mimicking constitutive acetylation , H3K64 to Q , can promote transcription of at least certain genes . We consider a twofold higher transcription due to a single histone mutation as biologically important . Additional lysines on the octamer’s lateral surface have been found to be acetylated , such as H3K56 , which is located closer to the ends of the nucleosomal DNA . Interestingly , these acetylations can mediate distinct effects on nucleosomes ( this work and Neumann et al . , 2009; Tropberger et al . , 2013 ) . Thus , specificity exists in the way that the biomechanical properties of nucleosomes are affected differentially by acetylation at different sites within the globular domain of H3 . Despite considerable efforts , we have so far been unable to detect specific readers , or binders of H3K64ac . Consequently , we prefer a model in which it acts via direct effects . However , H3K64ac may also function by preventing H3K64me3 from being laid down at key genomic regions . We have previously suggested that methylation of H3K64 is involved in the creation of repressive chromatin states ( Daujat et al . , 2009; Lange et al . , 2013 ) . Thus , H3K64ac might serve to prevent the nucleosome from adopting a repressive conformation , thereby maintaining chromatin in a transcriptionally competent state . Our working model ( Figure 7 ) proposes that opposing chromatin states can be defined via the mutually exclusive presence of H3K64ac or H3K64me3 . This model is reinforced by the observed distinct and largely mutually exclusive genomic distribution patterns of these two H3K64 modifications . 10 . 7554/eLife . 01632 . 025Figure 7 . Schematic model for H3K64ac function . H3K64ac ( bottom ) helps to create an open , permissive chromatin environment . In contrast to this H3K64me3 could lock the nucleosomes in a stable , repressive conformation ( top ) , resulting in an inactive chromatin environment . H3K64ac can directly act by affecting nucleosome dynamics , but also indirectly by blocking methylation of the same residue . DOI: http://dx . doi . org/10 . 7554/eLife . 01632 . 025 The data provided here strengthen the concept that lateral surface modifications play a key role in regulating chromatin function , as initially proposed by others ( Cosgrove et al . , 2004; Mersfelder and Parthun , 2006 ) and us ( Tropberger and Schneider , 2013 ) . They also extend our understanding of chromatin regulation and provide mechanistic insights into how chromatin organization is spatially and temporally controlled by globular domain modifications . The identification of novel pathways regulating chromatin function has also the potential to highlight druggable targets for diseases in which chromatin states have been distorted .
In vivo modified histone H3 was digested either with trypsin or chymotrypsin and analyzed by nanoLC-MS as previously published ( Waldmann et al . , 2011 ) . For both digests , STAGE tip-assisted sample purification was achieved essentially as described ( Rappsilber et al . , 2007 ) . Desalted samples were subsequently analyzed using nanoflow ( Agilent 1200 nanoLC , Germany ) LC-MS/MS on a linear ion trap ( LIT ) -Orbitrap ( LTQ-Orbitrap XL ) mass spectrometer ( ThermoFisher , Germany ) . Peptides were eluted with a linear gradient of 10–60% buffer B ( 80% ACN and 0 . 5% acetic acid ) at a flow rate of 250 nl/min over 40 or 60 min depending on the experiment . Data were acquired using a data-dependent ‘top 5’ method , dynamically choosing the five most abundant precursor ions from the survey scan ( mass range 250–1650 Th ) in order to isolate and fragment them in the LTQ . All data were acquired in the profile mode and dynamic exclusion was defined by a list size of 500 features and exclusion duration of 30 s with a MMD of 10 ppm . Early expiration was disabled to decrease the resequencing of isotope clusters . The isolation window for the precursor ion selection was set to 2 . 0 Th . Precursor ion charge state screening was enabled , and all unassigned charge states as well as singly charged ions were rejected . For the survey scan a target value of 10 , 00 , 000 ( 1000 ms maximal injection time ) and a resolution of 60 , 000 at m/z 400 were set ( with lock mass option enabled for the 445 . 120024 ion ) , whereas the target value for the fragment ion spectra was limited to 10 , 000 ions ( 200 ms maximal injection time ) . The general mass spectrometric conditions were: spray voltage , 2 . 3 kV; no sheath and auxiliary gas flow; ion transfer tube temperature , 150°C; collision gas pressure , 1 . 3 mTorr; normalized collision energy using wide-band activation mode; 35% for MS2 . Ion selection thresholds were 500 or 1000 counts for MS2 depending on the experiment . An activation q = 0 . 25 and activation time of 30 ms was applied . MS data were processed into peak lists by DTASuperCharge 2 . 0b1 ( part of the MSQuant 2 . 0b7 software environment Mortensen et al . , 2010 ) and searched with Mascot 2 . 2 against the human International Protein Index protein database ( IPI , version 3 . 65 ) combined with frequently observed contaminants and concatenated with the reversed versions of all sequences . The MMD for monoisotopic precursor ions and MS/MS peaks were restricted to 5 ppm and 0 . 8 Da , respectively . Enzyme specificity was set to trypsin ( with a maximum of two missed cleavages ) allowing cleavage N-terminal to proline and C-terminal to aspartate . Modifications were cysteine carbamidomethylation ( fixed ) and protein lysine acetylation , lysine butyrylation , lysine propionylation , lysine and arginine methylation ( all states ) , lysine formylation , serine/threonine/tyrosine phosphorylation , deamidation ( asparagine and glutamine ) and methionine oxidation ( variable ) . Alternatively , we searched the chymotrypsin digested sample with low chymotrypsin specificity ( cleavage carboxyterminal to L , M , W , F , and Y allowing a maximum of four missed cleavages ) . Protein and peptide identifications were further analyzed and manually verified by inspection of chromatograms and spectra . To confirm the identity of the tryptic peptide K ( Ac ) LPFQR ( Figure 1A ) a synthetic peptide ( Biosyntan , Berlin ) harbouring the sequence STELLIRK ( ac ) LPFQRLVGC was trypsin digested and demonstrated the generation of three tryptic peptides , in particular the peptide K ( ac ) LPFQR . The MS/MS fragment spectrum of the latter was virtually identical to the one derived from the endogenous tryptic peptide . Both spectra contain two significant immonium ion peaks ( at m/z 143 and m/z 126 ) that indicate the presence of acetylated lysine at the amino terminus ( Trelle and Jensen , 2008 ) . In addition two chymotryptic peptides with the sequence LIRK ( Ac ) LPF ( Figure 1A ) and IRK ( Ac ) LPF ( Figure 1—figure supplement 1A ) confirm the presence of an acetylated lysine at K64 of endogenous histone H3 . The acetylation of H3K64 was modelled on the 1KX5 Xenopus laevis nucleosome core particle structure ( Davey and Richmond , 2002 ) . Pictures and modifications were generated using PyMOL ( http://www . pymol . org ) . Mouse embryonic stem ( ES ) cells used for chromatin immunoprecipitation ( ChIP ) experiments were derived and cultivated as described previously ( Mohn et al . , 2008 ) . ES cells differentiation was carried out as described previously ( Bibel et al . , 2007 ) . Briefly , embryonic stem cells were cultured feeder-free for three to five passages after which LIF was withdrawn to allow formation of cellular aggregates in liquid culture . After 4 days , retinoic acid ( RA ) was added to induce neuronal progenitor ( NP ) formation for another 4 days before chromatin isolation . All the other mammalian cell lines were cultured in DMEM ( PAA , GE Healthcare Life Sciences , Sweden ) medium supplemented with 10% FCS , 2 mM L-glutamine , 100 U/ml Penicillin and 0 . 1 mg/ml Streptomycin . For the HDAC inhibitors treatment either Na-butyrate ( 10 mM , overnight ) or Trichostatin A ( TSA , 0 . 2 µM , 48 hr ) were used . For characterization of the H3K64ac antibody ( Active motif , Carlsbad , CA , cat no . 39545 , lot 32908001 ) immuno-dot blotting and peptide competitions were carried out as described previously ( Daujat et al . , 2009 ) . Histone H3 acetylated peptides corresponding to amino-acid residues 1–15 ( for K9 ) , 12–24 ( for K18 ) , 1–27 ( for K23 ) , 1–27 ( for K9-K14-K23 ) , and 51–62 ( for K56 ) and histone H3 methylated peptides corresponding to amino-acid residues 57–71 ( for K64me1 , me2 and me3 ) were used in these assays . 1 × 108 mouse NIH3T3 cells were harvested , washed in PBS , resuspended in cold hypotonic buffer ( 5 mM KCl , 1 . 5 mM MgCl2 , 20 mM Hepes pH 7 . 5 , 1x protease inhibitors ( Roche , Switzerland ) , 5 mM Na-butyrate ) at a density of 2 × 107 cells/ml and pelleted . Cells were resuspended in the same buffer at 4 × 107 cells/ml and incubated on ice for 10 min . Cell suspension was dounced 12 times with the S pestle and store on ice for 30 min . After 10 min of centrifugation at 500×g at 4°C , released nuclei were resuspended in 1 ml of isolation buffer-100 ( 0 . 25 M sucrose , 100 mM NaCl 1 . 5 mM MgCl2 , 1 mM CaCl2 , 10 mM Tris–HCl pH 7 . 5 , 1x protease inhibitors , 5 mM Na-butyrate ) and stored 10 min on ice . After 5 min of centrifugation at 4°C nuclei were resuspended in 1 ml of isolation buffer-250 ( same as previously with 250 mM NaCl ) and stored 10 min on ice . Nuclei were centrifuged at 4000×g for 10 min at 4°C and resuspended in 1 ml of isolation buffer-0 ( same as previously with no NaCl ) for DNA concentration determination . The chromatin was then digested in the same buffer for 30 min at 37°C ( 50 units/mg DNA of Micrococcal Nuclease in the presence of 3 mM CaCl2 ) and stopped by the addition of 10 mM EDTA . Nuclei were centrifuged at 10000×g for 10 min and resuspended in 1 ml of lysis buffer ( 650 mM NaCl , 5 mM EDTA pH 8 . 0 , 10 mM Tris–HCl pH 6 . 8 , 1x protease inhibitors , 5 mM Na-butyrate ) and incubated on ice for 30 min . Lysed nuclei were centrifuged at 16500×g for 10 min at 4°C and supernatant containing digested chromatin was loaded on a 5–40% sucrose gradient and centrifuged at 40 , 000 rpm ( SW41 rotor ) for 16 hr at 4°C . 600 µl fractions were then collected and the concentration of chromatin was determined at OD 260 nm . Chromatin of each fraction was either analysed by electrophoresis on a 1% agarose gel in 0 . 5x TBE or by SDS-PAGE to assess DNA size and core histones integrity . Fractions containing mononucleosomes were pooled and used for limited trypsin digestion . 10 µg of DNA corresponding to mononucleosomes were digested by 20 and 30 µg of trypsin for 30 min at 26°C in 250 µl final of PBS . This mild digestion cuts the N-terminal histones tails protruding out from the nucleosome and leaves the H3 core region ( protected by the DNA ) mainly intact . Intact or digested nucleosomes were analysed by SDS-PAGE and Western blotting for the presence of N-terminal H3 acetylations ( anti-H3K9ac; anti-H3K18ac; anti-H3K27ac ) and H3K64 acetylation . H3 digestion pattern was controlled with an anti-H3 antibody . ChIPs of histone modifications on native or crosslinked chromatin were performed with minor modifications of procedures described previously ( Cuthbert et al . , 2004; Daujat et al . , 2009 ) . For native ChIPs nuclei were extracted and isolated over a sucrose cushion , resuspended in MNase buffer ( 0 . 32 M sucrose , 50 mM Tris-HCl pH 7 . 5 , 4 mM MgCl2 , 1 mM CaCl2 ) and digested with 10 U MNase ( Thermo Fisher Scientific Inc . , Waltham , MA ) for at least 15 min at 37°C , so that mononuceosomes were released . This chromatin was incubated overnight with specific antibodies and then precipitated with protein A/G Sepharose 4 fast flow beads ( GE Healthcare Life Sciences ) . After washing three times with increasing amount of salt ( 50 mM Tris-HCl pH 7 . 5 , 10 mM EDTA , containing either 75 mM , or 125 mM , or 175 mM NaCl ) , immunoprecipitated DNA was eluted and purified . For ChIPs on cross-linked chromatin ES cells were lysed and sonicated to produce chromatin fragments of 0 . 5 Kb on average . Diluted chromatin equivalent to 2 × 106 cells was subjected to immunoprecipitation overnight . Sepharose beads were used to recover chromatin . Precipitates were washed 10 min at RT two times in lysis buffer ( 50 mM HEPES/KOH pH 7 . 5 , 500 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 0 . 1% DOC , 0 . 1% SDS ) , once in DOC buffer ( 10 mM Tris pH 8 , 0 . 25 mM LiCl , 0 . 5% NP40 , 0 . 5% DOC , 1 mM EDTA ) and once in TE and DNA was recovered as described previously ( Dedon et al . , 1991; Orlando et al . , 1997; Daujat et al . , 2002 ) . Quantitative real-time PCR were performed using SybrGreen ( Thermo Fisher Scientific Inc . ) on an ABI Prism 7300 apparatus ( Applied Biosystems , France ) . The 5′ to 3′ sequences of the ChIP primers used for different genomic regions in qPCR are listed in Supplementary file 1 . ChIP-on-chip of histone modifications were carried out as described previously ( Mohn et al . , 2008 ) . For ChIP followed by PCR-SSCP to distinguish the parental alleles , primary MEFs were derived from 13 . 5 d . p . c . ( C57BL/6 x JF1 ) F1 foetuses ( genotype: [C57BL/6 x M . m . molossinus]F1 ) . After ChIP , DNA was extracted from the immunoprecipitated ( B , bound ) and unbound ( U ) chromatin . As a control mock precipitation , we used a rabbit IgG antiserum against chicken IgI ( Sigma-Aldrich , St . Louis , MO , cat no . C2288 , lot 21K4851 ) . The maternal ( M ) and paternal ( P ) alleles were distinguished by radio-active PCR amplification across nucleotides that were polymorphic between the paternal JF1 ( M . m . molossinus ) and the maternal C57BL/6J ( B6 ) genome . Primers used for PCR are listed in Supplementary file 1 . Following denaturation of the amplification products , single-strand conformation polymorphisms ( SSCP ) were revealed by electrophoresis through a non-denaturing agarose gel ( Gregory and Feil , 1999; Umlauf et al . , 2004; Pannetier et al . , 2008 ) . ChIP samples were hybridized to custom tiling microarrays ( Lienert et al . , 2011 ) representing all well-annotated promoters and the complete chromosome 19 with an average probe spacing of 100 bp ( NimbleGen Systems Inc . , Madison , WI ) . Sample labelling , hybridization and array scanning were performed according to the manufacturer’s protocols using a MAUI hybridization station and a NimbleGen MS200 slide scanner in combination with NimbleScan software ( NimbleGen Systems Inc . ) . Further processing was done using R . For analysis raw fluorescent intensity values were used to calculate log2 of the bound/input ratios for each individual oligo . Subsequently , for comparison all arrays were normalized to a median log2 = 0 and scaled to have the same median absolute deviation using the ‘LIMMA’ R/Bioconductor package ( Smyth and Speed , 2003; Smyth , 2004 ) . For boxplots , all microarray probes that overlap the indicated genomic regions were grouped and log2 ChIP/Input values for every probe are displayed as one boxplot per class . Active and inactive TSS ( activity cutoff defined according to Lienert et al . , 2011 ) contain probes that are in a 2-kb window ( ± 1 kb ) around the respective transcription start sites , ‘enhancers’ contains all probes overlapping the enhancer regions as defined by Creyghton et al . ( 2010 ) , ‘genic’ represents all probes that map to annotated genes without the first 1000 bp to avoid overlap with the promoters , and ‘intergenic’ contains all remaining probes on chromosome 19 that are not in any of the other groups to avoid overlap between the individual classes . The data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus ( Edgar et al . , 2002 ) and are accessible through GEO Series accession number GSE35355 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE35355 ) . H3K27ac and p300 ChIPseq data were taken from Creyghton et al . ( 2010 ) , H3K4me1 data from Stadler et al . ( 2011 ) , mRNA expression data from Lienert et al . ( 2011 ) . Raw data was downloaded from the NCBI Gene Expression Omnibus and the reads were mapped to the Mus musculus genome ( mm9 ) and transcriptome ( RefSeq , downloaded on 07/17/2009 ) as previously described ( Lienert et al . , 2011 ) . For promoter analysis , mapped reads were counted in a 2-kb window centred around all transcription start sites without further normalization . Enhancers were classified according to the definitions previously reported ( Creyghton et al . , 2010 ) . In brief , active enhancer regions required a significant enrichment for both , H3K4me1 and H3K27ac , while inactive enhancers were defined as regions containing only H3K4me1 ( Creyghton et al . , 2010 ) . Total H3 ChIP-seq data to generate the meta plots around the TSSs and to calculate enrichments were extracted from Mikkelsen et al . , 2007 . Read densities ± 15 kb around TSSs were binned in 1 kb windows and plotted as mildly smoothed lines using R ( lowess , f = 0 . 05 ) . For histone extraction , nuclei were isolated by incubating the cell pellets 10 min in TEB buffer ( 0 . 5% Triton X-100 in PBS ) and histones were acid extracted with 0 . 2 N HCl overnight at 4°C . After extraction , protein concentration was measured according to the Bradford method and equal amounts of protein were separated on SDS-PAGE and transferred onto nitrocellulose membrane . For nuclear extracts preparation , HeLa cell pellets were incubated with TEB buffer as before , but then nuclei were incubated in extraction buffer ( 0 . 3% SDS , 0 . 8 M NaCl , 0 . 3% NP40 ) diluted in water to a final NaCl concentration of 0 . 4 M and sonicated . For H3 variants quantification , histones were acid extracted as described above from HEK293 cell lines stably expressing different FlagHA-H3 variants and immunoprecipitated using anti-Flag beads ( Sigma-Aldrich ) . Bound FlagHA-tagged H3s were eluted by boiling in Leammli buffer . Quantification was done using the ImageJ software and the ratio between specific H3K64ac and HA signals ( loading control ) was calculated . For Nap1 pull-down , His-tagged Nap1 was pre-incubated with different histone species in incubation buffer ( 10 mM Tris–HCl pH 8 . 0 , 100 mM NaCl , 100 ng/µl BSA , 0 . 05% NP-40 , 1 mM EDTA , 1 mM DTT , 5% glycerol ) for 30 min at 30°C 300 RPM . Samples were then supplied with nickel beads ( His-Select Nickel Affinity Gel , Sigma ) and incubated for 4 hr on the wheel at 4°C . Bound material was washed with wash buffer ( 10 mM Tris-HCl pH 8 , 250 mM NaCl , 0 . 1% NP-40 , 1 mM EDTA , 1 mM DTT , 5% glycerol ) and elution was achieved by boiling the beads in Leammli buffer . For HATs overexpression , a tagged p300 HAT domain was cloned in pcDNA3 vector and 5 μg of plasmid were transfected together with a carrier DNA plasmid ( 0 . 5 μg ) expressing mCherry . HEK293 cells were grown on poly-L-lysine-coated coverslips , transfected with the conventional calcium-phosphate method and processed for immunofluorescence ( see below for details ) after 24 hr . For HATs knock-down RNAiMAX ( Invitrogen , Carlsbad , CA ) was used and a reverse transfection protocol was applied following the supplier’s instructions . The siRNAs used were an ON-target plus SMART pool from Dharmacon ( Thermo Fisher Scientific Inc . ) : p300 , L-003486-00-0010; CBP , L-003477-00-0010; GCN5 , L-009722-00; PCAF , L-005055-00; scramble negative control , D-001810-10 . hMof siRNA was purchased from MWG ( Eurofins , Germany ) with sequence 5′-GUGAUCGAGUCUCGAGUGArUrU-3′ . Briefly , a mixture of 50 nM total of siRNA and lipofectamine were prepared in the wells and subsequently MCF7 cells were plated on it and grown for 72 hr . Full-length HATs were expressed and purified from Sf9 cells as described elsewhere ( Kraus and Kadonaga , 1998 ) . Recombinant GST-tagged HAT domains were expressed in E . coli and then purified . For the HAT assay , recombinant histones H3 ( wt and K64A ) and H4 were incubated with the specific HAT in the presence of acetyl-CoA ( without in case of mock reaction ) in HAT buffer ( 50 mM Tris–HCl pH8 . 0 , 7% glycerol , 25 mM NaCl , 0 . 1 mM EDTA , 5 mM DTT ) for 1 hr at 30°C . The reaction was then analysed by western blotting with the specific H3K64ac antibody and the activity of the enzyme was checked using antibodies against known specific targets . Histone modifications primary antibodies: H3K4me3 ( Millipore , Billerica , MA ) , H3K9ac ( Cell Signaling , Boston , MA ) , ChIP grade H3K9ac ( Abcam , UK ) , H3K9ac ( Millipore , Billerica , MA ) , H3K9me3 ( Millipore ) , H3K14ac ( Millipore ) , H3K18ac ( Cell Signaling ) , H3K18ac ( Abcam ) , ChIP grade H3K27ac ( Abcam ) , H3K27me3 ( Millipore ) , H3 ( Abcam ) , H4K16ac ( Santa Cruz Biotechnology Inc . , Dallas , Texas ) . Non-histone primary antibodies: HA ( Abcam ) . Cells were fixed in 4% paraformaldehyde/2% sucrose for 15 min , washed three times in cold PBS and permeabilised with 0 . 5% Triton X-100 in PBS for 20 min . After washing in PBS , cells were blocked in 3% BSA/PBS and then stained with specific primary antibodies overnight at 4°C . Fluorescent secondary antibodies ( FITC or RedX from Jackson Immunoresearch , UK ) were then used for detection . Coverslips were either directly mounted with Vectashield containing DAPI ( Vector laboratories , UK ) or stained with DAPI for 10 min and then mounted with DABCO ( Sigma-Aldrich ) . Confocal microscopy was performed in a Leica TCS SP2/MP inverted confocal microscope using a 63× or 100× oil objective . Images were acquired along the Z-axis every 0 . 5 μm . Tubules of mice were prepared as previously reported ( Kotaja et al . , 2004 ) . Cells were fixed in 90% ethanol for 10 min at RT . They were permeabilised with 0 . 2% Triton X-100 and 0 . 5% saponine in PBS for 15 min at RT and blocked with 5% milk in PBS . They were incubated with anti-H3K64me3 at 4°C overnight , followed by incubation with Alexa488 conjugated secondary antibody for 30 min at 37°C . Finally , cells were counterstained with DAPI and mounted in Vectashield with DAPI ( Vector laboratories ) . Imprints of mice testes were prepared and fixed in 90% ethanol for 10 min at RT . They were permeabilised with 0 . 6% Triton X-100 in PBS containing for 15 min at room temperature and blocked with 4% BSA and 0 . 2% Tween in PBS . They were incubated with anti-H3K64ac at 4°C overnight , followed by incubation with Alexa488 conjugated secondary antibody for 30 min at 37°C . AFA’s fixed , paraffin-embedded testicular sections ( 7 µm ) were deparaffinised using toluene and hydrated through graded series of ethanol ( 100% , 90% , 70% ) . The sections were then incubated with proteinase K ( 20 µg/ml ) in PBS containing 0 . 5% SDS for antigen retrieval , 15 min at 37°C . They were then blocked by incubation with PBS containing 5% milk for 30 min . Immunostaining was performed by three sequential experimental steps: incubation with anti-H3K64ac antibody overnight at 4°C , washed , and then incubated with a biotinylated secondary antibody for 30 min at RT . The slides were washed in PBS with 0 . 5% milk; the final detection was performed using the ABC Elite kit Vectastain and the DAB peroxidase Substrate kit ( Vector Laboratories ) according to the manufacturer’s instructions . The slides were washed and stained with PAS ( Periodic Acid Schiff; Sigma-Aldrich ) and hematoxylin eosin , dehydrated , and mounted in Eukitt ( Sigma-Aldrich ) . The testis tubules sections were observed with a transmission light microscope and staged according to the criteria previously described ( Russel et al . , 1990 ) . BL21 gold cells were transformed with plasmid pAcKRS-3 encoding for the tRNA acetyl-lysine synthetase and pCDF PylT-1 , and containing the amber suppressor tRNA and the H3 cDNA wild type , K64AMBER , or K9AMBER codon . The cells were then cultured as described before ( Neumann et al . , 2009 ) . To purify the recombinant histones , cells were resuspended in wash buffer ( 50 mM Tris-HCl pH7 . 5 , 100 mM NaCl , 1 mM β-mercaptoethanol ) supplemented with 20 mM Nicotinamide ( NAM ) and 10 mg/ml lysozyme and then sonicated . The inclusion bodies were washed with the same buffer supplemented with 1% Triton X-100 and then histones were extracted with unfolding buffer ( 6 M Guanidium HCl , 20 mM Na-acetate pH5 . 2 , 1 mM DTT ) for 1 hr at RT . The extracted proteins were dialysed against SAU200 buffer ( 7 M urea , 20 mM Na-acetate pH5 . 2 , 200 mM NaCl , 5 mM β-mercaptoethanol , 1 mM EDTA ) before loading the sample on a ResourceS ion exchange column ( GE Healthcare Life Sciences ) . Nucleosomes were assembled using DNA fragments derived from the MMTV-A positioning sequence ( Flaus and Richmond , 1998; Flaus and Owen-Hughes , 2003; Ferreira et al . , 2007 ) , and were purified as unmodified or acetylated H3/H4 tetramers and a stoichiometric amount of H2A/H2B dimers . All DNA fragments were made by PCR using fluorescently labelled primers . The PCR products were purified by anion exchange chromatography on a 1 . 8 ml SOURCE 15Q column ( GE Healthcare Life Sciences ) or native ( 0 . 5× TBE ) PAGE . The notation xAy denotes the 147-bp MMTV-A sequence with flanking DNA of x and y bp on the upstream and downstream side , respectively . Typically , assembly was performed by salt-gradient dialysis using a double-dialysis method ( unless specified otherwise ) , as follows—reactions ( 25–35 µl ) were placed in microdialysis buttons , which were placed inside a dialysis bag containing 30 ml 0 . 5× TE and 2 M NaCl , the dialysis bag was then dialysed overnight against 2 L of 0 . 5× TE at 4°C . The 147-bp DNA fragment 0A0 was internally labeled with the fluorescent dye AF488 and the dark quencher BHQ1 . The DNA was made by PCR using primers that contained an internally labeled thymidine 35 bp from the 5′ end of the primer ( indicated as bold and underlined ) . Forward AF488 primer: -ACTTGCAACAGTCCTAACATTCACCTCTTGTGTGTTTGTGTCT Reverse BHQ1 primer: -CAAAAAACTGTGCCGCAGTCGGCCGACCTGAGGGTCGCCGGGG Unmodified and acetylated nucleosomes were assembled on the doubly-labeled DNA and the integrity of reconstituted nucleosomes was checked by running equal amounts of each nucleosome reconstitution ( equivalent to 600 ng DNA ) on 0 . 5× TBE 5% native polyacrylamide gels . To measure qF the nucleosomes were diluted to 40 nM in the final dialysis buffer and then diluted with an equal volume of 2× NaCl stock solutions , yielding final NaCl concentrations from 0 M to 2 . 0 M . Reactions were left for at least 15 min to reach equilibrium , transferred to 96-well half-area black non-binding surface microplates ( Corning , Corning NY ) and AF488 fluorescence measured on a BMG Labtech FLUOstar Optima plate-reader using 488p and 520p excitation and emission filters , respectively . Titrations using DNA alone were performed and confirmed that NaCl-induced changes in qF of nucleosomes was due to nucleosome disruption and not solute-dependent effects on the fluorophore . Upper and lower plateaus for each titration were used to normalise the data on a scale from 1–0 . The plateau values were calculated by fitting the raw data using the sigmoid function described below in Microcal Origin 8 . 6 Software . Average titration curves were calculated from normalised data from three separate titrations ( using different nucleosome reconstitutions ) for unmodified and acetylated nucleosomes and fit again with the sigmoid function to determine the average NaCl concentration at which the normalised qF ( qFnorm ) is 0 . 5 ( C0 . 5 ) :qFnorm=A2+A1−A21+e ( C−C0 . 5 ) /dCA1 and A2 are the lower and upper plateau values , respectively , C the NaCl concentration and dC a constant . The competitive reconstitution procedure is based on that described in Thåström et al . ( 2004 ) and Manohar et al . ( 2009 ) . However , we found that using the dilution method described below gave more consistent results than the double-dialysis method . Reactions were initially set up in 14 . 5 µl 1× TE and 2 M NaCl and contained 5 µg UltraPure sheared salmon sperm DNA ( Invitrogen ) as non-specific competitor , 1 μg Cy3-labelled 0A0 DNA as the specific assembly fragment , 80 pmol H2A/H2B dimer , 40 pmol of unmodified or acetylated H3/H4 tetramer . A mastermix containing all components except the tetramer was assembled , split into separate reaction tubes , and the required amount of concentration-matched unmodified/acetylated tetramer added . This was to ensure the maximum amount of consistency between the reactions . The NaCl concentration was then decreased by addition of 1× TE 0 . 1 mM DTT to give concentrations of 1 . 5 , 1 . 0 , 0 . 75 , 0 . 6 , and 0 . 4 M , with sufficient time given for the reactions to equilibrate at each NaCl concentration ( 15–30 min ) . Once reactions had reached 0 . 4 M NaCl they were split into two separate microdialysis buttons and dialysed against 1× TE 0 . 1 mM DTT for several hours to reduce the NaCl to 0 M . The reactions were split in two to assess variability during dialysis; however , we found they did not vary by more than 3% . The reactions were then mixed with sucrose to a final concentration of 10% ( w/v ) and run on 0 . 2× TBE 6% native polyacrylamide gels run at 200 V for 75 min at 4°C . Reactions using unmodified and acetylated tetramer were always set up side-by-side . The data are the average of three different batches of assembly reactions performed on different days . Band intensities were quantified in AIDA software . The average assembly efficiency of the unmodified nucleosome reactions from each batch was defined as 1 . 0 and used as a reference for normalisation . Nucleosomes were assembled on Cy3- or Cy5-labelled 54A18 DNA for RSC remodelling reactions or 54A0 for Chd1 reactions . Recombinant Chd1 was produced as described in Ryan et al . ( 2011 ) . TAP-tagged RSC was produced as described in Ferreira et al . ( 2007 ) . The remodelling reactions were carried out essentially as described in Ferreira et al . ( 2007 ) ; Somers and Owen-Hughes ( 2009 ) and Ryan et al . ( 2011 ) . Briefly , the reactions contained 50 mM Tris pH 7 . 5 , 50 mM NaCl , 3 mM MgCl2 , and 1 mM ATP ( or 1 mM ATPγS where indicated ) , 50 nM each of H3K64-acetylated nucleosomes and unmodified nucleosomes on Cy3- and Cy5-labelled DNA , respectively , and either 1 nM Chd1 or 0 . 5 nM RSC . The reactions were incubated at 30°C and samples taken at 0 , 5 , 10 , 15 , 20 , 30 , and 45 min . The reactions were stopped by the addition of 500 ng HindIII digested λ DNA and sucrose to 5% w/v and placing on ice . Products were resolved on 0 . 2× TBE 5% native polyacrylamide gels run for 3 hr at 300 V at 4°C . Gels were imaged on an FLA-5100 fluorescence scanner . Band intensities were quantitated in AIDA software . Initial rates were obtained by fitting using the following equation in MicroCal Origin 7 . 0 software and solving the derivative at zero time:y=a ( 1−e−bx ) where , a describes the value of the asymptote of the curve and b the relative rate in terms of normalized fraction of nucleosomes repositioned per minute . The fraction is a normalised unitless value and so the units for the rate are just min−1 . Unmodified or H3K64 acetylated nucleosomes were assembled on the 147-bp DNA fragment 0A0 , which was labeled with fluorescent dyes attached to the same thymidine of the primer sequences shown before ( see also ‘Salt-dependent disassembly of nucleosomes monitored by quenched FRET of AlexaFluor488’ in ‘Materials and methods’ section ) , 6-Tamra ( forward primer ) and alexa647 ( reverse primer ) . PCR products were purified by DNA precipitation followed by size exclusion chromatography using a Superose 6 PC 3 . 2/30 column ( GE Healthcare Life Sciences ) . DNA and histone octamers containing either unmodified H3 or H3K64ac were assembled to mononucleosomes by salt-gradient dialysis according to the double-dialysis method described above , but placing microdialysis buttons into 300 ml 0 . 5× TE-NP40 and 2 M NaCl , and dialysing them overnight against 3 l of 0 . 5× TE-NP40 and 50 mM NaCl at 4°C . Finally , the buttons were placed in 1 L 0 . 5× TE-NP40 buffer for 1 hr . A 1:20 mixture of labelled and unlabelled DNA was used for the assembly to generate a mixture of labelled and unlabelled nucleosomes and thus avoid nucleosome disruption due to low nucleosome concentration during the single-molecule experiments . For smFRET measurements , nucleosomes were diluted in measurement buffer ( 10 mM Tris–HCl pH 7 . 6 , 1 mM EDTA , 0 . 05% NP40 , 0 . 1 mg/ml BSA and varying amounts of NaCl , ranging from 0 to 1 M ) to a concentration in the order of 500 pM , corresponding to a picomolar concentration of the fluorescent species . Samples were then kept at 4°C for at least 18 hr to allow for adaptation to the respective final salt concentration . A 20 µl drop was placed on a PEG-coated coverslip and measurements were performed for 10 min at room temperature . At least three measurements were performed for each salt concentration on a custom built confocal microscope using freely diffusing molecules and time-correlated single photon counting ( TCSPC ) , as described in Bönisch et al . ( 2012 ) . Photon arrival times were recorded by a Hydra Harp 400 ( PicoQuant GmbH , Germany ) . To select bursts of fluorescence , an all-photons burst search ( APBS , Nir et al . , 2006 ) was performed on the recorded data , requiring a minimum of five photons in 500 µs and at least 30 photons in total per burst . For each burst , FRET efficiency and stoichiometry were calculated from the recorded signals as explained in Bönisch et al . ( 2012 ) . Only bursts with more than 50 photons in total were considered for further analysis . The populations of singly-labeled molecules were excluded by a stoichiometry ( Sto ) threshold of 0 . 1<Sto<0 . 7 and rare multi-molecule events were excluded by limiting the time deviation signals ( TDS ) to TDS<1 and TDSred-PIE<0 . 4 as explained in Bönisch et al . ( 2012 ) . Because of the chosen position of the dyes , intact nucleosomes showed an intermediate to high FRET efficiency , while disassembled nucleosomes showed a very low FRET efficiency ( dyes position shown in Figure 5A ) . Thus , two FRET populations , represented by two Gaussian distance distributions , were fitted to the data by probability distribution analysis ( PDA , Antonik et al . , 2006 ) assuming a Förster radius of 60 nm . Additional measurement allowed us to estimate a background signal of 0 . 2 kHz in all channels , 4% crosstalk , and a γ-factor of 0 . 7 , as a correction factor for differences in the quantum yields of the dyes and the wavelength-dependent detection efficiencies of the setup . Measures of the higher FRET efficiency corresponded to the fraction of intact nucleosomes at the respective salt concentration . These values were normalized to the higher FRET efficiency value at 0 mM NaCl to allow comparison between unmodified and acetylated nucleosomes . The sigmoidal fit function described in the section ‘Salt-dependent disassembly of nucleosomes monitored by quenched fluorescence ( qF ) of AlexaFluor488 ( AF488 ) ’ was used to fit the data and retrieve C0 . 5 . The histone eviction assay was adapted from a protocol developed by Ito et al . ( 2000 ) and Sharma and Nyborg ( 2008 ) . Chromatin templates containing unmodified H3 , H3K64ac , or H3K64R were assembled on an ∼680-bp linear DNA fragment labeled with biotin on the 5′ end . Chromatin was assembled using the salt dilution method as described in Gutierrez et al . ( 2007 ) , where 20 µl reactions at 1 M NaCl were diluted with 1× TE , 1 mM DTT to 0 . 8 , 0 . 6 , 0 . 4 , 0 . 2 , and 0 . 1 M ( final dilution buffer 2×: 2× TE , 20% glycerol , 0 . 1% NP40 , 200 ng/µl BSA , 2 mM DTT , 1 mM Na-butyrate , protease inhibitors ) . Each dilution step was carried out at 4°C , 400 RPM , for 45 min . Assembled chromatin was analysed on native 5% polyacrylamide gel run in 0 . 2× TBE for 90 min at 150 V . Chromatin templates were bound on streptavidin beads for 3 hr at 4°C ( 10 mM Tris–HCl pH8 , 100 ng/µl BSA , 100 mM NaCl , 0 . 05% NP-40 , 1 mM EDTA , 1 mM DTT , 0 . 1 mM AEBSF , 0 . 5 mM Na-butyrate , 10% glycerol ) and the immobilised template was incubated ( 10 mM Tris–HCl pH 8 . 0 , 100 ng/µl BSA , 100 mM NaCl , 0 . 05% NP-40 , 1 mM EDTA , 1 mM DTT , 1 mM Na-butyrate , 5% glycerol ) first with Gal4-VP16 for 20 min at 30°C , 450 RPM , and then with full-length p300 and Nap1 ( Drosophila , full length , expressed in Sf9 cells ) for 15 min . Then , a mixture of radioactive 3H-acetyl-CoA and 100 µM cold acetyl-CoA was added to the samples and incubated for 40 min at 30°C . Before this last incubation , each reaction was supplemented with 1 µg of a supercoiled plasmid , on which the evicted octamers are assembled by Nap1 . The supernatant fraction was recovered and assayed in western blot for total H3 amounts . The bound fraction was washed three times and analysed on SDS-PAGE for p300 specific activity onto chromatin . NIH3T3 cell lines stably expressing different FlagHA-H3 . 3 species were seeded in selection medium containing 10% FBS and then starved for 2 days in starvation medium without FBS . Gene induction was stimulated supplementing the cells with fresh starvation medium supplemented with 25 nM TPA ( Cell Signaling ) . After the indicated times of induction , total RNA was isolated using Quick RNA Miniprep kit ( Zymo Research , Irvine , CA ) . Equal amounts of RNA were employed for first strand cDNA synthesis using RevertAid H Minus First Strand cDNA Synthesis Kit ( Thermo Fisher Scientific ) . cDNA samples were then used as templates to perform quantitative real-time PCR using SybrGreen ( Thermo Fisher Scientific ) on LightCycler 480 II ( Roche ) . Fold change values in gene expression were calculated relative to the untreated samples for each cell line using the—dCt method . Primers used for qPCR are listed in Supplementary file 1 . | DNA is a very long molecule , so it needs to be packaged carefully to fit into the nucleus of a cell . To achieve this , the DNA is wrapped around proteins called histones to form a structure termed a nucleosome , which is the building block of a more compacted substance called chromatin . However , to express the genes in the DNA it is necessary to open up parts of the chromatin to give various enzymes access to the DNA . Cells often chemically modify histones by adding acetyl or methyl groups , and these modifications are known to influence what proteins can bind to the nucleosomes , which ultimately influences what genes are expressed in the cell at a given time . It has been suspected for some time that histone modifications can also influence gene expression more directly , but there has been little evidence for this idea . Now Di Cerbo et al . have studied what happens when acetyl or methyl groups are added to a specific site within a histone called H3K64 , which is close to where the DNA wraps around this histone . These experiments showed that this site tends to be acetylated when a nearby gene is active , and to be unmodified or methylated when this gene is not active . It appears that the addition of the acetyl group makes this region of the chromatin less stable: this , in turn , makes it easier for the chromatin to be unpacked , thus giving access to the enzymes that transcribe the DNA and allowing transcription to take place . The work of Di Cerbo et al . shows that methylation and acetylation at the same site within a histone can define two opposing states of chromatin and DNA: an active state and a repressive state . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2014 | Acetylation of histone H3 at lysine 64 regulates nucleosome dynamics and facilitates transcription |
Combinatorial patterns of histone modifications regulate developmental and cell type-specific gene expression and underpin animal complexity , but it is unclear when this regulatory system evolved . By analysing histone modifications in a morphologically-simple , early branching animal , the sponge Amphimedonqueenslandica , we show that the regulatory landscape used by complex bilaterians was already in place at the dawn of animal multicellularity . This includes distal enhancers , repressive chromatin and transcriptional units marked by H3K4me3 that vary with levels of developmental regulation . Strikingly , Amphimedon enhancers are enriched in metazoan-specific microsyntenic units , suggesting that their genomic location is extremely ancient and likely to place constraints on the evolution of surrounding genes . These results suggest that the regulatory foundation for spatiotemporal gene expression evolved prior to the divergence of sponges and eumetazoans , and was necessary for the evolution of animal multicellularity .
Animals rely on genomic regulatory systems to direct the dynamic spatiotemporal and cell type-specific gene expression that is essential for the development and maintenance of a multicellular lifestyle . However , how such a system originated and evolved in animals remains unclear . As the last common ancestor of modern animals already possessed an extensive repertoire of regulatory genes , including most transcription factors and signaling pathways used in bilaterian development ( Srivastava et al . , 2010; Larroux et al . , 2008; Degnan et al . , 2009; Larroux et al . , 2006; Richards and Degnan , 2009; Ryan et al . , 2013; Moroz et al . , 2014; King et al . , 2008; Sebé-Pedrós et al . , 2011; de Mendoza et al . , 2013; King et al . , 2003; Richter and King , 2013 ) , the evolution of animal multicellularity likely required more than the origin of novel genes . Other regulatory features , such as cis-regulatory DNA and combinatorial patterns of histone covalent post-translational modifications ( PTMs ) ( Davidson and Peter , 2015 ) , would have been instrumental to direct differential gene expression in the first multicellular animals . For instance , recent analysis of the genome of Capsaspora , one of the closest unicellular relatives of animals , reveals a lack of chromatin repressive marks , developmental promoter types and distal cis-regulatory elements ( enhancers ) typically present in complex animals ( i . e . , eumetazoans ) ( Sebé-Pedrós et al . , 2016 ) . The development of high-throughput chromatin assays like chromatin immunoprecipitation coupled with massively parallel sequencing ( ChIP-seq ) ( Robertson et al . , 2007 ) has allowed the dissection of chromatin-encoded information beyond the primary DNA sequence , especially the systematic examination of histone PTMs and their role ( s ) in transcriptional regulation ( Zhou et al . , 2011; Thurman et al . , 2012; Kundaje et al . , 2015; ENCODE Project Consortium , 2012 ) . Although combinatorial patterns of histone acetylation and methylation are key components of gene regulatory mechanisms underpinning the formation and maintenance of eumetazoans ( Schwaiger et al . , 2014 ) , it remains unknown if this system is restricted to these animals or is indeed more ancient . Porifera ( sponges ) are considered one of the oldest surviving phyletic lineages of animals , diverging from other metazoans around 700 Mya ( Erwin et al . , 2011 ) . Despite being one of the morphologically simplest animals , lacking a gut , nerves and muscles , sponges possess an extensive gene repertoire for transcriptional regulation required in eumetazoan development and body patterning ( Srivastava et al . , 2010; Larroux et al . , 2008 , 2006; Adamska et al . , 2007; Gaiti et al . , 2015; Nakanishi et al . , 2014; Conaco et al . , 2012; Riesgo et al . , 2014; Grimson et al . , 2008; Richards et al . , 2008; Leininger et al . , 2014; Fortunato et al . , 2015 , 2014; Bråte et al . , 18212015 ) . Here , following on from our recent transcriptomic studies that revealed that the sponge Amphimedon queenslandica ( herein Amphimedon ) has dynamic developmental gene expression akin to eumetazoans ( Gaiti et al . , 2015; Fernandez-Valverde et al . , 2015; Levin et al . , 2016 ) , we set out to determine whether this transcriptional complexity is paralleled by regulatory complexity encoded by combinatorial histone PTM patterns . By analysing an extensive ChIP-seq compendium of histone H3 PTMs in this sponge , we show that a complex gene regulatory landscape comprised of combinatorial histone modifications was already in place at the dawn of animals . Moreover , we provide evidence for the evolution and expansion of distal cis-regulatory genomic capabilities at the origin of the animal kingdom .
We carried out chromatin immunoprecipitation ( ChIP ) on sexually reproducing Amphimedon adults and larvae using antibodies against specific histone H3 PTMs that have been used to define chromatin states in model bilaterians ( Zhou et al . , 2011; Ho et al . , 2014 ) ( Figure 1A ) . These analyses were undertaken on separate admixtures of adult and larval somatic cell types and , thus , a diversity of gene transcriptional states . Importantly , Amphimedon adults and larvae are comprised of different cell types with markedly different transcriptional profiles and regulatory states ( Gaiti et al . , 2015; Conaco et al . , 2012; Fernandez-Valverde et al . , 2015; Degnan et al . , 2015 ) . While our sampling strategy increases the biological complexity of chromatin states in toto , it may dilute cell type-specific signals . This contrasts with ChIP-seq analyses performed on cell lines , embryos with few cell types , or distinct tissue samples , which encapsulate more homogenous cellular populations and environments ( Sebé-Pedrós et al . , 2016; Kundaje et al . , 2015; Schwaiger et al . , 2014; Gerstein et al . , 2010; Pérez-Lluch et al . , 2015a ) . Given the current Amphimedon genome is a draft sequence , our analyses may also be incomplete in regions that have incomplete annotations and gaps in the assembly ( 13% of the total genome assembly ) ( Srivastava et al . , 2010 ) . 10 . 7554/eLife . 22194 . 003Figure 1 . Chromatin states in Amphimedon . ( A ) Schematic representation of Amphimedon life cycle . Larvae ( oval shaped , 300–500 µm long ) emerge from maternal brood chambers and then swim in the water column before they develop competence to settle and initiate metamorphosis into a juvenile . The juvenile body plan , which displays the hallmarks of the adult body plan , including an aquiferous system with canals , choanocytes chambers and oscula , is the outcome of the dramatic reorganization of the radially-symmetrical , bi- or trilayered larva . This juvenile will then grow and mature into a benthic adult ( ranging from 10–30 cm3 ) ( Degnan et al . , 2015; Edgar et al . , 2002 ) . ( B ) Definition and enrichments for a 9-state Hidden Markov Model based on five histone PTMs ( H3K4me3 , H3K27ac , H3K4me1 , H3K36me3 and H3K27me3 ) in adult Amphimedon . From left to right: chromatin state definitions , abbreviations , histone PTM probabilities , genomic coverage , protein-coding gene functional annotation enrichments , expressed ( Expr . ) and repressed ( Repr . ) protein-coding gene enrichments . Blue shading indicates intensity , scaled by column . ( C ) Adult chromatin state annotations on gene rich highly transcribed ( active ) scaffold ( contig13500 ) showing the predominance of ‘TssA’ , ‘TxFlnk’ , and ‘TxEnhA’ states . For the definition of chromatin states see panel ( A ) . Coding genes ( purple ) and long non-coding RNAs ( blue ) are shown , along with signal coverage tracks showing CEL-seq expression in adult . A grey scale indicates CEL-seq expression level: white ( no-expression ) ; black ( highest expression ) . ( D ) Adult chromatin state annotations on a predominantly silenced scaffold ( contig13522 from 500 , 000 to 1 , 500 , 000 bp ) showing the prevalence of ‘ReprPC’ and ‘ReprPCWk’ states . For the definition of chromatin states see panel ( A ) . Coding genes ( purple ) and long non-coding RNAs ( blue ) are shown , along with signal coverage tracks showing CEL-seq expression in adult . A grey scale indicates CEL-seq expression level: white ( no-expression ) ; black ( highest expression ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 00310 . 7554/eLife . 22194 . 004Figure 1—source data 1 . Histone H3 covalent post-translation modifications and RNA Polymerase II investigated in this study and their typical genomic localization relative to coding genes and regulatory regions in bilaterian model organisms . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 00410 . 7554/eLife . 22194 . 005Figure 1—source data 2 . Histone H3 sequences used to generate Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 00510 . 7554/eLife . 22194 . 006Figure 1—source data 3 . BLASTp search outcome of the relevant histone methyltransferases and acetyltransferases against Amphimedon queenslandica proteins ( NCBI nr database; E-value <1e-09 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 00610 . 7554/eLife . 22194 . 007Figure 1—source data 4 . Summary statistics and quality metrics of the ChIP-seq datasets used in this study . See also Materials and methods for preprocessing of ChIP-seq datasets procedure . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 00710 . 7554/eLife . 22194 . 008Figure 1—source data 5 . Validation of the ChIP-seq results by ChIP-quantitative PCRs ( ChIP-qPCRs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 00810 . 7554/eLife . 22194 . 009Figure 1—figure supplement 1 . Multiple sequence alignment of various eukaryotic histone H3 proteins ( 1–136 amino acids ) , produced by using ClustalO ( RRID:SCR_001591 ) ( Sievers et al . , 2011 ) . Note that the entire amino acid sequence of histone H3 is highly conserved across eukaryotes . Sponge sequence is highlighted . The amino acid sequences used to generate the alignment are also provided in Figure 1—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 00910 . 7554/eLife . 22194 . 010Figure 1—figure supplement 2 . Assessment of reproducibility for biological replicates between histone modifications andRNA Polymerase II . ( A ) Pearson correlation coefficients between histone modifications and RNA Polymerase II ( RNAPII ) . Adult experiments ( combined biological replicates ) are shown . Underlying colors indicate the similarity between the different datasets . Note that H3K36me3 was flagged for low signal to noise , potentially explaining the somewhat high correlation with H3K27me3 ( see Figure 1—source data 4 ) . However , this does not affect the conclusions of the paper in any way . ( B ) Adult chromatin state annotations on a predominantly silenced region . For the definition of chromatin states see Figure 1A . Coding genes ( purple ) are shown , along with input DNA-normalized coverage of each biological replicate ( R1 and R2 ) of different histone modifications and RNA-seq expression . ( C ) Same as ( B ) for highly transcribed regions . Apart from RNAPII replicate 1 , which did not pass the quality threshold required so it has been excluded from all further analyses ( see Figure 1—source data 4 ) , we obtained highly reproducible data sets . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 01010 . 7554/eLife . 22194 . 011Figure 1—figure supplement 3 . Neighborhood positional enrichment plots of adult chromatin states around transcription start site ( TSS ) and transcription end site ( TES ) of proteins-coding genes , produced by ChromHMM ( Ernst and Kellis , 2012 ) . For the definition of adult chromatin states see Figure 1A . ( A ) Positional enrichments in 100 bp genomic bins around the TSS and TES ( ±1 kb ) of expressed protein-coding genes in adult Amphimedon . ( B ) Same as ( A ) for repressed protein-coding genes in adult Amphimedon . Blue shading indicates intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 01110 . 7554/eLife . 22194 . 012Figure 1—figure supplement 4 . Chromatin states in Amphimedonlarva . ( A ) Definition and enrichments for a 9-state Hidden Markov Model based on four histone PTMs ( H3K4me3 , H3K27ac , H3K4me1 , H3K27me3 ) in Amphimedon larva . From left to right: chromatin state definitions , abbreviations , histone PTM probabilities , genomic coverage , protein-coding gene functional annotation enrichments , expressed ( Expr . ) and repressed ( Repr . ) protein-coding gene enrichments . Blue shading indicates intensity , scaled by column . ( B ) Chromatin state annotations on a gene rich highly transcribed ( active ) scaffold ( contig13500 ) as in Figure 1 . For the definition of chromatin states see panel ( A ) . Coding genes ( purple ) and long non-coding RNAs ( blue ) are shown , along with signal coverage tracks showing CEL-seq expression in larva . A grey scale indicates CEL-seq expression level: white ( no-expression ) ; black ( highest expression ) . ( C ) Chromatin state annotations on a predominantly silenced scaffold ( contig13522 from 500 , 000 to 1 , 500 , 000 bp ) as in Figure 1 . For the definition of chromatin states see panel ( A ) . Coding genes ( purple ) and long non- coding RNAs ( blue ) are shown , along with signal coverage tracks showing CEL-seq expression in adult . A grey scale indicates CEL-seq expression level: white ( no-expression ) ; black ( highest expression ) . ( D ) Neighborhood positional enrichments in 100 bp genomic bins around the TSS and TES ( ±1 kb ) of expressed protein-coding genes in larva . ( E ) Same as ( D ) for repressed protein-coding genes in larva . Blue shading indicates intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 012 The antibodies used target the following histone H3 PTMs: ( i ) monomethylated lysine 4 ( H3K4me1 ) , associated with distal cis-regulatory elements such as enhancers; ( ii ) trimethylated lysine 4 ( H3K4me3 ) , enriched in active promoters; ( iii ) trimethylated lysine 36 ( H3K36me3 ) , found with actively transcribed regions; ( iv ) trimethylated lysine 27 ( H3K27me3 ) , enriched in Polycomb-silenced regions; and ( v ) acetylated lysine 27 ( H3K27ac ) , which occurs around activated regulatory regions . We also used an antibody against total histone H3 ( Figure 1—source data 1 ) . An antibody against unphosphorylated Ser2 residues of RNA polymerase II ( RNAPII 8WG16 ) C-terminal domain also was included ( Brookes and Pombo , 2009 ) ( Figure 1—source data 1 ) . As the entire amino acid sequence of histone H3 is perfectly conserved in Amphimedon , along with the relevant histone methyltransferases and acetyltransferases , these antibodies are predicted to recognize the correct epitopes ( Figure 1—figure supplement 1; Figure 1—source data 2; Figure 1—source data 3 ) . These antibodies recognize the correct epitopes in even more distantly related organisms ( i . e . , non-metazoan eukaryotes ) ( e . g . , [Sebé-Pedrós et al . , 2016; Ercan et al . , 2009; Barraza et al . , 2015; Harmeyer et al . , 2015; Liu et al . , 2007; Eckalbar et al . , 2016] ) . ChIP-seq reads generated from immunoprecipitated and input ( whole-cell extract ) DNA were aligned to the Amphimedon genome ( Srivastava et al . , 2010 ) , resulting in highly reproducible data sets ( Figure 1—figure supplement 2; Figure 1—source data 4; Figure 1—source data 5 ) . Uniquely mapped reads were subsequently used to identify a set of distinct chromatin states based on the five different histone H3 PTMs we assayed . Specifically , chromatin states were predicted throughout the genome training a multivariate Hidden Markov Model with different a priori defined states ( from 5 to 15 ) ( Materials and methods ) . We elected to use a 9-state model for all further analyses as it covered all major gene coding and regulatory components ( promoter , enhancer , gene body ) that we expected to resolve with this selection of histone H3 PTMs . Despite the inherent cellular heterogeneity of our starting material , we were able to resolve specificities towards gene components between these nine chromatin states . They fell into two broad categories: one that correlated with actively transcribed genes that include active promoters ( ‘TssA’ ) and enhancers ( ‘TxEnhA’ , ‘EnhWk’ ) , and 5’ and 3’ boundaries of transcribed genes ( ‘TxFlnk’ ) ; and another category with genes with no or little detectable transcription; these include bivalent or poised regulatory ( ‘BivTx’ , ‘EnhP’ ) , repressed Polycomb ( ‘ReprPC’ , ‘ReprPCWk’ ) , and quiescent ( ‘Quies’ ) states ( Figure 1B–D ) . The nine chromatin states differentially associated with specific Amphimedon genomic features . For instance , the ‘TssA’ state ( defined by the presence of H3K4me3 ) was enriched around transcription start sites ( TSSs ) of active genes . ‘TxEnhA’ state ( defined by H3K4me1 , H3K27ac , and H3K36me3 enrichment ) associated with coding exons and introns that correspond to potential cis-regulatory elements and short intergenic regions , which are common in the Amphimedon genome ( Kundaje et al . , 2015; Fernandez-Valverde et al . , 2015; Kowalczyk et al . , 2012; Ritter et al . , 2012; Singer et al . , 2015; Birnbaum et al . , 2012; Zentner and Scacheri , 2012; Zentner et al . , 2011; Fernandez-Valverde and Degnan , 2016 ) . In contrast , the ‘ReprPC’ states ( defined by H3K27me3 enrichment ) were spread through the gene bodies of repressed genes , consistent with the known role of H3K27me3 in transcriptional silencing ( Zhou et al . , 2011; Ho et al . , 2014 ) ( Figure 1BD; Figure 1—figure supplement 2; Figure 1—figure supplement 3 ) . Despite being comprised of different cell types and having a distinct gene expression profile from the adult , the larval genome possesses a remarkably similar set of chromatin states ( Figure 1—figure supplement 4 ) . Obtaining consistent chromatin states based on histone PTMs ChIP-seq data from two markedly different stages of the Amphimedon life cycle provides corroborating evidence that this sponge possesses the same regulatory states as present in eumetazoans . To investigate the distribution of histone H3 PTMs in Amphimedon genes , we calculated the average enrichment of histone H3 PTMs and RNAPII relative to the TSSs of protein-coding genes . Input-normalized ChIP-seq read coverage revealed a strong unimodal H3K4me3 peak positioned immediately after the TSS of expressed genes that co-localizes with H3K27ac and RNAPII ( Figure 2A; Figure 2—figure supplement 1; Figure 2—figure supplement 2A ) . Additionally , H3K4me3 marked ( i ) genes with head-to-head orientation that may be under the control of a bidirectional promoter ( a common feature in the Amphimedon genome [Fernandez-Valverde and Degnan , 2016] ) , and ( ii ) alternative TSSs ( Figure 2—figure supplement 3 ) . This is consistent with H3K4me3 being promoter-proximal and positioned on the +1 nucleosome ( Zhou et al . , 2011; Ho et al . , 2014; Lenhard et al . , 2012 ) . A prominent nucleosome-depleted region was observed right upstream of the TSS of expressed genes ( likely corresponding to the proximal promoter ) followed by a narrowly localized nucleosome ( the +1 nucleosome ) ( see below Figure 2—figure supplement 4D ) , suggesting that the interplay between nucleosome positioning and transcription is conserved in sponge promoters ( Sebé-Pedrós et al . , 2016; Schwaiger et al . , 2014; Roy et al . , 2010; Bai and Morozov , 2010; Jiang and Pugh , 2009 ) . Overall , the distribution of histone H3 PTMs in Amphimedon correlated with the expression state of its genes , as in eumetazoans ( Schwaiger et al . , 2014; Roy et al . , 2010 ) ( Fisher’s exact test , FDR adjusted p-value<0 . 05 ) ( Figure 2B and C; Figure 2—figure supplement 2B–D ) . 10 . 7554/eLife . 22194 . 013Figure 2 . Histone PTMs are correlated with gene expression variations during development . ( A ) TSS-centred average input DNA normalised read coverage plot of H3K4me3 across Amphimedon protein-coding genes . The x-axis spans ± 3 kb around TSSs and represents the position within the gene relative to TSS . The y-axis represents the input DNA normalised enrichment for H3K4me3 ChIP-seq reads in adult Amphimedon . Pink line: Non-expressed genes . Blue line: Low expressed genes . Orange line: Medium expressed genes . Light blue line: High expressed genes . The shaded gray area represents the average size of Amphimedon coding sequences . ( B ) Example of coding genes marked by H3K4me3 peaks . The genomic window shows input DNA-normalized H3K4me3 coverage and RNA-seq expression in both larva and adult . ( C ) The association of regions of enrichment of five histone H3 PTMs ( H3K4me3 , H3K27ac , H3K4me1 , H3K36me3 and H3K27me3 ) and RNAPII with lists of various gene expression groups in adult is shown . The color key represents the log2 ( odds ratio ) and the significant adjusted P-values ( Fisher’s exact test ) are superimposed on the grids . A P-value of zero means the overlap is highly significant . N . S . : not significant . Odds ratio represents the strength of association . ( D ) TSS-centred average input DNA normalised read coverage plots of H3K4me3 and RNAPII across ‘high-variance’ and ‘low-variance’ protein-coding genes . The x-axis spans ± 3 kb around TSSs and represents the position within the gene relative to TSS . The y-axis represents the input DNA normalised enrichment for ChIP-seq reads in adult Amphimedon . Light blue: high-variance coding genes . Orange line: low-variance coding genes . The shaded gray area represents the average size of Amphimedon coding sequences . ( E ) Top five most significantly enriched Gene Ontology ( GO ) terms for high-variance and low-variance protein-coding genes ( adjusted P-values in brackets , Hypergeometric test ) . The full GO table is shown in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 01310 . 7554/eLife . 22194 . 014Figure 2—source data 1 . GO biological process term enrichment outcome for the high-variance and low-variance gene sets ( Hypergeometric test , FDR<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 01410 . 7554/eLife . 22194 . 015Figure 2—source data 2 . KEGG pathways significantly enriched in low-variance and high-variance genes . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 01510 . 7554/eLife . 22194 . 016Figure 2—figure supplement 1 . TSS-centred average input DNA normalised read coverage plots and heatmaps of RNAPII , H3K27ac , H3K36me3 , H3K4me1 and H3K27me3 across Amphimedon protein-coding genes . The x-axis spans ± 3 kb around TSSs and represents the position within the gene relative to TSS . The y-axis represents the input DNA normalised enrichment for ChIP-seq reads in adult . The shaded gray area represents the average size of Amphimedon coding sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 01610 . 7554/eLife . 22194 . 017Figure 2—figure supplement 2 . Histone PTMs and gene expression variations during development . ( A ) TSS-centred average input DNA normalised read coverage plot of H3K4me3 across Amphimedon protein-coding genes . The x-axis spans ± 3 kb around TSSs and represents the position within the gene relative to TSS . The y-axis represents the input DNA normalised enrichment for H3K4me3 ChIP-seq reads in larva . Blue line: Non-expressed genes . Orange line: second 500 expressed genes . Light blue line: first 500 expressed genes . The shaded gray area represents the average size of Amphimedon coding sequences . ( B ) The association of regions of enrichment of four histone H3 PTMs ( H3K4me3 , H3K27ac , H3K4me1 , and H3K27me3 ) and RNAPII with lists of various gene expression groups in larva is shown . The color key represents the log2 ( odds ratio ) and the significant adjusted P-values ( Fisher’s exact test ) are superimposed on the grids . A P-value of zero means the overlap is highly significant . N . S . : not significant . ( C ) Coding gene with larva-specific expression marked by H3K4me3 . The genomic window shows input-DNA normalized H3K4me3 coverage and RNA-seq expression in both larva and adult . ( D ) Same as ( C ) for coding genes with adult-specific expression . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 01710 . 7554/eLife . 22194 . 018Figure 2—figure supplement 3 . H3K4me3 enrichment at genes with head-to-head orientation and alternative TSSs . ( A ) Example of coding genes with alternative TSSs ( Aqu2 . 1 . 39785_001 and Aqu2 . 1 . 396786_001 ) marked by successive H3K4me3 peaks . The genomic window shows input DNA-normalized H3K4me3 coverage and RNA-seq expression in both larva and adult . ( B ) Same as ( A ) for closely located head-to-head genes ( Aqu2 . 1 . 30305_001 and Aqu2 . 1 . 30306_001 ) . Coding genes ( purple ) and coding gene isoforms ( light blue ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 01810 . 7554/eLife . 22194 . 019Figure 2—figure supplement 4 . ChIP-seq profiles of H3K4me3 and total histone H3 across high-variance and low-variance genes . ( A ) Developmental expression profile , from early cleavage to adult , of the highly expressed high-variance genes ( n = 1066 ) . Expression levels were measured by CEL-seq and rescaled by row . Red indicates high expression level , light blue low expression . PS , post-settlement postlarva . ( B ) TSS-centred average input DNA normalised read coverage plot of H3K4me3 across high-variance protein-coding genes . The x-axis spans ± 3 kb around TSSs and represents the position within the gene relative to TSS . The y-axis represents the input DNA normalised enrichment for H3K4me3 ChIP-seq reads in adult . Purple line: Low expressed genes . Orange line: Medium expressed genes . Light blue line: High expressed genes . ( C ) Same as ( B ) but for the low-variance genes . ( D ) TSS-centred average input DNA normalised read coverage plot of total histone H3 across high-variance and low-variance protein-coding genes . The x-axis spans ± 3 kb around TSSs and represents the position within the gene relative to TSS . The y-axis represents the input DNA normalised enrichment for ChIP-seq reads in adult . Light blue: High-variance coding genes . Orange line: Low-variance coding genes . The shaded gray area represents the average size of Amphimedon coding sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 019 To investigate the dynamics of histone PTMs in genes regulated throughout Amphimedon development , we analysed CEL-seq data ( Levin et al . , 2016; Hashimshony et al . , 2012; Anavy et al . , 2014 ) , comprising of 82 Amphimedon developmental samples from early cleavage to adult compressed into 17 stages , in the context of ChIP-seq profiles of total histone H3 , H3K4me3 , and RNAPII . We selected genes with the highest median absolute deviation for gene expression across these 17 Amphimedon developmental stages ( effectively measuring the amplitude of change in expression levels for a given gene ) , resulting in a set of 3 , 200 ‘high-variance’ expressed genes ( Figure 2—figure supplement 4A ) . The remaining expressed genes were defined as ‘low-variance’ genes ( 3 , 999 ) ( see Materials and methods for the complete list of selection criteria ) . It is noteworthy that the high-variance genes were , on average , also expressed at higher levels than the low-variance genes ( average adult expression of 51 vs 7 CEL-seq normalized counts , respectively ) . The TSSs of high-variance genes were strongly marked by H3K4me3 and occupied by RNAPII ( Figure 2D; Figure 2—figure supplement 4B ) . Additionally , they showed nucleosome depletion right upstream of the TSSs ( seen as lack of total histone H3 signal ) , consistent with the notion that H3K4me3 near TSSs destabilizes the interaction between histones and DNA to direct RNAPII to facilitate binding of promoter regulator elements and initiate transcription ( Jiang and Pugh , 2009; Ha et al . , 2011; Boeger et al . , 2003 ) ( Figure 2—figure supplement 4D ) . Conversely , lower levels of H3K4me3 or RNAPII ( Mann-Whitney U test , p-value=0 . 05287 and p-value<2 . 2e-16 , respectively; Figure 2D; Figure 2—figure supplement 4C ) but higher nucleosome occupancy characterized low-variance genes ( seen as lack of nucleosome depletion right upstream of the TSSs; Figure 2—figure supplement 4D ) . These results are consistent with H3K4me3 being predictive of gene expression levels ( Ha et al . , 2011; Karlić et al . , 2010 ) . The distinctive landscapes of histone PTMs in high-variance and low-variance genes also correlated with distinct functional related gene groups , as indicated by Gene Ontology ( GO ) and KEGG pathway analyses . High-variance genes , which also include a significantly higher number of transcription factor gene families ( e . g . , JUN and ATF6 Jindrich and Degnan [2016] ) compared to low-variance genes ( Fisher’s exact test , p-value=3 . 872e-08 ) , were predominantly enriched in signaling pathways ( Hypergeometric test , FDR adjusted p-value<0 . 01; Figure 2E; see Figure 2—source data 1 and Figure 2—source data 2 for the complete list ) . In contrast , low-variance genes were enriched for metabolic GO terms ( Figure 2E; see Figure 2—source data 1 and Figure 2—source data 2 for the complete list ) . This result is consistent with H3K4me3 being important for tuning the gene expression of dynamically expressed developmental genes , e . g . , transcription factor and signaling genes . However , it remains unclear whether H3K4me3 is needed for high levels of gene expression or if it is needed for , or associated with , frequent switching of transcriptional status . The recent finding that transcription of a subpopulation of extremely dynamically expressed genes – typically being expressed at only one stage of development – in Drosophila and C . elegans occurs in the absence of H3K4me3 challenged the canonical role of histone PTMs in the modulation of gene expression ( Pérez-Lluch et al . , 2015a ) . To test whether this newly-discovered feature is conserved in non-bilaterians , we interrogated above-mentioned CEL-seq data ( Levin et al . , 2016; Hashimshony et al . , 2012; Anavy et al . , 2014 ) , comprising 82 Amphimedon developmental samples from early cleavage to adult compressed into 17 stages , and arbitrarily selected , similarly to Pérez-Lluch et al . ( 2015a ) , the 1 , 000 genes with the lowest coefficients of variation ( ‘stable’ genes ) expressed with minor changes throughout development . Conversely , the 1 , 000 genes with the highest coefficients of variation were defined as ‘regulated’ genes . Notably , the ‘regulated’ genes consisted of a small population of genes that differed from the ‘high-variance’ genes described earlier in having much more restricted expression patterns , mainly expressed at late juvenile and/or adult stage ( Figure 3—figure supplement 1 ) . Although stable and regulated genes had similar levels of RNAPII and total histone H3 ( Figure 3—figure supplement 1B and C ) , the stable genes were strongly marked by H3K4me3 and the regulated genes had significantly lower levels of H3K4me3 ( Mann-Whitney U test , p-value=7 . 431e-05; Figure 3A ) , suggesting that reduction in H3K4me3 levels does not affect expression of the regulated genes ( Pérez-Lluch et al . , 2015a ) . 10 . 7554/eLife . 22194 . 020Figure 3 . Expression without H3K4me3 in strongly developmentally regulated genes . ( A ) TSS-centred average input DNA normalised read coverage plot of H3K4me3 across ‘regulated’ and ‘stable’ protein-coding genes during Amphimedon development . The x-axis spans ± 3 kb around TSSs and represents the position within the gene relative to TSS . The y-axis represents the input DNA normalised enrichment for H3K4me3 ChIP-seq reads in adult Amphimedon . Light blue line: first 500 regulated genes . Orange line: second 500 regulated genes . Purple line: first 500 stable genes . Pink line: second 500 stable genes . The shaded gray area represents the average size of Amphimedon coding sequences . ( B ) Input DNA-normalized H3K4me3 coverage and RNA-seq expression in adult for Aqu2 . 1 . 40735_001 , a gene stably expressed during Amphimedon development , Aqu2 . 1 . 39666_001 , a regulated gene with adult-specific expression , and Aqu2 . 1 . 34366_001 , a regulated gene with larva-specific expression . ( C ) TSS-centred average input DNA normalised read coverage plot of H3K4me3 across ‘regulated’ and ‘stable’ protein-coding genes during Nematostella vectensis development . The x-axis spans ± 3 kb around TSSs and represents the position within the gene relative to TSS . The y-axis represents the input DNA normalised enrichment for H3K4me3 ChIP-seq reads in Nematostella adult female polyps . The shaded gray area represents the average size of Nematostella coding sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 02010 . 7554/eLife . 22194 . 021Figure 3—figure supplement 1 . ChIP-seq profiles of RNAPII , total histone H3 , H3K36me3 and H3K27me3 across regulated and stable genes . ( A ) Developmental expression profile , from early cleavage to adult , of the regulated and stable protein-coding genes ( see main text and Materials and methods for details ) . Expression levels were measured by CEL-seq and rescaled by row . Red indicates high expression level , light blue low expression . Note that the ‘regulated’ genes show much more restricted expression patterns , being typically expressed at only one or two stage ( s ) of development ( oscula and/or adult ) , than both ‘stable’ and ‘high-variance’ genes ( see Figure 2—figure supplement 4A for a comparison ) . PS , post-settlement postlarva . ( B ) TSS-centred average input DNA normalised read coverage plot of RNAPII , ( C ) total H3 , ( D ) H3K36me3 and ( E ) H3K27me3 across ‘regulated’ and ‘stable’ protein-coding genes during Amphimedon development . The x-axis spans ± 3 kb around TSSs and represents the position within the gene relative to TSS . The y-axis represents the input DNA normalised enrichment for ChIP-seq reads in adult . Light blue line: first 1 , 000 regulated genes . Orange line: first 1 , 000 stable genes . The shaded gray area represents the average size of Amphimedon coding sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 021 We compared the pattern of H3K4me3 between one of the top three stably expressed genes during sponge development ( Aqu2 . 1 . 40735_001 , a E3 ubiquitin-protein ligase ) , and the gene with the highest coefficient of variation ( Aqu2 . 1 . 39666_001 , a putative sponge-specific gene specifically expressed in adult ) ( Figure 3B ) . The former showed a strong H3K4me3 enrichment at the TSS , whereas the latter lacked any marking , though its expression in the adult was ~70 times higher than the stable gene ( 33 vs 2361 CEL-seq normalized counts in adult , respectively ) . This lack of H3K4me3 at the TSS of regulated genes was similarly observed in the larva , exemplified here by a larva-specific regulated gene ( Aqu2 . 1 . 34366_001 ) expressed 3 . 5-fold higher than the above-mentioned stable gene ( Aqu2 . 1 . 40735_001 ) ( 147 vs 43 CEL-seq normalized counts in larva , respectively ) ( Figure 3B ) . Additionally , as shown in Drosophila ( Pérez-Lluch et al . , 2015a ) , regulated genes showed higher levels of H3K27me3 ( Mann-Whitney U test , p-value<6 . 517e-06 ) and lower levels of H3K36me3 ( Mann-Whitney U test , p-value<9 . 235e-08 ) than did stable genes ( Figure 3—figure supplement 1D and E ) . Analyzing RNA-seq–based gene expression through the development of the cnidarian Nematostella vectensis ( Helm et al . , 2013 ) and previously published ChIP-seq data sets in Nematostella adult female polyps ( Schwaiger et al . , 2014 ) , we obtained the same pattern ( Mann-Whitney U test , p-value<2 . 2e-16; Figure 3C ) . These results suggest that H3K4me3 might not be instrumental for extremely dynamic developmental expression and enforces our interpretation that it is required for tuning the levels of gene expression , a pattern that appears to be a conserved metazoan feature ( Pérez-Lluch et al . , 2015a ) . PRC2 is responsible for the trimethylation of lysine 27 of histone H3 ( H3K27me3 ) , one of the best-characterized repressive histone H3 PTMs ( Margueron and Reinberg , 2011 ) . As a step to investigate a putative mechanism of PRC2-mediated silencing in Amphimedon , we identified the sponge homologs of Drosophila PRC2 components and found that the Amphimedon genome contains four copies of E ( z ) homologs , two copies of ESC homologs and one copy for each of the remaining components , SU ( z ) 12 and Nurf55 ( Figure 4A; Figure 4—source data 1 ) . 10 . 7554/eLife . 22194 . 022Figure 4 . DNA motifs overrepresented in H3K27me3 transcriptionally silenced regions . ( A ) Diagram representing the composition of Drosophila PRC2 complex and its four core components: the catalytic subunit of the complex E ( z ) , the zinc finger protein SU ( z ) 12 , the WD-repeat protein ESC and the histone-binding protein Nurf55 . E ( z ) is responsible for the main enzymatic activity of PRC2 , which is to trimethylate histone H3 at lysine 27 , yielding H3K27me3 . Adapted from ( Vissers et al . , 2012 ) . The presence ( green ) or absence ( orange ) of PRC2 and its core components in the different opisthokont species represented in the phylogenetic tree ( left ) is shown . Amphimedon is highlighted in green . ( B ) Sequence logos of a subset of the DNA motifs determined by MEME-ChIP analysis to be significantly enriched in the transcriptionally silenced regions marked by H3K27me3 in adult Amphimedon . For each motif , the best TOMTOM match to a motif in the JASPAR CORE and UniPROBE mouse databases , the E-value and the number of sites contributing to the construction of the motif are shown , respectively . The matched motif is shown on the top and the query motif is shown on the bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 02210 . 7554/eLife . 22194 . 023Figure 4—source data 1 . Putative orthologs of Drosophila PcG components and associated factors in yeast , Capsaspora , sponge , nematode , and human genome . Table of PcG proteins is adapted from ( http://www . igh . cnrs . fr/equip/cavalli/link . PolycombTeaching . html ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 02310 . 7554/eLife . 22194 . 024Figure 4—figure supplement 1 . Matching sequence logos of the DNA motifs determined by MEME-ChIP analysis to be significantly enriched in the transcriptionally silenced regions marked by H3K27me3 in both adult and larva . The E-value ( log likelihood ratio of each motif ) and the number of sites contributing to the construction of the motif are shown , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 024 PRC2 recruitment has been best characterised in Drosophila where PRC2 proteins repress their target genes by recruitment to Polycomb Response Elements ( PREs ) , which contain binding sites for sequence-specific DNA binding proteins , including GAGA factor and members of the Krüppel-like factor family ( Müller and Kassis , 2006; Brown et al . , 2005; Strutt et al . , 1997; Simon and Kingston , 2009; Kassis and Brown , 2013 ) . To test whether Amphimedon PRC2 complexes might be recruited via a similar mechanism , we used the transcriptionally silenced regions marked by H3K27me3 in a de novo motif analysis ( Materials and methods ) . We searched for short motifs ( 6–15 bp ) on the basis that the known interaction sites of PREbinding proteins in Drosophila are of approximately this length ( ~8 bp ) . Conserved binding motifs similar to the GAGA and Krüppel-like factors , in addition to binding motifs similar to homeodomain-containing developmental regulators ( e . g . , Irx family members ) , were significantly enriched ( E-value<0 . 05 ) in the DNA associated with the H3K27me3 silenced regions in both adult and larva ( Figure 4B; Figure 4—figure supplement 1 ) . As in eumetazoans , this result suggests that Amphimedon PRC2 complexes are likely to be recruited through PRE-like sequences and may target developmental regulators for H3K27me3 deposition and transcriptional silencing ( Margueron and Reinberg , 2011; Di Croce and Helin , 2013; Boyer et al . , 2006 ) . An additional layer of regulatory complexity in eumetazoan development is provided by long intergenic non-coding RNAs ( lincRNAs ) ( Ulitsky , 2016; Hezroni et al . , 2015; Quinn and Chang , 2016 ) , which have been recently demonstrated to be developmentally expressed in sponges ( Gaiti et al . , 2015; Bråte et al . , 2015 ) . Here , we extended these analyses and analyzed the chromatin states of Amphimedon long intergenic ncRNAs ( lincRNAs ) ( Gaiti et al . , 2015 ) , avoiding lncRNAs in protein-coding sequence introns or antisense to coding genes , which may yield ambiguous signals . Previous studies have shown that the ratio of H3K4me1-to-H3K4me3 marks around TSSs can separate lincRNAs into enhancer-like lincRNAs ( elincRNAs; high H3K4me1-to-H3K4me3 ratio ) and canonical promoter-like lincRNAs ( plincRNAs; low H3K4me1-to-H3K4me3 ratio ) ( Sebé-Pedrós et al . , 2016; Marques et al . , 2013; IIott et al . , 2014 ) . Thus , to explore whether sponge lincRNAs might originate from enhancer regions , we interrogated our ChIP-seq data sets and calculated the relative ratio of H3K4me1-to-H3K4me3 in a 4 kb window centered on lincRNA TSSs . Only lincRNAs in scaffolds larger than 10 kb that overlapped with regions of enrichment of H3K4me1 , H3K4me3 , and RNAPII were used in this analysis ( n = 217 ) . Similarly to IIott et al . ( 2014 ) , we arbitrarily adopted a H3K4me1-to-H3K4me3 ratio of >1 . 2 and <0 . 8 to define elincRNAs and plincRNAs , respectively . Based on these criteria , we found 153 putative elincRNAs ( 70% ) significantly enriched for H3K4me1 over H3K4me3 ( Mann-Whitney U test , p-value=2 . 272e-05 ) and 21 ( 10% ) putative plincRNAs with canonical promoter signature , i . e . , higher enrichment of H3K4me3 over H3K4me1 ( Mann-Whitney U test , p-value=1 . 925e-07 ) . 43 ( 20% ) lincRNAs could not be assigned to either group , that is , 0 . 8 < H3K4me1-to-H3K4me3 < 1 . 2 ( Figure 5A–D; Figure 5—source data 1; Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 22194 . 025Figure 5 . Amphimedon lincRNA populations defined by histone PTM signatures . ( A ) Heatmap showing the average read normalised coverage of H3K4me1 , H3K4me3 and their ratio in adult Amphimedon across a 4 kb interval centred on TSSs of lincRNAs . Each line of the heatmaps represents a single lincRNA ( y-axis ) . Profiles are sorted based on the differences in enrichment between H3K4me1 and input DNA , and H3K4me3 and input DNA , respectively . Also provided is the H3K4me1:H3K4me3 log2 ( ratio ) around TSSs . ( B ) Enrichment of H3K4me1 ( left ) and H3K4me3 ( right ) ( ChIP versus input ) at plincRNAs and elincRNAs . P-values are indicated for Mann-Whitney U test . ( C ) Example of lincRNAs with promoter-like chromatin signature ( plincRNAs ) . For the definition of adult chromatin states see Figure 1A . Promoter-like lincRNAs ( blue ) are shown , along with input DNA-normalized coverage of different histone modifications and RNA-seq expression in adult . ( D ) Same as ( C ) but for lincRNAs with enhancer-like chromatin signature ( elincRNAs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 02510 . 7554/eLife . 22194 . 026Figure 5—source data 1 . Annotation of putative elincRNAs and plincRNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 02610 . 7554/eLife . 22194 . 027Figure 5—figure supplement 1 . Additional examples of plincRNAs and elincRNAs . ( A ) Example of a lincRNA with promoter-like chromatin signature ( plincRNA ) . For the definition of adult chromatin states see Figure 1A . Promoter-like lincRNA ( blue ) is shown , along with input DNA-normalized coverage of different histone modifications and RNA-seq expression in adult . ( B ) Same as ( A ) but for a lincRNA with enhancer-like chromatin signature ( elincRNA ) . ( C ) Same as ( B ) but for non-expressed lincRNAs with enhancer-like chromatin signature ( elincRNAs ) . Note the prevalence of ‘ReprPC’ , ‘ReprPCWk’ and ‘EnhP’ chromatin state at these regions . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 027 These results indicate that sponge lincRNAs can be separated in two distinct populations of poly ( A ) + transcripts based on the chromatin status at their TSSs . Although these two populations resemble those found in human , mouse and Capsaspora lincRNAs ( Sebé-Pedrós et al . , 2016; Marques et al . , 2013; IIott et al . , 2014 ) , their functional significance is yet to be determined . To identify putative enhancer elements in Amphimedon in silico , we selected distal H3K4me1 regions of enrichment ( high confidence regions , representing reproducible events across true biological replicates ) that did not overlap TSSs ( ±200 bp ) of protein-coding genes and lncRNAs , but overlapped with regions designated as being in an enhancer chromatin state based on the ChromHMM analysis ( ‘TxEnhA’ or ‘EnhWk’ or ‘EnhP’ state in adult; ‘TxEnhA1’ or ‘TxEnhA2’ or ‘EnhWk’ or ‘EnhP’ state in larva , which consist of typical eumetazoan enhancer histone H3 PTM patterns ) ( Figure 6A ) . A subset of these regions was also marked by H3K27ac , and therefore likely to be transcriptionally active ( Figure 6A and B; Figure 6—source data 1 ) . These predicted activated enhancer-like regions showed a significant enrichment of H3K4me1 and H3K27ac over H3K4me3 ( Mann-Whitney U test , p-value<2 . 2e-16; Figure 6C; Figure 6—figure supplement 1 ) , a biochemical signature typical of eumetazoan enhancers ( Schwaiger et al . , 2014 ) . Interestingly , RNAPII occupied some of these Amphimedon predicted activated enhancer-like elements ( 35% and 41% in adult and larva , respectively ) , suggesting poly ( A ) + enhancer RNAs could be transcribed from these regions ( Natoli and Andrau , 2012; Li et al . , 2016; Kim et al . , 2010 ) ( Figure 6A–D; Figure 6—figure supplement 2 ) . Alternatively , but not exclusively , this might represent the result of chromatin looping and the simultaneous pulldown of both enhancers and promoters with the RNAPII antibody ( Shlyueva et al . , 2014 ) . 10 . 7554/eLife . 22194 . 028Figure 6 . Distal enhancer regulation at the dawn of animals . ( A ) Overview of the computational filtering pipeline adopted to predict the putative Amphimedon activated enhancer-like elements . See main text and Materials and methods for details . ( B ) Heatmap showing different histone modifications enrichment at predicted activated enhancer-like elements ( ±2 kb of flanking regions ) . ( C ) Boxplot showing enrichment of different histone modifications ( ChIP versus input ) at predicted activated enhancer-like elements , showing that activated enhancer-like elements have higher H3K4me1 than H3K4me3 levels , a typical characteristic of eumetazoan enhancers . Four asterisks ( **** ) indicate p-values<2 . 2e-16 for Mann-Whitney U test between H3K4me3 and H3K27ac , between H3K4me3 and H3K4me1 , and between H3K4me3 and RNAPII , respectively . ( D ) Example of predicted activated enhancer-like elements . Protein coding genes ( purple ) are shown , along with input DNA-normalized coverage of different histone modifications and RNA-seq expression in adult . Regions of enrichments ( high confidence peaks , representing reproducible events across true biological replicates ) corresponding to the predicted activated enhancer-like elements are highlighted in grey . ( E ) Sequence logos of the DNA motifs determined by MEME-ChIP analysis enriched in the adult predicted activated enhancer-like sequences . For each motif , the best match to a motif in the JASPAR CORE and UniPROBE mouse databases , the E-value and the number of sites contributing to the construction of the motif are shown , respectively . The matched motif is shown on the top and the query motif is shown on the bottom . ( F ) Gene Ontology ( GO ) enrichment activities of the nearest neighbor protein-coding genes of the adult predicted activated enhancer-like elements are shown . Bar length indicates the significance of the enrichment ( Hypergeometric test; -log10[adjusted P- value] ) . Only the top ten GO biological process terms are shown . See Figure 6—source data 2 for the complete list . ( G ) Boxplot showing the size of introns that harbour adult activated enhancer-like elements versus all introns in the genome . The y-axis indicates the intron size ( bp ) in log scale . P-value is indicated for Mann–Whitney U test . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 02810 . 7554/eLife . 22194 . 029Figure 6—source data 1 . Genomic location of all the predicted activated enhancer-like elements and their distance to the closest TSS . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 02910 . 7554/eLife . 22194 . 030Figure 6—source data 2 . Functional annotation of nearest neighbors genes of the adult predicted activated enhancer-like elements . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 03010 . 7554/eLife . 22194 . 031Figure 6—source data 3 . Functional annotation of nearest neighbors genes of the larva predicted activated enhancer-like elements . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 03110 . 7554/eLife . 22194 . 032Figure 6—source data 4 . GO term enrichment outcome for the nearest neighbors genes of the adult predicted activated enhancer-like elements ( Hypergeometric test , FDR<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 03210 . 7554/eLife . 22194 . 033Figure 6—source data 5 . GO term enrichment outcome for the nearest neighbors genes of the larva predicted activated enhancer-like elements ( Hypergeometric test , FDR<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 03310 . 7554/eLife . 22194 . 034Figure 6—figure supplement 1 . Activated enhancer-like elements have higher H3K4me1 than H3K4me3 levels . ( A ) Enrichment of H3K4me1 ( red ) and H3K4me3 ( blue ) ( ChIP versus input ) at predicted activated enhancer-like elements ( left ) and TSSs ( right ) , showing that activated enhancer-like elements have higher H3K4me1 than H3K4me3 levels , a typical characteristic of eumetazoan enhancers . P-values are indicated for Mann-Whitney U test . ( B ) Relative distance distribution observed between H3K4me1 ( red ) and H3K4me3 ( blue ) regions of enrichment ( peaks ) and TSSs . If there is no spatial correlation between the two sets , one would expect the relative distances to be uniformly distributed among the relative distances ranging from 0 to 0 . 5 , as observed for H3K4me1 peaks and TSSs . If , however , the intervals tend to be much closer than expected by chance , the distribution of observed relative distances would be shifted towards low relative distance values , as observed for H3K4me3 peaks and TSSs . ( C ) Relative distance distribution observed between H3K4me1 ( red ) and H3K4me3 ( blue ) peaks and predicted activated enhancer-like elements , indicating spatial correlation between H3K4me1 peaks and predicted activated enhancer-like elements . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 03410 . 7554/eLife . 22194 . 035Figure 6—figure supplement 2 . Examples of CEL-seq or RNA-seq expression detected at putative activated enhancer-like sites , suggesting that 1D eRNAs , which are generally polyadenylated ( Natoli and Andrau , 2012; Li et al . , 2016 ) , might be transcribed from these regions . Protein-coding genes ( purple ) are shown , along with input DNA-normalized coverage tracks of different histone modifications . Regions of enrichments ( high confidence peaks , representing reproducible events across true biological replicates ) corresponding to putative activated enhancer-like sites are highlighted in grey . Unless otherwise specified , adult data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 03510 . 7554/eLife . 22194 . 036Figure 6—figure supplement 3 . Additional sequence logos of the DNA motifs determined by MEME-ChIP analysis to be significantly enriched in the adult predicted activated enhancer-like sequences . The E-value ( log likelihood ratio of each motif ) and the number of sites contributing to the construction of the motif are shown , respectively . The matched motif is shown on the top and the query motif is shown on the bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 03610 . 7554/eLife . 22194 . 037Figure 6—figure supplement 4 . Matching sequence logos of the DNA motifs determined by MEME-ChIP analysis to be significantly enriched in the predicted activated enhancer-like sequences in both adult and larva . The E-value ( log likelihood ratio of each motif ) and the number of sites contributing to the construction of the motif are shown , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 03710 . 7554/eLife . 22194 . 038Figure 6—figure supplement 5 . Examples of predicted enhancer-like elements in proximity of well-known developmental and transcription factor genes . Protein-coding genes ( purple ) are shown , along with input DNA-normalized coverage of different histone modifications and RNA-seq expression . Regions of enrichments ( high confidence peaks , representing reproducible events across true biological replicates ) corresponding to adult putative enhancer-like sites are highlighted in grey . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 038 In eumetazoans , genes encoding transcriptional regulators are themselves regulated by multiple enhancer elements ( Schwaiger et al . , 2014; Shlyueva et al . , 2014; Nègre et al . , 2011; Bogdanovic et al . , 2012; Woolfe et al . , 2005; Heintzman et al . , 2009 ) . We therefore performed de novo motif analysis and , despite the limited power of motif detection due to the inherent cellular heterogeneity of our starting material , we were able to show that consensus binding motifs of key developmental transcription factor families were over-represented in the adult predicted activated enhancer-like sequences , including Zinc finger , Irx , SOX and POU binding motifs ( Figure 6E; Figure 6—figure supplement 3 ) . It is noteworthy that Zinc fingers can also be involved in roles that might be unrelated to directly regulating gene expression per se , e . g . , chromatin remodeling ( Wysocka et al . , 2006 ) . Similar binding motifs were obtained analysing the larva predicted activated enhancer-like sequences ( Figure 6—figure supplement 4 ) . Next , we examined whether the sponge predicted activated enhancer-like elements were preferentially located next to protein-coding genes involved in development and/or transcriptional regulation . By searching for the closest located TSSs to each of the predicted activated enhancer-like elements in Amphimedon , we nominated putative target protein-coding genes . Akin to eumetazoans , these nearest neighbor genes were significantly enriched for Gene Ontology ( GO ) terms associated with transcription factor activity and developmental processes ( Hypergeometric test , FDR adjusted p-value<0 . 01 ) ( Figure 6F ) , and comprised several transcription factors , including SOX2 , FOS and NF-kB ( Figure 6D; Figure 6—figure supplement 5; Figure 6—source data 2–5 ) . Vertebrates exhibit expansive intergenic regions where the majority of predicted enhancers are located ( ENCODE Project Consortium , 2012; Djebali et al . , 2012 ) . In contrast , in Amphimedon , which has a highly compact genome with minimal intergenic regions ( Fernandez-Valverde and Degnan , 2016 ) , predicted activated enhancer-like elements were predominantly intragenic , with only a minority found in intergenic regions ( 9% and 20% in adult and larva , respectively ) ( Fernandez-Valverde and Degnan , 2016 ) . This , along with the strong enrichment of chromatin states typically associated with eumetazoan enhancers – ‘TxEnhA’ and ‘EnhWk’ – in introns ( Figure 1A; Figure 1—figure supplement 4 ) , suggests a similar overall genomic distribution between Amphimedon , Nematostella and Drosophila enhancer elements ( Schwaiger et al . , 2014; Nègre et al . , 2011; Arnold et al . , 2013 ) . Greater intron length often associates with the presence of highly conserved non-coding elements ( Irimia et al . , 2011 ) . We , therefore , extracted the introns that harbour predicted activated enhancer-like elements and compared their size distribution to the size of all intronic regions found across the genome . The former were significantly longer than the average genomic intron size , with a mean of 332 bp and 256 bp , and a median of 99 bp and 71 bp , respectively ( Ansari-Bradley test , p-value=0 . 06151; Mann-Whitney U test , p-value=1 . 927e-06 ) ( Figure 6G ) , suggesting that a cis-regulatory expansion appear to have occurred primarily in intronic rather than intergenic regions in Amphimedon . Highly conserved non-coding regulatory elements are often associated not only with greater intron length , but also with genes encoding developmental regulators ( Woolfe et al . , 2005; Vavouri et al . , 2007; Sandelin et al . , 2004 ) . Particularly interesting are the conserved ancestral microsyntenic pairs ( herein microsyntenic units ) that consist of either ( i ) two neighbor genes that share common cis-regulatory elements , or ( ii ) a developmental regulator and nearby functionally unrelated gene ( s ) , whose introns harbor conserved cis-regulatory elements ( Kikuta et al . , 2007; Irimia et al . , 2013; Engström et al . , 2007; Irimia et al . , 2012; Naville et al . , 2015 ) . Experimental evidence has been provided for the existence of this type of cis-regulation in vertebrates ( Irimia et al . , 2012; Naville et al . , 2015 ) . To test whether this is an ancient cis-regulatory mechanism maintained through animal evolution , we assessed the spatial relationship between the genes of each of the 80 microsyntenic units previously reported to be present in the Amphimedon genome ( Irimia et al . , 2012 ) and clarified their orthology , confirming the presence of 60 unambiguous microsyntenic units . Remarkably , 43 of these 60 evolutionary conserved metazoan microsyntenies contained putative enhancer-like signatures in Amphimedon adults ( Figure 7A; Figure 7—source data 1; Figure 7—figure supplement 1 ) . This was a much higher fraction relative to a control set consisting of 60 pairs of two randomly selected nonsyntenic neighbor genes ( 1 , 000 iterations; p-value<0 . 00001 ) . This pattern was substantiated by the finding of larva enhancer-like signatures in 16 of the 60 microsyntenic units , seven of which contained both larva and adult predicted enhancer-like elements ( Figure 7A; Figure 7—source data 1; Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 22194 . 039Figure 7 . Amphimedon enhancer-like elements are enriched in metazoan-specific microsyntenic units . ( A ) Putative adult and larva enhancer-like signatures identified in the 60 metazoan-specific microsyntenic pairs investigated . ( B ) The cladogram represents known phylogenetic distribution of the Isl2-Scaper microsyntenic gene pair across opisthokonts . The orientation of the arrow corresponds to gene orientation . Isl2-Scaper is not conserved in yeast , Capsaspora , Nematostella and C . elegans . ( C ) Enhancer elements in the Isl-Scaper microsyntenic gene pair locus in Amphimedon . Scaper and Isl genes ( purple ) are shown , along with input DNA-normalized coverage of H3K4me3 and H3K4me1 and RNA-seq expression in both adult and larva . Regions of enrichments ( high confidence peaks , representing reproducible events across true biological replicates ) corresponding to the predicted enhancer-like elements located within the introns of Scaper are highlighted in grey . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 03910 . 7554/eLife . 22194 . 040Figure 7—source data 1 . 60 microsyntenic units representing functional gene linkages and presence-absence of chromatin states containing typical eumetazoan enhancer histone PTM patterns ( ‘EnhP’ , ’ EnhWk’ , ’ TxEnhA’ ) ( adult only ) and/or in silico predicted enhancer-like elements ( both larva and adult ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 04010 . 7554/eLife . 22194 . 041Figure 7—figure supplement 1 . Additional examples of predicted enhancer-like elements in conserved microsyntenic units . Protein-coding genes ( purple ) are shown , along with input DNA-normalized coverage of different histone modifications and RNA-seq/CEL-seq expression . Regions of enrichments ( high confidence peaks , representing reproducible events across true biological replicates ) corresponding to putative enhancer-like sites located within the microsyntenic units are highlighted in grey . Unless otherwise specified , adult data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 041 A striking case of conserved gene linkage involves the Islet LIM homeobox gene ( Isl ) , which plays conserved roles in animal development ( Thor and Thomas , 1997; Liang et al . , 2011 ) , and Scaper ( S-phase cyclin A-associated protein in the ER ) ( Figure 7B ) . The Amphimedon Scaper contains 25 introns , some of which are considerably longer ( >1 kb ) than the mean intron size ( Fernandez-Valverde et al . , 2015 ) , and predicted enhancer-like elements located within its intron 10 , 17 and 21 ( Figure 7C ) . Likewise , the microsynteny of Tfap4 ( transcription factor AP-4 ) ( Simionato et al . , 2007 ) and Glis2 ( GLIS family zinc finger 2 ) is deeply conserved . Similar to an observation in vertebrates ( Abbasi et al . , 2007 ) , the sponge Glis2 contains two introns , of which the second harbors several adult predicted activated enhancer-like elements ( Figure 7—source data 1; Figure 7—figure supplement 1 ) . Together , these results suggest that the genomic location of some cis-regulatory elements likely places constraints on the evolution of nearby genes , leading to the occurrence of conserved microsyntenic gene blocks across the animal kingdom .
Despite Amphimedon’s morphological simplicity , we find strong evidence in this sponge for the existence of a range of regulatory states that underlie eumetazoan development . For instance , the genome-wide promoter analysis of H3K4me3 – the canonical and widespread eukaryotic histone H3 PTM of active transcription – reveals a complex correlation between H3K4me3-containing nucleosome occupancy and gene expression in Amphimedon adults and larvae , consistent with an active and finely tuned role for H3K4me3 in modulating transcriptional activity and expression variability of developmental genes . Unexpectedly , we identify a small subpopulation of highly and specifically expressed genes that challenge this premise and are transcribed in the absence of H3K4me3 in Amphimedon and Nematostella . This subpopulation of genes differs from most other developmentally-expressed genes that possess the H3K4me3 mark , in having much more stage-restricted expression profiles; in this analysis , most are expressed only in one stage of development . Although it could be argued that this apparent absence of H3K4me3 is the consequence of the expression of regulated genes being spatially confined to specific cell populations , thus potentially limiting our detection sensitivity with our cell admixture ChIP-seq , these results run parallel to the recent finding by Pérez-Lluch et al . ( 2015a ) that Drosophila and C . elegans exhibit the same pattern , suggesting that this newly-discovered feature is conserved across the animal kingdom . As the expression of the developmentally regulated genes is required only for a limited period , the absence of H3K4me3 mark would allow their rapid on-off switching . Alternative mechanisms , such as the transient binding of transcription factors , appear to play a major role in regulating the expression of these genes ( Pérez-Lluch et al . , 2015a , 2015b ) . Polycomb Repressive Complex 2 ( PRC2 ) primarily trimethylates histone H3 on lysine 27 and has been conserved throughout opisthokonts evolution , with its core subunits ( E ( z ) , SU ( z ) 12 , ESC and Nurf55 ) being present in animals , choanoflagellates and multicellular fungi , but absent in Capsaspora , and budding and fission yeast ( Sebé-Pedrós et al . , 2016; Margueron and Reinberg , 2011; Shaver et al . , 2010; Jamieson et al . , 2013; Connolly et al . , 2013; Ikeuchi et al . , 2015; Whitcomb et al . , 2007 ) ( Figure 4A ) . This is consistent with PRC2 complex being lost in several unicellular lineages . One of the ancestral roles of PCR2 in opisthokonts may have been in defense response against viruses and transposable elements , or insertion of new genes ( Jamieson et al . , 2013 ) , prior to being co-opted for cell-type specific developmental regulation in animals , where H3K27me3 and PRC2 are required for transmitting the memory of repression across generations and during development ( Margueron and Reinberg , 2011; Shaver et al . , 2010; Gaydos et al . , 2014; Barski et al . , 2007 ) . In fact , PRC2 often regulates deposition of H3K27me3 marks at loci encoding developmental regulators ( Ha et al . , 2011; Margueron and Reinberg , 2011; Barski et al . , 2007 ) . The finding of short conserved developmental transcription factor-binding-sites in Amphimedon H3K27me3 silenced regions is consistent with this evolutionary scenario . Analogous to recent findings in plants ( Deng et al . , 2013; Hecker et al . , 2015 ) , the identification of an enriched motif in the H3K27me3 silenced regions similar to the GAGA factor binding site , a component of the Drosophila Polycomb group response elements , suggests a role for the GAGA factor binding sites in strengthening PRC2 recruitment to target genes ( Müller and Kassis , 2006; Simon and Kingston , 2009; Kassis and Brown , 2013 ) . It is noteworthy that a sponge homolog of Drosophila GAGA factor was not identified in the current Amphimedon genome assembly ( Figure 4—source data 1 ) , suggesting the convergent co-option of other DNA binding proteins with analogous role ( s ) in the recruitment of PRC2 . Analysis of cis-regulatory DNA and histone PTMs have revealed that some cis-regulatory mechanisms , such as those associated with proximal promoters , are present in non-animal holozoans , while others appear to have evolved later on the stem leading to the crown metazoans , most notably distal enhancers ( Sebé-Pedrós et al . , 2016; Schwaiger et al . , 2014 ) . The latter has been posited to be one of the key contributing factors underlying the spatial and temporal coordination of cell differentiation that defines animal development ( Levine , 2010; Levine et al . , 2014; Levine and Tjian , 2003; Peter and Davidson , 2011 ) . Our in silico prediction of Amphimedon enhancer elements based on histone H3 PTM co-localization patterns is consistent with these elements evolving along the metazoan stem at the transition to multicellularity ( Sebé-Pedrós et al . , 2016 ) . Interestingly , promoter DNA regulatory elements to allow for context and cell type-specific gene expression also appeared to evolve in stem metazoans ( Fernandez-Valverde and Degnan , 2016 ) , suggesting these are also a critical component of the animal cis-regulatory landscape . Amphimedon predicted enhancer-like elements are characterized by the same combination of histone H3 PTMs as in eumetazoans , which appear to be lacking in unicellular holozoan relatives of animals ( Sebé-Pedrós et al . , 2016; Bulger and Groudine , 2011 ) . Their preferential association with developmental and transcriptional regulators suggests that Amphimedon enhancer elements are likely to regulate developmental genes in a manner akin to eumetazoans ( Schwaiger et al . , 2014; Shlyueva et al . , 2014; Nègre et al . , 2011; Bogdanovic et al . , 2012; Woolfe et al . , 2005; Heintzman et al . , 2009 ) . Enhancer elements are known to be associated with the transcription of both short poly ( A ) - and long poly ( A ) + enhancer RNAs ( 2D and 1D eRNAs , respectively ) ( Natoli and Andrau , 2012; Li et al . , 2016; Kim et al . , 2010 ) . The presence of RNAPII and the detection of expression at a subset of the Amphimedon activated enhancer-like elements is consistent with this notion ( Figure 6—figure supplement 2 ) . Although non-coding transcription at these enhancers will need to be investigated in detail , this co-occupancy of enhancer elements and RNAPII has also been observed in Nematostella and bilaterians ( Schwaiger et al . , 2014; Li et al . , 2016; Kim et al . , 2010; De Santa et al . , 2010; Chen et al . , 2013 ) , where these elements might be physically interacting with the transcription initiation complex at the TSS of their target gene ( s ) ( Schwaiger et al . , 2014 ) . Unlike bilaterians , where the transcriptional repressor CCCTC-binding factor ( CTCF ) localizes with cohesin genome-wide and is involved in enhancer-promoter long-range interactions and higher-order chromatin structure ( Lee and Iyer , 2012; Seitan et al . , 2013; Merkenschlager and Odom , 2013 ) , Amphimedon lacks CTCF ( Heger et al . , 2012 ) . This likely constrains Amphimedon enhancer interactions with the proximal promoter transcriptional machinery to short distances . Chromatin looping of enhancers to their target promoters in this sponge might therefore occur through a CTCF-independent cohesin binding mechanism , as proposed in cnidarians , which also lack CTCF ( Schwaiger et al . , 2014 ) . Alternatively , but not exclusively , RNAPII and its associated transcriptional machinery may track through the intervening DNA between enhancers and promoters ( Li et al . , 2016 ) , and might be the preferred mechanism of enhancer-promoter interactions in this sponge . The co-occupancy of Amphimedon enhancer-like elements and RNAPII supports this mechanism of transcriptional activation . Future studies of the 3D genome architecture will be crucial in elucidating the mechanism of enhancer-promoter interaction in this sponge and other early-branching non-bilaterian animals lacking this architectural protein ( Gaiti et al . , 2016 ) . Finally , we find strong evidence for cis-regulatory elements being important for the maintenance of metazoan-specific microsyntenic gene blocks over 700 Myr of evolution . The emergence of distal enhancer regulation prior to metazoan cladogenesis could explain the pervasiveness of conserved syntenic regulatory blocks in animal genomes and the absence of these blocks in their unicellular relatives ( Srivastava et al . , 2010; Sebé-Pedrós et al . , 2016; Irimia et al . , 2013 , 2012; Bulger and Groudine , 2011; Putnam et al . , 2007; Duan et al . , 2010 ) . The strong evidence for enhancer elements being enriched in deeply conserved metazoan-specific microsyntenic units suggests that their genomic location is likely to constraint genome architecture , leading to the occurrence of conserved microsyntenies across the animal kingdom ( Irimia et al . , 2013 , 2012 ) . In conclusion , a conserved gene regulatory landscape similar to that of morphologically-complex eumetazoans appears to have been already in place at the dawn of animals , and thus likely to have originated at least 700 Mya . Specifically , there appears to have been fundamental changes in the cis-regulatory architecture of the genome along the metazoan stem , concomitant with the evolution of animal multicellularity , including the apparent origin of distal enhancers and promoter types for cell-type-specificity and developmental regulation . With this in mind , we propose an evolutionary scenario in which quantitative rather than qualitative differences in regulatory mechanisms likely drive the evolution and diversification of eumetazoan body plans ( Figure 8 ) . 10 . 7554/eLife . 22194 . 042Figure 8 . Origin of animal cis-regulatory complexity . The phylogenetic relationship of representative animal lineages and unicellular holozoans is shown here . Highlighted are the major genomic innovations that correlate with the emergence and early diversification of animals . Some components of the metazoan regulatory landscape may predate the split of the metazoan and holozoan lineages , including core TF-TF regulatory interactions and long intergenic non-coding RNAs , which have been recently identified in unicellular relatives of animals ( Sebé-Pedrós et al . , 2016; de Mendoza et al . , 2015 ) but for which the evolutionary origin is still unclear . With a complex gene regulatory landscape already in place at the dawn of animals , the expansion of developmental gene families ( encoding transcription factors and components of signaling pathways ) , cis-regulatory DNA and non-coding RNAs , along with the emergence of the architectural protein CTCF to allow more complex enhancer-promoter interactions , appear to underlie the evolutionary diversification of the eumetazoan body plans . DOI: http://dx . doi . org/10 . 7554/eLife . 22194 . 042
Amphimedon queenslandica adults and larvae were collected from Heron Island Reef , Great Barrier Reef , Queensland , Australia , and reared as previously described ( Leys et al . , 2008 ) . We used a mouse monoclonal antibody against the unphosphorylated C-terminal repeat of RNA polymerase II ( RRID:AB_492629 ) ( clone 8WG16 , #05–952 , Merck Millipore , Billerica , MA ) , a rabbit polyclonal antibody against H3K4me3 ( RRID:AB_1977252 ) ( #07–473 , Merck Millipore ) , a rabbit polyclonal antibody against H3K27me3 ( RRID:AB_310624 ) ( #07–449 , Merck Millipore ) , a mouse monoclonal antibody against H3K4me1 ( RRID:AB_10806625 ) ( #17–676 , Merck Millipore ) , a rabbit polyclonal antibody against H3K27ac ( RRID:AB_310550 ) ( #07–360 , Merck Millipore ) , a rabbit monoclonal antibody against H3K36me3 ( RRID:AB_10615601 ) ( #17–10032 , Merck Millipore ) , and a rabbit polyclonal antibody against histone H3 ( RRID:AB_417398 ) ( #07–690 , Merck Millipore ) ( Figure 1—source data 1 ) . The entire amino acid sequence of histone H3 is perfectly conserved between Amphimedon and other eukaryotes where these antibodies have been used successfully ( Sebé-Pedrós et al . , 2016; Ercan et al . , 2009; Barraza et al . , 2015; Harmeyer et al . , 2015; Liu et al . , 2007; Eckalbar et al . , 2016 ) ( Figure 1—figure supplement 1 ) . Approximately a cm3 of adult sponge tissue was squeezed through a fine cloth and cells ( ~107 ) were crosslinked in 2% formaldehyde for 5 min at room temperature ( RT ) . Larvae ( ~350 ) were pooled , homogenized and crosslinked as above . A similar procedure was then adopted for both developmental stages . Specifically , crosslinking was quenched with 125 mM glycine for 5 min at RT . Cells were washed twice in 0 . 22 µm filtered seawater and centrifuged at 500 g for 5 min . Pelleted cells were lysed in SDS Lysis buffer ( 10 mM EDTA , 50 mM Tris-HCl at pH 8 . 0 , 1% SDS , plus protease and phosphatase inhibitors ) , incubated for at least 10 min on ice , and sonicated for 12 min ( 12 cycles , each one 30 s ‘ON’ , 30 s ‘OFF’ ) in a Bioruptor Sonicator ( Diagenode , Seraing , Belgium ) to generate 200–300 bp fragments . Optimal sonication conditions were previously determined by testing a range of sonication cycles ( from 5 to 30 ) ; 12 cycles were deemed as optimal . Non-soluble material was removed from the lysate by centrifugation at 12 , 000 g for 10 min at 4°C . An aliquot of the soluble material was removed for input DNA and stored at −20°C . To reduce the SDS concentration to 0 . 1% , the remaining soluble material was diluted 10-fold in ChIP dilution buffer ( 1 . 2 mM EDTA , 16 . 7 mM Tris-HCl at pH 8 . 0 , 167 mM NaCl , 1 . 1% Triton X-100 , 0 . 01% SDS , plus PhosSTOP phosphatase inhibitor and cOmplete protease inhibitor cocktail [Roche , Basil , Switzerland] ) . To reduce non-specific background , the diluted soluble material was pre-cleared with Dynabeads protein G beads ( #10003D , ThermoFisher , Waltham , MA ) , and , at the same time , the antibodies were linked to Dynabeads protein G beads ( #10003D , ThermoFisher ) by rotating for one hour at 4°C . At this point , the pre-cleared diluted soluble material was incubated with the antibody-bead mixtures , rotating at 4°C overnight . Immunoprecipitated material was washed three times with Low Salt Wash Buffer ( 2 mM EDTA , 20 mM Tris-HCl at pH 8 . 0 , 150 mM NaCl , 1% Triton X-100 , 0 . 1% SDS ) , three times with High Salt Wash Buffer ( 2 mM EDTA , 20 mM Tris-HCl at pH 8 . 0 , 500 mM NaCl , 1% Triton X-100 , 0 . 1% SDS ) , three times with LiCl Wash Buffer ( 1 mM EDTA , 1 mM Tris-HCl at pH 8 . 0 , 1% DOC , 1% NP-40 , 250 mM LiCl ) , and three times with TE buffer ( 10 mM Tris·Cl , pH 8 . 0; 1 mM EDTA ) . DNA complexes were eluted 30 min at 65°C with TE-SDS ( 10 mM Tris·Cl , pH 8 . 0; 1 mM EDTA; 1% SDS ) and decrosslinked overnight at 65°C , along with input DNA , with the addition of 125 mM NaCl . Decrosslinked DNA complexes and input DNA were treated with RNaseA , and subsequently with proteinase K . Finally , immunoprecipitated and input DNA were purified with phenol:chloroform:isoamyl extraction ( 25:24:1 ) , recovered by precipitation with ethanol in the presence of 300 mM NaOAc pH 5 . 2 and 2 µl of glycogen carrier ( 10 mg/ml ) , and resuspended in UltraPure DNase/RNase-Free Distilled Water ( ThermoFisher ) for later use . Libraries of immunoprecipitated DNA and input DNA were prepared using the NEBNext ChIP-seq Library Prep Master Mix Set for Illumina ( #E6240 , New England Biolabs , Ipswich , MA ) according to the manufacturer’s protocol . The quality and profile of the libraries was analyzed using Agilent High Sensitivity DNA Kit ( #5067–4626 , Agilent , Santa Clara , CA ) and quantified using KAPA Library Quantification Kit ( #KK4824 , Kapa Biosystems , Wilmington , MA ) . Deep sequencing ( 100 bp paired-end ) of the adult libraries – two biological replicates for H3K4me3 , H3K4me1 , H3K36me3 , H3K27me3 , RNAPII , input DNA and no biological replicates for H3K27ac and total histone H3 – was performed by the Macrogen Oceania NGS Unit on Illumina HiSeq 2000 instrument ( Illumina , San Diego , CA , United States ) . Deep sequencing ( 40 bp paired-end ) of the larva libraries – no biological replicates for H3K4me3 , H3K4me1 , H3K27me3 , H3K27ac , RNAPII , input DNA – was performed by the Central Analytical Research facility ( CARF ) , Brisbane , Queensland , Australia , on Illumina NextSeq 500 instrument ( Illumina , San Diego , CA , United States ) . Adult raw Illumina sequencing reads were checked using FastQC v0 . 52 ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) and quality filtered using Trimmomatic v1 . 0 . 0 ( SLIDINGWINDOW: 4:15 , LEADING: 3 , TRAILING: 3 , HEADCROP: 5 , MINLEN: 50 ) ( RRID:SCR_011848 ) ( Bolger et al . , 2014 ) . Quality filtered paired-end Illumina sequencing reads were then aligned to the Amphimedon genome ( Srivastava et al . , 2010 ) using Bowtie v1 . 1 . 2 ( RRID:SCR_005476 ) ( Langmead et al . , 2009 ) with -m 1 , -n 2 , -X 500 , --best parameters ( uniquely mapped reads and maximum of two mismatches within the seed ) . Non-aligned reads were removed using SAMtools v0 . 1 . 19 ( RRID:SCR_002105 ) ( Li et al . , 2009 ) . For all the ChIP-seq data sets , strand cross-correlation measures were used to estimate signal-to-noise ratios using SPP v1 . 11 . 0 ( RRID:SCR_001790 ) . ChIP-seq data sets for each mark were flagged if the scores were below a normalized strand cross-correlation coefficient ( NSC ) threshold of 1 . 05 , as described in the modENCODE and ENCODE guidelines ( ENCODE Project Consortium , 2012; Landt et al . , 2012; Kellis et al . , 2014; Kharchenko et al . , 2008 ) . These analyses were performed on Galaxy-qld server ( http://galaxy-qld . genome . edu . au/galaxy ) developed within the GVL project ( Afgan et al . , 2015 , 2016 ) and maintained by the Research Computing Centre , University of Queensland , Australia . Pearson’s correlation coefficients ( Pearson’s r ) of genome-wide fold enrichment ( FE ) signals ( see below ) was computed for biological replicates and a minimum threshold of 0 . 5 was required , as per Ho et al . ( 2014 ) . In addition , to ensure consistency between biological replicates , we further required an Irreproducible Discovery Rate ( IDR ) of at least 0 . 5 ( see below ) , as described in the modENCODE and ENCODE guidelines ( ENCODE Project Consortium , 2012; Landt et al . , 2012; Kellis et al . , 2014; Kharchenko et al . , 2008 ) . ChIP-seq data sets that met these criteria were then merged across biological replicates ( see , ) . Histone PTM regions of enrichment relative to corresponding sequenced input DNA controls were determined using MACS2 v2 . 1 . 0 ( RRID:SCR_013291 ) ( Zhang et al . , 2008 ) according to modENCODE , ENCODE and Roadmap Epigenomics consortiums guidelines ( Kundaje et al . , 2015; ENCODE Project Consortium , 2012; Landt et al . , 2012; Kellis et al . , 2014; Kharchenko et al . , 2008 ) . Specifically , MACS2 was used in broadpeak mode with a broadpeak P-value threshold of 0 . 1 and a narrowpeak threshold of 0 . 01 ( -p 0 . 01 , --broad , --nomodel , --extsize 146 , -g 1 . 45e8 ) . Enriched regions were scored on individual replicates ( R1 and R2 ) , pooled data ( reads pooled across biological replicates ) ( P ) and on subsampled pseudoreplicates ( obtained by pooling reads from biological replicates and randomly subsampling , without replacement , two pseudoreplicates with half the total number of pooled reads ) ( PR1 and PR2 ) . For each histone PTM , we defined ‘R’ as the set of peaks in P that overlap peaks in R1 and R2 , and ‘PR’ as the set of peaks in P that overlap peaks in PR1 and PR2 . Next , we defined ‘M’ as the set of peaks that match exactly in R and PR , and ‘T’ as the set of peaks that match exactly in R and PR as well as those that are unique to R or unique to PR . For a statement about reproducibility we required the M-to-T ratio to be at least 0 . 5 ( Figure 1—source data 4 ) . To obtain reliable regions of enrichment , we restricted all further analyses to enriched regions identified using pooled data that were also independently identified in both replicates and pseudoreplicates ( the ‘M’ set ) . These regions of enrichment can be interpreted as high confidence regions , representing reproducible events across true biological replicates . For H3K27ac , for which no replication was available , we used the P-value column to rank peaks and only retained peaks with a p-value<0 . 001 . We used the gappedPeak representation for the histone PTMs with relatively compact enrichment patterns , including H3K4me3 , H3K27ac and H3K4me1 . The gapped peaks are broad domains ( passing P-value 0 . 1 ) that contain at least one narrow peak passing a P-value of 0 . 01 . For the diffused histone PTMs – H3K36me3 and H3K27me3 – we used the broadPeak representation . RNAPII peaks were detected using the peakzilla software ( RRID:SCR_007471 ) ( Bardet et al . , 2013 ) , using input DNA reads as control ( -c 1 . 5 , –s 3 ) . The fraction of reads falling within peak regions ( FRiP ) was also calculated ( see Figure 1—source data 4 ) . In line with ENCODE guidelines ( ENCODE Project Consortium , 2012; Landt et al . , 2012; Kellis et al . , 2014; Kharchenko et al . , 2008 ) , all our data sets have a FRiP enrichment of 1% or more . For every pair of aligned ChIP and matching input DNA data sets , we also used MACS2 ( Zhang et al . , 2008 ) to generate genome-wide signal coverage tracks for every position in the Amphimedon genome ( Srivastava et al . , 2010 ) . Input DNA was used as a control for signal normalization for the histone ChIP-seq coverage . The three types of signal score statistics computed per base are as follows: ( i ) fold-enrichment ratio of ChIP-seq counts relative to expected background counts local ( FE ) ; ( ii ) negative log10 of the Poisson P-value of ChIP-seq counts relative to expected background counts local ( ppois ) ; and ( iii ) subtraction of noise from treatment sample ( subtract ) . Larva ChIP-seq data sets were analysed as described above , with the following minor modifications . Adapter contamination prior to read quality filtering was removed using Cutadapt ( RRID:SCR_011841 ) ( Martin , 2011 ) . Reads were then quality filtered using Trimmomatic v1 . 0 . 0 ( SLIDINGWINDOW: 4:15 , LEADING: 3 , TRAILING: 3 , HEADCROP: 3 , MINLEN: 20 ) ( RRID:SCR_011848 ) ( Bolger et al . , 2014 ) . Histone PTM and RNAPII regions of enrichment relative to sequenced input DNA controls were determined using MACS2 v2 . 1 . 0 ( RRID:SCR_013291 ) ( Zhang et al . , 2008 ) in broadpeak mode with a broadpeak q-value threshold of 0 . 1 and a narrowpeak threshold of 0 . 05 ( -q 0 . 05 , --broad , --nomodel , --extsize 146 , -g 1 . 45e8 ) . In both stages , chromatin states across the genome were defined using ChromHMM v1 . 10 ( Ernst and Kellis , 2012 ) , which is based on a multivariate Hidden Markov Model , using default parameters . For each ChIP-seq data set , read counts were computed in non-overlapping 200 bp bins across the Amphimedon genome ( Srivastava et al . , 2010 ) . Each bin was discretised into two levels , one indicating enrichment and 0 indicating no enrichment . The binarization was performed by comparing ChIP-seq read counts to corresponding input DNA control read counts within each bin and using a Poisson P-value threshold of 1e-4 ( the default discretization threshold in ChromHMM ) . We trained several models in parallel mode with the number of states ranging from 5 states to 15 states and chose a 9-state model as the best model that captures all the key interactions between the chromatin marks and cover all possible genomic locations ( promoter , enhancer , gene body ) that we expected to resolve given the selection of histone PTMs we used ( H3K4me3 , H3K27ac , H3K36me3 , H3K4me1 , H3K27me3 in adult; and H3K4me3 , H3K27ac , H3K4me1 , H3K27me3 in larva ) . To assign biologically meaningful mnemonics to the nine states , ChromHMM was used to compute the overlap and neighborhood enrichments of each state relative to various types of functional annotations ( Figure 1B; Figure 1—figure supplement 2; Figure 1—figure supplement 3; Figure 1—figure supplement 4 ) . State enrichment in different genomic features was calculated dividing the percentage of nucleotides occupied by a state in a particular genomic feature by the percentage of nucleotides that this genomic feature represents in the entire genome . For the overlap enrichment plots in the figures , the enrichments for each genomic feature ( column ) across all states is normalized by subtracting the minimum value from the column and then dividing by the max of the column . So , the values always range from 0 ( white ) to 1 ( dark blue ) ( i . e . , a column wise relative scale ) . For the neighborhood positional enrichment plots , the normalization is done across all columns ( i . e . , the minimum value over the entire matrix is subtracted from each value and divided by the maximum over the entire matrix ) . The functional annotations used were as follows: ( 1 ) CpG islands obtained using Hidden Markov Models as described in Wu et al . ( 2010 ) . ( 2 ) Exons , genes , introns , transcription start sites ( TSSs ) and transcription end sites ( TESs ) , 200 bp windows around TSSs and 200 bp windows around TESs based on Aqu2 . 1 gene model annotations ( Fernandez-Valverde et al . , 2015 ) . ( 3 ) Expressed and repressed genes , their TSSs and TESs . Genes were classified into expressed ( CEL-seq normalized counts > 0 . 5 ) and repressed ( CEL-seq normalized counts < 0 . 5 ) class based on their CEL-seq expression levels in the relevant stage ( larva or adult ) ( Levin et al . , 2016; Hashimshony et al . , 2012; Anavy et al . , 2014 ) . Regions of enrichment of the various histone H3 PTMs and RNAPII were overlapped with protein-coding genes and the Bioconductor R package GeneOverlap v1 . 14 . 0 ( https://www . bioconductor . org/packages/release/bioc/html/GeneOverlap . html ) was used to test and visualise their association with lists of various gene expression groups ( R Core Team , 2014 ) ( Figure 2B; Figure 2—figure supplement 2B ) . Protein-coding genes were classified into ‘high’ , ‘mid’ , ‘low’ and ‘non-expressed’ based on their CEL-seq expression levels in the relevant stage ( larva or adult ) ( Levin et al . , 2016; Hashimshony et al . , 2012; Anavy et al . , 2014 ) . Expressed genes were liberally defined as genes that had CEL-seq read counts > 0 in the relevant stage . Specifically , to define ‘high’ , ‘medium’ , ‘low’ expressed genes , all protein-coding genes expressed in the relevant stage were sorted based on CEL-seq data values and separated into three bins of an equal number of genes , similar to previous analyses ( Schwaiger et al . , 2014 ) . Enhancer elements were predicted as reliable H3K4me1 regions of enrichment , which did not overlap TSSs ( no intersection with 200 bp upstream or 200 bp downstream of the TSSs of protein-coding genes and lncRNAs ) , but overlapped with regions designated as being in an enhancer chromatin state ( ‘TxEnhA’ or ‘EnhWk’ or ‘EnhP’ state in adult; ‘TxEnhA1’ or ‘TxEnhA2’ or ‘EnhWk’ or ‘EnhP’ state in larva ) based on the ChromHMM analysis . The activated enhancer elements were predicted intersecting enhancer elements with H3K27ac significant peaks , requiring a 50% minimal overlap fraction . BEDTools v2 . 23 . 0 ( RRID:SCR_006646 ) ( Quinlan and Hall , 2010 ) was used to calculate overlaps between regions of enrichment and chromatin states with the different genomic features , as well as to identify the nearest TSS for each of the activated enhancer elements . De novo motif enrichment analyses were performed using MEME-ChIP against JASPAR CORE and UniPROBE Mouse databases ( -meme-minw 6 , -meme-maxw 15 , meme-nmotifs 20 , -dreme-e 0 . 05 , -meme-mod zoops ) ( RRID:SCR_001783 ) ( Machanick and Bailey , 2011 ) . Each motif was renamed according to their most similar motif in the TOMTOM database or literature , if any . Gene Ontology ( GO ) functional enrichment analyses were performed using the Cytoscape plugin BiNGO ( RRID:SCR_005736 ) ( Maere et al . , 2005; Shannon et al . , 2003 ) with custom annotation and a FDR adjusted P-value cut-off of 0 . 01 . All Amphimedon predicted peptides ( Fernandez-Valverde et al . , 2015 ) were annotated using BLASTp ( RRID:SCR_001010 ) ( Altschul et al . , 1990 ) ( E-value of 0 . 001 ) against the non-redundant ( nr ) NCBI protein database . All proteins were also searched for protein motifs and signal peptides using InterProScan 5 ( Jones et al . , 2014 ) with default parameters . KEGG pathway annotations were obtained on the webserver BlastKOALA for the taxonomic group ‘Animals’ against the ‘family_eukaryotes + genus_prokaryotes’ database file , using default settings . Pathway analyses were performed with the BlastKOALA annotation files using the KEGG Mapper – Reconstruct pathway tool ( Kanehisa et al . , 2016 ) . Transcription Start Site ( TSS ) input DNA-normalised coverage profiles and heatmaps were calculated using ngs . plot v2 . 61 ( RRID:SCR_011795 ) ( Shen et al . , 2014 ) and deepTools v2 . 4 . 1 ( Ramírez et al . , 2016 ) . As above , protein-coding genes were classified into ‘high’ , ‘mid’ , ‘low’ and ‘non-expressed’ based on their CEL-seq expression levels in the relevant stage ( larva or adult ) ( Levin et al . , 2016; Hashimshony et al . , 2012; Anavy et al . , 2014 ) . Expressed genes were liberally defined as genes that had CEL-seq read counts >0 in the relevant stage . Specifically , to define ‘high’ , ‘medium’ , ‘low’ expressed genes , protein-coding genes expressed in the relevant stage were sorted based on CEL-seq data values and separated into three bins of an equal number of genes , similar to previous analyses ( Schwaiger et al . , 2014 ) . Only lincRNAs found in scaffolds larger than 10 kb were used for all the analyses and , given the compact genome of Amphimedon ( Fernandez-Valverde and Degnan , 2016 ) , all the TSS analyses were restricted to non-overlapping protein-coding genes with an intergenic distance >1 kb that were found in scaffolds larger than 10 kb . All genome browser figures were generated using a local instance of the UCSC genome browser ( RRID:SCR_005780 ) ( Kuhn et al . , 2013 ) . ChIP-quantitative PCRs ( ChIP-qPCRs ) were performed using the LightCycler 480 platform ( Roche , Basil , Switzerland ) . ChIP ( H3K4me1 , H3K27ac , H3K4me3 , H3K27me3 ) and Input DNA libraries were diluted in water , combined with LightCycler 480 SYBR green I master mix ( Roche , Basil , Switzerland ) and 0 . 2 µM primers , then cycled with the following profile: 95°C for 10 min , 40 cycles of 95°C for 10 s , 60°C for 10 s , 72°C for 20 s . Primer sequences are available in Supplementary file 1 . Quantification cycle ( Cq ) values were extrapolated from manufacturers software ( version 1 . 5 . 1 . 6 . 1 SP2 ) using High Confidence settings . A melt curve and no template controls ( ntc ) were also run to ensure single amplicons were responsible for the fluorescent signal . The numerical value 3 . 32 ( log210 , representing 10% of input chromatin ) was subtracted from the Cq value of the input sample to generate the adjusted input Cq . Two different intergenic regions not bound by our histone PTMs of interest were used as negative controls . Double delta ( dd ) Cq analysis was computed ( see Figure 1—source data 5 ) . Specifically , the following formulas were used to calculate fold increase in signal over background: dCq_IP = Cq_IP - Cq_Intergenic dCq_Input = Cq_Input - Cq_intergenic ddCq = dCq_IP - dCq_Input Fold Change = 2∧ ( -ddCq ) CEL-seq raw reads were processed and mapped back to the Amphimedon genome using Bowtie ( RRID:SCR_005476 ) ( Langmead et al . , 2009 ) . We then compressed the 82 Amphimedon developmental samples , from early cleavage to adult , into 17 stages averaging the biological replicates for each developmental stage across them . Larval stages have been combined in two different groups ( Larvae 0–7 hr and Larvae 6–50 hr ) , as these developmental time points only have one replicate per time point . To reduce noise , the protein-coding genes and long non-coding RNAs with an overall expression of less than 100 CEL-seq raw counts throughout the whole developmental time course were discarded . The CEL-seq raw gene counts were then normalized using variance stabilizing transformation in DEseq2 1 . 6 . 3 ( RRID:SCR_000154 ) ( Love et al . , 2014 ) and the 15 , 000 most variable genes ( 14 , 698 protein coding genes + 301 lncRNAs ) were extracted using median absolute deviation . The 14 , 698 protein-coding genes were then filtered to retain only non-overlapping protein-coding genes with detectable expression at adult stage ( CEL-seq normalized counts > 0 ) with an intergenic distance >1 kb that were found in scaffolds larger than 10 kb . This resulted in a total number of 3 , 200 ‘high-variance’ genes . The remaining expressed ( CEL-seq normalized counts > 0 in adult ) non-overlapping protein-coding genes with an intergenic distance >1 kb that were found in scaffolds larger than 10 kb were considered ‘low-variance’ genes ( n = 3 , 999 ) . To define low , medium and high , the 3 , 200 high-variance genes and 3 , 999 low-variance genes were sorted based on CELseq data values and separated into three bins of an equal number of genes . CEL-seq raw reads were processed and mapped back to the Amphimedon genome using Bowtie ( RRID:SCR_005476 ) ( Langmead et al . , 2009 ) . Read counts were normalized by dividing by the total number of counted reads and multiplying by 106 . We then compressed the 82 Amphimedon developmental samples , from early cleavage to adult , into 17 stages averaging the biological replicates for each developmental stage across them . Larval stages have been combined in two different groups ( Larvae 0–7 hr and Larvae 6–50 hr ) , as these developmental time points only have one replicate per time point . To reduce noise , only the protein-coding genes with an expression of at least four CEL-seq normalised counts in at least two developmental time points were retained . To define the transcriptional stability of protein-coding genes , the coefficient of variation of gene expression was calculated for each protein-coding gene ( n = 15 , 146 ) , as reported by Pérez-Lluch et al . ( 2015a ) . For the TSS input DNA-normalised coverage plots , these 15 , 146 protein-coding genes were then filtered to retain only expressed ( CEL-seq normalized counts > 0 in the relevant stage [larva or adult] ) non-overlapping protein-coding genes with an intergenic distance >1 kb that were found in scaffolds larger than 10 kb . Finally , from the full ranking of these expressed protein-coding genes , we defined the bottom 1 , 000 genes with the lowest variation in expression during development as ‘stable’ genes and the top 1 , 000 genes with the highest variation in expression as strongly developmentally ‘regulated’ genes . Available ChIP-seq data sets on adult female polyps for H3K4me3 and corresponding input DNA controls were used ( Schwaiger et al . , 2014 ) . Aligned ChIP and matching input DNA data sets and developmentally stable and regulated genes were generated using the same procedures as in the sponge ( see above ) . To obtain gene and transcript quantifications , we mapped available RNA-seq data sets ( Helm et al . , 2013 ) to NveGenes2 . 0 gene models ( http://www . cnidariangenomes . org/ ) using kallisto ( Bray et al . , 2016 ) . Orthologs of Drosophila PcG components and associated factors were identified using BLASTp ( RRID:SCR_001010 ) ( Altschul et al . , 1990 ) searches against the predicted proteomes of the selected species ( Figure 4—source data 1 ) with a threshold E-value of 0 . 001 and taking a maximum of 5 hits per species . All the obtained protein hits were aligned using MAFFT with L-INS-i mode ( RRID:SCR_011811 ) ( Katoh and Standley , 2013 ) . The alignments were automatically trimmed with trimAl v1 . 4 ( 151 ) in -automated1 mode . Resulting trimmed alignments were then used for phylogenetic inference using FastTree2 ( Price et al . , 2010 ) with -wag -cat 8 -gamma parameters . The phylogenetic trees were inspected manually to discriminate which BLASTp hits formed monophyletic clades with the Drosophila query sequences . The same methodology was used to identify the conserved ancestral microsyntenic pairs taken from Irimia et al . ( 2012 ) , but using Homo sapiens sequences as query proteins . The phylogeny-validated Amphimedon ortholog pairs were manually checked for contiguity in the genome and those found in different scaffolds or with more than two intervening genes were removed from the subsequent analyses . Amphimedon ChIP-seq data sets have been deposited to the NCBI Gene Expression Omnibus ( GEO ) ( RRID:SCR_007303 ) ( Edgar et al . , 2002 ) under accession number GSE79645 . Amphimedon genome assembly ampQue1 was used throughout the study . CEL-seq data sets can be obtained from NCBI GEO ( GSE54364 ) ( Anavy et al . , 2014 ) . Amphimedon RNA-seq data sets can be downloaded at NCBI's SRA ( RRID:SCR_004891 ) with accession SRP044247 ( Fernandez-Valverde et al . , 2015 ) . Nematostella vectensis RNA-seq data sets can be downloaded at NCBI's SRA with accession SRP018739 ( Helm et al . , 2013 ) . N . vectensis ChIP-seq data sets can be obtained from NCBI GEO ( GSE46488 ) ( Schwaiger et al . , 2014 ) . We used the following gene model data sets for all analyses . A . queenslandica: Aqu2 . 1 models ( http://amphimedon . qcloud . qcif . edu . au/ ) ( last accessed February 25 , 2017 ) ( Fernandez-Valverde et al . , 2015 ) , lncRNAs ( http://amphimedon . qcloud . qcif . edu . au/lncRNAs/ ) ( last accessed February 25 , 2017 ) ( Gaiti et al . , 2015 ) ; N . vectensis: NveGenes2 . 0 models ( http://www . cnidariangenomes . org/ ) ( last accessed February 25 , 2017 ) . | Animals come in many shapes and sizes , and vary in how they move , grow and reproduce . The long-held thought that animal complexity is related to the number of genes that are in the animal’s DNA has now been largely dismissed; simple animals like sponges and cnidarians ( for example , jellyfish , anemones and corals ) have comparable gene numbers to vertebrates , insects , mollusks and other complicated bilaterians ( animals that feature a plane of symmetry , meaning that they have a top , a bottom , a front and a back ) . This observation led to the idea that gene regulation ( how and when genes are turned off and on ) is responsible for the evolution of animal diversity . Genomic DNA packs into cells by winding around proteins called histones . Histones themselves can bear certain chemical marks , which in turn determine if the genes contained in the DNA associated with the histones are going to be turned on or off . In bilaterians and cnidarians these marks substantially contribute to gene regulation . Some of these marks predate the evolution of multicellular animals from single-celled organisms . However , the origin of the marks that associate with the gene regulatory elements that are essential for animals to be multicellular remained unknown . In other words , does the evolution of histone marks underpin animal complexity ? Gaiti et al . turned to the marine sponge Amphimedon queenslandica to address this question . Sponges are one of the morphologically simplest animals , lacking a gut , nerves and muscles . By analyzing histone marks in this sponge , Gaiti et al . found they were remarkably similar to the networks of histone marks seen in more complex animals . This is consistent with this form of gene regulation being present at the dawn of the animal kingdom . Indeed , this mode of gene regulation may have been necessary for multicellular animals to first evolve . It now appears that most of the genes and regulatory mechanisms underlying the formation of complex animals , like ourselves , had an unexpected early origin – probably as early as the first steps in the evolution of multicellular animals from single-celled organisms . Further studies of animals that are close relatives of sponges , such as comb jellies , and their single-celled cousins , may further improve our understanding of how these simple single-celled organisms became multicellular animals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"evolutionary",
"biology"
] | 2017 | Landscape of histone modifications in a sponge reveals the origin of animal cis-regulatory complexity |
TFIID—a complex of TATA-binding protein ( TBP ) and TBP-associated factors ( TAFs ) —is a central component of the Pol II promoter recognition apparatus . Recent studies have revealed significant downregulation of TFIID subunits in terminally differentiated myocytes , hepatocytes and adipocytes . Here , we report that TBP protein levels are tightly regulated by the ubiquitin-proteasome system . Using an in vitro ubiquitination assay coupled with biochemical fractionation , we identified Huwe1 as an E3 ligase targeting TBP for K48-linked ubiquitination and proteasome-mediated degradation . Upregulation of Huwe1 expression during myogenesis induces TBP degradation and myotube differentiation . We found that Huwe1 activity on TBP is antagonized by the deubiquitinase USP10 , which protects TBP from degradation . Thus , modulating the levels of both Huwe1 and USP10 appears to fine-tune the requisite degradation of TBP during myogenesis . Together , our study unmasks a previously unknown interplay between an E3 ligase and a deubiquitinating enzyme regulating TBP levels during cellular differentiation .
The TATA-box binding protein ( TBP ) is one of the central players in eukaryotic transcription . TBP serves as a key subunit to facilitate transcription initiation by all three RNA polymerases in eukaryotic cells . As a part of the SL1 complex , TBP plays a role in recognizing Pol I-transcribed promoters ( Comai et al . , 1994 ) . Similarly , TBP and its associated factors ( TAFs ) that make up the TFIID complex are specific for Pol II-mediated mRNA transcription ( Dynlacht et al . , 1991 ) . Likewise , the TBP/B/BRF complex TFIIIB drives the transcription of small nuclear RNAs by Pol III ( Taggart et al . , 1992 ) . Given this pivotal role in transcription , TBP-mediated transcriptional regulation has been extensively studied by biochemical and genetic approaches in past decades ( Hernandez , 1993 ) . However , only limited studies have interrogated the regulation of TBP during cell-type specification . The main reason is that cell-type specific transcription programs have generally been considered to be dictated by classic sequence-specific transcription factors ( Farnham , 2009 ) , while the components of the core promoter recognition machinery such as TBP were thought to be largely invariant across different cell types ( Thomas and Chiang , 2006 ) . Recently , studies demonstrated that during a number of terminal differentiation processes , TBP protein levels become dramatically downregulated . Specifically , in terminally differentiated myocytes , hepatocytes and adipocytes , normally high concentrations of TBP protein as well as the canonical components of the TFIID complex become severely reduced while other core pre-initiation complex components such as RNA Pol II remain largely unaffected ( Deato and Tjian , 2007; D'Alessio et al . , 2011; Zhou et al . , 2013 ) . In conjunction with TBP downregulation , various cell-type specific ‘orphan’ TAFs are switched on to help direct developmental gene expression and terminal differentiation program ( D'Alessio et al . , 2009; Goodrich and Tjian , 2010 ) . Interestingly , it was found that reductions in TBP protein levels are much more pronounced than decreases in its mRNA levels , suggesting that an important level of TBP regulation occurs post-transcriptionally ( Deato and Tjian , 2007; D'Alessio et al . , 2011; Liu et al . , 2011; Zhou et al . , 2013; Herrera et al . , 2014 ) . We therefore set out to decipher the mechanisms by which TBP protein levels are regulated during differentiation . Here , we show that TBP is a substrate for ubiquitination by a specific E3 ligase in vivo and in vitro . By using a combination of in vitro assays and biochemical fractionation , we identified the 480 kDa Huwe1 protein as a major E3 ligase targeting TBP . We confirmed in vitro that purified recombinant Huwe1 targets TBP for ubiquitination , while loss-of-function experiments allowed us to probe TBP regulation by the E3 ligase in vivo . We also identified USP10 as a ubiquitin-specific protease ( USP ) that modulates Huwe1-mediated TBP ubiquitination and protects TBP from proteasome-mediated degradation . Importantly , we were able to show that the E3 ligase becomes significantly upregulated during myotube differentiation , while its paired USP becomes dramatically downregulated . Together , our results support a model in which coordinated ubiquitination and deubiquitination activities finely balance TBP protein levels during terminal differentiation .
Consistent with several earlier observations from our lab and other studies ( Perletti et al . , 1999; Deato and Tjian , 2007 ) , we found that upon myogenic differentiation of C2C12 cells TBP protein levels become significantly downregulated , while its mRNA levels remained largely unchanged ( Figure 1—figure supplement 1 ) , suggesting that TBP downregulation mainly occurs via modulation of its protein levels . Since the major protein degradation pathway in eukaryotic cells is mediated by the ubiquitin/proteasome system ( UPS ) ( Hochstrasser , 1996; Hershko and Ciechanover , 1998 ) , we treated post-mitotic myotubes with the proteasome inhibitor MG132 and checked whether TBP downregulation was affected . Strikingly , TBP protein levels were partially restored within 4 hr after MG132 treatment ( Figure 1A ) , suggesting that the UPS likely plays a dominant role in regulating TBP levels during terminal differentiation . 10 . 7554/eLife . 08536 . 003Figure 1 . The ubiquitin-proteasome system regulates TBP protein levels . ( A ) Treatment of differentiated C2C12 myotubes with the proteasome inhibitor MG132 rescues TBP protein levels . Western blot analysis of cytoplasmic and nuclear extracts from C2C12 myoblasts and myotubes was done using antibodies against TBP and myogenin . Loading control was done by Ponceau S staining . Myotubes were treated as indicated on differentiation day 6 . MG132 was used at a final concentration of 10 μM for 4 hr * , a modified version of myogenin was detected upon MG132 treatment . ( B ) TBP is ubiquitinated in vivo . 293T cells were transfected with indicated plasmids . Ubiquitin conjugates were purified using Ni-NTA resin under denatured conditions from MG132 treated cells , then subjected to western blot analysis using anti-TBP antibody . ( C ) TBP can be ubiquitinated by both C2C12 myoblast and myotube whole cell extracts in vitro . In vitro transcribed and translated HA-TBP was incubated with whole cell extracts from C2C12 myoblasts or myotubes for in vitro ubiquitination assays . After the reaction , immunopurified HA-TBP ( purified using an antibody against HA-tag ) was analyzed by western blot using an antibody against ubiquitin . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 00310 . 7554/eLife . 08536 . 004Figure 1—figure supplement 1 . Protein levels of TBP and TAF4 but not RNA polymerase II significantly decrease during C2C12 differentiation . Western blot analysis of whole-cell extracts prepared from C2C12 myoblasts ( D0 ) and myotubes collected on day 3 ( D3 ) and day 6 ( D6 ) post-differentiation was done using antibodies against TAF4 , TBP , myogenin and the carboxy terminal domain ( CTD ) of the large RNA polymerase II ( Pol II ) subunit RPB1 . Signals in Ponceau S staining were used as loading controls . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 004 To test whether TBP is indeed a substrate for ubiquitination , we next performed in vivo ubiquitination assays , in which we treated 293T cells expressing His-ubiquitin and Homologous to the E6-AP Carboxyl Terminus ( HA-TBP ) with MG132 overnight . Subsequently , all ubiquitinated protein species in the cells were pulled down under denaturing conditions by using Nickel resin ( See details in the ‘Materials and methods’ ) . Western blot analysis confirmed the presence of slower-migrating poly-ubiquitinated species of the TBP protein in the pull-downs , suggesting TBP is indeed ubiquitinated in cells ( Figure 1B ) . Ubiquitination is catalyzed by a cascade of three different enzymes , E1 activation enzyme , E2 conjugation enzymes and E3 ligases . The target specificity of the system is largely determined by E3 ligases , that recognize specific substrates for ubiquitination ( Nakayama and Nakayama , 2006 ) . To identify potential E3 ligases that specifically target TBP for ubiquitination , we used biochemical fractionation followed by an in vitro ubiquitination assay . As a first step towards establishing an in vitro biochemical assay for TBP ubiquitination , we incubated in vitro translated HA-tagged TBP protein with either myoblast or myotube lysates supplemented with proteasome inhibitor MG132 , a general deubiquitinase inhibitor ( ubiquitin aldehyde ) , ubiquitin and an energy regenerating mix ( See details in the ‘Materials and methods’ ) . After 90 min of reaction at 37°C , we purified the substrate protein by anti-HA antibody affinity chromatography and analyzed TBP ubiquitination by anti-ubiquitin western blots . Higher molecular weight ubiquitinated TBP species were detected in reactions with both myoblast and myotube lysates , suggesting TBP ubiquitination activities are present in both proliferating muscle progenitors and terminally differentiated muscle cells ( Figure 1C ) . Since it is impractical to scale up myoblast and myotube cultures to obtain sufficient materials for bulk biochemical fractionation , we then tested whether TBP can be ubiquitinated by lysates of Hela cells . We detected significant TBP ubiquitination activities in Hela whole cell extracts ( Figure 2A ) . We then separated the whole cell extract into cytoplasmic and nuclear fractions , and found that most of the TBP ubiquitination activities were present in the cytoplasmic fraction ( S100 ) ( Figure 2A ) . Using S100 as starting material , we designed a purification scheme for isolating putative TBP E3 ligases based on the in vitro ubiquitination assay . To simplify the ubiquitination assay , we used recombinant GST-tagged TBP as the substrate , which allows efficient isolation of the ubiquitinated TBP species with GST resin . It is worth noting that while GST-TBP can be efficiently ubiquitinated in the reaction , we did not detect any ubiquitination of the GST protein ( Figure 2—figure supplement 1A ) , suggesting that these ubiquitination activities are specific to TBP . We could also detect ubiquitinated TBP species as slower-migrating bands by anti-TBP western blots ( Figure 2—figure supplement 1B ) . 10 . 7554/eLife . 08536 . 005Figure 2 . Purification of TBP E3 ligase from HeLa cytoplasmic fraction ( S100 ) . ( A ) In vitro ubiquitination of TBP by HeLa cell lysates . In vitro transcribed and translated HA-TBP was incubated with indicated HeLa cellular fractions for ubiquitination assays . After the reaction , immunopurified HA-TBP was analyzed western blot using an antibody against ubiquitin . ( B ) Fractions eluted from the ion exchange ( D52 ) column were assayed in the presence of E1 ( UBE1 ) , E2 ( UbcH5b ) , ubiquitin , ubiquitin aldehyde ( deubiquitinase inhibitor ) and MG132 ( proteasome inhibitor ) . TBP ubiquitination was analyzed by western blots using anti-TBP antibody . ( C ) Chromatography scheme of purification of TBP E3 ligase . Hela cytoplasmic fraction ( S100 ) was subjected to a series of chromatographic fractionation as indicated . ( D ) TBP E3 ligase migrates at a size around 440–660 kDa . Input ( In ) and Superose 6 fractions were assayed as in ( B ) . Motilities of the peak activity ( 440–660 kDa ) are shown ( bottom ) . ( E ) TBP E3 ligase activity after the final Mono Q chromatography step . Reactions containing Input ( In ) and Mono Q fractions were performed as in ( B ) . ( F ) The abundance of Huwe1 protein positively correlates with levels of TBP E3 ligase activity within selected MonoQ fractions . Protein compositions of selected MonoQ fractions were determined by Mudpit ( Multidimensional Protein Identification Technology ) . Shown are numbers of detected peptides and percentage of coverage for Huwe1 protein in indicated fractions . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 00510 . 7554/eLife . 08536 . 006Figure 2—figure supplement 1 . GST-TBP can be used as a substrate for in vitro ubiquitination assays . ( A ) GST-TBP but not GST is ubiquitinated by HeLa S100 cytoplasmic fraction in vitro . Recombinant GST-TBP and GST proteins ( right ) were incubated with Hela S100 fraction for in vitro ubiquitination assay . After the reaction , proteins were purified using GST beads and analyzed by western blot using an antibody against ubiquitin ( left ) . ( B ) Ubiquitinated species of TBP can be detected by anti-TBP antibody as slow-migrating bands in western blot analysis . GST-TBP was incubated with in vitro ubiquitination reaction mixture in the presence ( + ) and absence ( − ) of S100 . After the reaction , purified GST-TBP protein was analyzed by western blot using an anti-TBP antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 006 To purify E3 ligases , we first fractionated the S100 lysate on an ion exchange column ( D52 ) into multiple fractions: flowthrough , proteins bound to the column and eluted at 0 . 2 M KCl , 0 . 3 M KCl and 0 . 5 M KCl , respectively . None of the fractions alone was sufficient to mediate TBP ubiquitination . Reasoning that E1 , E2 and E3 activities might be eluted at different salt concentrations , we then supplemented each fraction with recombinant E1 and a panel of E2s , and found that the 300 mM elution supplemented with several E2s ( UbcH5 family E2s and UbcH7 ) robustly catalyzed TBP ubiquitination ( Figure 2B ) . We then precipitated the 300 mM D52 eluate with ammonium sulfate and further fractionate it by Butyl Sepharose , Superose 6 gel filtration , Heparin and finally MonoQ columns to enrich for E3 ligases responsible for TBP ubiquitination ( Figure 2C ) ( See details in the ‘Materials and methods’ ) . Ubiquitination assays on the activities eluted from the Superose 6 sizing column suggest the molecular weight of the E3 ligase ( s ) to be about 440–660 kDa ( Figure 2D ) . The E3 activity was eluted from MonoQ at roughly 400 mM KCl ( Figure 2E ) . At this stage , the limited amount of remaining material prevented further purification so the sample was subjected to Mudpit ( Multidimensional Protein Identification Technology ) mass spectrometry analysis along with control fractions containing little or no detectable levels of TBP E3 activities . Of the three E3 ligases detected by Mudpit analysis , Huwe1 emerged as the most promising candidate , as its abundance correlated very well with TBP ubiquitination activities within each fraction ( Figure 2F ) . Specifically , a total of 81 Huwe1 peptides ( 18 . 4% coverage ) were detected in fractions with the highest E3 activities , and this number fell to 30 and 14 peptides in the fractions with intermediate and low E3 activities , respectively . Moreover , Huwe1 protein consists of 4374 amino acids with a molecular mass of around 480 kDa ( Chen et al . , 2005; Zhong et al . , 2005 ) , consistent with the molecular weight predicted for the TBP E3 ligase activity after Superose 6 gel filtration ( Figure 2D ) . To test whether Huwe1 indeed mediates TBP ubiquitination , we performed in vitro ubiquitination assays with purified recombinant Huwe1 proteins ( Figure 3—figure supplement 1A ) , which resulted in a dose-dependent poly-ubiquitination of TBP detected as high molecular weight species using anti-TBP antibody ( Figure 3A ) . As expected , the TBP ubiquitination activity of the purified recombinant Huwe1 protein was much more robust than that of partially purified fractions , presumably due to higher Huwe1 concentrations and/or the absence of potential inhibitors of ubiquitin polymerization . When using a methyl ubiquitin that only supports monoubiquitination together with purified recombinant Huwe1 in our assay ( Hershko and Heller , 1985 ) , we detected a ladder of shifted bands separated by around 8 kDa similar to the TBP ubiquitination patterns generated by the partially purified Mono Q fractions ( Figure 3—figure supplement 1B ) . Since these shifted bands likely correspond to different ubiquitination sites , the use of methyl ubiquitin also allowed us to estimate the number of distinct ubiquitination sites on TBP to be more than 10 ( Figure 3—figure supplement 1B ) . 10 . 7554/eLife . 08536 . 007Figure 3 . In vitro ubiquitination of TBP using recombinant Huwe1 protein . ( A ) Recombinant Huwe1 protein ubiquitinates TBP in vitro . Titrations ( twofold concentration range ) of recombinant His-tag Huwe1 protein are incubated with GST-TBP supplemented with E1 , UbcH5b , wild-type ubiquitin and ATP regenerating mix , ubiquitin aldehyde in in vitro ubiquitination assays . After the reaction , TBP ubiquitination was analyzed through western blot using the anti-TBP antibody . ( B ) Hect domain of Huwe1 is required for its TBP ubiquitination activity . Wildtype ( WT ) , Hect domain truncation ( ΔC ) and catalytic site mutant ( C → S ) Huwe1 were assayed as in ( A ) to test their activities to ubiquitinate TBP . ( C ) Hect domain alone is insufficient for Huwe1's TBP ubiquitination activity . Two different doses of His-tag Huwe1 ( twofold range ) and recombinant E6AP protein ( twofold range ) are used in the ubiquitination assay as in ( A ) . ( D ) Huwe1 E3 activity can be supported by UbcH5 family E2s and UbcH7 . A panel of different E2 conjugating enzymes is used in the in vitro ubiquitination assays . ( E ) Huwe1 mediated the K48-linked ubiquitination of TBP . Wild-type ubiquitin ( WT ) , lysine 11 to arginine mutant ( K11R ) , lysine48 to arginine mutant ( K48R ) , lysine63 to arginine mutant ( K63R ) and lysine-methylated ubiquitin ( Methyl-Ub ) are used in the ubiquitination assays . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 00710 . 7554/eLife . 08536 . 008Figure 3—figure supplement 1 . Huwe1 mediates TBP ubiquitination at multiple sites . ( A ) PageBlue SDS-PAGE Gel of purified recombinant wild-type ( WT ) , catalytic domain truncation ( ΔC ) and catalytic site C4341S mutant ( C → S ) Huwe1 proteins are shown . HiMark pre-stained protein marker ( Invitrogen , highest molecular weight 460 kDa ) was used to ensure the purification of full-length proteins . ( B ) Huwe1 ubiquitinates TBP in a dose-dependent manner . Titrations ( twofold concentration range ) of recombinant His-tag Huwe1 protein were incubated with GST-TBP supplemented with E1 , UbcH5b , methyl-ubiquitin , ATP regenerating mix , and ubiquitin aldehyde . After the reaction , TBP ubiquitination was analyzed through western blot using an anti-TBP antibody . ( C ) Hect domain of Huwe1 is required for the TBP ubiquitination activity . Wildtype ( WT ) , Hect domain truncated ( ΔC ) and catalytic site C4341S ( C → S ) Huwe1 mutants were used in the ubiquitination assay as in ( B ) . ( D ) Hect domain alone is insufficient for Huwe1's TBP ubiquitination activity . Two different doses of His-tag Huwe1 ( over twofold range ) and Hect-domain containing E3 E6AP ( over twofold range ) were used in the ubiquitination assay . ( E ) Huwe1 E3 activity can be activated by UbcH5 family E2s and UbcH7 . A panel of different E2 conjugating enzymes were tested by the in vitro ubiquitination assays . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 00810 . 7554/eLife . 08536 . 009Figure 3—figure supplement 2 . Huwe1 directly interacts with TBP in vitro . GST-tagged full length ( FL ) TBP , TBP C-terminal DNA binding domain ( DBD ) and N-terminal transactivation domain ( TAD ) were used to pull-down purified recombinant Huwe1 protein . Pull down materials were analyzed by western blot using anti-Huwe1 ( upper panel ) and anti-GST ( lower panel ) antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 009 To confirm that this ubiquitination reaction requires direct physical interaction between TBP and Huwe1 , we tested whether these two proteins coprecipitate through GST pull down experiments . Indeed , purified His-Huwe1 bound to full-length TBP protein ( Figure 3—figure supplement 2 ) . Moreover , this interaction is dependent on the C-terminal DNA binding domain of TBP , which is a highly conserved and highly structured region ( Figure 3—figure supplement 2 ) . Previous studies showed that Huwe1 contains a HECT ( Homologous to the E6-AP Carboxyl Terminus ) domain , which accepts ubiquitin from an E2 ubiquitin-conjugating enzyme in the form of a thioester , and then directly transfers the ubiquitin to targeted substrates ( Rotin and Kumar , 2009 ) . To confirm that the HECT domain of Huwe1 is required for its TBP ubiquitination activity , we generated two mutants of Huwe1: one with the HECT domain truncated ( Δ4341–4374 ) , the other with the active site cysteine residue mutated to serine . Both mutant proteins were purified through the same procedures as the wild type protein ( Figure 3—figure supplement 1A ) . Neither of the mutants was able to mediate TBP ubiquitination in vitro ( Figure 3B , Figure 3—figure supplement 1C ) . Apparently , the N-terminal region of Huwe1 outside the catalytic domain is also required for efficient TBP ubiquitination , as the homologous HECT domain from E3 ligase E6AP was unable to carry out TBP ubiquitination ( Figure 3C , Figure 3—figure supplement 1D ) . Having established that Huwe1 mediates TBP ubiquitination in vitro , we then asked which E2 conjugating enzymes work best with Huwe1 in this reaction . Many recent studies have revealed the importance of E2 conjugating enzymes in determining the length and topology of ubiquitin chains and hence the cellular consequences of ubiquitination ( Ye and Rape , 2009 ) . We therefore tested the ability of a panel of E2s to collaborate with Huwe1 in mediating TBP ubiquitination , and found that both the UbcH5 family E2s and UbcH7 are active in the in vitro ubiquitination assay with recombinant purified Huwe1 ( Figure 3D , Figure 3—figure supplement 1E ) , consistent with the E2 enzymes that worked with the crude 0 . 3 M KCl fraction from the D52 column . We also asked whether TBP ubiquitination has specific poly-ubiquitin chain topologies . It was shown that poly-ubiquitin chains on a substrate can be formed using any of the seven lysines in ubiquitin , and specific ubiquitin chain topologies are linked to distinct biological pathways ( Komander and Rape , 2012 ) . To map the topology , we performed in vitro TBP ubiquitination assays using ubiquitin mutants with individual lysines mutated to arginines . Strikingly , when K48R ubiquitin was used , poly-ubiquitinated TBP species were significantly diminished ( Figure 3E , lane 4 ) , suggesting that Huwe1 likely mediates K48-linked ubiquitin chain formation on TBP . Ubiquitin chains of this topology usually lead to proteasome-mediated degradation ( Komander and Rape , 2012 ) . Having established that Huwe1 supports TBP K48 linkage ubiquitination in vitro , we next tested whether Huwe1 regulates TBP protein levels in vivo . Knocking-down Huwe1 in 293T cells with two different shRNAs resulted in increased levels of endogenous TBP protein ( Figure 4A ) . To determine relative TBP protein stability more quantitatively in cells , we developed an in vivo TBP half-life measurement system based on a fluorescence timer assay ( Khmelinskii et al . , 2012 ) . In this assay , TBP is fused to both a monomeric red fluorescent protein ( mCherry ) , which takes hours to complete its maturation , and a monomeric superfolder green fluorescent protein ( sfGFP ) , which becomes fluorescent within minutes after synthesis . Due to the slow maturation rate of mCherry relative to sfGFP , the mCherry/sfGFP intensity ratio at steady state correlates with the half-life of the fusion protein , independent of the protein synthesis rate ( Figure 4—figure supplement 1A ) . We co-transfected 293T cells with plasmids expressing Huwe1 shRNAs and TBP-mCherry-sfGFP fusion protein . Knockdown of Huwe1 by 2 independent shRNAs led to significant increases in mCherry/GFP intensity ratios , evident in both imaging and flow cytometry analysis ( Figure 4B , Figure 4—figure supplement 1B ) . Similarly , lentivirus-mediated Huwe1 shRNA knockdown in C2C12 myoblasts leads to elevated endogenous TBP protein levels ( Figure 5A ) . We further tested how Huwe1 affects TBP protein half-life by cycloheximide ( CHX ) chase experiments . Consistent with results from fluorescent timer experiments , Huwe1 knockdown by two independent shRNAs both increased TBP protein half-life in myoblasts ( Figure 5—figure supplement 1 ) . Taken together , our results strongly suggest that Huwe1 modulates TBP protein levels in living cells . 10 . 7554/eLife . 08536 . 010Figure 4 . Huwe1 regulates TBP protein levels in proliferating cells . ( A ) Huwe1 knockdown increases TBP protein levels in 293T cells . Western blot analysis of whole cell extracts from control ( NT ) and Huwe1 knockdown 293T cells was done using antibodies against Huwe1 and TBP . In this assay , Tubulin signals were used as loading controls . ( B ) Huwe1 knockdown increases protein half-life of TBP-fluorescent-timer fusion protein . Fluorescent images of 293T cells cotransfected with a plasmid expressing the TBP-fluorescent-timer fusion protein and control pLKO . 1 ( NT ) or two different pLKO . 1 plasmids expressing shRNA against Huwe1 ( Huwe1 KD#1 and Huwe1 KD#2 ) are taken using the same exposure time , then processed in parallel with ImageJ . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 01010 . 7554/eLife . 08536 . 011Figure 4—figure supplement 1 . Huwe1 knockdown prolongs half-life of fluorescent-timer TBP fusion protein in 273T cells as measured by fluoresent-timer assays . ( A ) Schematic representation of the principle for fluorescent timer assay . ( B ) Fluorescent-timer TBP fusion protein exhibits higher mCherry/GFP ratio in Huwe1 knockdown 293T cells . 293T cells were cotransfected with a plasmid expressing the TBP-fluorescent-timer fusion protein and control pLKO . 1 ( NT ) or pLKO . 1 plasmid expressing shRNA against Huwe1 ( Huwe1 KD#1 or Huwe1 KD#2 ) . mCherry/GFP signal ratios in single cells were measured using DB FACSAria III Cell Sorter . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 01110 . 7554/eLife . 08536 . 012Figure 5 . Huwe1 regulates TBP protein level in myoblasts and myotubes . ( A ) Lentivirus-mediated depletion of Huwe1 elevates TBP protein level in C2C12 myotubes . Cytoplasmic and nuclear extracts of C2C12 myoblasts infected with control ( NT ) lentiviruses or two different lentiviruses targeting Huwe1 ( 1 , 2 ) were analyzed by western blots using antibodies against Huwe1 and TBP . Signals in Ponceus S staining were used as loading controls . ( B ) Up-regulation of Huwe1 protein levels during C2C12 differentiation . Cytoplasm and nuclear extracts of proliferating C2C12 myoblasts ( D0 ) and differentiated C2C12 myotubes ( D6 ) were analyzed by western blots using antibodies against Huwe1 and TBP . Signals in Ponceus S staining were used as loading controls . Results from two biological replicates are shown . ( C ) Huwe1 knockdown in differentiated myotubes partly rescues TBP protein levels . Cytoplasmic and nuclear extracts of differentiation day 8 myotubes infected with either control lentiviruses ( NT ) or two different lentiviruses targeting Huwe1 ( 1 , 2 ) were analyzed by western blots using antibodies against Huwe1 and TBP . Signals in Ponceus S staining were used as loading controls . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 01210 . 7554/eLife . 08536 . 013Figure 5—figure supplement 1 . Huwe1 knockdown increases TBP protein half-life in C2C12 myoblasts . ( A ) Control and Huwe1 knockdown C2C12 myoblasts were treated with 50 μg/ml cycloheximide . Cells were collected at indicated time points and analyzed by western blot using anti-TBP and anti-Tubulin ( control ) antibody . ( B ) Quantification of western blots in ( A ) using ImageJ . Signal intensities were normalized to 0 time point in each blots . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 013 Although our results indicated that Huwe1 regulates TBP protein levels in vivo , it remained unclear what causes the dramatic downregulation of TBP protein levels upon myogenesis . One possibility is that Huwe1 E3 ligase activities are upregulated during muscle differentiation . To test this hypothesis , we examined Huwe1 protein levels in myoblasts and myotubes . Strikingly , in contrast to dramatic decreases in TBP protein levels , Huwe1 protein levels are significantly upregulated in myotubes ( Figure 5B ) . This observation is consistent with previously published results showing that Huwe1 transcripts are enriched in muscle tissues ( Schwarz et al . , 1998; Chen et al . , 2005 ) . To confirm that Huwe1 is indeed responsible for active TBP degradation in myotubes , we treated myotubes with high titer viruses expressing shRNAs against Huwe1 on differentiation Day 4 . This treatment resulted in abnormal morphologies of the polynucleated myotubes ( Figure 6D ) , accompanied by increased TBP protein levels ( Figure 5C ) . These results support a model wherein the upregulation of Huwe1 protein levels during myogenesis likely leads to enhanced TBP ubiquitination and subsequent degradation of TBP protein . 10 . 7554/eLife . 08536 . 014Figure 6 . Huwe1 is required for muscle differentiation and maintenance of normal muscle morphologies . ( A ) Huwe1 knockdown impaired C2C12 differentiation efficiency . Shown are phase contrast images of control and Huwe1 knockdown C2C12 myoblasts after 6 days of differentiation in 2% horse serum . ( B ) Muscle marker genes were not efficiently induced in Huwe1 knockdown C2C12 cells after 6 days of differentiation . Differential gene expression analysis was used to characterize the effects of Huwe1 knockdown on muscle differentiation using mRNA-seq data . The log2 scale values of fragments per kilobase per million ( FPKM ) for genes in both control and Huwe1 knockdown samples are plotted in the graph . Dark green , twofold down; light green , eightfold down; orange , twofold up; red , eightfold up . ( C ) Goseq gene ontology analysis of genes down-regulated in Huwe1 knockdown samples . The list shows top categories ranked by p-value . ( D ) Knockdown of Huwe1 in differentiated muscle cells affected normal muscle morphology . Shown are phase contrast images of control and Huwe1 knockdown C2C12 myotubes . ( E ) Huwe1 knockdown in terminally differentiated myotubes reduced transcription levels of muscle marker genes . Differential gene expression analysis to characterize the effects of Huwe1 knockdown in terminally differentiated muscle cells using mRNA-seq data . Figure was labelled the same way as ( B ) . ( F ) Goseq gene ontology analysis of genes down regulated by Huwe1 knockdown in myotubes . List showes top categories ranked by p-value . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 01410 . 7554/eLife . 08536 . 015Figure 6—source data 1 . Differentiation gene expression analysis of control and Huwe1 knockdown C2C12 cells on differentiation Day 6 by RNA-seq . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 01510 . 7554/eLife . 08536 . 016Figure 6—source data 2 . Differentiation gene expression analysis of control and Huwe1 knockdown myotubes on differentiation Day 8 by RNA-seq . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 01610 . 7554/eLife . 08536 . 017Figure 6—source data 3 . Differentiation gene expression analysis of control and Huwe1 knockdown C2C12 cells on differentiation Day 0 by RNA-seq . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 01710 . 7554/eLife . 08536 . 018Figure 6—figure supplement 1 . Huwe1 knockdown causes modest changes in the transcriptional program of C2C12 myoblasts . ( A ) Huwe1 knockdown C2C12 myoblasts did not exhibit significant morphological differences compared to control C2C12 s when cultured in proliferation medium ( 10% FBS ) . ( B ) Huwe1 knockdown results in modest changes in the transcriptional program of C2C12 myoblasts . Differential gene expression analysis was used to characterize the effects of Huwe1 knockdown in proliferating myoblasts using mRNA-seq data . The log2 scale values of fragments per kilobase per million ( FPKM ) for genes in both WT and Huwe1 knockdown samples are plotted in the graph . Dark gene , twofold down; light green , eightfold down; orange , twofold up; red , eightfold up . ( C ) Goseq gene ontology analysis of genes affected by knockdown of Huwe1 using RNA-seq data . Enriched gene categories obtained are ranked by p-value . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 01810 . 7554/eLife . 08536 . 019Figure 6—figure supplement 2 . Huwe1 knockdown inhibits the induction of master myogenic factors during C2C12 differentiation . ( A ) Huwe1 , Myod1 and Mef2c mRNA levels in control and two different Huwe1 knockdown C2C12 myoblast lines were analyzed by RT-qPCR and normalized to Gapdh . Shown are values relative to control cells and from representative experiments; error bars represent standard deviation ( n = 3 ) . ( B ) Myod1 , Myogenin and Mef2c mRNA levels in control and two different Huwe1 knockdown C2C12 lines during differentiation were analyzed by RT-qPCR and normalized to Gapdh . Shown are values relative to control cells on differentiation Day 0 ( D0 ) and from representative experiments; error bars represent standard deviation ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 019 Based on the observations that Huwe1 regulates TBP protein levels in both proliferating and differentiated muscle cells and that Huwe1 protein levels increase during myogenesis , we investigated the effects of Huwe1 knockdown on C2C12 myoblast proliferation and differentiation . When cultured in proliferation medium , the Huwe1 knockdown cells are morphologically similar to control cells ( Figure 6—figure supplement 1A ) . Consistent with this observation , genome-wide transcriptome analysis revealed no significant gene expression changes for cell-cycle and house-keeping genes upon Huwe1 knockdown in C2C12 cells ( Figure 6—figure supplement 1B; Figure 6—source data 3 ) . However , it is worth noting that two important myogenic regulator genes , Myod1 and Mef2c , showed a ∼twofold decrease in expression levels in Huwe1 knockdown cells ( Figure 6—figure supplement 1B ) . We next performed myogenic differentiation experiments to evaluate the lineage commitment potential of Huwe1 depleted myoblasts . Interestingly , Huwe1 knockdown cells failed to differentiate into polynucleated myotubes efficiently even after 6 days of differentiation ( Figure 6A ) . Consistent with this phenotypic observation , the induction of several master myogenic regulators ( Myod1 , Myog and Mef2c ) was inhibited in the Huwe1 knockdown cells during differentiation ( Figure 6—figure supplement 2 ) . To gain more information about potential changes in transcription between control and Huwe1 knockdown cell populations , we compared gene expression profiles of these two cell types by RNA-seq analysis . Differential gene expression followed by gene ontology analysis revealed that many genes critical for muscle development have lower expression levels in Huwe1 knockdown cells ( e . g . , Myod1 , Myog , Mef2a , Mef2c , Ckm and Myh3 ) on differentiation Day 6 ( Figure 6B , C; Figure 6—source data 1 ) , suggesting the functional importance of Huwe1 during in vitro muscle differentiation . We further tested this notion by directly knocking down Huwe1 in differentiated myotubes . Loss of Huwe1 induced abnormal morphological changes of myotubes ( Figure 6D ) . Consistent with these morphological changes , differential gene expression followed by gene ontology analysis revealed that many genes ( e . g . , Myog , Mef2a , Mef2c , Ckm and Myh3 ) important for muscle development are downregulated in Huwe1 depleted myotubes ( Figure 6E , F; Figure 6—source data 2 ) . In C2C12 myoblasts , TBP protein is present at substantial levels despite the presence of Huwe1 E3 ligase . We therefore reasoned that there must be activities in these cells that may be antagonizing the ubiquitination of TBP by Huwe1 and thereby protect it from degradation . Interestingly , a previous study reported that in Saccharomyces cerevisiae , Ubp3 deubiquitinates Tbp1 both in vivo and in vitro and prevents Tbp1 from proteasome-mediated degradation ( Chew et al . , 2010 ) . A protein sequence homology comparison suggests that ubiquitin-specific protease 10 ( USP10 ) is the likely mammalian homolog of Ubp3 ( Cohen et al . , 2003 ) . To test whether USP10 deubiquitinates and stabilizes TBP in mammalian cells , we overexpressed USP10 in 293T cells and monitored TBP ubiquitination levels in these cells . Strikingly , overexpression of USP10 significantly inhibited TBP ubiquitination ( Figure 7A ) . To test whether USP10 deubiquitinates TBP through direct protein–protein interactions , we used HA-TBP to co-IP with Flag-USP10 co-expressed in 293T cells . Our experiment revealed that TBP can co-immunoprecipitate with USP10 ( Figure 7B ) . Importantly , both western blot and flow cytometry analysis confirmed modestly increased TBP protein levels in a C2C12 stable cell line overexpressing USP10 ( Figure 7C , Figure 7—figure supplement 1A ) . Moreover , cycloheximide chase experiments ( CHX ) revealed that USP10 overexpression increased TBP protein half-life in C2C12 myoblasts ( Figure 7—figure supplement 2 ) . These results suggest that USP10-mediated TBP deubiquitination might also contribute to regulating TBP levels during myogenesis . To further test this possibility , we compared the protein levels of USP10 before and after myotube differentiation . Interestingly , similar to TBP , the protein levels of USP10 decrease during in vitro muscle differentiation ( Figure 7D ) . We then examined whether increasing USP10 protein levels in post mitotic myotubes can rescue TBP protein levels . To do this , we generated a stable cell line with an inducible USP10 transgene . Upon addition of the inducer ( cumate ) , USP10 can be induced to higher levels in both myoblasts and myotubes ( Figure 7E ) . USP10 overexpression in myotubes resulted in higher levels of TBP protein ( Figure 7E ) . To probe whether UPS10 downregulation during myogenesis is of functional significance , we evaluated the differentiation potential of the C2C12 cell line overexpressing USP10 . These cells failed to differentiate into polynucleated myotubes in differentiation medium ( Figure 7F ) . Consistent with this observation , transcriptome analysis revealed that several important myogenic markers ( e . g . , Myod1 , Myog , Myh3 and Ckm ) failed to be induced in cells overexpressing USP10 even on differentiation Day 6 ( Figure 7—figure supplement 1B ) . Taken together , our results suggest that a combination of Huwe1 upregulation and USP10 downregulation may be important for modulating TBP degradation during myogenesis . 10 . 7554/eLife . 08536 . 020Figure 7 . Ubiquitin-specific protease 10 ( USP10 ) regulates TBP protein level during myogenesis . ( A ) USP10 overexpression inhibits TBP ubiquitination in vivo . 293T cells were transfected with indicated plasmids as well as a plasmid expressing HA-TBP . Ubiquitin conjugates were purified using Ni-NTA resin under denatured conditions from MG132 treated cells , then subjected to western blot analysis using anti-TBP antibody . Input lysates ( 1% ) are analyzed using antibodies against TBP and USP10 . ( B ) USP10 interacts with TBP . Coimmunoprecipitation using antibodies against HA epitope was performed in 293T cell lysates transfected with indicated plasmids . Input ( 5% ) and precipitated proteins were analyzed using antibodies as indicated . ( C ) USP10 overexpression results in modest increases of TBP protein levels in C2C12 myoblasts . Western blot analysis of whole cell extracts from control or USP10 overexpressing C2C12 myoblasts ( stable cell line ) were done using antibodies against USP10 and TBP . Signals in Ponceau S staining were used as loading controls . ( D ) USP10 protein level decreases during C2C12 differentiation . Whole cell extracts of myoblasts ( D0 ) and myotubes on differentiation Day 3 ( D3 ) and Day 6 ( D6 ) were analyzed by western blots using antibodies against USP10 , TBP and tubulin . ( E ) Induction of USP10 rescues TBP protein levels in differentiated myotubes . Whole cell extracts from myoblasts ( D0 ) and myotubes ( D6 ) of control and Usp10 inducible ( Cumate-USP10 ) C2C12 cells incubated with 10X cumate solution were analyzed by western blot using antibodies against USP10 and TBP . Signals in Ponceau S staining were used as loading controls . ( F ) USP10 overexpression inhibits C2C12 myoblasts differentiation . Shown are phase contrast images of control and USP10 overexpressing C2C12 myoblasts after 6 days of differentiation . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 02010 . 7554/eLife . 08536 . 021Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 02110 . 7554/eLife . 08536 . 022Figure 7—figure supplement 1 . USP10 overexpression elevates TBP protein levels and impairs differentiation capacity . ( A ) USP10 overexpression increases TBP protein level . Wild type and USP10 overexpressing C2C12 ( stable cell line ) were double-stained using rabbit anti-USP10 and mouse anti-TBP primary antibodies , followed by incubation with Dylight594 goat-anti-rabbit and Dylight488 goat-anti-mouse secondary antibodies . Fluorescent intensities were then analyzed by DB FACSAria III Cell Sorter . ( B ) Muscle marker genes were not efficiently induced in USP10 overexpressing C2C12 cells after 6 days of differentiation . Differential mRNA-seq gene expression analysis was used to characterize the effects of USP10 overexpression upon C2C12 differentiation . The log2 scale values of fragments per kilobase per million ( FPKM ) for genes in both WT and USP10 overexpressed differentiation Day 6 samples are plotted in the graph . Dark gene , twofold down; light green , eightfold down; orange , twofold up; red , eightfold up . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 02210 . 7554/eLife . 08536 . 023Figure 7—figure supplement 2 . USP10 overexpression increases TBP protein half-life in C2C12 myoblasts . ( A ) USP10 overexpressing C2C12 myoblasts were treated with 50 μg/ml cycloheximide . Cells were collected at indicated time points and analyzed by western blot using anti-TBP and anti-Tubulin ( control ) antibodies . ( B ) Quantification of western blots in ( A ) using ImageJ . Signal intensities were normalized to 0 time point in each blots . Control values from Figure 5—figure supplement 1 were plotted as comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 023
Given its central role in transcription regulated by all three classes of RNA polymerases , TBP has been long considered to be a universally required component of cellular transcription . However , recent studies have shown that the protein levels of TBP as well as other canonical components of the TFIID complex can become significantly down-regulated in terminally differentiated myocytes , hepatocytes and adipocytes ( Deato and Tjian , 2007; D'Alessio et al . , 2011; Zhou et al . , 2013; Herrera et al . , 2014 ) . Moreover , decreases in TBP protein levels are much more pronounced than changes in its mRNA levels . However , the mechanism for regulating cellular TBP protein levels remained largely unknown . Here , we show that the ubiquitin-proteasome system plays a dominant role in mediating TBP protein levels in terminally differentiated myotubes . Interestingly , we found that an E3 ligase activity responsible for ubiquitinating TBP is also present in proliferating cell types such as myoblasts and Hela cells , where TBP is highly expressed . These results suggest that cellular TBP protein levels are under active surveillance by the ubiquitin-proteasome system and that TBP protein levels may be tightly regulated in a cell-type specific manner . Using a combination of biochemical fractionation and in vitro ubiquitination assays , we identified the Hect domain containing Huwe1 as a key E3 ligase capable of mediating TBP ubiquitination in vitro ( Figure 3 ) . These in vitro biochemical studies subsequently led us to a series of experiments confirming that Huwe1 indeed also regulates TBP levels during myogenesis of C2C12 cells ( Figure 5 ) . Huwe1 is essential for animal development . Specifically , deletion of Huwe1 results in embryonic lethality , with knock-out embryos displaying hemorrhage in the abdominal region by embryonic day 14 . 5 ( E14 . 5 ) , followed by growth impairment , necrosis , and eventual death ( Kon et al . , 2012 ) . In addition to TBP , Huwe1 has also been suggested to target several other proteins involved in cell-cycle check point and apoptosis , including p53 , MCL-1 , N-MYC , C-MYC and CDC6 . Interestingly , although Huwe1 mRNA is ubiquitously expressed in various tissues , it is particularly enriched in skeletal muscle , which is the tissue where a dramatic decrease in TBP protein levels during terminal differentiation was first reported ( Schwarz et al . , 1998; Chen et al . , 2005; Deato and Tjian , 2007 ) . Consistent with this observation , we found that Huwe1 is significantly upregulated during in vitro muscle differentiation of C2C12 cells and that up-regulation of this E3 ligase appears to be functionally important for myogenesis and maintenance of normal muscle morphology ( Figure 6 ) . As part of the TBP surveillance system , we also found that a deubiquitinase , UPS10 , contributes to the regulation of TBP ubiquitination and degradation by counteracting the Huwe1 E3 ligase activity . USP10 is a ubiquitously expressed deubiquitinase , whose substrates include tumor suppressor p53 ( Yuan et al . , 2010 ) . The exact role of USP10 during development remains unclear due to the absence of mouse models . However , we found that a stable cell line ( C2C12 ) overexpressing USP10 is impaired in myotube formation , suggesting that down-regulation of USP10 may also be a prerequisite for efficient differentiation of myoblasts into myotubes in culture . Deubiquitinases achieve their target specificities through either direct recognition of their substrates or targeting specific ubiquitin chain topologies . Our immune-precipitation experiments suggest that USP10 can recognize ubiquitinated TBP through direct protein–protein interactions ( Figure 7B ) . It remains unclear at this point whether there are other deubiquitinases that can also recognize ubiquitinated TBP . It is worth noting that due to the promiscuous ‘one-to-many’ relationship between E3 ligase , Deubiqutinase ( DUBs ) and their substrates , it is difficult to directly test whether the myogenic defects we observed after the loss of Huwe1 or USP10 over-expression are directly due to the failure of down-regulating TBP during differentiation or some other consequences of depleting an E3 ligase or over-expressing a deubiquitinase . However , given the seminal role of TFIID/TBP in promoting the transcription of cell cycle and DNA replication genes ( Um et al . , 2001 ) , it is reasonable to speculate that down-regulation of TBP should at least influence cell cycle exit of myoblasts , a key step during myotube differentiation . In the future , it may be interesting to study the functional role of Huwe1 and UPS10 during muscle development in vivo and how these two enzymes may regulate TBP protein levels in mouse models . Since TBP protein levels also become dramatically reduced in terminally differentiated hepatocytes and adipocytes ( D'Alessio et al . , 2011; Zhou et al . , 2013 ) , it will also be interesting to test whether Huwe1 and UPS10 contribute to TBP downregulation in these other cell types . It also remains unclear whether there are other E3s that would work together with Huwe1 to facilitate TBP protein degradation in terminally differentiated muscle cells . In addition to TBP , other components of the TFIID complex also become down-regulated during terminal differentiation , and in the future it will be worth investigating whether they are targeted by the same or different E3/deubiquitinase pairs . Our results suggest that significant up-regulation of Huwe1 and simultaneous down-regulation of UPS10 during myotube differentiation ( Figure 5B , Figure 7D ) may play an important role in regulating proper TBP protein levels during muscle differentiation ( Figure 8 ) . One striking observation is that although the protein levels of TBP and TAFs are significantly down-regulated , the protein levels of other basal transcription factors like RNA polymerase II remain largely unchanged ( Figure 1—figure supplement 1 ) . We still have not fully sorted out the functional importance of this selective down-regulation of TBP and TAFs during terminal differentiation . 10 . 7554/eLife . 08536 . 024Figure 8 . Schematic representation of coordinated regulation of TBP protein level in proliferating and differentiated cells by E3 ligase Huwe1 and deubiquitinase USP10 . Both Huwe1 E3 ligase and USP10 deubiquitinase are present in proliferating myoblasts to maintain steady state TBP protein levels . Upon stimulation by differentiation signals , Huwe1 protein levels increase while USP10 protein levels decrease , resulting in increased TBP ubiquitination and degradation by the proteasome . DOI: http://dx . doi . org/10 . 7554/eLife . 08536 . 024 In several previous studies , we observed that in addition to the loss or depletion of TBP during terminal differentiation of cell-types including myotubes and adipocytes , one of the so-called ‘orphan TAFs’ such as TAF3 and TAF7l , respectively , becomes up-regulated or remains highly expressed while the other prototypic TFIID subunits become down-regulated ( Deato and Tjian , 2007; Yao et al . , 2011; Zhou et al . , 2013 ) . In the case of myoblast to myotube formation , we also detected the presence of a TBP related factor , TRF3 , that appeared to play some role in possibly either substituting for TBP , as seen by in vitro reactions , or otherwise remaining detectable during terminal differentiation ( Deato and Tjian , 2007; Deato et al . , 2008 ) . This led us , initially , to propose that the loss of TBP might be accompanied by the assembly of different initiation complexes lacking TBP but containing TRFs and orphan TAFs . However , in several subsequent studies including ESC's , adipocytes and motor neurons , although we consistently observed the emergence and up-regulation of orphan TAFs ( i . e . TAF3 , TAF7l and TAF9b ) , we did not detect any up-regulation of TRFs ( Liu et al . , 2011; Zhou et al . , 2013; Herrera et al . , 2014 ) . Instead , in each of these cases , ChIP-seq experiments suggested that these orphan TAFs were associated both with core promoters ( i . e . TSS ) as well as either distal or proximal enhancers of target genes ( Liu et al . , 2011; Zhou et al . , 2013; Herrera et al . , 2014 ) . It was also reported that KO mice lacking TRF3 showed no obvious muscle development phenotypes casting doubt on the importance of TRF3 in muscle development in vivo ( Gazdag et al . , 2009 ) . It is therefore possible that both the loss of TBP and TFIID components in certain cell-types is not necessarily accompanied by a change in the core promoter recognition complex as initially postulated ( i . e . substituting various TBP/TAFs/TRFs ) but a more elaborate and as yet unclear mechanism may come into play potentially involving new interactions between certain TAFs and cell-type specific transcription factors bound to enhancers . In any case , in all our studies of terminal differentiating cell types , we observed a significant , sometimes dramatic loss of TBP and TFIID components that is often accompanied by persistent or differential up-regulation of an orphan TAF . An interesting possibility is that down-regulation of TBP protein levels could differentially affect genes with distinct TBP-promoter binding affinities . Recent single molecule studies suggest that transcription factor temporal occupancies at high affinity sites are less sensitive to changes in concentration than those at low-affinity sites ( Chen et al . , 2014 ) . Thus , it is possible that , by fine-tuning TBP protein levels , genes with weak TBP/TFIID binding promoters become selectively turned down while genes with strong TBP/TFIID binding promoters remain largely unaffected thereby executing another mechanism for differential gene regulation .
C2C12 myoblasts were obtained from ATCC ( CRL-1772 , Manassas , VA ) , and cultured in DMEM medium ( 11995-065 , Life technologies , Frederick , MD ) containing 10% FBS ( 26140-079 , Life technologies ) at sub-confluent densities ( <70% ) . For differentiation , C2C12 cells were grown to 100% confluence and shifted to differentiation medium DMEM ( 11995-065 , Life technologies ) containing 2% horse serum ( 16050-130 , Life technologies ) . Since decreases in protein levels of TBP and TAFs are generally more dramatic starting from differentiation day 6 , we collected myotubes on this day for most analysis unless otherwise noted . Cytoplasmic and Nuclear extraction were prepared as previously described ( Dignam et al . , 1983 ) . For whole cell extraction preparation , cell pellets were incubated in lysis buffer ( 1% NP-40 , 150 mM NaCl , 50 mM , pH 8 . 0 Tris–HCl , 0 . 5% sodium deoxycholate , protease inhibitor cocktails ( 05892791001 , Roche , Nutley , NJ ) ) for 30 min on ice , passed through needles ( 20 Gauge ) 7–8 times , then centrifuged at 15 , 000 rpm for 30 min . Supernatant was kept as the cell lysate . Protein concentration of soluble cell lysate was measure by Bradford assay . Monoclonal antibodies against TBP ( ab61411 ) , Huwe1 ( ab78397 ) and polyclonal antibodies against USP10 ( ab72486 ) , RNA polymerase II ( ab52202 ) , HA-tag ( ab9110 ) from abcam ( Cambridge , MA ) ; polyclonal antibody against ubiquitin ( 07–375 ) from EMD MILLIPORE ( Billerica , MA ) ; monoclonal antibodies against TAF4 ( 612054 , BD Transduction Laboratories ) and Myogenin ( 556358 , BD Pharmingen ) from BD Biosciences ( Franklin Lakes , NJ ) ; anti-Flag tag ( M2 ) monoclonal antibody from Sigma Aldrich ( St . Louis , MO ) . 293T cells were transfected with His6-ubiquitin ( 6 μg ) and HA-TBP ( 2 μg ) as indicated in Figure 1B and Figure 7A by calcium phosphate transfection . 24 hr after transfection , cells with treated with 1 μM MG132 overnight to accumulate ubiquitinated proteins . Cells were harvested for in vivo ubiquitination assays , which were performed as described previously ( Jin et al . , 2012 ) . Western blots were performed using anti-TBP antibody . In a 30-µl reaction , 2 μl of in vitro translated HA-TBP ( generated using TNT SP6 Coupled Reticulocyte Lysate System ( L4600 , Promega , Madison , WI ) ) or 200 ng of recombinant GST-TBP was incubated with an ATP regenerating system ( 37 . 5 mM creatine phosphate , 5 mM ATP , pH 8 . 0 , 5 mM MgCl2 ) . 2 μg of ubiquitin ( U-100 , Boston Biochem , Cambridge , MA ) , 50 ng E1 ( E-305 , Boston Biochem ) , 100 ng UbcH5b ( E2-622 , Boston Biochem ) , 2 μM ubiquitin aldehyde ( U-201 , Boston Biochem ) , 10 μM MG132 ( 474790 , EMD Millipore ) and 10 μg of S100 , partial purified fractions or purified recombinant Huwe1 protein for one and a half hour . For detection of ubiquitinated species using the anti-ubiquitin antibody , after the reaction , reaction mixtures were diluted in buffer ( 1% NP40 , 150 mM NaCl , 50 mM Tris–HCl pH 7 . 5 , 0 . 5% sodium deoxycholate ) , followed by immunoprecipitation using anti-HA antibody/protein G sepharose beads or pull-down using glutathione sepharose beads . Purified TBP proteins were then subjected to three washes with the same buffer , eluted by boiling in 2×SDS loading buffer and analyzed by western blots . For detection of ubiquitinated species using the anti-TBP antibody , reactions were terminated with 2×SDS sample buffer , and then analyzed by western blots . All steps were performed at 4°C . Cytoplasmic extracts were prepared from 200 L of Hela cells using Buffer A ( 10 mM HEPES pH7 . 9 , 0 . 5 mM MgCl2 , 100 mM KCl , 0 . 5 mM DTT ) , then applied to a D52 DEAE cellulose ( Whatman ) column , washed extensively at 100 mM KCl , 200 mM KCl and eluted at 300 mM KCl . This 300 mM KCl fraction was then precipitated with ammonium sulfate ( 40% saturation ) , and re-suspended in Buffer A containing 100 mM KCl and 800 mM ( NH4 ) 2SO4 . The soluble fraction was applied to Hitrap Butyl Sepharose FF hydrophobic column ( 17-5197-01 , GE Healthcare Life Sciences , Pittsburgh , PA ) , subjected to 10 column volume ( CV ) washes using Buffer A containing 800 mM ( NH4 ) 2SO4 , eluted with a 10 CV linear gradient from Buffer A containing and 800 mM ( NH4 ) 2SO4 to Buffer A . Active fractions were then pooled and separated on a Superose6 10/300 GL ( 17-5172-01 , GE Healthcare Life Sciences ) equilibrated with Buffer A . Active Superose 6 fractions with an approximate molecular mass of 440–660 kDa were pooled and supplemented with 0 . 1 mg/ml insulin ( 11376497001 , Roche ) . Pooled fractions were applied to a Hitrap Heparin HP column ( 17-0406-01 , GE Healthcare Life Sciences ) , washed with 10 CV of Buffer A , and eluted with a 10 CV linear gradient from 0 . 1M KCl to 0 . 5 M KCl . Active Heparin fractions were pooled and dialyzed against Buffer A . Dialyzed fractions were then applied to a Mono Q GL column ( 17-5166-01 , GE Healthcare Life Sciences ) , washed with 10 CV of Buffer A , then developed with a 20 CV linear gradient from 100 mM KCl to 500 mM KCl . Active TBP E3 ligase fractions eluted from 300 mM to 400 mM KCl . pFastBac plasmid for his-tag wild-type Huwe1 expression was generously provided by Dr . Qing Zhong ( UTSW ) . Huwe1 catalytic domain deletion was created by replacing the a ApaI-NotI fragment from pCI-neo wild-type Huwe1 plamids ( provided by Dr . Qing Zhong ( UTSW ) ) with a ApaI-NotI fragment containing Huwe1 deletion ( amplified using primers 5′-CGGGGTCGGGCCCGCCTCCTGGTAGGCAAC-3′ and 5′-ACGATGCGGCCGCTTATGTGTGAGCTGAAGGCAGGCG-3′ ) . The full-length Huwe1 cDNA with C-terminal deletion was then cloned into pFastBac as a SalI-NotI fragment . pFastbac plasmid for Huwe1 catalytic site mutant was created in a similar way , except for that a different reverse primer ( 5′-CGGGGTCGGGCCCGCCTTAGGCCAGCCCAAAGCCTTCAGAGC ACTCCTGGATAGCCAACAGTAGCATGTGGCGGAGCTTCTCAAAGCTCTCATAGGCAGGCAGATCCAGCTGATTAAAACTTGTGTGAGCTGAA-3′ ) was used for ApaI-NotI fragment generation . pFastBac plasmids were transfected into Sf9 using Cellfectin II reagent ( 10362-100 , Life technologies ) following manufacture's instructions . P3 viruses were used to infect Sf9 cells for protein production . The cells were harvested 48 hr post-infection . Cell pellets were re-suspended in Buffer A ( 50 ml Buffer A for cell pellets from 1 L culture ) , and homogenized with a glass douncer . Cytoplasmic fraction was collected by centrifuging at 10 , 000 rpm for 30 min , added in imidazole to final concentration of 10 mM , and then loaded onto a Ni-NTA gravity column . The column was washed with 20 CV Buffer A containing 10 mM imidazole , and the bound His- Huwe1 was eluted using Buffer A containing 100 mM imidazole . A Superdex200 Gl gel filtration column was used to further purify the His-Mule elution . PageBlue staining was used to check Huwe1 protein in each fraction . Glycerol was added to the fractions containing Huwe1 to a final concentration of 5% , and the proteins were flash frozen in liquid nitrogen and stored at −80°C . Huwe1 knockdown in C2C12 myoblasts and myotubes were performed using lentiviruses . Viruses were produced by transfecting pLKO . 1 shRNA plasmids targeting Huwe1 ( targeting sequences: #1: 5′-GCACTCTTCATAACTCACTTT-3′; #2: 5′-GCACTGCTCATCAAAGATGTT-3′ ) ( CloneID:TRCN0000092553; TRCN0000092556; GE Dharmacon , Lafayette , CO ) with packaging vectors into 293T cells using FuGENE HD ( E2311 , Promega ) . Supernatants were concentrated using Fast-Trap Lentivirus Purification and Concentration Kit ( FTLV00003 , Millipore ) . Huwe1 knockdown in myoblasts were performed in the presence of 8 µg/ml polybrene at around 70% confluency followed puromycin ( A11138-03 , Life technologies ) selection ( 1 µg/ml ) . Extensive passages of surviving cells were avoided in order to maintain knockdown efficiency . Huwe1 knockdown in myotubes was performed by incubating lentiviral concentrates in the presence of 8 µg/ml polybrene on differentiation day 4 for two days , followed by puromycin selection for another 2 days ( 1 µg/ml ) to eliminate uninfected myotubes . The plasmid expressing wild type TBP-mCherry-sfGFP fusion protein under a CMV promoter were transfected into 293T cells by Fugene HD reagent following manufacturer's instructions . Cell imaging was done using Zeiss Axio Observer microscope . sfGFP signal was detected using excitation filter TBP 495/20 , and mCherry signal was detected using excitation filter TBP 570/30 . Same exposure time was used for different samples . Images were then further analyzed using ImageJ through the same processes . FACS analysis of TBP-mCherry-sfGFP signal was done in DB FACSAria III Cell Sorter . sfGFP was detected using 488 nm laser , and mCherry was detected using 561 nm laser . mCherry/sfGFP was set as a parameter and detected simultaneously with mCherry and sfGFP signals . Total RNA from wild type and Huwe1 knockdown myoblasts or myotubes was isolated using RNeasy kit ( 74106 , Qiagen , Valencia , CA ) . mRNA was then purified using Dynabeads Oligo ( dT ) 25 ( 25-61002 , Life technologies ) . RNA-seq library was prepared using ScriptSeq v2 RNA-seq Library Preparation Kit ( SSV21106 , illumina , San Diego , CA ) , and then sequenced using an Illumina Hiseq 2000 sequencing platform . Reads were mapped to the genome using Tophat , and differentiation expression analysis was done using Cuffdiff ( Trapnell et al . , 2009 , 2012 ) . Results were plotted out using R package Plotrix , and gene ontology analysis was done using R package Goseq ( Young et al . , 2010 ) . The following data set was generated: Li L . A specific E3 ligase/deubiquitinase pair modulates TBP protein levels during muscle differentiation http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE72105 . Publicly available at NCBI Gene Expression Omnibus . | Most animal cells specialize to perform particular roles that contribute to the survival of the animal in different ways . For example , the cells that form our muscles are able to contract , while other cells in the body are efficient at storing fat . The different types of cells develop from unspecialized cells , but it is not clear what controls this process to form a particular type of cell in the right place at the right time . The TATA-box binding protein ( TBP ) is one of a group of proteins that helps to activate the expression of genes in animal cells . Recent studies have revealed that TBP is deliberately destroyed by a group of proteins called the proteasome in muscle cells , in a type of liver cell , and in fat cells . Here , Li et al . used biochemical techniques to study the regulation of TBP during the formation of muscle cells from less specialist mouse cells called myoblasts . The experiments show that an enzyme called Huwe1 selectively adds a tag to TBP that marks TBP for destruction by the proteasome . Another protein called USP10 acts to remove the tags to prevent TBP from being destroyed . Therefore , it appears that changes in the levels of Huwe1 and USP10 fine-tune the amount of TBP that is degraded during the formation of muscle cells . Li et al . 's findings suggest that other proteins that are also involved in activating gene expression may also be destroyed as muscle cells form . The next step is to understand how important the degradation of these proteins is to the formation of other types of specialist cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"biochemistry",
"and",
"chemical",
"biology"
] | 2015 | A specific E3 ligase/deubiquitinase pair modulates TBP protein levels during muscle differentiation |
The central complex is a highly conserved insect brain region composed of morphologically stereotyped neurons that arborize in distinctively shaped substructures . The region is implicated in a wide range of behaviors and several modeling studies have explored its circuit computations . Most studies have relied on assumptions about connectivity between neurons based on their overlap in light microscopy images . Here , we present an extensive functional connectome of Drosophila melanogaster’s central complex at cell-type resolution . Using simultaneous optogenetic stimulation , calcium imaging and pharmacology , we tested the connectivity between 70 presynaptic-to-postsynaptic cell-type pairs . We identified numerous inputs to the central complex , but only a small number of output channels . Additionally , the connectivity of this highly recurrent circuit appears to be sparser than anticipated from light microscopy images . Finally , the connectivity matrix highlights the potentially critical role of a class of bottleneck interneurons . All data are provided for interactive exploration on a website .
The central complex consists of four main neuropiles — the protocerebral bridge ( PB ) , the ellipsoid body ( EB , Central Body Lower in other insects ) , the fan-shaped body ( FB , Central Body Upper in other insects ) and the noduli ( NO ) — and at least three accessory neuropiles ( also known as the lateral complex ) — the gall ( GA ) , the lateral accessory lobe ( LAL ) and the bulbs ( BU ) ( Figure 1A and [Wolff et al . , 2015; Lin et al . , 2013; Hanesch et al . , 1989] ) . Throughout this manuscript , we denote output ( resp . input ) neurons that link central complex neuropiles to neuropiles outside the central complex . Some of the most striking neural elements of the central complex are the columnar neurons , which innervate one of the 18 ( in Drosophila ) glomeruli of the PB , one vertical section of either the FB or EB , and one accessory neuropile — a column being constituted by the PB glomeruli and FB/EB section . A total of 12 different columnar cell types have been described , with stereotypical correspondences between the PB glomerulus and the EB/FB section . In addition to these ‘principal cells’ , there are a number of neurons innervating multiple columns of one neuropile . These neurons often innervate subdivisions orthogonal to the columns . Moreover , they sometimes also project to neuropiles outside the central complex . This set of neurons includes the ring neurons , which innervate a ring within the EB and an accessory neuropile , and a collection of inputs and interneurons with processes in the FB and PB . From this light level anatomy and putative synaptic polarity , one can derive a hypothetical picture of information flow through the central complex ( Figure 1Bi ) : We show that this overall flow of information is generally supported functionally for the parts we have tested so far , but with a few potentially important differences ( Figure 1Bii ) : the observed connectivity in the PB is sparse , rendering the function of PB interneurons possibly critical; accessory structures are usually input rather than output areas; and , consequently , output channels of the central complex are scarce .
We picked driver lines for functional connectivity mapping by visually inspecting the Janelia Gal4-driver collection ( Jenett et al . , 2012 ) for strength of expression in the cell types of interest , and sparseness of the expression pattern in the central complex . The 37 driver lines ( for 24 cell types ) cover the main columnar neuron types ( 8 of the 11 types described in Wolff et al . ( 2015 ) ) and PB interneurons ( 3 out of the five in Wolff et al . ( 2015 ) ) , a LAL-FB neuron , three types of ring neurons , a Gall-EB neuron columnar in the EB and neurons innervating accessory structures , namely four types of LAL interneuron and three types of neurons connecting the LAL to the noduli . Drivers are listed in Table 1 and Table 2 . Neuron types are schematized in Figure 2A and Figure 1—figure supplement 1 . The dataset includes inputs to the EB system , connections between EB columnar neurons , connections in the PB as well as potential inputs and outputs in the LAL , Gall and noduli . Among the types tested , 43 of the 59 anatomically possible connections could be tested with the reagents available . The connectivity of the multitude of cell types within the FB has not been explored: neither FB interneurons ( also known as pontine cells ) , nor FB input neurons are part of this study . Cell-type pairs to be tested were chosen based on overlaps between their expression patterns in light microscopy images . For each combination selected , we expressed CsChrimson and GCaMP6m in potential pre- and post-synaptic partners , respectively ( Figure 2B , C ) , and probed their connection in an ex vivo preparation using a standardized protocol ( see Figure 2D , and Materials and methods ) . Whenever large responses were observed , we used pharmacology to both check that observed transients were synaptically mediated , and to narrow down the neurotransmitters involved ( Figure 4—figure supplement 2 and Figure 4—figure supplement 3 ) . Effects of stimulation ranged from very large and reliable transients ( Figure 3Ai ) to undetectable changes ( Figure 3Aiii ) . In between those extremes , we observed transients of variable size and reliability ( Figure 3B ) . To our surprise , we could also detect clear inhibitory responses ( Figure 3Aii ) . This was possible because—at least in some cell types—fluctuations in baseline activity occasionally elevated GCaMP levels during the experiment ( see Discussion and Figure 3—figure supplement 3 ) . Therefore , even though hyperpolarization below resting potential is likely not detectable through calcium imaging , we could detect inhibition from an excited state as a dip in the fluorescence trace . Since no single characteristic of the responses could adequately capture their variety , variability and complexity , we chose to characterize the transients by using a battery of statistics reflecting response amplitude , shape , reliability and stimulus sensitivity ( see Figure 3C , Figure 3—figure supplement 2 , and Materials and methods ) . Responses of control pairs with non-overlapping processes were then used to form the null-hypothesis distributions of two metrics that capture response amplitude and reliability ( see Figure 3D ) . For every data point , the Mahalanobis distance ( a covariance corrected measure in a multidimensional space , see Materials and methods ) to the null distribution was computed and used as a connection strength metric in summary diagrams like Figure 4 and Figure 5 . Non-overlapping pairs usually showed no fluctuations upon stimulation , and when they did , they were small and unreliable ( see Figure 3—figure supplement 1 ) , likely reflecting effects of indirect connections . Not surprisingly , responses were always detected with same-cell-type-stimulation controls , where CsChrimson and GCaMP6m were expressed in the same neuron type ( see Figure 3D ) . All the individual responses and statistics , in the context of the overall connectivity diagram , are available at http://romainfr . github . io/CX-Functional-Website/romainfr . github . io/CX-Functional-Website/ , a website that enables an interactive exploration of the results of this study . We plan to update this website as further experiments are performed . The website can also be expanded to accommodate other sources of data , which would make it an exhaustive source of information about the central complex . The connectivity matrix resulting from our experiments is shown in Figure 4 in two alternative visualizations , namely a network diagram ( Figure 4 ) and a matrix of connection strengths . Figure 5 outlines some of the connectivity patterns we observed . We focus in particular on inputs and outputs to the ellipsoid body , protocerebral bridge and paired noduli , connectivity within the protocerebral bridge , and components of the ring attractor network within the central complex . Figure 5Dii shows the subpart of the network that has been proposed to sustain the ring attractor representation of heading ( Green et al . , 2017; Turner-Evans et al . , 2017 ) . One hypothesized feature of such a circuit is a large degree of recurrence between the different EB columnar types . In particular , P-EN to E-PG reciprocal connections are important for models of the rotation of the bump . While we found strong support for the P-EN1 to E-PG connection , the E-PG to P-EN1 connection that we reported functionally under a stronger stimulation protocol ( Turner-Evans et al . , 2017 ) may be mediated through the Δ7 interneurons . A few other connections were found in the EB ( for example , P-EN1 to P-EN2 ) , but it is important to stress that not all combinations could be tested due to limitations in the genetic reagents available . For example , the role of the P-EG neurons in this circuit , remains unclear . A key additional type that our results suggest may contribute in important ways to the persistence of activity in this circuit is the AMPG-E neuron , a columnar Gall-EB neuron not innervating the PB , which appears to provide localized excitatory feedback to the E-PG neurons .
The connectivity technique we applied has several limitations that are important to keep in mind . First , connections detected using CsChrimson and GCaMP cannot be guaranteed to be direct and monosynaptic . However , the large set of controls with cell-type pairs whose processes do not overlap provides a statistical framework to interpret the results — not surprisingly , uncertainty is highest for weak connections . We believe that the metric we used to assess connectivity – distance to a control set rather than just response strength – makes the resulting network more interpretable . Additionally , by releasing the entire dataset rather than just the derived network , we hope to provide interested central complex researchers the opportunity to explore the data and to potentially reinterpret it in the light of other findings . A more fundamental issue concerns the sensitivity of our protocol , which is limited by the stimulation protocol and the sensitivity of GCaMP6m . Specifically , an absence of a post-synaptic response cannot be interpreted as an absence of a connection . The fact that some inhibitory responses are visible , and that strong responses saturate with the range of stimulations used ( see Figure 4—figure supplement 4 ) is reassuring . However , it is likely that EM reconstructions of central complex circuits will reveal that some weak synaptic connections have been missed by our technique . Their functional importance will need to be investigated using more sensitive methods , for example , intracellular electrophysiology . For example , we may be underestimating the level of connectivity in the PB: our finding of sparseness in the structure should thus be interpreted as sparseness at the resolution of our technique , because this network may be dominated by weak connections below our detection threshold . Further , we relied on full-field stimulation of populations of specific neuronal types , which comes with its own drawbacks: this approach provides no access to connectivity between neurons of the same class , and does not account for potential non-physiological network effects . One such effect would be the recruitment of global inhibitory networks that could mask an otherwise excitatory connection . However , whenever we suspected this could be a possibility , we controlled for it by blocking inhibition with picrotoxin , and never saw evidence of a significant effect ( Figure 4—figure supplement 3 ) . Even though picrotoxin was effective in blocking the inhibitory responses we observed when activating ring neurons or Δ7 neurons ( Figure 4—figure supplement 3 ) , we cannot exclude the possibility that picrotoxin-insensitive inhibition might be present in the network . Possible candidates mediating such effect would be GABA-B ( Olsen and Wilson , 2008 ) , metabotropic glutamatergic or peptidergic transmission ( Kahsai and Winther , 2011 ) . For example , the PB shows both GABA-B and metabotropic glutamatergic receptor immunoreactivity ( Kahsai et al . , 2012 ) , which could make the observed connectivity seem sparser than it actually is . Interestingly , whereas in other insect species the PB displays peptidergic immunoreactivity ( in particular Allatostatin-A , see[Vitzthum et al . , 1996] ) , this seems not to be the case in Drosophila ( with the exception of SIFamide , see [Kahsai et al . , 2012] ) . The fact that we stimulate entire presynaptic populations also means that the strength of connections we report is influenced both by neuron-to-neuron transmission strength and the degree of convergence in the network . Caution should therefore be exercised when comparing connections between columnar neuron population and other types ( e . g . from PF-L and SMPL-L or P-EG to G-Eo ) , which are potentially highly convergent , to connections between columnar neuron types , or from accessory structure neuron types to columnar populations ( like GL-N1 to P-EN1 ) , which are both mediated by a single or small handful of presynaptic neurons for any given post-synaptic neuron . Furthermore , even though several of the neurons described have complex morphologies that are suggestive of local processing within specific neuropiles , we never found neuropile-specific responses: when a neuron responded , the response seemed to invade the entire neuron . This may be the result of our broad and artificial stimulation protocol . Given that our protocol is limited to the activation of one cell type , we might also have missed connections gated by other inputs . For example , we failed to find any PB input for the PF-LCre neurons – the sole output neurons we identified in this study . It is possible that this neuron requires convergent inputs from the PB and FB to be activated , as was suggested in ( Stone et al . , 2017 ) . Finally , all our experiments were performed in ex vivo brain preparations . Given the variety of neuromodulators that operate in the central complex ( Kahsai and Winther , 2011 ) , it is likely that functional connectivity within this region is modulated by brain state ( Homberg , 1994; Seelig and Jayaraman , 2013; Weir et al . , 2014; Weir and Dickinson , 2015 ) . Consistent with this possibility , we saw that the fluorescence baseline tended to fluctuate spontaneously during the course of our experiments in most types recorded ( as shown in Figure 3—figure supplement 3A ) . Two neuron types had relatively small baseline fluctuations that could have make it difficult to detect inhibition when those were used as post-synaptic targets: P-EG and PF-LCre . Intriguingly , although increases in baseline activity allowed us to detect inhibitory responses , we noticed that excitatory responses also occasionally depended on this baseline fluctuation ( Figure 3—figure supplement 3C ) . It is conceivable that such baseline fluctuations reflect a kind of artificial brain state upon which the response amplitudes depend . From our pharmacology experiments , we propose that columnar neurons and IS-P neurons are likely cholinergic ( see Figure 4—figure supplement 2 ) , whereas the LAL and Gall ring neurons , as well as the Δ7 neurons are either glutamatergic or GABAergic ( see Figure 4—figure supplement 3 ) . A large fraction of ‘canonical’ BU-ring neurons have been shown previously to be GABAergic ( Zhang et al . , 2013 ) , which makes this likely for the LAL and Gall-ring neurons described here . We argue below that the response profiles of Δ7 neurons suggests that they are glutamatergic , in accordance with ( Daniels et al . , 2008 ) . The connectivity matrix we obtained is sparser than that predicted by light level anatomy . Our results suggest that the Δ7 interneurons are a bottleneck for information processing in the PB . This is all the more interesting given the range of responses evoked by Δ7 stimulation ( Figure 5—figure supplement 1 ) . Properties of the synapses that Δ7 neurons make with their post-synaptic partners may play a primary role in the way that a heading signal is generated and maintained in the EB columnar system , and also in how it may be transferred to the FB columnar system . This has also been suggested in other insect species ( Stone et al . , 2017 ) for the homologous TB1 neuron , which was a key component of the proposed compass circuit model in that study . Every Δ7 neuron innervates all columns of the PB , and has presynaptic-looking processes in two columns . The fact that a neuron with such extensive arbors participates in a circuit where representations are spatially restricted ( heading-related activity is limited to a few neighboring columns at any given time ) suggests that understanding local processing at the single neuron level might be critical to a complete understanding of how the circuit as a whole operates . This may also be the case for some of the ring neurons that provide input to the ellipsoid body . The observation that Δ7 neuron stimulation can excite or inhibit its post-synaptic partners can have several explanations . Either the population of Δ7 is not homogeneous , and contain several functionally distinct types responsible for the different types , or the responses reflect a variety of receptors on the post-synaptic side . This latter hypothesis would be compatible with the previously discussed hypothesis that Δ7 are glutamatergic , as the diversity of ionotropic and metabotropic glutamate receptors would allow such response diversity . The fact that several sources of input are inhibitory raises the question of how activity is maintained in the region . Candidate mechanisms are the uncovered excitatory inputs into the PB and EB , recurrent connections in the EB and intrinsic properties of neurons ( Egorov et al . , 2002; Yoshida and Hasselmo , 2009; Russell and Hartline , 1982 ) — some cell types , for example , showed robust post-stimulation rebounds ( see Figure 3Bii ) . It is also possible that our selection of cell types and our methods missed some sources of excitation . The range of inputs revealed here opens many avenues for investigation . Whereas some ring neuron subtypes have received considerable attention ( Sun et al . , 2017; Shiozaki and Kazama , 2017; Seelig and Jayaraman , 2013 ) , most PB inputs and LAL-noduli interneurons have not yet been characterized . A recent study in the sweat bee ( Stone et al . , 2017 ) , for example , reported that one of the LAL-noduli interneurons — a likely input to the FB system — carries forward and backward translational optic flow signals . This is all the more interesting given that we show that one of the LAL-NO interneuron types ( GL-N1 ) provides input to the P-EN neurons , known to encode rotational signals ( Turner-Evans et al . , 2017; Green et al . , 2017 ) . The specific functions subserved by the network motifs that we have uncovered may only become clear with functional studies in behaving animals . A key puzzle set up by our findings is the small number of output channels of the central complex . Our results are consistent with the LAL being the primary output structure for the central complex ( Chiang et al . , 2011; Hanesch et al . , 1989 ) , although the structure also acts as an input region ( via ring neurons and potentially via IMPF-L neurons ) . While it is possible that our selection of Gal4 lines was unintentionally biased against output neurons , or that our technique otherwise missed a number of output pathways , the picture of the central complex that emerges is of a densely recurrent sensorimotor hub with relatively low dimensional output ( much as proposed by some models for example [Stone et al . , 2017; Fiore et al . , 2015; Strauss and Berg , 2010] ) . The implications of this bottleneck for motor control remains a challenge for future studies to resolve .
Drivers were chosen based on relatively sparse expression within the central complex . For any given pair of neurons , the overlap between pre- and post-synaptic looking regions was assessed based on publicly available expression patterns ( [Tirian and Dickson , 2017; Jenett et al . , 2012] , see Figure 1—figure supplement 1 ) digitally aligned on a common reference brain ( as described in [Aso et al . , 2014] ) . For every LexA driver used , we prepared two stocks containing GCaMP6m ( Chen et al . , 2013 ) and CsChrimson ( Klapoetke et al . , 2014 ) under LexAop ( resp . UAS ) or UAS ( resp . LexAop ) control: XXX-LexA;13XLexAop2-IVS-p10-GCaMP6m 50 . 629 in VK00005 , 20xUAS-CsChrimson-mCherry-trafficked in us ( How ) attP1 and XXX-LexA;20xUAS-IVS-GCaMP6m 15 . 629 in attP2 , 13XLexAop2-CsChrimson-tdTomato in VK00005 . Those stocks were then crossed to a Gal4 driver or a split-Gal4 ( Luan et al . , 2006 ) driver for the experiment . For split-Gal4s , the two split halves were inserted in attP40 and attP2 respectively . To avoid transection between the split and the LexA driver ( Mellert and Truman , 2012 ) , we inserted the LexA drivers in alternative sites , either su ( Hw ) attP5 ( Pfeiffer et al . , 2010 ) or VK22 ( Venken et al . , 2006 ) , and used the splits exclusively in combination with those lines after checking their expression patterns . The list of drivers used and the corresponding cell types are given in Table 1 . Throughout this paper , we follow the naming convention set out in Wolff et al . ( 2015 ) for full names , and the scheme described in Kakaria and de Bivort ( 2017 ) and used in Green et al . , ( 2017 ) and Turner-Evans et al . , ( 2017 ) for abbreviations . For each cell type , we labeled every region innervated as presynaptic or postsynaptic ( or both ) : this was done at the resolution of the glomerulus for the PB , the layer for the FB and the individual nodulus . We divided the LAL into three zones based on the overlap between the lines used . Existing subdivisions for the EB and Gall were preserved . This labeling was used to evaluate whether the arbors of a given cell-type-pair overlapped . The brains of 5 to 9 days old female flies were extracted and laid on a poly-D-lysine coated coverslip ( Corning , Corning , NY ) . In most experiments , both the brain and the ventral nerve chord ( VNC ) were dissected out , as we found that having the VNC attached to the brain increased the mechanical stability of the preparation . Dissection was performed using the minimum level of illumination possible to avoid spurious activation of CsChrimson . The preparation was bathed throughout in saline containing ( in mM ) : 103 NaCl , 3 KCl , 5 TES , 8 trehalose dihydrate , 10 glucose , 26 NaHCO3 , 1 NaH2PO4 , 2 CaCl2 , 4 MgCl2 , bubbled with carbogen ( 95% O2 , 5% CO2 ) . Brains were positioned anterior-side-up , except when the connection tested was thought to be in the PB , in which case they were positioned posterior-side-up to maximize light access close to the assumed synaptic site . Trachea were removed . Only for experiments involving pharmacology , the glial sheath was gently torn with tweezers to enhance drug access to the neuropiles . Imaging was performed on an Ultima II two-photon scanning microscope ( Bruker , Billerica , MA ) with a Vision II laser ( Coherent , Santa Clara , CA ) . Brains were continuously perfused in the saline used for dissection at 60 mL/hr . Once the sample was placed and centered under the objective , we waited 5 min before starting the experiment to avoid any lingering network activation from the dissection or transmission lights . Two-photon excitation wavelength was 920 nm , and power at the sample varied between 3 and 10 mW . CsChrimson was excited with trains of 2 ms long 590 nm light pulses via an LED ( M590L3-C1 , Thorlabs , Newton , NJ ) shone through the objective . The excitation light path contained a 605/55 nm bandpass filter and was delivered to the objective with a custom dichroic ( zt488-568tpc , reflecting between 568 nm and 700 nm ) . A 575 nm dichroic beam splitter and bandpass filters ( 525/70 nm and 607/45 nm for the green and red respectively ) were placed in the detection arm before photons reached the PMTs ( Hamamatsu multi-alkali ) . Instantaneous power measured out of the objective was roughly 50 μW/mm2 . Stimulus pulse trains were delivered at 30 Hz and the number of pulses varied between 1 , 5 , 10 , 20 and 30 — corresponding to total stimulation durations ranging from 2 ms to 1 s . Imaging fields of view were chosen as to avoid scanning regions containing CsChrimson-expressing neuropil while being as close as possible to the putative connection site , as we observed occasional two-photon-evoked slow activation of CsChrimson-expressing cells ( high-intensity two-photon stimulation of CsChrimson was used for spatially precise neuronal activation in [Kim et al . , 2017] ) . When this was impossible — for example , in self-activation controls or for completely overlapping cell types — we chose a large ROI of which the CsChrimson/GCaMP6m-expressing neuropile represented a small fraction , so as to minimize duty cycle . ROIs were kept constant throughout the experiment . Each experimental run consisted of four repeats , each approximately 16s long . Runs were themselves repeated every 2 min . All experiments started with five runs corresponding to the five stimulation strengths , in a random order . This was sometimes followed by pharmacological testing . At the end of the experiment , a high intensity 3D stack was acquired to check that the expression patterns were as expected , and that the neutrophil imaged was the targeted one in cases where fluorescence levels during the experiments were very low . At least six flies were tested for every pair considered . For blocking nicotinergic or inhibitory ( GABAergic or glutamatergic ) transmission , mecamylamine ( 50 μM ) or picrotoxin ( 10 μM ) ( Sigma-Aldrich , St Louis , MO ) were perfused by switching to a different line for 3 min , followed by a wash period during which the perfusion was drug-free again . Thirty pulses stimulation runs were repeated every 2 min , starting 4 min before the drug application and throughout the wash . Prior to use , solutions were kept frozen in 25 mM and 0 . 3 M aliquots , respectively . All analyses were performed in http://julialang . org/Julia , using custom-written routines . All data and code are available as an OpenScienceFramework project at https://osf . io/vsa3z/ ( Franconville , 2018a ) . Code is also centralized in a Github repository ( [Franconville , 2018b] , https://github . com/romainFr/CX-Functional-Analysis; copy archived at https://github . com/elifesciences-publications/CX-Functional-Analysis ) and notebooks recapitulating the analysis can be run directly from the browser at https://mybinder . org/v2/gh/romainFr/CX-Functional-Analysis/master ( using https://mybinder . org/Binder ) . An earlier version of this manuscript is available as a preprint at https://www . authorea . com/155729/_TsHpd9reMuWijjossgt6Q ( DOI: 10 . 22541/au . 151537454 . 41878908 ) . | Some of the most evocative discoveries in neuroscience have been those of internal representations , such as neural activity patterns that represent which direction an animal is facing and its place in its surroundings . Understanding how neurons connect to one another to form ‘circuits’ is crucial to understanding how these circuits maintain such representations . Many of the design principles that underlie circuit function in the brains of fruit flies apply to other animals . However , fly brains are easier to study because genetic tools can be used on them to selectively activate and image the activity of specific types of neurons . By activating one type of neuron and imaging the activity of another that may be connected to it , we obtain what is called a functional ‘connectome’: a map of neural connectivity that identifies different pathways that information can flow along . A region of the fly brain called the central complex is involved in many important behaviors , including navigation and sleep . Researchers know about the types of neurons in the region and about how the activity of some of them changes during different behaviors . However , obtaining the connectome of the central complex would make it easier to understand how the central complex works . A technique called optogenetics allows specific types of neurons to be activated one at a time by shining light onto them . By imaging the activity of neurons that might be connected to an optogenetically activated neuron , Franconville et al . have now built an extensive – albeit still incomplete – map of the connections within the central complex of fruit flies . The map reveals two key bottlenecks in the central complex circuit . Firstly , a neuron type in a substructure called the protocerebral bridge controls a lot of the information flowing through the circuit . Secondly , the circuit appears to have very few true ‘output’ neuron types – Franconville et al . identified only one . These results suggest that however complicated the computations performed by the central complex circuit might be , the output of the circuit , which likely guides the fly’s actions , may be much simpler . Franconville et al . have compiled the mapping results into an interactive website that makes the neuroscientific data both freely available and easily explorable . As researchers perform more such experiments , the new data can be added to the map . This information can be used to constrain theories and inspire new ideas about how the central complex does what it does . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Building a functional connectome of the Drosophila central complex |
The molecular underpinnings of synaptic vesicle fusion for fast neurotransmitter release are still unclear . Here , we used a single vesicle–vesicle system with reconstituted SNARE and synaptotagmin-1 proteoliposomes to decipher the temporal sequence of membrane states upon Ca2+-injection at 250–500 μM on a 100-ms timescale . Furthermore , detailed membrane morphologies were imaged with cryo-electron microscopy before and after Ca2+-injection . We discovered a heterogeneous network of immediate and delayed fusion pathways . Remarkably , all instances of Ca2+-triggered immediate fusion started from a membrane–membrane point-contact and proceeded to complete fusion without discernible hemifusion intermediates . In contrast , pathways that involved a stable hemifusion diaphragm only resulted in fusion after many seconds , if at all . When complexin was included , the Ca2+-triggered fusion network shifted towards the immediate pathway , effectively synchronizing fusion , especially at lower Ca2+-concentration . Synaptic proteins may have evolved to select this immediate pathway out of a heterogeneous network of possible membrane fusion pathways .
There are many examples where protein-mediated membrane fusion plays a key role , such as entry of enveloped viruses , fertilization , development , carcinogenesis , intracellular trafficking , secretion , and neurotransmitter release ( Rothman , 1994; Jahn et al . , 2003; Jahn and Scheller , 2006; Harrison , 2008; Rizo and Rosenmund , 2008; Moreau et al . , 2011 ) . Among these biological processes , synaptic vesicle fusion is unique in that it is both regulated and fast , in the order of a millisecond ( Meinrenken et al . , 2003 ) . Upon an action potential , at most one synaptic vesicle undergoes exocytosis in a synapse among the readily releasable pool ( Dobrunz and Stevens , 1997 ) , that is , only a small subset of docked synaptic vesicles undergo fusion . How synaptic proteins can achieve fast fusion in such a precisely regulated and Ca2+-dependent fashion is a matter of intense investigation and considerable controversy ( Sørensen , 2009; Collins et al . , 2012 ) . In this context , reductionist in vitro systems play an important role in uncovering the mechanism of action since they allow manipulations and observations not possible in vivo , and they establish if a particular subset of factors represents a minimal system for promoting fast Ca2+-triggered release . Protein-mediated fusion between biological membranes is thought to proceed from a pre-fusion contact to a fusion pore through one or more intermediates ( Chernomordik and Kozlov , 2008; Jackson and Chapman , 2008 ) . A hemifusion intermediate is defined as a membrane state where the outer leaflets of the two juxtaposed membranes exchange lipids; examples include an hourglass shaped ‘stalk’ and an elongated hemifusion diaphragm . Investigations of both protein-free and protein-mediated liposome fusion in model systems , along with macroscopic continuum models and atomistic computer simulations , revealed at least two pathways: a direct pathway from pre-fusion contact to complete fusion pore via a stalk , and the ‘classical’ pathway with a hemifusion diaphragm intermediate between the stalk and the fusion pore . The energetics and kinetics of these processes are dependent on many factors , including lipid composition and membrane curvature . Model calculations suggested that a hemifusion diaphragm is an energetically relatively stable , long-lived intermediate , and that the direct pathway from stalk to fusion pore may be faster ( Kuzmin et al . , 2001 ) . Several key proteins cooperate in Ca2+-triggered synaptic vesicle fusion , including the soluble N-ethylmaleimide-sensitive factor attachment protein receptors ( SNAREs ) ( Söllner et al . , 1993; Sutton et al . , 1998; Jahn and Scheller , 2006 ) , the Ca2+ sensor synaptotagmin 1 , the activator/inhibitor complexin , and SM proteins ( Rizo and Rosenmund , 2008; Südhof and Rothman , 2009 ) . Exocytosis can proceed by full-collapse fusion , where the opening of a small pore continues to expand to a large pore , and ‘kiss-and-run’ , where a small pore opens transiently but closes again as the vesicle stays associated with the plasma membrane ( Zhang et al . , 2009 ) . There are two macroscopic models for fusion pore formation: lipid-lined where the synaptic proteins act as scaffolds for pore formation and protein-lined where the transmembrane domains of some of the synaptic proteins play a more active role by lining the pore ( Jackson and Chapman , 2006; Ngatchou et al . , 2010 ) . However , atomistic computer simulations suggested that the molecular mechanism of pore formation is likely more complex than implied by simple macroscopic models ( Risselada et al . , 2011; Lindau et al . , 2012 ) . SNARE proteins can promote membrane fusion via hemifusion intermediates , as inferred from lipid-mixing studies with proteoliposomes where fluorescent labels measure the exchange of components between lipid bilayers ( Weber et al . , 1998; Reese and Mayer , 2005; Xu et al . , 2005; Schaub et al . , 2006; Liu et al . , 2008 ) . However , it is content mixing , that is , the diffusion between compartments of small molecules with size comparable or slightly larger to that of neurotransmitter molecules that is the correlate for neurotransmitter release . Several previous studies implied that content mixing is not necessarily correlated with lipid mixing for biological membrane fusion: neuronal SNAREs alone do not produce much content mixing , although they readily induce lipid mixing ( Bowen et al . , 2004; Kyoung et al . , 2011 ) , influenza virus-induced fusion content mixing occurs seconds after initial lipid mixing ( Floyd et al . , 2008 ) , and content mixing occurs with minutes delay after lipid mixing in vacuolar fusion ( Jun and Wickner , 2007 ) . Even inner leaflet mixing can occur without content mixing when fusion is induced by DNA-zippering ( Chan et al . , 2009 ) , although this intriguing observation will require verification with synaptic-protein induced fusion . The results presented in this work reveal an even more striking example of the difference between content mixing and lipid mixing for Ca2+-triggered fusion with synaptic proteins . Which pathway is used for fast synchronous neurotransmitter release ? Is it the classical hemifusion diaphragm pathway or a more direct pathway ? We investigated this fundamentally important question with our recently developed optical microscopy method that simultaneously monitors the temporal sequence of both content and lipid exchange upon Ca2+-triggering between single pairs of donor and acceptor vesicles on a 100-ms time scale ( Kyoung et al . , 2011; Kyoung et al . , 2012 ) . Donor and acceptor vesicles mimicked synaptic vesicles and the plasma membrane , respectively . Our system can discriminate between docking , hemifusion , and complete fusion . After improvements in optical instrumentation as well as protein expression and purification , here we achieved a Ca2+ sensitivity in the 250–500 μM range . This Ca2+ range is reasonably close to the physiological concentration range ( starting at 10 μM and saturating at several 100 μM ) ( Heidelberger et al . , 1994 ) and comparable to other recent in vitro experiments ( Wang et al . , 2011; Hernandez et al . , 2012 ) . Our system mimics a stepwise Ca2+ concentration increase that acts on the readily-releasable pool of primed synaptic vesicles , and it is thus reminiscent to experiments with live neurons using photolysis of caged Ca2+ compounds ( Schneggenburger and Neher , 2000; Sun et al . , 2007 ) . We anticipate that a different Ca2+ delivery technique for our system will allow us to mimic the transient Ca2+ pulses ( or trains of such pulses ) that are more typical for physiological neurotransmitter release . We combined our optical microscopy experiments with cryo-electron microscopy ( cryo-EM ) image analysis of mixtures of the same types of donor/acceptor vesicles before and after Ca2+-addition . We observed a variety of morphologies of membrane-membrane interfaces , as well as changes in the distribution of membrane–membrane interfaces upon Ca2+ addition . Taken together , we deciphered the pathway that allows synaptic vesicles to fuse immediately upon Ca2+-triggering . We found that all fast ( immediate ) fusion events start from a hemifusion-free state and proceed to full fusion upon Ca2+ injection without discernible hemifusion intermediates . In contrast , stable , initially hemifused states are slow to fuse , if at all . Moreover , we found that complexin dramatically increases the number of immediate fusion events , with a more pronounced effect at lower Ca2+ concentration ( 250 μM ) .
We modified our original single vesicle–vesicle microscopy system ( Kyoung et al . , 2011; Kyoung et al . , 2012 ) and improved protein quality in order to enable studies at 250–500 μM Ca2+-concentration ( for details , see ‘Materials and methods’ ) . As before , we used membrane compositions and protein number densities that mimic synaptic vesicles and the plasma membrane , respectively , and reconstituted full-length synaptotagmin 1 together with synaptobrevin ( also referred to as VAMP , Vesicle Associated Membrane Protein ) , and syntaxin together with SNAP-25 into two separate populations of liposomes , termed donor and acceptor vesicles , respectively . We had previously established the homogeneity of our vesicle preparations ( Kyoung et al . , 2011 ) . Unlabeled acceptor vesicles were tethered to a PEG-coated glass surface , and the donor vesicles were labeled with self-quenched lipid ( DiD ) and content ( sulforhodamine B ) fluorophores ( Figure 1A ) . After a defined incubation period our system started from a metastable state of docked vesicles at zero Ca2+ concentration . Upon Ca2+ injection we monitored ‘changes’ in membrane state in real-time by visual inspection of individual content and lipid mixing fluorescence intensity time traces and identification of significant jumps in these traces . 10 . 7554/eLife . 00109 . 003Figure 1 . Single vesicle–vesicle microscopy for monitoring changes in the membrane and contents upon Ca2+-injection . ( A ) Labeling and reconstitution scheme of our single-vesicle lipid/content mixing system . Other factors and proteins can be added to the system . The detailed experimental protocol is described in ( Kyoung et al . , 2012 ) with modifications as described in ‘Materials and methods’ . ( B ) Representative real-time fluorescence intensity traces with donor ( synaptobrevin and synaptotagmin 1 ) and acceptor ( syntaxin and SNAP-25 ) vesicles ( red/upper traces: lipid dye fluorescence intensity , green/lower traces: content dye fluorescence intensity ) . Time point 0 indicates the instance of Ca2+ injection at 500 μM ( blue vertical lines ) . Shown are instances of ‘immediate’ fusion , as defined by a content dye fluorescence intensity jump during the first 600 ms time bin upon Ca2+-injection along with a simultaneous jump in lipid dye fluorescence intensity , ‘delayed’ fusion , as defined by a content dye fluorescence intensity jump during a subsequent time bin , and ‘hemifusion-only’ , as defined by a lipid dye fluorescence intensity jump without a content dye fluorescence intensity jump during the observation period of 50 s . For illustration purposes , only 30 s of the observation period are shown since there was no change in the subsequent 20 s . ( C ) Bar graph of the percentage of immediate , delayed fusion , and hemifusion-only events involving SNAREs and synaptotagmin 1 . The bar graph was normalized with respect to the number of fluorescent spots that exhibited at least one lipid-mixing event during the observation period . Note , that there are fewer immediate fusion events compared to our previous experiments that were carried out at higher Ca2+ concentration ( Kyoung et al . , 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00109 . 003 Representative examples for Ca2+-injection at 500 μM are shown in Figure 1B for experiments with full-length SNAREs and full-length synaptotagmin 1 . We observed three characteristic changes in membrane state , including instances of immediate fusion ( a correlate with synchronous neurotransmitter release ) , delayed fusion ( a correlate with asynchronous release ) , and hemifusion-only ( i . e . , without complete fusion ) during the observation period of 50 s . Overall , for this very minimal system consisting of only SNAREs and synaptotagmin 1 , immediate fusion instances were in the minority , followed by delayed fusion instances , and then hemifusion-only instances ( Figure 1C ) . As we had shown previously , Ca2+ triggered fusion depended on the presence of both functional SNAREs and synaptotagmin 1 , for example , mutation of one of the Ca2+ binding sites of synaptotagmin 1 greatly reduced the amount of immediate fusion and disruption of SNARE complex formation prevented both docking and fusion ( Kyoung et al . , 2011 ) . In order to correlate the temporal sequence of changes in membrane state observed by single vesicle–vesicle microscopy with membrane morphology , we imaged mixtures of the same types of donor/acceptor vesicles before ( Figure 2A–C ) and approximately 35 s after addition of Ca2+ ( Figure 2D–F ) by cryo-EM . We used the same defined incubation period for the donor/acceptor vesicle mixture prior to Ca2+ addition as for the single vesicle–vesicle microscopy experiments ( 30 min ) . Many ‘point contacts’ ( see ‘Materials and methods’ for definitions of all contact and interface types ) between donor and acceptor vesicles were observed before Ca2+ addition ( Figure 2B ) , as well as hemifusion diaphragms ( Figure 2C ) . Both point contacts and hemifusion diaphragms are likely between donor and acceptor vesicles since we did not observe interactions between vesicles of the same type in previous cryo-EM experiments ( Kyoung et al . , 2011 ) . After Ca2+ addition , besides point contacts and hemifusion diaphragms ( Figure 2D , E ) , we also observed a smaller number of ‘extended’ close contacts without hemifusion , defined as a triple-layered feature extending over more than 10 nm in length ( Figure 2F ) . Interestingly , similar extended close contacts were induced between liposomes by denatured Munc18 ( Xu et al . , 2011 ) , and by SNAREs alone when the membrane-proximal layer of the SNARE was disrupted by mutation ( Hernandez et al . , 2012 ) . Since we observed these hemifusion diaphragms and extended close contacts by cryo-EM imaging approximately 35 s after Ca2+-injection , such interfaces are expected to be part of ‘slow’ , that is , delayed fusion pathways . 10 . 7554/eLife . 00109 . 004Figure 2 . Imaging of donor/acceptor interface morphologies by cryo-EM before and after Ca2+ addition . ( A–F ) Cryo-EM images of mixtures of donor ( synaptobrevin and synaptotagmin 1 ) and acceptor ( syntaxin and SNAP-25 ) vesicles before ( A–C ) and approximately 35 s after ( D–F ) 500 μM Ca2+ addition ( panels B , C , E , and F are close-up views ) . Vesicles were imaged in the holes of the substrate carbon film , visible as the darker areas in the image , in conditions that clearly show the lipid bilayers ( ‘Materials and methods’ ) . The particular images shown in this figure were selected out of total of 16 ( before Ca2+ addition ) and 21 ( after Ca2+ addition ) EM micrographs , respectively , with emphasis on showing point contacts before Ca2+ addition and extended interfaces after Ca2+ addition in order to illustrate the variety of these particular states . Arrows indicate interfaces between vesicles that are approximately perpendicular to the direction of the projection ( see ‘Materials and methods’ for definitions of all contact and interface types ) . Large black arrow: point contact ( representative close-up in B ) , small black/white arrow: hemifusion diaphragm ( representative close-ups in C and E ) , and large white arrow: extended close contact ( representative close-up in F ) . Scale bars in A and D are 100 nm , and 20 nm in B , C , E , and F . ( G ) Distribution of vesicle sizes before and after Ca2+ addition . Vesicle diameters were calculated from all cryo-EM images in both conditions ( ‘Materials and methods’ ) . ( H ) Bar graph of the percentage of various vesicle interfaces , that is , point contacts , hemifusion diaphragms , and extended close contacts ( including a few instances of mixed , i . e . , extended/hemifused , interfaces ) , normalized with respect to the total number of interfaces observed before and after addition of 500 μM Ca2+ , respectively ( ‘Materials and methods’ ) . In addition to the changes in the distribution of vesicle interfaces upon Ca2+ addition , some amount of complete fusion between vesicles occurred as indicated by the shift of the diameter distribution in panel G towards larger values . Source files of all cryo-EM micrographs used for the quantitative analysis are available in the Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00109 . 00410 . 7554/eLife . 00109 . 005Figure 2—source data 1 . Source files for cryo-EM data . This zip archive contains all cryo-EM images used for the quantitative analyses shown in Fig . 2 . The folder named “No_Ca++” contains the images before Ca++ addition ( individual files are named P3_1_** . tif or jpg ) , and folder named “With_Ca++” contains the images ∼35s after Ca++ addition ( individual files are named P3_3_** . tif or jpg ) . Images were collected in low dose conditions at 200 kV acceleration voltage on a CM200 FEG electron microscope ( FEI ) with a 2k × 2k Gatan UltraScan 1000 camera , at 50 , 000× magnification and 1 . 5 mm underfocus . The full resolution data were exported as 16 bit “tif” files ( 2048 × 2048 pixels , scale 0 . 2 nm/pixel at specimen ( the corresponding files have the extension “tif” ) . Note that these files cannot not be viewed with a standard picture viewer , but must be viewed with a program , such as “ImageJ” . To facilitate easier viewing , the original images were converted to smaller ( 1024×1024 , 0 . 4 nm/pixel ) , contrast adjusted jpeg images ( 8 bits ) for easy and immediate visualization with commonly used picture viewers ( the corresponding files have the extension “jpg” ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00109 . 005 We performed a quantitative analysis of all observed cryo-EM images . As expected from our single-vesicle microscopy fusion experiments with identical vesicle preparations , some fusion occurred after Ca2+-addition , shifting the diameter distribution of vesicles observed in the cryo-EM images towards larger vesicles than before Ca2+ addition , that is , there are fewer small vesicles and more larger vesicles than before Ca2+-addition ( Figure 2G ) . The mean diameter increased from 61 . 3 ± 18 . 6 ( s . d . ) nm to 65 . 1 ± 23 . 9 ( s . d . ) nm ( a significant increase with 97% confidence from a Student's t-test ) , corresponding to an increase of mean volume from 1 . 6 × 105 nm3 to 2 . 2 × 105 nm3 . Moreover , the distribution of vesicle interfaces changed after Ca2+ addition ( Figure 2H ) : the percentage of point contacts decreased from 55% to 40% , the percentage of hemifused diaphragms increased from 45% to 49% , and the percentage of extended interfaces jumped from zero to 11% . We analyzed the image density of point contacts at zero Ca2+; two representative examples are shown in Figure 3A . The cryo-EM images of most point contacts indicated that membranes are close to each other at the tip of the contact but not forming a membrane stalk ( an hourglass shaped connection between two bilayers that involves the merger of outer leaflets ) since the observed minimal separation between membranes ( ranging from 4 . 9 to 2 . 3 nm ) was larger than the critical distance for membrane stalk formation ( 0 . 9 nm ) ( Aeffner et al . , 2012 ) . Interestingly , a similar membrane apposition between liposomes could be induced by the soluble C2AB domain of synaptotagmin 1 and Ca2+ , but without SNAREs ( Araç et al . , 2006 ) , as well as with SNAREs alone ( Hernandez et al . , 2012 ) ; in this context it has been suggested that synaptotagmin may act as a distance regulator ( van den Bogaart et al . , 2011b ) . 10 . 7554/eLife . 00109 . 006Figure 3 . Image density profile analysis of selected vesicle–vesicle interfaces . ( A ) Image density profile analysis of two representative point contacts between vesicles at zero Ca2+ . Dotted lines indicate 2 nm thick sections selected from the cryo-EM images . These sections were used to generate the profiles shown to the left and the right of the close-up views in the center ( for details , see ‘Materials and methods’ ) . ( B ) Image density profile analysis of a transition from extended close contact to hemifusion diaphragm after addition of 500 μM Ca2+ . As in panel A , selected sections are indicated by dotted lines , and the corresponding profiles are shown the left and right of the close-up view in the center . DOI: http://dx . doi . org/10 . 7554/eLife . 00109 . 006 We also analyzed the images density of a mixed interface ( Figure 3B ) , which likely represents a transition state from extended contact to hemifusion diaphragm along a particular fusion pathway . The estimated distance between the two bilayers in the extended contact region is less than 1 nm ( Figure 3B , middle panel , left half of the image ) , which is representative for the distances observed for other extended close contacts that are not hemifused . Considering the size of synaptic proteins ( SNARE complex: approximately 11 . 6 × 2 . 6 nm , synaptotagmin C2 domains: approximately 3 × 5 nm ) , they would have to act at the periphery of many of the observed vesicle interfaces . The point contact is a special case since it is the only interface type where the proteins would be close to a putative transitory stalk to initiate fusion . Indeed , as mentioned above , quantitative analysis of all cryo-EM images revealed that the fraction of point contacts decreased upon Ca2+ addition ( Figure 2H ) . Inspired by this relative decrease after addition of Ca2+ , we hypothesized that some of these initial point contacts are the likely starting point for immediate fusion upon Ca2+ injection; those that did not fuse might be ready to undergo fusion during subsequent pulses . Our single vesicle–vesicle microscopy system provided the tool to test if point contacts are the starting points for immediate fusion events . With an extension of our system , we were able to deduce the initial state of the membrane interface ( docked or hemifused ) at the time of Ca2+-injection for each individual fusion event; a change in the single vesicle–vesicle membrane interface during the incubation period would result in a significant difference between the initial lipid dye fluorescence intensity and that after the incubation period ( Figure 4 and ‘Materials and methods’ ) . Remarkably , no significant lipid fluorescence intensity changes ( i . e . , changes larger than the noise level ) were observed during the incubation period for ‘all’ instances of immediate fusion events ( Figure 4A ) , in other words , these events did not start from a hemifused state . In contrast , Ca2+-triggered content mixing rarely occurred during the observation period of 50 s when starting from a stable hemifused state ( Figure 4B ) . Since cryo-EM imaging of the membrane interfaces prior to Ca2+-addition indicated that there were only two classes , point contacts and hemifusion diaphragms , it follows that all immediate Ca2+-triggered fusion events must start from point contacts that are not hemifused . The ability to characterize the initial membrane state and the temporal sequence of events upon Ca2+ injection , along with the ability of controlling the constituents , is another illustration of the power of our single vesicle–vesicle microscopy system . 10 . 7554/eLife . 00109 . 007Figure 4 . Upon Ca2+-injection , immediate fusion events start form hemifusion-free point contacts whereas initially hemifused states are slow to fuse . ( A ) Representative real-time fluorescence intensity traces of instances of immediate fusion with imaging of initial florescence intensity levels right after addition of donor ( synaptobrevin and synaptotagmin 1 ) to acceptor vesicles ( syntaxin and SNAP-25 ) ( red/upper traces: lipid dye fluorescence intensity , green/lower traces: content dye fluorescence intensity ) . Time point 0 indicates the instance of Ca2+ injection at 500 μM ( blue vertical lines ) . As in Figure 1 , immediate fusion is defined as a content dye fluorescence intensity jump that occurs during the first 600 ms time bin upon Ca2+ injection . In all cases , no significant ( i . e . , above noise level ) lipid fluorescence intensity change was observed during the incubation period , which excludes the possibility of a hemifused initial state before Ca2+ triggering . Moreover , ‘all’ ( seven ) observed traces with immediate fusion upon Ca2+ injection exhibited this behavior . ( B ) Representative real-time fluorescence intensity traces of cases where the membranes are hemifused prior to Ca2+-injection . Initial hemifusion was characterized by a significant increase ( i . e . , above noise level ) in only lipid dye fluorescence intensity between the initial recording and at the end of incubation period at 0 Ca2+ . For approximately 400 fluorescent spots co-localized in both the initial recording and at the end of the incubation period , we observed 51 traces showing ( 1 ) an increase in lipid dye fluorescence and ( 2 ) no change in content dye fluorescence intensity during the incubation period but before the Ca2+ injection . For most of the 51 cases , no content mixing was observed upon Ca2+-injection during the observation period of 50 s except for two cases ( see one example in the bottom panel ) where delayed content mixing occurred at a later time . For illustration purposes only 20 s of the 50 s observation period are shown all traces but one since there was no change in the subsequent 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 00109 . 007 As mentioned above , the observed cryo-EM images of point contacts indicate that there is space for synaptic proteins close the tip of the contact between the membranes , although less likely right at the tip . However , few significant densities ( greater than a factor of two above noise level ) were actually visible near the point contacts , with some notable exceptions ( e . g . red arrow in Figure 5A ) . This lack of ubiquitous protein densities could be attributed to the low number of synaptic proteins that are expected to be involved in docking and exocytosis ( van den Bogaart and Jahn , 2011 ) ; proteins could also be overlapping with the interface in projection and be masked by the high-contrast lipids . To support the notion that the few observed instances of significant densities could be related to synaptic proteins , we overlaid two alternative models of the synaptotagmin 1 • SNARE complex to the image of a point contact with the highest contrast , assuming that the marked density is related to the globular synaptotagmin C2 domains ( Figure 5B , C ) . The apparent lack of density for the SNARE complex agrees with the notion of a partially folded trans-SNARE complex that is stabilized by repulsive forces between solvated membranes ( Gao et al . , 2012 ) , as approximately indicated in our models . The single molecule FRET efficiency data of the synaptotagmin 1 • SNARE complex indicated multiple binding modes ( Choi et al . , 2010 ) , with the top solution shown in Figure 5B . An alterative model of this complex ( Figure 5C ) was based on NMR chemical shift perturbation experiments of the complex between synaptotagmin 1 and the SNARE complex ( Dai et al . , 2007 ) . Both models predict that it is the C2B domain that interacts with the SNARE complex , although the exact interface and orientation between the molecules is different . While approximate , these conceptual models may serve as guide to speculate about the involvement of synaptic proteins in forming the point contact , and setting the stage for Ca2+-triggered fusion . For example , the position of synaptotagmin 1 in our models would enable Ca2+-triggered membrane juxtaposition ( Araç et al . , 2006 ) and membrane bending ( Hui et al . , 2009; McMahon et al . , 2010 ) . Membrane bending or buckling could in turn destabilize the membrane , and in conjunction with full zippering of the SNARE complex , lead to fusion pore opening . 10 . 7554/eLife . 00109 . 008Figure 5 . Modeling of putative protein density in a pre-fusion point contact state . ( A ) Close up view of Figure 3A , top middle panel , around a significant density feature ( greater than a factor of two above noise level ) that is not part of the membranes ( red arrow ) . Only the density features on the left-hand side of the panel were considered since the carbon grid ( dark feature in the lower right corner in Figure 3A , top middle panel ) may have affected protein interactions . ( B ) Overlay of the model of the complex of the C2A–C2B fragment of synaptotagmin 1 ( cyan ) and the neuronal SNARE complex ( synaptobrevin—blue , syntaxin—red , SNAP-25—green ) based on 34 single-molecule FRET experiments ( Choi et al . , 2010 ) . The last four C-terminal helical turns of synaptobrevin were modeled as a random coil . ( C ) Overlay of a model of the complex of the C2B domain of synaptotagmin 1 and the neuronal SNARE complex based on proximity between the C2B domain polybasic region and residues D186 , D193 of SNAP-25 , both of which show sizeable chemical shift perturbations in NMR experiments upon complex formation , among other perturbations ( Dai et al . , 2007 ) . Six C-terminal α-helical turns of synaptobrevin and one α-helical turn of syntaxin were modeled as a random coil . The transmembrane domains were modeled as α-helices , and the linkers as random coils . See ‘Materials and methods’ for details . DOI: http://dx . doi . org/10 . 7554/eLife . 00109 . 008 The variety of observed fusion pathways induced by the minimal system of full-length SNAREs and full-length synaptotagmin 1 upon Ca2+ addition suggest a key role for other synaptic factors to enhance the immediate ( fast ) fusion pathway from point contact to fusion pore . To test this hypothesis , we included in our system the synaptic protein complexin which has both activating and inhibiting roles for Ca2+-evoked and spontaneous neurotransmitter release in vivo , respectively ( Giraudo et al . , 2009; Maximov et al . , 2009 ) . By design , our in vitro system with SNAREs and synaptotagmin 1 probes the function of complexin for Ca2+-evoked release . Strikingly , the content mixing histogram showed more pronounced short-time scale content mixing upon Ca2+ injection than in the absence of complexin ( Figure 6A ) . The content mixing histogram is related to the probability of fusion vs time , a mimic for evoked postsynaptic currents . Indeed , the content mixing histogram in the presence of complexin could be fitted to a biphasic exponential decay function where the fast fraction with a time constant of 0 . 36 s accounts for 68% of all events . At lower Ca2+ concentration ( 250 μM ) , the effect of complexin on the fusion probability time trace became even more pronounced ( Figure 6B ) with 90% in the fast fraction . In contrast , the fusion probability time trace was best fit to a single exponential decay function with a time constant of 3 s in the absence of complexin , which is a significant difference to the content mixing histogram in the presence of complexin . 10 . 7554/eLife . 00109 . 009Figure 6 . Effect of complexin on Ca2+-triggered fusion probability . Probability of fusion vs time upon 500 μM ( A ) or 250 μM ( B ) Ca2+-injection ( i . e . , occurrence of Ca2+-triggered content-mixing events vs time ) with and without 5 μM complexin . The same donor ( synaptobrevin and synaptotagmin 1 ) and acceptor ( syntaxin and SNAP-25 ) vesicles were used as in Figure 1 . Histograms were normalized to the number of respective fluorescence intensity time traces showing at least one lipid-mixing event during the observation period of 50 s ( see ‘Materials and methods’ for details ) . For illustration purposes only 20 s of the 50 s observation period are shown since there were few events in the subsequent 30 s; a time binning of 600 ms was used . Black lines are fits to exponential decay functions over the entire 50 secs . For 500 μM Ca2+ , in the absence of complexin the fitted function is f ( t ) = 0 . 0015 + 0 . 096e−t/1 . 93 and it is 10 times more likely compared to a two-exponential fit , and in the presence of 5 μM complexin the fitted function is f ( t ) = 0 . 0021 + 0 . 084e−t/2 . 38 + 0 . 18e−t/0 . 36 and it is 107 times more likely compared to a single exponential fit . For 250 μM Ca2+ , in the absence of complexin the fitted function is f ( t ) = 0 . 0026 + 0 . 027e−t/2 . 98 and it is three times more likely compared to a two-exponential fit , and in the presence of 5 μM complexin the fitted function is f ( t ) = 0 . 0005 + 0 . 10e−t/0 . 89 + 0 . 011e−t/14 . 5 and it is 1014 times more likely compared to a single exponential fit . DOI: http://dx . doi . org/10 . 7554/eLife . 00109 . 009
As mentioned in the ‘Introduction’ , lipid mixing ( exchange between membranes ) is necessary for subsequent content mixing ( complete fusion ) , but it is not sufficient . Moreover , it is content mixing that is relevant for neurotransmitter release , not lipid mixing . As a further demonstration of this principle , complexin had little effect on the lipid mixing histogram upon 500 μM Ca2+ injection ( Figure 8A ) . In marked contrast , complexin had a large effect on the corresponding content mixing histograms ( Figure 6A ) . Likewise , at 250 μM Ca2 the effect of complexin is more pronounced on the content mixing histogram compared to the lipid mixing histogram ( compare Figures 6b and 8b ) . This large difference in complexin-induced lipid mixing and content mixing effects greatly amplifies previous observations for biological membrane fusion ( Bowen et al . , 2004; Jun and Wickner , 2007; Floyd et al . , 2008; Chan et al . , 2009; Kyoung et al . , 2011 ) . Taken together , these observations , and especially the results presented in this work for Ca2+ triggered fusion , call into question conclusions in the literature drawn from lipid-mixing assays when they were not validated by observations related to content mixing . It is thus essential that biological fusion experiments monitor a quantity that is directly related to content mixing in order to distinguish between different membrane states , such as docking , hemifusion , and full fusion . The combination of single vesicle–vesicle microscopy and cryo-EM imaging experiments allowed us to decipher fusion pathways that were induced by the combination of full-length SNAREs , full-length synaptotagmin 1 , and , in some experiments , complexin . Before Ca2+ addition , the majority ( 55% ) of interacting vesicles formed point contacts , while the remaining vesicle interfaces consisted of hemifusion diaphragms ( Figure 2H ) with SNAREs and synaptotagmin 1 . Only a small fraction of vesicles underwent spontaneous ( complete ) fusion during the incubation period . Starting from hemifused vesicles , little complete fusion was observed upon Ca2+-injection ( Figure 4B ) . In contrast , starting from point contacts , we observed a heterogeneous network of both immediate and delayed fusion pathways upon Ca2+ injection ( Figure 7 ) . It is remarkable that the very minimal system of just SNAREs and synaptotagmin 1 exhibited Ca2+ triggered activity , including instances of immediate fusion ( Figure 1C ) . The delayed fusion pathways proceeded on a second to minute time scale via long-lived extended contact or hemifusion intermediates ( Figure 7 ) . 10 . 7554/eLife . 00109 . 010Figure 7 . Model of multiple fusion pathways upon Ca2+ injection starting from a point contact . Complexin synchronizes Ca2+-triggered fusion by increasing the number of immediate fusion processes from point contact to fusion pore opening , relative to delayed fusion pathways involving stable hemifusion diaphragms and other long-lived intermediates . DOI: http://dx . doi . org/10 . 7554/eLife . 00109 . 010 Our experiments revealed the pathway for immediate Ca2+-triggered fusion by synaptic proteins ( Figure 7 ) . It starts from a hemifusion-free point contact and proceeds to full fusion without discernible intermediates on the 100-ms time scale . This pathway is based on: ( 1 ) observation of only two membrane interface classes prior to Ca2+-injection by cryo-EM , point contacts and hemifusion diaphragms ( Figure 2H ) . ( 2 ) Single vesicle–vesicle fluorescence intensity time traces that recorded both the initial state of the system as well as the temporal sequence of events upon Ca2+ injection ( Figure 4A ) . We found that all instances of immediate fusion events upon Ca2+-triggering started from interacting membranes that were not hemifused . Taken together , it follows that all immediate fusion events started from hemifusion-free point contacts . It is of course possible that this immediate pathway proceeds from point contact to fusion pore via a transient stalk , however , the lifetime of such a transient stalk would have to be faster than the 100-ms time resolution of our instrument . Our experiments definitely rule out the existence of a stable hemifusion intermediate for fast Ca2+-triggered fusion . Our discovery of the immediate fusion pathway ( point contact to full fusion ) raised the hypothesis that for neurotransmitter release , other synaptic proteins introduce a preference for this immediate pathway upon Ca2+-triggering . We confirmed this hypothesis by observing that complexin ( in combination with SNAREs and synaptotagmin 1 ) favors this immediate fusion pathway upon Ca2+ triggering , a result that correlates well with complexin's activating role in vivo for synchronous release upon an action potential ( Maximov et al . , 2009 ) . Specifically , we observed that complexin significantly increased the number of immediate fusion events that occur right upon Ca2+-injection ( Figure 6 ) . This effect is especially pronounced at the lowest Ca2+ concentration that we tested ( 250 μM ) . Thus , complexin suppresses the emergence of long-lived hemifusion intermediates , effectively leading to a higher probability of immediate fusion upon Ca2+-triggering . Moreover , complexin was recently found to prevent spontaneous fusion in the absence of Ca2+ in another model system with proteoliposomes ( Malsam et al . , 2012 ) . There are two distinct N-terminal sequence elements that required for complexin's activating role in vivo , one of which ( the accessory α-helix ) clamps SNARE complexes ( Krishnakumar et al . , 2011 ) . Our results suggest a possible explanation activating function of complexin by imposing geometric restraints on trans SNARE complexes through inter-SNARE complex interactions ( Kümmel et al . , 2011 ) , and thereby possibly preventing the formation of long-lived hemifusion intermediates upon Ca2+-triggering . Of course , other explanations are also possible , and such molecular mechanisms could be tested in the future by combining single vesicle–vesicle fusion experiments with single molecule observations . A recent study showed that SNAREs alone can produce a variety of spontaneous ( i . e . , without Ca2+-triggering ) fusion pathways on the minute time scale at elevated temperature ( 30°C ) ( Hernandez et al . , 2012 ) , including point contacts , as well as extended close contacts , and hemifusion diagrams . Thus , biological membranes are poised to undergo fusion via a number of different pathways once they are brought into close proximity by the action of SNAREs . This variety of pathways could have provided a ‘noisy’ background for evolutionary selection of the immediate pathway that we discovered here . Factors such as complexin may have evolved to select this immediate pathway out of all possible pathways , and may have offered a distinct advantage for fast Ca2+-evoked release and efficient communication between neurons .
Full-length rat proteins of syntaxin 1A , SNAP-25A , synaptobrevin 2 , synaptotagmin 1 , and complexin 1 were expressed and purified essentially as described previously ( Kyoung et al . , 2011 ) . In order to achieve higher yield and purity for the single vesicle–vesicle microscopy experiments , we applied considerable improvements as detailed below . As before , we used a cysteine-free mutant of SNAP-25A ( C84S , C85S , C90S , and C92S ) , and the single site mutants of syntaxin ( S193C ) and synaptobrevin ( S28C ) ; the latter mutants offered the option for fluorophore labeling and single molecule number density experiments ( Kyoung et al . , 2011 ) . As before , full-length synaptotagmin 1 was expressed in Sf9 insect cells ( Invitrogen , Grand Island , NY ) , purified by Ni2+-nitrilotriacetic acid ( NTA ) sepharose ( Qiagen , Hilden , Germany ) affinity chromatography , followed by his-tag cleavage , size exclusion chromatography , and , finally , ion exchange chromatography . SNAP-25 was expressed with an N-terminal TEV cleavable his-tag from plasmid pTEV5 ( Rocco et al . , 2008 ) in BL21 ( DE3 ) Escherichia coli cells ( Novagen , EMD Chemicals , Gibbstown , NJ ) and purified by Ni2+-NTA sepharose affinity chromatography . After removal of the his-tag by overnight cleavage with TEV protease , the sample was further purified by size exclusion chromatography using a Superdex 200 10/300 column ( GE Healthcare , Uppsala , Sweden ) in buffer containing 20 mM HEPES , pH 7 . 5 , 100 mM NaCl , and 4 mM dithiothreitol ( DTT ) . Complexin was expressed as a his-tagged protein from vector pET28a ( Novagen , EMD Chemicals , Gibbstown , NJ , USA ) in BL21 ( DE3 ) . Overnight cultures ( 4 l ) were grown in autoinducing LB medium ( Studier , 2005 ) at 30°C , harvested , lysed , and then purified using the protocol described in reference ( Kyoung et al . , 2011 ) . Full-length rat syntaxin and synaptobrevin were expressed with a N-terminal , TEV protease cleavable , hexa-histidine tag from plasmid pTEV5 ( Rocco et al . , 2008 ) . Proteins were expressed overnight at 25°C in autoinducing media ( Studier , 2005 ) in strain C43 ( Miroux and Walker , 1996 ) . Cell pellets from a 8 l of culture were suspended in 400 ml of 50 mM NaPi pH 8 , 1 M NaCl , 5 mM EDTA , and 1 mM PMSF supplemented with Complete Protease Inhibitor Cocktail tablets ( Roche , Basel , Switzerland ) , and broken by three passes through a M-110-EH microfluidizer ( Mircrofluidics Corp . , Newton , MA ) at 15 , 000 PSI . Inclusion bodies were removed by two consecutive 10 min 10 , 000 RPM spins in a JA-14 ( Beckman Coulter , Brea , CA ) rotor , and the membrane fraction collected by centrifugation at 40 , 000 RPM for 2 hr in a Ti-45 ( Beckman Coulter , Brea , CA ) rotor . Membranes containing syntaxin were washed with 10 mM Tris-H2SO4 , pH 7 . 5 , 10 mM EDTA , 10% ( wt/vol ) glycerol , centrifuged at 40 , 000 RPM for 1 . 5 hr in a Ti-45 rotor , the pellet resuspended in 20 mM HEPES , pH 7 . 5 , 500 mM NaCl , and 1 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) , and centrifuged for an additional 1 . 5 hr in the same rotor . Membranes were suspended to a concentration of 5 mg/ml in 20 mM HEPES , pH 7 . 5 , 500 mM NaCl , 1 mM TCEP , 10 mM imidazole , 1 mM PMSF and EDTA-free Complete Protease Inhibitor Cocktail ( Roche , Basel , Switzerland ) . Dodecylmaltoside was added to 2% , and after incubation at 4°C for 1 hr , the sample was centrifuged for 35 min at 55 , 000 RPM in a Ti-70 ( Beckman Coulter , Brea , CA ) rotor , and the supernatant loaded onto a 0 . 75 ml column of Ni-NTA agarose ( Qiagen , Hilden , Germany ) . The column was washed with 20 mM HEPES , pH 7 . 5 , 300 mM NaCl , 1 mM TCEP , 20 mM imidazole , 110 mM octylglucoside ( OG ) , and the protein eluted in that buffer containing 500 mM imidazole and 1 M NaCl . 1 mM EDTA was immediately added to each fraction , and those fractions containing protein were loaded onto a Superdex 200 HR 10/30 ( GE Healthcare , Uppsala , Sweden ) that was equilibrated with 20 mM HEPES , pH 7 . 5 , 300 mM NaCl , 1 mM TCEP , 110 mM OG . Protein fractions were pooled , and digested with TEV protease for 1 hr at ambient temperature , after which the reaction was complete and the TEV protease had precipitated . TEV was removed by centrifugation at 5000 RPM in an Eppendorf ( Hamburg , Germany ) model 5804 R tabletop centrifuge . All proteins were reconstituted into vesicles within 1 to 2 hr after the final purification steps in order to prevent potential degradation and aggregation . Proteins were reconstituted into acceptor ( syntaxin/SNAP-25 ) and donor ( synaptobrevin/synaptotagmin 1 ) vesicles with a detergent depletion method as previously described ( Kyoung et al . , 2011; Kyoung et al . , 2012 ) . Briefly , lipid films and other membrane components with compositions that mimic synaptic vesicle and active zone membranes were dissolved in 110 mM OG buffer and purified proteins ( synaptobrevin/synaptotagmin 1 and syntaxin for donor and acceptor vesicles , respectively ) were added . The synaptobrevin and syntaxin protein to lipid ratio was 1:200; the synaptotagmin 1 to synaptobrevin molar ratio was 1:4 . 6 , consistent with observations of purified synaptic vesicles ( Takamori et al . , 2006 ) . For acceptor vesicles , SNAP-25 solution ( five times the concentration of syntaxin ) was added to the protein–lipid mixture in order to prevent formation of ‘dead-end’ 2:1 syntaxin/SNAP-25 complexes . 3 . 5 mol% phosphatidylinositol 4 , 5-bisphosphate ( PIP2 ) was added to the acceptor vesicles only; this concentration is within the range of that observed in the plasma membrane of PC12 cells ( van den Bogaart et al . , 2011a ) . Detergent free buffer ( 20 mM HEPES , pH 7 . 4 , 90 mM NaCl , 1% 2-mercaptoethanol ) was added to the protein–lipid mixture until the detergent concentration was at the critical micelle concentration and the solutions were purified with a CL4B column and dialyzed overnight with Bio-beads SM2 ( Bio-rad , Hercules , CA ) in detergent-free ‘Vesicle Buffer’ ( 20 mM HEPES , pH 7 . 4 , 90 mM NaCl , 20 µM EGTA , 1% 2-mercaptoethanol ) . Donor vesicles were labeled with DiD and formed in the presence of 50 mM sulforhodamine B ( Invitrogen , Grand Island , NY ) , prior to size exclusion chromatography and dialysis . Reconstituted vesicles were used within 1 to 2 days for optical microscopy and cryo-EM experiments . The homogeneity of the reconstituted donor and acceptor vesicles obtained from our reconstitution protocol was extensively tested with cryo-EM , light scattering , single molecule counting experiments in order to determine the protein number distributions in single vesicles , SDS gel electrophoresis of reconstituted vesicles , and determination of the directionality of reconstituted proteins as previously described ( Kyoung et al . , 2011 ) . Furthermore , in this work , for each experimental condition we performed multiple experiments from different protein preparations and reconstitutions to ensure that the results are not dependent on a particular preparation . In vitro ensemble assays with reconstituted proteoliposomes have been widely used to study SNARE-mediated membrane fusion ( Weber et al . , 1998 ) . They were important to establish that SNAREs have some fusogenic activity . However , nearly all of these studies monitored lipid mixing , that is the exchange of lipids between membranes . Despite the ease of such ensemble lipid mixing assays , conclusions drawn in the absence of content mixing indicators were often incomplete , or in some cases , they did not correlate well with physiological observations ( Sørensen , 2009 ) . This is not surprising since lipid mixing is necessary for content mixing , but it is not sufficient ( compare Figures 6 and 8 ) . 10 . 7554/eLife . 00109 . 011Figure 8 . Histograms of the occurrence of Ca2+-triggered lipid-mixing events with and without 5 μM complexin , corresponding to the experiments shown in Figure 6 . ( A ) Ca2+ injection at 500 μM . ( B ) Ca2+ injection at 250 μM . Black lines are fits to exponential decay functions over the entire 50 s . For 500 μM Ca2+ , in the absence of complexin the fitted function is f ( t ) = 0 . 0014 + 2 . 66e−t/0 . 11 + 0 . 22e−t/1 . 94 and it is 1018 times more likely compared to a single exponential fit , and in the presence of 5 μM complexin the fitted function is f ( t ) = 0 . 0006 + 0 . 42e−t/0 . 81 + 0 . 05e−t/4 . 76 and it is 1010 times more likely compared to a single exponential fit . For 250 μM Ca2+ , in the absence of complexin the fitted function is f ( t ) = 0 . 0023 + 0 . 13e−t/0 . 61 + 0 . 05e−t/9 . 05 and it is 106 times more likely compared to a single exponential fit , and in the presence of 5 μM complexin the fitted function is f ( t ) = 0 . 003 + 0 . 33e−t/0 . 51 + 0 . 05e−t/6 . 34 and it is 1014 times more likely compared to a single exponential fit . DOI: http://dx . doi . org/10 . 7554/eLife . 00109 . 011 Content mixing measurements have been notoriously difficult to achieve for ensemble-based assays because of a number of technical hurdles , including potential leakiness of proteoliposomes , aggregation , and vesicle rupture that may plague ensemble experiments . Such phenomena may produce a large fluorescence intensity change that cannot easily be distinguished from genuine fusion . Another issue with commonly used lipid-mixing ensemble experiments is that they cannot distinguish between docking and hemifusion/fusion since the observed lipid-mixing signal depends on both processes ( Cypionka et al . , 2009 ) . Even measuring content mixing in an ensemble experiment would not be able to discern effects caused by differences in docking and fusion since the ensemble-average fluorescence signal depends on both docking and complete fusion . On a different note , some of the first ensemble lipid mixing studies used an unreasonably high protein to lipid ratio ( e . g . , 1:10 for synaptobrevin-reconstituted vesicles [Weber et al . , 1998] ) ; this is a concern since high protein concentrations are known to cause vesicle instabilities . Finally , ensemble measurements may obscure heterogeneous fusion pathways since they only observe averages rather than individual fusion events . This is indeed a problem for ensemble experiments since single-vesicle lipid-mixing experiments revealed multiple intermediates in SNARE-mediated fusion ( Yoon et al . , 2008; Karatekin et al . , 2010 ) , and the single vesicle–vesicle content/lipid mixing results presented in this work uncovered heterogeneous fusion pathways for Ca2+-triggered fusion in the presence SNAREs , synaptotagmin , and complexin . Even at a single vesicle level , lipid mixing is not necessarily indicative of content mixing , especially since it is often not possible to resolve two distinct lipid mixing events that would correspond to outer and inner leaflet mixing of a single interacting vesicle pair , respectively . Furthermore , many of these experiments were typically performed on mixtures of donor/acceptor vesicles in solution at a constant Ca2+ concentration , rather than starting from the metastable state of interacting vesicle pairs at zero Ca2+ used in our system ( see below ) . This introduces the possibility of synaptotagmin C2 cis-interactions with its own membrane when Ca2+ is present prior to docking of vesicles ( Vennekate et al . , 2012 ) , and thereby reducing synaptotagmin trans-interactions with the acceptor vesicles . The single vesicle lipid mixing ( that is , without content monitoring ) experiments by ( Lee et al . , 2010 ) are a case in point . In their experiments a paradoxical ‘decrease’ of Ca2+-triggered lipid mixing from 10 to 100 μM was observed , assuming background level at 100 μM . In contrast , our system exhibits the more expected increase in fusion ( content mixing ) as the Ca2+ concentration is increased ( Figure 6 ) . A more detailed comparison of single vesicle systems is available in ( Kyoung et al . , 2012 ) . To overcome the shortcomings of both ensemble and single-vesicle lipid-mixing-only assays , we developed a single vesicle–vesicle method that simultaneously monitors the temporal sequence of changes in both lipid and content mixing using characteristic dequenching ‘jumps’ of two spectrally distinct fluorescent dyes , DiD and sulforhodamine B , respectively ( Kyoung et al . , 2011; Kyoung et al . , 2012 ) . The size of the soluble content dye sulforhodamine B is only slightly larger to that of certain neurotransmitters , such as glutamate or serotonin . For the content dye , dequenching may occur by content mixing , leakage , or photobleaching . Complete fusion is characterized by a rapid jump in content dye fluorescence followed by relatively steady fluorescence intensity . Leakage leads to a quick spike , followed by rapid disappearance of fluorescence intensity . Photobleaching leads to slow increases and then decreases of fluorescence intensity . Thus , events that do not lead to content mixing can be easily distinguished from leakage and photobleaching for each individual vesicle pair . Our single vesicle–vesicle system starts from a metastable state of docked vesicle pairs that is established by an incubation period at zero Ca2+ concentration . The incubation period at zero Ca2+ is followed by Ca2+ injection at a defined concentration . Our system discriminates between docking and fusion since , by definition , we only monitor lipid and content mixing for vesicles that are interacting , that is , docked vesicles . During the incubation period at zero Ca2+ concentration , some vesicles already undergo hemifusion , but a sizeable population ( typically , more than 50% ) will be available after the incubation period that has not yet hemifused or fused ( Figures 2 and 4 ) . Extensive characterizations and controls were performed to validate our system ( Kyoung et al . , 2011 ) . To our knowledge this is the only in vitro system that monitors the temporal sequence of both content and lipid mixing events between single vesicle pairs on a 100-ms time scale , differentiates between docking , hemifusion , complete fusion , and vesicle bursting , that triggers the fusion process by injection of Ca2+ starting from a metastable state of interacting vesicle pairs that have been incubated at zero Ca2+ , and that allows the addition of other factors during the incubation and observation stages . In our initial studies we applied a relatively high Ca2+-concentration in order to test our system under a variety of conditions ( Kyoung et al . , 2011 ) . This was a reasonable strategy at an early stage of technique development and proof-of-principle studies . After improvements in optical instrumentation ( see below ) as well as protein expression and purification ( described above ) , we are now able to routinely perform experiments at 250–500 μM Ca2+ . For the experiments described in this work , we used a prism type total internal reflection ( TIR ) setup in order to maximize the useable field of view . In addition , we employed one excitation wavelength ( 532 nm ) to excite both content and lipid dyes simultaneously rather than using two lasers as in our previous setup . This approach produced a better signal-to-noise ratio of the content dye fluorescence intensity time traces , and reduced photobleaching of the lipid dyes . The surface preparation , acceptor vesicle immobilization , and donor vesicle incubation were performed essentially as previously described ( Kyoung et al . , 2011; Kyoung et al . , 2012 ) with modifications as detailed here . A PEG-coated glass surface was prepared as described in Diao et al . , 2012 and then incubated with a neutravidin solution ( pH 7 . 5 , 50 µg/ml , 20 mM HEPES , 90 mM NaCl ) for 10 min followed by a buffer wash in order to prevent non-specific surface binding . Control experiments with acceptor-free PEG-coated surface were performed in order to ensure that non-specific binding of donor vesicles was rare; furthermore , we had previously performed a control experiment with a soluble synaptobrevin fragment to disrupt SNARE complex formation which resulted in loss of donor vesicle binding to surface-tethered acceptor vesicles ( Kyoung et al . , 2011 ) . The effect of surface tethering on vesicle interactions and fusion has been shown to be negligible ( see Figure S7 in [Yoon et al . , 2006] and Figure 3 in [Diao et al . , 2010] ) . Unlabeled acceptor ( syntaxin/SNAP-25 ) vesicles were immobilized on the modified surface , and excess vesicles were removed by thoroughly washing with Vesicle Buffer . Donor vesicles ( approximately 500× dilution ) encapsulating sulforhodamine B and labeled with DiD were introduced into the sample chamber . After 20–30 min incubation , excess vesicles were removed by extensive washing with Vesicle Buffer followed by Ca2+ injection using a motorized syringe pump with a flow rate of 33 µl/s . For the experiments that included complexin , 5 μM complexin and approximately 500× diluted donor vesicles were incubated at ambient temperature for 20–30 min followed by a Vesicle Buffer wash in the presence of complexin . All experiments were performed at ambient temperature . It should be noted that experiments in live neurons by photolysis of caged Ca2+ compounds were also carried out at ambient temperatures ( Schneggenburger and Neher , 2000; Sun et al . , 2007 ) . In principle , our system is suitable for studies at other temperatures as well , although the stability of the vesicles would have to be carefully checked at elevated temperatures . For the experiments shown in Figure 4 we performed fluorescence intensity acquisition right at the start of the experiment . We determined the state of the membrane interface ( docked or hemifused ) right before Ca2+ injection by assessing if changes occurred during the incubation period ( laser illumination was interrupted during the incubation period to prevent photobleaching ) . We injected donor vesicles ( at 50× dilution ) for 30 s followed by a buffer wash , and then immediately recorded the fluorescence intensities of the lipid dyes and content dyes for short periods ( 1–4 s ) . A 20 min period followed without laser illumination to prevent photobleaching . The laser illumination resumed 5 s prior to injection of Ca2+ into the sample chamber . Since the sample remained on the microscope stage during the 20 min incubation period , the positions of the docked vesicles remained unchanged . Content and lipid dye fluorescence intensity time traces from individual single vesicle pairs were analyzed using the single molecule software developed in Dr . Taekjip Ha's lab at University of Illinois ( Diao et al . , 2012 ) . A 200 ms time binning was used for recording and a 600 ms moving average was used for data analysis . The lipid and content mixing events are characterized by significant fluorescent intensity jumps in their respective recorded time traces ( see representative examples in Figures 1B and 4 ) . Typically five to ten time points immediately before and after the jump were used to evaluate the fluorescence intensity change as well as the noise level . A jump was considered ‘significant’ if the change in fluorescence intensity ΔI was greater than the average noise level σ = √ ( σ12 + σ22 ) where σ1 and σ2 are the standard deviations of I before and after the jump . For the data used in this work , the typical signal to noise ratio ( SNR = ΔI/σ ) was in the range 4 to 10 . For the histograms shown in Figures 6 and 8 , time differences between Ca2+ injection and instances of lipid mixing and content mixing were determined by inspection of the individual time traces; the time stamp of Ca2+ injection was defined as the instance of the first lipid-mixing event among all docked vesicles . Histograms of these time differences were generated for lipid and content mixing instances , respectively . Histograms were normalized with respect to the number of docked vesicles that exhibited at least one lipid dye fluorescence intensity jump during the observation period . Histograms were fitted to decay functions with one exponential and a sum of two exponentials , respectively , and the fit used that is more likely correct based on the Akaike information criterion implemented in OriginPro 8 . 6 ( OriginLab , Inc . ) . For the experiments shown in Figures 6 and 8 at 500 μM Ca2+-injection in the absence of complexin we observed 245 lipid mixing events and 106 content mixing events ( out of a total of 1794 spots ) , at 500 μM Ca2+-injection in the presence of 5 μM complexin we observed 248 lipid mixing events and 150 content mixing events ( out of a total of 1852 spots ) , at 250 μM Ca2+-injection in the absence of complexin we observed 157 lipid mixing events and 55 content mixing events ( out of a total of 3085 spots ) , and at 250 μM Ca2+-injection in the presence of 5 μM complexin we observed 215 lipid mixing events and 100 content mixing events ( out of a total of 3726 spots ) . The histograms are combinations of a total of five independent experiments for each of the four conditions . Frozen-hydrated samples of SNARE-containing vesicles prior to , and approximately 35 s after , Ca2+ addition were prepared using the procedures for observation in cryo-EM as described previously ( Kyoung et al . , 2011 ) . Briefly , samples were observed by cryo-EM in low dose conditions using a CM200F electron microscope ( FEI , Hillsboro , OR ) operating at 200 kV . Images of both conditions were collected at a 50 , 000× magnification and 1 . 5 µm under-focus on a 2k × 2k Gatan UltraScan 1000 camera ( Gatan Inc . , Pleasanton , CA ) . We observed many vesicle–vesicle interactions that were away from the holey carbon grid ( Figure 2 ) . Cryo-EM images of individual donor and acceptor vesicle populations ( in Supplementary figure S3 , panels A , B in [Kyoung et al . , 2011] ) showed no pronounced interactions between same vesicles , so the grid has a negligible effect , if any , on vesicle interface formation . Analyses were performed from a total of 16 ( without Ca2+ ) and 21 ( with Ca2+ ) cryo-EM micrographs with good contrast ( selected images are shown in Figure 2 and all raw images are available in Figure 2—source data 1 ) . Vesicle diameters were marked with two points for each vesicle using the ‘boxer’ feature in EMAN ( Ludtke et al . , 1999 ) . Diameters in nm were calculated from the pairs of coordinates using Spider ( Frank et al . , 1996 ) . The vesicle diameter distribution histograms ( Figure 2G ) were calculated with Matlab ( MathWorks , Inc . ) using 10 nm binning for all vesicles for which an entire section was visible in a particular image ( a total of 228 and 237 vesicles in the images at zero Ca2+ and after Ca2+ addition , respectively ) . A Student's t-test was performed to assess the significance of the increase of the mean of the vesicle diameter distribution after Ca2+ addition . In the observed cryo-EM images the most prominent features of lipid membranes are the electron-dense head groups , rich in phosphorous atoms , which appear as high contrast dark lines when observed on end ( along the direction of the electron beam—see for example [Tahara and Fujiyoshi , 1994; Lambert et al . , 1997] ) . A liposome bilayer membrane appears as two parallel black lines at its circumference . Hence , contacts between liposomes can be classified depending on the number of dark lines and their respective distance at the interface . All types of interfaces in our data could be divided into four classes , defined as follows: For image density measurements , small 80 × 80 pixel images of vesicle interface regions were selected and cut out of micrographs using the ‘boxer’ feature in EMAN ( Ludtke et al . , 1999 ) , then processed in Spider ( Frank et al . , 1996 ) in order to generate average density profiles at selected points along the interface . Briefly , images were rotated so that the lipid bilayers at the interface were aligned with the Y axis , the contrast was flattened ( using ramp removal ) and inverted , and then average density was calculated over 10 pixel lines , corresponding to 2-nm thick sections perpendicular to the interface , to improve the signal to noise ratio . Average densities were plotted out as function of distance ( nm ) to measure the separation between lipid layers ( maxima in density plots ) of the vesicle interface regions . For the models presented in Figures 5B , C , we assumed that the significant non-bilayer density feature corresponds to the C2 domains of synaptotagmin 1 , that the Ca2+ binding loops of synaptotagmin 1 are close to the bilayer , that the C-terminal end of the first α-helix of SNAP-25 is close to bilayer in order to allow interaction of the adjacent palmitoylated cysteine residues with the membrane , and that the transmembrane domains of syntaxin and synaptobrevin are close to the point contact between vesicles . These assumptions produced a reasonable overlay with the cryo-EM image . For the model shown in Figure 5B , we estimated that four α-helical turns would be able to bridge a 3 . 0 nm distance ( Rrms ) in a random coil conformation , using a persistence length lp of 0 . 9 based on single molecule FRET experiments of synaptobrevin ( Choi et al . , 2011 ) ( Rrms2 = 2lpn × 0 . 36 nm , where n is the number of residues which is 14 for two heptad coiled coil repeats ) . Thus , the last four C-terminal helical turns of synaptobrevin were modeled as a random coil in order to approximately account for the distance that a portion of the synaptobrevin polypeptide chain would have to extend in order to reach the vesicle membrane . For the model shown in Figure 5C , six C-terminal α-helical turns of synaptobrevin and one α-helical turn of syntaxin were modeled as a random coil in order to approximately account for the required extended linkers to the transmembrane domains . | The central nervous system relies on electrical signals travelling along neurons and through synapses at high speeds . Signals often have to pass between two neurons , or from a neuron to a muscle fiber , and the nervous system relies on a process called membrane fusion to ensure that the neurotransmitter molecules that carry the signal across the synapses are released as quickly as possible . Membrane fusion is an important process in many areas of biology , including intracellular transport and fertilization , but it occurs much faster ( millisecond time scale ) in the nervous system than anywhere else in the body . The reasons for this have long been a mystery , although calcium ions are known to trigger the fusion process . The fusion of two biological membranes is similar in many regards to the way that small soap bubbles merge together to form large bubbles . Just as soap bubbles can form a variety of discernible intermediate structures when they merge , so can biological membranes . This means that it is possible to produce a so-called hemifusion intermediate in which the outer layers of the membranes have merged , but the inner layers have not , so it is not possible for anything—such as serotonin , dopamine and other neurotransmitter molecules—to transfer from one membrane to the other . Diao et al . have used a combination of advanced optical imaging and cryogenic electron microscopy to explore membrane fusion between synthetic membranes that contained reconstituted synaptic proteins , including synaptotagmin and a family of protein receptors called SNAREs . When calcium ions were injected into the synthetic system , the basic characteristics of neurotransmitter release—such as membrane fusion on a millisecond time scale—was observed . Contrary to some theories of membrane fusion , the fastest fusion events did not begin or proceed via a discernible hemifusion intermediate state . Rather , these events proceeded from a ‘point contact’ state in which the membranes were close to each other ( just 1–5 nm apart ) without being fused , and were ready to undergo fast fusion once the calcium ions had been injected . And when Diao et al . introduced a protein called complexin , which is known to be important for fast neurotransmitter release in vivo , they observed more immediate fusion events and fewer events that involved a hemifusion intermediate . With a synthetic system it is possible to perform experiments that are currently not possible with live neurons , and this has allowed Diao et al . to clarify the roles of the individual components in the process of membrane fusion , and could prove useful in efforts to develop novel therapeutic treatments to combat neurological disorders . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"neuroscience"
] | 2012 | Synaptic proteins promote calcium-triggered fast transition from point contact to full fusion |
The kinesin I family of motor proteins are crucial for axonal transport , but their roles in dendritic transport and postsynaptic function are not well-defined . Gene duplication and subsequent diversification give rise to three homologous kinesin I proteins ( KIF5A , KIF5B and KIF5C ) in vertebrates , but it is not clear whether and how they exhibit functional specificity . Here we show that knockdown of KIF5A or KIF5B differentially affects excitatory synapses and dendritic transport in hippocampal neurons . The functional specificities of the two kinesins are determined by their diverse carboxyl-termini , where arginine methylation occurs in KIF5B and regulates its function . KIF5B conditional knockout mice exhibit deficits in dendritic spine morphogenesis , synaptic plasticity and memory formation . Our findings provide insights into how expansion of the kinesin I family during evolution leads to diversification and specialization of motor proteins in regulating postsynaptic function .
Synapse maturation and remodeling are crucial for brain functions including learning and memory . The postsynaptic sites of excitatory synapses are located on the dendritic spines , which undergo dynamic structural changes that are essential for experience-driven wiring of the neuronal network ( Trachtenberg et al . , 2002 ) . More than 1000 proteins with diverse structures and functions have been identified in the postsynaptic density ( PSD ) ( Bayés et al . , 2011 ) , and a tight regulation of their abundance and localization is essential for proper synapse development and plasticity . Many of the postsynaptic proteins are locally translated in dendrites , which allows spatial and temporal regulation of molecular composition of individual synapses in response to local extracellular stimuli ( Holt and Schuman , 2013 ) . To achieve protein synthesis in dendrites , mRNAs synthesized in the soma need to be assembled in ribonucleoproteins ( RNPs ) and transported over long distances by molecular motors along microtubule ( Doyle and Kiebler , 2011 ) . Kinesin and dynein superfamilies of proteins are microtubule-dependent molecular motors that mediate long-distance transport of materials in neuron . The kinesin superfamily is very diverse and contains 45 members in mammal . It is sub-divided into 14 different families based on structural similarity ( Hirokawa et al . , 2010 ) . The kinesin I family ( encoded by the Kif5 genes ) contains the founding kinesin protein kinesin heavy chain ( KHC ) ( Brady , 1985; Vale et al . , 1985 ) . While only one single KIF5 is present in invertebrates such as Drosophila , C . elegans and Aplysia , gene duplication events give rise to three homologous KIF5 genes ( Kif5a , Kif5b and Kif5c ) in vertebrates ( Miki et al . , 2001 ) . Unlike KIF5B which is ubiquitously expressed , KIF5A and KIF5C are mostly expressed in neuron ( Kanai et al . , 2000 ) . Functional redundancy has been demonstrated among the three KIF5s , as exogenous expression of KIF5A or KIF5C can rescue the impaired mitochondrial transport in cells lacking KIF5B ( Kanai et al . , 2000 ) . In contrast , specific function of individual KIF5 has been reported in zebrafish , in which axonal transport of mitochondria depends only on KIF5A but not the other two KIF5s ( Campbell et al . , 2014 ) . Furthermore , only KIF5A dysfunction leads to seizure and the neuromuscular disorder Hereditary Spastic Paraplegia ( Fink , 2013; Nakajima et al . , 2012 ) . It is therefore plausible that the expansion of the Kif5 gene family during evolution enables functional specificity of individual KIF5 in the vertebrate brain , although the molecular basis of the specificity has not been identified . The three KIF5s contain motor , stalk , and tail domains ( Friedman and Vale , 1999 ) , and they all bind to kinesin light chain ( KLC ) which mediates interaction with some of the cargoes ( Kamal et al . , 2000; Morfini et al . , 2016 ) . Despite the overall structural similarity , the carboxyl-termini ( starting from around amino acid 934 until the last amino acid ) of the three KIF5s are very different , which may confer the individual KIF5 distinctive functions in neurons . Previous studies have mostly focused on KIF5 function in axonal transport because the motor domain of KIF5 preferentially moves out of dendrites into axons , and KIF5 function is negatively regulated by the dendritic protein MAP2 ( Gumy et al . , 2017; Huang and Banker , 2012; Kapitein et al . , 2010; Tas et al . , 2017 ) . However , all three KIF5s are co-purified with RNPs , and dominant-negative KIF5 disrupts the dendritic localization of RNA-binding proteins ( Kanai et al . , 2004 ) . Additional dendritic cargoes for KIF5 , including the AMPA receptor/GRIP1 complex and GABAA receptor , have also been identified ( Heisler et al . , 2014; Nakajima et al . , 2012; Setou et al . , 2002; Twelvetrees et al . , 2010 ) . KIF5s therefore likely participate in both axonal and dendritic transport . Despite previous studies on its importance on AMPA receptor trafficking ( Kim and Lisman , 2001; Setou et al . , 2002; Hoerndli et al . , 2013; Heisler et al . , 2014 ) , the role of KIF5 on dendritic spine morphogenesis and synaptic plasticity has not been comprehensively examined . In this study , we aim to investigate whether the three KIF5s have specific roles in the development and function of excitatory synapses on the postsynaptic neuron , and what might underlie the functional specificity . Here we report that KIF5B but not KIF5A is specifically involved in the development of excitatory synapses of postsynaptic neurons and dendritic transport of the RNA-binding protein fragile X mental retardation protein ( FMRP ) . The diverse carboxyl-termini of KIF5A and KIF5B determine their functional specificity , and we further identified arginine methylation of KIF5B as a novel post-translational modification ( PTM ) in regulating cargo binding . Because of the embryonic lethality of KIF5B knockout mice that precludes their use to study the synaptic and cognitive functions of adult brain in vivo , we generate mice with KIF5B conditional knockout in CaMKIIα-expressing neurons . The KIF5B conditional knockout mice exhibit altered dendritic spine structural plasticity in vivo , as well as deficits in synaptic plasticity and memory formation . Our study strongly suggests that homologous motor proteins of the kinesin I family have non-redundant functions in regulating the development and function of excitatory synapses that is crucial for learning and memory .
To compare the synaptic functions of different KIF5s , we mainly focus on neurons from the hippocampus , a brain region that is important for learning and memory and where the development of excitatory synapses is well-studied . We first determined the expression of different KIF5s in the hippocampus along development . Although KIF5C was previous reported to be expressed exclusively in medulla and spinal cord ( Kanai et al . , 2000 ) , Kif5c mRNA is detected in the developing hippocampus in Allen Brain Atlas . Expression data for Kif5a and Kif5b transcripts in the developing brain is not available , but transcripts encoding the three KIF5s are detected in the adult mouse hippocampus in the atlas . Previous study has reported that Kif5 mRNAs expression is unchanged in cultured hippocampal neurons along maturation in vitro ( Silverman et al . , 2010 ) . On the other hand , we found that all three KIF5 proteins showed similar developmental expression profiles in the hippocampus , with the expression more prominent at early postnatal stages and significantly reduced at later postnatal and adult stages ( Figure 1A ) . Next , we examined the distribution of KIF5 protein in the brain by fractionation . All three KIF5s were detected in the synaptic plasma membrane fraction ( Figure 1B ) , which is consistent with the proteomic study reporting the presence of three KIF5s in the PSD ( Bayés et al . , 2011 ) . Many functional studies on KIF5s employ over-expression of dominant-negative constructs , which contain cargo-binding domains of the kinesin but lacking motor domains , thereby disrupting cargo movement through competitive binding . Here we attempt to address the role of individual KIF5 by specifically depleting each KIF5 homolog in neurons using RNA-interference . Three short hairpin RNAs ( shRNAs ) were created that specifically targeted KIF5A , KIF5B , and KIF5C . The knockdown efficiency and specificity of each shRNA in neuron were confirmed by Western blot and immunofluorescence staining ( Figure 1—figure supplement 1 ) . To examine the effect on excitatory synaptic transmission , whole-cell patch recording was performed in hippocampal neurons transfected with shRNAs targeting different KIF5s together with GFP construct . We found that knockdown of individual KIF5 differentially affected excitatory synaptic transmission . Compared to control shRNA , knockdown of KIF5B resulted in the most profound and significant reduction in the frequency of miniature excitatory synaptic current ( mEPSC ) , while knockdown of KIF5C did not affect mEPSC frequency or amplitude . Notably , introduction of KIF5A-shRNA did not change the mEPSC frequency but instead significantly increased the mEPSC amplitude ( Figure 1C ) . Since the shRNA and GFP constructs were introduced to the neurons using calcium phosphate precipitation which has very low transfection efficiency , the reduction of mEPSC frequency in the GFP-positive neuron was likely due to cell-autonomous decrease in synapse number on the postsynaptic neuron instead of change in presynaptic release . To test this hypothesis , the density of different types of dendritic spines was examined . Although knockdown of either KIF5B or KIF5C caused a significant reduction in the density of mushroom spines , only the introduction of KIF5B-shRNA increased the density of filopodia . On the other hand , knockdown of KIF5A did not cause any change in the density of mushroom spines or filopodia when compared to control neurons ( Figure 1D ) . The differential effect of KIF5A and KIF5B knockdown on spine morphogenesis and synaptic transmission is not attributed to differences in knockdown efficiency , as either shRNA reduced the target KIF5 expression by similar levels ( Figure 1—figure supplement 1 ) . Taken together , knockdown of KIF5B in hippocampal neurons leads to more profound changes in mEPSC and dendritic spine morphogenesis than knockdown of KIF5C , while knockdown of KIF5A has no effect on dendritic spines . To confirm that KIF5A and KIF5B indeed differentially regulate dendritic spine morphogenesis and to exclude potential off-target effect of the KIF5B-shRNA , rescue experiments using different KIF5s were performed . We focus on mushroom spines instead of the other three spine types in subsequent experiments because mushroom spines are regarded as mature spines that are more stable and possess the excitatory PSD ( Bourne and Harris , 2007; Berry and Nedivi , 2017 ) . Moreover , among the different spine types only mushroom spines were reduced after KIF5B knockdown , and the fewer mushroom spines correlated well with the decrease in mEPSC frequency . As expected , co-expression of KIF5B reversed the loss of mushroom spines induced by the KIF5B-shRNA . However , co-expression of KIF5A with the KIF5B-shRNA failed to rescue the loss of mushroom spines ( Figure 2A ) . In contrast , co-expression of KIF5C fully reversed the mushroom spine defects induced by the KIF5B-shRNA ( Figure 2B ) , suggesting that KIF5B and KIF5C share similar function on excitatory synapse development . Both endogenous and exogenously expressed KIF5A and KIF5B were present in dendrites and dendritic spines , and the percentage of dendritic spines containing endogenous KIF5A was even higher than that of KIF5B ( Figure 2—figure supplement 1 ) . These findings indicate that KIF5A and KIF5B have intrinsically distinct functions on excitatory synapses , although both KIF5A and KIF5B can be found in dendritic spines . KIF5 protein structure is divided into three domains: a motor domain , two coiled-coil domains which together form the stalk , and the tail domain ( Friedman and Vale , 1999 ) . Since the carboxyl termini , the most diverse regions between the KIF5s , represent part of the cargo-binding tail domain ( Morfini et al . , 2016; Nakajima et al . , 2012 ) ( Figure 3A ) , we next ask whether the three KIF5s might bind to cargoes differentially . We examine several different dendritic cargoes including the RNA-binding proteins ( RBPs ) FMRP and Ras GTPase-activating protein-binding protein ( G3BP1 and G3BP2 ) , which have been shown to regulate dendritic spine maturation ( Dictenberg et al . , 2008 ) , as well as the AMPA receptor subunit GluA2 . Pull-down assay using carboxyl-terminal fragments of individual KIF5s revealed that FMRP was preferentially pulled down by KIF5B and KIF5C but not KIF5A , while all three KIF5s could pull down G3BPs and GluA2 ( Figure 3B ) . Next , we examined whether knockdown of KIF5A and KIF5B differentially affects the dendritic localization and transport of FMRP . Neurons were co-transfected with GFP-FMRP and tdTomato , which labels the dendritic arbors and spines , together with the control shRNA , KIF5A-shRNA , or KIF5B-shRNA , followed by spinning disk confocal live imaging . Consistent with previous study on the trafficking of RBPs ( Mitsumori et al . , 2017 ) , most FMRP granules were either stationary or exhibiting oscillatory movement , while a small proportion showing unidirectional or bidirectional movement . Compared to control shRNA , knockdown of KIF5B significantly reduced the density of FMRP granules on dendrites . Interestingly , KIF5B shRNA only significantly decreased the density of stationary but not motile granules . In contrast , knockdown of KIF5A caused a general increase in the density of motile granules while decreasing the stationary granules , resulting in no net change in the density of total granules ( Figure 3C–D ) . There was no effect on the motility of the unidirectional and bidirectional granules after knocking down either KIF5A or KIF5B ( Figure 3—figure supplement 1 ) . To further characterize the effect on FMRP function in dendrite , the localization of two FMRP-cargoes , CaMKIIα and Grin2b mRNAs , was examined using fluorescent in situ hybridization ( FISH ) upon knockdown of KIF5A or KIF5B , and the distribution of mRNA puncta along individual dendrites was analyzed . Consistent with the reduced density of GFP-FMRP granules , knockdown of KIF5B also significantly reduced the density of both CaMKIIα and Grin2b mRNA puncta on dendrites ( Figure 3E ) . In contrast , knockdown of KIF5A did not affect CaMKIIα and Grin2b mRNA density on dendrite . Together these findings indicate that KIF5A and KIF5B differentially regulate the dendritic transport of FMRP and its mRNA cargoes . What is the molecular basis of the functional specialization of KIF5A and KIF5B ? The presence of a longer carboxyl-terminus in KIF5A which is very diverse from the corresponding regions of KIF5B and KIF5C ( Figure 3A ) prompt us to explore if it represents an inhibitory constraint for cargo binding . Towards this end , we created a truncated KIF5A construct with the carboxyl-terminal lacking the last 88 amino acids , as well as a chimeric KIF5A in which the last 88 amino acids were substituted by the shorter carboxyl-terminus of KIF5B . Either one of these constructs but not the wild-type KIF5A was able to pull down FMRP from the synaptoneurosome ( SNS ) , suggesting that the carboxyl-terminus of KIF5A indeed inhibits binding of specific cargoes ( Figure 4A ) . Remarkably , when shRNA targeting KIF5B was introduced into hippocampal neurons to induce loss of mushroom spines , co-expression of the chimeric KIF5A that contained the carboxyl-terminus of KIF5B was able to reverse the spine phenotype ( Figure 4B ) . These findings indicate that the last 88 amino acids of KIF5A prevent the motor protein to promote dendritic spine maturation , while its substitution by the shorter carboxyl terminus of KIF5B is sufficient to regain its synaptic function . Amino acid sequence alignment of the carboxyl termini of different KIF5s revealed the presence of two arginine residues ( Arg-941 and Arg-956 ) followed by glycine residues ( the RGG motif ) in KIF5B that are conserved across different vertebrates . KIF5C contains only the Arg-941 but not Arg-956 , while these two RGG motifs are absent in the KIF5A carboxyl-terminus ( Figure 5A ) . The RGG motifs often undergo arginine methylation , which involves the addition of methyl group to the guanidine nitrogen atom of arginine and is catalyzed by the protein arginine methyltransferases ( PRMT ) ( Najbauer et al . , 1993 ) . Hundreds of arginine-methylated proteins in the adult mouse brain have recently been identified by mass spectrometry ( Guo et al . , 2014 ) , and our data mining results indicated that KIF5B was one of the methylated proteins . Although arginine methylation is a well-established mechanism in the regulation of gene transcription and splicing in the nucleus ( Bedford and Clarke , 2009 ) , emerging studies have indicated their function outside the nucleus , in particular their importance in synaptic functions ( Penney et al . , 2017 ) . We therefore investigate whether arginine methylation represents a novel post-translational mechanism in regulating kinesin functions . We first confirmed the arginine methylation of KIF5B and KIF5C but not KIF5A when exogenously expressed in 293 T cells ( Figure 5B ) . Using reciprocal immunoprecipitation with antibodies that recognize the mono-arginine methylation within glycine-rich region or KIF5B , we confirmed that KIF5B was methylated in the synaptoneurosome ( Figure 5C ) . To determine whether the two conserved RGG sequences within the carboxyl-terminus of KIF5B are indeed the major methylation sites , we substituted the two arginine residues to histidine by site-directed mutagenesis , which retained the positive charges of the residues but could not undergo PRMT-mediated methylation . The KIF5B R941H or R956H mutant showed reduced methylation , whereas arginine methylation was absent in the double mutant ( R941/956H ) in which both arginine residues were substituted by histidine ( Figure 5D ) . These results indicate that R-941 and R-956 are the two major methylation sites of KIF5B . To ask whether and how arginine methylation affects KIF5B function , pull-down experiments were performed using the wild-type or methylation-deficient mutant ( R941/956H ) of KIF5B . The amount of FMRP and G3BP1 pulled down by the methylation-deficient mutant was significantly reduced when compared to wild-type KIF5B ( Figure 5E ) . To address whether arginine methylation is required for the synaptic function of KIF5B , we first compared the activity of wild-type and methylation-deficient mutant in the formation of mushroom spines using the KIF5B-shRNA rescue experiments . Co-expression of wild-type KIF5B reversed the loss of mushroom spines induced by the knockdown of KIF5B , while the methylation-deficient KIF5B failed to rescue the mushroom spine loss ( Figure 5F ) . Moreover , co-expression of wild-type but not the methylation-deficient KIF5B with the KIF5B-shRNA significantly increased the mEPSC frequency , ( Figure 5G ) . These results are consistent with the hypothesis that arginine methylation at the carboxyl-terminus is essential for KIF5B function on dendritic spine development and synaptic transmission , and suggesting a mechanism through regulating cargo-binding . Since KIF5B homozygous knockout is embryonic lethal ( Tanaka et al . , 1998 ) , we generated a KIF5B conditional knockout ( Kif5b-/- ) mice using the Cre/loxP gene-targeting strategy to study the function of KIF5B in vivo . CaMKIIα promoter-driven Cre transgenic line ( CaMKIIα-Cre ) ( Tsien et al . , 1996 ) and Kif5bfl/fl mice ( Cui et al . , 2011 ) were used to generate heterozygous ( CaMKIIα-Cre;Kif5bfl/+ , Hetero ) and homozygous ( CaMKIIα-Cre;Kif5bfl/fl , Homo ) conditional knockout mice in CaMKIIα-expressing neurons , which started the expression of Cre-recombinase after birth ( Dragatsis and Zeitlin , 2000; Tsien et al . , 1996 ) ( Figure 6A ) . Both homozygous and heterozygous mice were viable , and the homozygous mice did not differ in the general appearance or brain size from the wild-type ( Figure 6—figure supplement 1 ) . Analysis of whole-brain lysate showed a significant reduction of KIF5B protein level in homozygous knockout mice when compared to wild-type , and importantly there were no significant changes in the expression of KIF5A and KIF5C ( Figure 6B ) , or the dendritic kinesin KIF17 which is crucial for synaptic plasticity and memory formation ( Song et al . , 2009; Yin et al . , 2011; Franker et al . , 2016 ) ( Figure 6—figure supplement 2 ) . We also examined the levels of KIF5B expression by immunohistochemistry in excitatory neurons using neurogranin ( NRGN ) as a marker in the neocortex . We found that homozygous mice showed a significant reduction of cells that were positive for both KIF5B and NRGN in the frontal association cortex ( FrA ) when compared to wild-type mice , without significant change in the number of neurons in this region ( Figure 6C , Figure 6—figure supplement 1 ) . To determine the effect of KIF5B knockout on dendritic spines in adult neurons , the conditional knockout mice were crossbred with Thy1-YFP H line mice to enable sparse neuronal labeling for isolated dendrite imaging , followed by three-dimensional reconstruction for the analysis of spine number ( Figure 7A ) . Conditional knockout of KIF5B at postnatal stages resulted in a significant reduction of dendritic spines in CA1 hippocampal neurons of homozygous mice ( Figure 7B ) . However , the effect of KIF5B on spine density is region-specific , since the dendritic spine number was not significantly different between control and knockout mice in neurons of the FrA ( Figure 7C ) . To examine the excitatory synaptic transmission of CA1 hippocampal neurons , whole-cell patch recording was conducted on hippocampal slices from the wild-type and KIF5B conditional knockout mice . CA1 hippocampal neurons of the KIF5B conditional knockout mice showed a significant reduction in both the frequency and amplitude of mEPSC as compared with wild-type neurons ( Figure 7D ) . Therefore , the KIF5B conditional knockout mice showed a reduction of dendritic spine density that is associated with deficient excitatory synaptic transmission in hippocampal neurons . Although there was no significant difference in terms of dendritic spine density in FrA in homozygous conditional knockout , this region was chosen to examine dendritic spine plasticity based on its involvement in associative learning and accessibility for in vivo transcranial imaging ( Lai et al . , 2012; Nakayama et al . , 2015 ) . Using two-photon microscopy , we monitored the baseline dendritic spine plasticity of adolescent mice ( P31 ±1 ) over 7 days . Imaging sessions were performed on Day 0 , 2 , and 7 ( Figure 7E–F ) . We found that both heterozygous and homozygous mice showed a significant increase in dendritic spine elimination compared to wild-type mice over 2 days ( Figure 7G ) . However , when we examined the spine plasticity in the next time window from Day 2 to Day 7 over 5 days , both heterozygous and homozygous mice showed an increase in dendritic spine formation ( Figure 7H ) . Overall , both heterozygous and homozygous KIF5B conditional knockout mice showed an increase of dendritic spine turnover rate when compared to wild-type , but only that in homozygous was statistically significant ( Figure 7I ) . Although we did not observe significant difference in the survival rate of newly formed spines ( Figure 7—figure supplement 1 ) , we found that the increase in spine formation during the second time window was caused by the significant increase in re-formation of spines in close proximity to eliminated spines from first time window ( Figure 7J ) . These data suggest that KIF5B knockout in excitatory pyramidal neurons alters normal dendritic spine plasticity with an increase of synaptic instability in the neural circuitry . Based on the role of KIF5B on dendritic spine density and plasticity , we next investigated the impact of KIF5B conditional knockout on animal behavior . A series of behavioral tests were performed , including open field test , elevated plus maze , marble burying test , 3-chamber social interaction test , novel object recognition test , auditory-cued fear conditioning , and Barnes maze . We found that there was no significant difference in open field test , elevated plus maze , and marble burying test in heterozygous and homozygous mice when compared to wild-type , indicating that conditional knockout of KIF5B did not lead to hyperactivity , anxiety-like or repetitive behaviors ( Figure 8—figure supplement 1 ) . On the other hand , homozygous mice exhibited memory deficits in a variety of learning-related behaviors . In 3-chamber social interaction test , homozygous mice showed a significant reduction of social memory index ( Figure 8A ) , but no significant difference in total interaction time from wild-type ( Figure 8—figure supplement 2A–B ) . These data showed that KIF5B homozygous conditional knockout leads to deficits in social memory . In novel object recognition test , mice were presented with a novel object 14–16 hr after the mouse was exposed to the familiar objects for testing short-term object recognition memory . Homozygous mice showed a significantly reduced preference to the novel object ( Figure 8B ) , suggesting a deficit in short-term memory recall . Next , we used auditory-cued fear conditioning to test fear associative memory . The freezing response of KIF5B conditional knockout mice was similar to wild-type in the acquisition phase ( Figure 8—figure supplement 2C ) , but homozygous mice showed a significant decrease of freezing response to the conditioned stimulus ( CS , auditory cue ) during the recall test 48 hr after fear acquisition ( Figure 8C ) . Since there was no significant difference in the trend of fear acquisition , this data indicates that homozygous mice show deficit in fear memory recall . The absence of significant deficits in heterozygous mice in these memory tests suggests the dose-dependent role of KIF5B in memory formation and retrieval . We next investigated the effect of KIF5B conditional knockout in spatial memory without using heterozygous conditional knockout mice . In Barnes maze test , mice were trained to locate the escape chamber among the 20 holes in the maze during the acquisition phase based on contextual cues . Wild-type mice showed a learning progress during the acquisition phase as indicated by a decreasing trend of primary errors they made during training , but such learning progress was not observed in homozygous mice . Homozygous mice also tended to stay in the wrong target hole instead of exploring the environment which in turn showing fewer primary errors when compared to wild-type ( Figure 8—figure supplement 2D ) . Nonetheless , homozygous mice showed a significantly higher primary latency to locate the escape hole when compared to wild-type during the recall test five days after the training ( Figure 8D ) . Since the results of the behavioral tests strongly suggest deficits in hippocampal-dependent functions , we examined the expression of long-term potentiation ( LTP ) at the Schaffer collateral ( SC ) -CA1 synapses from acutely prepared hippocampal slices of the control and KIF5B homozygous conditional knockout mice . While LTP could be induced in both control and homozygous hippocampal slices , the LTP decayed faster at the homozygous SC–CA1 synapses , and the field EPSP amplitude during the last 10 min recording was significantly reduced at the homozygous SC–CA1 synapses when compared to wild-type ( Figure 8E ) . There was no significant difference in input/output relationship between wild-type and homozygous mice , indicating the baseline synaptic response was not affected ( Figure 8F ) . To test whether presynaptic function was altered , the pair-pulse ratios were measured with several different inter-pulse durations . We found no significant difference in pair-pulse ratio between wild-type and homozygous mice , indicating similar presynaptic responses ( Figure 8G ) . Taken together , these findings suggest that conditional knockout of Kif5b causes memory recall deficits in social memory , object recognition memory , fear associative memory , and spatial memory , showing the important role of KIF5B in memory formation and retrieval . These memory deficits are associated with impaired long-term synaptic plasticity in the hippocampus . Moreover , the synaptic and memory deficits in the KIF5B conditional knockout mice cannot be compensated by the presence of the other two homologous KIF5s .
Many studies have examined the functional significance of individual kinesin through exogenous expression of dominant-negative construct , which usually contains the tail domain of the kinesin of interest without the motor domain and hence does not move along microtubule . This approach is useful to demonstrate the effect of competitive binding between the dominant negative protein and endogenous motors for the cargoes . Using an alternative approach to delineate the function of individual KIF by RNAi or gene knockout , we demonstrate the important roles of KIF5B in regulating dendritic spine development and maintenance both in dissociated neurons in vitro and in the animal . Through the generation of conditional knockout mice in which the Kif5b gene is ablated only after birth in order to avoid lethality , we are able to demonstrate the physiological significance of KIF5B in regulating excitatory synaptic plasticity as well as learning and memory . Our findings provide compelling evidences that the function of KIF5B in neuron cannot be compensated by the other two neuron-specific KIF5s . Since only one KIF5 is expressed in invertebrates , it appears that the neuronal-specific KIF5A and KIF5C evolve specifically for higher brain function in vertebrates . We found that knock down of either KIF5B or KIF5C , but not KIF5A , reduced mushroom spines . On the other hand , co-expression of KIF5C but not KIF5A can rescue the loss of mushroom spines caused by KIF5B-shRNA . These finding indicates that KIF5B and KIF5C share functional similarity in dendritic spine morphogenesis and their roles cannot be replaced by the functionally distinct KIF5A . However , in the KIF5B conditional knockout mice in which KIF5C expression remains unaffected , reduction in both spine density and mEPSC is observed in hippocampal neurons . Therefore , the impaired synaptic function due to KIF5B deficiency cannot be compensated by KIF5C in the postnatal brain . One possible explanation is that KIF5B is the more prominently expressed kinesin compared to KIF5C in the adult hippocampus as shown by quantitative immunoblot ( Kanai et al . , 2000 ) . The presence of KIF5C in the conditional knockout mice may not be sufficient to compensate for the shortage of motor proteins after ~50% reduction of KIF5B expression . The carboxyl termini of the three KIF5s share little amino acid sequence similarity . The carboxyl terminus may not bind to cargo directly since GST-KIF5B constructs without this region ( a . a . 936–963 ) still pull down various cargoes ( Setou et al . , 2002; Cho et al . , 2007; Xu et al . , 2010; Barry et al . , 2014; Lin et al . , 2019 ) . Our present findings also suggest that the carboxyl-terminus is not directly involved in FMRP binding because removing it ( amino acid residues 939–1027 ) from KIF5A increases , rather than decreases , the pull-down of FMRP . Furthermore , although replacement of the KIF5A carboxyl-terminus by the KIF5B counterpart increases the binding to FMRP and G3BP1 , given that the input of KIF5B is much less than the chimeric KIF5A ( Figure 4A ) , it is likely that equal amount of KIF5B would pull down much more FMRP and G3BP1 . This again points to the involvement of KIF5 sequence besides the carboxyl-terminus in cargo-binding . Nonetheless , in the rescue experiments with chimeric KIF5A , swapping the carboxyl-terminus with KIF5B is sufficient to transform KIF5A into a kinesin motor that enhances spine maturation , therefore unraveling a new function of the carboxyl-terminus in determining functional specficity of KIF5s . In this regard , it is noteworthy that the longer carboxyl-terminus of KIF5A binds directly to GABAA receptor-associated protein for the development of inhibitory synapses ( Nakajima et al . , 2012 ) . Our findings together raise the interesting possibility that there is a division of labor among the two KIF5s in regulating excitatory and inhibitory synapses , and the evolution of their diverse carboxyl-termini confer them functional specificities . Many axonal cargoes , such as syntabulin , SNAP25 and amyloid precursor protein ( APP ) have been identified for KIF5s ( Hirokawa and Tanaka , 2015; Kamal et al . , 2000 ) . KIF5 motor domain also predominately recognizes axonal rather than dendritic microtubules , which highlight its functional significance in axon ( Kapitein et al . , 2010 ) . However , KIF5 is also implicated in the transport of cargoes such as GABAA receptor ( Twelvetrees et al . , 2010; Nakajima et al . , 2012 ) , AMPA receptor ( Heisler et al . , 2014; Setou et al . , 2002 ) and RNPs ( Kanai et al . , 2004 ) , which are believed to be mainly carried to the dendrites of mature neurons . We found that KIF5B is localized not only in the axons , but is also present in the dendrites and dendritic spines of dissociated hippocampal neurons , supporting the role of dendritic KIF5B in the development of excitatory postsynaptic sites . Although it was originally thought that microtubule is not present in dendritic spines , emerging study has revealed the invasion of microtubule and kinesin to the spine heads from dendritic shaft , which are crucial for dendritic spine plasticity ( Jaworski et al . , 2009; McVicker et al . , 2016 ) . Our findings suggest that KIF5B might represent one of the kinesin motors that deliver synaptic proteins to the dendritic spines . Dendritic spines exist as heterogeneous morphologies , which are usually classified into short stubby spines with no apparent spine neck , thin spines with elongated necks and small heads , mushroom-shaped spines with large bulbous heads , and filopodia which are long and thin and do not possess a PSD ( Ziv and Smith , 1996; Bourne and Harris , 2007; Lai and Ip , 2013; Berry and Nedivi , 2017 ) . Stubby and filopodia are regarded as immature dendritic protrusions because they are relatively scarce in the mature brain ( Harris et al . , 1992 ) . The distinct morphologies are critical to determine the properties and functions of dendritic spines . These include signal compartmentalization , calcium dynamics , capacity of local translation , and turnover ( McKinney , 2010 ) . Mushroom spines possess larger PSD which are correlated with greater synaptic strength and stability for information storage; while the dynamic thin spines are transient , but they may become persistent in response to a learning paradigm and contribute to the remodeling of neural circuits ( Bourne and Harris , 2007; Berry and Nedivi , 2017 ) . It is interesting that knockdown of KIF5B specifically decreases mushroom spines in cultured hippocampal neurons while increasing the abundance of the other three types of spines . Emerging studies have demonstrated that different spine types can be regulated differentially and independently ( Sanders et al . , 2012; Spiga et al . , 2014 ) . At the molecular level , we have also identified the postsynaptic scaffolding protein STRN4 , which is encoded by a dendritic mRNA and its expression depends on NMDA receptor activity , is involved specifically in the maintenance of mushroom spines ( Lin et al . , 2017 ) . It is tempting to speculate that a subset of proteins and/or mRNAs may depend on KIF5B for the delivery to mushroom spines that confer their selective maintenance . Since KIF5s can pull down RNPs from the brain ( Kanai et al . , 2004 ) , one possible mechanism by which KIF5B promotes the maintenance of mushroom spines is through the dendritic transport of mRNAs and RNA-binding proteins . We have found that knockdown of KIF5B reduced the dendritic localization of FMRP and two associated RNA transcripts as compared to knockdown of KIF5A , indicating their differential functions in dendritic transport of mRNAs . This may explain the altered spine morphology after knockdown of KIF5B , since the depletion of FMRP in mouse brain also resulted in an increase of dendritic filopodia ( Comery et al . , 1997 ) . The local translation of CaMKIIα and Grin2b mRNAs is critical to synaptic plasticity ( Kuklin et al . , 2017; Williams et al . , 2007 ) , which may contribute to the disrupted LTP in mouse hippocampus upon KIF5B depletion . FMRP and associated mRNA transport involves interaction with KLC ( Dictenberg et al . , 2008 ) . Since both KIF5A and KIF5B contain the conserved KLC binding domain , there could be additional mechanism that underlies the specific role of KIF5B in FMRP transport , which may involve the preferential interaction between FMRP and the KIF5B tail domain as revealed by our pull-down assay . It is also intriguing that KIF5B-shRNA only leads to fewer stationary granules on dendrites without affecting the motile oscillatory , unidirectional and bidirectional granules . Since other kinesins besides KIF5 can also bind to FMRP ( Charalambous et al . , 2013; Davidovic et al . , 2007 ) , we speculate that different pools of FMRP granules are carried by different KIFs , with KIF5B mainly responsible for the less motile granules while other KIFs transport the more motile pools of FMRP . It was recently reported that different KIFs transport cargoes with different velocities and MAP2 inhibits KIF5B activity in dendrites by interacting with the coiled-coil region and blocking microtubule binding ( Gumy et al . , 2017 ) . This study therefore also suggests that KIF5B-mediated transport in dendrites is ineffective as compared to other kinesins . Alternatively , since microtubule and dynein are required for mRNA anchoring in Drosophila embryos ( Delanoue and Davis , 2005 ) , it is possible that besides a conventional transport function , KIF5s may help anchoring the dendritically localized FMRP and mRNAs near synapses for local translation in response to extracellular stimuli such as BDNF or synaptic activity ( Schratt , 2004 ) . The function of kinesin is regulated by post-translational modification . Previous studies on the Kinesin-2 motor protein KIF17 revealed a novel mechanism of cargo release through calmodulin-dependent protein kinase ( CaMKII ) -mediated phosphorylation , which disrupts the interaction with the adaptor protein LIN10 and unloads the NMDA receptor subunit 2B ( GluN2B ) containing vesicles ( Guillaud et al . , 2008 ) . On the other hand , the association between synaptotagmin-containing vesicles and the motor adaptor UNC76 of KIF5 in Drosophila is strengthened by phosphorylation ( Toda et al . , 2008 ) . In the present study , we have characterized the methylation of two RGG motifs within the carboxyl-terminus of KIF5B involving Arg-941 and Arg-956 . Invertebrates such as C . elegans have shorter carboxyl-terminus of KIF5 that lacks the RGG motif , while Drosophila has one RGG motif containing Arg-956 , same as the mammalian KIF5C . The two RGG motifs in KIF5B are conserved across many vertebrates , indicating the importance of arginine methylation . Here we show that the KIF5B methylation is essential for the formation of mushroom spines and it modulates the interaction of KIF5B with FMRP , therefore unraveling a previously unidentified PTM in regulating kinesin function . There are extensive cross-talks between arginine methylation and other PTMs , such as phosphorylation , ubiquitination , and acetylation ( Basso and Pennuto , 2015; Yang et al . , 2018 ) . Future studies are needed to investigate how arginine methylation of KIF5B may interact with other forms of PTM in regulating cargo-binding of the motor protein . Does KIF5B play any specific role in learning and memory ? To answer this question , we generated the KIF5B conditional knockout mouse line in CaMKIIα-expressing neurons . Since the expression of CaMKIIα is developmentally regulated and is restricted to the forebrain with high levels in the pyramidal neurons of the neocortex and hippocampus ( Dragatsis and Zeitlin , 2000; Tsien et al . , 1996 ) , we can specifically knockout KIF5B postnatally without affecting early neurodevelopment . Here we demonstrated that specific knockout of Kif5b in CaMKIIα-expressing neurons leads to deficits in memory recall in social memory , novel object recognition , auditory-cued fear conditioning , and spatial memory tests , with no significant deficit during initial memory acquisition phase . Furthermore , the KIF5B conditional knockout mice show deficits in the maintenance of LTP in CA1 hippocampal neurons and the loss of dendritic spines . Although there is no significant decrease of dendritic spine density in the frontal association cortex of conditional knockout mice , the rates of dendritic spine formation and elimination are significantly higher at different time points in two-photon in vivo imaging , suggesting the increase of dendritic spine instability in this region . Increase in dendritic spine instability has been commonly found in various disease models , such as Fragile X syndrome ( Nagaoka et al . , 2016; Pan et al . , 2010 ) , schizophrenia ( Fénelon et al . , 2013 ) , spinocerebellar ataxia type 1 ( Hatanaka et al . , 2015 ) and Huntington disease ( Murmu et al . , 2013 ) . It has been found that a small fraction of the population of transient spines grows after experience or behavioral training over days can be stabilized over the animal's lifetime , contributing to long-lasting circuit remodeling associated with new experience ( Yang et al . , 2009 ) . The enhanced dendritic spine instability in KIF5B conditional knockout mice could contribute to brain dysfunction and deficits in learning and memory . Since the frontal cortex maturation happens at later developmental stage ( Caballero et al . , 2016; Gogtay et al . , 2004; Zuo et al . , 2005 ) , the lack of dendritic spine density difference in the frontal association cortex between wild-type and KIF5B conditional knockout mice could be due to the delay of frontal cortex maturation and pruning in the conditional knockout mutant . Nonetheless , the impairments in memory recall , LTP maintenance , and dendritic spine deficits in KIF5B conditional knockout demonstrate the crucial role of KIF5B in learning and memory that cannot be compensated by KIF5A and KIF5C in vivo . The process of memory storage is not a random event . The synaptic tagging and capture hypothesis proposes that the synapses activated during LTP induction become ‘tagged’ ( Rogerson et al . , 2014 ) . These tagged synapses become a target for subsequent plasticity-related product ( PRP ) trafficking . The capture of these PRPs by specific synapses is essential for their structural modification , as well as the maintenance of LTP and long-term memory formation . The deficits that we observed in KIF5B conditional knockout mice could be stemmed from the impairment of PRP trafficking specifically delivered by KIF5B in dendrites in response to activity-dependent plasticity . Taken together , our findings have revealed the significance of KIF5B in regulating excitatory synapse development and function of neuron both in vitro and in vivo , and support the notion that the three homologous KIF5s have non-redundant functions in the brain . It is plausible that homologous members of the other kinesin families also exhibit functional specificity in the brain , an interesting research area which warrants further study in the future .
Antibody against KIF5B was previously described ( Cui et al . , 2011 ) , while others were purchased commercially , including antibodies against KIF5A , KIF5C , FMRP ( Abcam ) , KIF17 , FLAG ( Sigma ) , G3BP1 , G3BP2 ( Bethyl ) , GluA2 , RNMT , NRGN and NeuN ( Millipore ) , PSD-95 ( NeuroMab ) , methylated mono-arginine R*GG ( Cell Signaling ) , GFP ( Invitrogen ) , and RFP ( Rockland ) . Alexa-conjugated secondary antibodies ( Invitrogen ) were used for immunofluorescence and horseradish peroxidase-conjugated goat anti-rabbit IgG or anti-mouse IgG ( Cell Signaling ) were used for western blot analysis . For the specific knockdown of KIF5A , KIF5B , and KIF5C , a 19-nucleotide ( KIF5A: 5’-TGGAAACGCCACAGATATC-3’ , KIF5B: 5’-GGACAGATGAAGTATAAAT-3’ , KIF5C: 5’-GACCCTGGCAGATGTGAAT-3’ ) sequence derived from the rat KIF5A mRNA , mouse KIF5B mRNA at the 3’-UTR and rat KIF5C mRNA were used to create the shRNA constructs after subcloning into the pSUPER vector ( Oligoengine ) . The sequence of control shRNA is 5’-GGCTACCTCCATTTAGTGT-3’ . Full-length mouse KIF5A and KIF5C constructs were obtained from Quan Hao ( The University of Hong Kong ) , and the coding sequence was amplified and subcloned into pcDNA3 backbone . Full-length mouse KIF5B was amplified by PCR using the plasmid pcDNA3-FLAG-KIF5B as template , which contains the insert of full-length mouse KIF5B coding region . Methylation-deficient R941H , R956H and R941/956H constructs were created by site-directed mutagenesis and the PCR products were digested by DpnI ( NEB ) at 37°C water bath for 3 hr before transformation into E . coli competent cells . The nucleotide sequence was verified by Sanger sequencing . For GFP-FMRP construct , the human FMRP coding sequence was amplified from the plasmid pFRT-TODestFLAGHAhFMRPiso1 that was from Thomas Tuschl ( Addgene #48690 ) and cloned into the pEGFP-C1 backbone using SacI and EcoRI . KIF5A-GFP and KIF5B-GFP were constructed by inserting PCR-amplified mouse KIF5A and KIF5B coding sequences into the pEGFP-N1 plasmid using KpnI and BamHI . All PCR in this study was performed using high-fidelity Pfu DNA polymerase ( Agilent Technologies , Inc ) . Mice were group housed under a 12 hr light/dark cycle , with food and water available ad libitum . C57BL/6 mice expressing CaMKIIα-Cre and yellow fluorescent protein ( YFP ) in layer V pyramidal neurons ( Thy1-YFP-H ) and CaMKIIα promoter-driven Cre transgenic mice were purchased from the Jackson Laboratory . Kif5bfl/fl mice were described previously ( Cui et al . , 2011 ) . CaMKIIα promoter-driven Cre transgenic mice were used to conditionally delete exons flanked by loxP . Mice were then further crossed with Thy1-YFP-H line to allow imaging of layer V pyramidal neurons . Sample size was decided based on experiments in previous studies ( Lai et al . , 2012; Yang et al . , 2014 ) . For animal behavioral tests and in vivo imaging experiments , results from at least two independent experiments were pooled together for analysis . Mice were group housed in The Laboratory Animal Unit , The University of Hong Kong , accredited by Association for Assessment and Accreditation of Laboratory Animal Care International . Four to five weeks old mice were used in this study unless stated otherwise . All experiments were approved and performed in accordance with University of Hong Kong Committee on the Use of Live Animals in Teaching and Research guidelines . Whole-cell recordings were obtained by the MultiClamp 700B amplifier ( Molecular Devices ) . For cultured hippocampal neurons , which were recorded at DIV 16–17 , the pipettes with a resistance of 3–5 MΩ were filled with the internal solution consisting of 115 mM CsCl , 10 mM HEPES , 2 mM MgCl2 , 4 mM NaATP , 0 . 4 mM NaGTP , 0 . 5 mM EGTA , and pH was adjusted to 7 . 2–7 . 4 by CsOH . The neurons were perfused with the external solution of the following composition: 110 mM NaCl , 5 mM KCl , 2 mM CaCl2 , 0 . 8 mM MgCl2 , 10 mM HEPES , 10 mM Glucose , and pH was adjusted to 7 . 2–7 . 4 by NaOH . For miniature excitatory postsynaptic currents ( mEPSCs ) recording , tetrodotoxin ( 1 µM ) and bicuculline ( 20 µM ) were added into the external solution to block action potentials and the inhibitory current from GABA receptor , respectively . The signals were filtered at 2 kHz and sampled at 20 kHz using the Digidata 1440A ( Molecular Devices ) . The holding potential is at −70 mV , and the recording lasts for 5 to 10 min . The data were analyzed by the commercial software MiniAnalysis ( Synaptosoft ) . For recording mEPSCs in dorsal hippocampal CA1 brain slices , postnatal day ( P ) 45 ± 3 wild-type and KIF5B conditional knockout mice were perfused by ice-cold dissection buffer ( 92 mM NMDG , 2 . 5 mM KCl , 1 . 25 mM NaH2PO4 , 30 mM NaHCO3 , 25 mM glucose , 20 mM HEPES , 5 mM Na-ascorbate , 3 mM Na-pyruvate , 2 mM thiourea , 10 mM MgSO4 and 0 . 5 mM CaCl2pH = 7 . 1–7 . 3 ) after euthanized . The brains were taken out immediately and submerged in ice-cold dissection buffer . Coronal brain slices containing CA1 were sectioned in 250 µm by vibratome . Slices were recovered in warm artificial cerebral spinal fluid ( ACSF ) at 32°C for 15 min , followed by room temperature incubation . The recordings were performed in ACSF at room temperature . The ACSF consisted of the following ( in mM ) : 119 NaCl , 2 . 5 KCl , 1 MgCl2 , 2 CaCl2 , 26 NaHCO3 , 1 . 23 NaH2PO4 and 10 glucose . All solutions were oxygenated by 95% O2/5% CO2 . Internal solution consisted of the following ( in mM ) 131 Cs-methanesulfonate , 20 CsCl , 8 NaCl , 10 HEPES , 2 EGTA , 2 NaATP and 0 . 3 NaGTP , pH7 . 3 , osmolarity 290 mOsm . The glass micropipette was filled with internal solution ( resistance 4–6 MΩ ) and connected to the electrode for recording . The mEPSCs were recorded with the presence of 1 µM tetrodotoxin , 10 µM bicuculline and 1 µM strychnine . For recording LTP , hippocampal slices from the wild-type and the KIF5B conditional knockout mice ( 3 months old ) were prepared . A planar multi-electrode recording setup ( MED64 system , Alpha Med Sciences Co . , Ltd , Japan ) was employed to record the field excitatory postsynaptic potential ( fEPSP ) , and to study LTP . Briefly , hippocampal slices were placed on special probes that were fabricated with 8 × 8 electrode arrays and pre-coated with polyethylenimine ( PEI , Sigma ) . The P210A probes ( Alpha Med Sciences ) with an inter-electrode distance of 100 μm were routinely used . Correct placement of the electrodes at the CA3–CA1 region was done manually , monitored by a microscope ( MIC-D , Olympus Ltd . , Japan ) . To increase the efficiency of the experiments and to minimize the variation in the results arising from differences in incubation times , a maximum of 4 slices were studied simultaneously . Each slice was superfused by oxygenated ACSF . fEPSPs were recorded from the dendritic layer of CA1 neurons by choosing an electrode in the Schaffer collateral pathway as the stimulating electrode . Based on the stimulus–response curve , we chose a stimulation intensity that evoked the fEPSP with a magnitude of 30–40% of the maximum response . After allowing a stable baseline of 30 min , an induction protocol consisting of 1 train of 100 Hz stimulus that lasted for 1 s was applied , and the field potential response for 1 hr after the tetanus was recorded . The magnitude of the LTP was quantified as % change in the average amplitude of the fEPSP taken from 50 to 60 min interval after induction . To assess basal synaptic transmission , the input-output relationship was generated by delivering 10–100-μA electrical stimuli , and the amplitude of the peak fEPSPs was measured . To characterize the paired-pulse ratio , twin stimuli that were separated by a variable time interval ( 50 , 100 , 150 , 200 or 400 ms ) were delivered to the CA3-CA1 pathway ten times each , and the average ratio of the amplitude of the second evoked fEPSPs over the first one was determined . All the electrophysiology experiments were performed and analyzed blinded . Primary hippocampal neurons and cortical neurons were prepared from embryonic day 18–19 embryos of Sprague Dawley rats according to our previous study ( Lin et al . , 2017 ) . Hippocampal neurons were cultured on 18 mm coverslips or 35 mm MatTek dishes ( with 14 mm central glass , MatTek corp ) dishes coated with poly-D-lysine ( 1 mg/ml , Sigma P0899 ) at high density ( 1 . 4 × 105 cells per coverslip for dendritic spine analysis; 2 × 105 per cover glass on MatTek dish for live cell imaging of GFP-FMRP ) or low density ( 0 . 4 × 105 cells per coverslip for FISH and immunofluorescence staining ) in Neurobasal medium supplemented with 2% B27% and 0 . 5% L-glutamate . Hippocampal neurons were transfected with different plasmids using calcium phosphate precipitation as previously described ( Lai et al . , 2008 ) . Cortical neurons were transfected by electroporation using the Neon transfection system ( ThermoFisher Scientific ) , in which a total of 1 × 106 cells in suspension were electroporated in each reaction with the parameter of 1500V pulse voltage and 20 ms pulse width . After electroporation , cells were plated on 35 mm dishes and cultured for 5 days before Western blot analysis . Images were taken using Perkin Elmer UltraView Vox Spinning Disk Confocal Microscope 60x oil-immersion objective ( NA 1 . 40 ) at a resolution of 512 × 512 pixels , one frame per second for 100 s . Images were exported using Volocity software and processed using FIJI software . Kymographs of selected dendrites were generated in FIJI software using the ‘KymoResliceWide’ plugin . The kymographs were randomized and reviewed blindly , and images with low signal-to-noise ratio were excluded due to the difficulty in quantification . The movement of individual granules in selected kymographs was then traced manually by drawing polygonal lines as overlays on the image and the traces were reviewed by an experimenter blind to the conditions . Minimum and maximum values of the kymographs were constantly adjusted during manual tracing due to uneven intensity on different segments of the dendrite but were limited to a range that was considered appropriate for that batch of images . The traces were then exported with information of the x and y coordinates of each point on the polygonal lines . To classify the type of movement exhibited by each granule , the net displacement ( ND ) and lateral maximal displacement ( LMD ) were measured . ND is defined as the difference in x coordinates of the first point and the last point of the trace . Lateral maximal displacement ( LMD ) is defined as the absolute value of maximal difference in x coordinates of all the points on the trace ( the difference between the most proximal point and the most distal point ) . Granules with ND ≧2 μm are defined as unidirectional , within which the granules with ND >0 are defined as anterograde ( from soma towards distal dendrite ) and granules with ND <0 are defined as retrograde ( from distal dendrite to soma ) . For granules with ND <2 μm , granules with LMD <1 μm are defined as stationary; granules with 1 μm≦LMD <2 μm are defined as oscillatory; while granules with LMD ≧ 2 μm are defined as bidirectional . The motility of each granule in unidirectional or bidirectional movements was further quantified in terms of travel distance , maximal run length and maximal velocity . Travel distance is defined as the total length of the granule trajectory within 100 s . The maximal run length is defined as the largest x-axis distance of a period of movement with constant velocity . The maximal velocity is defined as the maximum of all velocity values of a granule , which is calculated by dividing the x-axis distance of each segment of the trajectory by the y-axis distance ( the time covered by this segment ) . SPM fraction was prepared using sucrose gradient method as described ( Bermejo et al . , 2014 ) . Briefly , mouse brains ( ~P20 ) were homogenized in 0 . 32M HEPES-buffered sucrose solution . The homogenate was either centrifuged at 13000 rpm for 10 min ( min ) yielding the supernatant ( Homo ) for western blot analysis , or subjected to fractionation . The homogenate was centrifuged at 900 x g for 10 min ( min ) to remove nuclear fraction and the crude synaptosomal fraction ( P2 ) was enriched from the supernatant using two times of centrifugation at 10000 x g for 15 min . The P2 pellet was later subjected to hypo-osmotic shock and centrifugation at 25 , 000 x g for 20 min to yield the Synaptosomal Membrane Fraction ( P3 ) . The obtained pellet was then resuspended and loaded to a 0 . 8M/1 . 0M/1 . 2M HEPES-buffered sucrose gradient and centrifuged at 150 , 000 x g for 2 hr , separating fractions in different layers . The SPM fraction was collected at the 1 . 0M/1 . 2M interface , further centrifuged at 160 , 000 x g for 30 min , and resuspended in 50 mM HEPES/2 mM EDTA solution . For Western blot analysis , homogenate and SPM samples were diluted with RIPA buffer or 50 mM HEPES/2 mM EDTA and denatured in sample buffer ( 5x sample buffer: 300 mM Tris-HCl buffer pH 6 . 8 10% ( w/v ) DSD , 25% ( v/v ) beta-mercaptoethanol , 50% ( v/v ) SDS , 25% ( v/v ) glycerol , 0 . 05% ( w/v ) bromophenol blue ) . The preparation of SNS was performed as previously described with modification ( Scheetz et al . , 2000 ) . In; brief , P15 mice were decapitated , and cerebellum together with the superficial , retinorecipient layers of the superior colliculus were removed . The rest of the brain tissues were homogenized in ice-cold homogenized buffer ( 5M NaCl , 1M KCl , 1M MgSO4 , 0 . 5M CaCl2 , 1M KH2PO4 , 212 . 7 mM glucose , pH 7 . 4 ) supplemented with protease inhibitor cocktail ( Roche ) . All subsequent steps were carried out at 4°C . Samples were passed through a series of nylon filters of descending pore size . The final pass was through Millipore filter with a 10 μm pore size . Samples were then centrifuged for 15 min at 1000 x g at 4°C . The supernatant was discarded , and the pellet was resuspended in 100 μl homogenization buffer for immunoprecipitation . To test whether KIF5B was methylated in SNS , equal amount of SNS fraction lysate ( 800 μg ) was incubated with KIF5B or mono-methyl-arginine ( Cell Signaling ) antibody at 4°C with rocking overnight . Immunoprecipitate was obtained with RIPA buffer after incubation with Protein A-Sepharose beads ( GE Healthcare ) for 1 hr in cold room with rocking . Beads were washed four times with RIPA buffer containing various protease and phosphatase inhibitors ( 10 µg/ml soybean trypsin inhibitor , 10 µg/ml leupeptin , 10 µg/ml aprotinin , 2 µg/ml antipain , 30 nM okadaic acid , 5 mM benzamidine , 1 mM sodium orthovanadate , 1 mM PMSF , 1 mM sodium fluoride , 100 mM beta-glycerophosphate ) . Proteins were eluted by boiling in sample buffer for 6 min . The eluate was collected by centrifugation at 13000 rpm for 1 min at 4°C and then subjected to SDS-PAGE and Western blot analysis . The protein extract was boiled in sample buffer for 5 min , separated by SDS-PAGE , and transferred onto PVDF membranes , followed by blocking with 5% skim milk in TBS with 0 . 1% Tween 20 ( TBST ) for 1 hr at room temperature ( RT ) . The membrane was incubated at 4°C with primary antibody diluted in TBST containing 5% BSA overnight . After washing three times with TBST , membranes were incubated for 1 hr at RT with HRP-conjugated secondary antibody diluted in 5% skim milk in TBST . The HRP signal was detected by ECL ( Thermo Scientific ) and quantified by densitometry using Photoshop software . To map the methylation sites of KIF5B , HEK-293T cells cultured in 100 mm dishes with 80% confluence were transfected with various KIF5B plasmids using Lipofectamine ( ThermoFisher Scientific ) . Twenty-four hours after transfection , the cells were washed by ice-cold D-PBS and lysed by RIPA containing various protease and phosphatase inhibitors . Lysate was incubated at 4°C for 45 min and the cell debris was cleared by centrifugation at 13000 rpm for 10 min at 4°C . Equal amount of lysate ( 1 mg ) was incubated with FLAG beads ( Sigma ) in cold room for 1 hr with rocking . The FLAG-beads were centrifuged at 3000 x g for 1 min at 4°C and washed for three times with RIPA buffer containing various protease and phosphatase inhibitors , and proteins were eluted by boiling in sample buffer for 6 min . The eluate was collected by centrifugation at 13000 rpm for 1 min at 4°C and then subjected to SDS-PAGE and Western blot analysis . For pull-down experiments , different FLAG-tagged segments from KIF5s were transfected into HKE293T cells using Lipofectamine ( ThermoFisher Scientific ) . Twenty-four hours after transfection , cell lysate was collected by RIPA buffer with various protease and phosphatase inhibitors as described above . Equal amount of lysate ( 1 mg ) was incubated with FLAG beads ( Sigma ) for immunoprecipitation . SNS pellet was collected and lysed with Tris buffer ( 20 mM Tris , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 5 mM NaF , 0 . 5% NP40 ) . Equal amount of SNS fraction ( 1 mg ) was incubated with the immunoprecipitation from FLAG beads at 4°C with rocking overnight . FLAG beads were centrifuged at 3000 x g for 1 min at 4°C and washed for three times with Tris buffer containing various protease and phosphatase inhibitors . Protein were eluted by boiling in sample buffer for 6 min and then subjected to SDS-PAGE and Western blot analysis . To validate KIFs expression in the KIF5B conditional knockout mice , brain lysate was obtained from mice ( P44 ) . Protein levels were determined by blotting with anti-KIF5A , anti-KIF5B , anti-KIF5C , anti-KIF17 ( all 1: 1000 ) and anti-β-actin ( 1:3000 ) antibodies . The recombinant GST-fused proteins were expressed by E . coli BL21 ( DE3 ) grown in LB culture medium . Isopropyl β-D-1-thiogalactopyranoside ( 0 . 1 mM ) was used to induce expression of GST-fused KIF5A ( a . a . 677–1027 ) at 28°C for 5 hr , while 0 . 5 mM isopropyl β-D-1-thiogalactopyranoside was used to express all other GST-fused proteins at 37°C for 3 hr . Mice ( ~6 week old ) were sacrificed and forebrains were homogenized and lysed with Tris buffer . The brain lysate was pre-cleared by glutathione sepharose four fast flow beads ( GE health ) and GST proteins with rocking at 4°C for 1 hr . Equal amount of pre-cleared brain lysate and beads were incubated with equimolar GST-fused proteins at 4°C for 2 hr with end-over-end mixing . Then , the beads were washed with Tris buffer for three times . Proteins were eluted by boiling in sample buffer for 6 min and then subjected to Western blot or silver staining using SilverQuest Silver Staining Kit ( Life technologies ) . FISH was performed using ViewRNA ISH Cell Assay Kit ( ThermoFisher ) following manufactural instructions . In brief , cells were fixed using 4% formaldehyde for 30 min and rinsed in 1 x PBS . Cells were then treated with detergent and incubated with custom designed probe sets against Grin2b transcript ( NM_012574 . 1 , type 1 ) and CaMKIIα transcript ( NM_012920 . 1 , type 6 ) for 3–4 hr , preamplifier mix for 30 min , amplifier mix for 30 min , and label probe sets for 30 min , all in 40°C . Coverslips were washed with wash buffer for three times in between . Anti-GFP antibody was subsequently used for immunostaining . To stain GFP-transfected neurons for dendritic spine analysis , neurons were incubated with GFP antibody ( 1:2000 ) in GDB buffer at 4°C overnight . After washing three times with phosphate washing buffer ( 20 mM phosphate buffer and 0 . 5M NaCl ) , neurons were incubated with Alexa488-conjugated anti-mouse IgG2a secondary antibody ( 1:2000 diluted in GDB buffer ) at RT for 1 hr , followed by washing three times by the phosphate washing buffer before mounting . For other immunocytochemistry experiments , cells were fixed by 4% PFA/4% sucrose in D-PBS for 15 min at RT . After washing with D-PBS , cells were incubated with blocking buffer ( 0 . 4% Triton X-100 ( vol/vol ) and 1% BSA ) for 45 min at RT , and incubated with primary antibodies in blocking buffer at 4°C overnight . Cells were washed three times with washing buffer ( 0 . 02% Triton X-100% and 1% BSA in PBS ) , incubated with anti-mouse IgG2a Alexa 488 conjugate and anti-rabbit IgG Alexa 546 conjugate at RT for 1 hr , followed by washing twice in washing buffer and once by D-PBS before mounting with Hydromount ( National Diagnosis ) . Carl Zeiss LSM 700 confocal laser-scanning microscopes installed with Zen digital imaging software were used to acquire z-stack fluorescent images using a 63x oil-immersion objective ( NA 1 . 40 ) with the following parameters: 1 AU or smaller pinhole , 0 . 5x optical zoom , scan speed 6–8 , interval 0 . 35 μm with 16-bit dynamic range . The images from the same experiment were captured using identical acquisition settings , except for GFP or tdTomato ( RFP ) staining which served to visualize dendritic arbors and spines . Images from 2 to 3 coverslips were acquired for each experimental condition , and results from three independent experiments were pooled together for analysis . Sample size was decided based on experiments in previous study ( Lin et al . , 2017 ) . For dendritic spine quantification in dissociated hippocampal neurons , images of the whole neuron were captured by confocal microscope and assigned a random number , and dendrites with length more than 50 μm were selected by another blinded experimenter for quantification . Dendritic spines were classified based on our previous study ( Lin et al . , 2017 ) . The length ( L ) , head width ( H ) and neck width ( N ) of each individual spine were measured manually using the MetaMorph software . Mushroom spines were defined as those having H/N ≧1 . 5; stubby spines were defined as those having H/N ≦ 1 and L/N ≦1; thin spines had the ratio of 1≦ H/N < 1 . 5 and 1 . 5 ≦ L/N ≦3 . Filopodia were defined as those with the ratio of H/N < 1 . 2 and L/N > 3 . For each neuron , one to three isolated dendrites were selected and quantified , and the average spine density would be calculated . The ‘n’ number is defined as the number of neurons analyzed . For quantification of KIF5A-GFP and KIF5B-GFP puncta , images after maximal projection of multiple z-layers were intensity-adjusted to the same minimum and maximum values using FIJI software ‘Brightness/Contrast’ function before manual counting of puncta . To quantify the localization of endogenous KIF5A and KIF5B by immunostaining , images after maximal projection of z-layers were intensity-adjusted to remove signals below the threshold , which is determined by the negative control without primary antibody as reference . The spine density and percentage of puncta-positive spines were quantified by manual counting . For quantification of KIF5A and KIF5B knockdown efficiency by immunofluorescence , areas of cell soma were outlined based on GFP signals on images after maximal projection of multiple layers . The signal intensity of the target protein within selected area was measured using FIJI software . For quantification of FISH images , selected dendrites from maximal projected images were straightened using ‘Straighten’ ( Kocsis et al . , 1991 ) plug-in in FIJI . For each channel of interest , a threshold was determined based on a negative control image and the puncta information was extracted using ‘Analyze particle’ function in FIJI within the region of the dendrite ( outlined based on GFP signals ) . For the analysis of granule distribution along dendrites , the number of granules within each bin ( 5 μm ) was determined for every dendrite , and the number in the first bin was normalized as 1 . For the analysis of dendritic spines in hippocampus in vivo , mouse brains were fixed at P44 and coronally sectioned at 50 μm on a vibratome ( Leica ) . Confocal images of secondary dendrites from apical branches of CA1 hippocampal neurons and prefrontal cortex neurons were captured as described above . 3D reconstruction of individual dendrites was performed . The dendritic spine number was analyzed by Neuron Studio software . For neuronal nuclei ( NeuN ) , neurogranin ( NRGN ) and KIF5B staining , mouse brains were sacrificed at P44 and post-fixed with 4% paraformaldehyde . The samples were then sectioned to 50 μm per slice using vibratome . Brain sections were blocked with 1 . 5% normal goat serum ( NGS ) in PBST ( 0 . 3% Triton X-100 ) and incubated with a 1:1000 diluted primary antibody against KIF5B at 4°C overnight . Alexa 488-conjugated goat anti-rabbit IgG secondary antibody was used to probe the anti-KIF5B signals . Since both anti-KIF5B and anti-NRGN were from rabbit host , the sections were blocked again with 5% normal rabbit serum ( NRS ) in PBST ( Wessel and McClay , 1986 ) . Next , sections were incubated with 1:1000 anti-NRGN primary antibody overnight at 4°C . Another secondary antibody , goat anti-rabbit Alexa 546 , was used to probe anti-NRGN signals . For NeuN staining , brain sections were incubated with anti-NeuN antibody ( 1:1000 ) after blocking . Goat anti-mouse IgG Alexa 546 conjugate was used to probe the anti-NeuN signals . Imaging was carried out under LSM700 confocal microscope . Quantification of fluorescence images was performed using ImageJ software . All behavioral tests were performed in the chronological order of open field test ( OFT ) , elevated plus maze ( EPM ) , marble burying test ( MBT ) , 3-chamber social interaction ( SI ) and fear conditioning ( FC ) . Barnes maze ( BM ) , novel object recognition ( NOR ) and rotarod training were done in separate sets of animals . Open field test . Mice were placed in the center of a square open field chamber ( 40 × 40 × 40 cm ) surrounded by walls . Tracing was performed using ANY-maze software . The time of the mouse spent in the center area was measured over the course of 15 min ( Shin Yim et al . , 2017 ) . Elevated plus maze . Mice were placed in the center of a plus-shaped chamber that stands 38 cm above ground . Mice were then allowed to explore freely for 5 min . The duration of the mouse spent in either arm was recorded and tracked using ANY-maze software ( Walf and Frye , 2007 ) . Marble burying test . Mice were placed into testing arenas ( arena size: 42 . 5 cm ×27 . 6 cm × 15 . 3 cm , bedding depth: 5 cm ) each containing 20 glass marbles ( laid out in four rows of five marbles equidistant from one another ) . At the end of the 30 min exploration period , mice were carefully removed from the testing cages and the number of marbles buried was recorded . The marble burying score was arbitrarily defined as the following: four for completely buried marbles , three for marbles covered >50% with bedding , two for marbles covered 50% with bedding , one for marbles covered <50% with bedding , or 0 for anything less . The final marble burying score for each mouse was the sum of the scores of the 20 marbles ( Shin Yim et al . , 2017 ) . Novel object recognition . Mice were placed into a training chamber ( 25 cm x 25 cm x 40 cm ) containing two identical objects . Mice were allowed to freely explore in the chamber for 10 min . In the recall session , mice were put back to the same chamber while one of the two identical objects were replaced with a novel object with different color and slightly different shape 14–16 hr after the training session . The movement of the mice was tracked with ANY-maze software for its interaction with both the familiar and novel objects . Discrimination index = interaction time with novel object/total interaction time with both objects ( Leger et al . , 2013 ) . Three-chamber social interaction . Two empty object-containment cages ( shape of a cup with evenly spaced metallic bars ) were each placed into the left and right chamber of a 3-chamber arena ( 20 cm ×42 cm × 26 cm ) . In the adaptation period , a mouse was shut within the center chamber for 5 min . In stage 1 , a stranger mouse of same sex , similar age and size as the test mouse was put into the left cage . The test mouse in the center was released then to freely explore all of the three chambers for a 10 min period . After stage 1 , the test mouse was shut within the center again when the experimenter put another stranger mouse to the right cage . At stage 2 , the test mouse was allowed to explore all the three chambers again for 10 min . Approach behaviour within 2 cm with targets was defined as interaction time . Sessions were video-recorded . Approach behaviour and total distance travelled were analyzed using ANY-maze tracking system ( Shin Yim et al . , 2017 ) . Sociability index = ( percentage time of interaction with stranger ) - ( percentage time of interaction with empty cage ) /percentage of interaction time with both objects . Social memory index = ( percentage time of interaction with novel stranger ) - ( percentage time of interaction with familiar ) /percentage of interaction time with both strangers . Auditory-cued fear conditioning . FreezeFrame system ( Coulbourn Instruments ) was used to train and test mice . For training , the chamber was equipped with stainless-steel shocking grids , which were connecting to a precision feedback current-regulated shocker . Each chamber was contained in a sound-attenuating enclosure . Animal behaviour was recorded using low-light video cameras . Actimetrics FreezeFrame software ( version 2 . 2; Coulbourn Instruments ) was used to control the stimulus presentation by a preset program . All equipment was thoroughly cleaned with water followed by ethanol between sessions to avoid residue of scents from mouse feces and urine . Mice were habituated for 1 min on a shocking grid ( cage set-up A: shocking floor grids , ethanol scent ) . Fear conditioning was conducted with three pairings of a 30 s , 4000 Hz , 80 dB auditory cue ( CS ) co-terminating with a 2 s , 0 . 5-mA scrambled footshock ( US ) . The inter-trial interval was 20 s . One minute after conditioning , mice were returned to their home cages . For the recall test , mice were placed in a different context ( cage set-up B: test floor grids , 1% lemon scent detergent ) for an initial 2 min ( pre-tone ) period and this was followed by tone presentation for 2 min ( CS ) ( Lai et al . , 2012 ) . Rotarod . An EZRod system ( Omnitech Electronics , Inc ) was used as a motor training model . Mice were placed on the motorized rod ( 30 mm in diameter ) in the chamber . The rotation speed gradually increased from 0 to 100 r . p . m . over the course of 3 min . Rotarod training was performed for 20 trials , each trial lasts until the subjects dropped and the system would automatically complete that trial ( Deacon , 2013; Yang et al . , 2014 ) . Barnes maze . Mice were placed on a white circular table ( 92 cm in diameter , 1 m tall ) , which had a total of 20 holes ( 5 cm in diameter ) separated evenly along the edge of the table . During the test , strong light with an intensity of 1500 lux and repetitive noise from metronome of 80 dB were given to serve as aversive stimuli to induce escape behaviour . On acquisition day , mice were first guided manually to the escape hole for adaptation purpose . Then , mice received 5 trials of training , with each separated from one another by 15 min . Each trial would last for 3 min . If mice were not able to find the target escape hole by the end of the each trial , mice will be guided to the target escape hole as a part of training . A 3 min recall session was carried out 5 days after acquisition day . Mice were subjected to the same maze except the escape hole was also blocked . The number of errors and latency to reach the original escape hole were measured manually to confirm the result generated by ANY-maze . Heat maps were obtained by ANY-maze . ( Sunyer et al . , 2007 ) . Spine formation and elimination were examined in longitudinal studies by imaging the mouse cortex through a thinned-skull window as described previously ( Lai et al . , 2012; Yang et al . , 2009 ) . Briefly , one-month-old mice expressing YFP were anesthetized with ketamine/xylazine ( i . p . , 20 mg/ml , 3 mg/ml respectively in saline , 6 μl/g body weight ) . Thinned skull windows were made with high-speed microdrills in head-fixed mice . Skull thickness was reduced to about 20 μm . A two-photon microscope tuned to 920 nm ( 25x water immersion lens , N . A . 1 . 05 ) was used to acquire images . For re-imaging of the same region , thinned regions were identified based on the maps of the brain vasculature . Microsurgical blades were used to re-thin the region of interest until a clear image could be obtained . The area of the imaging region is 216 μm × 216 μm . The center of imaging region is located at the frontal association cortex ( +2 . 8 mm bregma , +1 . 0 mm midline ) . All data analysis was performed blind to treatment conditions . For imaging of dendritic spines , dendritic branches were randomly sampled within a 216 µm × 216 µm area imaged at 0–100 µm distance below the pia surface . The same dendritic segments were identified from three-dimensional image stacks taken at different time points with high image quality ( ratio of signal to background noise >4:1 ) . The number and location of dendritic protrusions ( protrusion lengths were more than one-third of the dendritic shaft diameter ) were identified . Filopodia were identified as long , thin structures ( generally larger than twice the average spine length , ratio of head diameter to neck diameter <1 . 2:1 and ratio of length to neck diameter >3:1 ) . The remaining protrusions were classified as spines ( Ng et al . , 2018; Lai et al . , 2012 ) . The percentage of spine formation and elimination represented the number of spines formed or eliminated between the first and second view divided by the total number of spines counted at the first view in each individual mouse . For dendrite image display , fluorescent structures near and out of the focal plane of the dendrites of interest were removed manually from image stacks using Adobe Photoshop . The modified image stacks were then projected to generate two-dimensional images and adjusted for contrast and brightness . Data are represented as mean + SEM/SD in quantitative analysis . Statistical analysis was performed with Student’s t test or One-way ANOVA followed by Tukey post-hoc test . If comparison was made across grouped data , Two-way ANOVA with Tukey post-hoc test was used . If dataset did not follow a normal distribution as detected by Shapiro-Wilk normality test , Mann-Whitney test or Kruskal-Wallis test with post-hoc Dunnett's multiple comparison test was used . Statistical significances were defined as p<0 . 05 . | Transporting molecules within a cell becomes a daunting task when the cell is a neuron , with fibers called axons and dendrites that can stretch as long as a meter . Neurons use many different molecules to send messages across the body and store memories in the brain . If the right molecules cannot be delivered along the length of nerve cells , connections to neighboring neurons may decay , which may impair learning and memory . Motor proteins are responsible for transporting molecules within cells . Kinesins are a type of motor protein that typically transports materials from the body of a neuron to the cell’s periphery , including the dendrites , which is where a neuron receives messages from other nerve cells . Each cell has up to 45 different kinesin motors , but it is not known whether each one performs a distinct task or if they have overlapping roles . Now , Zhao , Fok et al . have studied two similar kinesins , called KIF5A and KIF5B , in rodent neurons to determine their roles . First , it was shown that both proteins were found at dendritic spines , which are small outgrowths on dendrites where contact with other cells occurs . Next , KIF5A and KIF5B were depleted , one at a time , from neurons extracted from a brain region called the hippocampus . Removing KIF5B interfered with the formation of dendritic spines , but removing KIF5A did not have an effect . Dendritic spines are essential for learning and memory , so several behavioral tests were conducted on mice that had been genetically modified to express less KIF5B in the forebrain . These tests revealed that the mice performed poorly in tasks that tested their memory recall . This work opens a new area of research studying the specific roles of different kinesin motor proteins in nerve cells . This could have important implications because certain kinesin motor proteins such as KIF5A are known to be defective in some inherited neurodegenerative diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2020 | Specific depletion of the motor protein KIF5B leads to deficits in dendritic transport, synaptic plasticity and memory |
Synovial joints are the lubricated connections between the bones of our body that are commonly affected in arthritis . It is assumed that synovial joints first evolved as vertebrates came to land , with ray-finned fishes lacking lubricated joints . Here , we examine the expression and function of a critical lubricating protein of mammalian synovial joints , Prg4/Lubricin , in diverse ray-finned fishes . We find that Prg4 homologs are specifically enriched at the jaw and pectoral fin joints of zebrafish , stickleback , and gar , with genetic deletion of the zebrafish prg4b gene resulting in the same age-related degeneration of joints as seen in lubricin-deficient mice and humans . Our data support lubricated synovial joints evolving much earlier than currently accepted , at least in the common ancestor of all bony vertebrates . Establishment of the first arthritis model in the highly regenerative zebrafish will offer unique opportunities to understand the aetiology and possible treatment of synovial joint disease .
Synovial joints allow for free movement between adjacent bones and are characterized by a fluid-filled cavity separating layers of hyaline articular cartilage . The synovial cavity is enclosed by a membrane , which is often strengthened externally by a fibrous capsule and contains lubricating molecules , such as hyaluronic acid and lubricin , that reduce friction at the joint surface ( Koyama et al . , 2014; Rhee et al . , 2005 ) . A prevailing hypothesis is that lubricated synovial joints first evolved in tetrapods in response to newfound mechanical challenges imposed on the weight-bearing joints of nascent limbs ( van der Kraan , 2013a , 2013b ) ( Figure 1A ) . Whereas previous histological studies had suggested that the jaw joints of lungfish ( a lobe-finned fish like humans ) ( Bemis , 1986 ) , and potentially longnose gar and sturgeon ( ray-finned fishes ) ( Haines , 1942 ) , have synovial-like morphology , the extent to which these joints are molecularly and functionally similar to tetrapod synovial joints had remained untested . In particular , the assumption that ray-finned fishes lack the sophisticated types of lubricated joints found in humans has hampered the use of the zebrafish model for the study of synovial joint diseases such as arthritis . By examining the expression and function of homologs of a critical lubricating protein of mammalian joints , Lubricin , we provide evidence that certain joints of adult zebrafish are indeed true synovial joints . 10 . 7554/eLife . 16415 . 003Figure 1 . Synovial-like morphology of jaw and fin joints in ray-finned fish . ( A ) Phylogenetic tree contrasts the old model of synovial joint evolution ( grey asterisk ) with the new model of synovial joints evolving in a common precursor of all bony vertebrates ( black asterisk ) . ( B ) Alcian Blue-stained adult zebrafish and accompanying diagrams show the pectoral fin joints , and jaw joints in open and closed positions . ( C–E , G–I ) Sections of 14 dpf ( n = 4 ) , 1 mpf ( n = 3 ) , and adult ( n = 6 ) zebrafish jaw joints; ray-scapula joint in the adult zebrafish pectoral fin ( n = 4 ) ; and stickleback ( 1 mpf , n = 3 ) and spotted gar ( 10 . 2 cm , n = 3 ) jaw joints . Sections are stained by H&E ( C , D , G–I ) or trichrome ( E ) . Articular chondrocytes ( black arrowheads ) line the cavity . ( F ) Schematic of adult jaw joint shows bone ( red ) , cartilage ( blue , lighter shade indicates articular ) , and synovial membrane ( green ) . ( J , K ) Magnifications of ( I ) show the synovial membrane ( arrow ) , fibrous capsule ( asterisk ) and multilayered articular cartilage ( K ) . Scale bar in h , 100 μm; all other panels , 50 μm . aa: anguloarticular; q: quadrate; sc: scapula; r: ray; pr: proximal radial; dr: distal radial; M: Meckel’s; pq: palatoquadrate; s: superficial; t: transitional; rd: radial layer; c: calcified cartilage; b: bone . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 003
Given the suggested synovial-like morphology of jaw joints in several fishes , we examined whether joints of the widely used teleost species , the zebrafish ( Danio rerio ) , also display synovial morphology ( Figure 1A ) . Bone μCT of adult zebrafish shows that the jaw joint , an articulation between the anguloarticular and quadrate bones , resembles a hinge joint ( Video 1 ) , with manual opening and closing of the mouth in fixed Alcian-Blue-stained animals revealing movement in a single plane ( Figure 1B ) . While study of zebrafish has contributed to our understanding of the embryonic development of the jaw joint , zebrafish larvae at the most commonly studied stage , 6 days post-fertilization ( dpf ) , show little evidence of a synovial cavity ( Miller et al . , 2003 ) . However , whether this joint acquires synovial characteristics later had not been described . Our histological investigations revealed the variable presence of a partial cavity as early as 14 dpf , and a prominent and consistently present cavity by 28 dpf ( Figure 1C , D ) . In adult zebrafish ( 12 months post-fertilization , mpf ) , distinct layers of flattened articular chondrocytes line the jaw joint cavity , with hypertrophic chondrocytes located beneath the articular surface ( Figure 1E , F ) . We also observed similar cavities lined by flattened chondrocytes in joints of the pectoral fin in adult zebrafish , in particular between the proximal and distal radials and between the marginal ray and scapula ( Figure 1B , G ) . We next used sox10 and tricho-rhino-phalangeal syndrome 1 ( trps1 ) transgenes ( Askary et al . , 2015 ) to visualize jaw joint cavitation over time in living zebrafish . At 3 , 7 , and 14 dpf , the early chondrocyte marker trps1:GFP labels the subset of sox10:dsRed+ chondrocytes at the joint surface , with a partial cavity variably apparent at 14 dpf ( Figure 2A–C ) . At 1 , 2 and 8 mpf , trps1:GFP is maintained in articular chondrocytes as the cavity expands , with a subset of cells marked by sox10:dsRed ( Figure 2D–F ) . In the jaw of the relatively small zebrafish , the presence of a cavity and joint-lining cells , with a minimal fibrous capsule , is consistent with it being a synovial-like joint in miniature , similar to the homologous incudomalleolar synovial joint of the mammalian middle ear ( Whyte et al . , 2002 ) . 10 . 7554/eLife . 16415 . 004Video 1 . Bone μCT of the adult zebrafish head . 3D reconstruction of bone shows the structure and relative position of the jaw joint within the adult zebrafish head ( 12 mpf ) . Color-coding at the end of the movie illustrates the anguloarticular ( pink ) and quadrate ( yellow ) bones that articulate at the jaw joint . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 00410 . 7554/eLife . 16415 . 005Figure 2 . Live imaging of jaw joint cavitation . ( A–F ) In double transgenic zebrafish , sox10:DsRed marks all chondrocytes and trps1:GFP labels nascent joint chondrocytes ( arrow ) and perichondrial cells at 3 and 7 dpf . By 14 dpf , a partial cavity ( asterisk ) was evident in 1/6 animals . In 3/3 animals at 1 mpf , 4/4 at 2 mpf , and 4/4 at 8 mpf , a fully formed cavity is evident and sox10:DsRed is expressed in a subset of trps1:GFP+ articular chondrocytes . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 005 To examine whether synovial-like morphology is a conserved feature of ray-finned fish , we also examined juveniles of the distantly related teleost fish , the three-spined stickleback ( Gasterosteus aculeatus ) , and an outgroup of teleost fish , the ray-finned spotted gar ( Lepisosteus oculatus ) ( Braasch et al . , 2016 ) . Both stickleback and spotted gar jaw joints displayed synovial cavities lined by flattened cells ( Figure 1H , I ) . In the jaw joint of the larger spotted gar ( standard length 10 . 2 cm ) , an internal one-cell-thick membrane and thick external fibrous capsule enclosed the cavity , with joint cartilage divided into the same superficial , transitional , radial , and calcified layers seen in mammalian synovial joints ( Figure 1J , K ) . The presence of these additional morphological features in the gar jaw supports synovial-like features being an ancestral property of bony vertebrates . Given the synovial-like morphology of several types of joints in fish , we next examined whether the chondrocytes lining these joints share a common molecular signature with those of mammalian joints . Chondrocytes lining mammalian synovial joints differ from those in the growth plate by expressing Prg4 , which encodes a lubricin proteoglycan that forms a cross-linked network with hyaluronan and Aggrecan to reduce friction across the joint surface ( Jay and Waller , 2014 ) . Consistent with the appearance of a synovial cavity in juvenile stages , we find that the joint-lining cells of the zebrafish jaw express prg4b starting from 15 dpf and continuing throughout adulthood ( Figure 3A–D ) . We observed much weaker levels of expression at other joints of the face , such as the midline ceratohyal-ceratohyal joint ( Figure 3A , arrowhead ) and hyoid joint ( Figure 3E ) , which lack synovial morphology . Expression of prg4b was not detected in the jaw joint at earlier stages ( 7 dpf , data not shown ) , consistent with the late onset of Prg4 expression at mammalian joints ( Rhee et al . , 2005 ) . In addition , prg4b expression appeared in joint-lining cells of the synovial-like ray-scapula articulation of the pectoral fin at 3 mpf , but not in the non-synovial intervertebral discs ( Figure 3F , G ) . Zebrafish also expressed prg4b outside of joints , including conserved expression with mammalian Prg4 in liver ( Ikegawa et al . , 2000 ) and possibly ligaments ( Sun et al . , 2006 ) ( Figure 3A , G ) . Similar to zebrafish , stickleback expressed prg4b and gar expressed prg4 in joint-lining cells of the juvenile jaw ( Figure 3H–J ) . In contrast , the related prg4a gene is not enriched at the jaw joint in either zebrafish or stickleback , instead showing expression throughout cartilage ( Figure 3—figure supplement 1 ) . Of note , the lineage leading to the spotted gar diverged before the teleost genome duplication ( Amores et al . , 2011 ) , resulting in zebrafish and stickleback having two Prg4 co-orthologs and gar a single ortholog ( Figure 3—figure supplement 2 ) . Analysis of the single prg4 gene in gar therefore reveals that enriched expression of Prg4 within articular chondrocytes existed before the divergence of ray-finned and lobe-finned vertebrates . 10 . 7554/eLife . 16415 . 006Figure 3 . Expression of Prg4 genes in articular chondrocytes of ray-finned fish . ( A–G ) prg4b expression in articular chondrocytes of the zebrafish jaw joint ( boxed region in A , B–D ) , hyoid joint ( E ) ; ray-scapula joint ( F ) ; and vertebral column ( G ) . prg4b is also expressed in the liver ( asterisk ) , possibly in ligaments above the vertebrae , and weakly at the ceratohyal-ceratohyal joint ( arrowhead ) . n = 3 each . ( H–J ) Expression of stickleback prg4b ( 1 mpf , n = 3 ) and gar prg4 ( 10 . 2 cm , n = 3 ) in jaw joint articular chondrocytes ( J , magnification of I ) . ( K–M ) Exclusion of acana , col10a1 , and matn1 expression from articular chondrocytes of the zebrafish jaw . n = 3 each . Scale bar , 50 μm . ih: interhyal . See also Figure 3—figure supplement 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 00610 . 7554/eLife . 16415 . 007Figure 3—figure supplement 1 . Gene expression within the zebrafish and stickleback jaw joints . ( A–D ) In situ hybridization reveals broad chondrocyte expression of prg4a but no enrichment within jaw joint articular chondrocytes ( C–D , magnified views ) . ( E ) Zebrafish has3 is expressed in chondrocytes just underneath the articular surface ( arrow ) and in a small number of cells within the growth plate ( arrowhead ) . ( F ) Expression of matn1 is excluded from superficial chondrocytes of the zebrafish jaw joint at 1 mpf . ( G , H ) Immunofluorescence staining for Col2a1 and Aggrecan protein reveals broad cartilage expression yet exclusion from jaw articular chondrocytes . n = 3 each panel . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 00710 . 7554/eLife . 16415 . 008Figure 3—figure supplement 2 . Evolution of vertebrate Proteoglycan 4 ( Lubricin ) . ( A ) Maximum likelihood phylogeny of vertebrate Prg4 proteins generated using the LG model in PhyML ( Guindon et al . , 2009 ) . The tree was rooted with spotted gar and human Vitronectin ( Vtn ) proteins . In the ray-finned lineage , after divergence of the holostean lineage ( spotted gar and bowfin ) , the teleost genome duplication ( TGD ) generated Prg4a and Prg4b found in extant teleosts . GenBank/Ensembl accession numbers for sequences in the MAFFT ( Katoh and Standley , 2014 ) alignment: zebrafish: Prg4a , NP_997918 , Prg4b , XP_005160745; stickleback: Prg4a , ENSGACP00000019616 , Prg4b , concatenate of ENSGACP00000009961/ENSGACP00000009951/ENSGACP00000009946; medaka: Prg4a , ENSORLP00000017289 , Prg4b , ENSORLP00000011688; Spotted gar: Prg4 , XP_015210531 , Vtn , XP_006641129; bowfin: Prg4 , AAC_TPR . 1 . 1 [http://phylofish . sigenae . org/]; human: PRG4 , NP_005798 , VTN , NP_000629; mouse: Prg4 , NP_067375; chicken: Prg4 , ENSGALP00000008204; Anole lizard: Prg4 , XP_008112908; Western clawed frog: Prg4 , XP_012817383; coelacanth: Prg4 , ENSLACP00000017445 . ( B ) Orthologous pairwise synteny cluster from the Synteny Database ( Catchen et al . , 2009 ) ( window size: 50 genes ) shows extensive conserved synteny of Prg4 gene regions on human chromosome Hsa1 and spotted gar Loc10 , supporting their orthology . ( C ) Composite cluster from the Synteny Database ( window size: 50 genes ) compares the prg4 gene regions in zebrafish to the TGD outgroup spotted gar . Double conserved synteny of prg4 on gar Loc10 and zebrafish prg4a on Dre2 and prg4b on Dre20 , supports the prg4 duplication as result of the TGD . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 008 We next examined whether articular chondrocytes of the zebrafish jaw display other features of chondrocytes lining mammalian synovial joints , including enriched expression of hyaluronan synthase ( Has ) enzymes in the radial layer ( Hiscock et al . , 2000 ) and lower levels of types II and X Collagen , Aggrecan , and Matrilin compared to growth plate chondrocytes ( Khan et al . , 2007 ) . Consistently , we observed relative depletion of col10a1 , acana , and matrilin1 mRNA and Col2a1a and Aggrecan protein in articular versus deeper chondrocytes , as well as enriched expression of has3 in the radial layer of the juvenile zebrafish jaw joint ( Figure 3K–M , and Figure 3—figure supplement 1 ) . Although none of these markers are exclusive to synovial joints , the shared expression of Lubricin and hyaluronan synthase enzymes and relative absence of cartilage maturation genes ( e . g . Collagen II/X , Aggrecan , Matrilin1 ) demonstrates a common molecular signature between articular chondrocytes of the zebrafish jaw and mammalian synovial joints . A major function of the synovial cavity is to lubricate the joint , with loss of lubrication resulting in age-related joint degeneration . We therefore asked whether expression of Prg4 orthologs by cells lining synovial-like joints in ray-finned fishes reflects a conserved requirement of lubricin protein in lubricating and hence maintaining these joints . To do so , we used TALE nucleases ( Huang et al . , 2011; Sander et al . , 2011 ) to generate loss-of-function deletion alleles for zebrafish prg4a and prg4b ( Figure 4A ) . Mice lacking Prg4 function ( Koyama et al . , 2014; Rhee et al . , 2005 ) and humans with homozygous loss of PRG4 in Camptodactyly-arthropathy-coxa vara-pericarditis syndrome ( Marcelino et al . , 1999 ) have a progressive deterioration of joint surfaces that includes loss of articular chondrocytes , accumulation of acellular matrix in the cavity , synovial hyperplasia , and thickening of the deep chondrocyte layer . Consistent with the lack of early joint defects in Prg4-/- mice , zebrafish doubly mutant for prg4a and prg4b had no defects in the jaw joint at either 6 dpf or 1 mpf and no gross defects of the adult skeleton ( Figure 4B–D ) . However , prg4a-/-; prg4b-/- zebrafish began to display a weak accumulation of acellular matrix in limited domains of the joint surface at 2 mpf ( Figure 4E ) , a phenotype that became more severe by 6 mpf ( Figure 4F ) . By 12 mpf , prg4a-/-; prg4b-/- zebrafish had multiple jaw joint abnormalities , including an acellular matrix at the joint surface , synovial hyperplasia , increased numbers of deep chondrocytes , and in some cases complete erosion of the joint surface accompanied by underlying bone defects ( Figure 4G , Figure 4—figure supplement 1 , and additional examples in Figure 4—figure supplement 2 ) . Quantification of jaw joint defects using an Osteoarthritis Research Society International ( OARSI ) scoring system ( Pritzker et al . , 2006 ) that we modified for zebrafish ( Figure 4—figure supplement 2 ) confirmed that joint defects increased in severity during aging ( Figure 4H ) , consistent with the progressive arthritis seen in Prg4-/- mice ( Koyama et al . , 2014; Rhee et al . , 2005 ) and patients with CACP ( Marcelino et al . , 1999 ) . Defects found in prg4a-/-; prg4b-/- double mutants were also not confined to the jaw joint , appearing in the ray-scapula and inter-radial joints of the pectoral fin ( Figure 5A–C ) . Further , consistent with only prg4b being enriched in jaw joint chondrocytes , prg4b but not prg4a single mutants displayed ray-scapula fin joint defects and a similar severity of jaw joint defects as double mutants at 12 mpf ( Figure 4G , H and Figure 5B ) . In contrast , we detected no changes in the hyoid joints of the face and the intervertebral discs in 12 mpf prg4a-/-; prg4b-/- mutants ( Figure 5D , E ) , consistent with these joints lacking synovial cavities and high-level prg4b expression . Our data therefore support a specific requirement for zebrafish Prg4 homologs in the adult maintenance of joints with synovial-like morphology . 10 . 7554/eLife . 16415 . 009Figure 4 . Progressive deterioration of the jaw joint in zebrafish lacking prg4b . ( A ) Schematics of prg4a and prg4b TALEN mutants show deleted sequences . ( B ) X-ray imaging shows no gross morphological bone defects of prg4a-/-; prg4b-/- mutants ( DKO ) at 12 mpf . ( C ) Alcian Blue staining shows normal facial cartilages in DKO at 6 dpf . ( D ) SafraninO staining shows a normal jaw joint between Meckel’s ( M ) and palatoquadrate ( pq ) cartilages in DKO at 1 mpf . ( E–G ) SafraninO staining at 2 , 6 and 12 mpf shows increasingly abnormal joints between the jaw anguloarticular ( aa ) and quadrate ( q ) bones in DKO . Defects include acellular matrix accumulation in the cavity ( asterisks , see magnified regions in E’–G’ ) at all stages , and synovial hyperplasia ( arrowheads ) and expanded deep chondrocytes ( arrows ) at 12 mpf . Single prg4b but not prg4a mutants showed similar jaw joint defects to DKO at 12 mpf . ( H ) Quantification of jaw joint defects using our modified OARSI system for zebrafish . Scale bar 50 μm , except in ( B ) . See also Figure 4—figure supplements 1 , 2 , and Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 00910 . 7554/eLife . 16415 . 010Figure 4—source data 1 . Quantification of joint defects in zebrafish lacking prg4 genes . This file includes the average OARSI scores for wild types and mutants related to Figure 4H . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 01010 . 7554/eLife . 16415 . 011Figure 4—figure supplement 1 . Serial sections through a representative wild-type and prg4a-/-; prg4b-/- mutant jaw joint . SafraninO staining of a serial series of 5 μm sections were imaged every third section to capture a complete wild-type and prg4a-/-; prg4b-/- ( DKO ) jaw joint at 12 mpf . See also Figure 4G . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 01110 . 7554/eLife . 16415 . 012Figure 4—figure supplement 2 . OARSI scoring system for zebrafish . SafraninO-stained histological sections demonstrate the defining features for each grade of joint damage at the zebrafish jaw . Grade 0 – a smooth intact surface with normal superficial and deeper cartilage layers ( 12 mpf , wild-type ) . Grade 1 – an uneven surface cartilage with small fibrillations limited to the superficial layer ( 12 mpf , wild-type ) . Grade 2 – the superficial layer is disrupted with focal fibrillations and some matrix loss ( 12 mpf , DKO ) . Grade 3 – vertical clefts or fissures extend beyond the superficial layer , disrupting the deeper cartilage ( 12 mpf , DKO ) . Grade 4 – erosion of the superficial layer of cartilage with matrix loss extending into deeper cartilage layers ( 12 mpf , DKO ) . Grade 5 – cartilage is completely lost , exposing the bone surface ( 12 mpf , prg4b-/- ) . Grade 6 – the exposed bone surface is deformed , showing altered contour of the joint ( 12 mpf , DKO ) . Individual grade and stage values were determined for both the anterior and posterior sides of the left and right jaw joints from three histological sections per joint . Overall joint OA scores were generated as an average of ( grade ) x ( stage ) for each animal . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 01210 . 7554/eLife . 16415 . 013Figure 5 . Requirement of Prg4 gene function for fin joints of zebrafish . ( A ) Schematic of pectoral fin joints . ( B–E ) SafraninO staining at 12 mpf shows abnormal joints between the ray ( r ) and scapula ( sc ) joint of the pectoral fin ( B ) in prg4b but not prg4a single mutants and prg4a; prg4b double mutants ( DKO ) , and defects in the proximal radial ( pr ) and distal radial ( dr ) joint of the pectoral fin ( C ) in DKO . Defects include acellular matrix accumulation in the cavity ( asterisks ) , synovial hyperplasia ( arrowheads ) , and expanded deep chondrocytes ( arrow ) . In contrast , the hyoid joint ( D ) and intervertebral discs ( E ) were normal in DKO . Phenotypes were consistently observed in three animals of each genotype . Scale bar 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16415 . 013
Although we do not know if Prg4b protein is secreted into joint cavities in zebrafish , or whether the cavities are fully enclosed as in mammals , the finding that the jaw and fin joints of zebrafish require the molecular lubricant lubricin for their maintenance supports zebrafish having synovial-like lubricated joints . Contrary to existing dogma ( van der Kraan , 2013b ) , our data suggest that lubricated synovial joints evolved before vertebrates moved onto land in at least the common ancestor to all bony vertebrates . Although fish are not subject to gravity in the same way as tetrapods , the evolution of synovial joints in fish-like ancestors may have facilitated movement of first the jaw , and then the fins , against water resistance . Interestingly , recent studies show that the radial joints of the pectoral fin , which we find to have synovial morphology and prg4b dependency in zebrafish , are homologous to those of the tetrapod wrist ( Gehrke et al . , 2015 ) . The existence of these synovial joints at the base of the pectoral fins of ancestral bony fish may therefore have facilitated their later functional evolution into the larger synovial joints of tetrapod limbs . Our findings show that zebrafish can be a relevant model to understand the development of synovial specializations , including the poorly understood process of cavitation . The establishment of the first genetic model of arthritis in zebrafish will also allow a better understanding of the developmental progression of synovial joint disease . In the future , it will be exciting to test whether the articular cartilage lining synovial joints , which is affected in arthritis , displays the same regenerative potential as many of the other tissues in zebrafish ( Knopf et al . , 2011; Morgan , 1900; Poss et al . , 2002 ) .
The Institutional Animal Care and Use Committees of the University of Southern California , the University of California , Berkeley , and the University of Oregon approved all zebrafish ( Danio rerio ) , three-spined stickleback ( Gasterosteus aculeatus ) and spotted gar ( Lepisosteus oculatus ) procedures , respectively – ( IACUC #10885 ) . Previously reported zebrafish lines used in this study include Tg ( sox10:dsRED ) el10 ( Das and Crump , 2012 ) , trps1j127aGt ( RRID:ZFIN_ZBD-GENO-100809-11 ) ( Talbot et al . , 2010 ) , and Tg ( col2a1aBAC:GFP ) el483 ( Askary et al . , 2015 ) . The phylogenetic tree was generated using phyloT ( http://phylot . biobyte . de/ ) with NCBI taxonomy , and visualized in iTOL:Interactive Tree of Life ( http://itol . embl . de/ ) . Adult zebrafish samples were fixed in 4% PFA at 4°C for 7 days . Following fixation , animals were cut into smaller pieces to facilitate embedding and sectioning of the desired structures . The samples were then decalcified in 20% EDTA solution for 10 days at room temperature . For embedding , the tissue was first dehydrated through a series of ethanol washes ( 30 , 50 , 70 , 95 , and 100% ) for 20 min each . Then ethanol was replaced with xylene substitute Hemo-De ( Electron Microscopy Sciences , Hatfield , PA ) in a series of 15 min washes ( 50 , 75 , and 100% Hemo-De ) . The samples were then incubated in a 1:1 Hemo-De:paraffin ( Paraplast X-tra , VWR , Radnor , PA ) solution at 65°C for an hour before an overnight incubation in 100% paraffin at 65°C . The following day , the samples were embedded in freshly melted paraffin . Larval and juvenile samples were prepared following the same general procedure with shorter fixation and decalcification steps ( 2 days fixation and 4 days decalcification ) . Animals younger than 3 weeks did not require decalcification . Juvenile stickleback and spotted gar samples were processed as above for adult zebrafish with the following changes: stickleback were decalcified in 20% EDTA for 7 days at room temperature; isolated spotted gar heads were fixed in 4% PFA at 4°C for 10–14 days and decalcified in 20% EDTA for 14 days at room temperature . For hematoxylin & eosin ( H&E ) staining , 5 μm paraffin sections were deparaffinized with xylene and re-hydrated through an ethanol series to distilled water . Sections were then stained in hematoxylin ( VWR ) for 2 min followed by brief acetic acid rinse and 2 min in Blueing Reagent solution ( VWR ) ; 2 × 30 s washes in water , 2 × 30 s in 95% ethanol . The sections were stained in Eosin ( VWR ) for 30 s followed by 3 × 1 min washes in 95% ethanol and 2 × 1 min washes in 100% ethanol . Following 2 × 2 min in Hemo-De , samples were mounted with cytoseal ( Richard-Allan Scientific , Kalamazoo , MI ) for imaging . For Trichrome staining , 5 μm paraffin sections were deparaffinized with xylene and re-hydrated through an ethanol series to distilled water . Trichrome stain was performed according to manufacturer’s instructions using the Trichrome , Gomori One-Step , Aniline Blue Stain ( Newcomer Supply , Middleton , WI ) . For SafraninO staining , 5 μm paraffin sections were deparaffinized with xylene and re-hydrated to distilled water . They were then stained in Weigert’s Iron Hematoxylin ( Newcomer Supply ) for 5 min , washed in distilled water for 5 changes , differentiated in 0 . 06 N HCl solution in 70% ethanol for 2 s followed by 3 more washes in water . Sections were then stained in 0 . 02% Fast Green FCF ( Sigma-Aldrich , St . Louis , MO ) for 1 min and rinsed for 30 s in 1% acetic acid . Staining in 1% SafraninO ( Newcomer Supply ) was then performed for 30 min , followed by 3 × 1 min washes in 95% ethanol . Slides were then washed 2 × 1 min in 100% ethanol and 2 × 2 min in Hemo-De , before mounting with cytoseal for imaging . For Alcian staining , juvenile zebrafish were processed as described ( Walker and Kimmel , 2007 ) , and adult zebrafish were processed using a modified version of an online protocol ( https://wiki . zfin . org/pages/viewpage . action ? pageId=13107375 ) . Probes were generated for zebrafish prg4a , prg4b , has3 , acana , col10a1 , and matn1 using PCR amplification from cDNA . See Supplementary file 1A for primers used in amplification of zebrafish probe templates . Probes for stickleback prg4a and prg4b and gar prg4 were synthesized from gBlocks Gene Fragments ( Integrated DNA Technologies , Coralville , IA ) . The DNA sequences used for probe generation are listed in Supplementary file 1B . Stickleback prg4b sequence was determined through sequence homology comparison with zebrafish Prg4b protein sequence using BlastP search in Ensembl . Two independent probes were designed for the putative stickleback prg4b ( ENSGACG00000007505; ENSGACG00000007501 ) and both showed the same expression pattern in three independent animals . gBlocks or PCR products from cDNA amplification were subcloned into pCR-Blunt II TOPO vector ( Invitrogen , Carlsbad , CA ) . Following sequence confirmation , digoxigenin ( DIG ) -labeled antisense probes were synthesized using T7 or SP6 RNA polymerase ( Roche , Switzerland ) . In situ hybridization protocol was modified from ( Lien et al . , 2006 ) . In brief , after de-paraffinization , slides were digested in 7 . 5 µg/ml proteinase K for 5 min and fixed in 4% PFA/0 . 2% gluteraldehyde for 20 min . Each slide was incubated overnight at 65°C with 1 μg of DIG-labeled riboprobe diluted in hybridization buffer . After hybridization , slides were washed three times in 1x SSC/Formamide at 65°C for 30 min , and three times in MABT for 15 min . Following 1 hr of blocking in 2% Blocking Buffer ( Roche ) , hybridization was detected with anti-DIG-AP antibody ( Roche , RRID:AB_514497 ) and developed with NBT/BCIP substrate colorimetric reaction ( Roche ) . Slides were counterstained with nuclear fast red ( Vector Laboratories , Burlingame , CA ) prior to mounting . Paraffin sections ( 5 µm ) were de-paraffinized with xylene and rehydrated through an ethanol series to 1xPBST . Antigen retrieval was performed using pH 6 . 0 sodium citrate buffer in a steamer for 35 min . The sections were blocked with 2% donkey serum in PBST for 30 min at room temperature . The sections were incubated with primary antibodies against Col2a1 ( goat polyclonal , SC7763 , Santa Cruz Biotechnology , Santa Cruz , CA , RRID:AB_2229686 ) and Aggrecan ( Cat# 13880-1-AP , Proteintech , Rosemont , IL ) overnight at 4°C . Sections were further incubated in secondary antibodies and Hoechst 33342 nuclear stain for 1 hr at room temperature prior to mounting with Fluoromount-G ( Southern Biotech , Birmingham , AL ) . Fluorescence imaging of live animals and sections was performed using a Zeiss LSM5 confocal microscope . Histology and colorimetric in situ hybridization slides were photographed using a Leica D8 2500 microscope . Alcian Blue-stained samples were imaged using a Leica S8APO microscope . For microCT ( µCT ) , adult zebrafish were euthanized and fixed in 4% PFA overnight and rinsed twice in 1x PBS . The fish head was dissected and glued to a Pasteur pipette to place it next to the scan head . The scans were performed in air on a XT H 225S T µCT scanner ( Nikon Metrology , Brighton , MI ) with a PerkinElmer 1621 detector at 120 kVp , 26 uA 500 ms exposure time resulting in an isotropic 3 micron voxel volume . A molybdenum target was used with no additional filtration of the beam . Raw data were reconstructed in CT Pro 3D v4 . 3 . 4 ( XT Software Suite , Nikon Metrology ) and video rendering performed on VG StudioMax v2 . 2 ( Volume Graphics GmbH , Germany ) . X-ray imaging of fixed adult zebrafish utilized an UltraFocus60 x-ray cabinet ( Faxitron Bioptics , Tucson , AZ ) . To make knockout alleles for prg4a and prg4b , two pairs of Transcription Activator-Like Effector Nucleases ( TALENs ) were used per gene to remove most of the coding sequence of the target genes . TALENs were designed and constructed as described previously ( Sanjana et al . , 2012 ) . TALEN pairs were targeted toward the following sequences in the zebrafish genome: TTGGTCTCTTCTGGCTCTGC and TCGTCTGCTGCTCAGGGTGA for prg4a 5’ TALENs , TAGGCGTCCCGTCACCCATT and CGCTGCAACTGCCAGGGCAA for prg4a 3’ TALENs , TGCTGTTTGTGTGGGTCTCC and CACGTAAGCCAACAGATCGA for prg4b 5’ TALENs , and TCCCCCAGCTGCAGCACTGG and TCTCACGAACCTGGAGAGGA for prg4b 3’ TALENs . ARCA-capped RNA for each TALEN was transcribed in vitro using the mMESSAGE mMACHINE T7 Ultra Kit ( Life Technologies , Carlsbad , CA ) . The RNA for four TALENs corresponding to each gene were mixed and injected into one-cell stage embryos to generate founder lines . Founders were then outcrossed to wild-type fish and the progenies were tested for genomic deletions . Deletion alleles were identified in the F1 generation by cloning and sequencing of amplicons that span the TALEN target sites for each gene . We obtained single prg4ael687 ( 7686 bp deletion ) and prg4bel594 ( 11 , 638 bp deletion ) alleles by screening the progeny of 6 and 21 injected animals , respectively . For the list of genotyping primers see Supplementary file 1C . 5 µm serial sections throughout the jaw joint were stained with SafraninO and imaged as described above . For both the left and right jaw joint , three representative images were selected for each sample . Care was taken to select images that represent equivalent sectioning planes in wild-type and mutant joints . Grade ( 0–6 ) and stage ( 0–4 ) values were assigned to the anterior ( anguloarticular ) and posterior ( quadrate ) surfaces in each of the six joint images for each sample . The grade and stage value for each individual articular surface were multiplied to generate the score for that joint surface . Scores for both surfaces of the left and right joints were then averaged for each animal . Data are presented as a scatter plot with a line for the mean . Statistical analysis of average OARSI scores was performed using GraphPad Prism 7 ( RRID:SCR_002798 ) , with a two-tailed Student’s t-test used to generate the p values . | We owe our flexibility to the lubricated joints that connect the bones of our body . Unfortunately , these joints tend to deteriorate over time , leading to a condition called osteoarthritis that affects millions of people . Scientists had thought that lubricated joints first evolved when backboned animals started walking on land , with fish lacking these types of joints . However , by studying zebrafish , Askary , Smeeton et al . now show that fish do have lubricated joints; in fact , the joints in the jaw and fins of zebrafish have a similar structure to those in humans . These zebrafish joints make an important protein called Lubricin that is known to lubricate joints in mice and humans . Furthermore , analyzing two other fish species – a stickleback and a primitive fish called a spotted gar – revealed that fish joints in general produce Lubricin . This pushes back the evolutionary origins of lubricated joints millions of years , to at least the common ancestor of all backboned animals . Next , Askary , Smeeton et al . used a new type of molecular scissors to eliminate the ability of zebrafish to produce Lubricin . These mutant fish developed the same early onset arthritis as mice and humans that lack Lubricin . Studying such fish should allow new approaches to be developed that will help us to understand how debilitating joint diseases progress . As zebrafish are highly regenerative , future studies could also explore whether they can regenerate damaged joints , which could spur new strategies for treating and reversing arthritis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2016 | Ancient origin of lubricated joints in bony vertebrates |
Soluble karyopherins of the importin-β ( impβ ) family use RanGTP to transport cargos directionally through the nuclear pore complex ( NPC ) . Whether impβ or RanGTP regulate the permeability of the NPC itself has been unknown . In this study , we identify a stable pool of impβ at the NPC . A subpopulation of this pool is rapidly turned-over by RanGTP , likely at Nup153 . Impβ , but not transportin-1 ( TRN1 ) , alters the pore's permeability in a Ran-dependent manner , suggesting that impβ is a functional component of the NPC . Upon reduction of Nup153 levels , inert cargos more readily equilibrate across the NPC yet active transport is impaired . When purified impβ or TRN1 are mixed with Nup153 in vitro , higher-order , multivalent complexes form . RanGTP dissolves the impβ•Nup153 complexes but not those of TRN1•Nup153 . We propose that impβ and Nup153 interact at the NPC's nuclear face to form a Ran-regulated mesh that modulates NPC permeability .
The nuclear pore complex ( NPC ) is a very large cellular transport channel conserved among all eukaryotes . The NPC controls the nuclear entry and exit of cargos ranging from single proteins to large ribonucleoprotein complexes ( Stewart , 2007; Peters , 2009 ) . Cargos smaller than ∼40 kDa can passively equilibrate across the nuclear envelope while larger cargos must bind special transport receptors to move from the cytoplasm into the nucleus and accumulate there ( Stewart , 2007; Peters , 2009 ) . These transport receptors are able to bind cargos but can also interact with unstructured phenylalanine–glycine repeat proteins ( FG nucleoporins ) within the pore . Directional transport of cargos is powered by the small GTPase Ran and a system of compartment-specific GTP hydrolysis and GDP-to-GTP exchange , which establishes a sharp concentration gradient of RanGTP across the nuclear envelope ( Izaurralde et al . , 1997; Kalab et al . , 2002 ) . Importin-β ( impβ ) and other members of the karyopherin family of nuclear transport receptors form a complex with their cognate cargos in the RanGTP-low cytoplasmic environment and release cargos upon binding to RanGTP in the nucleus . Contemporary transport models ( ‘selective phase’ hydrogel [Ribbeck and Gorlich , 2001; Frey et al . , 2006] , ‘virtual gate’ [Rout et al . , 2003] , ‘reduction of dimensionality’ [Peters , 2005] , and ‘polymer brush’ [Lim et al . , 2007] ) address the behavior of FG nucleoporins and transport receptors to explain the NPC's selectivity and ability to facilitate cargo diffusion . A tacit implication of these models is that the diffusive movement of cargos through the NPC and the overall directionality of active transport are fundamentally distinct and separate processes . In this perspective , cargo–receptor complexes are expected to equilibrate freely across the nuclear envelope in the absence of an energy bias such as the RanGTP gradient , and efficient cargo accumulation against a concentration gradient requires only the Ran-driven unbinding of cargo molecules from their transport receptors . However , several intersecting lines of evidence raise the possibility that RanGTP influences the permeability of the NPC itself , rather than only acting on cargos once they have completely entered the nuclear compartment . First , early studies have suggested that Ran is needed for cargos to move through the NPC ( Moore and Blobel , 1993; Gorlich et al . , 1994; Moore and Blobel , 1994 ) . Second , it was shown that Ran plays an important role in dissociating impβ from the nuclear face of the NPC , in addition to displacing impβ from cargos ( Gorlich et al . , 1996 ) . Third , extended tracking of cargos within single pores revealed a substantial RanGTP-dependent asymmetry in the cargo's exit step . Without RanGTP , cargos entered the pore but had an ∼100-fold higher probability of exiting the pore at the cytoplasmic face than the nuclear face , suggesting that RanGTP influences barriers felt by cargo-impβ complexes within the pore ( Lowe et al . , 2010 ) . The Ran-sensitive exit asymmetry of large cargo-receptor complexes suggests that a currently unexplained Ran-dependent process takes place inside the channel near the nuclear face of the pore at about ∼70 nm along the transport axis ( Lowe et al . , 2010 ) ( Figure 1A ) . Fourth , the overall transport success of large cargos appears to be more sensitive to RanGTP levels than other cargos ( Lyman et al . , 2002; Snow et al . , 2013 ) and thus RanGTP somehow influences the interplay of cargo size and active transport . This latter connection is not necessarily mediated by cargo multivalency ( Snow et al . , 2013 ) . Together , these observations hint at additional Ran-driven processes within the pore that are not addressed by current models of active nucleocytoplasmic transport . 10 . 7554/eLife . 04052 . 003Figure 1 . Effect of Ran on impβ binding affinity and turnover at the NPC . ( A ) Schematic of the NPC showing the location of the Ran-dependent exit step for cargo-receptor complexes . ( B ) Representative images of cargo-receptor complexes and impβ-YFP stalled within the pore and forming bright nuclear rims in the absence of Ran . Fluorescence intensity profiles are plotted for the yellow lines showing the nuclear rim intensity drop when Ran is added . Scale bar ( white ) : 10 μm . ( C ) Nuclear rim fluorescence intensities of nuclei in ( B ) normalized to −Ran condition . Error bars represent the standard deviation about the mean . Asterisks ( * ) indicate a significant p < 10–20 using the Mann–Whitney U test ( N ≥ 97 for all conditions ) . ( D ) Representative FRAP recovery curves of impβ-YFP at the nuclear envelope showing rapid initial recovery of impβ in the presence of Ran with evidence of a second , slower pool of impβ . Recovery in the absence of Ran ( blue trace ) is considerably slower . ( E ) Photo-conversion based characterization of the slowly dissociating impβ pool . Ran reduces the initial fluorescence signal but does not clear all impβ's from the NPC as shown by the residual fluorescence at the nuclear rim lasting hundreds of seconds . Shaded regions indicate the standard error of the mean ( N = 20 for both conditions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 00310 . 7554/eLife . 04052 . 004Figure 1—figure supplement 1 . Effect of impβ concentration on active transport . The optimum concentration was ∼1 μM . At low impβ concentrations , cargo•impβ complex formation is limited by impβ availability , leading to low import levels . At high impβ concentrations , competition for NPC binding sites between free impβ and the cargo•impβ complexes leads to reduced import efficiency . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 00410 . 7554/eLife . 04052 . 005Figure 1—figure supplement 2 . Schematic of the FRAP microscope . Lasers are combined and passed through a Polarizing Beam Splitter ( PBS ) whereupon the polarized light is split into two paths . By adjusting the half-wave plates , one can adjust the amount of light through either of the paths . A pair of identical plano-convex lenses ( f1 = f2 ) allows the adjustment of the focus of the diffraction-limited spot along the optical axis . A shutter in this path controls temporal activation of the spot . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 00510 . 7554/eLife . 04052 . 006Figure 1—figure supplement 3 . Photoconversion experiment details . ( A ) False-color image of the photoswitched region within a single nucleus . ( B ) Mask generated for photoswitched region of the nuclear envelope . ( C ) Timecourse of the green channel at time points: pre-photoswitch , 0 , 50 , 100 , and 300 s . ( D ) Corresponding timecourse of photoswitched molecules in the red channel . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 006 The spatial and temporal arrangement of factor ( s ) that allows the NPC to almost perfectly prevent large cargo translocation in the absence of RanGTP ( error rate <1% [Lowe et al . , 2010] ) but allows efficient directional transport in its presence remains unknown . In this study , we investigate the interaction of impβ and Ran within single NPCs using quantitative biophysical measurements , and we relate those interactions to cargo translocation . We identify Nup153 as an important impβ binding partner at the nuclear face of the pore . The impβ•Nup153 interaction is Ran-sensitive and contributes to the NPC permeability barrier in vitro . Ran's effect on impβ turnover , stoichiometry , and spatial distribution at the nuclear pore is characterized , and the impβ•Nup153 binding behavior is examined . We propose a mechanism for how Ran-dependent modulation of impβ at Nup153 may contribute to the NPC's selective permeability .
We employed the commonly used in vitro nuclear transport assay ( digitonin permeabilized HeLa cells supplemented with exogenous recombinant transport factors ) ( Adam et al . , 1990; Lowe et al . , 2010 ) to characterize impβ-mediated nuclear transport . This assay yields nuclei with functional NPCs while allowing us to control the composition and concentrations of transport factors and cargos . A model cargo consisting of a fluorescently labeled tetravalent streptavidin ( SA ) bound to biotinylated impβ binding ( IBB ) domains was used to examine cargo binding at the NPC . The streptavidin-IBB tetramer ( SA-IBB4 ) cargo is large ( ∼218 kDa ) and contains multiple import signals , as do many natural large cargos . In the presence of physiological levels of impβ ( 1 μM , Figure 1—figure supplement 1 ) , SA-IBB4 strongly stains the nuclear envelope but does not efficiently enter the nuclear interior ( Figure 1B , C ) . This indicates that cargo molecules accumulate within the NPC , presumably due to their inability to complete their translocation into the nucleus in the absence of Ran . However , when RanGDP ( 5 μM ) + GTP ( 2 mM ) ( henceforth referred to as RanGTP ) is added , the fluorescence intensity of the nuclear rim drops while that of the nuclear interior increases , showing that the cargos then efficiently exit the NPC and accumulate in the nucleus . To further characterize this RanGTP dependence , we fluorescently labeled impβ with a YFP tag , yielding impβ-YFP , and examined how this transport receptor binds the NPC in the presence or absence of RanGTP . As expected for a FG-binding karyopherin , impβ-YFP formed a bright nuclear rim when added without RanGTP . However , as with the SA-IBB4 cargos , the impβ-YFP signal at the rim was substantially reduced but not completely eliminated by RanGTP ( Figure 1B , C ) . Ran thus modulates the way in which the NPC interacts with both cargo-bound impβ and free impβ . Moreover , since RanGTP reduces the impβ-YFP rim fluorescence , at least a subset of non-cargo engaged but NPC-bound transport receptors must be RanGTP-sensitive . We used fluorescence recovery after photobleaching ( FRAP ) to characterize the turnover kinetics of impβ at the NPC and to examine the binding affinity of impβ for the pore . For the FRAP experiments , impβ-YFP was allowed to form a fluorescent rim at the nuclear envelope and a section of that rim was then photobleached ( the custom hardware is described in Figure 1—figure supplement 2 ) . We subsequently monitored the recovery of rim fluorescence in the photobleached region . In the absence of RanGTP , the initial recovery of the impβ signal after bleaching took several seconds ( time to 20% recovery = 1 . 6 ± 0 . 1 s , N = 20 , Figure 1D , blue ) . However , with RanGTP , the initial recovery was 16-fold more rapid ( time to 20% recovery = 0 . 1 ± 0 . 1 s , N = 20 , Figure 1D , red ) . Therefore , as already indicated by the simple rim fluorescence experiments , RanGTP is able to accelerate the cargo-independent turnover of impβ bound to the NPC . Inspection of the recovery traces hinted at a long-lived population of NPC-bound impβ with little or no turnover . Consistent with this , previous single-molecule titration experiments have suggested the presence of two types of impβ binding sites within the NPC ( Tokunaga et al . , 2008 ) . To directly observe the slow turnover impβ subpopulation , we used a two-color photo-conversion approach ( Figure 1—figure supplement 3 ) . The photo-conversion hardware and geometry was optimized for quantification of subpopulations with extremely slow or no turnover , at the expense of introducing a multi-second dead time ( Figure 1E , arrow ) immediately following photo-conversion . In these photo-conversion experiments , impβ was tagged with the photo-convertible fluorescent protein mEos2 ( impβ-mEos2 ) , initially yielding a green signal at the nuclear envelope . A small portion of the rim was then photo-converted to a red state . As bound ( red ) molecules are replaced by fresh non-converted ( green ) impβ-mEos2 from solution , the red rim signal gradually fades and the green rim signal gradually recovers , revealing the dissociation rate of bound impβ transport receptors . This red-to-green replacement process can be quantified for long times with good signal-to-noise . In the absence of RanGTP , the photo-converted ( red ) impβ-mEos2 signal decayed to half its initial value within 3–4 min ( Figure 1E ) , showing that some impβ molecules remain at the pore for long times . In the presence of RanGTP , the initial red signal was lower than without RanGTP ( ∼50 AU vs ∼75 AU ) , consistent with the previously detected RanGTP-dependent rapid turnover of impβ within the pore . Strikingly , however , some impβ molecules remained at the pore for several minutes , even in the presence of RanGTP ( Figure 1E ) . Summarizing , the FRAP experiments allowed us to quantify fast reactions within the pore and the photo-conversion experiments permitted quantification of slow reactions within the pore . Together , these experiments suggest that there are at least two pools of impβ within the NPC . One pool is stably bound to the NPC for many minutes , with and without RanGTP . The other pool is stably bound to the NPC only in the absence of RanGTP . Having detected two kinetically distinct impβ pools within the pore , we sought to characterize their spatial distribution and identify the nucleoporins they were binding . We were especially interested in the RanGTP-sensitive impβ pool , since RanGTP drives active transport and RanGTP-induced alterations of pore organization could therefore be relevant to active transport . We directly imaged and localized individual Cy5 or Alexa647 dye-labeled impβ molecules within the pore using dSTORM super-resolution localization microscopy ( Heilemann et al . , 2008; van de Linde et al . , 2011 ) ( Figure 2A–D; mean spatial precision , σx , y of 12 nm , Figure 2—figure supplement 1 ) . By directly labeling impβ with a fluorescent reporter , we removed additional localization uncertainty error ( commonly referred to as linkage error ) associated with the antibody labeling methods normally used for super-resolution or electron microscopy . 10 . 7554/eLife . 04052 . 007Figure 2 . Super-resolution imaging of Alexa647-labeled impβ in digitonin-permeabilized HeLa cells . ( A ) Simulated widefield impβ localization at the equatorial plane of the nucleus and ( B ) corresponding dSTORM image . Mean localization precision is 12 nm . ( C and D ) Corresponding widefield and dSTORM images taken at the basal surface of the nucleus showing characteristic punctate NPC structures . ( E ) Zoom of the dSTORM image in ( B ) , showing discrete NPC structures ( examples , blue arrows ) . Putative NPC structures ( green ) markers are automatically identified using the linearized envelope localization histogram ( shown in white , Supplementary methods ) . ( F ) Method for putative NPC structure isolation and alignment . Peaks in the envelope histogram ( black line ) are identified as potential locations for putative NPC structures ( black circles ) . Localizations falling into a window ( width , w , length , l ) centred at these locations ( p ) are cropped out and rotated to a common frame , c , by the angle θ , maintaining the cytoplasm-nucleus vector , n . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 00710 . 7554/eLife . 04052 . 008Figure 2—figure supplement 1 . Localization precision . ( A ) Calibration of EMCCD camera for photon conversion factor . ( B ) Histogram of calculated localization precisions from dSTORM image data . The median value of localization precision ( positional error ) is 12 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 008 The dSTORM images taken at the equatorial plane of the nucleus show discrete elongated structures oriented normal to the nuclear envelope ( Figure 2B , E ) . Viewed from the bottom of the nucleus , we observe radially symmetric , punctate NPC structures ( Figure 2D ) . We did not see a recently reported ( Ma et al . , 2012 ) ‘ring’-like distribution of impβ , although this could be a consequence of our spatial precision . To visualize the axial distribution of impβ , we developed approaches for comparing hundreds to thousands of individual NPCs . Individual NPCs were identified by calculating an ‘envelope histogram’ of the number of localizations in a window normal to the nuclear envelope path ( Figure 2E ) . Well-separated peaks within this histogram , containing a threshold number of localizations , indicate the position of putative NPCs and were selected for further study . The impβ localizations belonging to these NPCs were then extracted , rotated according to the interpolated envelope normal vector ( Figure 2F ) and aligned along the transport axis . Those structures requiring very large alignment shifts , or having poor correlation with the remainder of the data set , were removed . Having extracted , rotated , and aligned the NPCs , we averaged the impβ localizations ( Figure 3A ) . Viewed along the transport axis , there are two pools of impβ localizations separated by approximately 90 nm and occupying a footprint and spatial arrangement consistent with structural studies ( Frenkiel-Krispin et al . , 2010 ) . Antibody labeling of Nup358/RanBP2 ( located on the NPC cytoplasmic filaments ) with a second fluorescent dye was used to confirm that the outermost pool of the impβ signal spatially overlapped with the cytoplasmic face of the NPC ( Figure 3—figure supplement 2 ) . The central channel measured ∼50 nm at the narrowest point , consistent with single quantum dot transport studies ( Lowe et al . , 2010 ) and other super-resolution measurements ( Loschberger et al . , 2012 ) . The addition of RanGTP , which produces a transport-competent pore with active impβ turnover , markedly decreased the total number of impβ localizations ( Figure 3A , ‘RanGTP’ and Figure 3—figure supplements 1 , 3 ) but also changed the shape of the probability density function ( PDF ) of impβ molecules within the pore . In the presence of RanGTP , the PDF is bimodal and shows a depletion of impβ from the nucleoplasmic face of the NPC ( Figure 3C , compare red trace to black trace ) . RanGTP was therefore not simply displacing impβ from the pore but was displacing impβ preferentially from specific sites within the pore . 10 . 7554/eLife . 04052 . 009Figure 3 . Localization microscopy of impβ spatial organisation and its Ran-dependence . ( A ) 2D histograms of impβ density in the NPC under different conditions . To generate these panels , we sum all localizations for a given condition and divide by the number of NPCs per condition . Figure 3—figure supplement 1 shows examples of the raw localization images for each of the conditions . Anti-Nup358 antibody was localized using a second dye pair , as shown in Figure 3—figure supplement 2 . Histograms of the number of localizations per NPC , for each condition are shown in Figure 3—figure supplement 3 . ( B ) Schematic of the NPC showing the tethering locations of Nup358 and Nup153 . ( C ) Probability density functions ( PDF ) of impβ localizations , showing the relative redistribution of impβ localizations in the presence of RanGTP and with Nup153 knockdowns . ( D ) 'Waterfall' plots showing PDF projections of NPC structures in each of the conditions tested . Each column of the plot represents a single NPC structure and is arranged from left to right according to the ratio of cytoplasmic to nuclear localizations . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 00910 . 7554/eLife . 04052 . 010Figure 3—figure supplement 1 . Examples of raw localization data for each of the conditions . ( A ) Total raw localizations for each of the conditions presented in Figure 3 . Note the strongly reduced number of localizations in the Ran+GTP case . Each image is plotted on the same intensity scale . ( B ) 30 randomly selected examples of individual putative NPC structures identified from each of the data sets presented in ( A ) and rotated and aligned to a common frame . The data are presented unnormalised but plotted on a common intensity scale . Again , note the strongly reduced number of localizations in the Ran+GTP and ∆Nup153 case . Also note the increased heterogeneity amongst the structures in the knockdown data . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 01010 . 7554/eLife . 04052 . 011Figure 3—figure supplement 2 . Two color STORM imaging . ( A ) Widefield imaging of STORM labeled abNup358 and impβ . ( B ) Two-color STORM showing abNup358 labeled in blue and impβ labeled in red . The antibody again Nup358 localizes toward the cytoplasmic face of each NPC structure . The image is undersampled due to activation cross-talk problems outlined in the discussion . ( C ) Histogram of average counts for abNup358 and impβ localizations , showing the cytoplasmic bias in position of abNup358 . The difference between the two distributions is found mainly in the NPC channel . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 01110 . 7554/eLife . 04052 . 012Figure 3—figure supplement 3 . Histogram of the number of raw localizations per NPC structure . Measured from the dSTORM data in Figure 3—figure supplement 1 . Each distribution is fitted to a Gaussian distribution ( plotted as a bold line ) , to show the decrease in the number of localizations between the three conditions tested ( Impβ > Impβ , Δ15370% > Impβ+RanGTP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 01210 . 7554/eLife . 04052 . 013Figure 3—figure supplement 4 . Quantification of siRNA knockdown of Nup153 . ( A ) Immunofluorescence labeling of Nup153 in mock transfected and Nup153 siRNA-transfected HeLa cells . ( B ) Nup153 Western blot assay for HeLa cells that were not transfected , mock transfected , or transfected with Nup153 siRNA . β-actin was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 013 Based on the dSTORM data and particle-tracking studies that suggest that the end of the channel is the functional site of Ran action ( Lowe et al . , 2010 ) , we hypothesized that Nup153 might be a site of RanGTP-sensitive impβ binding . Nup153 is an important terminal binding site for the impβ transport pathway ( Shah et al . , 1998; Walther et al . , 2001 ) , can bind as many as seven impβ molecules ( Milles and Lemke , 2014 ) , and interacts with Ran ( Saitoh et al . , 1996; Ball and Ullman , 2005; Schrader et al . , 2008 ) . Since Nup153 is essential for cell viability , we used partial RNAi knockdown to study how alterations of Nup153 levels influence the organization of impβ within the pore . siRNA treatment led to ∼70% protein reduction of Nup153 ( Δ15370% , Figure 3—figure supplement 4 ) . Indeed , when Nup153 was reduced by siRNA knockdown , the dSTORM signal changed . The impβ map shows fewer impβ localizations overall and marked reduction of signal from the entire nuclear side of the NPC , creating an asymmetric , teardrop-like pattern ( Figure 3A , ‘impβ Δ15370%’ ) . Therefore , both addition of RanGTP and reduction of Nup153 alter the arrangement and loading of impβ within the pore , especially towards the NPC's nuclear face . Although population averaging allows major differences to be detected , such averaging can obscure more subtle changes . To compare many NPCs without population averaging , we represented each NPC as a single vertical line whose color corresponds to impβ concentration , ranging from blue to red . We placed each of those lines side-by-side , giving a ‘waterfall’ plot ( Figure 3E ) . The pores with most of the impβ at the nuclear face are at the left of the plots , while the pores with most of the impβ at the cytoplasmic face are to the right . As can be seen , there is considerable pore-to-pore heterogeneity in the axial distribution of impβ in all studied conditions . Some NPCs had a strong impβ signal only at the cytoplasmic face , other NPCs had similar levels of impβ at both faces , and finally , some NPCs had most of their impβ at their nuclear face . The most notable difference was that when Nup153 was reduced , there were very few pores with a dominant impβ signal at the nuclear face . Summarizing , in the absence of RanGTP , the pore is loaded with impβ . The addition of RanGTP increases impβ turnover , depleting transport receptors from the pore . Partial knockdown of Nup153 reduces the overall impβ counts and depletes transport receptors from the nuclear side of the pore , resulting in an asymmetric , cytoplasm-biased impβ distribution . Together , these results raise the possibility that Ran can modulate the interactions between impβ and the NPC . We then turned to a quantitative assessment of impβ levels within the pore , to determine whether Nup153 is a dominant site of RanGTP-sensitive impβ binding . Although dSTORM microscopy is able to localize populations of molecules and detect relative changes in their spatial arrangements , it does not allow absolute numbers of molecules to be estimated . We thus used a single-molecule photobleach step-counting assay ( Leake et al . , 2006 ) to estimate the numbers of impβ molecules displaced by nuclear RanGTP at the NPC . Digitonin-permeabilized nuclei from wild type and Δ15370% cells were incubated with impβ-mCherry with or without RanGTP and then fixed , yielding nuclear pores that can be imaged at the basal surface of the nucleus as bright spots ( Figure 4A ) . Under the appropriate imaging conditions , discrete single-molecule photobleaching steps can be resolved in the fluorescence bleaching traces of impβ-mCherry at the pores ( Figure 4B ) . The photobleaching fluorescence step-size , x , for a specific NPC , can be calculated by taking the first peak of the power spectrum of the pairwise difference distribution of the bleaching trace ( Figure 4C ) . From the fluorescence step-size and initial intensity , ΔI , of the pore , the relative amount of impβ molecules at a single NPC can be measured . Because of potential systematic errors in determining absolute numbers of impβ molecules with this technique ( e . g . , homo-FRET , incomplete mCherry maturation ) , relative analysis of impβ levels was performed by defining the impβ signal of the wild type–RanGTP condition as 100% ( Figure 4D ) . Wild type NPCs in the absence of RanGTP contained the greatest number of impβ molecules ( 70 bleach steps counted ) , whereas RanGTP caused a 27% decrease in impβ levels ( Figure 4A , Figure 4—figure supplement 1 , Table 1 ) . In Δ15370% nuclei , we found a 34% drop in the amount of impβ per pore without RanGTP and a 33% drop with RanGTP . These results suggest that most of the impβ molecules that are displaced from the NPC by nuclear RanGTP are those that are bound to Nup153 . 10 . 7554/eLife . 04052 . 014Figure 4 . Photobleach step-counting of impβ at the NPC . ( A ) A 200-frame average image of impβ-mCherry at the basal envelope of the nucleus . Individual NPCs can be identified ( example highlighted in red ) . ( B ) Fluorescence intensity vs time trace for the NPC highlighted in ( A ) under continuous illumination . The raw intensity signal is shown in gray and the Cheung-Kennedy filtered signal is shown in red . Inset: individual photobleaching steps ( x ) can be clearly identified . ( C ) Pairwise difference distribution function calculated from the intensity trace shown in ( B ) . Characteristic step sizes can be identified from the peaks in the distribution . Inset: the power spectrum of the pairwise difference distribution , showing the characteristic intensity signal for a single mCherry , x . We can then calculate the number of molecules by dividing the total intensity change in the trace in ( B ) by the fluorescence intensity of a single molecule calculated from ( C ) . ( D ) Impβ counts as a function of RanGTP and Nup153 . Addition of RanGTP and reduction of Nup153 decrease the impβ counts by approximately equal amounts . Normalized impβ abundance , number of pores analyzed , mean , and standard deviations are denoted . Error bars represent standard deviations about the means . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 01410 . 7554/eLife . 04052 . 015Figure 4—figure supplement 1 . Distribution of count values . Histogram of the count values for four conditions ( wild type or Δ15370% cells with and without RanGTP ) with a bin size of five molecules . Numbers of pores analyzed , means , and standard deviations are given . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 01510 . 7554/eLife . 04052 . 016Table 1 . Mann–Whitney confidence intervals of impβ changes seen in ±RanGTP and ±Δ15370% conditionsDOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 016WT − RanGTPWT + RanGTPΔ15370% − RanGTPΔ15370% + RanGTPWT − RanGTP–[17 . 8 , 18 . 9][22 . 5 , 23 . 7][21 . 1 , 22 . 2]WT + RanGTP[−17 . 8 , −18 . 9]–[4 . 5 , 5 . 4][2 . 9 , 3 . 8]Δ15370% − RanGTP[−22 . 5 , −23 . 7][−4 . 5 , −5 . 4]–[−0 . 9 , −1 . 8]Δ15370% + RanGTP[−21 . 1 , −22 . 2][−2 . 9 , −3 . 8][0 . 9 , 1 . 8]–For example , the addition of RanGTP compared to baseline results in a drop of about 18 impβ counts ( column 1 , row 2 ) ; The Mann–Whitney confidence interval is [−17 . 8 , −18 . 9] . The detection of a stable RanGTP-sensitive pool of impβ in the pore , the tentative identification of a binding partner , and the quantification of the energy-dependent changes within the pore motivated functional studies seeking to detect possible impβ/RanGTP/Nup153-mediated alterations of passive facilitated diffusion and active transport . We first investigated Nup153's relevance to impβ-mediated transport using the SA-IBB4 cargo . The cargo was added to digitonin-permeabilized nuclei that either contained impβ only ( to monitor passive equilibration of the cargo across the nuclear envelope ) or impβ , Ran , NTF2 ( the RanGDP importer ) , and GTP ( to monitor active transport ) . Impβ was added to the nuclei before the cargo , allowing us to examine how cargo molecules translocate through pores that already contain transport receptors . As shown earlier ( Figure 1 ) , little cargo was able to enter the nucleus under conditions of passive equilibration ( i . e . , in the absence of RanGTP ) in wild-type cells . In contrast , the cargo translocated the NPC faster in Δ15370% nuclei , indicating that the transport channel had become leakier to large cargos and translocation became less dependent on the presence of RanGTP ( Figure 5A , B ) . Interestingly , the opposite was observed for active transport in the presence of RanGTP , where net nuclear cargo accumulation was reduced for Δ15370% nuclei ( Figure 5A , B ) . The nucleoporin Nup153 therefore affects both the ease of passive impβ-mediated movement of large cargos through the pore and the efficiency of active transport into the nucleus . 10 . 7554/eLife . 04052 . 017Figure 5 . Effect of Nup153 reduction on active transport and passive equilibration . ( A ) Confocal fluorescence microscopy images showing the change in distribution of a fluorescently labeled 220 kDa SA-IBB4 cargo as a function of Nup153 knockdown . ( B ) Reduction of Nup153 enables the cargo to passively equilibrate ( pale blue region ) more rapidly through NPCs loaded with impβ . However , reduction of Nup153 impairs the active transport of SA-IBB4 ( pale yellow region ) . ( C ) Passive equilibration of the inert GFP1 and GFP3 probes as a function of impβ ( 1 μM ) and RanGTP ( 5 μM ) . For GFP1 , note the rate decrease by impβ and the rate increase with RanGTP . For GFP3 , passive equilibration is slow in all conditions . ( D ) Passive equilibration of GFP2 as a function of impβ , RanGTP , and Nup153 . For wild-type cells , impβ and RanGTP have similar effects on GFP2 as on GFP1 . For Δ15370% cells , impβ no longer slows passive diffusion of GFP2 . RanGTP , however , still facilitates equilibration . ( E ) TRN1 does not significantly slow GFP2 diffusion through the NPC at 1 or even 2 μM . In all plots , shaded regions indicate the standard error of the mean ( N ≥ 3 for all conditions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 01710 . 7554/eLife . 04052 . 018Figure 5—figure supplement 1 . RanGTP and a Ran ‘wash’ increase the passive equilibration of GFP2 into wild type nuclei even when no impβ is present . In the GFP2 passive equilibration assays , it was intriguing that RanGTP made the pore more permeable than when impβ was excluded . We hypothesized that perhaps there was a significant amount of endogenous impβ , as well as other transport receptors , still residing at the pore that were not removed even after several wash steps during the digitonin permeabilization procedure . To test this , we performed an extra RanGTP wash step , in which Ran , GTP , and an energy-regeneration system were incubated with the nuclei for 10 min before being washed away . During this step , RanGTP will bind to and dissociate any endogenous transport receptors still bound to the NPC which are then subsequently washed away . In comparison to the condition in which impβ and RanGTP are not added to the nuclei , the RanGTP wash did indeed make the NPCs more permeable ( red curve ) , although they were still not as permeable as when impβ and RanGTP are added together or when RanGTP is added alone ( green and purple curves ) . This trend suggests that a subpopulation of endogenous transport receptors still reside at the NPC even after digitonin permeabilization . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 018 To further characterize the permeability barrier within the NPC , we determined whether impβ and RanGTP affect the free diffusion of cargos through the pore under both normal and reduced levels of Nup153 . We employed a series of inert probes consisting of single GFPs , GFP dimers , and GFP trimers ( GFP1 , GFP2 , and GFP3 ) with molecular masses of 27 , 54 , and 83 kDa respectively . Because these probes do not contain an IBB and cannot bind impβ , they exclusively undergo passive transport . We found that 1 μM impβ significantly decreased the permeability of the NPC for the GFP1 and GFP2 probes ( Figure 5C , D ) . By contrast , when RanGTP was also added , the permeability was greatly increased . The GFP3 probe translocated across the NPC at a relatively slow rate with or without impβ and RanGTP , likely because GFP3's size is considerably larger than the passive diffusion size cutoff of the pore ( Figure 5C ) . We therefore decided to focus on the GFP2 probe and we used it to further explore the effects of impβ and Nup153 on the permeability of the NPC ( Figure 5D , E ) . First , we tested the possibility that general molecular crowding , for instance due to widespread transport receptor-FG nucleoporin interactions , was responsible for the observed permeability modulation by impβ . We therefore repeated the previous experiments with the related transport receptor transportin-1 ( TRN1 ) . TRN1 is the transport receptor for M9 signal peptide-containing cargos such as hnRNPs and belongs to the same class of karyopherins as impβ ( Pollard et al . , 1996 ) . At 1 and even 2 μM TRN1 , there was no strong effect on NPC permeability ( Figure 5E ) , suggesting that general molecular crowding is not responsible for the changes to NPC permeability . Importantly , for Δ15370% nuclei , addition of impβ no longer restricted the NPC ( Figure 5D ) . Together , these results suggest that specifically the impβ•Nup153 interaction causes the nuclear pore to become less permeable . Furthermore , because the inert probes undergo purely passive translocation across the NPC , the reduced permeability must be due to a specific steric ‘barrier’ within the pore and not due to a block of transport receptor-specific binding sites . This steric barrier appears to involve impβ•Nup153 interactions that are very stable and long-lived in the absence of RanGTP . Indeed RanGTP causes the pore to become more permeable even when no exogenous impβ is first added . This is likely due to endogenous impβ and other transport receptors residing in the pore that were not washed away during digitonin permeabilization ( Figure 5—figure supplement 1 ) . Along with the observation that impβ can persist in the pore for minutes or tens of minutes , these functional studies suggest that impβ could be considered a bona fide functional component of the pore and not only a soluble transport receptor . Moreover , the impβ•Nup153 interaction may be responsible for the permeability differences detected in our inert probe passive diffusion assays and may contribute to the permeability barrier function of the NPC . To explore the notion that impβ and Nup153 act together to form a Ran-sensitive permeability barrier , we investigated their interaction in vitro . Upon co-incubation of recombinant impβ and Nup153FG ( the FG domain of Nup153 comprising amino acids 874–1475 [Lim et al . , 2006] ) , large , micron-sized structures formed on a timescale of minutes ( Figure 6A ) . We turned to fluorescence fluctuation spectroscopy ( Chen et al . , 2000; Tetin , 2013 ) to examine the structure's assembly and disassembly behaviors and requirements . The fluorescence intensity signal of diffusing impβ-YFP molecules ( 50 nM ) showed a fluctuation pattern characteristic of freely diffusing proteins ( Figure 6B ) . However , when Nup153FG ( 0 . 5 µM ) was added , large intensity bursts appeared within tens of seconds . The appearance of these spikes in intensity ( along with their corresponding long tails in the photon counting histograms ) indicated the formation of large impβ•Nup153 complexes ( Figure 6B , C , red traces ) . These higher-order complexes were orders of magnitude brighter than the freely diffusing impβ-YFP , suggesting that they are comprised of tens or even hundreds of impβ molecules . The formation of large complexes can be explained by the many FG motifs found in Nup153's FG domain as well as the multiple sites on impβ's surface that may bind FG repeats . Notably , the addition of RanQ69L•GTP ( 2 µM ) , which does not hydrolyze GTP ( Bischoff et al . , 1994 ) and is therefore stably in the GTP-bound form , entirely inhibited formation of the complexes . RanQ69L•GTP even dissolved existing large impβ•Nup153FG complexes ( Figure 6—figure supplement 1A ) . This Ran action occurred specifically through impβ ( and not Nup153FG ) binding since Nup153FG in complex with an impβ truncation lacking the Ran-binding domain , impβ ( ΔN70 ) , became insensitive to RanQ69L•GTP ( Figure 6—figure supplement 1B ) . 10 . 7554/eLife . 04052 . 019Figure 6 . In vitro formation of large RanGTP-reversible impβ•Nup153FG complexes . ( A ) Confocal images of impβ•Nup153FG complexes . Brightfield and YFP fluorescence images of Nup153FG ( left ) , impβ-YFP ( center ) , and Nup153FG + impβ-YFP ( right ) . Complexes form only when both proteins are co-incubated . ( B ) Fluorescence fluctuation intensity traces of impβ-YFP in the presence of Nup153FG and RanQ69L•GTP . Large intensity bursts appear with the addition of Nup153FG but are inhibited in the presence of RanQ69L•GTP . ( C ) Photon counting histograms for the experiments shown in ( B ) . Impβ and Nup153FG form large , bright complexes ( note the extended tail with Nup153FG ) ; however , these complexes disappear when RanQ69L•GTP is added ( see inset schematic ) . ( D ) Photon counting histograms of fluorescence fluctuations for TRN1-GFP in the presence of Nup153FG and RanQ69L•GTP . Large complexes also form between TRN1 and Nup153FG but are not affected by RanQ69L•GTP ( see inset schematic ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 01910 . 7554/eLife . 04052 . 020Figure 6—figure supplement 1 . Additional fluctuation traces . ( A ) RanQ69L•GTP can dissolve existing aggregates . To determine whether RanGTP can dissolve aggregates that have already formed and not just prevent aggregate formation , impβ-YFP•Nup153 aggregates were allowed to form for 5 min at room temperature and their fluorescence fluctuations were measured ( blue traces ) . RanQ69L•GTP was then added to the reaction for 30 min and the fluctuations were measured again ( green traces ) . The disappearance of the majority of large fluorescent spikes indicate that RanQ69L•GTP can dissolve existing aggregates . ( B ) Test of Ran-reversibility of aggregates formed by an impβ truncation unable to bind Ran . To test the hypothesis that RanQ69L•GTP prevents aggregate formation by displacing impβ that is bound to Nup153FG , we characterized the aggregation capacity of an impβ truncation missing the first 70 N-terminal amino acids , impβ ( ΔN70 ) , corresponding to impβ's Ran-binding domain . Impβ ( ΔN70 ) -YFP did not form aggregates on its own ( blue trace ) but did so when mixed with Nup153FG ( green trace ) . Interestingly , RanQ69L•GTP was now unable to abolish aggregate formation ( red trace ) , indicating that RanGTP dissolves the aggregates through impβ binding and not Nup153FG binding . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 020 Performing similar experiments with TRN1-GFP ( 100 nM ) , we again detected spikes in intensity upon addition of Nup153FG , indicating the formation of large complexes ( Figure 6D ) . This is not surprising given that TRN1 is structurally similar to impβ and likely also contains multiple FG-binding sites ( Chook and Blobel , 1999 ) . However , although TRN1 and RanGTP are known binding partners ( Chook and Blobel , 1999 ) , RanQ69L•GTP had no observable effect on the TRN1•Nup153FG complexes , suggesting a functional difference between TRN1 and impβ . Indeed , it has been previously reported that TRN1-mediated nuclear import is less dependent upon the RanGTP gradient than impβ-mediated import ( Ribbeck et al . , 1999 ) . Moreover , the passive diffusion assays ( Figure 5E ) did not detect alterations of pore permeability in the presence of TRN1 , suggesting that TRN1 either does not form a meshwork inside the pore or that a hypothetical TRN1-mediated barrier has significantly different biophysical characteristics ( e . g . , effective pore size ) compared to the one formed by impβ . In support of both those possibilities , TRN1 has been shown to bind different sites within the Nup153 FG domain relative to impβ ( Shah and Forbes , 1998 ) . Therefore , despite the fact that impβ and TRN1 are both transport receptors and that both can bind Nup153 and form higher-order complexes with it , previous reports ( Shah and Forbes , 1998; Ribbeck et al . , 1999 ) and the data shown in Figures 5E , 6D point to TRN1 being functionally and biophysically distinct from impβ .
The importance of members of the impβ family of transport receptors and Ran in active nuclear transport has been firmly established for many years . The basic import reaction involves the RanGTP-driven displacement of impβ family members from their cargo in the correct compartment . However , there are multiple reports that RanGTP and impβ may have critical additional roles in passive equilibration and active transport . Specifically , it was proposed that Ran is necessary for impβ-bearing cargos to move past a barrier located at 70 nm along the transport axis ( Gorlich et al . , 1996; Lowe et al . , 2010 ) . Based on the experiments reported here , we propose the existence of a Ran-sensitive network of interactions between impβ and Nup153 centered at the nuclear face and including the central channel of the NPC , which contributes to the permeability of the pore ( Figure 7 ) . Whilst other nuclear basket localized Nups , such as Nup50 , have been shown to promote cargo dissociation from the pore in active transport ( Sun et al . , 2008 ) , Nup153 appears to also have a role in controlling bulk permeability of the NPC . The impβ•Nup153 interaction significantly restricts the ability of inert cargos to diffuse across the NPC , indicating the presence of a non-specific physical barrier that cannot be explained by simple molecular crowding . This barrier may take the form of a highly cross-linked ‘meshwork’ of long , flexible Nup153 FG domains fastened to each other by impβ molecules , which we characterized in vitro at physiological pH and salt concentrations . The multiply cross-linked nature of these impβ•Nup153 structures may be reminiscent of the FG gel materials reported by others ( Frey et al . , 2006; Schmidt and Gorlich , 2015 ) . However , these FG gels are held together by homotypic interactions between the FG domains; here , materials form via specific coordination between impβ and the FG domains . Moreover , the resulting impβ•Nup153 material is dynamic , in the sense that its formation and final stability is sensitive to RanGTP , which can even dissolve existing large impβ•Nup153 complexes ( Figure 6—figure supplement 1 ) . The in vitro fluorescence fluctuation data obtained through spectroscopic studies of purified proteins correlate well with our localization microscopy studies of impβ's spatial distribution within the pore , where we see a sub-population of impβ in the channel that is significantly reorganized by RanGTP . Furthermore , the photobleach counting experiments with wild type and reduced Nup153 NPCs suggest that this Ran-sensitive pool is predominantly located at Nup153 , although the counting experiments do not rule out other RanGTP-sensitive impβ binding sites within the pore . The sub-second turnover kinetics of the RanGTP-sensitive impβ pool ( Figure 1 ) are similar to kinetic values for impβ turnover inside living cells ( Rabut et al . , 2004 ) , suggesting that our reconstituted ‘in vitro’ permeabilized cell transport assay recapitulates key features of transport in intact living cells . 10 . 7554/eLife . 04052 . 021Figure 7 . Model of Ran-sensitive impβ•Nup153 interactions at the nuclear face of the NPC . In this model , multivalent interaction of impβ with Nup153 yields a cross-linked mesh that restricts the movement of inert molecules and cargo-receptor complexes . This impβ•Nup153 barrier is modulated by Ran . DOI: http://dx . doi . org/10 . 7554/eLife . 04052 . 021 Two quantitative imaging approaches , a two-color photo-conversion approach and single-protein counting , were used to investigate an extremely stable subpopulation of impβ which cannot be easily detected by other methods due to the reaction timescale , photo-bleaching effects , and limitations of instrument stability . We estimate that on average , 73 ± 16% of impβ molecules in each pore are insensitive to nuclear RanGTP and bound stably to the NPC for many minutes . These results , coupled with our observation that impβ contributes to the NPC's permeability , suggest that impβ is a functional component of the pore and does not just facilitate cargo translocation . Indeed , other theoretical and experimental studies have suggested that transport receptor binding at the NPC plays a critical role in the non-specific occlusion of inert cargos from entering the pore ( Zilman et al . , 2007; Jovanovic-Talisman et al . , 2009; Zilman and Bel , 2010; Schleicher et al . , 2014 , Kapinos et al . , 2014 ) . A functional role of impβ within the NPC is particularly intriguing in light of the structural and functional relationship between karyopherins and scaffold nucleoporins ( Andersen et al . , 2013; Sampathkumar et al . , 2013; Stuwe et al . , 2014 ) , suggesting that these two classes of proteins may share a common evolutionary ancestry . The possible functional and structural roles of the Ran-insensitive impβ pool remain to be discovered . At present , we only know ( 1 ) that the bulk of these Ran-insensitive impβ molecules are located near the cytoplasmic face of the pore and ( 2 ) that they are not bound to Nup153 , since reduction of Nup153 ( i . e . , Δ15370% ) did not reduce the impβ counts relative to the RanGTP condition . Although we have emphasized the ‘average’ characteristics of the NPC vis-à-vis changes in RanGTP levels and other experimental manipulations , the single-pore resolution dSTORM and counting experiments revealed significant NPC-to-NPC variation . The distributions for our counting experiments give an idea of the heterogeneity of impβ binding amongst NPCs even within one nuclear envelope . The variability of NPC composition may reflect variation of the instantaneous functional state of the cellular pool of several hundred NPCs; perhaps , not all NPCs are functionally equivalent at all times . In the future , it will be interesting to directly relate NPC-to-NPC variation of molecular composition to possible variation of functional transport characteristics . The observation that the permeability of the NPC , specifically its size-filtering , is sensitive to RanGTP levels is interesting in light of recent results that suggest that cells might actively regulate the RanGTP gradient . The RanGTP gradient is generated and maintained by the chromatin-associated RCC1 exchange factor ( Bischoff and Ponstingl , 1991 ) whose activity as well as the local concentration of its RanGDP substrate is subject to multi-tiered regulation ( Li and Zheng , 2004; Hood and Clarke , 2007; Yoon et al . , 2008; Hitakomate et al . , 2010 ) . Overall , our studies raise the possibility that the cell might have an extra layer of control over nucleocytoplasmic transport processes by regulating the spatiotemporal characteristics of the RanGTP gradient , which would then modulate both the size-cutoff of passive permeability and the extent of active transport . Beyond clarifying the RanGTP-dependent composition and organization of the intact pore , the ability to form the Ran-reversible impβ/Nup153 material in vitro will allow the interplay of energy , impβ , and Ran to be directly investigated and should also allow efficient and selective molecular rectifiers to be created in vitro , not just for biological cargos but also for other substrates .
The stepwise-photobleaching method was adapted from the approach of Leake et al . ( 2006 ) . The stepwise-photobleaching method relies on the irreversible and stochastic bleaching of fluorescent proteins upon repeated exposure . The sample is illuminated with an excitation light intensity low enough to slowly bleach it until background emission is reached . Plotting the intensity of a spot of interest over time results in an exponential decay function . Ideally , this function contains discernible steps . Each step corresponds to a bleaching event of a single molecule . When the number of molecules to count increases , the chance of having several bleaching steps at the same time also increases , resulting in steps having sizes that are multiples of a single bleach step . We added mCherry tagged impβ to permeabilized cells , fixed them , and imaged the basal envelope of the nuclei . The angle of the beam was chosen in such a way to minimize the background ( normally just slightly higher than the optimal TIRF angle ) . The desired focal plane was found using the lowest laser power possible ( ∼150 μW ) to avoid bleaching during focusing . As soon as the correct focal plane was found the laser power was increased to about 3 mW and movie recording was started . Movies were acquired at 50 Hz and normally have a length of about 180 s ( or until background emission was reached ) . Four different conditions were tested: WT +Ran+GTP , WT −Ran−GTP , R70%Nup153 +Ran+GTP , R70%Nup153 −Ran−GTP . | In our cells , genetic material is contained within the nucleus , which is separated from the rest of the cell by a double-layered membrane called the nuclear envelope . Within this membrane there are pores that allow proteins and other molecules to enter and exit the nucleus . Small molecules can pass through these pores unaided , which is known as ‘passive’ transport . However , larger cargos need help from transport receptor proteins in a process called ‘active’ transport . Large cargos bind to transport receptors , such as importin-β , in the cytoplasm and are then guided through the pore . Once the cargo and importin-β are inside the nucleus , a protein called RanGTP binds to importin-β to release the cargo . It is thought that importin-β and RanGTP are only important for the active transport of cargo . Here , Lowe et al . studied how importin-β interacts with the pore . The experiments show that in the absence of RanGTP , importin-β accumulates inside the pore and binds to a protein called Nup153 , which is part of the complex of proteins that makes up the pore . However , when RanGTP is present , some of the importin-β is displaced from Nup153 and leaves the pore , which makes it easier for cargo to pass through . Further experiments show that when Nup153 and importin-β are mixed , they associate into a gel-like material that can be ‘melted’ by RanGTP . Lowe et al . propose a model for how RanGTP may control the flow of cargo through the nuclear pore by affecting the binding of importin-β to Nup153 . Lowe et al . 's findings suggest that passive and active transport of cargo across the nuclear pore are fundamentally connected and suggest that RanGTP provides the cell with an additional layer of control over nucleocytoplasmic transport . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Importin-β modulates the permeability of the nuclear pore complex in a Ran-dependent manner |
During infection chlamydial pathogens form an intracellular membrane-bound replicative niche termed the inclusion , which is enriched with bacterial transmembrane proteins called Incs . Incs bind and manipulate host cell proteins to promote inclusion expansion and provide camouflage against innate immune responses . Sorting nexin ( SNX ) proteins that normally function in endosomal membrane trafficking are a major class of inclusion-associated host proteins , and are recruited by IncE/CT116 . Crystal structures of the SNX5 phox-homology ( PX ) domain in complex with IncE define the precise molecular basis for these interactions . The binding site is unique to SNX5 and related family members SNX6 and SNX32 . Intriguingly the site is also conserved in SNX5 homologues throughout evolution , suggesting that IncE captures SNX5-related proteins by mimicking a native host protein interaction . These findings thus provide the first mechanistic insights both into how chlamydial Incs hijack host proteins , and how SNX5-related PX domains function as scaffolds in protein complex assembly .
To counter host defence mechanisms intracellular bacterial pathogens have evolved numerous strategies to evade immune detection , replicate and cause infection . Many pathogens manipulate endocytic pathways to gain entry into host cells and generate a membrane-enclosed replicative niche . This frequently involves hijacking or inhibiting the host cell trafficking machinery , first to generate the pathogen containing vacuole ( PCV ) and subsequently to prevent fusion with lysosomal degradative compartments . Concomitantly the pathogen often endeavors to decorate the PCV with host proteins and lipids that mimic other host cell organelles in order to circumvent innate immune detection , expand the replicative niche and acquire nutrients to support intracellular replication ( Di Russo Case and Samuel , 2016; Personnic et al . , 2016 ) . This process is often orchestrated through the action of molecular syringe-like secretion systems that deliver bacterial effector proteins directly into the host cell cytoplasm . Chlamydia trachomatis is arguably one of the most successful human bacterial pathogens by virtue of its capacity to hijack host cell intracellular trafficking and lipid transport pathways to promote infection ( Bastidas et al . , 2013; Derré , 2015; Elwell et al . , 2016; Moore and Ouellette , 2014 ) . C . trachomatis causes nearly 100 million sexually transmitted infections annually worldwide , and if left unchecked leads to various human diseases including infection-induced blindness , pelvic inflammatory disease , infertility and ectopic pregnancy ( Aral et al . , 2006; Newman et al . , 2015 ) . Even though chlamydial infections can generally be treated with antibiotics , persistent infections remain a challenge ( Kohlhoff and Hammerschlag , 2015; Mpiga and Ravaoarinoro , 2006 ) . All Chlamydiae share a common dimorphic life cycle , where the bacteria alternates between the infectious but non-dividing elementary body ( EB ) form , and the non-infectious but replicative reticulate body ( RB ) form . Following internalization of EBs through a poorly defined endocytic process , the bacteria reside in a membrane-bound vacuole termed the inclusion , where they convert into RBs and replication occurs over 24–72 hr . RBs eventually redifferentiate back to EBs in an asynchronous manner , and are then released to infect neighboring cells ( Di Russo Case and Samuel , 2016; Hybiske , 2015; Ward , 1983 ) . The encapsulating inclusion membrane provides the primary interface between the bacteria and the host cell’s cytoplasm and organelles . From the initial stages of invasion until eventual bacterial egress the chlamydial inclusion is extensively modified by insertion of numerous Type-III secreted bacterial effector proteins called inclusion membrane proteins or ‘Incs’ . The Incs modulate host trafficking and signaling pathways to promote bacterial survival at different stages , including cell invasion , inclusion membrane remodeling , avoidance of the host cell innate immune defense system , nutrient acquisition and interactions with other host cell organelles ( Elwell et al . , 2016; Moore and Ouellette , 2014; Rockey et al . , 2002 ) . Chlamydiae secrete more than fifty different Inc proteins . While Incs possess little sequence similarity , they share a common membrane topology with cytoplasmic N- and C-terminal domains , separated by two closely spaced transmembrane regions with a short intra-vacuolar loop ( Dehoux et al . , 2011; Kostriukova et al . , 2008; Li et al . , 2008; Lutter et al . , 2012; Rockey et al . , 2002 ) ( Figure 1A ) . The cytoplasmic N- and C-terminal sequences of the Inc proteins act to bind and manipulate host cell proteins . Reported examples include the binding of the small GTPase Rab4A by CT229 ( Rzomp et al . , 2006 ) , Rab11A by Cpn0585 ( Cortes et al . , 2007 ) , SNARE proteins by IncA ( Delevoye et al . , 2008 ) , centrosomal and cytoskeletal proteins by Inc850 and inclusion protein acting on microtubules ( IPAM ) ( Dumoux et al . , 2015; Mital et al . , 2015 , 2010 ) , myosin phosphatase by CT228 ( Lutter et al . , 2013 ) , 14-3-3 and Arf family proteins by IncG and InaC ( Kokes et al . , 2015; Scidmore and Hackstadt , 2001 ) , and the lipid transfer protein CERT by IncD ( Derré et al . , 2011; Elwell et al . , 2011 ) . Despite these reports , there are no known structures of Inc family members either alone or in complex with host effectors . 10 . 7554/eLife . 22311 . 003Figure 1 . SNX5 , SNX6 and SNX32 tare recruited to C . trachomatis inclusions . ( A ) HeLa cells stably expressing the mCherry-Rab25 inclusion membrane marker ( red ) were infected with C . trachomatis serovar L2 ( 24 hr ) and transfected with myc-tagged SNX expression constructs . The samples were fixed and immunolabeled with anti-myc ( green ) and anti-chlamydial HtrA antibodies ( white ) and counterstained with DAPI ( blue ) . Similar experiments using GFP-tagged proteins are shown in Figure 1—figure supplement 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 00310 . 7554/eLife . 22311 . 004Figure 1—figure supplement 1 . SNX5 , SNX32 and SNX1 are recruited to C . trachomatis inclusions and membrane tubules . ( A ) Hela cells were transiently transfected with GFP-tagged SNX and mCherry-Rab25 proteins as indicated , and infected with C . trachomatis serovar L2 . Cells were imaged by confocal fluorescence microscopy for GFP-tagged proteins ( green ) , endogenous SNX1 ( blue ) , mCherry-Rab25 ( red ) and DAPI-stained nuclear material ( white ) . Both GFP-SNX5 and GFP-SNX32 are recruited to inclusion membranes , but the distantly related SNX-BAR protein SNX8 is not . The images are maximum projections . ( B ) An example of SNX1-decorated tubules ( green ) often observed emanating from inclusion membranes ( mCherry-Rab25 in red; DAPI staining in blue ) . The image is a maximum projection . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 00410 . 7554/eLife . 22311 . 005Figure 1—figure supplement 2 . Recruitment of SNX1 , SNX2 and SNX5 to inclusions is not dependent on 3-phosphoinositides . HeLa cells stably expressing mCherry-Rab25 were infected with C . trachomatis serovar L2 ( MOI ~0 . 5 ) for 24 hr and imaged by immunofluorescence microscopy using antibodies to SNX1 , SNX2 and SNX5 . mCherry-Rab25 provides marker for the inclusion membrane . The upper panels show control infections and lower panels show cells treated with wortmannin or Vps34-IN1 with concomitant loss of SNX association with endosomal compartments while inclusion localisation is unaffected . The images are maximum projections . Endosomal compartments are labeled with antibodies to endogenous Rab5 or Vps35 and Pearson’s correlation coefficients used to quantify loss of endosomal recruitment ( 100 cells per group; error bars , S . D ) . A movie showing the effect of wortmannin on GFP-SNX5 is shown in Video 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 005 Two recent studies have greatly expanded the repertoire of host cell proteins known to associate with chlamydial inclusions and Inc proteins ( Aeberhard et al . , 2015; Mirrashidi et al . , 2015 ) . Both reports confirmed that membrane trafficking proteins are major components of the inclusion proteome; and in particular members of the endosomal sorting nexin ( SNX ) family are highly enriched . Specifically it was shown that the C . trachomatis IncE/CT116 protein could recruit SNX proteins containing bin-amphiphysin-Rvs ( BAR ) domains SNX1 , SNX2 , SNX5 and SNX6 ( Mirrashidi et al . , 2015 ) . SNX1 and SNX2 are highly homologous and form heterodimeric assemblies with either SNX5 or SNX6 to promote endosomal membrane tubulation and trafficking ( van Weering et al . , 2012 ) . A fifth protein SNX32 is highly similar to SNX5 and SNX6 but is almost exclusively expressed in the brain and has not yet been characterized . SNX recruitment to the inclusion occurs via the C-terminal region of IncE interacting with the phox-homology ( PX ) domains of SNX5 or SNX6 ( Mirrashidi et al . , 2015 ) ( Figure 1A ) . Interestingly , RNAi-mediated depletion of SNX5/SNX6 does not slow infection but rather increases the production of infectious C . trachomatis progeny suggesting that the SNX recruitment is not done to enable bacterial infection . Instead it was proposed that because SNX proteins regulate endocytic and lysosomal degradation , the manipulation by IncE could be an attempt to circumvent SNX-enhanced bacterial destruction ( Aeberhard et al . , 2015; Mirrashidi et al . , 2015 ) . Here we use X-ray crystallographic structure determination to define the molecular mechanism of SNX5-IncE interaction , and confirm this mechanism using mutagenesis both in vitro and in cells . When bound to SNX5 , IncE adopts an elongated β-hairpin structure , with key hydrophobic residues docked into a complementary binding groove encompassing a helix-turn-helix structural extension that is unique to SNX5 , SNX6 , and the brain-specific homologue SNX32 . A striking degree of evolutionary conservation in the IncE-binding groove suggests that IncE co-opts the SNX5-related molecules by displacing a host protein ( as yet unidentified ) that normally binds to this site . Our work thus provides both the first mechanistic insights into how protein hijacking is mediated by inclusion membrane proteins , and also sheds light on the functional role of the SNX5-related PX domains as scaffolds for protein complex assembly .
It was previously shown that the sorting nexins SNX1 , SNX2 , SNX5 and SNX6 are recruited to the inclusion membrane in C . trachomatis infected cells ( Aeberhard et al . , 2015; Mirrashidi et al . , 2015 ) . We first confirmed this for myc-tagged SNX1 , SNX2 and SNX5 in HeLa cells infected with C . trachomatis serovar L2 ( MOI ~0 . 5 ) for 18 hr . All three proteins were recruited to the inclusion membrane as assessed by co-localisation with the inclusion marker mCherry-Rab25 ( Figure 1B ) ( Teo et al . , 2016 ) , as were GFP-tagged SNX1 and SNX5 but not the more distantly homologous GFP-SNX8 ( Figure 1—figure supplement 1A ) . We also observed localization of the SNX proteins to extensive inclusion-associated membrane tubules in a subset of infected cells as described previously ( Figure 1—figure supplement 1B ) ( Aeberhard et al . , 2015; Mirrashidi et al . , 2015 ) . Interestingly , when infected cells are treated with wortmannin , a pan-specific inhibitor of phosphoinositide-3-kinase ( PI3K ) activity , we see a loss of the SNX proteins from punctate endosomes , but not from the inclusion membrane ( Figure 1—figure supplement 2; Video 1 ) . A similar result is seen for specific inhibition of PtdIns3P production by Vps34 using Vps34-IN1 ( Figure 1—figure supplement 2 ) . This offers two possibilities; that either SNX recruitment to the inclusion occurs via protein-protein interactions , and does not depend on the presence of 3-phosphoinositide lipids that typically recruit SNX proteins to endosomal membranes , or alternatively that PI3Ks are not present at the inclusion and therefore wortmannin treatment has no effect at this particular compartment . Given our structural and mutagenesis studies below we favor the former explanation . 10 . 7554/eLife . 22311 . 006Video 1 . Movie showing that wortmannin disrupts SNX5 recruitment to endosomes but not the chlamydial inclusion . HeLa cells stably expressing mCherry-Rab25 ( red ) were transfected transiently with GFP-SNX5 ( green ) and infected with Chlamydia trachomatis L2 for 24 hr . Time-lapse videomicroscopy was performed using an interval of 1 min on an inverted Nikon Ti-E deconvolution microscope with environmental control at 40 x magnification . 10 min into recording 100 nM wortmannin was added . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 006 Mirrashidi et al . ( 2015 ) , demonstrated an in vitro interaction between IncE and the SNX5 and SNX6 PX domains . To confirm their direct association we assessed the binding affinities using isothermal titration calorimetry ( ITC ) ( Figure 2A; Table 1 ) . Initial experiments with the human SNX5 and SNX6 PX domains showed robust interactions with the IncE C-terminal domain ( residues 107–132 ) . The affinities ( Kd ) for SNX5 and SNX6 were essentially indistinguishable ( 0 . 9 and 1 . 1 µM respectively ) , but we detected no interaction with the PX domain of SNX1 confirming the binding specificity . The PX domains of SNX5 and SNX6 possess a helix-turn-helix structural insert ( Koharudin et al . , 2009 ) , which is not found in any other SNX family members except for SNX32 ( Figure 2B ) , a homologue expressed primarily in neurons ( BioGPS ( Wu et al . , 2009 ) ) . Confirming a common recruitment motif in the SNX5-related proteins , ITC showed a strong interaction between IncE and the SNX32 PX domain similar to SNX5 and SNX6 ( Kd = 1 . 0 µM ) ( Figure 2A; Table 1 ) , and SNX32 was robustly recruited to inclusion membranes in infected cells ( Figure 1B; Figure 1—figure supplement 1A ) . Overall , our data indicates that a common structure within the SNX5 , SNX6 and SNX32 PX domains is required for IncE interaction . 10 . 7554/eLife . 22311 . 007Figure 2 . IncE from C . trachomatis binds the PX domains of SNX5 , SNX6 and SNX32 . ( A ) Binding affinity between IncE peptide ( residues 107–132 ) and SNX PX domains by ITC . Top panels show raw data and lower panels show normalised integrated data . See Table 1 for the calculated binding parameters . Truncation analyses of the IncE peptide by ITC are shown in Figure 3 , Table 2 . ( B ) Sequence alignment of human SNX1 , SNX5 , SNX6 and SNX32 PX domains . Conserved residues are indicated in red . Side-chains that directly contact IncE in the crystal structure are indicated with black circles . Mutations that block IncE binding are highlighted with red triangles , and mutations that do not affect binding indicated with green circles . Secondary structure elements derived from the SNX5 crystal structure are indicated above . ( C ) Sequence alignment of IncE from C . trachomatis and putative homologues from C . muridarum and C . suis . IncE side-chains that directly contact SNX5 in the crystal structure are indicated with black circles . Mutations that block SNX5 binding are highlighted with red triangles , and mutations that do not affect binding indicated with green circles . Predicted transmembrane regions are indicated and C-terminal IncE sequences that form β-strands in complex with SNX5 are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 00710 . 7554/eLife . 22311 . 008Table 1 . Thermodynamic parameters of IncE binding to SNX PX domains* . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 008Sample cellTitrantKd ( µM ) △H ( kcal/mol ) T△S ( kcal/mol ) △G ( kcal/mol ) NSNX5 PXIncE peptide†0 . 95 ± 0 . 07−6 . 9 ± 0 . 3−1 . 9 ± 0 . 05−8 . 2 ± 0 . 011 . 01 ± 0 . 01SNX6 PXIncE peptide1 . 13 ± 0 . 08−5 . 0 ± 0 . 9−3 . 0 ± 1−8 . 0 ± 0 . 071 . 01 ± 0 . 08SNX32 PXIncE peptide1 . 15 ± 0 . 07−6 . 9 ± 0 . 4−1 . 3 ± 0 . 8−8 . 2 ± 0 . 41 . 06 ± 0 . 005SNX1 PXIncE peptideNo binding*Values are the mean from three experiments ±SEM . b . †IncE synthetic peptide sequence PANGPAVQFFKGKNGSADQVILVTQ . Finally we tested a series of IncE truncation mutants for their binding to the SNX5 PX domain ( Figure 3A , B and C; Table 2 ) . Synthetic peptides were used with single amino acids removed sequentially from the N and C-terminus to determine the minimal sequence required for binding . These experiments showed that the shortest region of IncE able to support full binding to SNX5 encompasses residues 110–131 ( GPAVQFFKGKNGSADQVILVT ) , while a shorter fragment containing residues 113–130 ( VQFFKGKNGSADQVILV ) can bind to SNX5 with a slightly reduced affinity . While variations are observed across the different C . trachomatis serovars ( Harris et al . , 2012 ) the SNX5-binding sequence appears to be preserved in all detected variants ( Figure 3D ) . A comparison with other chlamydial species suggests that IncE is not very widely conserved in this Genus , being clearly identifiable only in the closely related mouse pathogen C . muridarum and swine pathogen C . suis ( Figure 2C ) . Residues required for binding to SNX5 are preserved in these IncE homologues , but whether SNX proteins are also recruited during infection by these other chlamydial species remains to be determined . 10 . 7554/eLife . 22311 . 009Figure 3 . IncE residues 110–131 are sufficient for full recognition of the SNX5 PX domain . ( A ) Representative ITC experiments for truncated IncE peptides . These experiments were conducted using a single batch of SNX5 PX domain to minimize batch-to-batch protein variation . ( B ) Plots of the affinity constants for selected peptides to highlight the progressive loss of binding with N and C-terminal truncations . ( C ) Sequences of the truncated IncE peptides are given , with a qualitative indication of binding strength relative to the IncE_1 peptide containing residues 107–132 . Full binding is indicated by ‘++’ reduced binding by ‘+’ and lack of binding by ‘−‘ . All sequence information and their Kd values are given in Table 2 . When compared to the reference ITC experiment the binding affinity of peptides was unaffected when the first three N-terminal residues were removed ( IncE_2 , IncE_3 and IncE_4 ) and gradually became weaker until IncE_7 , after which binding was abolished . Results from IncE_6 are inconclusive due to the difficulty in successfully dissolving the peptides in buffer ( n . d . ) . C-terminal truncations showed that IncE_14 and IncE_15 had similar high binding affinities to the reference , while the binding of IncE_16 and IncE_17 became progressively weaker and peptides shorter than IncE_17 showed no binding . This data indicates that the minimal IncE binding sequence retaining full SNX5 binding is GPAVQFFKGKNGSADQVILVT , and a shorter fragment VQFFKGKNGSADQVIL can bind to SNX5 , albeit with a slightly reduced affinity . ( D ) Sequence alignment of IncE from different C . trachomatis serovars . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 00910 . 7554/eLife . 22311 . 010Table 2 . ITC data for SNX5 PX domain binding to truncated and mutated IncE peptides* . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 010ProteinPeptideSequenceKd ( µM ) △H ( kcal/mol ) T△S ( kcal/mol ) △G ( kcal/mol ) NSNX5 PXIncE_1PANGPAVQFFKGKNGSADQVILVTQ0 . 95 ± 0 . 07−6 . 9 ± 0 . 3−1 . 9 ± 0 . 05−8 . 2 ± 0 . 011 . 01 ± 0 . 01IncE_2 ANGPAVQFFKGKNGSADQVILVTQ1−5 . 0−2 . 6−8 . 10 . 98IncE_3 NGPAVQFFKGKNGSADQVILVTQ0 . 93−6 . 7−1 . 4−8 . 11 . 03IncE_4 GPAVQFFKGKNGSADQVILVTQ0 . 87−6 . 8−1 . 2−8 . 21 . 03IncE_5 PAVQFFKGKNGSADQVILVTQ2−5 . 9−1 . 2−8 . 30 . 99IncE_6 AVQFFKGKNGSADQVILVTQ/////IncE_7 VQFFKGKNGSADQVILVTQ2 . 2−6 . 9−1 . 1−7 . 70 . 99IncE_8 QFFKGKNGSADQVILVTQNo binding////IncE_9 FFKGKNGSADQVILVTQNo binding////IncE_10 FKGKNGSADQVILVTQNo binding////IncE_11 KGKNGSADQVILVTQNo binding////IncE_12 GKNGSADQVILVTQNo binding////IncE_13 KNGSADQVILVTQNo binding////IncE_14PANGPAVQFFKGKNGSADQVILVT0 . 72−5 . 1−1 . 6−8 . 41IncE_15PANGPAVQFFKGKNGSADQVILV0 . 97−6 . 5−1 . 3−8 . 20 . 98IncE_16PANGPAVQFFKGKNGSADQVIL1 . 1−5 . 6−1 . 4−8 . 120 . 99IncE_17PANGPAVQFFKGKNGSADQVI8 . 7−2 . 7−2 . 5−6 . 90 . 99IncE_18PANGPAVQFFKGKNGSADQVNo binding////IncE_19PANGPAVQFFKGKNGSADQNo binding////IncE_20PANGPAVQFFKGKNGSADNo binding////IncE_21PANGPAVQFFKGKNGSANo binding////IncE_22PANGPAVQFFKGKNGSNo binding////IncE_23PANGPAVQFFKGKNGNo binding////IncE_24PANGPAVQFFKGKNNo binding////IncE Q115APANGPAVAFFKGKNGSADQVILVTQ6 . 3−5 . 3−1 . 6−6 . 90 . 90IncE F116DPANGPAVQAFKGKNGSADQVILVTQNo bindingIncE K118APANGPAVQFFAGKNGSADQVILVTQ2 . 8−6 . 0−1 . 5−7 . 50 . 91IncE V127DPANGPAVQFFKGKNGSADQDILVTQNo bindingSNX5 PX L133DIncE_1PANGPAVQFFKGKNGSADQVILVTQNo bindingSNX5 PX F136AIncE_1PANGPAVQFFKGKNGSADQVILVTQNo bindingSNX5 PX E144AIncE_1PANGPAVQFFKGKNGSADQVILVTQ15−9 . 9−3 . 1−130 . 99*Except for IncE_1 all other peptide-binding experiments were performed only once . The canonical PX domain structure is composed of a three-stranded β-sheet ( β1 , β2 and β3 ) followed by three close-packed α-helices . The first and second α-helices are connected by an extended proline-rich sequence . Typically PX domains have been found to bind to the endosome-enriched lipid phosphatidylinositol-3-phosphate ( PtdIns3P ) via a basic pocket formed at the junction between the β3 strand , α1 helix and Pro-rich loop . In contrast SNX5 , SNX6 and SNX32 possess major alterations in the PtdIns3P-binding pocket that preclude canonical lipid head-group docking ( see below ) . In addition they possess a unique extended helix-turn-helix insert between the Pro-rich loop and α2 helix of unknown function ( Figure 2B ) ( Koharudin et al . , 2009 ) . To determine the structure of the SNX5-IncE complex we generated a fusion protein encoding the human SNX5 PX domain ( residues 22–170 ) and C . trachomatis IncE C-terminal sequence ( residues 108–132 ) attached at the PX domain C-terminus Figure 4—figure supplement 1A ) . This construct readily crystallised in several crystal forms , and we were able to determine the structure of the complex in three different spacegroups ( Figure 4; Table 3; Figure 4—figure supplement 1B ) . Confirming that the fusion does not alter complex formation , the short linker region is disordered , and the mode of IncE-binding to SNX5 is identical in all three structures ( Figure 4—figure supplement 1C and D ) . Because of the higher resolution , we focus our discussions on the structure of the SNX5 PX-IncE complex observed in the P212121 crystal form . The first three IncE N-terminal residues ( Pro107 , Ala108 , Asn109 ) and the last three IncE C-terminal residues ( Val130 , Thr131 , Gln132 ) were not modeled due to lack of electron density , suggesting disorder and matching precisely with our mapping experiments showing these residues are not necessary for SNX5 association . 10 . 7554/eLife . 22311 . 011Figure 4 . Structure of the SNX5 PX domain in complex with the IncE C-terminal domain . ( A ) Crystal structure of the SNX5 PX domain ( yellow ) in complex with IncE residues 107–132 ( magenta ) shown in cartoon representation . ( B ) Backbone atoms of the SNX5 and IncE proteins are shown to highlight the prominent β-sheet augmentation mediating the association between the two molecules . ( C ) Close up view of the SNX5-IncE interface highlighting specific contact areas at the N-terminus of the IncE peptide . ( D ) Close up of the SNX5-IncE interface highlighting specific contact areas at the hairpin loop of the IncE peptide shown at 90° to Figure 4C . ( E ) . Close up of the SNX5-IncE interface highlighting contact areas at the C-terminus of the IncE peptide in approximately the same orientation as Figure 4C . Residues in SNX5 ( Phe136 ) and IncE ( Phe116 ) that are critical for binding based on mutagenesis are boxed . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 01110 . 7554/eLife . 22311 . 012Figure 4—figure supplement 1 . Supplementary images for SNX5-IncE crystal structures . ( A ) Sequence of the SNX5 PX domain fusion protein with the IncE C-terminal peptide . ( B ) Refined 2fo-fc electron density contoured at 1 . 5σ for the SNX5-IncE structure in spacegroup P212121 . ( C ) Overlay of each independent SNX5-IncE complex observed in the three crystal forms . ( D ) Ribbon structures indicating the locations of the linker regions in each crystal form . The C-terminal SNX5 residues and the N-terminal IncE residues are shown by spheres with distances indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 01210 . 7554/eLife . 22311 . 013Table 3 . Summary of crystallographic structure determination statistics* . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 013CrystalSNX5 PX-IncE Form 1SNX5 PX-IncE Form 2SNX5 PX-IncE Form 3PDB ID5TGI5TGJ5TGHData collectionWavelength ( Å ) 0 . 953700 . 953700 . 95370Space groupP212121I2P32Cell dimensions a , b , c ( Å ) 60 . 7 , 67 . 5 , 88 . 258 . 4 , 80 . 3 , 94 . 6100 . 6 , 100 . 6 , 71 . 7 α , β , γ ( ° ) 90 , 90 , 9090 , 97 . 2 , 9090 , 90 , 120Resolution ( Å ) 60 . 7–1 . 98 ( 2 . 03–1 . 98 ) 31 . 9–2 . 6 ( 2 . 72–2 . 60 ) 50 . 3–2 . 80 ( 2 . 95–2 . 80 ) Rmerge0 . 104 ( 0 . 525 ) 0 . 153 ( 0 . 659 ) 0 . 101 ( 0 . 713 ) Rmeas0 . 112 ( 0 . 572 ) 0 . 18 ( 0 . 777 ) 0 . 124 ( 0 . 873 ) Rpim0 . 042 ( 0 . 225 ) 0 . 096 ( 0 . 408 ) 0 . 051 ( 0 . 363 ) <I> / σI12 . 4 ( 3 . 4 ) 39 . 6 ( 3 . 2 ) 11 . 7 ( 2 . 3 ) Total number reflections178868 ( 11000 ) 46691 ( 5757 ) 115149 ( 16861 ) Total unique reflections26075 ( 1805 ) 13432 ( 1632 ) 20001 ( 2923 ) Completeness ( % ) 100 ( 100 ) 99 . 9 ( 100 . 0 ) 100 ( 100 ) Multiplicity6 . 9 ( 6 . 1 ) 3 . 5 ( 3 . 5 ) 5 . 8 ( 5 . 8 ) Half-set correlation ( CC ( 1/2 ) ) 0 . 997 ( 0 . 868 ) 0 . 986 ( 0 . 55 ) 0 . 997 ( 0 . 683 ) RefinementResolution ( Å ) 45 . 1–1 . 98 ( 2 . 02–1 . 98 ) 31 . 9–2 . 6 ( 2 . 69–2 . 60 ) 41 . 2–2 . 8 ( 2 . 87–2 . 80 ) No . reflections/No . Rfree26021/200013421/1342 ( 1208/134 ) 19975/1972 ( 1301/144 ) Rwork/Rfree0 . 192/0 . 214 ( 0 . 221/0 . 246 ) 0 . 199/0 . 242 ( 0 . 276/0 . 332 ) 0 . 236/0 . 254 ( 0 . 329/0 . 372 ) No . atoms Protein257926195189 Solvent281690Average B-factor ( Å2 ) 31 . 842 . 556 . 0R . m . s deviations Bond lengths ( Å ) 0 . 0120 . 0110 . 015 Bond angles ( ° ) 1 . 271 . 151 . 27*Highest resolution shell is shown in parentheses . The IncE sequence forms a long β-hairpin structure that binds within a complementary groove at the base of the extended α-helical insertion of the SNX5 PX domain and adjacent to the β-sheet sub-domain ( Figure 4A; Video 2 ) . The β-hairpin structure of IncE ( N-terminal βA and C-terminal βB strands ) is directly incorporated as a β-sheet augmentation of the β1 , β2 and β3 strands of SNX5 ( Figure 4B ) . The N-terminal βA strand of the IncE sequence ( Gly111-Lys118 ) forms the primary interface with SNX5 , making main-chain hydrogen bonds with the β1 strand of the SNX5 PX domain for the stable positioning of the IncE structure . The two anti-parallel β-strands of IncE are connected by a short loop ( Gly119-Ala124 ) that makes no direct contact with the SNX5 protein , and the C-terminal IncE βB strand ( Asp125-Val130 ) forms an interface with the extended α-helical region of the SNX5 PX domain . 10 . 7554/eLife . 22311 . 014Video 2 . Animation highlighting the mechanism of interaction between SNX5 and IncE . The SNX5 PX domain is shown in yellow ribbons and the IncE peptide is shown in magenta . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 014 Detailed views of the SNX5-IncE interface are shown in Figure 4C , D and E . Aside from main-chain hydrogen bonding to form the extended β-sheet , IncE engages in several critical side-chain interactions with the relatively hydrophobic SNX5 binding groove . At the N-terminus of the βA strand Val114 of IncE inserts into a pocket formed primarily by Tyr132 and Phe136 on the SNX5 α’’ helix ( Figure 4C ) . A major contribution comes from IncE Phe116 , with π-stacking occurring with the Phe136 side-chain and hydrophobic docking with Val140 of SNX5 ( Figure 4D ) . Adjacent to IncE Phe116 at the end of the βA strand Lys118 makes an electrostatic contact with SNX5 Glu144 . Finally , at the C-terminal end of the IncE βB strand Val127 and Leu129 contact an extended SNX5 surface composed of Leu133 , Tyr132 and Met106 ( Figure 4E ) . To verify the crystal structure we mutated residues from both SNX5 and IncE and measured their affinities using ITC ( Figure 5A and B; Table 2 ) . At the interface between SNX5 and IncE several side chains make key contributions to peptide recognition . Because Leu133 and Phe136 residues in SNX5 are located at the core of the IncE-binding interface , and also due to the structural rearrangements these residues make on binding ( see below ) , we reasoned that L133D and F136A mutations would inhibit the interaction . Indeed these mutants abolished association with the IncE peptide ( Figure 5A ) . The reciprocal mutations in IncE residues F116A and V127D also abolished binding to the SNX5 PX domain ( Figure 5B ) , and the SNX6 and SNX32 PX domains ( Figure 5—figure supplement 1 ) , demonstrating the importance of these hydrophobic and π-stacking interactions for stable complex formation . In contrast mutations predicted to disrupt an observed electrostatic contact ( IncE K118A or SNX5 E144A ) had little effect on binding . Thus the core hydrophobic interactions are critical for IncE binding but the peripheral electrostatic contact is not essential . 10 . 7554/eLife . 22311 . 015Figure 5 . Mutations in the SNX5 and IncE proteins prevent complex formation in vitro . ( A ) ITC experiments testing the effect of SNX5 mutations on IncE binding . Both L133D and F136A mutations prevented IncE binding , but the A144A mutation had little effect . ( B ) ITC experiments testing the effect of IncE mutations on SNX5 binding . Both F116A and V127D blocked SNX5 interaction , while Q115A had a partial effect and K118A had no effect on association . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 01510 . 7554/eLife . 22311 . 016Figure 5—figure supplement 1 . SNX6 and SNX32 PX domains bind IncE at the same site as SNX5 . ITC experiments testing the effect of IncE peptide mutations on binding to SNX6 and SNX32 PX domains . The IncE F116A mutation blocks interaction with both PX domains similarly to SNX5 ( Figure 5B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 016 To confirm the role of IncE in direct SNX5 protein recruitment to the chlamydial inclusion we examined the localisation of GFP-tagged SNX5 in HeLa cells infected with C . trachomatis L2 ( CTL2 ) for 24 hr ( MOI ~0 . 5 ) . Cells expressing the GFP-SNX5 protein showed clear and uniform recruitment to the limiting membrane of the inclusion as defined by mCherry-Rab25 ( Figure 6A ) , which is consistent with the localisation observed by others ( Aeberhard et al . , 2015; Mirrashidi et al . , 2015 ) . In contrast , the GFP-SNX5 ( F136A ) mutant protein showed no recruitment to the chlamydial inclusion . The change in relative distribution of these GFP-SNX5 proteins on the inclusion was quantified for wildtype SNX5 ( Mander’s coefficient 0 . 67 ± 0 . 14 ) and GFP-SNX5 ( F136A ) ( 0 . 041 ± 0 . 051 ) ( Figure 6—figure supplement 1A ) . Like wild-type GFP-SNX5 the GFP-SNX5 ( F136A ) mutant was recruited to punctate endosomal structures throughout the cytoplasm of these cells , and in addition was able to co-immunoprecipate endogenous SNX1 in heterodimeric complexes identically to the wild-type GFP-SNX5 protein ( Figure 6—figure supplement 1B ) . This implies that BAR-domain mediated heterodimer formation with SNX1 or SNX2 is required for endosomal recruitment , and is not perturbed by the IncE-binding mutation in the PX domain . Lastly , we tested the importance of IncE residues for SNX interaction in situ by expressing the GFP-tagged IncE C-terminal domain . The wild-type GFP-IncE ( 91-132 ) was recruited to endosomal structures via its interaction with SNX5-related proteins in both uninfected and infected HeLa cells ( Figure 6B; Figure 6—figure supplement 1C ) . In contrast however , GFP-IncE ( 91-132 ) ( F116D ) , a SNX5-binding mutant , was exclusively cytosolic . Note that neither IncE construct is localised to the inclusion , as expected due to lack of signal peptides and transmembrane domains ( Figure 6—figure supplement 1C ) . 10 . 7554/eLife . 22311 . 017Figure 6 . Mutations in the SNX5 and IncE proteins prevents their interaction in cells . ( A ) Single amino-acid mutation in the PX domain of the SNX5 ( F136A ) abolishes recruitment to the chlamydial inclusion . HeLa cells stably expressing mCherry-Rab25 ( red ) were transfected transiently with GFP-SNX5 or GFP-SNX5 ( F136A ) ( green ) and infected with Chlamydia trachomatis L2 for 18–24 hr . The cells were fixed and the nucleic materials were counter-stained with DAPI ( blue ) . ( B ) HeLa cells were transfected transiently with GFP-IncE ( 91-132 ) or GFP-IncE ( 91-132 ) ( F116D ) ( green ) and co-labelled for the early endosomal marker EEA1 ( red ) . Mutation in the SNX5 binding IncE peptide ( F116D ) abolishes recruitment to endosomal structures . *Represents the inclusion . Scale bar 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 01710 . 7554/eLife . 22311 . 018Figure 6—figure supplement 1 . GFP-IncE C-terminal domain is localised to endosomes but not inclusions . ( A ) Quantitation of the degree of overlap between the GFP-SNX5 constructs and the mCherry-Rab25 inclusion membrane marker from Figure 6A . Mander’s correlation coefficient of mChRab25 signal over GFP-SNX5 signals ( 10 cells per group; error bars , S . D ) . ( B ) Co-immunoprecipitation of GFP-SNX5 from HeLa cells shows that both the wild-type and mutant protein ( F136A ) interact equally with endogenous SNX1 . This indicates that both proteins are correctly folded and otherwise functional . ( C ) HeLa cells stably expressing mCherry-Rab25 ( red ) were transfected transiently with GFP-IncE ( 91-132 ) or GFP-IncE ( 91-132 ) ( F116D ) ( green ) and infected with C . trachomatis L2 for 18–24 hr . Mutation in the SNX5 binding IncE peptide ( F116D ) abolishes recruitment to endosomal structures . Neither construct is recruited to inclusions , which is consistent with the lack of transmembrane regions . *Represents the inclusion . Scale bar 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 018 Superposition of the SNX5-IncE complex with the SNX5 PX domain in the apo state ( Koharudin et al . , 2009 ) reveals a significant conformational change in the α-helical extension , as well as localized alterations in the loop between the β1 and β2 strands to accommodate peptide binding ( Figure 7A ) . In essence the IncE β-hairpin acts as a tether between the core PX fold and extended α-helical hairpin , pulling the two sub-structures closer together . Overall the α-helical extension undergoes a maximal movement of ~8–10 Å at the furthest tip , facilitated by the flexibility of the structure following the Pro-rich linker and an apparent hinge-point at Pro97 ( Figure 7A upper panel ) . In the immediate vicinity of Pro97 the SNX5 loop that encompasses Asp43 is significantly shifted and stabilized by the repositioning of Arg103 . At both the start of the first α’ helix of the extension and the end of the second α’’ helix more subtle changes occur in the positions of Met106 , Leu128 , Tyr132 , Leu133 and Phe136 . These changes result in formation of the hydrophobic pocket that engages the IncE side-chains Val114 , Phe116 , Val127 and Leu129 ( Figure 7A , middle and lower panels ) . 10 . 7554/eLife . 22311 . 019Figure 7 . Conformational changes in SNX5 and a model for SNX-BAR recruitment to inclusion membranes . ( A ) Comparison of the SNX5-IncE complex ( yellow-magenta ) with the previously reported apo- SNX5 PX domain crystal structure ( blue ) ( PDB ID 3HPB ) ( Koharudin et al . , 2009 ) . The α-helical extension undergoes a significant displacement in the bound state . The enlarged panels to the right show several close-up views of the binding pocket highlighting conformational changes that are required to accommodate IncE . ( B ) A model for the SNX5-SNX1 PX-BAR heterodimer and its interaction with IncE at the inclusion membrane . The PX-BAR structure was modeled in silico ( see methods ) . The left panel shows cartoon representations of the structure , viewed from the side and from the membrane surface . Middle panels show the same structures in electrostatic surface representation ( red , negative; blue positive ) . The right panels show close ups of the putative PtdIns3P-binding pocket in SNX1 and SNX5 , with a PtdIns3P head-group ( shown in spheres ) docked by aligning the previous SNX9 crystal structure ( Pylypenko et al . , 2007 ) . SNX1 has a canonical PtdIns3P pocket , while SNX5 lacks a clear site for lipid head-group binding . ( C ) Sequence conservation of SNX5-related proteins was calculated and plotted using CONSURF . The surface representation indicates exposed side-chains that are evolutionarily conserved in green . The IncE peptide binds to a highly conserved surface groove , while the putative phosphoinositide binding region ( Koharudin et al . , 2009 ) on the opposite face is neither highly conserved nor poised to allow docking . ( D ) Cartoon model depicting the recruitment of SNX5 and related proteins to the inclusion membrane . Heterodimers with SNX1 or SNX2 will be recruited via IncE in infected cells , and this recruitment will be in competition with the binding of SNX1 and SNX2 to PtdIns3P for normal endosomal association , as well as interactions with other proteins including retromer and unidentified molecules that potentially bind to the conserved groove of the SNX5 PX domain . DOI: http://dx . doi . org/10 . 7554/eLife . 22311 . 019 To better understand how IncE can recruit the SNX5-containing SNX-BAR complexes to inclusion membranes we constructed an in silico model of the SNX5-SNX1 heterodimeric PX-BAR proteins ( Figure 7B ) . Consistent with the length of the IncE C-terminal cytoplasmic sequence the model predicts that the IncE sequence will bind to the surface of SNX5 close to , but oriented away from , the inclusion membrane . While PX domains are commonly able to recognise PtdIns3P lipid headgroups , SNX5-related proteins lack the typical binding pocket ( Figure 7B right panel ) , and there is some controversy regarding their ability to mediate specific membrane interactions ( Koharudin et al . , 2009; Teasdale and Collins , 2012 ) . We propose that in the context of C . trachomatis infection , SNX5-related proteins are directly associated with the inclusion via IncE protein-protein interactions in a phosphoinositide-independent manner , and are able to recruit their heterodimeric partners SNX1 and SNX2 ( Sierecki et al . , 2014; van Weering et al . , 2012; Wassmer et al . , 2009 ) . The PX-BAR-domain containing complexes are then localised to the inclusion in a retromer-independent manner ( Mirrashidi et al . , 2015 ) , and may contribute to the formation of the dynamic inclusion-associated membrane tubules . Interestingly , when a cross-species evolutionary analysis of side-chain conservation in the SNX5-related proteins is performed it is clear that the IncE peptide binds a hydrophobic surface groove that is strictly conserved in this protein family ( Figure 7C ) . This very strongly implies that the site is normally engaged in a protein-protein interaction with an as yet unidentified binding partner ( s ) required for SNX5’s regular biological function , and that IncE is directly competing for this interface .
Although more than fifty putative Incs have been identified in C . trachomatis , the exact roles of these inclusion membrane proteins are still poorly understood . Chlamydiae manipulate the host cellular and signaling networks via interactions between the cytoplasmic region of Incs and numerous host cell proteins . Recent studies reported retrograde trafficking proteins as significant components of the inclusion , with sorting nexin family members being particularly enriched ( Aeberhard et al . , 2015; Mirrashidi et al . , 2015 ) . In this study , we present the first reported crystal structure of a chlamydial inclusion protein ( IncE ) binding to its host effector protein ( SNX5 ) . While the detailed mechanism of IncE-mediated protein recruitment will be specific to this family member , the principle of extended cytoplasmic Inc sequences engaging with cellular host proteins on the inclusion is certain to be a general one . A simple analogy would be to consider the Inc proteins as being like a molecular ‘velcro’ that recognises and attaches host machinery needed for bacterial replication and survival . The manipulation of endocytic transport machinery is clearly critical for the obligate intracellular survival of C . trachomatis ( Aeberhard et al . , 2015; Mirrashidi et al . , 2015; Moore and Ouellette , 2014 ) . In addition to C . trachomatis , SNX1 , SNX2 , SNX5 , SNX6 and the associated retromer complex have also been directly implicated in the cellular pathogenesis of Coxiella burnetii ( McDonough et al . , 2013 ) , Salmonella enterica serovar Typhimurium ( Bujny et al . , 2008 ) , hepatitis C virus ( Yin et al . , 2016 ) , human papilloma virus ( Ganti et al . , 2016; Popa et al . , 2015 ) , and Legionella pneumophila ( Finsel et al . , 2013 ) . Broadly then the manipulation of SNX proteins and endosomal trafficking machinery by viral and bacterial pathogens is a common occurrence during intracellular infection , and points to a wide-ranging role in host-pathogen interactions . Typically PX domains of sorting nexins , including SNX1 and SNX2 ( Cozier et al . , 2002; Zhong et al . , 2005 ) , play an important role in endosomal membrane recruitment by binding the endosome-enriched lipid PtdIns3P through four conserved residues ( Mas et al . , 2014; Teasdale and Collins , 2012 ) . These residues are conserved in most PX domains including in SNX1 and SNX2 , but are entirely absent in SNX5 , SNX6 and SNX32 . Although there is evidence for the weak association of the SNX5 PX domain with the lipid PtdIns ( 4 , 5 ) P2 from nuclear magnetic resonance ( NMR ) spectroscopy experiments ( Koharudin et al . , 2009 ) , the crystal structure does not point to a clear binding mechanism . A second feature that sets SNX5-related proteins apart from the rest of the SNX family is the presence of an extended α-helical insertion . Our work confirms the central importance of this unique insert for the binding of the IncE inclusion protein , and provides the first clear description of how a PX domain can function as a protein-protein interaction scaffold as opposed to a lipid-binding domain . The high degree of conservation in the IncE binding surface of SNX5 implies that this site is critical for the normal function of SNX5 and its homologs . Previously , the expression of a GFP-tagged IncE C-terminal domain was shown to interfere with the SNX5/SNX6-dependent retrograde trafficking of the cation-independent mannose-6-phosphate receptor ( CI-MPR ) ( Mirrashidi et al . , 2015 ) . Combined with our structural data , this infers that IncE is mimicking and interfering with SNX5/SNX6-mediated protein interactions , with a ligand ( s ) required for normal endosomal trafficking that remains to be discovered . Once recruited to the inclusion , SNX-BAR proteins are localized to the bulk membrane and dynamic tubules . While it is logical to imagine they could play a positive role in the sculpting of the inclusion , this is somewhat difficult to reconcile with the effect of SNX5 and SNX6 knockdown , which results in an increased production of C . trachomatis infectious progeny . Alternatively , although a pool of SNX5/SNX6 and associated SNX1/SNX2 proteins remain on endosomes in C . trachomatis infected cells , their sequestering by the chlamydial inclusion may interfere with normal endosomal trafficking ( Figure 7D ) . It was thus proposed that the role of IncE could be to compete for SNX-retromer endosomal interactions , resulting in the breakdown of normal trafficking of the CI-MPR and lysosomal hydrolases and hence perturbation of the endolysosomal system’s capacity for bacterial destruction ( Aeberhard et al . , 2015; Mirrashidi et al . , 2015 ) . Defining the precise role of SNX proteins and other endocytic machinery in chlamydial infection will clearly require further study . In conclusion , our work provides novel molecular insights into the mechanism of SNX protein coercion by the IncE chlamydial effector , and presents a blueprint for future studies of other inclusion protein activities . In addition , our results provide a possible clue to understanding how SNX5-related molecules mediate protein interactions required for canonical cell trafficking pathways .
All synthetic peptides used for isothermal titration were purchased from Genscript ( USA ) . For ITC experiments , peptides were weighed and dissolved in 50 mM Tris ( pH 8 . 0 ) and 100 mM NaCl ( ITC buffer ) to make a stock peptide concentration of 2 mM , which was diluted to 0 . 75 mM before use . Polyclonal antibodies against C . trachomatis HtrA were generated previously ( Huston et al . , 2008 ) . Monoclonal antibodies against EEA1 ( 610457 , 1:100 ) , SNX1 ( 611483 , 1:200 ) and SNX2 ( 611308 , 1:200 ) were supplied by BD Bioscience . Monoclonal antibodies against the myc epiptope ( 9B11 , 1:2000 ) were supplied by Abcam . Rabbit polyclonal antibodies against GFP ( A-6455 , 1:500 ) were purchased from Molecular Probes ( Invitrogen ) . Rabbit polyclonal antibodies against Rab5 ( C8B1 , 1:100 ) were from Cell Signaling Technology . Goat polyclonal antibodies against Vps35 ( IMG-3575 , 1:400 ) were from Imgenex . Secondary antibodies were purchased from Molecular Probes ( Life Technologies ) and Li-Cor Bioscience . Wortmannin was supplied by Sigma-Aldrich ( W1628 ) . VPS34-In1 was from Merck Millipore ( 532628 ) . The IncE sequence used in this study is from the L3 serovar L3/404/LN ( NCBI reference WP_015506602 ) ( Harris et al . , 2012 ) . The pGEX-4T-2 bacterial expression plasmid encoding the human SNX5 PX domain ( residues 22–170 ) was generated using a standard PCR-based cloning strategy , and its identity confirmed by sequencing . All other bacterial expression constructs for human SNX proteins were synthesized and cloned into pGEX-4T-2 by Genscript ( USA ) . These included the SNX5 PX domain IncE fusion ( SNX5 residues 22–170 with IncE residues 108–132 fused at the C-terminus ( Figure 4—figure supplement 1A ) , SNX6 ( residues 29–170 ) , SNX32 ( residues 17–166 ) , and SNX5 PX domain mutants . The pcDNA3 . 1-N-eGFP mammalian expression constructs encoding full-length human SNX5 , SNX5 ( F136A ) , IncE ( 91-132 ) and IncE ( 91-132 ) ( F116D ) with N-terminal GFP-tags were generated by Genscript ( USA ) . The pCMU-myc-SNX5 was as described previously ( Kerr et al . , 2006 ) , and the SNX6 and SNX32 genes cloned into the pcDNA3 . 1-nMyc vector at BamHI and XhoI restriction sites ( Kerr et al . , 2012 ) . SNX5 , SNX32 and SNX8 were also cloned by polymerase chain reaction , restriction digest and ligation into pEGFP-C1 for expression with N-terminal GFP tags as described previously ( Wang et al . , 2010 ) . All proteins except SNX5 PX domain mutants were expressed in Escherichia coli Rosetta cells , whereas mutant constructs were expressed in BL21 Codon Plus supplemented with appropriate antibiotics . Single colonies from cultures grown on LB agar plates were inoculated into 50 mL LB2+ with ampicillin ( 0 . 1 mg/mL ) and chloramphenicol ( 0 . 1 mg/mL ) , and grown at 37°C with shaking overnight . The following day , 30 mL from the overnight culture was used to inoculate 1 L LB media containing ampicillin ( 0 . 1 mg/mL ) and chloramphenicol ( 0 . 1 mg/mL ) and incubated at 37°C . Cells were grown to an optical density ( OD ) of 0 . 5–0 . 6 at 600 nm and induced with 0 . 5 mM isopropyl-β-D-thiogalactopyranoside ( IPTG ) ( except for the SNX5-IncE fusion , where expression was induced at OD600 of 0 . 8 with 1 mM IPTG ) . Cultures were incubated with shaking overnight at 18°C until the cells reach an O . D of 3 . 0 ( ~24 hr ) . Cells were harvested using a Beckman rotor JLA 8 . 1000 at 4000 RPM for 30 min at 4°C . Pellets were resuspended in 10 mL lysis buffer ( 50 mM Tris ( pH 8 . 0 ) , 100 mM NaCl , 5% glycerol , 1 mM DTT , 0 . 1 mg/ml benzamidine , 0 . 1 mg/ml DNase ) per litre of culture . The cells were subjected to cell disruption and centrifugation at 18 , 000 RPM for 30 min at 4°C . The soluble fractions were first purified using affinity chromatography with glutathione-sepharose , and when required the GST tags were cleaved by thrombin while still bound to the column . The proteins were eluted in 50 mM Tris ( pH 8 . 0 ) , 100 mM NaCl , 5% glycerol , and 1 mM DTT , and then further polished using gel filtration chromatography ( Superdex 200 , GE healthcare ) in a buffer containing 50 mM Tris ( pH 8 . 0 ) , 100 mM NaCl . The fractions corresponding to the respective proteins were then pooled and used directly for ITC or were further concentrated for crystallization . ITC experiments were performed on a Microcal iTC200 instrument at 25°C . The proteins were buffer exchanged into ITC buffer ( 50 mM Tris ( pH 8 . 0 ) and 100 mM NaCl ) by gel filtration prior to ITC experiments . IncE peptides at 750 µM were titrated into 50 µM PX domain samples . The binding data was processed using ORIGIN 7 . 0 with a single site binding model to determine the stoichiometry ( n ) , the equilibrium association constant Ka ( 1/Kd ) , and the enthalpy ( △H ) . The Gibbs free energy ( △G ) was calculated using the equation △G = −RTIn ( Ka ) ; binding entropy ( △S ) was calculated by △G = △H – T△S . Three experiments were performed for each set of samples to determine the average ± standard error of the mean ( SEM ) for thermodynamic quantities , except for the peptide truncation experiments where only single experiments were performed . For these truncated peptide experiments , all experiments were performed using the same batch of protein to allow direct comparions to be made . The SNX5 PX domain fusion with IncE was concentrated to 15 mg/ml for crystallization . Eight 96-well crystallization hanging-drop screens were set up using a Mosquito Liquid Handling robot ( TTP LabTech ) at 20°C . Optimized diffraction-quality crystals were obtained using streak seeding in sitting drop vapor diffusion plates . The crystallisation solution for crystal form 1 was 0 . 2 M KSCN , 25% PEG 2K MME , 100 mM sodium acetate ( pH 5 . 5 ) , for crystal form 2 was 0 . 1 M NaCl , 0 . 1 M MgCl2 , 0 . 1 M Nacitrate ( pH 3 . 5 ) , 12 % PEG 4000 , and for crystal form 3 was 1 . 26 M ( NH4 ) 2SO4 , acetate ( pH 4 . 5 ) , 0 . 2 M NaCl . Data were collected at the Australian Synchrotron MX1 and MX2 Beamlines , integrated with iMOSFLM ( Battye et al . , 2011 ) , and scaled with AIMLESS ( Evans and Murshudov , 2013 ) in the CCP4 suite ( Winn et al . , 2011 ) . The structures were initially solved by molecular replacement with PHASER ( McCoy et al . , 2007 ) using the apo-SNX5 PX domain crystal structure as the input model ( PDB code 3HPB ) , minus the extended α-helical domain . The resulting model was rebuilt with COOT ( Emsley et al . , 2010 ) , followed by repeated rounds of refinement with PHENIX ( Adams et al . , 2011 ) . All structural figures were generated using PyMOL ( DeLano scientific ) . HeLa cells stably expressing mCherry-Rab25 were previously generated within the lab ( Teo et al . , 2016 ) and were maintained in DMEM ( Gibco ) supplemented with 10% ( v/v ) FCS ( Bovogen ) and 2 mM L-glutamine ( Invitrogen ) in a humidified air/atmosphere ( 5% CO2 ) at 37°C . Cells were transfected at 70% confluence with pcDNA3 . 1-N-eGFP plasmid constructs using Lipofectamine 2000 as per manufacturer’s protocol ( Invitrogen ) and examined 18–24 hr later . The HeLa cell line used in this study was from America Type Culture Collection ( #ATCC CCL2 ) . Parental and stable cells lines were negative for mycoplasma by DAPI staining , and authenticated by STR profiling ( Cell Bank Australia ) . For inhibitor treatments , cells were treated with either 100 nM wortmannin or 1 µM Vps34-IN1 for 1 hr . C . C . trachomatis serovar L2 ( ATCC VR-902B ) was used to infect cells at a multiplicity of infection ( MOI ) of ~0 . 5 . Cells were infected 2 hr post-transfection in normal DMEM ( Gibco ) supplemented with 10% ( v/v ) FCS ( Bovogen ) and 2 mM L-glutamine ( Invitrogen ) in a humidified air/atmosphere ( 5% CO2 ) incubator at 37°C . After 2 hr media was replaced with fresh media . Transfected and infected cells ( 18–24 hr post-infection ) were fixed with 4% paraformaldehyde , permeabilised using TritonX-100 ( Sigma ) and immunolabeled as described previously ( Teo et al . , 2016 ) and counter-stained with DAPI . The coverslips were imaged using a confocal laser-scanning microscope ( LSM 710 meta , Zeiss ) with 63x oil immersion objective . Time-lapse videomicroscopy was carried out on individual live cells using a Nikon Ti-E inverted deconvolution microscope using a 40x , 0 . 9 Plan Apo DIC objective , a Hamamatsu Flash 4 . 0 4Mp sCMOS monochrome camera and 37°C incubated chamber with 5% CO2 . GFP was excited with a 485/20 nm LED and captured using a 525/30 nm emission filter , and mCherry was excited using a 560/25 nm LED and captured using a 607/36 nm emission filter . Data was processed using ImageJ ( https://imagej . nih . gov/ij/ ) and compiled using Adobe Illustrator CS6 . The immunofluorescence colocalisation of GFP-SNX5 with chlamydial inclusion membranes ( Figure 6A; Figure 6—figure supplement 1A ) imaged on a confocal microscope was measured by Mander’s correlation coefficient of red pixel ( EEA1 or mCherry-Rab25 ) over green pixel ( GFP-SNX5 ) signals , which were determined using ImageJ ( https://imagej . nih . gov/ij/ ) with the JACoP plugin ( Bolte and Cordelières , 2006 ) . Punctate structures were automatically counted using ImageJ analyse particle tool across total of 10 cells from two biological replicates . To quantify the effect of PI3K inhibitors on SNX recruitment ( Figure 1—figure supplement 2 ) , Z-stacks were captured with a Zeiss 710 confocal laser scanning microscope using a 40x objective . Maximum projections were generated with FIJI ( https://fiji . sc/ ) and Pearson’s correlation coefficients for individual cells determined using the FIJI ‘Coloc 2’ function with Costes threshold regression and 100 Costes randomisations . Co-localization analyses were conducted on two independent experiments from five images per condition each containing at least 20 cells ( >100 cells analysed per condition ) . HeLa cells were transfected with pcDNA3 . 1-N-eGFP plasmid constructs overnight at 70% confluence and the cells were lysed using lysis buffer ( H2O , 50 mM HEPES , 150 mM NaCl , 1% Triton-X100 , 10 mM Na4P2O7 , 30 mM NaF , 2 mM Na3VO4 , 10 mM EDTA , 0 . 5 mM AEBSF and protease inhibitor cocktail ) . Cell lysates were incubated with GFP nano-trap agarose beads ( Protein Expression Facility , UQ ) after preclear using protein G-agarose beads ( Invitrogen ) . Protein complexes attached to the beads were detached by boiling for 5 min with 5x denaturing and reducing buffer ( 0 . 625 M Tris pH 6 . 8 , 50% glycerol , 10% SDS , 0 . 25% Bromophenol blue and 500 mM DTT ) . Denatured and reduced proteins were separated by molecular mass using SDS-PAGE . Proteins were transferred onto PVDF-FL membrane ( Immobilon ) and were detected by immunoblotting with polyclonal anti-GFP and monoclonal SNX1 antibodies , and near-infrared fluorescent dyes ( LI-COR ) . Immunolabelled proteins were visualised using LI-COR Odyssey imaging system . Human SNX5 and SNX1 sequences were submitted to the PHYRE2 server for automated homology-based model building ( Kelley et al . , 2015 ) . For both proteins the top scoring modelling template was the crystal structure of the SNX9 PX-BAR domains ( PDB ID 2RAJ ) ( Pylypenko et al . , 2007 ) with Confidence Scores of 100% ( and sequence identities of 19% and 16% respectively ) . The PX domain of the SNX5 model generated using this structural template was missing the extended α-helical insert , so to complete the model the SNX5 PX domain-IncE complex was substituted and a dimer of SNX5 and SNX1 PX-BAR domains generated by overlaying with the SNX9 dimer in the PtdIns3P-bound state ( PDB ID 2RAK ) ( Pylypenko et al . , 2007 ) . The resulting model was subjected to simple energy minimisation in PHENIX ( Adams et al . , 2011 ) . Conservation of surface residues was computed using the CONSURF server ( Ashkenazy et al . , 2016 ) . Structural data are deposited in the protein data bank ( PDB ) under accession numbers 5TGI , 5TGJ , and 5TGH . Raw diffraction images are available on the University of Queensland eSPACE server ( http://espace . library . uq . edu . au/view/UQ:409277 ) . | The bacterium Chlamydia trachomatis , commonly known as chlamydia , is a frequent cause of sexually transmitted infections , and a leading cause of blindness due to infection . The bacteria must directly enter the cells of its human host to grow and multiply . Inside a human cell , the bacteria form and then develop within specialized compartments called inclusions that are surrounded by membrane . The outside of the inclusion membrane becomes coated with dozens of unique bacterial proteins . The major role of these bacterial proteins is to hijack other proteins in the human cell to generate and maintain the membrane of the inclusion compartments . One bacterial protein in particular , called IncE , is able to bind to specific host proteins called sorting nexins . These host proteins normally control the formation of tube-like membrane structures , which transport fatty molecules and proteins throughout the cell . The IncE protein is thought to recruit sorting nexins to help shape the inclusion membrane and perhaps control which types of proteins and fatty molecules associate with it . However , until now it was unknown how IncE , or any similar protein for that matter , could specifically hijack a host cell protein . Now , Paul et al . have revealed the three-dimensional structure of a human sorting nexin protein , called SNX5 , bound to a small fragment of the IncE protein from chlamydia . The structure shows that the part of SNX5 that associates with IncE is the part of the protein normally thought to interact with specific fatty molecules rather than proteins . Further experiments showed that SNX5 was still recruited to the inclusion compartment when the amount of these fatty molecules in human cells was reduced . However , this was not the case if SNX5 was prevented from interaction with the IncE protein . Paul et al . also observed that the site on SNX5 where IncE binds is almost identical in related proteins from many other species , including zebrafish and worms , most of which are not hosts for chlamydia . This lead them to suspect that IncE hijacks the sorting nexin proteins by mimicking an important host protein that is yet to be discovered . Proteins in the inclusion membrane play many important roles , and so this work on IncE only provides the first glimpse at how these proteins are able to manipulate the machinery of the host cell to their own ends . Further studies will therefore be needed to understand how these proteins exploit their host environment at the molecular level , and might be targeted in new antibacterial approaches . The findings also show how studying bacteria that live within host cells , like chlamydia , can provide insight into how other molecules are normally transported within cells: a process that is fundamental to all living cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"microbiology",
"and",
"infectious",
"disease"
] | 2017 | Structural basis for the hijacking of endosomal sorting nexin proteins by Chlamydia trachomatis |
During infection , CD8+ T cells initially expand then contract , leaving a small memory pool providing long lasting immunity . While it has been described that CD8+ T cell memory formation becomes defective in old age , the cellular mechanism is largely unknown . Autophagy is a major cellular lysosomal degradation pathway of bulk material , and levels are known to fall with age . In this study , we describe a novel role for autophagy in CD8+ T cell memory formation . Mice lacking the autophagy gene Atg7 in T cells failed to establish CD8+ T cell memory to influenza and MCMV infection . Interestingly , autophagy levels were diminished in CD8+ T cells from aged mice . We could rejuvenate CD8+ T cell responses in elderly mice in an autophagy dependent manner using the compound spermidine . This study reveals a cell intrinsic explanation for poor CD8+ T cell memory in the elderly and potentially offers novel immune modulators to improve aged immunity .
Upon successful clearance of a pathogen , the majority of short-lived effector T cells die and the remaining cells differentiate into a memory T cell population that provides long lasting immunity . While the cytokines , surface molecules , and signaling components involved in T cell memory formation have been extensively studied , the molecular pathways supporting these cell fate decisions are poorly understood . The T memory population is ( 1 ) quiescent , ( 2 ) long-lived and actively maintained , and ( 3 ) relies on mitochondrial respiration ( Pearce and Pearce , 2013 ) . Both reduced cell cycling and longevity of this stem cell-like population demand rigorous maintenance of the cytoplasm as debris cannot be diluted to daughter cells , reminiscent of true stem cells ( Mortensen et al . , 2011; Guan et al . , 2013; Warr et al . , 2013 ) . Across mammalian cell types , clearance of debris and damaged organelles such as mitochondria is typically executed via autophagy ( Choi et al . , 2013 ) . A recent study showed that formation of the influenza-specific B cell memory pool requires autophagy ( Chen et al . , 2014 ) . However , while the role of autophagy in the homeostasis of naïve T cells is well studied ( Puleston and Simon , 2014 ) , nothing is known about the requirement of autophagy in antigen-experienced T cells . The T cell memory pool has previously been shown to be controlled by PI3K/Akt and AMPK signaling as well as mTOR ( mammalian Target Of Rapamycin ) inhibition ( Araki et al . , 2009; Kim et al . , 2012; Rolf et al . , 2013 ) , all of which also control autophagy ( Jung et al . , 2010 ) . Indeed , T cell memory responses can be improved with the mTOR inhibitor rapamycin ( Araki et al . , 2009 ) . Interestingly , rapamycin also rejuvenates HSCs in old mice ( Chen et al . , 2009 ) . Neither of these studies addressed autophagy as a contributing mechanism . Evidence for age-related declining levels of autophagy stems largely from lower organisms such as yeast , flies , and worms ( Rubinsztein et al . , 2011 ) . A recent study showed that the polyamine spermidine induces autophagy and thereby prolongs life span in model organisms ( Eisenberg et al . , 2009 ) . Our own work demonstrated decreased levels of autophagy in CD8+ T cells of old individuals ( Phadwal et al . , 2012 ) . In this study , investigating a mouse model lacking the essential autophagy gene Atg7 specifically in T cells , we find peripheral T cell lymphopenia , leading to proliferation and an activated phenotype within the CD8+ T cell compartment . While Atg7−/− T cells respond normally during the early stages of live viral challenge , a severely compromised memory CD8+ T cell compartment was found in response to influenza and murine cytomegalovirus ( MCMV ) . Using bone marrow ( BM ) chimeras , we excluded that this is due the effects of lymphopenia; poor CD4+ T cell help; exhaustion , or altered cytokine receptor expression . Moreover , autophagy was found to be highest in antigen-specific CD8+ T cells when compared to naïve cells . Antigen-specific Atg7−/− CD8+ T cells also underwent more cell death at the time of memory formation , display compromised mitochondrial health , and increased expression of the glucose receptor GLUT1 , a marker for glycolysis . Furthermore , recall CD8+ T cell responses to repeat immunizations and vaccination protocols were greatly diminished . This being reminiscent of the human ageing immune system ( Haq and McElhaney , 2014 ) , we confirmed reduced autophagy at the transcriptional and functional level in murine T cells from old mice . Importantly , we were able to restore the CD8+ T cell memory response in old mice with the autophagy-inducing compound spermidine , but not in autophagy-deficient mice . Finally , we found that spermidine induces autophagy independently of mTOR in T cells . Enhancing autophagy in an mTOR-independent manner may provide a safe way to improve vaccine responses in the elderly .
Atg7flox/flox mice were bred with CD4-Cre mice to generate mice with defective autophagy in both CD4+ and CD8+ T lymphocytes ( T-Atg7−/− ) . Successful excision and thereby absence of Atg7 mRNA and Atg7 protein was confirmed in purified T cells ( Figure 1—figure supplement 1A and B , respectively ) . Using the imaging flow cytometer ( ImageStream ) to count LC3 puncta in CD4+ and CD8+ T cells ( Phadwal et al . , 2012 ) , we demonstrated that functional autophagy was significantly diminished in Atg7−/− CD8+ T cells ( Figure 1—figure supplement 1C with examples of ImageStream images in right panel ) . In addition , using a classical technique to detect lipidated LC3 , we confirmed that basal autophagy was diminished in the presence and absence of the autophagy flux inhibitor Bafilomycin A ( Figure 1—figure supplement 1D ) . Previous reports have noted a number of changes to the naïve CD8+ T cell compartment in the absence of autophagy , with T cell lymphopenia , a consistent observation ( Pua et al . , 2007; Puleston and Simon , 2014 ) . We set out to investigate if an altered naïve CD8+ T cell compartment exists in T-Atg7−/− mice . We confirmed observations from previous reports using similar autophagy-deficient mouse models ( Pua et al . , 2007 , 2009 ) that thymic development of CD4+ and CD8+ T cells was normal in 6-week old T-Atg7−/− mice ( Figure 1A ) . However , mice were lymphopenic for both CD4+ and CD8+ T cells in the lymph nodes and blood ( Figure 1B , C ) . Moreover , Atg7−/− CD8+ T cells exhibited an activated phenotype with increased CD44 expression ( Figure 1D ) and decreased CD62L expression ( Figure 1E ) , resembling a ‘virtual memory’ compartment ( Akue et al . , 2012 ) . We observed similar frequencies of central effector memory CD62L+CD44hi , however , T-Atg7−/− mice accumulated CD8+ T cells with an effector memory phenotype ( CD62L−CD44hi ) ( Figure 1—figure supplement 1E ) . Similarly , data from a mouse model in which another essential autophagy gene , Atg5 , was deleted under the hematopoietic stem cell-specific promoter Vav , showed T cell lymphopenia and the expanded virtual memory T cell compartment ( CD8+CD44+ ) suggesting this phenotype is not Atg7 specific ( Figure 1—figure supplement 2A and B ) . Next , we established that proliferation was increased in the activated CD44hi CD8+ T cell compartment by Ki-67 staining ( Figure 1F ) . The observed activated phenotype and increased cell turnover in Atg7−/− CD8+ T cells are likely driven by homeostatic proliferation in an attempt to fill the depleted T cell niche . Indeed , the expression of the homeostatic proliferation marker CD24 ( Li et al . , 2006 ) was found to be significantly increased on Atg7−/− CD8+ T cells ( Figure 1G ) . To investigate whether lymphopenia drives this activated phenotype in the CD8+ T cell compartment , we generated 1:1 mixed bone marrow ( BM ) chimeras from CD45 . 2+ T-Atg7−/− BM mixed with CD45 . 1+ wild-type BM . Both BMs contributed equally to form the new hematological system ( Figure 1—figure supplement 2C ) . Atg7−/− CD8+ T cells were still diminished even in the presence of a replete T cell niche , suggesting the survival defect previously described for autophagy-deficient T cells ( Pua et al . , 2007; Mortensen et al . , 2010 ) is cell-intrinsic ( Figure 1—figure supplement 2D ) . However , the activated Atg7−/− CD8+ T cell phenotype was no longer detected in BM chimeras as measured by the frequency of donor CD45 . 2+ CD8+ T cells found to be CD62L+ ( Figure 1H ) and CD44hi ( Figure 1I ) . These data indicate that the observed homeostatic proliferation and the change in surface phenotype of Atg7−/− CD8+ T cells are driven by lymphopenia and are not cell-intrinsic . 10 . 7554/eLife . 03706 . 003Figure 1 . Lymphopenia induces homeostatic proliferation and an activated CD8+ T cell phenotype in T-Atg7−/− mice . ( A ) Frequency of mature single positive CD4+ and CD8+ T cells in thymi of 6 week old mice ( n = 4 ) , representative FACS plot of three independent experiments . ( B ) Flow cytometric analysis of CD4+ and CD8+ T cell frequencies in blood and lymph nodes of T-Atg7−/− and wild-type mice ( dot plots depict staining in blood ) . Quantitative analyses are representative of six independent experiments , *p < 0 . 05 ( n = 4 ) . ( C ) Absolute counts of CD4+ and CD8+ T cells in blood of T-Atg7−/− and WT mice over time ( n = 4 ) . ( D ) Percentage of CD44hi cells in the splenic CD8+ T cell compartment of T-Atg7−/− and WT mice . Bar graphs depict the frequency of gated cells ( representative of seven independent experiments ) , *p < 0 . 05 ( n = 4 ) . ( E ) Percentage of splenic CD8+ T cells positive for CD62L in T-Atg7−/− and WT mice . Data are representative of three independent experiments , *p < 0 . 05 , ( n = 4 ) . ( F ) Percentage of CD44lo and CD44hi CD8+ T cells expressing Ki-67 in the spleen of T-Atg7−/− and WT mice . Bar graphs are representative of three independent experiments , *p < 0 . 05 ( n = 4 ) . ( G ) Frequency of splenic CD8+ T cells expressing CD24 in T-Atg7−/− and WT mice . Bar graph is representative of two independent experiments , *p < 0 . 05 ( n = 4 ) . ( H ) Percentage of donor-derived CD62L+CD8+ T cells . Lethally irradiated CD45 . 1 hosts reconstituted with a 1:1 mix of T-Atg7−/− or WT BM ( both CD45 . 2 ) with CD45 . 1 wild-type BM . Controls were WT and T-Atg7−/− mice; *p < 0 . 05 ( n = 4 ) . ( I ) Frequency of donor-derived CD45 . 2+ CD8+ T cells in the spleen that is CD44hi . Controls were normal WT and T-Atg7−/− mice , *p < 0 . 05 ( n = 4 ) . All values are mean ± s . e . m , and all statistical analyses are Mann–Whitney U-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 00310 . 7554/eLife . 03706 . 004Figure 1—figure supplement 1 . Atg7 is efficiently excised in CD8+ T cells from T-Atg7−/− mice resulting in loss of functional autophagy . ( A ) Atg7 gene expression in Atg7+/+ and Atg7−/− T cells . T cells were isolated from the spleens of WT and T-Atg7−/− mice by flow cytometry . mRNA was extracted , and Atg7 gene expression was measured by q-PCR . Bar graph shows relative Atg7 expression in T cells from T-Atg7−/− mice compared to T cells from WT mice . ****p < 0 . 0001 by Student's t test . ( B ) Protein was purified from pooled MACS-sorted CD8+ T cells from WT and T-Atg7−/− mice and western blotted for ATG7 . GAPDH was used as a loading control ( n = 6 pooled mice ) . ( C ) LC3 Spot count in WT and Atg7−/− CD8+ T cells . Splenocytes from WT and T-Atg7−/− mice were untreated or treated for 2 hr with an autophagy flux inhibitor before staining for relevant cell surface markers and LC3-II . LC3 spot count was assessed using ImageStream software . Quantification is gated on CD8+ T cells . Right panel depicts example LC3 spot count images ( ×60 magnification ) . *p = 0 . 0132 , ***p = 0 . 0002 as determined by Student t test ( n = 4 ) ( D ) Autophagy flux in WT and Atg7−/− CD8+ T cells . Purified CD8+ T cells were untreated or treated with bafilomycin for 6 hr before whole protein extraction and western blot for LC3 . ( E ) Percentage of CD44hi CD62L− and CD44hi CD62L+ cells in the splenic CD8+ T cell compartment of T-Atg7−/− and WT mice . Representative FACS plots of three independent experiments are shown , ***p = 0 . 0003 as determined by Mann–Whitney U-test ( n = 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 00410 . 7554/eLife . 03706 . 005Figure 1—figure supplement 2 . The T cell phenotype of Vav-Atg5−/− mice is similar to that of T-Atg7−/− mice and the survival defect of Atg7−/− CD8+ T cells is cell intrinsic . ( A ) Flow cytometric analysis of CD4+ and CD8+ T cell frequencies in the spleen of WT and Vav-Atg5−/− mice . Quantitative analyses are representative of two independent experiments , **p < 0 . 01 ( n = 5 ) . ( B ) Frequency of CD44hi cells within the CD8+ T cell subset of WT and Vav-Atg5−/− mice . Data are representative of two independent experiments , **p < 0 . 01 ( n = 5 ) . ( C ) BM reconstitution in BM chimera mice . 9 weeks after marrow transplantation , the frequency of donor CD45 . 2 cells was assessed in the spleen by flow cytometry . Quantified is the frequency of splenocytes expressing CD45 . 2 in BM chimera mice and in normal WT and T-Atg7−/− mice that acted as controls . ( D ) Frequency of CD8+ T cells that are CD45 . 2 donor-derived in BM chimera mice . Example dot plots are gated on CD8+ T cells . Quantified is the frequency of CD8+ T cells that express CD45 . 2 in BM chimera and in normal WT and T-Atg7−/− mice that acted as controls . *p < 0 . 05 ( n = 4 ) . Data are mean ± s . e . m and all statistical analyses are Mann–Whitney U-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 005 Previous studies have shown a role for autophagy in naive T cell organelle homeostasis and survival ( Pua et al . , 2009; Jia and He , 2011 ) . However , the importance of autophagy in CD8+ T cells responding to infection is unknown . To investigate this , we challenged T-Atg7−/− mice with influenza ( PR8 strain ) and MCMV . In T-Atg7−/− mice , we found normal expansion of the antigen-specific effector CD8+ T cell ( CD8+ Teff ) compartment using influenza-specific tetramers on day 10 ( peak response ) in the lungs ( Figure 2A ) . However , due to the pre-existing lymphopenia observed in naïve mice , the absolute counts were diminished ( Figure 2—figure supplement 1A ) . The CD8+ Teff response to MCMV in the blood on day 7 was also normal ( Figure 2B ) ( Hutchinson et al . , 2011 ) . However , the ability of T-Atg7−/− mice to form memory CD8+ T cells ( CD8+ Tmem ) to both influenza and MCMV is severely compromised ( Figure 2C , D ) . Performing serial bleeding on influenza infected mice over time , we demonstrated a catastrophic collapse of the antigen-specific CD8+ T cell pool at day 21 , resulting in a failure to retain CD8+ Tmem in T-Atg7−/− mice ( Figure 2E ) . Interestingly , the response to the ‘inflationary’ epitopes m38 and IE3 from MCMV is also compromised ( Figure 2F ) . In response to these epitopes , wild-type CD8+ T cells continue to expand throughout MCMV chronic infection . However , Atg7−/− CD8+ T cells fail to inflate and instead undergo a dramatic contraction leading to a failure to form a MCMV-specific CD8+ T cell memory pool . This profound CD8+ T cell contraction in T-Atg7−/− mice is also observed in response to a conventional , non-inflating MCMV epitope ( m45 , Figure 2—figure supplement 1B ) . While viral titers against influenza in the lungs of wild-type and T-Atg7−/− mice were comparable at day 3 of infection , they were significantly higher on day 6 in T-Atg7−/− mice ( Figure 2G ) . This is most likely due to the significantly lower absolute number of antigen-specific effector CD8+ T cells . As T-Atg7−/− mice survived influenza challenge , we expect that the virus is eventually cleared in all mice . In keeping with these findings , autophagy levels ( CytoID ) are significantly increased in the antigen-specific CD8+ T cell compartment compared to naïve T cells ( CD44lo ) in response to influenza in both spleen ( Figure 2H ) and lungs ( Figure 2I ) , indicating autophagy is induced upon antigen stimulation in vivo . 10 . 7554/eLife . 03706 . 006Figure 2 . Normal effector CD8+ T cell responses to viral infection but defective memory CD8+ T cell formation in T-Atg7−/− mice . ( A ) Effector CD8+ T cell response to influenza in WT and T-Atg7−/− mice . Mice were immunized intra-nasally with 0 . 00032 HAU PR8 influenza . On day 10 , antigen-specific CD8+ T cells to nucleoprotein ( NP ) was assessed with tetramer in lungs . Dot plots show examples of tetramer staining gated on CD8+ T cells . Bar graph indicates percentage of CD8+ T cells specific for NP ( n = 5–6 ) and is representative of three independent experiments . ( B ) Effector CD8+ T cell's response to MCMV in WT and T-Atg7−/− mice . CD8+ T cells from blood were stained with m45 tetramer on day 7 post-infection . Dot plots show m45-tetramer+ cells gated on CD8+ T cells . Bar graph indicates % CD8+ T cells specific for m45 ( n = 4–5 ) and is representative of three independent experiments . ( C ) CD8+ Tmem response to influenza . WT and T-Atg7−/− mice immunized as in ( A ) and the antigen-specific CD8+ T cell response was assessed in lungs on day 50 by tetramer . *p < 0 . 05 , by Mann–Whitney U-test ( n = 4 ) . Dot plots are gated on CD8+ T cells . Bar graph is representative of three independent experiments . ( D ) CD8+ Tmem's response to MCMV . Lung CD8+ T cells on day 65 post-infection were stained with IE3-tetramer . Dot plots are gated on CD8+ T cells . Quantitation depicts frequency of IE3 specific CD8+ T cells . *p < 0 . 05 , by Mann–Whitney U-test ( n = 4 ) . Data are representative of two independent experiments . ( E ) CD8+ T cell kinetics to influenza infection . WT and T-Atg7−/− mice were immunized as in ( A ) and CD8+ T cell response tracked over time in blood by tetramer . Y-axis shows frequency of NP-specific CD8+ T cells . ( F ) CD8+ T cell kinetics to MCMV infection . CD8+ T cell response to epitopes m38 ( left panel ) and IE3 ( right panel ) were tracked over time in blood by tetramer in WT and T-Atg7−/− mice . Y-axis indicates the percentage of CD8+ T cells that are m38-specific . ( G ) Influenza virus titres . WT and T-Atg7−/− mice were culled at days 3 and 6 post-immunization with PR8 , and lungs were collected and snap frozen in liquid nitrogen . Virus titres were determined using MDCK-SIAT1 cells . ***p = 0 . 0002 as determined by Student t-test ( n = 4 ) ( H ) WT mice were immunized with PR8 influenza as in ( A ) . CD8+ T cells from spleen were stained with CytoID at day 9 post-infection and assessed by flow cytometry . Histograms show examples of CD44lo CD8+ T cells from unimmunized mice ( filled grey line ) and in NP-specific CD8+ T cells from immunized mice ( open black line ) . Quantification is by mean fluorescence intensity ( MFI ) of CytoID on gated indicated cell population and representative of two independent experiments . ****p < 0 . 0001 by Student t test ( n = 6 ) . ( I ) Autophagy levels by CytoID staining on CD8+ T cells from the lungs on day 9 post-infection as in ( G ) . Histograms provide examples of CytoID staining , gated on CD8+ T cells in lungs of immunized mice ( solid grey lines ) , and gated on NP-tetramer-specific CD8+ T cells from immunized mice ( open black lines ) . Quantification is by mean fluorescence intensity ( MFI ) of CytoID on gated indicated cell population and representative of two independent experiments . ****p < 0 . 0001 by Student t test ( n = 4–8 ) . All values are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 00610 . 7554/eLife . 03706 . 007Figure 2—figure supplement 1 . T-Atg7−/− mice fail to form memory CD8+ T cells to a conventional MCMV epitope . ( A ) Effector CD8+ T cell absolute counts in WT and T-Atg7−/− mice . Quantification of the absolute number of CD8+ NP-tetramer+ T cells was determined in the lungs on day 10 of PR8 influenza infection . **p = 0 . 0084 by Student t test ( n = 6–9 ) . ( B ) Frequency of m45-specific CD8+ T cells in the liver of MCMV-immunized WT and T-Atg7−/− mice on day 100 post-infection . Dot plots are gated on CD8+ T cells; bar graphs depict the percentage of CD8+ T cells that are m45-tetramer+ . All values are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 007 To exclude that a failure to maintain a CD8+ Tmem pool is caused by lymphopenia in T-Atg7−/− mice , we performed viral challenge experiments in 1:1 mixed BM chimeras , facilitating the observation of antigen-specific Atg7−/− CD8+ T cell responses in a replete T cell environment . Even in the context of a BM chimera , Atg7−/− CD8+ T cells fail to generate a CD8+ Tmem pool following influenza infection , suggesting this defect is cell intrinsic ( Figure 3A ) . By using BM chimeras to observe antigen-specific Atg7−/− CD8+ T cell responses , we excluded a number of explanations as to why T-Atg7−/− mice fail to form CD8+ Tmem: ( 1 ) lymphopenia , ( 2 ) defective CD4+ T cell help , and ( 3 ) the excision of Atg7 in CD4-expressing antigen-presenting cells affecting CD8+ T cell priming . Furthermore , a defect in the frequency of memory precursor cells might also impact upon CD8+ Tmem formation in T-Atg7−/− mice . Mixed BM chimeras demonstrated that the frequency of short-lived effector ( SLEC , CD127+KLRG1− ) and memory precursor ( MPEC , CD127−KLRG1+ ) CD8+ T cells was in fact normal in T-Atg7−/− mice ( Figure 3B ) . We then tested whether the lack of memory response is due to exhaustion of autophagy-deficient antigen-specific CD8+ T cells . While in T-Atg7−/− mice , we found a sharp increase of antigen-specific CD8+ T cells co-expressing the exhaustion markers PD-1+ and TIM-3+; this was not found in the mixed BM chimeras . This indicates that without lymphopenia , exhaustion is not a hallmark of Atg7−/− CD8+ T cell responses and cannot explain the diminished CD8+ Tmem compartment ( Figure 3C ) . We then sought to determine if altered expression of cytokine receptors crucial for CD8+ Tmem maintenance could explain the defective CD8+ Tmem compartment . However , both Il-7Rα ( CD127 ) and Il-15Rα were normally expressed on Atg7−/− MCMV-specific CD8+ T cells ( Figure 3D , E ) . 10 . 7554/eLife . 03706 . 008Figure 3 . Loss of memory CD8+ T cell formation in the absence of Atg7 is not due to lymphopenia , poor CD4+ T cell help , exhaustion , or defective cytokine receptor expression . ( A ) CD8+ Tmem response to influenza in mixed BM chimeras , generated as in Figure 1H . Mice were immunized with PR8 influenza , and the CD8+ Tmem response of the CD45 . 2 donor to NP was assessed in lungs by tetramer on day 40 . Quantitation shows frequency of ( donor ) CD45 . 2+ CD8+ T cells that are NP-specific ( n = 5–6 ) . **p < 0 . 01 , by Mann–Whitney U-test ( n = 4–7 ) . ( B ) SLEC and MPEC populations in the Atg7+/+ and Atg7−/− antigen-specific CD8+ T cell pool . Mixed BM chimera generated as in Figure 1H , were immunized with MCMV 8 weeks after transplantation . Dot plots show example of KLRG1 and CD127 expression on gated CD45 . 2+ m45-tetramer+ CD8+ T cells on day 10 post-infection . Upper bar graph depicts the % of CD45 . 2+ m45-tetramer+ CD8+ T cells that are CD127− KLRG1+ ( SLECs ) . Lower bar graph shows the % of CD127+ KLRG1− ( MPECs ) in the same population . *p < 0 . 05 , by Mann–Whitney U-test ( n = 4–7 ) . ( C ) Markers of exhaustion on Atg7−/− MCMV-specific CD8+ T cells on MCMV challenged BM chimera . Dot plots depict example of PD-1 and TIM-3 staining on gated CD45 . 2+ m45-tetramer+ CD8+ T cells . Bar graph quantifies the percentage of ( donor ) CD45 . 2+ m45-tetramer+ CD8+ T cells that are PD-1+ TIM-3+ at day 10 post-infection . *p < 0 . 05 , by Mann–Whitney U-test ( n = 4–7 ) . ( D ) CD127 expression on Atg7−/− MCMV-specific CD8+ T cells in MCMV challenged BM chimeras . Examples of CD127 staining on gated CD45 . 2+ m45-tetramer+ CD8+ T cells from spleen on day 10 post-infection are shown . *p < 0 . 05 , by Mann–Whitney U-test ( n = 4–7 ) . ( E ) IL-15Rα expression on splenic Atg7−/− MCMV-specific CD8+ T cells in MCMV challenged BM chimeras . Histograms depict IL-15Rα expression in CD44lo CD8+ T cells from unimmunized mice ( grey dotted line ) , Atg7+/+ CD45 . 2+ m45-tetramer CD8+ T cells ( grey filled line ) , and Atg7−/− CD45 . 2+ m45-tetramer CD8+ T cells ( black line ) . The left histogram shows expression in control normal WT and T-Atg7−/− mice , the right histogram indicates staining in donor CD45 . 2+ cells from BM chimera mice . Quantified is IL-15Rα mean fluorescence intensity on gated CD45 . 2+ m45-tetramer CD8+ T cells ( n = 4–7 ) . All values are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 00810 . 7554/eLife . 03706 . 009Figure 3—figure supplement 1 . Failure of to form memory CD8+ T cells in T-Atg7−/− mice is not the result of defective IRF4 or EOMES expression . ( A ) IRF4 expression in Atg7−/− and Atg7+/+ splenic m45-specific CD8+ T cells on day 9 , 15 , and 22 post-infection . As a control , IRF4 was also measured in CD44lo CD8+ T cells from unimmunized mice ( naïve ) . Quantification shows IRF4 mean fluorescence intensity from gated m45-tetramer+ CD8+ T cells and CD44lo CD8+ T cells ( naïve ) . Statistics—Student's t test ( n = 4–5 ) . ( B ) EOMES expression in Atg7−/− and Atg7+/+ antigen-specific CD8+ T cells . WT and T-Atg7−/− mice were immunized with MCMV and EOMES expression was measured in m45-specific CD8+ T cells on day 9 , 15 , and 22 . As a control , EOMES was also measured in CD44lo CD8+ T cells from unimmunized mice ( naïve ) . Quantification shows EOMES mean fluorescence intensity on gated m45-tetramer+ CD8+ T cells and CD44lo CD8+ T cells ( naïve ) ( n = 4–5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 009 Finally , we tested whether autophagy compromises effector/memory differentiation and measured two transcription factors typically required during this differentiation . We found that the expression of IRF4 , responsible for sustaining the expansion and differentiation of CD8+ Teff ( Yao et al . , 2013 ) , and EOMES , a factor that promotes CD8+ Tmem formation ( Intlekofer et al . , 2005 ) , behaved normally over time in Atg7−/− CD8+ T cells ( Figure 3—figure supplement 1A and B ) . To understand the mechanism that leads to the failure to maintain a memory compartment in the absence of autophagy , we next checked if Atg7−/− CD8+ T cells were undergoing more cell death than wild-type cells . While this is well described for naïve autophagy deficient T cells ( Puleston and Simon , 2014 ) , it had not been analyzed in antigen-specific T cells . As expected , in wild-type mice challenged with MCMV , we found highest levels of cell death among antigen-specific CD8+ T cells just after the peak of the effector phase on day 9 , whereas in T-Atg7−/− mice , cell death was found to increase over time ( Figure 4A ) . This could not be explained by a change in levels of the anti-apoptotic protein Bcl-2 ( Figure 4—figure supplement 1A ) . The control of mitochondrial quality and reactive oxygen species ( ROS ) via mitophagy , the degradation of mitochondria , has been found to prevent cell death in T cells ( Pua et al . , 2009 ) . In keeping with this , mitochondrial content ( Figure 4B ) and mitochondrial ROS ( Figure 4C ) were significantly increased in Atg7−/− antigen-specific CD8+ T cells . While this likely explains the increased cell death , as shown for other hematopoietic cells ( Mortensen et al . , 2010 ) and T cells ( Pua et al . , 2009 ) , mitophagy is also thought to be essential for maintenance of healthy mitochondrial energy generation . This notion is consistent with our data in other autophagy-deficient hematopoietic cell types ( macrophages and primary leukemic lines , submitted ) demonstrating increased glycolytic enzymes by proteomic analysis , increased lactate production and decreased oxygen consumption using metabolic seahorse measurements . Indeed , studies by Pearce et al showed that formation of the CD8+ Tmem pool is accompanied by a switch to mitochondrial respiration ( Pearce et al . , 2009; Sukumar et al . , 2013 ) . However , in the physiological viral infection models used here , aiming to mimic human infection , the scarcity of antigen-specific CD8+ T cells prevented us from performing these metabolic measurements . In addition , due to the loss of CD8+ Tmem in the absence of Atg7 by day 30 , any metabolic analysis was restricted to early time points when antigen-specific CD8+ T cells are still present in T-Atg7−/− mice . We resorted to the measurement of a surrogate marker of glycolysis , the glucose transporter GLUT-1 on antigen-specific CD8+ T cells by two different techniques , ( a ) the fluorescently labeled GLUT-1 binding domain of HTLV ( human T cell leukemia virus ) , which specifically detects GLUT-1 ( Manel et al . , 2003; Kinet et al . , 2007 ) , and ( b ) a GLUT-1 antibody staining . As expected , GLUT-1 is upregulated on CD8+ Teff ( day 9 ) and then downregulated on CD8+ Tmem ( day 22 ) in wild-type mice as CD8+ T cells switch to mitochondrial respiration ( Figure 4D ) . However , antigen-specific Atg7−/− CD8+ Teff expresses more GLUT-1 and downregulation does not occur to the same extent at the later time points compared to WT cells ( day 22 ) ( Figure 4D ) . This suggests that autophagy supports the metabolic switch during the differentiation from Teff to Tmem . This was confirmed using the GLUT-1 antibody ( Figure 4—figure supplement 1B ) . However , treatment with the anti-diabetic drug metformin that induces mitochondrial β-oxidation metabolism in T cells ( Pearce et al . , 2009 ) did not rescue the memory T cell compartment in this model ( data not shown ) . 10 . 7554/eLife . 03706 . 010Figure 4 . Atg7−/− memory CD8+ T cells show increased mitochondrial content , reactive oxygen species , apoptosis and fail to down-regulate GLUT-1 . ( A ) The spleens of unimmunized and MCMV-infected mice were stained with the apoptotic marker Annexin V and a dead cell dye that stains cells with disrupted membranes on the time points indicated . Apoptotic cells were defined as dead cell dye-negative Annexin V+ . Dot plots are gated on either CD44lo CD8+ T cells ( naïve ) or m45-tetramer+ CD8+ T cells . *p < 0 . 05 , by Mann–Whitney U-test ( n = 4–5 ) . ( B ) Mitochondrial volume by MitoTracker Green in Tetramer+ CD8+ T cells from WT and T-Atg7−/− mice . Spleens from MCMV-immunized mice were stained with MitoTracker Green on day 15 post-infection . Quantification depicts mean fluorescence intensity on m45-tetramer+ CD8+ T cells and is representative of three independent experiments . **p < 0 . 01 , Student t test ( n = 4–5 ) . ( C ) Mitochondrial superoxide production in Tetramer+ CD8+ T cells by MitoSox . Spleens from MCMV-immunized WT and T-Atg7−/− mice were stained with MitoSox Red on day 15 post-infection and analyzed by flow cytometry . Bar graph depicts mean fluorescence intensity on m45-tetramer+ CD8+ T cells and is representative of three independent experiments . **p < 0 . 01 , by Student t test ( n = 4–5 ) . ( D ) GLUT-1 expression . Tetramer+ CD8+ T cells from MCMV-immunized WT and T-Atg7−/− mice were stained with the GFP-tagged HTLV receptor binding domain ( eGFP-HRBD ) , that binds GLUT-1 , at the time points indicated . As a control , GLUT-1 was also measured on CD44lo CD8+ T cells from unimmunized mice ( naïve ) . Bar graph shows the percentage of cells expressing GLUT-1 . *p < 0 . 05 , as determined by Mann–Whitney U-test ( n = 4–5 ) . All values are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 01010 . 7554/eLife . 03706 . 011Figure 4—figure supplement 1 . Normal Bcl-2 levels and altered GLUT-1 expression on antigen-specific Atg7−/− CD8+ T cells . ( A ) MCMV-immunized WT and T-Atg7−/− mice were assessed for Bcl-2 expression in splenic m45-specific CD8+ T cells on day 9 and 22 post-infection . As a control , Bcl-2 was measured in CD44lo CD8+ T cells from unimmunized mice ( naïve ) . Quantification shows Bcl-2 mean fluorescence intensity . Statistics—Student's t test ( n = 4–5 ) . ( B ) GLUT-1 antibody staining on day 9 and day 22 post-infection in MCMV-immunized WT and T-Atg7−/− mice . As a control , GLUT-1 was assessed in CD44lo CD8+ T cells from unimmunized mice ( naïve ) . Bar graphs indicate the frequency of m45-tetramer+ CD8+ T cells and CD44lo CD8+ T cells ( naïve ) that express GLUT-1 . Statistics—Mann Whitney U-test ( n = 4–5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 011 We then tested whether T-Atg7−/− mice were able to mount a recall CD8+ T cell response to a secondary infection . We primed T-Atg7−/− mice with PR8 ( H1N1 ) followed by heterologous challenge with the X31 strain of influenza ( H3N2 ) . By challenging with a heterologous virus strain expressing different surface antigens , it is possible to significantly diminish the influence of antibodies in mediating immunity to secondary infection . Thus , heterotypic immunity relies heavily on cross-reactive CD8+ T cell responses ( Zweerink et al . , 1977; Townsend et al . , 1986 ) , as opposed to homotypic immunity ( for example PR8 primed , PR8 challenged ) to which influenza-specific antibodies contribute ( Townsend et al . , 1986; Epstein and Price , 2010 ) . While the wild-type mice mounted a fast and strong recall response upon X31 challenge in PR8-primed mice ( PR8 + X31 ) , this was drastically diminished in T-Atg7−/− mice in the lungs on day 5 ( Figure 5A , absolute counts in Figure 5—figure supplement 1A ) . Even when three live virus immunizations were administered ( PR8 primed , challenged with X31 then with PR8 ) , T-Atg7−/− mice were unable to mount recall responses . Surprisingly , T-Atg7−/− mice survived the heterotypic viral challenges despite the vastly diminished numbers of secondary CD8+ Teff . 10 . 7554/eLife . 03706 . 012Figure 5 . Atg7−/− memory CD8+ T cells mount a significantly reduced recall response to secondary immunization . ( A ) Recall responses to influenza . Here , WT and T-Atg7−/− mice were split into three groups . In one , mice were immunized with 0 . 00032 HAU PR8 influenza and the NP-specific CD8+ T cell response was measured in lungs on day 24 ( PR8 , n = 4 ) . In another group , mice were immunized with 0 . 00032 HAU PR8 influenza followed by 0 . 32 HAU X31 influenza challenge on day 24 . The recall response to this challenge was measured on day 5 post–challenge using NP-specific tetramers ( PR8+X31 , n = 6 ) . In the third group , mice were immunized with 0 . 00032 HAU PR8 influenza , followed by 0 . 32 HAU X31 on day 24 , and 30 days later immunized for a third time with 32 HAU PR8 . The recall response to this third infection was measured in the lungs on day 5 post–challenge by tetramer ( PR8 + X31 + PR8 , n = 4 ) . Quantification shows frequency of CD8+ T cells that are NP-specific and are representative of two independent experiments . *p < 0 . 05 , **p < 0 . 01 , by Mann–Whitney U-test . ( B ) The Atg7−/− CD8+ T cell response to influenza vaccination . WT and T-Atg7−/− mice were vaccinated twice , 22 days apart with 32 HAU of the live attenuated H1N1 vaccine , S-Flu . 30 days after the last vaccination , mice were challenged with 32 HAU X31 influenza . The CD8+ T cell response to NP was measured in lungs on day 23 post–challenge by tetramer . As a control , unvaccinated mice were challenged with 32 HAU X31 and culled on day 4 due to weight loss and morbidity . Quantification indicates the percentage of CD8+ T cells in the lung that are specific for NP on day 23 post–challenge . Example dot plots are shown . Data are representative of two independent experiments . *p < 0 . 05 by Mann–Whitney U-test ( n = 4–5 ) . ( C ) Atg7−/− CD8+ T cell kinetics to influenza vaccination . WT and T-Atg7−/− mice were vaccinated with S-Flu twice , 22 days apart , and the CD8+ T cell response to NP was tracked over time in blood by tetramer ( n = 4–6 ) . ( D ) Atg7−/− CD8+ T cell kinetics to influenza vaccination and challenge . WT and T-Atg7−/− mice were vaccinated as described in ( C ) . 30 days after the last vaccination , mice were challenged with 32 HAU X31 influenza . Using NP-specific tetramers , the CD8+ T cell response to the vaccine regime and the challenge was tracked in the blood over time ( n = 4–6 ) . All values are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 01210 . 7554/eLife . 03706 . 013Figure 5—figure supplement 1 . Atg7−/− CD8+ T cells fail to mount robust re-call responses to secondary infection . ( A ) Recall CD8+ T cell responses in WT and T-Atg7−/− mice . Mice were immunized once with PR8 influenza ( n = 4 ) ; twice with PR8+X31 influenza ( n = 6 ) ; or three times with PR8 + X31 + PR8 ( n = 4 ) . Absolute counts of CD8+ NP-tetramer+ T cells were determined in the lungs at either day 24 ( PR8 ) , or day 5 post–challenge ( PR8 + X31 & PR8 + X31 + PR8 ) . All values are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 013 Using an established non-replicating pseudotyped influenza vaccination protocol ( Powell et al . , 2012 ) , we next tested whether H1N1 vaccinated T-Atg7−/− mice could mount secondary CD8+ T responses to heterotypic viral challenge . Mice were vaccinated twice , 22 days apart then challenged with X31 influenza . Large numbers of influenza NP-specific CD8+ T cells were detected in the lungs of vaccinated wild-type mice 5 days post-challenge , but not in T-Atg7−/− mice ( Figure 5B ) . When the CD8+ T cell kinetics to the vaccine regimen were followed over time in the blood of T-Atg7−/− mice , influenza-specific CD8+ T cells were detected at normal frequencies to primary immunization but the secondary ‘booster’ vaccine was unable to induce an increase in NP-specific CD8+ T cell frequency like it could in wild-type mice ( Figure 5C ) . Similarly , vaccinated T-Atg7−/− mice failed to generate significant CD8+ T cell responses to viral challenge in the blood ( Figure 5D ) . As before , all mice survived the live viral challenge ( data not shown ) . In summary , these data indicate that autophagy is required for the CD8+ T cell recall response to either repeated immunizations with live virus or vaccination followed by live viral challenge . However , these results were not repeated in a mixed bone marrow chimera setting , making it difficult to exclude that lymphopenia , CD4+ T cell help or CD4-expressing antigen-presenting cells might be responsible for defective recall response in T-Atg7−/− mice . We reasoned that a natural setting where these findings might be applicable is the aged organism , where reduced autophagy levels are found in many organs and cell types ( Rubinsztein et al . , 2011 ) . Indeed , we have previously shown that autophagy levels are significantly reduced in the CD8+ T cells of aged healthy human donors ( >65 years ) as compared to young donors ( <30 years ) ( Phadwal et al . , 2012 ) . Taking results presented here so far into account , we hypothesized that these low levels may contribute to the poor CD8+ Tmem formation and inefficient influenza vaccination observed in the elderly ( Kapasi et al . , 2002; Haynes et al . , 2003; Kedzierska et al . , 2012 ) . We addressed whether diminished CD8+ Tmem formation can be improved in the elderly by modulating autophagy . First , we showed that mRNA levels of essential autophagy genes are decreased in sorted CD8+ T cells from naïve old mice ( 2 years ) as compared to young mice ( 8 weeks ) . The CD8+ CD44hi memory compartment is particularly affected ( Figure 6A ) . CD8+ T cells from old mice also showed significantly decreased autophagic flux detected by counting LC3 spots in NP-specific CD8+ T cells from young and old mice both in the presence and absence of an autophagy flux inhibitor ( Figure 6B , C ) . This was confirmed by using two flow cytometry based autophagy detection , also in NP-specific CD8+ T cells ( Figure 6—figure supplement 1A and B ) . 10 . 7554/eLife . 03706 . 014Figure 6 . Boosting autophagy restores CD8+ T cell responses to vaccination in elderly mice . ( A ) Autophagy gene expression in CD8+ T cells from young and elderly mice . CD44lo and CD44hi CD8+ T cells were purified from 6 week old and 2 year old mice using fluorescent activated cell sorting . mRNA was extracted and the expression of essential autophagy genes was measured by q-PCR . Shown is the fold change in expression in CD8+ T cells from old mice relative to expression in young mice ( normalized to gapdh and hprt ) . ( B ) LC3 Spot count in CD8+ T cells from young and old mice . Splenic CD8+ T cells from 8 week old and 2 year old mice were treated with an autophagy flux inhibitor for 2 hr , as a control cells were left untreated . LC3 spot count was determined on CD8+ CD44hi T cells using ImageStream . Representative images are shown ( ×60 magnification ) . **p = 0 . 0082 , ***p = 0 . 0004 as determined by Student t-test ( n = 4–5 ) . ( C ) Quantification for images shown in ( B ) . ( D ) Human T cell line Jurkat was incubated with 100 µM spermidine for 2 , 4 , or 6 hr or left untreated followed by whole protein extraction for LC3 Western Blot ( upper panel ) . In the lower panel , Jurkat cells were incubated either with no spermidine ( control ) , 100 µM , or 500 µM spermidine for 6 hr and then Western blotted for LC3 . GAPDH was used as loading control for all Western blots . ( E ) Jurkat cells were treated with 1 mM DFMO or 1 mM DFMO with 1 µM spermidine for 48 hr or left untreated ( control ) . In the final 6 hr of incubation , all cells were treated with 10 nM bafilomycin A1 and LC3-I to LC3-II conversion was assessed by Western Blot . ( F ) Jurkat cells were treated with 50 nM rapamycin , 100 µM or 500 µM spermidine for 6 hr followed by detection of phosphorylated S6 ( Ser235/236 ) by Western Blot . As a control , cells were left untreated . ( G ) CD8+ T cell kinetics to influenza vaccination in aged mice in the presence of spermidine . 8-week-old young WT and 23 month old WT and T-Atg7−/− mice were vaccinated 22 days apart with S-Flu . 21 days prior to the first vaccination , aged WT and T-Atg7−/− mice were administered spermidine in the drinking water at a concentration of 5 mM through to the experimental endpoint . As a control , 23 month old WT mice were administered water alone . The CD8+ T cell response to NP was tracked over time in the blood by tetramer ( n = 4–5 ) . Y-axis depicts the frequency of CD8+ T cells that are specific for NP . **p < 0 . 01 by Mann–Whitney U-test . ( H ) CD8+ T cell kinetics to influenza vaccination and challenge in aged mice in the presence of spermidine . 8-week-old young WT and T-Atg7−/− and 23 month old WT and T-Atg7−/− mice were vaccinated as described in ( G ) . 30 days after the last vaccination , mice were challenged with 32 HAU X31 and the CD8+ T cell response to challenge was measured 9 days later in the lungs by tetramer . From 21 days prior to the first vaccination , through to the experimental end point , aged WT and T-Atg7−/− mice were administered spermidine as before . Y-axis indicated the percentage of CD8+ T cells that are specific for NP ( n = 4–5 ) . ( I ) CD8+ T cell response to influenza challenge in vaccinated aged mice in the presence of spermidine . 9 days post–challenge , lungs were harvested and the NP-specific CD8+ T cell response to influenza challenge was measured by tetramer . Example dot plots are gated on CD8+ T cells . Bar chart shows absolute counts of NP-specific CD8+ T cells in the lung per lobe . **p < 0 . 01 , ****p < 0 . 0001 by Student t test ( n = 4–5 ) . All values are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 01410 . 7554/eLife . 03706 . 015Figure 6—figure supplement 1 . Autophagy levels are significantly diminished in antigen-specific CD8+ T cells from aged mice . Autophagy levels by CytoID ( A ) and LC3-II staining ( B ) in antigen-specific CD8+ T cells from young and old mice . 8-week-old and 2 year old mice were immunized with PR8 influenza . On day 10 post-infection lungs were harvested and stained with CytoID or for LC3-II to assess autophagy levels in NP-Tetramer+ CD8+ T cells by flow cytometry . ( A ) ****p < 0 . 0001 ( n = 7 ) . ( B ) **p = 0 . 0056 , ( n = 7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03706 . 015 Next , we chose the autophagy-inducing , naturally occurring compound spermidine to modulate autophagy in vivo , as it had been safely and effectively administered to mice previously ( Eisenberg et al . , 2009 ) . We first confirmed that spermidine induces autophagy in T cells in vitro in a dose and time-dependent manner shown by increased levels of LC3-II ( Figure 6D ) . Spermidine levels in the blood are known to decrease with age ( Pucciarelli et al . , 2012 ) . To mimic this , we used the inhibitor of the natural spermidine synthesis pathway DFMO in the human Jurkat T cell line and found that autophagy levels detected by LC3 western blot significantly dropped in BafA treated samples and untreated samples ( Figure 6E ) . Autophagy levels could be rescued by the addition of very low levels of spermidine in T cells ( Figure 6E ) . As mTOR inhibition , such as achieved with rapamycin , is well known for its multiple side effects ( Lamming et al . , 2013 ) , we confirmed that spermidine induces autophagy in T cells in an mTOR-independent manner . Rapamycin completely abolished the phosphorylation of S6K , a major target downstream of activated mTOR , whereas spermidine had no such effect ( Figure 6F ) . Finally , we administered spermidine to old and young mice via the drinking water at concentrations known to induce autophagy ( Eisenberg et al . , 2009 ) during the influenza vaccination protocols as in Figure 5B–D . In response to first and second vaccination , and as observed in ageing humans , old mice do not mount a robust CD8+ T cell response as compared to young mice . Spermidine dramatically enhanced influenza-specific CD8+ T cell responses in old vaccinated wild-type mice but not in old vaccinated T-Atg7−/− mice as tracked over time in blood ( Figure 6G ) . Similarly , the response to live influenza challenge in the blood was improved twofold to threefold ( back to levels similar to young mice ) when spermidine was administered to old mice , but not in the absence of autophagy ( Figure 6H ) . In the lungs , the number of NP-specific CD8+ T cells responding to influenza challenge in vaccinated old mice were improved fivefold to sixfold in the presence of spermidine , but had little effect on the response of aged T-Atg7−/− mice ( Figure 6I ) .
Hematopoietic and immune health both decline significantly with age ( Beerman et al . , 2010 ) . With the ageing population , this has become a substantial health and socio-economic problem for the developed world . Our work sheds light on the cellular mechanisms of immune ageing and how to improve responses to vaccination in the elderly . The naïve T cell compartment in the absence of several different essential autophagy genes has been extensively analysed ( Puleston and Simon , 2014 ) . These models uniformly demonstrated increased damaged mitochondria leading to cell death and peripheral lymphopenia . In this study , we show how this lymphopenia drives the remaining T cells to take on an activated/memory phenotype as they proliferate in an attempt to re-fill the T cell niche , often termed ‘virtual memory’ ( Jia and He , 2011 ) . Thus , we provide a link between lymphopenia and the formation of the ‘virtual memory’ compartment and highlight the significant changes in T cell surface phenotype that can occur in a lymphopenic environment . Interestingly , the virtual memory compartment is also a hallmark of the ageing adaptive immune system . This observed phenotype in naïve mice would have precluded a coherent analysis of antigen-activated T cells; however , the use of mixed BM chimeras crucially excluded a contribution of the skewed naïve repertoire to the CD8+ Tmem phenotype . In this study , we showed a novel and essential link for autophagy in the formation of CD8+ Tmem to infection . The absence of autophagy also dramatically altered the ability of CD8+ T cells to respond to secondary infection . Interestingly , the severely depleted CD8+ T cell recall response did not prevent T-Atg7−/− mice from surviving a lethal secondary heterotypic influenza challenge . Influenza virus is typically controlled by both antibodies and cellular immunity . However , in a heterotypic challenge , the surface hemagglutinin and neuraminidase ( recognized by antibodies ) differ between the two heterotypic strains , while the viral internal proteins ( presented by MHC class I to CD8+ T cells ) remain the same , leaving the CD8+ T cells to provide protection ( Rimmelzwaan et al . , 2007 ) . The unique protective role of CD8+ T cells in heterotypic immunity was described several decades ago in mice ( Rimmelzwaan et al . , 2007 ) . The observation that the majority of CD8+ T cell epitopes are cross-reactive between subtypes and are located in the most conserved regions of the internal proteins ( nucleoprotein and matrix protein ) confirms their role . In humans , the evidence , typical for human studies , is sparse and more circumstantial . A study from 1983 by McMichael et al demonstrated that in experimentally infected individuals , virus-specific cytotoxicity inversely correlated with the extent of virus shedding in the absence of virus-specific antibodies for the strain that was used for infection ( McMichael et al . , 1983 ) . More recently , Lalvani et al found higher frequencies of pre-existing T cells to conserved epitopes in individuals who developed less severe illness in the 2009 H1N1 flu pandemic in the absence of cross-protective antibodies ( Sridhar et al . , 2013 ) . We would postulate therefore that the survival of the T-Atg7−/− mice after the secondary heterotypic challenge relies on the very few effector CD8+ T cells generated . These data also suggest that eliciting CD8+ T cell memory responses in influenza vaccination is highly desirable , particularly to protect from pandemic flu . This also raises interesting questions as to why such robust secondary T cell responses have evolved when survival can be achieved with a fraction of what is observed normally in nature . Although no CD8+ Tmem cells can be detected by conventional tetramer technology 20–30 days post-immunization in T-Atg7−/− mice , some memory cells must remain below the detection limit at very low frequency . In addition , these Atg7−/− CD8+ T cells must maintain normal functional and proliferative capacity as they are detected at early time points after secondary immunization , characteristic of true memory T cell kinetics . To explain how autophagy affects the maintenance of CD8+ Tmem cells , we addressed two alternative hypotheses experimentally: ( 1 ) accumulation of cellular damage affects the long-lived CD8+ Tmem in particular and ( 2 ) autophagy is required for the switch from glycolysis to mitochondrial respiration important for the survival of CD8+ Tmem cells . Regarding the first one , we show evidence for the accumulation of mitochondria and ROS in the absence of autophagy . However , a limitation of our study is , we did not test for accumulation of other organelles and protein aggregates . As for the second hypothesis , initial studies involved TRAF6 ( TNFR-associated factor 6 ) in the regulation of CD8+ Tmem development by modulating fatty acid metabolism and thereby increasing mitochondrial respiration ( Pearce et al . , 2009; van der Windt et al . , 2012 ) . Interestingly , TRAF6 , as an E3 ubiquitin ligase , has recently been shown to stabilize essential proteins in the autophagy pathway such as Ulk1 ( Nazio et al . , 2013 ) and Beclin-1 ( Shi and Kehrl , 2010 ) . We therefore hypothesize that TRAF6 controls metabolism via autophagy and propose three mechanisms contributing to autophagy's role in mitochondrial respiration: ( 1 ) through lipophagy , the degradation of lipids ( Weidberg et al . , 2009 ) , autophagy provides the fatty acids that fuel the Krebs cycle with Acetyl-CoA though β-oxidation , ( 2 ) through degradation of glycolytic enzymes ( Xiong et al . , 2011 ) , and ( 3 ) through mitophagy , autophagy provides mitochondrial quality control that is required for functional mitochondrial ATP generation . We provide evidence for the latter mechanism here . This suggests that autophagy is important in the control of metabolic pathways in T cells . Interestingly both the accumulation of damaged mitochondria/oxidative stress as well as a bias to preferentially use glycolysis are hallmarks of the ageing cell ( Hipkiss , 2006 ) . In this study , we showed that autophagy levels are impaired in CD8+ T cells from aged mice . Using spermidine , we could dramatically improve the CD8+ T cell response to vaccination and infection in elderly mice in an autophagy-dependent manner . However , aged CD4+ T cells also undergo immune senescence . While inducing autophagy with spermidine in the CD8+ T cell compartment clearly contributes to the increased numbers of antigen-specific CD8+ T cells , the limitation of these experiments is that we cannot gauge the contribution of improved CD4+ T cell function to CD8+ T cell numbers . This will be included in future studies . These results offer the first cell-intrinsic explanation as to why T cell memory is defective in old age . Spermidine is a pleiotropic compound and its major function has been described in yeast as an inhibitor of histone acetyl transferases , inducing epigenetic changes to autophagy-related gene expression ( Eisenberg et al . , 2009 ) . We showed that spermidine operates independently of mTOR to induce autophagy . These findings offer the prospect of improving vaccine T cell responses in the elderly through spermidine in an mTOR-independent fashion . However , no other ways of inducing autophagy were tested here . We expect that fasting or other mTOR-independent drugs such as resveratrol ( Morselli et al . , 2009; Morselli et al . , 2010 ) may also be beneficial for CD8+ Tmem maintenance . However , spermidine is significantly more attractive than previously published Tmem-boosting compounds such as metformin ( Pearce et al . , 2009 ) and rapamycin ( Araki et al . , 2009 ) , that come with unwelcome side effects and toxicity in humans . The elucidation of the precise role of spermidine will be the subject of future work and will enable its development as a safe , readily administrable immune-modulating drug .
CD4-Cre mice ( Lee et al . , 2001 ) were from Adeline Hajjar ( University of Washington ) and crossed with Atg7flox/flox ( Komatsu et al . , 2005 ) to obtain T-Atg7−/− mice ( CD4-Cre+ Atg7+/+ ) on a C57BL/6 background . Unless otherwise stated , all mice were 6–12 weeks of age at the start of each experiment and were age and sex matched . CD4-Cre− Atg7+/+ mice were used as wild-type controls and were littermates where possible . No phenotype was observed in CD4-Cre+ Atg7+/− mice . Old mice and young control mice were purchased from Charles River , UK . C57BL/6 SJL CD45 . 1 mice for bone marrow chimeras were purchased from Biomedical Services , Oxford . All mice were housed in Biomedical Services , Oxford and animal experiments were approved by the local ethical review committee and performed under UK project license ( PPL 39/2809 ) . The following antibodies were used for flow cytometry ( antibody clone in brackets ) : CD8 ( Ly2 ) PE/PE-CY7; CD8 ( 53-6 . 7 ) FITC/eF450/PE-Cy7; CD4 ( GK1 . 5 ) PE/FITC/APC; TCRβ ( H57-597 ) FITC/PE/PE-Cy7/APC; CD3 ( 145-2C11 ) eF450/APC; CD62L ( MEL-14 ) FITC/PE-Cy7; CD44 ( IM7 ) FITC/PE-Cy7; KLRG1 ( 2F1 ) FITC/PE-Cy7; IRF4 ( 3E4 ) FITC; EOMES ( Dan11mag ) PE-Cy7; CD127 ( A7R34 ) FITC/PE/eF450; Ki-67 ( SolA15 ) eF450; CD24 ( M1/69 ) APC; CD16/32 ( 93 ) purified Fc block; CD45 . 2 ( 104 ) eF450; all from eBioscience ( San Diego , CA , USA ) . PD-1 ( 29F . 1A12 ) PE-Cy7; PD-1 ( RMP1-30 ) PE; CD45 . 2 ( 104 ) PE-Cy7; TIM-3 ( B8 . 2C12 ) PE/APC; Bcl-2 ( BCL/10C4 ) PE; all from Biolegend ( San Diego , CA , USA ) . Glut1 ( q ) PerCP; IL-15Rα ( BAF551 ) biotinylated; both from R&D Systems ( Minneapolis , USA ) . For surface staining , cells were suspended in PBS , 2% FCS , 5 mM EDTA and stained at 4°C . Anti-CD16/32 ( Fc Block , eBioscience ) was generally added to antibody mix to minimize non-specific staining . LIVE/DEAD Fixable Violet Dead Cell Stain Kit ( Life Technologies ) was used prior to surface staining to exclude dead cells . For cytoplasmic intracellular staining ( Bcl-2 ) , cells were stained for surface antibodies then fixed in IC fixation buffer before permeabilisation with perm buffer ( eBioscience ) . Cells were then suspended in perm buffer at RT for antibody staining . For nuclear targets ( IRF , EOMES , Ki-67 ) , after surface staining cells were fixed and permeabilised with FoxP3 Fixation/Permeabilisation kit ( eBioscience ) before resuspension in perm buffer for intracellular staining at RT . Absolute cell counts were performed on peripheral blood taken from the lateral tail vein of live animals , collected in heparin-coated tubes ( Microvette 300 , Sarstedt [Nümbrecht , Germany] ) to avoid coagulation . Cell counts were calculated with BD TruCount tubes ( BD Bioscience , NJ , USA ) according to the manufacturer's instructions . Tetramers were generated as previously described ( Altman et al . , 1996 ) and MHC-Class I monomers stored at −80°C . Biotinylated monomers were tetramerized with Streptavidin PE or APC at the right concentration to achieve 1:1 ratio with biotin binding sites and added in 1/10th volumes waiting 10 min between additions . Tetramerized complexes were stored at 4°C . Peptide sequences for MCMV tetramers were as follows: m45 985HGIRNASFI993 , H-2Db-restricted; m38 316SSPPMFRV323 , H-2Kb-restricted; IE3 416RALEYKNL423 , H-2Kb-restricted . The peptide sequence for the influenza tetramer was as follows: NP 366ASNENMETM374 , H-2Db-restricted . Cells were always stained with tetramers prior to surface antibody staining in PBS , 2% FCS at 37°C for 15 min . Tracking tetramer-positive cells over time in the blood was performed by serially bleeding the lateral tail vein . After erythrocyte lysis with red cell lysis buffer ( eBioscience ) , remaining white cells were stained with tetramer and surface antibodies as described above . Autophagy detection by flow cytometry was measured using the CytoID Autophagy Detection Kit ( Enzo Life Science , Exeter , UK ) . Splenocytes and cells isolated from the lung were first stained with CytoID according to the manufacturer's instructions , prior to tetramer and surface antibody staining . Autophagy was also measured using flow cytometry by quantifying LC3-II mean fluorescence intensity using the FlowCellect Autophagy LC3 Antibody-based Assay Kit ( FCCH100171 , Merk-Millipore , MA , USA ) according to the manufacture's instructions and always following cell surface antibody staining . Use of this kit includes a step where cytosolic LC3-I is washed from the cell , leaving only membrane bound LC3-II prior to staining . For apoptosis detection , splenocytes were stained with m45-tetramer and surface antibody as described above and then stained with LIVE/DEAD Fixable Violet Dead Cell Stain Kit . Cells were finally stained with Annexin V PE-Cy7 ( eBioscience ) in Annexin V binding buffer at room temperature . Apoptotic cells were determined as LIVE/DEAD cell dye negative ( live cells ) , Annexin V positive . GLUT-1 was measured through binding to its ligand , the receptor binding domain ( RBD ) of a recombinant glycoprotein from the human T lymphotrophic virus ( HTLV ) fused to eGFP ( HRBD-eGFP ) ( Montel-Hagen et al . , 2008 ) . For mitochondrial mass analysis , cells were stained with MitoTracker Green ( Life Technologies , Carlsbad , CA , USA ) at 150 nM in PBS , 2% FCS for 30 min at 37°C after tetramer and surface antibody staining . To measure mitochondrial superoxide production , MitoSOX Red ( Life Technologies ) was used at final concentration of 5 µM from a 5 mM stock in PBS , 2% FCS for 30 min at 37°C following tetramer and cell surface staining . All flow cytometry experiments were performed on a Cyan flow cytometer ( Beckman Coulter , Brea , CA , USA ) . BM cells were extracted from a single 8-week old wild-type , T-Atg7−/− ( both CD45 . 2+ ) , and C57BL/6 SJL mouse expressing CD45 . 1 . After erythrocyte lysis , 3 × 106 wild-type or T-Atg7−/− BM cells were added to 3 × 106 CD45 . 1+ BM cells ( 1:1 CD45 . 2+:CD45 . 1+ ) in a total volume of 200 µl PBS . The 1:1 BM mix was injected i . v 2 hr after lethal irradiation ( 450 cGy twice , 4 hr apart ) to C57BL/6 SJL CD45 . 1+ recipients ( total of 6 × 106 BM cells per mouse in 200 µl ) . After 8 weeks of reconstitution , mixed BM chimera was immunized with MCMV or PR8 influenza , or left unimmunized . WT and T-Atg7−/− mice were always used as controls . Mice were immunized i . v . with 1 × 106 p . f . u MCMV in 100 µl ( Smith strain ATCC: VR194 ) . Unimmunized controls were injected with 100 µl PBS i . v . For influenza infections , mice were administered intra-nasally with either A/PR/8/34 ( PR8 , H1N1 Cambridge ) influenza or X31 ( H3N2 ) influenza at the stated dose in 50 µl viral dilution media ( VDM; DMEM , 0 . 1% BSA , 10 mM HEPES , 100 U/ml penicillin , 100 µg/ml streptomycin , and 2 mM glutamine; all from Sigma , St Louis , MO , USA ) . Mice were anaesthetized with isofluorane and droplets of VDM-containing virus were applied to the nares until the total 50 µl was inhaled . Unimmunized mice were always used as controls ( 50 µl of VDM alone ) . For influenza vaccination , mice were immunized intra-nasally with 32 HAU pseudotyped H1N1 influenza ( S-Flu ) ( Powell et al . , 2012 ) , using the influenza infection protocol described above , and challenged with 32 HAU X31 . Unvaccinated and fully naïve mice were used as controls in all influenza vaccine experiments . In unvaccinated mice that received X31 challenge , a combination of weight loss and clinical score was used as a humane endpoint . Cells were purified with a MoFlo cell sorter ( Beckman Coulter ) by their surface markers . RNA was extracted using RNeasy Kit ( Qiagen , Hilden , Germany ) and quantified using a Nanodrop spectrophotometer ( Thermo Scientific , Waltham , MA , USA ) . RNA was reverse transcribed ( RT ) using a High Capacity RNA to cDNA kit ( Applied Biosystems ( AB ) , Foster City , CA , USA ) . Resulting cDNA was stored at −20°C . Real-time quantitative PCR using comparative Ct method ( ΔΔCt ) was utilized to evaluate gene expression using validated TaqMan probes ( AB ) on a 7500 Fast Real-time PCR machine ( AB ) . Conditions: ( 1 ) 50°C , 2 min; ( 2 ) 95°C , 10 min; ( 3 ) 95°C 15 s; ( 4 ) 60°C 1 min; 40 cycles of 3–4 . The assay IDs for the primers of the analyzed genes are as follows: Mm00504340_m1 ( Atg5 ) , Mm00512209_m1 ( Atg7 ) , Mm01264428_m1 ( Atg9 ) , Mm00470550_m1 ( Atg10 ) , Mm00503201_m1 ( Atg12 ) , Mm0051717_m1 ( beclin1 ) , Mm00458724_m1 ( Map1lc3a ) , Mm01545399_m1 ( hprt ) , Mm99999915_g1 ( gapdh ) . ImageStream ( Amnis imaging flow cytometer , MA , USA ) has previously been used to determine autophagic flux ( Phadwal et al . , 2012 ) . To determine LC3 spot count , we stained cells for LC3-II ( following cell surface staining ) using the FlowCellect Autophagy LC3 Antibody-based Assay Kit ( FCCH100171 , Merk-Millipore ) according to the manufacture's instructions . Before LC3 detection , to assess autophagic flux , cells were incubated for 2 hr with the autophagy flux inhibitor provided by the kit specified and according to the manufacturer's instructions . LC3 spot count was quantified using Ideas software ( Amnis ) , which contains a specialised objective spot counting feature . Images collected were all at ×60 magnification . 3 × 106 Jurkat cells ( human CD4+ T cell line ) or 5 × 106 purified primary CD8+ T cells were lysed on ice using 100 μl 1 × NP-40 Lysis Buffer . Protein concentration in supernatant was measured using BCA Protein Assay Kit ( PI-23227 , Thermo Scientific ) , and reducing Laemmli Sample Buffer was added to make protein samples for SDS-PAGE . Proteins of 30–50 μg per lane were separated on 16% SDS-PAGE and transferred to PVDF membrane ( IPFL00010 , Millipore , MA , USA ) . After blocking with 5% skim milk , the membrane was blotted using the following primary antibodies: LC3 ( L8918 , Sigma ) ( 1:1000 ) , GAPDH ( MAB374 , Millipore ) ( 1:10 , 000 ) , pS6 ( 2211 , Cell Signaling Technology , Danvers , MA , USA ) ( 1:5000 ) . For S6 blotting , membranes were blotted with anti-pS6 first , then stripped for re-blot using primary antibody against S6 ( 2217 , Cell Signaling Technologies ) ( 1:1000 ) . IRDye secondary antibodies were bought from LI-COR ( Lincoln , NE , USA ) : ( 926-32 , 211 , 926-68 , 020 ) ( 1:15 , 000 ) . For the purification of primary CD8+ T cells , cells were negatively selected from splenocytes using the CD8+ T cell Isolation kit ( 130-104-075 , Miltenyi , Bergisch Gladbach , Germany ) according to the manufacturer's instructions . Determination of the 50% tissue culture infective dose ( TCID50 ) was performed on MDCK-SIAT1 cells in a 96-well flat-bottom plate performed by serial dilution of lung homogenates onto 3 × 104 MDCK-SIAT1 cells followed by incubation for 72 hr at 37°C . Virus was then detected by hemagglutination , where 50 µl 1% ( vol/vol ) human erythrocytes ( adjusted so that a 1:2 dilution gave an optical density at 600 nm ) was added to 50 µl of the serially diluted lung homogenates/MDCK-SIAT1 supernatant in a 96-well V-bottom plate . Hemagglutination was analyzed by the loss of teardrop formation after tilting the plate . TCID50 was calculated as described by Reed and Muench ( 1938 ) . All data were presented as mean ± s . e . m . p values were determined using either a Mann–Whitney U-test or two-tailed Student's t test using GraphPad Prism software . Significant statistical differences are indicated in the figure legends or on the graphs themselves . Mice were administered spermidine in the drinking water 21 days prior to immunization through to the experimental endpoint . Spermidine was added at a concentration of 5 mM from a 1 M aqueous stock . Spermidine-containing drinking water was replenished every 2–3 days . | In the face of an infection , the immune system mounts an aggressive response by producing many copies of killer immune cells called CD8+ T cells that recognize and destroy any cells infected with the invading pathogen . The number of killer cells produced depends on the extent of the infection . Once the infection has been brought under control , most of the CD8+ T cells die off . The small numbers that are retained—called memory cells—‘remember’ the pathogen , so that if it invades the body again , they can help the immune system to respond more quickly and effectively . Memory cells are also critical to the effectiveness of vaccines , many of which introduce a dead or weakened pathogen into the body . This does not cause an infection , but does allow the immune system to create memory cells that are able to fend off the same pathogen in the future . However , vaccines only work in individuals that are able to produce and maintain memory cells , which many older people are less able to do . An important system that maintains cells , called autophagy , destroys and removes the ‘junk’ and toxic by products that all cells accumulate over time as a result of normal cell functions . Without autophagy , cells become less able to produce energy and they may die . Puleston et al . show that autophagy begins to fail in old mice , which prevents the formation of a proper memory response . In addition , mice that lack an important gene needed for autophagy are unable to produce memory cells after being infected with viruses such as influenza . Puleston et al . found that boosting autophagy in older mice using a chemical called spermidine—which is also found naturally in many tissues—helped to restore the mice's ability to create and maintain memory cells . Spermidine-treated mice developed a stronger immunity to influenza after vaccination compared with other mice of a similar age . Further research is required to better understand how spermidine works to see if it could be developed into a drug that safely boosts the immune system of humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"immunology",
"and",
"inflammation"
] | 2014 | Autophagy is a critical regulator of memory CD8+ T cell formation |
DNA methylation , especially CpG methylation at promoter regions , has been generally considered as a potent epigenetic modification that prohibits transcription factor ( TF ) recruitment , resulting in transcription suppression . Here , we used a protein microarray-based approach to systematically survey the entire human TF family and found numerous purified TFs with methylated CpG ( mCpG ) -dependent DNA-binding activities . Interestingly , some TFs exhibit specific binding activity to methylated and unmethylated DNA motifs of distinct sequences . To elucidate the underlying mechanism , we focused on Kruppel-like factor 4 ( KLF4 ) , and decoupled its mCpG- and CpG-binding activities via site-directed mutagenesis . Furthermore , KLF4 binds specific methylated or unmethylated motifs in human embryonic stem cells in vivo . Our study suggests that mCpG-dependent TF binding activity is a widespread phenomenon and provides a new framework to understand the role and mechanism of TFs in epigenetic regulation of gene transcription .
DNA methylation is an ancient and major epigenetic modification that plays an important role in key biological processes , including genomic imprinting , X-chromosome inactivation , suppression of transposable elements , and carcinogenesis ( Jaenisch and Bird , 2003; Egger et al . , 2004; Robertson , 2005; Feinberg , 2007; Reik , 2007 ) . In higher eukaryotes , methylation of CpG sites , especially at promoter regions , is generally considered as the hallmark of gene silencing ( Baylin , 2005 ) . The molecular consequence of CpG methylation is generally believed to disrupt TF–DNA interactions either directly ( Nan et al . , 1998 ) , or indirectly by recruiting sequence-independent methylated DNA-binding proteins that occupy the methylated promoters and compete for the TF binding sites ( Boyes and Bird , 1991 ) . So far , only MeCP2 , MBD1 , MBD2 , and a few zinc finger proteins have been identified as bona fide methylated DNA-binding proteins ( Lewis et al . , 1992; Meehan et al . , 1989; Filion et al . , 2006; Bartke et al . , 2010; Bartels et al . , 2011; Quenneville et al . , 2011; Spruijt et al . , 2013 ) . It is unclear whether the methylated DNA binding activity is widespread among different TF subfamilies . Furthermore , the transcriptional regulatory activity of these methylation-dependent TF–DNA interactions has not been explored . Finally , the structural basis of these methylation-dependent TF–DNA interactions remains elusive . Here we employed a protein microarray-based approach to characterize the entire human TF repertoire for their direct binding capacity to a large set of methylated DNA motifs . In this work , we simultaneously tested the methylated DNA-binding activity for the majority of human TFs . Since we examined each methylated DNA motif individually on the protein microarray , we were able to determine the specific sequences surrounding the methylated CpG that is recognized by TFs . Furthermore , we demonstrated that the methylated- and unmethylated-DNA binding activities of KLF4 could be decoupled by protein mutagenesis .
We selected a total of 154 DNA motifs ( Supplementary file 1A ) with the following characteristics: ( 1 ) predicted to be potential TF-binding sites in promoter regions of the human genome ( Xie et al . , 2005; Yu et al . , 2006; Elemento et al . , 2007 ) ; ( 2 ) representative of a subset of the 460 DNA motifs that have been examined for protein-binding activity in our previous study ( Hu et al . , 2009; Xie et al . , 2010 ) ; and ( 3 ) carrying at least one CpG site . Because our goal was to identify DNA methylation-dependent binding activity , we developed a competition assay on a protein microarray to identify TFs that prefer DNA motifs carrying mCpGs ( Figure 1A ) . Each synthesized double-stranded DNA motif was end-labeled with Cy5 ( Hu et al . , 2009 ) and converted to the double-stranded mCpG form using a bacterial CpG DNA methylase SssI . Each methylated motif was incubated with the protein microarray in the presence of its unlabeled ( cold ) , unmethylated counterpart in 10-fold excess . The competition assay was first tested on a pilot array , comprised of MBD1 , MBD2 , MeCP2 , and several negative control proteins . As illustrated in Figure 1B , strong binding signals were observed in a protein concentration-dependent manner , while control proteins , such as GST and BSA , did not show any detectable signals , confirming the reliability of the competition assay . 10 . 7554/eLife . 00726 . 003Figure 1 . Protein microarray-based approach identified mCpG-dependent DNA-binding activity among human TFs and cofactors . ( A ) A competition assay was used to identify proteins that preferentially bind to methylated DNA motifs . SCAPER ( S-phase cyclin A-associated protein in the ER ) and E2F3 ( E2F transcription factor 3 ) were shown here as two examples of methylated DNA-binding proteins . ( B ) A proof-of-principle assay was conducted using known methylated DNA-binding proteins on a pilot protein microarray . ( C ) Binding profiles of 41 TFs and 6 co-factors against 150 of the 154 tested methylated DNA motifs are summarized in the interaction map . TFs are color-coded based on the subfamilies . ( D ) EMSA assays validated DNA-binding activity for four selected TF candidates . Representative images from three independent experiments with similar results are shown . ( E ) Competition EMSA assays confirmed mCpG-dependent DNA-binding activities . As expected , 10-fold unlabeled , methylated DNA motif readily abolished the protein–DNA complex formation of the tested TFs with the biotinylated and methylated DNA motifs ( Lane 1 in each image ) . However , 10-fold cold unmethylated DNA counterparts could not compete off methylated DNA binding , consistent with the protein microarray results . ( F ) HOXA5 and DIDO1 showed mCpG-dependent activation of luciferase activity in GT1-7 cells . Values represent mean ± SD ( n = 3; **: p<0 . 01; t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 00310 . 7554/eLife . 00726 . 004Figure 1—figure supplement 1 . Data analysis of the protein microarray assays . ( A ) Workflow of data normalization . ( B ) Local normalization ( window size 9 × 9 ) . ( C ) Extrapolation of background noise distribution . Noise distribution of N2 is mirrored from distribution of N1 . Standard deviation ( SD ) was calculated based on distribution N ( Lower panel ) . ( D ) Distribution of Z scores of all proteins on a microarray . Z = 3 was selected as the cutoff in our study to identify the positives . The Z scores of some methylated DNA-binding proteins identified previously ( Spruijt et al . , 2013 ) are below our cutoff , while KLF4's Z score for one DNA motif is >6 . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 00410 . 7554/eLife . 00726 . 005Figure 1—figure supplement 2 . Reproducibility of protein microarray data . Left panel: signal comparison between a duplicated binding-assay with motif M303 shows a high correlation , confirming the reproducibility of the assay . Right panel: comparison between two random binding assays exhibited non-correlation between motifs M303 and M259 . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 00510 . 7554/eLife . 00726 . 006Figure 1—figure supplement 3 . Distribution of number of mCpG-binding TFs/co-factors in a given motif-bind assay . The median value of TFs/cofactors binding to one methylated CpG-containing motif is 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 00610 . 7554/eLife . 00726 . 007Figure 1—figure supplement 4 . Distribution of number of methylated motifs recognized by a given TF/co-factor . Most TFs/cofactors bind to very few methylated DNA motif ( s ) ; whereas 7 TFs bind to more than 77 of the 154 motifs tested in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 00710 . 7554/eLife . 00726 . 008Figure 1—figure supplement 5 . Distribution of TF subfamily members . ( A ) Distribution of TF subfamily members that showed mCpG-binding activity . ( B ) Distribution of all annotated TF subfamily members presented on the TF protein microarrays . Statistic analysis showed that none of the TF subfamilies was significantly enriched ( p<0 . 01 ) for methylated motif-binding activities . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 00810 . 7554/eLife . 00726 . 009Figure 1—figure supplement 6 . Four additional EMSA assays ( A ) and competition EMSA assays ( B ) . The results confirmed specificity of mCpG-dependent DNA-binding activities . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 00910 . 7554/eLife . 00726 . 010Figure 1—figure supplement 7 . Methylation level of the KLF4 and HOXA5 luciferase reporter constructs . Eight units of KLF4 ( TCCCGCCCA ) and HOXA5 ( AAACGCTGCC ) binding motifs were separately cloned into the promoter region of a CpG-free luciferase reporter vector , and methylated with SssI before transfected into GT1-7 cells . Bisulfite sequencing confirmed that the CpG methylation levels of both motifs reached ∼100% after SssI treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 01010 . 7554/eLife . 00726 . 011Figure 1—figure supplement 8 . Number of unique mCpG-binding TFs/co-factors in function of number of tested methylated DNA motifs . The curve is far from saturation , suggesting that more such TFs/co-factors remain to be discovered . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 011 We then probed for direct interactions between 154 methylated motifs and the human TF repertoire using the TF protein microarray , which includes 1321 TF and 210 co-factor proteins ( Supplementary file 1B ) . These full-length human proteins were overexpressed and individually purified in yeast as N-terminal GST fusion proteins as previously reported ( Hu et al . , 2009 ) . The quantity and purity of the purified proteins were examined with Coomassie and silver stains against a random set of purified proteins ( data not shown ) . Each of the labeled 154 methylated motifs was separately incubated on the human TF protein microarray in the presence of the unlabeled counterparts in 10-fold excess . Each reaction was performed in duplicate to ensure reproducibility . After a washing step , binding signals were acquired and processed using the GenePix software ( Figure 1—figure supplement 1 ) . After the binding signals were further normalized by correcting local background signals with a 9 × 9 sub-grid , Z-scores of each protein on the array were determined as previously described ( Hu et al . , 2009 ) . A Z-score of 3 was used as the cutoff to identify the positives . Correlation analysis of the binding signals showed high reproducibility between each duplicated assay ( Figure 1—figure supplement 2 ) . Of the 154 methylated motifs examined , 150 ( 97% ) showed significant binding signals to at least one protein on the microarrays ( median = 8 ) , suggesting binding to methylated DNA is prevalent , at least in this in vitro competition assay ( see ‘Materials and methods’ , Figure 1—figure supplement 3 ) . In addition , 41 TFs and 6 TF co-factors ( 3% of all factors examined ) showed mCpG-dependent binding activity ( Figure 1C ) . Among these , 15% showed a broad binding activity to over 50% of all methylated motifs tested , suggesting that these proteins are not sensitive to the sequence context surrounding the mCpG site . For example , two zinc finger TFs , ZNF114 and ZNF416 , could recognize almost all of the motifs tested in the competition assay . However , the majority of the 47 mCpG-binding proteins showed mCpG- and sequence-dependent binding activity , with 22 TFs exhibiting binding signaling to fewer than three motifs ( Figure 1—figure supplement 4 ) . Notably , the mCpG binding activity is widespread among various TF subfamilies , such as the zf-C2H2 , Homeobox , bHLH , Forkhead , bZIP , and HMG box subfamilies ( Figure 1C ) ; no TF subfamily was found significantly enriched in the hit list ( Figure 1—figure supplement 5 ) . On the other hand , consistent with the important role of DNA methylation in the development of cancer ( Egger et al . , 2004; Reik , 2007 ) , there was a significant enrichment for known oncogenes and tumor suppressors ( 10/47; p<0 . 01; hypergeometric model ) and factors involved in tissue development ( 25/47; p<0 . 015; hypergeometric model; Figure 1C ) . Together , these results suggest that methylation-dependent direct TF–DNA interaction is a widespread phenomenon among various TF subfamilies in humans . To validate the protein microarray results , we selected 11 TFs with different mCpG-dependent DNA-binding behaviors ( i . e . , sequence-dependent or -independent ) . Among them , six ( e . g . , ARNT2 ) bound to fewer than three methylated motifs , while the others ( e . g . , HOXA5 ) bound to at least 66 motifs ( Supplementary file 1C ) . Using an electrophoretic mobility shift assay ( EMSA ) , we confirmed that 8 of the 11 tested TFs could readily form a protein–DNA complex with their corresponding methylated motifs , but not with the unmethylated form ( Figure 1D; Figure 1—figure supplement 6 ) . This result indicates either a 27% of false positive rate of the protein microarray assay or a higher sensitivity of protein microarray than EMSA . While the binding activities to the mCpG-carrying motifs could be readily competed off with the unlabeled mCpG-containing analogs in 10-fold excess , the unmethylated counterparts showed no obvious impact , indicating that the complex formation between these TFs and specific DNA motifs is not non-specific but requires CpG methylation ( Figure 1E ) . To determine whether mCpG-dependent TF–DNA interactions could convey transcriptional activity in vivo , we selected HOXA5 and DIDO1 to perform cell-based luciferase assays . Motifs M305 and M24 recognized by HOXA5 and DIDO1 , respectively , were separately cloned into the promoter region of a CpG-free luciferase construct ( pCpGL ) ( Klug and Rehli , 2006 ) , which was then methylated with SssI and transfected with the HOXA5 and DIDO1 expression constructs at different concentrations into a mammalian cell line . The complete methylation of the motifs before transfection was confirmed using Sanger-bisulfite sequencing ( Figure 1—figure supplement 7 ) . Unmethylated luciferase construct was used as a negative control . Comparing with the luciferase activity of the negative controls , we found that both HOXA5 and DIDO1 demonstrated dose- and methylation-dependent enhancement of the luciferase activity ( Figure 1F ) , indicating that the two TFs could specifically bind to the methylated promoters and enhance the downstream gene transcription . We then asked whether these mCpG-binding TFs also bound to the same consensus motifs in an unmethylated form . Since our previous genome-wide analysis of human TF–DNA interactions included these 150 unmodified CpG-containing motifs ( Hu et al . , 2009 ) , we integrated both protein microarray datasets . We discovered that 17 TFs recognize both methylated and unmethylated motifs ( Figure 2A ) . Of the 435 protein–DNA interactions ( PDIs ) examined , only a small fraction ( 4% ) exhibited motif-binding activity irrespective of the CpG methylation status ( Figure 2A; yellow bars ) . Even in these cases , TF binding to methylated forms is much stronger than to the unmethylated forms based on the inability of non-methylated analogs to compete effectively even when present in 10-fold excess . Interestingly , 321 ( 74% ) and 97 ( 22% ) PDIs exhibit specificity to methylated ( Figure 2A; red bars ) and unmethylated motifs ( Figure 2A; blue bars ) , respectively . To evaluate sequence specificity for the newly discovered methylation-dependent PDIs , we determined the consensus methylated motif sequences for the 17 TFs . The comparison with their known consensus sequences indicated that , in most cases , the sequences of methylated and unmethylated motifs recognized by the same TFs are distinct ( Figure 2A; right panel; ‘Materials and methods’ ) . For example , nuclear factor I/C ( NFIC ) binds to a methylated motif M209 , GTmCpGCC , while its known consensus sequence is TTGGC , suggesting that many TFs are of dual-specificity , recognizing methylated and unmethylated DNA motifs of distinct sequences . 10 . 7554/eLife . 00726 . 012Figure 2 . A group of 17 TFs can bind to both methylated and unmethylated motifs of distinct sequences . ( A ) Our previous PDI dataset was compiled with the dataset in this study to generate binding preference of the 17 TFs . Methylated consensus motifs of the 17 TFs identified based on the protein microarray results are compared with their known consensus motifs . ( B ) EMSA assays confirmed that four TFs could specifically interact with both methylated and unmethylated motifs of distinct sequences . Representative images from three independent experiments with similar results are shown . ( C ) and ( D ) Two possible scenarios are proposed to distinguish the mode of interactions between these TFs and their corresponding motifs . ( E ) and ( F ) Competition EMSA assays showed that both scenarios are possible . Representative images from two independent experiments with similar results are shown in each panel . ( G ) OIRD sensorgrams for ZMYM3 and KLF4 binding to methylated motifs M203 and M197 , and their unmethylated counterparts , respectively . The OIRD measurements were performed at two concentrations of each protein . Solid lines represent the OIRD signals . Dashed lines are fitted On- and Off-curves . Red arrows indicate the starting point when a TF protein was introduced to the OIRD reaction chamber . Blue arrows indicate the time points when wash buffer was added . ( H ) Summary of average KD values measured at two concentrations of each protein . ‘NB’ indicates no observed binding signals . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 01210 . 7554/eLife . 00726 . 013Figure 2—figure supplement 1 . Competition EMSA assays for ARID3B and ZMYM3 . As expected , unlabeled and methylated motif M319 showed dose-dependent competition against the labeled , methylated motif M319; whereas unlabeled and unmethylated motif M47 could readily compete off the binding signals . Same results were observed for ZMYM3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 01310 . 7554/eLife . 00726 . 014Figure 2—figure supplement 2 . Competition EMSA assays for KLF4 and TFAP2A . Complex formation between KLF4 and methylated mM197 and between KLF4 and unmethylated umM412 is not affected by either umM412 or mM917 , respectively . However , when both methylated and non-methylated competitor DNA was added , the complex formation was abolished . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 01410 . 7554/eLife . 00726 . 015Figure 2—figure supplement 3 . Summary of KLF4’s dual-specificity . Competition EMSA assays confirm KLF4's binding specificity to methylated motif M197 ( mM197 ) and unmethylated motif M412 ( umM412 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 01510 . 7554/eLife . 00726 . 016Figure 2—figure supplement 4 . OIRD sensorgrams for three TFs and MBD2b binding to three methylated DNA motifs . ( A ) MBD2b with a reported KD value of 330 nM was used as a benchmark in the OIRD system , showing the sensorgrams of MBD2b binding to methylated M203 , M213 and M197 . ( B ) – ( D ) OIRD sensorgrams for ZMYM3 , TFAP2A and KLF4 binding to methylated motifs M203 , M213 and M197 , and their unmethylated counterparts , respectively . The OIRD measurements were performed at two concentrations of each protein . Solid lines represent OIRD signals . Dashed lines are fitted On- and Off-curves . Red arrows indicate the starting point when a TF protein was introduced to the OIRD reaction chamber . Blue arrows indicate the time points when wash buffer was added . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 016 To experimentally confirm this apparent dual-specificity , we performed EMSA assays with two motifs , recognized by the same TF , in both methylated and unmethylated forms ( Figure 2B ) . For instance , KLF4 could form a protein–DNA complex with the methylated motif M197 ( TCCmCpGCCC ) , but not with its unmethylated form ( Figure 2B ) . In contrast , methylation on motif M412 ( GCTTTTACG ) disrupted its interaction with KLF4 ( Figure 2B ) . The same phenomenon was confirmed for three additional TFs , namely TFAP2A , ARID3B , and ZMYM3 ( Figure 2B ) . These results raised an interesting question: does binding by a dual-specificity TF to one motif affect binding to the other of different methylation status ? By performing a competition EMSA assay we observed two scenarios for different TFs ( Figure 2C–F ) . In the first scenario , binding to one motif interferes the binding activity to the other motif . For example , the formation of protein–DNA complex between ARID3B and a methylated motif M319 ( AAAmCpGCTTCC ) could be readily competed off with the unmethylated motif M47 ( GTGGGCGAAA ) in 10-fold excess , and vice versa ( Figure 2E ) , suggesting that ARID3B either uses the same DNA-binding domain to interact with both motifs , or binding to one motif inhibits binding to the other , presumably via conformational changes or steric hindrance ( Figure 2C ) . The specificity of these assays was further confirmed with a competition EMSA assay with ARID3B and ZMYM3 ( Figure 2—figure supplement 1 ) . In the second scenario , binding activities of a dual-specificity TF are independent of each other ( Figure 2D ) . For instance , KLF4–DNA complex formation with the methylated motif M197 was not affected by adding unmethylated motif M412 in 10-fold excess in the competition assay , suggesting that it may use two different domains to distinguish methylated from unmethylated motifs ( Figure 2F; left panel ) . As expected , addition of 10-fold excess of methylated M197 could readily compete off the complex formation of KLF4 with labeled M197 ( Figure 2F; left panel ) . The same results were observed for TFAP2A ( Figure 2F; right panel ) . The binding specificities of KLF4 and TFAP2A were further confirmed by adding both methylated and non-methylated competitor DNA , resulting in disruption of the complex formation ( Figure 2—figure supplement 2 ) . To facilitate direct comparison , we reproduced all combinations of the competition EMSA assays for KLF4 in parallel on the same gel . The results confirmed its dual-specificity in a non-competitive fashion ( Figure 2—figure supplement 3 ) . The fact that KLF4 could form protein–DNA complexes with both motifs equally well , regardless of the presence of the competitors , also suggested that it is highly unlikely that these two motif sequences could bind to the same domain at different affinities . We next employed an oblique incidence reflectivity difference ( OIRD ) system ( Landry et al . , 2004; Zhu et al . , 2007; Fei et al . , 2011 ) to determine binding affinity ( i . e . , KD values ) of ZMYM3 , TFAP2A , and KLF4 with their corresponding methylated DNA motifs M203 , 213 , and M197 , respectively . As a comparison , the well known methylated DNA-binding protein , MBD2b , was also included . Because the OIRD system can monitor binding events in a real-time , label-free fashion , we therefore obtained the Kon and Koff values , and determined the KD values of ZMYM3 , TFAP2A , and KLF4 as 460 nM , 399 nM , and 479 nM , respectively ( Figure 2G; Figure 2—figure supplement 4A–D ) . As expected , none of them showed any significant binding activity to the unmethylated DNA motifs in the OIRD measurements , confirming our previous observations . On the other hand , the KD values of MBD2b measured against the same motifs are in close range to the three TFs tested ( Figure 2—figure supplement 4A ) , suggesting that these TFs could bind to methylated DNA motifs almost as well as MBD2b ( Figure 2H ) . To further dissect the molecular mechanism of this phenomenon , we focused on KLF4 to identify the key residue ( s ) responsible for interacting with the methylated cytosine , in order to decouple its dual-specificity . Intriguingly , KLF4 encodes a classic zf-C2H2 domain , a truncated and a full zf-H2C2 domain at its very C-terminus ( Figure 3A and Figure 3—figure supplement 1 ) . Because the X-ray structure of KLF4 bound to methylated CpG sequences is not available , we used a modeling approach . These efforts were facilitated by crystal structures of other , unrelated 5mC-binding proteins ( e . g . , MeCP2 and ZFP57 ) ( Ho et al . , 2008; Liu et al . , 2012 ) , as well as the adduct composed of mouse KLF4 and unmethylated DNA , which was employed as a modeling template . Using the existing KLF4 structure ( Schuetz et al . , 2011 ) , we identified Arg458 and Asp460 in the last H2C2 zinc finger as a candidate 5mCpG recognition motif . In the KLF4 model , DNA binding is stabilized by a 5mC-Arg-G via improved van der Waals interaction and an Arg458-Asp460 salt-bridge ( Figure 3A ) . Binding is further stabilized by a CH•••O ( H2O-5mC ) H-bond contact of the methylated cytosine on the complementary strand ( Figure 3B ) . Interestingly , the local atomic environment of the KLF4 model is rather similar to those of MeCP2 and ZFP57 in their mCpG-bound forms ( Figure 3—figure supplement 2 ) . Therefore , Arg458 and Asp460 were predicted to be crucial for KLF4 and mCpG-dependent interaction ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 00726 . 017Figure 3 . KLF4’s mCpG-dependent binding activity is decoupled from its binding activity to unmethylated motifs . ( A ) Simulation of KLF4–DNA interactions predicted that two residues , Arg458 and Asp460 , are involved in the interactions with methylated cytosine . Double arrow indicates van der Waals interactions between Arg458 and methyl group on the cytosine in one strand ( 5mCA ) . Red balls represent water molecules . ( B ) Asp460 further stabilizes binding to 5 mC on the other strand ( 5mCB ) via a CH•••O ( H2O—5mC ) H-bond contact . ( C ) EMSA assays using KLF4 mutated proteins demonstrated that both R458 and D460 are crucial for mCpG-dependent binding activity . Representative images from three independent experiments with similar results are shown . ( D ) In cell-based luciferase assays for M197 , WT KLF4 showed mCpG-dependent activation of downstream gene expression ( red bars in the upper panel ) , while both R458A and D460A mutations abolished this activity ( red bars in the middle and lower panels ) . ( E ) In cell-based luciferase assays with M412 ( blue bars ) , both WT and mutants can activate the expression of unmethylated M412 ( blue bars ) , but have no effect on methylated M412 ( red bars ) . In ( D ) and ( E ) , values represent mean ± SD ( n = 3; **: p<0 . 01; t-test ) DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 01710 . 7554/eLife . 00726 . 018Figure 3—figure supplement 1 . Architecture of KLF4 DNA-binding domain . KLF4 encodes two and half zinc finger DNA-binding domains at its C-terminus . Residues R458 and D460 , which were predicted to interact with the 5-methyl group in the cytosine , are located in the zf-H2C2 domain . D432 indicates where a truncated KLF4 construct ends . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 01810 . 7554/eLife . 00726 . 019Figure 3—figure supplement 2 . Known crystal structures of MeCP2 and ZFP57 in complex with methylated DNA . The pink and blue double arrows represent van der Waals force between the arginine and methyl groups . Red balls are water molecules . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 01910 . 7554/eLife . 00726 . 020Figure 3—figure supplement 3 . EMSA assays to evaluate impacts of KLF4 R458K , R458A::D460A mutations , and Δ432 truncation on its binding activity to motifs M412 and M197 . These results clearly demonstrated that both the single- and double-mutations , as well as the truncation , abolished KLF4's ability to form a complex with methylated motif M197 , while neither showed detectable impact on complex formation with unmethylated motif M412 . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 02010 . 7554/eLife . 00726 . 021Figure 3—figure supplement 4 . Western blot analysis of overexpression of KLF4WT , KLF4R458A and KLF4D460A proteins in GT1-7 cells . Using GAPDH as a control , these results demonstrated equal transfection efficiency of the constructs . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 021 To test our prediction , we generated a series of KLF4 mutants , including three single ( i . e . , R458A , R458K , D460A ) , one double ( R458A::D460A ) mutants , and one truncation ( Δ432 ) . EMSA assay with purified KLF4 mutant proteins showed that all point mutations at either or both residues completely abolished KLF4’s binding activity to methylated motif M197 ( CCmCpGCC ) , while none of them had any detectable impact on the binding activity to the unmethylated motif M412 ( TACpGCC; Figure 3C and Figure 3—figure supplement 3 ) . The R458A and R458K mutations showed similar impact on mCpG-dependent binding activity , suggesting that charge on the residue is not critical . Similarly , KLF4 truncation showed no binding activity to methylated motif M197 ( CCmCpGCC ) , but still maintained its binding activity to the unmethylated motif M412 ( TACpGCC; Figure 3—figure supplement 3 ) , indicating that the C2H2 zinc finger in KLF4 mediates KLF4’s interaction with unmethylated motif M412 . To examine whether KLF4’s binding activity to M197 ( CCmCpGCC ) is important for transcription regulation , we employed cell-based luciferase assays using both wild-type ( WT ) and mutated KLF4 expression constructs . Indeed , when co-transfected with the pCpGL construct carrying methylated motif M197 in the promoter , WT KLF4 showed dose-dependent increase of the luciferase activity ( Figure 3D ) . This enhancement of transcription is also methylation-dependent , because WT KLF4 could not increase luciferase activity when motif M197 was not methylated in the construct ( Figure 3D ) . More importantly , both KLF4R458A and KLF4D460A constructs completely lost their ability to modulate transcription from the pCpGL construct carrying methylated motif M197; whereas they both showed dose-dependent enhancement of gene transcription in the luciferase assay , like WT KLF4 , when co-transfected with the pCpGL construct carrying unmethylated motif M412 ( Figure 3E ) . Consistent with the EMSA results , neither WT nor mutant KLF4 constructs showed any significant increase of transcription when motif M412 was methylated in the construct ( Figure 3E ) . Western blot analysis demonstrated an equal transfection efficiency of WT KLF4 , KLF4R458A and KLF4D460A constructs ( Figure 3—figure supplement 4 ) . Altogether , these results suggested that KLF4 could bind to methylated and unmethylated CpG sites in different sequence contexts , resulting in activation of downstream gene expression in heterologous cells in vivo , and that these dual specificities are independent and achieved via two functionally separable domains encoded by KLF4 . Finally , we asked whether KLF4 binding preferences to both methylated and unmethylated motifs that we identified in vitro reflect sequences preferred in vivo . Since KLF4 is well known to play an important role in embryonic stem cell ( ESC ) maintenance , we therefore examined whether mCpG-dependent KLF4–DNA interactions occur in human ESCs ( H1 ) . First , we performed bioinformatics analysis of in vivo KLF4-binding sites in H1 cells based on their methylation status ( Figure 4—figure supplement 1 ) . We superimposed the published in vivo KLF4 binding sites , as determined by the ChIP-seq approach ( Lister et al . , 2009 ) , with the published methylome dataset of single-base resolution obtained with whole-genome bisulfite sequencing in H1 hESCs ( Lister et al . , 2009 ) . We identified numerous cases in which KLF4 tends to bind to those genomic regions containing a methylated CCCpGCC sequence , consistent with our protein microarray results ( Figure 4—figure supplement 1; lower panel ) . Among the KLF4-binding sites containing at least one CpG , we found that the majority of these sites could be categorized into two groups across the whole-genome: 48% have high methylation levels of over 80% and 38% have methylation levels lower than 20% ( Figure 4A ) . We identified statistically significant consensus motifs in methylated group ( ‘Materials and methods’; Supplementary file 1D ) . The top consensus motif , CmCpGC , discovered in the high methylation group is embedded in motif M197 ( CCmCpGCC ) that was recognized by KLF4 in the protein microarray assays ( Figure 4A ) . Taken together , these integration analyses demonstrate that KLF4’s preference to methylated consensus sequence identified in vivo is highly similar to that determined in vitro . 10 . 7554/eLife . 00726 . 022Figure 4 . Endogenous KLF4 binds to methylated loci in human embryonic stem cells ( H1 ) in vivo . ( A ) Bioinformatics analysis to derive methylated DNA motif logo binding to KLF4 by integrating of KLF4 ChIP-Seq and methylome data in H1 cells . Based on the distribution of methylation level at the KLF4 binding sites , a top methylated consensus motif boxed in red was discovered in the highly methylated sites . As a comparison , M197 sequence recognized by KLF4 in the protein microarray assays is shown below . ( B ) Experimental procedure of KLF4 ChIP-bisulfite sequencing to confirm that KLF4 preferentially interacts with hyper-methylated motifs in H1 cells . ( C ) The gel images of KLF4 ChIP’ed loci ( L1: chr1: 559311-559516; L2: chr5: 44424678-44424792; L3: chr16: 4681299-4681481; L4: chr2: 132747088-132747377; L5: chr12: 81828301-81828506 ) demonstrate specific and direct binding of KLF4 to its target regions . Negative controls were performed in the absence of the anti-KLF4 monoclonal antibodies . ( D ) Analysis of KLF4-ChIP against the five loci using the quantitative real-time PCR ( qPCR ) method . Fold change at each locus was obtained by taking the ratio of KLF4-ChIP qPCR signals over the negative control signals . Statistics analysis was based on three technical replicates . ( E ) Sanger bisulfite sequencing reads of input and KLF4-ChIP’ed DNA . Filled and blank circles indicate methylated and unmethylated CpG sites , respectively . Blue and red arrows indicate CpGs in the context of motifs M412 and M917 , respectively . ( F ) For relatively lower methylation input , KLF4 methylated binding sites tend to have a higher methylation level after KLF4 ChIP . The lower panel in ( F ) shows the methylation differences at each CpG site between the input and KLF4 ChIP’ed DNA . p values were determined by binominal probability density function . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 02210 . 7554/eLife . 00726 . 023Figure 4—figure supplement 1 . Integration of KLF4 ChIP-seq and methylome data in H1 cell . KLF4 ChIP-Seq and methylome data in H1 were compiled to assign the methylation levels in KLF4 ChIP'ed segments ( upper panel ) . Lower panel was schematic plot for KLF4 binding summits . The pink ovals indicate KLF4 binding summits as determined in the KLF4 ChIP-seq experiments . The short vertical lines in red and blue indicate the CpG sites in the contexts of motif M197 and M421 , respectively . Other CpG sites are annotated with gray lines . The thin green lines underneath represent the observed methylation level . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 02310 . 7554/eLife . 00726 . 024Figure 4—figure supplement 2 . Five selected KLF4-binding loci for further analyses . The chromosome positions and KLF4 ChIP-seq peaks ( GSM447584 ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 02410 . 7554/eLife . 00726 . 025Figure 4—figure supplement 3 . An example of KLF4 ChIP-bisulfite sequencing assay . The sequencing results confirmed that KLF4 bound to hyper-methylated loci in the sequence context of CCmCGCC ( arrows ) in H1 cell . Upper and lower panels represent bisulfite sequencing results of the input and KLF4 ChIP'ed loci . DOI: http://dx . doi . org/10 . 7554/eLife . 00726 . 025 To provide direct experimental evidence that KLF4 binds to endogenous CCmCpGCC in H1 hESCs in vivo , we performed KLF4 ChIP coupled with bisulfite sequencing ( Figure 4B ) , using primer pairs designed on the basis of the above bioinformatics analysis . First , chromatin IP ( ChIP ) demonstrates specific and direct binding of KLF4 to five selected regulatory loci based on KLF4 ChIP-seq signals ( Figure 4C; Figure 4—figure supplement 2 ) . Next , quantitative real-time PCR ( q-PCR ) also showed the high enrichment ( ≥9-fold ) on the specific loci after KLF4 ChIP ( Figure 4D ) . KLF4 ChIP-bisulfite sequencing analyses showed that KLF4 could readily ChIP those loci ( i . e . , L1 and L2 in Figure 4E ) with TACpGCC at two completely unmethylated fragments , and with CCCpGCC at ∼100% methylation level in other two loci tested ( L3 in Figure 4E , and Figure 4—figure supplement 3 ) , indicating that KLF4 binds to both unmethylated and highly methylated CpGs in different sequence contexts in vivo . To rule out the possibility that other sites within the ChIP’ed fragments might be responsible for KLF4 binding , we searched the published dataset ( Lister et al . , 2009 ) and did not find any other KLF4 binding peaks within the surrounding 500 bp , supporting that KLF4 indeed binds to these unmethylated or methylated sites . We also examined KLF4 methylated target sites that exhibit medium methylation levels ( e . g . , ∼50% ) based on the published methylome dataset . Importantly , KLF4 ChIP-bisulfite sequencing analyses showed that the methylation level of CCCpGCC consensus was significantly increased from 30% and 55% of the genomics inputs to 65% ( p=0 . 001 ) and 73% ( p=0 . 04 ) in the ChIP’ed samples for two examined regions , respectively ( L4 and L5 in Figure 4F; upper panel ) . As predicted , those nearby CpG sites not embedded in the KLF4 consensus did not show any significant increase in methylation ( Figure 4F; lower panel ) . Together , the above analyses provided direct evidence that KLF4 preferentially binds to CCmCpGCC consensus sequence in vivo .
In this study we have identified numerous human TFs across various subfamilies that showed mCpG- and sequence-dependent binding activity , a much more prevalent phenomenon than previously appreciated ( Karlsson et al . , 2008; Rishi et al . , 2010; Quenneville et al . , 2011; Liu et al . , 2012 ) . It is likely that more proteins with the similar activity are yet to be discovered ( Figure 1—figure supplement 8 ) , because ( 1 ) we only surveyed a tiny fraction of methylated CpG space; ( 2 ) we did not screen for TFs that can recognize methylated CHG or CHH sequences in this study , which are present in several cell types including ESCs , induced pluirpotent stem cells , germline and neural cells ( Ramsahoye et al . , 2000; Ziller et al . , 2011 ) ; ( 3 ) we did not include in the screen any hemimethylated DNA motifs , which occurs in vivo on newly synthesized DNAs during replication; and ( 4 ) some of the unconventional DNA-binding proteins we discovered in an earlier study might also encode this property ( Hu et al . , 2009 ) . We estimated false positive and false negative rate for our protein microarray assays . Since 8 of the 11 methylated DNA-binding TFs were confirmed by EMSAs , we estimated that this protein microarray strategy might produce ∼27% of false positives . Our approach also showed some false negatives because some known methylated DNA binding proteins were not identified in this study ( Mann et al . , 2013; Spruijt et al . , 2013 ) . There are several possible reasons to explain this discrepancy . First , since we employed a competition assay on the human TF protein microarrays , only those that preferentially bind to methylated DNA motifs will be identified . In other words , if a protein binds to a DNA sequence irrespective of its methylation status , it will not show strong signals in our screen . Second , the number of DNA sequences we tested is rather small . Based on our simulation analysis ( Figure 1—figure supplement 8 ) , we expect that there exist more such human TFs yet to be discovered with more DNA motifs tested . Third , because we do not have any TF heterodimers on our arrays , it may also generate false negatives . For example , a recent study ( Spruijt et al . , 2013 ) employed a generic methylated DNA probe to pull down potential methylated DNA-binding proteins in mouse cell lysates , resulted in the identification of 19 proteins . Among them , 11 have human orthologs and are presented on our human protein microarrays . Five of them , namely MeCP2 , NRF1 , MBD1/4 , and KLF4 , showed methylated-specific binding activities in our study . The other six proteins showed weak binding signals in our competition assay , presumably due to the reasons discussed above ( Figure 1—figure supplement 1E ) . Regardless of all the above limitations , we still discovered many novel methylation-dependent DNA-binding activities . For example , Znf114 and Znf416 were identified as new generic methylated-DNA binding proteins , with binding activities to almost all methylated DNA motifs tested in this study . The reason that they were not discovered in previous studies might be due to the fact that most of the previous studies were not performed in a systematic way . In our systematic survey , we identified a total of 47 novel methylated DNA-binding proteins , which significantly expanded the methylation-dependent protein–DNA interaction landscape . Our findings provide a new framework to better understand mechanisms of DNA methylation-dependent regulation of gene expression , especially in cancer , stem cell and development biology . For example , recent DNA methylome studies in cancer have identified dynamic epigenetic changes , resulting in global reprogramming in gene expression ( Jones and Baylin , 2007; Figueroa et al . , 2010; Noushmehr et al . , 2010 ) . However , CpG methylation does not always correlate with transcription repression in cancers ( Everhard et al . , 2009 ) . Our unbiased screen of the entire human TF family identified many sequence-specific mCpG ‘readers’ , which can presumably interpret the DNA methylation changes and in turn , regulate gene expression in a dynamic environment . Therefore , by presenting distinct binding sites for TFs , methylated cytosines may serve as the fifth alphabet that changes the landscape of TF–DNA interactions and increases the complexity and diversity of gene regulation .
The protein annotation of transcription factors and cofactors was obtained from our previous study ( Hu et al . , 2009 ) with some minor manual corrections . The information about TF binding domains was obtained from the Pfam ( Bateman et al . , 2004 ) . The oncogenes and tumor repressor genes were downloaded from UniProtKB ( http://www . uniprot . org/ ) . A gene was categorized as a development gene if it has a GO ( gene ontology ) function which includes a keyword of ‘development’ . Supplementary file 1B lists all transcription factors and co-factors on our protein microarray . Double-stranded DNA probes were generated according to a protocol described previously ( Hu et al . , 2011 ) . Methylated CpG probes or luciferase reporter constructs were prepared as previously described ( Guo et al . , 2011 ) . Briefly , 1 μg of DNA were incubated with 1 μl of M . SssI CpG methyltransferase ( 4 U/μl , NEB , Ipswich , MA ) , 1 μl of S-adenosylmethionine ( SAM , 32 mM ) and 2 μl of 10×NEBuffer 2 in a 20 μl reaction volume at 37°C overnight , followed by incubation with freshly added 1 μl of M . SssI ( 4 U/μl ) , 1 μl of SAM ( 32 mM ) , 0 . 5 μl of 10xNEBuffer 2 and 2 . 5 μl of water at 37°C for 4 hr , then at 65°C for 20 min . The methylation status was confirmed by bisulfite sequencing ( Figure 1—figure supplement 7 ) . Human proteins were purified from yeast as GST fusion and arrayed on FAST slides ( Whatman , Maidstone , UK ) in duplicate as described previously ( Hu et al . , 2009 ) . The protein microarrays were probed with Cy5-labeled methylated DNA motifs in the presence of 10-fold unlabeled ( cold ) , unmethylated competitors using a similar protocol described previously ( Hu et al . , 2009 ) . We selected 154 DNA motifs ( Supplementary file 1A ) and performed tests in duplicate protein microarrays . The slides were then washed and scanned with a GenePix 4000B scanner ( Molecular Devices , Sunnyvale , CA ) and the binding signals were acquired using the GenePix 6 . 0 software . We used GenePix 6 . 0 to align the spot-calling grid and record the foreground and background intensities for every protein spot . Figure 1—figure supplement 1 shows detailed workflow for protein microarray analysis . The raw binding intensity ( Rij ) of each probe was defined as Fij/Bij , where Fij and Bij are the median values of foreground and background signals of the probe at site ( i , j ) on the microarray , respectively . We first normalized the raw signal of each probe based on median value of raw signals of its neighboring probes determined by the window size ( window size = 9 × 9 in our study , Figure 1—figure supplement 1B ) . Most probes showed no binding signal ( R′ = 1 ) and had variations around R′ = 1 due to background noise distribution as seen in Figure 1—figure supplement 1C . In order to evaluate the background noise of each microarray , we selected the shadow part ( N1: R′ , in Figure 1—figure supplement 1D ) and artificially combined with its symmetrical part around 1 ( N2: = 2 − N1 ) to obtain the standard deviation ( SD ) of the noise distribution ( N = N1 + N2 , Figure 1—figure supplement 1D ) . Then the Z-score of each probe was calculated byZi , j=R′i , j−N¯std ( N ) , where R′i , j is the locally normalized intensity of probe ( i , j ) on the microarray , N¯ and std ( N ) are mean value and standard deviation , respectively , of noise distribution on the microarray . Since each protein is printed in duplicate on a microarray and each motif binding assay was performed in duplicate , a protein was identified as a positive hit only when all of its four spots produced a Z-score ≥ 3 . Supplementary file 1C lists all transcription factor and cofactor hits showing binding to at least one methylated DNA motif . Each binding reaction was carried out with 100 fmol of biotinylated dsDNA probe and 1 pmol of purified protein in 20 µl of binding buffer as described previously ( Hu et al . , 2009 ) . 10-fold ( 1 pmol ) of unlabeled ( cold ) DNA motifs were added in the competition assays . All the expression clones for proteins used in EMSA were verified by DNA sequencing . Binding affinity between a transcriptional factor and a methylated DNA motif was determined by the oblique incidence reflection difference ( OIRD ) method ( Landry et al . , 2004; Zhu et al . , 2007; Fei et al . , 2011 ) . Synthesized DNA motifs were methylated using SssI enzyme as described above ( ‘Materials and methods’ ) , then printed together with their unmethylated counterparts onto Superamine 2 slides ( ArrayIt , Sunnyvale , CA ) at a concentration of 2 . 5 μM in 50% DMSO , exposed to 600 μJ UV for 3 min , and dried in a 37°C incubator for 1 hr . We selected ZMYM3 , TFAP2A and KLF4 to measure their affinity . We also used MBD2b ( Millipore , Billerica , MA ) as a bench marker . For each OIRD measurement , a DNA motif slide was first washed with EMSA binding buffer ( Hu et al . , 2009 ) at 3 ml/min for 5 min , and then flooded with the corresponding TF protein solution at various concentrations as indicated in Figure 2—figure supplement 3 . The observed on curve was then measured in real time until the OIRD signals reached saturation . To determine the off curve , the EMSA buffer was then pumped through the reaction chamber at a speed of 200 μl/min until the OIRD signals stabilized . Binding and dissociation signals were recorded every 10 s . Data analysis was carried out using Origin 9 . 0 following Majka J and Speck C’s method ( Majka and Speck , 2007 ) . The CpG-free luciferase reporter vector used in our study , pCpGL , was a gift from Dr Michael Rehli ( Klug and Rehli , 2006 ) . Eight units of a DNA motif were subcloned into pCpGL promoter region , and the resulting plasmids ( pCpGL-8X-motif ) were grown in Escherichia coli PIR 1 strain using Zeocin as a selection marker . The CpG sites in the motifs were methylated using M . SssI as described previously . GT1-7 cells were co-transfected with three constructs: pCpGL-8X-motif , pCAGIG expressing the corresponding TF constructs at various concentrations , and pTK-RL ( Promega , Madison , WI ) using the FugeneHD reagent ( Roche , Basel , Switzerland ) . Cells were harvested 48 hr post-transfection for luciferase reporter assay using the Dual-Luciferase reporter assay system ( Promega ) . All assays were performed in triplicate . We reason that the methylated CpG is the key and conserved position for binding because the unmethylated counterpart shows no binding . Therefore , we first aligned the motif sequences around methylated CpG . For each TF binding to multiple methylated motifs , we then hierarchically clustered all 6-mer binding sequences centered on mCpG . With a selected cutoff for sequence similarity , all the binding motifs were separated into several groups . Each group of sequences was combined to derive the consensus sequences . The consensus sequence for the largest sequence group is shown in Figure 2A . If a TF binds to only one motif , the consensus sequence is the same as the binding motif . We collected the known consensus sequences for the TFs from the literature and databases ( TRANSFAC and JASPAR ) . Some TFs have multiple known consensus sequences . In such cases , we prefer consensus sequences derived from in vitro interaction assays to those derived from in vivo experiments such as ChIP-seq , because the binding sequences derived from ChIP-seq could include methylated sequences . For example , one of the KLF4 consensus sequences in TRANSFAC was based on a ChIP-seq experiment ( Chen et al . , 2008 ) , while another KLF4 binding motif was based on an in vitro oligonucleotide library ( Shields and Yang , 1998 ) . The known consensus for KLF4 shown in Figure 2A was created based on the dataset from in vitro oligonucleotide library using MDscan ( Liu et al . , 2002 ) . Site-directed mutagenesis was carried out using the QuikChange Multi Site-Directed Mutagenesis Kit ( Agilent , Santa Clara , CA ) . We integrated human methylome data and KLF4 ChIP-seq data to obtain the KLF4 binding landscape in H1 human embryonic stem cells ( Lister et al . , 2009 ) . First , we adopted one popular software MACS ( Model-based Analysis for ChIP-Seq ) to determine KLF4 binding peaks and summits . The lengths of peaks determined by MACS were from 152 bp to 6062 bp . We selected 95% of these peaks which were shorter than 371 bp for further analysis . To remove the effect of sequence composition difference in genomic regions , we focused on the binding peaks but extended the peaks to ±50 bp to achieve enough ‘background’ sequences . The ‘foreground’ was sequences within ±60 bp around KLF4 binding summits . Since we are interested in motifs that could be affected by methylation , we obtained the occurrences of all possible 6-mers having CpG at the center position , e . g . AACGCT . The 6-mers significantly enriched in highly methylated summits ( M ≥ 0 . 8 ) than in the background were predicted as KLF4 methylated binding motifs . The p-value was calculated based on hypergeometric cumulative distribution function . At the next step , we clustered significant 6-mer sequences ( p<0 . 01 after Bonferroni multiple-test correction , Supplementary file 1D ) from methylated group to obtain KLF4 methylated DNA binding consensus motif . One consensus sequence is shown in Figure 4A . For feeder-free culture of H1 , cells were cultured on Matrigel ( BD , 1:80 dilution ) -coated plates with MEF-condition media ( D-MEM/F12 supplemented with 20% KSR , 2 mM L-glutamine , 100 mM MEM NEAA , and 100 mM beta-mercaptoethanol ) supplemented with 4 ng/ml bFGF as previously described ( Chiang et al . , 2011 ) . Media were changed daily . Chromatin immunoprecipitation ( ChIP ) was carried out in H1 cells using a rabbit anti-KLF4 antibody ( H180; Santa Cruz , Dallas , TX ) according to a protocol described previously ( Nelson et al . , 2006 ) , except that the protein A-Sepharose was replaced with Dynabeads Protein A ( Life Technologies , Grand Island , NY ) . Normal rabbit IgG was used for mock IP as a negative control . Primers shown in Supplementary file 1E were used in ChIP-PCR and qPCR to detect the binding and enrichment of KlF4 regulatory regions after ChIP . Sanger bisulfite sequencing was performed as previously described ( Guo et al . , 2011; Ma et al . , 2009 ) . Purified genomic DNA or ChIP DNA from H1 cells were treated by EZ DNA Methylation-Direct Kit ( Zymo Research , Irvine , CA ) . After bisulfite conversion , regions of interest were PCR-amplified . The primers used in Figure 4D , E were listed in Supplementary file 1E . PCR products were gel-purified and cloned into a TA vector ( Life Technologies ) . Individual clones were sequenced and aligned with the reference sequence . | DNA methylation—the addition of a methyl group to a cytosine or adenine base within DNA—has a key role in regulating the expression of genes as proteins . It contributes to processes such as X-inactivation , in which one copy of the X chromosome is silenced in females , and genomic imprinting , in which the expression of a gene depends upon which parent it was inherited from . DNA methylation has also been implicated in the development of cancer . However , the molecular mechanisms by which it produces these effects are not fully understood . In mammals , the methylation of CpG sites—which consist of a cytosine base next to a guanine base—is typically thought to reduce gene expression by preventing proteins called transcription factors from binding to regions of DNA called promoters . This can occur directly if methylation disrupts interactions between the DNA and the transcription factors , or indirectly if other proteins that bind to the methylated DNA compete with the transcription factors for binding sites . However , only a small number of proteins that bind to methylated DNA have so far been identified . Now , Hu et al . have screened the entire family of roughly 1300 human transcription factors and 210 co-factors ( proteins that interact with transcription factors ) for their ability to bind to some 150 different stretches of methylated DNA . They found that 47 of the proteins could bind to methylated CpG sites , with the majority showing a preference for specific DNA sequences . Moreover , some transcription factors and co-factors bind to methylated and non-methylated DNA targets with distinct sequences . These two types of binding are largely independent , as illustrated by the fact that mutations that prevent a transcription factor called KLF4 from binding to methylated DNA do not prevent it binding to unmethylated DNA , and vice versa . The work of Hu et al . suggests that methylated cytosine can effectively act as a ‘fifth base’—in addition to adenine , cytosine , guanine and thymine—and emphasizes the importance of DNA methylation for regulating gene expression . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"genetics",
"and",
"genomics"
] | 2013 | DNA methylation presents distinct binding sites for human transcription factors |
The pre-placodal ectoderm , marked by the expression of the transcription factor Six1 and its co-activator Eya1 , develops into placodes and ultimately into many cranial sensory organs and ganglia . Using RNA-Seq in Xenopus laevis we screened for presumptive direct placodal target genes of Six1 and Eya1 by overexpressing hormone-inducible constructs of Six1 and Eya1 in pre-placodal explants , and blocking protein synthesis before hormone-inducing nuclear translocation of Six1 or Eya1 . Comparing the transcriptome of explants with non-induced controls , we identified hundreds of novel Six1/Eya1 target genes with potentially important roles for placode development . Loss-of-function studies confirmed that target genes encoding known transcriptional regulators of progenitor fates ( e . g . Sox2 , Hes8 ) and neuronal/sensory differentiation ( e . g . Ngn1 , Atoh1 , Pou4f1 , Gfi1 ) require Six1 and Eya1 for their placodal expression . Our findings provide insights into the gene regulatory network regulating placodal neurogenesis downstream of Six1 and Eya1 suggesting new avenues of research into placode development and disease .
The cranial placodes give rise to many sense organs of the vertebrate head ( including nose , ear and lateral line ) and contribute to the anterior pituitary and sensory ganglia of the cranial nerves . Together with the neural crest , which also contributes to cranial ganglia as well as the head skeleton , they originated as an evolutionary novelty in stem vertebrates , on the adoption of a more active and exploratory life style ( Northcutt and Gans , 1983; Schlosser , 2015 ) . Defects in placode development underlie many congenital diseases of sensory organs and the endocrine system ( Petit et al . , 2001; Davis et al . , 2013; Xu , 2013 ) , however , despite this central importance of placodes in the evolution and development of the vertebrate head , they have been much less well studied than the neural crest , and little is known about the gene regulatory networks ( GRNs ) driving early placode development . Fate mapping studies have shown that all cranial placodes develop from a common precursor region , the pre-placodal ectoderm ( PPE ) ( Streit , 2002; Bhattacharyya et al . , 2004; Xu et al . , 2008; Pieper et al . , 2011 ) . In neural plate stage embryos , the PPE is located as a horseshoe-shaped domain around the anterior neural plate ( and abutting the cranial neural crest laterally ) which subsequently breaks up into individual placodes ( Schlosser , 2010; Grocott et al . , 2012; Saint-Jeannet and Moody , 2014 ) . Molecularly , the PPE is characterised by the expression of Six1 and Eya1 , which also continues in most placodes derived from the PPE ( Schlosser and Ahrens , 2004 ) . Whereas Six1 encodes a transcription factor , Eya1 encodes a transcriptional co-activator that also has phosphatase activity ( Kumar , 2009; Tadjuidje and Hegde , 2013 ) , and Six1 and Eya1 have been shown to form a protein complex and synergistically activate transcription ( Ohto et al . , 1999; Li et al . , 2013 ) . However , both Six1 and Eya1 also interact with other protein interaction partners; Six1 , for example , has been shown to act as a transcriptional repressor after binding to the co-repressor Groucho ( Brugmann et al . , 2004 ) whereas Eya1 is known to form protein complexes with other binding partners including the transcription factor Sox2 ( Ahmed et al . , 2012a; Tadjuidje and Hegde , 2013 ) . Loss of Six1 or Eya1 function in mouse , zebrafish , chick or Xenopus embryos leads to a similar spectrum of PPE and placodal defects , with altered expression of other PPE genes , decreased proliferation and increased apoptosis in many placodes , compromised morphogenetic movements ( invagination or cell delamination ) and a decreased production of sensory cells and neurons ( Xu et al . , 1999; Laclef et al . , 2003; Zheng et al . , 2003; Brugmann et al . , 2004; Zou et al . , 2004; Kozlowski et al . , 2005; Schlosser et al . , 2008; Christophorou et al . , 2009; Ahmed et al . , 2012a , 2012b ) . In human patients , mutations in both Six1 and Eya1 lead to branchio-oto-renal ( BOR ) and branchio-otic ( BO ) syndromes with congenital hearing loss ( Kochhar et al . , 2007 ) . These findings suggest that these proteins are core regulators of placode development and promote multiple aspects of placode development synergistically , although Eya1-independent roles of Six1 have also been reported ( Brugmann et al . , 2004; Bricaud and Collazo , 2011 ) . Specifically , Six1 and Eya1 have been shown to play central roles , during multiple steps , in the development of sensory cells ( e . g . hair cells in the inner ear ) as well as sensory neurons , and promote both the proliferation of sensory/neuronal progenitors as well as sensory and neuronal differentiation in a dosage dependent fashion ( Zou et al . , 2004; Schlosser et al . , 2008; Zou et al . , 2008; Ahmed et al . , 2012b , 2012a ) . Recently Atoh1 , an essential determination gene for hair cell development , has been shown to be directly transcriptionally activated by Six1/Eya1 binding to its enhancer ( Ahmed et al . , 2012a ) . Moreover , the neuronal progenitor genes Sox2 and Sox3 have been shown to be up-regulated by Six1 and Eya1 in the absence of protein synthesis , suggesting that they are also direct target genes ( Schlosser et al . , 2008 ) . Several other direct target genes of Six1 have been identified ( Kumar , 2009; Xu , 2013 ) , but no specific screen for direct target genes of Six1 and Eya1 in the PPE and the developing placodes has yet been conducted . Here , using RNA-Seq in Xenopus laevis , we present the first comprehensive screen for presumptive direct target genes of Six1 and Eya1 in the developing placodes in any vertebrate . Hormone-inducible constructs of Six1 and Eya1 ( fused with the human glucocorticoid receptor [GR] ) were overexpressed , either alone or in combination , in Xenopus embryos . We then explanted the PPE at neural fold stages and activated nuclear translocation of Six1 or Eya1 in these explants by the addition of dexamethasone ( DEX ) after blocking protein synthesis by cycloheximide ( CHX ) . This approach has previously been shown to reliably activate direct targets of GR-fusion constructs only in the presence of DEX ( Kolm and Sive , 1995; Seo et al . , 2007 ) . We then analysed the transcriptome of placodal explants by RNA-Seq and compared this to control explants which were not hormone induced , in order to specifically survey target genes directly activated or repressed by Six1 or Eya1 in the PPE and developing placodes . Using this method , we were able to identify a large number of novel target genes with potentially important roles for cranial placode development . We were further able to show in loss of function studies that several target genes encoding known regulators of progenitor fates ( e . g . Sox2 , Hes8 ) and neuronal/sensory differentiation ( e . g . Ngn1 , Atoh1 , Pou4f1 . 2 , Gfi1a ) required both Six1 and Eya1 for their expression in the developing placodes . Our findings provide pioneering insights into the GRNs regulating placode development downstream of Six1 and Eya1 , and suggest exciting new avenues of research for understanding placode development and disease .
RNA was extracted from explants cut from the PPE of un-injected embryos and characterised using RNA-Seq to provide a complete transcriptome of the PPE . After removing genes expressed at low levels ( FPKM < 1 ) and annotation against a Xenopus mRNA database ( see Materials and methods ) , we assembled a transcriptome comprising 15 , 794 transcripts , and the top 1000 expressed genes are shown in Supplementary file 1 . Gene Set Enrichment Analysis ( GSEA ) on these genes revealed that RNA processing/splicing was very highly enriched in the PPE transcriptome ( enrichment score [E]: 43 ) , suggesting that RNA-binding proteins and mRNA splicing mechanisms may play an important role in placodal development as has also been reported for the neural crest ( Simões-Costa et al . , 2014 ) . Translation elongation and ribosomal proteins were also highly enriched ( E: 32 ) , perhaps reflecting the high rate of protein turnover in the rapidly changing PPE ( McCabe et al . , 2004 ) . To identify presumptive direct targets of Six1 and Eya1 , Six1-GR and Eya1-GR fusion proteins were overexpressed either alone or together in the PPE . In combination with a protein synthesis inhibitor ( CHX ) , nuclear translocation of Six1 and Eya1 was induced by adding DEX for 2 . 5 hr , and gene expression was analysed using RNA-Seq ( Figure 1 ) . Presumptive direct targets of Six1 and Eya1 were determined by comparing Six1-GR- , Eya1-GR- or Six1-GR+Eya1-GR-injected embryos treated with CHX alone ( as controls ) against CHX+DEX-treated samples . Resultant data sets from such individual treatment groups ( each with two biological replicates ) are henceforth referred to as Six1i , Eya1i and Six1+Eya1i . In this paradigm , the expression of target genes for which either Six1 or Eya1 concentrations are limiting in the PPE should be affected in Six1i and Eya1i treatment groups , respectively ( and potentially also in Six1+Eya1i ) , while the expression of target genes limited by both Eya1 and Six1 concentrations in the PPE should be modulated only in the Six1+Eya1i treatment group . 10 . 7554/eLife . 17666 . 003Figure 1 . Experimental pipeline and overview of bioinformatic analysis . ( A ) Both blastomeres of two-cell stage embryos were injected with Six1-GR , Eya1-GR or Six1-GR+Eya1-GR and explants were cut from pre-placodal ectoderm . Explants were incubated in CHX for 30 min before being split into two groups; 50% were kept in CHX for 2 . 5 hr and 50% were transferred to CHX+DEX for 2 . 5 hr . RNA was extracted from both treatment groups and submitted to RNA-Sequencing . ( B ) On average , 80 million reads were generated in sequencing for each treatment group , and 65 million quality-trimmed reads were successfully mapped to the Xenopus genome . An average of 49 , 000 transcript models were assembled , of which 80% ( 39 , 000 ) were successfully annotated against a Xenopus mRNA database . Annotated transcript models were then filtered to condense duplicate annotations into 15 , 794 uniquely annotated transcript models , and differential expression analysis was then performed using CHX treated explants as a control for those treated with CHX+DEX . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 003 Using this approach , we identified 365 genes up-regulated at least twofold that satisfied all criteria for differential expression ( log2 fold change [FC] ≥ 1; FPKM ≥ 1; FC < 0 . 5 in un-injected control ) in Six1i , 508 in Eya1i and 836 in Six1+Eya1i , treatment groups , while 292 genes were down-regulated in Six1i , 218 in Eya1i and 490 in Six1+Eya1i treatment groups ( Figure 2A and B; Supplementary file 2 ) . As an initial means of estimating data quality , we searched for targets of Six1 established in previous studies ( Atoh1 ( Ahmed et al . , 2012a ) ; Slc12a2 ( Ando et al . , 2005 ) ; CyclinA1 ( Coletta et al . , 2004 ) ; CyclinD1 ( Li et al . , 2013 ) ; c-Myc ( Li et al . , 2003 ) ; Ezrin ( Yu et al . , 2006 ) ; Gdnf ( Li et al . , 2003 ) ; Sox3 ( Schlosser et al . , 2008 ) ; Sox2 ( Schlosser et al . , 2008 ) ; Sall1 ( Chai et al . , 2006 ) ; and MyoD1 ( Liu et al . , 2013 ) ) in the Six1i and Six1+Eya1i data sets . With the exception of c-Myc , all genes were present in the transcriptome , and most were found in either Six1i ( CyclinD1 [ccndx] FC: 7 . 48; Slc12a2 , FC: -2 . 75; CyclinA1 , FC: -3 . 68; Sox2 , FC: 1 . 2; MyoD , FC: 3 . 4 ) or Six1+Eya1i ( Sox3 , FC: 0 . 9; Atoh1 , FC: 1 . 4; Sall1 , FC: 0 . 99 ) data sets , confirming the utility of our approach in identifying direct targets . Moreover , Atoh1 , Sox2 and MyoD1 were found both in our Six1+Eya1i and Eya1i datasets as expected based on the known coregulation of these Six1 target genes by Eya1 ( Ahmed et al . , 2012a; Grifone et al . , 2007; Schlosser et al . , 2008 ) . We suggest that overexpression of Eya1 alone may upregulate such genes in those parts of the ectoderm where Six1 is already expressed at high levels but Eya1 at relatively low levels in vivo . 10 . 7554/eLife . 17666 . 004Figure 2 . Differentially expressed genes after overexpression of Six1 and Eya1 in the PPE . Plots A and B show number of genes differentially regulated after overexpression of Six1 alone ( Six1i; yellow ) , Eya1 alone ( Eya1i; blue ) or Six1 and Eya1 combined ( Six1+Eya1i; green ) . Each Venn diagram shows the number of genes ( red ) unique for each treatment group or shared between them . ( A ) Number of genes up-regulated and ( B ) down-regulated after injection with Six1 , Eya1 or Six1+Eya1 . ( C ) The merged analysis resulted in hundreds of significantly differentially expressed genes in the PPE data set . Plot shows log2 transformed ( FPKM+1 ) values after overexpression of Six1 or Eya1 ( combination of all treatment groups; Six1+Eya1m ) . Green points represent significantly ( q<0 . 05 ) up-regulated genes and red points show significantly down-regulated genes . Plot D shows the enrichment of molecular function terms after overexpression of Six1 or Eya1 based on significantly differentially expressed genes from the merged data set ( Six1+Eya1m; Supplementary file 3 , Table 5 ) . The area of the pie represents the total number of functional terms contained in the analysis , with each slice representing the percentage of genes against this total . Molecular functions shown can be broadly divided into five categories: Green slices are related to binding functions ( 53% ) ; purple/blue represents enzyme activity ( 30% ) ; pink/red shows transmembrane proteins ( 13% ) ; orange cytoskeleton ( 3% ) and yellow anti-oxidant ( 1% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 004 Comparison between our different treatment groups allows us to distinguish genes likely co-regulated by Six1 and Eya1 from those that are not and , thus , may be regulated by Six1 or Eya1 alone or in conjunction with other protein-binding partners . Since ectodermal expression of Six1 and Eya1 is widely overlapping in vivo but not completely congruent , genes co-regulated by Six1 and Eya1 may be differentially expressed not only after coinjection of Six1 and Eya1 ( Six1+Eya1i treatment group ) but also after injection of Six1 or Eya1 alone ( Six1i and Eya1i treatment groups , respectively ) because elevation of Six1 or Eya1 levels will produce higher levels of the coregulatory complex in those parts of the ectoderm where the respective protein is expressed at much lower levels than its binding partner . Hence , a subset of target genes with high response thresholds to the Six1-Eya1 coregulatory complex ( e . g . due to low affinity binding sites ) will respond to overexpression of Six1 or Eya1 alone with differential expression in these parts of the ectoderm while another subset of genes with low response thresholds ( e . g . due to high affinity binding sites ) will not . The latter subset will , thus , only be differentially expressed after overexpression of both Six1 and Eya1 , creating expanded areas of Six1 and Eya1 coexpression in the ectoderm . Notably , the false discovery rate is expected to be lower for the former subset , which is supported by three independent treatment groups ( Six1i , Eya1i and Six1+Eya1i ) , than in the latter subset , supported only by one ( Six1+Eya1i ) . About half of all genes differentially expressed in the PPE in our various treatment groups show evidence of co-regulation by Six1 and Eya1 . This includes 690 ( 633+57 ) up-regulated and 444 ( 440+4 ) down-regulated genes ( Figure 2A and B ) . Indeed , the top 10% of transcripts ( ranked by FC; post DEX-filtering ) up-regulated in Six1i , Eya1i or Six+Eya1i treatment groups were each highly enriched for the top 10% of transcripts up-regulated in any of the other experimental treatment groups ( p<0 . 0001; Fisher’s exact test ) . More genes co-regulated by Six1 and Eya1 were up-regulated than were down-regulated ( 690/1134 = 60 . 8% for all co-regulated genes , 57/61 = 93 . 4% for co-regulated genes identified in each treatment group; Figure 2A and B ) , corroborating previous findings that Six1 and Eya1 typically act synergistically to activate transcription ( Ahmed et al . , 2012b , 2012a; Brugmann et al . , 2004; Christophorou et al . , 2009; Li et al . , 2003; Ruf et al . , 2004 ) . However , our identification of a subset of genes synergistically down-regulated by Six1 and Eya1 suggests that Eya1 may not always act as a co-activator of Six1 . In contrast , there is no support for co-regulation for genes that are differentially expressed only in Six1i but not Eya1i treatment groups ( and vice versa ) even for those genes that are also differentially expressed after Six1+Eya1i treatment . We identified 283 ( 190+93 ) genes up-regulated and 270 ( 233+37 ) genes down-regulated by Six1 but not Eya1 , indicating that these are regulated by Six1 in an Eya1 independent way but possibly dependent on other co-factors . Conversely , we identified 426 ( 373+53 ) genes up-regulated and 196 ( 187+9 ) genes down-regulated by Eya1 but not Six1 ( Figure 2A , B ) suggesting that these are regulated by Eya1 in conjunction with transcription factors other than Six1 . To add statistical power to our analysis , we next merged treatment groups and determined significantly differentially expressed genes ( q<0 . 05 ) in these merged groups . We first created a data set Six1+Eya1m in which all replicates that involved overexpression of either Six1 or Eya1 were considered as equivalent ( injection of Six1-GR , Eya1-GR or Six1-GR+Eya1-GR; 6 replicates in total ) . This allowed us to identify genes that are significantly differentially expressed across all treatment groups . We also created a data set Six1m , in which all replicates that involved Six1 overexpression were considered as equivalent ( injection of Six1-GR or Six1-GR+Eya1-GR; 4 replicates ) . This allowed us to identify genes with significant differential expression after Six1 upregulation . Similarly , we created data set Eya1m based on all replicates that involved Eya1 overexpression ( injection of Eya1-GR or Six1-GR+Eya1-GR; 4 replicates ) allowing us to identify genes differentially expressed after Eya1 upregulation . We found 181 significantly ( q<0 . 05 ) up-regulated genes in the Six1+Eya1m group , 149 in Six1m and 112 in Eya1m ( Supplementary file 3 , Tables 1–3 ) . Substantially fewer genes were negatively regulated in these merged groups , with only 14 significantly down-regulated genes found in Six1+Eya1m , 11 in Six1m and 13 in Eya1m ( Supplementary file 3 , Tables 4–6 ) , re-enforcing the notion that together , Six1 and Eya1 act primarily as transcriptional activators ( Figure 2C ) . Presumptive direct targets that were significantly up-regulated in our merged data set ( Six1+Eya1m ) were analysed using Panther ( Mi et al . , 2013 ) to examine the representation of genes grouped by molecular function ( Figure 2D ) . Transcription factors and protein binding together accounted for the largest fraction of up-regulated genes ( 53% in total ) , followed by enzymes ( 30% ) and transporter molecules ( 13% ) , suggesting a developmental function of many of the genes up-regulated by either Six1 or Eya1 . GSEA was then conducted using DAVID ( Huang et al . , 2009 ) on the sets of significantly up- or down-regulated genes in our merged data sets , as well as in various combinations of subsets of differentially expressed genes from our individual treatment groups ( Figure 3 and Figure 3—figure supplement 1 ) . This analysis showed that genes directly up-regulated by Six1 , Eya1 or Six1+Eya1 were highly enriched for terms associated with sense organ development , inner-ear development , mechanoreceptor differentiation , eye morphogenesis , neurogenesis and axon guidance consistent with their synergistic role in sensory development ( Grocott et al . , 2012; Schlosser , 2010 ) and neurogenesis ( Maier et al . , 2014; Schlosser and Northcutt , 2000 ) . Apart from genes encoding transcription factors involved in sensory development ( see below ) , genes encoding cell cycle regulators ( CyclinD , RGCC ) , cell surface receptors and adhesion molecules ( e . g . CXCR7 , EDAR , Protocadherin11 , Claudin3 , Fzd1 , Fzd4 , ) , secreted proteins ( e . g . FGF3 , FGF19 , Dkk1 , Neurotrophin3 ) and cytoskeletal regulators ( e . g . RhoV , Espin ) with known or potential roles in placode development were also up-regulated . 10 . 7554/eLife . 17666 . 005Figure 3 . Gene set enrichment analysis ( GSEA ) showing enriched clusters of functional terms for up-regulated genes in different treatment groups . In each case , treatment groups considered are highlighted and outlined in bold in the accompanying Venn diagram . Yellow colouring indicates Six1 treatment; blue shows Eya1 and green Six1+Eya1 . Enrichment scores ≥1 . 5 are reported for individual treatment groups ( Ind . ) and , where available , ≥0 . 5 for merged treatment groups ( Merg . ) . ( A ) Up-regulated genes from all treatment groups included in analysis; ( B ) Six1 overexpression only; ( C ) Eya1 overexpression only . ( D ) Genes differentially expressed after overexpression of both Six1 and Eya1; ( E ) exclusively after Six1 overexpression; ( F ) exclusively after Eya1 overexpression . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 00510 . 7554/eLife . 17666 . 006Figure 3—figure supplement 1 . Gene set enrichment analysis ( GSEA ) showing enriched clusters of functional terms for down-regulated genes in different treatment groups . In each case , treatment groups considered are highlighted and outlined in bold in the accompanying Venn diagram . Yellow colouring indicates Six1 treatment; blue shows Eya1 and green Six1+Eya1 . Enrichment scores ≥1 . 5 are reported for individual treatment groups ( Ind . ) and , where available , ≥0 . 5 for merged treatment groups ( Merg . ) . ( A ) Down-regulated genes from all treatment groups included in analysis; ( B ) Six1 overexpression only; ( C ) Eya1 overexpression only . ( D ) Genes differentially expressed after overexpression of both Six1 and Eya1; ( E ) exclusively after Six1 overexpression . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 006 GSEA analysis of discrete subsets of genes exclusively regulated by Six1 or Eya1 suggested that they also regulate some categories of genes independently of one another . A particularly interesting finding was the extreme enrichment of Hox genes ( specifically of the Antennapedia-type ) in the Eya1-specific subset of up-regulated genes ( Figure 3F ) , suggesting that Eya1 may play a previously un-identified role in regulating Hox gene expression independently of Six1 . To verify our RNA-Seq data , we selected a number of target genes for further characterisation and , in order to gain insight into the GRN downstream of Six1 and Eya1 , we restricted candidates to transcription factors or co-factors up-regulated by Six1 or Eya1 . First , we generated a list of well-supported target genes containing all genes with at least a two-fold up-regulation in at least two of our three treatment groups ( Table 1 ) . From the 228 genes in this list we selected all 30 transcription factors or co-factors for further analysis . However , we were unable to amplify two genes from this list ( Egr3 , Fbxo41 ) from cDNA and therefore omitted these genes from further characterisation . We additionally included Sox3 and Ngn1 - which were found to be slightly below our threshold of twofold up-regulation in at least two treatment groups - because previous studies have implicated these genes in the regulation of placodal neurogenesis downstream of Six1 and Eya1 ( Ma et al . , 1996 , 1998; Schlosser et al . , 2008; Ahmed et al . , 2012b ) ( Table 2 ) . 10 . 7554/eLife . 17666 . 007Table 1 . Genes with at least two-fold up-regulation in at least two out of three individual treatment groups ( Six1i; Eya1i; Six1+Eya1i ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 007Annotation*AccessionSix1 FC†Eya1 FC‡Six1+Eya1 FC§Chromosome unknown open reading frameXM_002938866 . 26 . 2-7 . 9cDNA clone IMAGE:7022272BC094950 . 15 . 65 . 17 . 5X . laevis cyclin Dx ( ccndx ) NP_001087887 . 17 . 5-5 . 2Calcitonin gene-related peptide-likeXM_002941675 . 27-3X . laevis tripartite motif containing 63 , E3 ubiquitin protein ligase ( trim63 ) NM_001093214 . 15 . 33 . 66 . 3ATP-sensitive inward rectifier potassium channel 11-likeXM_004916278 . 15 . 1-6 . 1Leucine rich repeat containing 52 ( lrrc52 ) XM_002933773 . 26 . 1-2 . 8#SIX homeobox 2 ( six2 ) NM_001100275 . 153 . 55 . 9Potassium voltage-gated channel shaker-related subfamily member 2 ( kcna2 ) XM_004910736 . 15 . 1-4 . 9Butyrophilin subfamily 2 member A1 ( btn2a1 ) NM_001094508 . 1-1 . 24 . 9Glutathione peroxidase 2 ( gpx2 ) NM_001256315 . 1-2 . 54 . 8#X . laevis for Xsox17-alpha proteinAJ001730 . 13 . 62 . 64 . 8X . laevis ectodysplasin A receptor ( edar ) NM_001087047 . 12 . 82 . 54 . 7Uncharacterized ( LOC101734405 ) XM_004918247 . 14 . 40 . 83 . 5Glutathione peroxidase 2 ( gpx2 ) NM_001256315 . 13 . 42 . 14 . 4X . laevis cytochrome P450 , family 2 , subfamily D , polypeptide 6 ( cyp2d6 ) NM_001093574 . 11 . 1-4 . 4Calcium/calmodulin-dependent protein kinase kinase 2beta ( camkk2 ) XM_002937701 . 24 . 42 . 63 . 5Cytochrome P450 family 26 subfamily B polypeptide 1 ( cyp26b1 ) NM_001079187 . 23 . 344 . 3Troponin I type 1 ( skeletal , slow ) BC0612681 . 8-4 . 372 kDa inositol polyphosphate 5-phosphatase-like ( LOC101734556 ) XM_004916572 . 1-4 . 21 . 3Chemokine ( C-X-C motif ) receptor 7 ( cxcr7 ) NM_001030434 . 132 . 84 . 1#X . laevis xSox17 alpha 2AB052691 . 11 . 71 . 44Espin ( espn ) transcript variant X3XM_004916193 . 1-2 . 14B-cell CLL/lymphoma 11B ( zinc finger protein ) ( bcl11b ) XM_004917116 . 1-1 . 94C-X-C motif chemokine 10-likeXM_002940578 . 21 . 943 . 5X . laevis hedgehog-interacting proteinBC046952 . 1-2 . 74X-linked inhibitor of apoptosis ( xiap ) NM_001030412 . 143 . 12 . 3X . laevis uncharacterized ( LOC496300 ) NM_001095458 . 11 . 43 . 91 . 1X . laevis RDC1 like proteinBC098974 . 13 . 62 . 13 . 9X . laevis for frizzled 4 protein ( fz4 gene ) AJ251750 . 11 . 30 . 63 . 8Espin ( espn ) transcript variant X1XM_002933856 . 23 . 12 . 73 . 7Paired box 1 ( pax1 ) transcript variant X1JQ929179 . 1-33 . 7Potassium voltage-gated channel subfamily F member 1 ( kcnf1 ) NM_001102926 . 13 . 6-2 . 1Echinoderm microtubule-associated protein-like 1-likeXM_004917169 . 1-3 . 62 . 8Leucine rich adaptor protein 1-like ( lurap1l ) XM_002940127 . 23 . 61 . 42 . 4Sine oculis binding protein homolog ( Drosophila ) BC154687 . 12 . 71 . 63 . 4RNA-directed DNA polymerase homologXM_004916122 . 13 . 4-2 . 1Kinesin family member 3C ( kif3c ) transcript variant X1XM_004914940 . 11 . 40 . 83 . 4Anoctamin 2 ( ano2 ) XM_002932297 . 22 . 21 . 33 . 4X . laevis natriuretic peptide C ( nppc ) NM_001112924 . 12 . 1-3 . 3Uncharacterized ( LOC101734952 ) XM_004916172 . 12 . 5-3 . 3Poly ( ADP-ribose ) polymerase 14-like ( LOC101731378 ) XM_004920062 . 13 . 1-3 . 3Protocadherin-11 X-linked-like ( LOC100493938 ) XM_004916890 . 1-3 . 21 . 4Uncharacterized ( LOC101733225 ) XM_004919937 . 12 . 53 . 21 . 6Calcium channel voltage-dependent beta 4 subunit ( cacnb4 ) NM_001142151 . 13 . 1-1 . 9F-box protein 32 ( fbxo32 ) transcript variant X1XM_002941397 . 21 . 8-3 . 1cDNA clone TEgg026p17CR761997 . 232 . 6-X . laevis transforming growth factor beta-induced ( tgfbi ) NM_001095238 . 11 . 3-3Mucin-2-like ( LOC100494747 ) XM_002936043 . 2321 . 7X . laevis uncharacterized protein ( MGC68450 ) NM_001089841 . 12 . 2-2 . 8X . laevis neuregulin alpha-1AF076618 . 11 . 40 . 82 . 7Potassium voltage-gated channel Isk-related ( kcne1 ) XM_004912135 . 12 . 21 . 52 . 7Olfactory receptor 5G3-like ( LOC100492086 ) XM_002942220 . 11 . 9-2 . 7Alpha-kinase 2 ( alpk2 ) XM_004910401 . 11 . 12 . 22 . 7X . laevis arginyl amino peptidase ( amino peptidase B ) b ( rnpep-b ) NM_001092079 . 1-2 . 71 . 7#X . laevis SRY-box containing protein ( Sox1 ) EF672727 . 1-2 . 62 . 1Copine II ( cpne2 ) transcript variant X1XM_004913481 . 111 . 22 . 6X . laevis hemoglobin , gamma A ( hbg1 ) NM_0010963471 . 2-2 . 6KIAA0895 protein ( kiaa0895 ) NM_001114073 . 11 . 62 . 6-#X . laevis empty spiracles homeobox 1gene 2 ( emx1 . 2 ) NM_001093430 . 12 . 61 . 91 . 1Homeobox B8 ( hoxb8 ) transcript variant X1XM_002938021 . 21 . 12 . 5-Monocyte to macrophage differentiation-associated ( mmd ) XM_004918560 . 1-1 . 22 . 5X . laevis uncharacterized ( LOC100036933 ) NM_001097704 . 11 . 51 . 52 . 5Finished cDNA clone TNeu143f19CR760056 . 22 . 22 . 5-Chromosome unknown open reading frame C2orf80XM_002937119 . 21 . 42 . 12 . 4#Single-minded homolog 1 ( sim1 ) transcript variant X2XM_004914545 . 1-1 . 42 . 4Transmembrane protein 2-like ( LOC100491930 ) XM_002932255 . 22 . 41 . 91 . 3PX domain containing 1 ( pxdc1 ) NM_001130262 . 11 . 4-2 . 4Aldehyde dehydrogenase 1 family member L2 ( aldh1l2 ) XM_002938070 . 20 . 91 . 32 . 3Uncharacterized ( LOC100490228 ) XM_002942932 . 21 . 8-2 . 3Beta-1 3-galactosyltransferase 2-like ( LOC101732799 ) XM_004918863 . 11 . 62 . 3-Alpha-2 3-sialyltransferase ST3Gal V ( st3gal5 ) FN550108 . 11 . 8-2 . 3X . laevis uncharacterized protein ( MGC64538 ) NM_001086337 . 1-1 . 62 . 3Transmembrane channel-like protein 7-like ( LOC100493700 ) XM_002932222 . 21 . 40 . 92 . 3Kinase insert domain receptor ( a type III receptor tyrosine kinase ) ( kdr ) XM_002934669 . 21 . 90 . 92 . 3Serine/threonine kinase 32A ( stk32a ) XM_002936707 . 21 . 32 . 2-Pancreatic lipase-related protein 2 ( pnliprp2 ) NM_001089647 . 12 . 10 . 72 . 2X . laevis nephrin ( NPHS1 ) AY902238 . 1-2 . 21 . 1Poly ( ADP-ribose ) polymerase 14-like ( LOC100485144 ) XM_002943546 . 222 . 21 . 2Frizzled family receptor 4 ( fzd4 ) XM_002936543 . 21 . 40 . 72 . 1Neuropeptide Y receptor Y2 ( npy2r ) XM_004911153 . 12 . 1-1 . 6Deoxyribonuclease gamma-like ( LOC100497175 ) XM_002938386 . 21 . 82 . 12X . laevis dehydrogenase/reductase ( SDR family ) member 11 ( dhrs11 ) NM_001094963 . 1-1 . 52 . 1X . laevis gamma-glutamyl hydrolase ( ggh ) NM_001092691 . 12 . 11 . 32Opsin-3-likeXM_002932623 . 211 . 22X . laevis transmembrane protein 56 ( tmem56-b ) NM_001086447 . 1-1 . 12X . laevis pyruvate dehyrogenase phosphatase catalytic subunit 1 ( pdp1 ) NM_001094221 . 11 . 521ArfGAP with SH3 domain ankyrin repeat and PH domain 3 ( asap3 ) XM_002939360 . 21 . 7-1 . 9#Early growth response 3 ( egr3 ) XM_002932703 . 21 . 60 . 81 . 9#POU class 4 homeobox 1 ( pou4f1 . 2 ) NM_001097307 . 11 . 311 . 9Activin beta B subunitS61773 . 1-1 . 71 . 8Monocyte to macrophage differentiation-associated ( mmd ) XM_002937811 . 21 . 71 . 11 . 8X . laevis ribosomal protein S2eBC130122 . 1-1 . 81 . 7X . laevis ras homolog family member V ( rhov ) NM_001128659 . 11 . 20 . 81 . 6X . laevis adenomatosis polyposis coli down-regulated 1 ( apcdd1 ) NM_001094109 . 11 . 211 . 6#X . laevis zinc finger protein 214 ( znf214 ) NM_001097042 . 11 . 20 . 81 . 5X . laevis cdc25BaAB363840 . 11 . 2-1 . 5X . laevis apelin ( apln-a ) NM_001097924 . 10 . 91 . 31 . 5Suppressor of cytokine signaling 2 ( socs2 ) NM_001095760 . 1-1 . 11 . 5#cAMP responsive element modulator ( crem ) XM_002935162 . 2-1 . 41 . 5X . laevis clone IMAGE:4684003BC042305 . 11 . 4-1 . 2#X . laevis ets-2a proto-oncogeneBC133183 . 11 . 311 . 4X . laevis similar to envoplakinBC045116 . 11 . 41 . 4-Ras homolog family member V ( rhov ) NM_001095566 . 11 . 411 . 2Samd9l protein ( samd9l ) XM_002943522 . 2-1 . 21 . 3Flocculation protein FLO11-like ( LOC100490389 ) XM_002942555 . 21 . 2-1 . 3c-Jun-amino-terminal kinase-interacting protein 4-like ( LOC100493724 ) XM_002939963 . 21 . 1-1 . 2X . laevis Dickkopf-1 ( Xdkk-1 ) AF030434 . 111 . 21 . 1X . laevis ectoderm neural cortex related-3 ( Encr-3 ) AY216793 . 11 . 10 . 81 . 2Uncharacterized ( LOC101730819 ) XM_004915204 . 10 . 91 . 11 . 2#X . laevis LIM class homeodomain proteinBC084744 . 11 . 10 . 71 . 1Ceramide kinase-like ( cerkl ) XM_002932015 . 21 . 41 . 32Mannosyl ( alpha-1 3- ) -glycoprotein beta-1 4-N-acetylglucosaminyltransferase ( mgat4b ) NM_001091975 . 12-1 . 8Fibroblast growth factor 19 ( fgf19 ) NM_001142825 . 1-21 . 5#F-box protein 41 ( fbxo41 ) NM_001079043 . 11 . 30 . 62Avidin-like ( LOC100487365 ) XM_002939983 . 221 . 6-Four and a half LIM domains 2 ( fhl2 ) NM_001126761 . 1-1 . 11 . 9Metalloprotease TIKI1-like ( LOC100491951 ) XM_002936336 . 21 . 11 . 41 . 9X . laevis Kazal-type serine peptidase inhibitor domain 1 ( kazald1 ) NM_001092073 . 11 . 61 . 11 . 9Uncharacterized ( LOC101734664 ) XM_004910525 . 11 . 20 . 61 . 9X . laevis similar to calsequestrin 2 ( cardiac muscle ) BC097545 . 11 . 81 . 51 . 9X . laevis COMM domain containing 3 ( commd3 ) NM_001095386 . 11 . 11 . 90 . 6X . laevis alcohol dehydrogenase iron containing1 ( adhfe1 ) NM_001127802 . 1-1 . 91 . 2X . laevis ectonucleoside triphosphate diphosphohydrolase 1 ( entpd1 ) NM_001092268 . 11 . 80 . 61 . 3#Protein fosB-like transcript variant X2XM_004916957 . 1-1 . 71 . 4Tocopherol ( alpha ) transfer protein ( ttpa ) NM_001008184 . 1-1 . 71 . 6X . laevis tetratricopeptide repeat domain 39B ( ttc39b ) NM_001094701 . 11 . 1-1 . 7#X . laevis Tbx6 ( Tbx6 ) DQ355794 . 11 . 41 . 71X . laevis uncharacterized ( LOC100036989 ) NM_001097746 . 1-1 . 31 . 7X . laevis cDNA clone IMAGE:6947552BC093552 . 11 . 31 . 7-B-cell CLL/lymphoma 10 ( bcl10 ) NM_001015777 . 21 . 7-1 . 2Uncharacterized ( LOC100494710 ) XM_002939048 . 21 . 41 . 6-X . laevis keratin 17 ( krt17 ) NM_001094941 . 1-1 . 21 . 6Membrane metallo-endopeptidase-like 1 ( mmel1 ) NM_001127095 . 10 . 91 . 11 . 6Putative methyltransferase KIAA1456 homologXM_002934674 . 21 . 1-1 . 6Phospholipase Cdelta 3 ( plcd3 ) XM_002935518 . 21 . 11 . 51 . 6IdnK gluconokinase homolog ( E . coli ) ( idnk ) NM_001126592 . 11 . 40 . 91 . 5Uncharacterized ( LOC100486093 ) transcript variant X2XM_002939117 . 21 . 5-1 . 5X . laevis similar to calsequestrin 2 ( cardiac muscle ) BC041283 . 11 . 1-1 . 5Piwi-like RNA-mediated gene silencing 2 ( piwil2 ) NM_001112999 . 11 . 1-1 . 5Zinc finger and BTB domain containing 20 ( zbtb20 ) XM_002939649 . 21 . 4-1 . 1#V-maf musculoaponeurotic fibrosarcoma oncogene homolog A ( mafa ) NM_001032304 . 11 . 40 . 91 . 1X . laevis uncharacterized protein ( MGC81120 ) NM_001091225 . 11 . 40 . 91 . 3#Single-minded homolog 1 ( Drosophila ) ( sim1 ) transcript variant X3XM_004914546 . 11 . 11 . 31 . 2Xenopus laevis alpha-2-macroglobulin-like 1 ( a2ml1 ) NM_001135077 . 11 . 1-1 . 1X . laevis chromogranin A ( parathyroid secretory protein 1 ) ( chga ) NM_001094724 . 11 . 61 . 42 . 2X . laevis lipaseendothelial ( lipg ) NM_001090061 . 11 . 21 . 30 . 6G protein-coupled receptor 56 ( gpr56 ) XM_002931653 . 21 . 7-1 . 6X . laevis family with sequence similarity 101member B ( fam101b ) NM_001093870 . 11 . 50 . 81 . 5X . laevis CD81 protein ( cd81-a ) NM_001086613 . 10 . 71 . 11 . 9X . laevis calbindin D28kBC170542 . 12 . 2-3 . 1X . laevis ATPaseNa+/K+ transportingbeta 1 polypeptide ( atp1b1 ) NM_001086759 . 11 . 211 . 7X . laevis 7-transmembrane receptor frizzled-1AF231711 . 11 . 412X . laevis prostaglandin reductase 2 ( ptgr2 ) NM_001079334 . 11 . 41 . 5-X . laevis TGF-beta2 for transforming growth factor-beta2X51817 . 11 . 3-1 . 1#SRY ( sex determining region Y ) -box 2 ( sox2 ) NM_213704 . 31 . 11 . 31 . 9#X . laevis for enhancer of split related 9 ( esr9 gene ) AJ009282 . 11 . 71 . 6-X . laevis mal T-cell differentiation protein ( mal ) NM_001086577 . 1-1 . 21 . 4Transmembrane proteaseserine 13 ( tmprss13 ) XM_002932904 . 21 . 51 . 11 . 9X . laevis Ras-related associated with diabetes ( rrad ) NM_001092750 . 184 . 64 . 2Integrin beta 4 ( itgb4 ) transcript variant X1XM_002939974 . 21 . 4-2 . 2Xenopus ( Silurana ) tropicalis FERM domain containing 4A ( frmd4a ) XM_002935243 . 21 . 10 . 61 . 3X . laevis complement factor I ( cfi-a ) NM_001085952 . 11 . 41 . 21 . 6#X . laevis SIX homeobox 1 ( six1 ) NP_001082027 . 11 . 41 . 22 . 3FH2 domain-containing protein 1-like ( LOC100496216 ) XM_002934907 . 21 . 90 . 91 . 9#X . laevis mab-21-like 2 ( mab21l2-b ) NM_001096770 . 1-2 . 82 . 9X . laevis regulator of cell cycle ( rgcc ) NM_001093976 . 11 . 31 . 11 . 7X . laevis Cep63FJ464988 . 1-1 . 42 . 3X . laevis CD81 antigen ( target of anti proliferative antibody 1 ) BC041217 . 11 . 71 . 12Transmembrane serine protease 9BC087611 . 11 . 11 . 11 . 2#X . laevis POU class 3 homeobox 2 ( pou3f2-b ) NM_001096751 . 132 . 32 . 9G protein-coupled receptor 153 ( gpr153 ) NM_001128052 . 12 . 51 . 11 . 5#X . laevis Myoblast determination protein 1 homolog ABC041190 . 13 . 52 . 74 . 7#T-cell leukemia homeobox 1 ( tlx1 ) transcript variant 1XM_002936768 . 22 . 62 . 32 . 6X . laevis neurotrophin 3 ( ntf3 ) NM_001092740 . 11 . 41 . 51 . 9X . laevis p21 GTPase-associated kinase 1 ( PAK1 ) AF000239 . 11 . 2-2 . 1#X . laevis hairy and enhancer of split 9 , gene 1 ( hes9 . 1-b ) NP_001089097 . 11 . 81 . 51 . 6X . laevis tetraspanin 1 ( tspan1 ) NM_001095473 . 11 . 20 . 71 . 3X . laevis uncharacterized protein ( MGC83079 ) NM_001091250 . 121 . 5-X . laevis cDNA clone IMAGE:5085355BC073731 . 11 . 3-1 . 4Family with sequence similarity 198member A ( fam198a ) XM_002937853 . 21 . 70 . 71 . 3Progestin and adipoQ receptor family member IX ( paqr9 ) XM_004914351 . 11 . 7-1 . 2#Hairy and enhancer of split 8 ( hes8 ) XM_002933849 . 22 . 81 . 73 . 6X . laevis p21 GTPase-associated kinase 1BC081113 . 11 . 30 . 81 . 7Finished cDNA clone TNeu008g03CR761907 . 21 . 21 . 10 . 7WD repeat domain 27 ( wdr27 ) XM_002931515 . 21 . 22 . 21 . 1#Growth factor independent 1 transcription repressor ( gfi1 ) XM_002933803 . 21 . 81 . 83 . 2Protein phosphatase 2 regulatory subunit B'beta ( ppp2r5b ) NM_001100279 . 12 . 41 . 44 . 2Ornithine decarboxylase antizyme 2 ( oaz2 ) , transcript variant 2NP_001106583 . 21 . 8-1 . 5X . laevis fast troponin T ( TNNT3 ) AY114144 . 1-1 . 11 . 5#X . laevis xRipply3 for xRipply3 proteinAB455086 . 10 . 91 . 12RAS-like family 11member B ( rasl11b ) NM_001015774 . 1-1 . 21 . 4X . laevis for thimet oligopeptidaseBC070748 . 13 . 8-2X . laevis fibroblast growth factor 3 ( fgf3 ) NM_001008153 . 121 . 22X . laevis cDNA clone IMAGE:8332229BC155363 . 11 . 50 . 91 . 4Proline rich 15 ( prr15 ) XM_002933381 . 21 . 6-1 . 3Integrin beta 6 ( itgb6 ) NM_001097306 . 12 . 30 . 62 . 8#Xenopus laevis empty spiracles homeobox 1 , gene 2 ( emx1 . 2 ) NM_001093430 . 12 . 61 . 42 . 1X . laevis p21-activated kinase ( PAK1 ) AF169794 . 11 . 41 . 82 . 6#ISL LIM homeobox 2 ( isl2 ) NM_001166041 . 11 . 5-1 . 7#Atonal homolog 1 ( Drosophila ) ( atoh1 ) XM_004911085 . 10 . 91 . 11 . 5Ectodysplasin A receptor ( edar ) NM_001087047 . 14 . 3-3 . 3X . laevis degr03DQ096846 . 12 . 12 . 22Calcyphosine ( caps ) NM_001097320 . 1-1 . 43 . 7X . laevis kiaa0930NM_001086221 . 11 . 511 . 6Putative N-acetyltransferase 16-like ( LOC100490742 ) XM_002943189 . 12 . 111 . 7#T-box 15 ( tbx15 ) XM_002940981 . 2211 . 8#SRY ( sex determining region Y ) -box 1 ( sox1 ) NM_001080996 . 10 . 61 . 51 . 2Cytochrome P450 family 2 subfamily C polypeptide 18 ( cyp2c18 ) NM_001091776 . 12 . 11 . 41 . 6X . laevis calcitonin receptor-like ( calcrl ) NM_001086737 . 11 . 10 . 81 . 6X . laevis claudin 3 ( cldn3 ) NM_001005709 . 12 . 11 . 31 . 5Atlastin GTPase 1 ( atl1 ) NM_001078754 . 11 . 821 . 7Rho GTPase activating protein 9 ( arhgap9 ) , transcript variant X2XM_0129578291 . 81 . 23 . 4#X . laevis Hes2BC084134 . 11 . 70 . 91 . 3X . laevis U3 snRNAX07318 . 112 . 81 . 1Uncharacterized ( LOC101732195 ) XM_004912378 . 12-1 . 5Tumor necrosis factor receptor superfamilymember 21 ( tnfrsf21 ) NM_001079136 . 11 . 10 . 81 . 2X . laevis arginase 3U08408 . 1-1 . 31 . 8ChaC cation transport regulator homolog 1 ( chac1 ) XM_002939546 . 21 . 21 . 31 . 5X . laevis DIRAS familyGTP-binding RAS-like 3 ( diras3 ) NM_001095243 . 10 . 81 . 71 . 4X . laevis DnaJ ( Hsp40 ) homolog subfamily C member 27 ( dnajc27-b ) NM_001095422 . 11 . 10 . 81 . 1* Genes are ranked by FC value , using the highest FC in each of the three treatment groups . Genes included must have FC ≥ 1 in at least two out of the three treatment groups as well as showing at least a two-fold difference in FC to the un-injected control ( not shown ) . Corresponding values ≥0 . 5 are shown for all treatments . † Log2 Fold change values after Six1-GR overexpression . ‡ Log2 Fold change values after Eya1-GR overexpression . § Log2 Fold change values after Six1-GR+Eya1-GR overexpression . # Denotes transcription factors with at least a two-fold change in at least two treatment groups selected for further analysis . 10 . 7554/eLife . 17666 . 008Table 2 . Transcription factors and co-factors selected for characterisation by in-situ-hybridisation ranked by FC value in individual treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 008AnnotationGeneAccessionIndividualMergedSix1*Eya1†Six1+Eya1‡Six1§Eya1#Six1+Eya1¶SIX homeobox 2 ( Six2 ) Six2NM_001100275 . 153 . 55 . 95 . 4**5**4 . 9**X . laevis for Xsox17-alpha proteinSox17AJ001730 . 13 . 62 . 64 . 84 . 4**3 . 3**3 . 5**X . laevis Myoblast determination protein 1 homolog AMyoD1BC041190 . 13 . 52 . 74 . 74 . 1**4 . 2**3 . 9**Hairy and enhancer of split 8 ( Hes8 ) Hes8XM_002933849 . 22 . 81 . 73 . 63 . 2**3 . 2**3 . 1**Growth factor independent 1 transcription repressor ( Gfi1 ) Gfi1aXM_002933803 . 21 . 81 . 83 . 22 . 4**2 . 6**4 . 1**X . laevis POU class 3 homeobox 2 ( Pou3f2-b ) Pou3f2bNM_001096751 . 132 . 32 . 93**2 . 6**2 . 7**X . laevis Mab-21-like 2 ( Mab21l2-b ) Mab21l2bNM_001096770 . 1-2 . 82 . 9---T-cell leukemia homeobox 1 ( Tlx1 ) transcript variant 1Tlx1XM_002936768 . 22 . 62 . 32 . 62 . 4**2 . 4**2 . 4**X . laevis empty spiracles homeobox 1 gene 2 ( Emx1 . 2 ) Emx1 . 2NM_001093430 . 12 . 61 . 91 . 1--1 . 7**X . laevis SRY-box containing protein ( Sox1 ) Sox1EF672727 . 1-2 . 62 . 1-2**-Single-minded homolog 1 ( Sim1 ) transcript variant X2Sim1XM_004914545 . 1-1 . 42 . 4---X . laevis SIX homeobox 1 ( Six1 ) Six1AF279254 . 11 . 41 . 22 . 31 . 9**1 . 6**1 . 6**F-box protein 41 ( Fbxo41 ) Fbxo41NM_001079043 . 11 . 30 . 62---T-box 15 ( Tbx15 ) Tbx15XM_002940981 . 2211 . 82**1 . 4**1 . 7**X . laevis xRipply3 for xRipply3 proteinRipply3AB455086 . 10 . 91 . 121 . 6**1 . 4**1 . 3**Early growth response 3 ( Egr3 ) Egr3XM_002932703 . 21 . 60 . 81 . 91 . 7**1 . 3**1 . 9**SRY ( sex determining region Y ) -box 2 ( Sox2 ) Sox2NM_213704 . 31 . 11 . 31 . 91 . 6**1 . 6**1 . 5**POU class 4 homeobox 1 ( Pou4f1 . 2 ) Pou4f1 . 2NM_001097307 . 11 . 311 . 91 . 6**1 . 5**1 . 5**X . laevis for enhancer of split related 9 ( Esr9 gene ) Hes9 . 1aAJ009282 . 11 . 71 . 6----ISL LIM homeobox 2 ( Isl2 ) Isl2NM_001166041 . 11 . 5-1 . 71 . 6**1 . 1**1 . 4**X . laevis Tbx6 ( Tbx6 ) Tbx6DQ355794 . 11 . 41 . 71---Protein FosB-like transcript variant X2FosBXM_004916957 . 1-1 . 71 . 4-1 . 4**1 . 2**X . laevis Hes2Hes2BC084134 . 11 . 70 . 91 . 3---cAMP responsive element modulator ( Crem ) CremXM_002935162 . 2-1 . 41 . 5-1 . 4**1 . 2**X . laevis zinc finger protein 214 ( Znf214 ) Znf214NM_001097042 . 11 . 20 . 81 . 51 . 2**5 . 9**5 . 8**Xenopus laevis SRY ( sex determining region Y ) -box 21 ( Sox21 ) Sox21NM_001172213 . 11 . 20 . 61 . 51 . 4**1 . 2**1 . 2**Atonal homolog 1 ( Drosophila ) ( Atoh1 ) Atoh1XM_004911085 . 10 . 91 . 11 . 5111X . laevis Ets-2a proto-oncogeneEts2aBC133183 . 11 . 311 . 41 . 3**1 . 2**1 . 2**V-maf musculoaponeurotic fibrosarcoma oncogene homolog A ( Mafa ) MafaNM_001032304 . 11 . 40 . 91 . 11 . 9**-1 . 8**X . laevis LIM class homeodomain proteinLhx5BC084744 . 11 . 1-1 . 1---Xenopus ( Silurana ) tropicalis neurogenin 1 ( Neurog1 ) Ngn1NM_001123423 . 10 . 80 . 90 . 80 . 80 . 80 . 8**Xenopus laevis SOX3 proteinSox3BC072222 . 10 . 5-0 . 90 . 70 . 70 . 6*Log2 fold change values after Six1 overexpression ( Six1i ) . † Log2 fold change values after Eya1 overexpression ( Eya1i ) . ‡ Log2 fold change values after Six1+Eya1 overexpression ( Six1+ Eya1i ) . § Log2 fold change values after overexpression of Six1 or Six1+Eya1 ( Six1m ) . # Log2 fold change values after overexpression of Eya1 or Six1+Eya1 ( Eya1m ) . ¶ Log2 fold change values after overexpression of Six1 or Eya1 or Six1+Eya1 ( Six1+Eya1m ) . ** Denotes statistically supported data ( q < 0 . 05 ) . The expression of genes previously undescribed in Xenopus ( Crem , FosB , Hes8 , Isl2 , Tbx15 , Znf214 ) was fully characterised in neural fold , and early and late tail bud stages , along with those for which expression has been described for relatively few stages ( Atoh1 , Emx1 . 2 , Gfi1a , Hes2 , Hes9 , Lhx5 , Mab21l2b , Pou3f2b , Pou4f1 . 2 , Ripply3 , Sim1 , Sox21 , Tbx6 , Tlx1 ) ( summarised in Figure 4A–T; Figure 4—figure supplements 1–5 ) . Genes with extensively characterised expression patterns ( Ngn1 ( Nieber et al . , 2009 ) ; Six1 ( Pandur and Moody , 2000 ) ; Six2 ( Ghanbari et al . , 2001 ) ; Sox2 ( Mizuseki et al . , 1998 ) ; Sox3 ( Penzel et al . , 1997 ) ; Sox17 ( Hudson et al . , 1997 ) , MyoD1 ( Hopwood et al . , 1989 ) , Sox1 ( Nitta et al . , 2006 ) , Ets2a ( Salanga et al . , 2010 ) , Mafa ( Coolen et al . , 2005 ) ) are not shown here . 10 . 7554/eLife . 17666 . 009Figure 4 . Expression of selected presumptive direct targets of Six1/Eya1 in whole-mount Xenopus embryos . Genes expressed at neural plate stages ( stages 14–18 ) are shown in panels A–L , and those only expressed at later stages are shown at mid/late tail bud stage ( stages 28–32 ) in panels M–T . Several of the genes surveyed ( Lhx5 , Pou3f2b , Tbx15 , Tbx6 , Emx1 . 2 and Sim1 ( A–D , M and N ) , were not expressed in the PPE , nor any placodal derivatives in later stages . Instead , such genes were expressed in the adjacent neural folds ( Lhx5 , Pou3f2b and Tbx15 ) , ectoderm ( Tbx6 ) , or in the forebrain at later stages ( Emx1 . 2 ) . Several other genes were expressed broadly across the cranial ectoderm , at least partially overlapping with the PPE at neural plate stages ( Ripply3 , Crem , FosB and Znf214; E , O–Q ) , some of which are also maintained in placodal derivatives such as Znf214 in the otic vesicle ( Q ) . The remaining genes ( F–L , R–T ) are expressed in parts of the PPE and maintained in some placodes ( Hes2 , Hes8 , Hes9 , Mab21l2b , Sox21 , Isl2 , Pou4f1 . 2 , and Tlx1 ) or are expressed in a subset of placodes only ( Atoh1 , Gfi1a ) ( see Figure 4—figure supplements 1–4 for additional stages ) . Yellow arrows mark placodal expression . Arrowheads mark non-placodal expression . Abbreviations: pA: anterior placodal region; pAD: anterior lateral line placode; pE: epibranchial placode; pL: lens placode; L: lens; pM: middle lateral line placode; pO: olfactory placode; pP: posterior placodal region; pPrV: profundal/trigenimal placodes; vOt: otic vesicle . Plots U and V show qPCR after Six1 or Eya1 overexpression . Log2 fold change values were calculated from qPCR data obtained after overexpression of Six1-GR ( U ) or Eya1-GR ( V ) in placodal explants and are shown next to corresponding fold change values obtained from the RNA-Seq data . In all cases shown , qPCR values broadly corroborate those from the RNA-Seq data - showing up-regulation of target genes after either Six1 or Eya1 overexpression . Vertical error bars show the standard deviation of the mean of biological triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 00910 . 7554/eLife . 17666 . 010Figure 4—figure supplement 1 . Expression of targets not expressed in placodes in whole-mount Xenopus embryos . Expression dynamics for each target are shown across a range of developmental stages: A1–F1 show expression in neural plate stage embryos , A2–F2 show early tail bud stage embryos and A3–F3 show late tail bud stage embryos . ( A ) Expression of PPE marker gene Six1 is shown as reference for placodal domains ( for details see Pandur and Moody , 2000; Schlosser and Ahrens , 2004 ) . ( B ) Emx1 . 2 is expressed broadly in the neural plate in neural plate stages ( B1 ) , and becomes restricted to the forebrain in late tail bud stages ( B3; asterisk ) . ( C ) Lhx5 is expressed in the forebrain at all developmental stages ( C1–C3; asterisk ) , and at early and late tail bud stages Lhx5 is also expressed in the hindbrain and spinal cord ( C3; arrowhead ) . ( D ) Pou3f2b is expressed in the neural plate and developing neural tube ( D1; asterisk ) at neural plate stages . Expression in the brain and spinal cord is maintained during early and late tail bud stages ( D2 and D3 ; arrowhead and asterisk , respectively ) . ( E ) Tbx15 is expressed in a restricted domain of the anterolateral neural folds in neural plate stages ( E1; asterisk ) . At tail bud stages expression is prominent in somites ( E2 and E3; arrowhead ) and migrating neural crest cells of the hyoid and first branchial neural crest streams ( Nc ) . Both of these expression domains are maintained into late tail bud stages ( E3 and E4 ) . E4 shows section at the level indicated in E3 ( dotted line ) . Bar in E4: 100 μm . ( F ) Throughout all developmental stages ( F1–F3 ) Tbx6 is expressed strongly in the posterior paraxial and lateral plate mesoderm ( F1 and F2; asterisk ) with weaker expression in the pharyngeal arches ( F2; arrowhead ) . Subsequently , it’s expressed in somites , as indicated by a diamond in F2 . Abbreviations: pA: anterior placodal region; pAD: anterior lateral line placode; pE: epibranchial placode; pM: middle lateral line placode; pO: olfactory placode; vOt: otic vesicle; pPrV: profundal/trigeminal placodes . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 01010 . 7554/eLife . 17666 . 011Figure 4—figure supplement 2 . Expression of targets that broadly overlap with PPE in whole-mount Xenopus embryos . Expression dynamics for each target are shown across a range of developmental stages: A1–E1 show expression in neural plate stage embryos , A2–E2 show early tail bud stage embryos and A3–E3 show late tail bud stage embryos . ( A ) Expression of PPE marker gene Six1 is shown as reference for placodal domains ( for details see Pandur and Moody , 2000; Schlosser and Ahrens , 2004 ) . ( B ) Crem is initially expressed broadly in paraxial mesoderm ( B1; asterisk ) and cranial ectoderm ( B1; arrowhead ) at neural plate stages and in pharyngeal arches and overlying ectoderm at early and late tail bud stages ( B2 and B3 ; diamond ) . ( C ) FosB is expressed in a broad pattern across the cranial ectoderm and trunk mesoderm at both neural fold and early tail bud stages ( C1 and C2 ) . At late tail bud stages expression is maintained in cranial ectoderm as well as becoming apparent in the migrating neural crest cells ( Nc ) and weakly in the somites ( C3; arrowhead ) . ( D ) Znf214 is expressed broadly across the ectoderm at all developmental stages ( D1–D3 ) . At both early and late tail bud stages there is expression in the migrating neural crest cells ( Nc ) as well as in the retina ( D2 and D3; asterisk ) , and in late tail bud stages Znf214 is expressed in the otic vesicle and lens . ( E ) Ripply3 is expressed broadly in the posterior placodal region at neural fold stages ( E1 ) . At both early and late tail bud stages expression is confined to posterior cranial ectoderm ( E2 and E3; asterisk ) . Yellow arrows mark placodal expression . Abbreviations: pA: anterior placodal region; pAD: anterior lateral line placode; pE: epibranchial placode; L: lens; pM: middle lateral line placode; pO: olfactory placode; vOt: otic vesicle; pP: posterior placodal region; pPrV: profundal/trigeminal placodes . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 01110 . 7554/eLife . 17666 . 012Figure 4—figure supplement 3 . Expression of targets with dynamic/transient expression pattern in placodes in whole-mount Xenopus embryos . Expression dynamics for each target are shown across a range of developmental stages: A1–E1 show expression in neural plate stage embryos , A2–E2 show early tail bud stage embryos and A3–E3 show late tail bud stage embryos . ( A ) Expression of PPE marker gene Six1 is shown as reference for placodal domains ( for details see Pandur and Moody , 2000; Schlosser and Ahrens , 2004 ) . ( B ) Hes2 is expressed very strongly in a broad region corresponding to the posterior placodal domain including the prospective otic and lateral line placodes , as well as weakly in a scattered subset of neuroectodermal cells ( B1; asterisk ) at neural plate stages ( see Figure 4—figure supplement 5 for section ) . Expression is later restricted to the otic vesicle and a new expression domain becomes established in the developing retina at early and late tail bud stages ( B2 and B3; arrowhead ) . ( C ) During neural plate stages ( C1 ) , Hes8 is expressed in the developing profundal and trigeminal placodes as well as in the anterior placodal region , the anterior neural plate ( C1 ; asterisk ) and the primary neurons ( motor neurons , intermediary neurons and sensory neurons ) of the posterior neural plate ( C1; arrowheads; see Figure 4— figure supplement 5 for section ) . In early tail bud stages ( C2 ) trigeminal expression is lost and replaced by expression in the otic vesicle , as well as lateral line , epibranchial and olfactory placodes . Throughout late tail bud stages ( C3 ) , expression is maintained in these regions and the brain ( C3; cross ) and is initiated in the retina ( C3; diamond ) . ( D ) During neural plate stages ( D1 ) , Hes9 is expressed in the developing profundal and trigeminal placodes as well as in the anterior placodal region , the anterior neural plate ( D1; asterisk ) and the primary neurons ( motor neurons , intermediary neurons and sensory neurons ) of the posterior neural plate ( D1; arrowheads; see Figure 4—figure supplement 5 for section ) . In early tail bud stages ( D2 ) trigeminal expression is lost but expression is apparent in the olfactory placodes , as well as in the otic and lateral line placodes and retina ( D2; diamond ) . In late tail bud stages ( D3 ) Hes9 is expressed broadly thoughout the brain ( D3 ; cross ) , and is maintained in the lateral line and olfactory placodes as well as the otic vesicle . ( E ) During neural plate stages ( E1 ) , Mab21l2b is expressed in the prospective lens placode , as well as in the eye field ( prospective retina ) of the forebrain ( E1; asterisk ) and the prospective midbrain ( E1; arrowhead ) . At early tail bud stages expression in the lens and midbrain is maintained and its expression becomes apparent in the hindbrain ( E2; diamond ) . In late tail bud stages Mab21l2b is additionally prominently expressed in migrating neural crest cells ( E3 ; Nc ) . ( F ) During neural plate stages Sox21 is expressed broadly throughout the anterior neural plate ( F1; asterisk ) . At early tail bud stages ( F2 ) , this expression becomes confined to the forebrain ( asterisk in F2 ) and midbrain-hindbrain boundary ( arrowhead in F2 ) and is maintained into late tail bud stages , ( F3 ) . In late tail bud stages , Sox21 is also expressed in the olfactory placode , otic vesicle and becomes up-regulated in the hindbrain ( F3; diamond ) . Yellow arrows mark placodal expression . Dotted lines in B1–D1 indicate levels of sections shown in Figure 4—figure supplement 5 . Abbreviations: pA: anterior placodal region; pAD: anterior lateral line placode; pE: epibranchial placode; pLl: lateral line placodes; L: lens; pL: lens placode; pM: middle lateral line placode; pO: olfactory placode; vOt: otic vesicle; pP: posterior placodal region; pPrV: profundal/trigeminal placodes . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 01210 . 7554/eLife . 17666 . 013Figure 4—figure supplement 4 . Expression of targets with persistent expression in placodes in whole-mount Xenopus embryos . Expression dynamics for each target are shown across a range of developmental stages: A1–F1 show expression in neural plate stage embryos , A2–F2 show early tail bud stage embryos and A3–F3 show late tail bud stage embryos . B4–D4 and B5 –D5 shows sections at level indicated in B3–D3 , respectively ( dotted lines ) . ( A ) Expression of PPE marker gene Six1 is shown as reference for placodal domains ( for details see Pandur and Moody , 2000; Schlosser and Ahrens , 2004 ) . ( B ) Atoh1 is initially expressed at very low levels in presumptive otic placodes at neural plate stages ( B1 ) . This expression becomes more pronounced in the otic vesicle at early tail bud stages ( B2 ) concomitant with the initiation of expression in lateral line ganglia and strong expression in the hindbrain ( B2; asterisk ) . Expression becomes more pronounced in all three regions at late tail bud stages ( B3 –B5 ) . ( C ) Gfi1a is expressed at high levels in haematopoietic cells during neural plate stages ( C1; asterisk ) . At early tail bud stages ( C2 ) expression becomes more pronounced and diffuse , and expression is also initiated in the otic vesicle . At late tail bud stages Gfi1a is expressed in lateral line placodes as well as otic vesicles as the haematopoietic expression begins to subside ( C3–C5 ) . ( D ) During neural plate stages Isl2 is expressed in the profundal and trigeminal placodes and in the anterior placodal region along the anterior edge of the neural plate ( D1 ) . At early tail bud stages Isl2 expression is maintained in the profundal and trigeminal placodes/ganglia as well as in otic and lateral line placodes/ganglia and primary neurons in the spinal cord ( D2; asterisk ) . Expression is maintained in cranial ganglia at late tail bud stages ( D3–D5 ) and becomes apparent in the forebrain and lens ( D3; arrowhead ) . ( E ) During neural plate stages Pou4f1 . 2 is expressed in the profundal and trigeminal placodes as well as in a stripe of primary sensory neurons ( E1; asterisk; see Figure 4—figure supplement 5 for section ) . In early tail bud stages ( E2 ) expression in the profundal/trigeminal placodes/ganglia and primary neurons is maintained , and expression in the otic and lateral line placodes is strengthened . Expression is maintained in all domains as well as in the cranial ganglia derived from placodes into late tail bud stages when expression becomes up-regulated in the retina ( E3; diamond ) . Dotted line in E1 indicates the level of section shown in Figure 4— figure supplement 5 . ( F ) Tlx1 is expressed in the presumptive ventral visceral arches at neural plate stages ( F1; asterisk ) . This is maintained into early and late tail bud stages ( F2 and F3 ) , which also exhibit prominent expression in the profundal/trigeminal placodes and ganglia and the otic vesicle . Yellow and black arrows mark placodal expression . Bar in B4 , C4 and D4: 100 μm ( also for B5 , C5 and D5 , respectively ) . Abbreviations: pA: anterior placodal region; pAD: anterior lateral line placode; gAD: ganglion of the anterodorsal lateral line nerve; pE: epibranchial placode; pLl: lateral line placodes; L: lens; pL: lens placode; pM: middle lateral line placode; pO: olfactory placode; vOt: otic vesicle; pOt: presumptive otic placode; pPr: profundal placode; pP: posterior placodal region; pPL: posterior lateral line placode; pPrV: profundal/trigeminal placodes; pV: trigeminal placode; gV: ganglion of the trigeminal nerve . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 01310 . 7554/eLife . 17666 . 014Figure 4—figure supplement 5 . Expression of selected targets in pre-placodal ectoderm ( PPE ) in sections through neural plate stage Xenopus embryos . Neural crest and ( NC ) and neural plate ( NP ) domains are indicated . ( A–D ) : Hes2 ( A ) , Hes8 ( B ) , Hes9 ( C ) and Pou4f1 . 2 ( D ) are all expressed in parts of the PPE ( for level of sections see Figure 4 —figure supplements 3 and 4 ) . While the exact boundaries of NC and NP cannot be determined in these sections , comparisons with sections through embryos stained by double in-situ-hybridisation for the PPE marker Six1 and the NP marker Sox3 ( E ) or Six1 and the NC marker FoxD3 ( F ) indicate that expression of each of these target genes is largely confined to the PPE although some overlap with the lateral NC region cannot be ruled out ( E and F modified from Schlosser and Ahrens , 2004; Figure 6 ) . Bar in A: 100 μm ( also for B–F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 014 We found 19/30 ( 63 . 3% ) of these transcriptional regulators to be expressed in PPE or placodal derivatives , while 11/30 ( Sox17 , MyoD1 , Sox1 , Ets2a , Mafa , Emx1 . 2 , Lhx5 , Pou3f2b , Tbx6 , Tbx15 , Sim1 ) were not expressed in either the PPE or any placodal derivatives . However , many of the genes in the latter group were expressed in the adjacent neural folds or other tissues . Thus , it is possible that such genes may be direct targets of Six1 or Eya1 in domains surrounding the PPE , likely to have been included in our screen as a result of non-PPE contamination during dissection . Of the transcriptional regulators identified in our list of well-supported targets and expressed in the PPE or placodal derivatives 79% ( 15/19 ) were statistically supported in the analyses of merged datasets ( Table 2 ) . These included genes broadly expressed in cranial ectoderm including the PPE ( Crem , FosB , Znf214 , Ripply3 ) , and genes expressed in the PPE or subdomains of the PPE and subsequently in some placodes ( Hes2 , Hes8 , Hes9 , Mab21l2b , Six1 , Six2 , Sox2 , Sox3 , Sox21 , Atoh1 , Ngn1 , Gfi1a , Isl2 , Pou4f1 . 2 , Tlx1 ) ( Figure 4A–T ) . To begin to elucidate the GRN downstream of Six1 and Eya1 we chose ten transcription factors showing expression in posterior placodes ( i . e . those derived from the posterior placodal area; the lateral line , otic and epibranchial placodes ) for additional functional studies including genes implicated in the maintenance of neuronal progenitors ( Sox2 , Sox3 , Hes8 and Hes9 ) as well as genes implicated in the regulation of sensory or neuronal differentiation ( Atoh1 , Gfi1a , Isl2 , Ngn1 , Pou4f1 . 2 and Tlx1 ) . Selected genes were independently verified as being direct targets of either Six1 ( Isl2 ) or of both Six1 and Eya1 ( all other targets; Sox3 not analysed ) in the PPE by qPCR , and the results were broadly consistent with the RNA-Seq data ( Figure 4U and V ) . To explore whether Six1 or Eya1 were required for the expression of selected target genes , the expression of each target was analysed by in-situ-hybridisation after MO-mediated knockdown of Six1 or Eya1 . The efficacy and specificity of both co-injected Six1-MOs ( Six1-MO1 and Six1-MO2; Brugmann et al . , 2004 ) and Eya1-MOs ( Eya1-MO1 and Eya1-MO2; Schlosser et al . , 2008 ) have been previously reported . Compared to injection with a control MO ( Eya1-mmMO with 5 mismatches relative to Eya1-MO2 ) , knockdown of either Six1 or Eya1 significantly reduced the expression of all direct target genes in PPE or placodes , demonstrating that both Six1 and Eya1 are required for their expression ( Figure 5 and Figure 5—figure supplement 1; Table 3 ) . To control for off-target effects associated with MO use , target gene expression was also analysed after overexpression of a dominant-negative version of Six1 ( Six1-EnR; Brugmann et al . , 2004 ) . Expression patterns of all target genes were highly similar to those seen after MO-knockdown of either Six1 or Eya1 , suggesting that the observed reductions in expression were caused by Six1 or Eya1 knockdown as opposed to being an artefact of MO use ( Figure 5—figure supplement 2 ) . Taken together , these findings show that Six1 and Eya1 are essential direct upstream regulators of multiple genes encoding transcription factors that promote neuro- and sensorigenesis in the PPE and placodes . 10 . 7554/eLife . 17666 . 015Figure 5 . Effects of Eya1 knockdown on target genes . Tail bud ( A–G ) and neural plate ( H–I ) stage embryos after unilateral injection of Eya1-MO1+2 . In each case , lacZ was co-injected as a lineage tracer and panels A1–G1 show the control ( un-injected ) side and A2–G2 show the injected side ( lacZ staining out of frame in most specimens ) . The injected side is positioned to the right in H–J , as marked by blue lacZ staining . Arrows and arrowheads mark reductions in marker gene expression in placodal and non-placodal derivatives , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 01510 . 7554/eLife . 17666 . 016Figure 5—figure supplement 1 . Effects of Six1 knockdown on target genes . Tail bud stage ( A–G ) and neural plate stage ( H–J ) embryos after unilateral injection of Six1-MO1+2 . In each case , lacZ was co-injected as a lineage tracer and panels A1–G1 show the control ( un-injected ) side and A2–G2 show the injected side . The injected side is positioned to the right in H–J , as marked by blue lacZ staining ( lacZ staining out of frame in some specimens ) . Arrows mark reductions in marker gene expression in placodal derivatives , and asterisks indicate increased expression . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 01610 . 7554/eLife . 17666 . 017Figure 5—figure supplement 2 . Repression of Six1 target genes by Six1-EnR injection . Tail bud stage ( A–C ) and neural plate stage ( D–I ) embryos after unilateral injection of Six1-EnR . In each case , lacZ was co-injected as a lineage tracer and panels A1–C1 show the control ( un-injected ) side and A2–C2 show the injected side . The injected side is positioned to the right in D–I , as marked by blue lacZ staining ( lacZ staining out of frame in some specimens ) . Arrows mark reductions in marker gene expression in placodal derivatives , and asterisks indicate increased expression . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 01710 . 7554/eLife . 17666 . 018Table 3 . Changes in marker gene expression in the placodes after injection of various constructs . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 018InjectionSix1-MO*Eya1-MO*Six1-EnREya1-mmMOSix1-GR§Eya1-GR§Phenotype%%%%%% ( n ) ( n ) ( n ) ( n ) ( n ) ( n ) Atoh1Reduced77**90‡94102642 ( 26 ) ( 20 ) ( 18 ) ( 21 ) ( 19 ) ( 12 ) Increased00003542 ( 26 ) ( 20 ) ( 18 ) ( 21 ) ( 17 ) ( 12 ) Gfi1aReduced82†67†69315736 ( 27 ) ( 17 ) ( 16 ) ( 26 ) ( 14 ) ( 14 ) Increased0000743 ( 27 ) ( 17 ) ( 16 ) ( 26 ) ( 14 ) ( 14 ) Hes8Reduced74‡83‡70176057 ( 35 ) ( 35 ) ( 46 ) ( 24 ) ( 40 ) ( 56 ) Increased002401529 ( 35 ) ( 35 ) ( 46 ) ( 24 ) ( 40 ) ( 56 ) Hes9Reduced73‡76‡84117529 ( 45 ) ( 33 ) ( 38 ) ( 27 ) ( 12 ) ( 29 ) Increased008000 ( 45 ) ( 33 ) ( 38 ) ( 27 ) ( 12 ) ( 29 ) Isl2Reduced66†100‡nd275024 ( 38 ) ( 17 ) nd ( 22 ) ( 18 ) ( 17 ) Increased60nd03141 ( 38 ) ( 17 ) nd ( 22 ) ( 16 ) ( 17 ) Ngn1Reduced65‡49†84171736 ( 51 ) ( 43 ) ( 31 ) ( 24 ) ( 30 ) ( 59 ) Increased016642341 ( 51 ) ( 43 ) ( 31 ) ( 24 ) ( 30 ) ( 59 ) Pou4f1 . 2Reduced67‡63†71164781 ( 48 ) ( 30 ) ( 35 ) ( 19 ) ( 15 ) ( 37 ) Increased0000130 ( 48 ) ( 30 ) ( 35 ) ( 19 ) ( 15 ) ( 37 ) Sox2Reduced74‡78‡8769048 ( 19 ) ( 18 ) ( 30 ) ( 16 ) ( 21 ) ( 33 ) Increased0023#0012 ( 19 ) ( 18 ) ( 30 ) ( 16 ) ( 21 ) ( 33 ) Sox3Reduced68‡54†3994940 ( 25 ) ( 26 ) ( 31 ) ( 22 ) ( 25 ) ( 23 ) Increased0071#01617 ( 25 ) ( 26 ) ( 31 ) ( 22 ) ( 25 ) ( 23 ) Tlx1Reduced84†91‡10033407 ( 31 ) ( 32 ) ( 13 ) ( 15 ) ( 10 ) ( 15 ) Increased60004073 ( 31 ) ( 32 ) ( 13 ) ( 15 ) ( 10 ) ( 15 ) * Significant differences ( Fisher’s exact test ) ;† p<0 . 05 , ‡ p<0 . 001 ) to Eya1-mmMO injections are indicated . § Dexamethasone treatment from stages 16–18 on . # Expression ectopic in epidermis . n: Number of embryos analysed at both neural plate ( stage 14–16 ) and tail bud ( stage 21–26 ) stage . nd: Not determined . To complement the loss-of-function studies , and to examine the spatial distribution of presumptive direct targets of Six1 and Eya1 in gain-of-function experiments , we injected Six1-GR and Eya1-GR individually and , to ensure that overexpression did not affect early embryogenesis , induced their nuclear translocation by adding DEX at neural fold stage ( stages 16–18 ) , after PPE commitment ( Ahrens and Schlosser , 2005 ) . Surprisingly , although injection of Six1-GR or Eya1-GR resulted in up-regulation of direct targets in a minority of cases ( Table 3; Figures 6 and 7 ) , the dominant observed phenotype was down-regulation of target gene expression in the PPE or placodes ( Table 3; Figures 8 and 9 ) . Considering that here , unlike in the initial RNA-Seq screen and qPCR experiments , CHX was not used to block protein synthesis , these results indicate that Six1 and Eya1 additionally affect expression of many of their direct target genes in indirect and partly opposing ways . 10 . 7554/eLife . 17666 . 019Figure 6 . Up-regulation of target gene expression domains after overexpression of Six1 . Tail bud stage embryos ( A–F ) after unilateral injection of Six1-GR and DEX induction at neural plate stage ( 16–18 ) . In each case , lacZ was co-injected as a lineage tracer and panels A1–F1 show the control ( un-injected ) side and A2–F2 show the injected side . Arrows and arrowheads mark expansions in marker gene expression in placodal and non-placodal derivatives , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 01910 . 7554/eLife . 17666 . 020Figure 7 . Up-regulation of target gene expression domains after overexpression of Eya1 . Tail bud stage embryos ( A–G ) after unilateral injection of Eya1-GR and DEX induction at neural plate stage ( 16–18 ) . In each case , lacZ was co-injected as a lineage tracer and panels A1–G1 show the control ( un-injected ) side and A2–G2 show the injected side . Arrows and arrowheads mark expansions in marker gene expression in placodal and non-placodal derivatives , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 02010 . 7554/eLife . 17666 . 021Figure 8 . Down-regulation of target gene expression domains after overexpression of Six1 . Tail bud stage embryos ( A–H ) after unilateral injection of Six1-GR and DEX induction at neural plate stage ( 16–18 ) . In each case , lacZ was co-injected as a lineage tracer and panels A1–H1 show the control ( un-injected ) side and A2–H2 show the injected side . Arrows mark reductions in marker gene expression in placodal derivatives . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 02110 . 7554/eLife . 17666 . 022Figure 9 . Down-regulation of target gene expression domains after overexpression of Eya1 . Tail bud stage embryos ( A–J ) after unilateral injection of Eya1-GR and DEX induction at neural plate stage ( 16–18 ) . In each case , lacZ was co-injected as a lineage tracer and panels A1–J1 show the control ( un-injected ) side and A2–J2 show the injected side . Arrows mark reductions in marker gene expression in placodal derivatives . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 022
Overexpression of GR-fusion constructs followed by DEX-induced nuclear translocation in the presence of protein synthesis inhibitors has been previously used successfully to screen for direct target genes of transcription factors or cofactors in Xenopus ( Kolm and Sive , 1995; Taverner et al . , 2005; Seo et al . , 2007 ) . Here , we combine this approach with high-throughput sequencing of tissue-specific RNA to identify several hundred novel presumptive direct target genes of Six1 and Eya1 in the PPE . We show that this strategy indeed recovers the majority of direct Six1 target genes known from previous studies , indicating its reliability . Our in situ and qPCR analyses of target genes predicted from the RNA-Seq screen also provided independent verification of selected target genes suggesting a low false discovery rate . Moreover , the expression of all genes selected for detailed analysis proved to be dependent on Six1 and Eya1 in the PPE , indicating that many genes our screen predicted as Six1/Eya1 targets are also functionally dependent on these upstream regulators in the PPE . A comparison of our data set with recently identified direct target genes of sine oculis , the Six1 orthologue in the developing eye of Drosophila ( Jusiak et al . , 2014; Jin et al . , 2016 ) also reveals that homologues to six out of the 12 sine oculis target genes identified with high confidence in Jin et al . ( 2016 ) are differentially expressed in our Sixi or Six1+Eya1i ( and often also in Eya1i ) treatment groups , viz . Six1 and Six2; Eya4; Shh; various matrix metalloproteases ( e . g . MMP9 ) ; Ets2; and Frizzled1 and Frizzled4 . This suggests that a relatively high proportion of Six1 target genes may be evolutionarily conserved . Our finding that many of the presumptive direct target genes of Six1 or Eya1 are not up-regulated in the absence of CHX indicates that without blocking protein synthesis it is not possible to reliably identify direct target genes , presumably due to the existence of indirect interactions with such targets . We believe that this is one of the reasons why our findings differ substantially from the study of Yan et al . ( 2015 ) , which analysed differentially expressed genes in Xenopus animal cap explants after overexpression of Six1 without first blocking protein synthesis . None of the transcription factors in our prioritised list was identified in the study by Yan et al . ( 2015 ) ; and we found none of the transcription factors differentially expressed in their study . A second likely reason for the discrepancy between the results presented here and in Yan et al . ( 2015 ) is that , while we specifically analysed PPE tissue ( presumably containing tissue-specific cofactors required for the activation or repression of Six1 and Eya1 target genes specific for the developing placodes ) , they analysed target genes in animal cap tissue , known to be composed of pluripotent cells . Previous studies have shown that Six1 and Eya1 are essential for both the establishment of the PPE ( Brugmann et al . , 2004; Christophorou et al . , 2009 ) , as well as for the subsequent development of placode-derived sense organs ( Xu et al . , 1999; Laclef et al . , 2003; Zheng et al . , 2003; Brugmann et al . , 2004; Zou et al . , 2004; Kozlowski et al . , 2005; Schlosser et al . , 2008; Ahmed et al . , 2012b , 2012a ) but the mechanisms through which they act are poorly understood . The continued expression of both genes in almost all placodes developing from the PPE ( Schlosser and Ahrens , 2004 ) , combined with the observed deficiencies in derivatives from most placodes after loss-of-function of either Six1 or Eya1 , indicates that they play a role in generic aspects of placode development shared by all placodes . Indeed , our data show that genes revealed as presumptive direct targets of Six1 and Eya1 were highly enriched for GO terms associated with neurogenesis and placode development . Our screen also confirms previous studies suggesting that Six1 and Eya1 synergistically regulate many genes in the PPE , and that the Six1-Eya1 protein complex predominantly acts by activating transcription ( Li et al . , 2003; Brugmann et al . , 2004 ) . However , we also find support for independent action of Six1 and Eya1 in the PPE , possibly in conjunction with other interacting partners ( Brugmann et al . , 2004; Ahmed et al . , 2012a ) . Surprisingly , we found Hox genes to be strongly enriched in the list of target genes activated by Eya1 only . This deserves further study since Eya1 has not previously been recognised as an upstream regulator of Hox genes . It has previously been suggested ( Schlosser , 2006 ) that a generic role of Six1 and Eya1 for all placodes could be implemented in two possible ways: ( 1 ) By the direct contribution to the activation of genes regulating developmental processes shared between different placodes such as proliferation , morphogenetic movements and neuronal or sensory cytodifferentiation; or ( 2 ) by direct contribution to the activation of genes defining the identity of different individual placodes within the PPE . Our data strongly suggest that Six1 and Eya1 act predominantly in the first rather than in the second mode . A large number of transcription factor encoding genes , including several Pax , Pitx , ANF and FoxI genes , have been implicated in conferring identity to individual placodes , or groups of placodes , within the PPE ( reviewed in Schlosser , 2006 , 2010; Grocott et al . , 2012; Saint-Jeannet and Moody , 2014 ) however only a few of these genes were recovered as targets of Six1 or Eya1 , e . g . Gbx2 ( FC 1 . 7 in Six1+Eya1i ) and FoxI4 ( FC 1 . 09 in Six1i ) . In contrast , we found a large number of genes encoding transcription factors with roles in neuronal/sensory cytodifferentiation but also other proteins with likely roles for the maintenance of proliferating progenitors ( e . g . Cyclin D , RGCC ) , the regulation of cell adhesion and morphogenetic movements ( e . g . EDAR , CXCR7 , Protocadherin11 , RhoV , Claudin3 ) and cytodifferentiation ( e . g . Espin , Neurotrophin3 ) . This suggests that , similar to Hox or Pax genes , Six1 and Eya1 act as both master genes and micro-managers ( Akam , 1998; Thompson and Ziman , 2011; Rezsohazy et al . , 2015 ) , acting upstream of a GRN co-ordinating cell differentiation in the PPE as well as directly activating terminal differentiation gene batteries . Considering that Six1 and Eya1 have previously been shown to promote a proliferative progenitor state at high doses but neuronal and sensory differentiation at lower doses ( Schlosser et al . , 2008 ) , it is particularly interesting that we identified presumptive direct target genes encoding transcription factors previously implicated in progenitor maintenance ( Sox2 , Sox3 , Hes8 , Hes9 ) and differentiation ( Ngn1 , Atoh1 , POU4f1 , Gfi1a , Isl2 , Tlx1 ) . Both Hes ( Hes8 , Hes9 ) and SoxB1 ( Sox2 , Sox3 ) proteins are known to keep progenitor cells in an undifferentiated state , and must be down-regulated for neuronal differentiation to proceed . While Sox2 and Sox3 play multiple roles including activity as pioneer factors , which open up chromatin for transcription ( Bylund et al . , 2003; Graham et al . , 2003; Pevny and Placzek , 2005; Bergsland et al . , 2011 ) , Hes proteins generally repress neuronal/sensory determination genes such as Ngn1 or Atoh1 as effectors of Notch signalling ( Kobayashi and Kageyama , 2014; Su et al . , 2015; Abdolazimi et al . , 2016 ) . Conversely , Ngn1 and Atoh1 are known to act as proneural factors that initiate differentiation of sensory neurons and hair cells , respectively ( Ma et al . , 1996 , 1998; Bermingham et al . , 1999; Millimaki et al . , 2007 ) , whereas POU4f1 ( previously known as Brn3a ) , Gfi1a , Isl2 and Tlx1 act further downstream in differentiation of sensory neurons ( Patterson and Krieg , 1999; Wallis et al . , 2003; Cheng et al . , 2004; Eng et al . , 2004; Lanier et al . , 2009; Dykes et al . , 2011 ) , and Gfi1a and the related POU domain factor POU4f3 ( or Brn3c ) are required for hair cell maintenance and survival ( Xiang et al . , 1998; Wallis et al . , 2003 ) . Our findings strongly indicate that Six1 and Eya1 directly promote multiple steps during sensory and neuronal development , and act to drive both progenitor maintenance and neuronal differentiation programmes in placodes ( summarised in Figure 10 ) , although further functional studies are needed to clarify the mechanism allowing Six1 and Eya1 to maintain the balance between activation of progenitor and differentiation genes . Additionally , direct binding of Six1 to regulatory regions of targets identified in this study should be confirmed by methods such as ChIP-Seq . 10 . 7554/eLife . 17666 . 023Figure 10 . Network summary for Six1/Eya1-activated gene regulation in the PPE . Six1/Eya1 act to promote neuronal differentiation , by activation of pro-neural genes ( Ngn1 , Atoh1 ) , as well as progenitor state maintenance , by activation of genes such as SoxB1 and Hes genes . Arrows indicate direct ( solid line ) and indirect ( dotted line ) activation; barred lines show direct ( solid line ) and indirect ( dotted line ) repression . Evidence for interactions: Six1 positively autoregulates ( Sato et al . , 2012 ) ; Six1/Eya1 directly activate Sox2 , Sox3 , Hes8 , Hes9 , Ngn1 , Atoh1 , Isl2 , Pou4f1 . 2 , Tlx1 and Gfi1a ( this study ) ; Sox2 synergises with Six1/Eya1 ( Ahmed et al . , 2012b , 2012a ) ; Sox2 directly activates Atoh1 ( Ahmed et al . , 2012a ) and Ngn1 ( Cimadamore et al . , 2011 ) ; Atoh1 and Ngn1 indirectly repress each other ( Gowan et al . , 2001 ) ; Ngn1 indirectly represses Sox2 ( Evsen et al . , 2013 ) ; Ngn1 directly activates NeuroD1 ( Seo et al . , 2007 ) ; Atoh1 positively autoregulates ( Helms et al . , 2000 ) ; Atoh1 indirectly represses Sox2 ( Neves et al . , 2012 ) and activates Gfi1 ( Wallis et al . , 2003 ) ; NeuroD1 directly activates Pou4f1 . 2 ( Hutcheson and Vetter , 2001 ) and Isl1 ( Lee et al . , 1995 ) ; Pou4f1 . 2 directly activates Gfi1 ( Hertzano et al . , 2004 ) and indirectly activates Tlx1 ( Hutcheson and Vetter , 2001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17666 . 023 The analysis of Six1 and Eya1 presumptive direct target genes presented here establishes a GRN regulating the development of cranial vertebrate sensory organs and neurons from the PPE ( Figure 10 ) , and identifies a large number of novel putative direct target genes encoding a diverse array of proteins . Among these are many promising candidates potentially involved in mediating the effects of Six1 or Eya1 on proliferation , morphogenesis and cytodifferentiation in developing placodes . This makes our data an invaluable repository of information for designing further functional studies on early sensory development in vertebrates . Finally , while our study focussed on the role of Six1 and Eya1 during sensory development , cell proliferation , morphogenesis and cytodifferentiation are also known to be affected in human patients in which Six1 and Eya1 are dysregulated , leading to sensory deficits after Six1 or Eya1 loss of function mutations ( Kochhar et al . , 2007 ) or enhanced tumour progression after Six1 or Eya1 up-regulation ( Blevins et al . , 2015; Liu et al . , 2016 ) . This suggests that many target genes identified in our study may also be misregulated in these diseases , potentially opening up exciting new avenues for therapeutic intervention .
Capped RNAs of Xenopus Six1-GR , Eya1-GR and Six1-EnR were made by in vitro transcription using the mMessage mMachine SP6 kit ( Ambion , Austin , Texas ) from the following templates: pCS2+-GR-myc-Six1 , pCS2+-GR-myc-Eya1α ( Schlosser et al . , 2008 ) and pCS2-EnR-Six1 ( Brugmann et al . , 2004 ) . Translation blocking morpholinos ( MO ) for Six1 ( Six1-MO1: 5’-GGAAGGCAGCATAGACATGGCTCAG-3’; Six1-MO2: 5’-CGCACACGCAAACACATACACGGG-3’ ) and Eya1 ( Eya1-MO1: 5’-TACTATGTGGACTGGTTAGATCCTG-3’; Eya1-MO2: 5’-ATATTTGTTCTGTCAGTGGCAAGTC-3’ ) were previously described ( Brugmann et al . , 2004; Schlosser et al . , 2008 ) . An Eya1-MO carrying 5 mismatches was used as a control ( Eya1-mmMO; mismatches shown in lower case: 5’-ATtTTaGTTCTGaCAGTGGgAAcTC-3’ ) . Six1-GR ( 500 pg ) , Eya1-GR , ( 500 pg ) , Six1-EnR ( 100 pg ) mRNAs and Six1-MO1+2 ( 2 ng ) , Eya1-MO1+2 ( 2 ng ) , and Eya1-mismatch-MO ( 2 ng ) were freshly prepared before each injection . lacZ ( 250 pg ) or myc-GFP ( 125 pg ) mRNAs were co-injected to mark the injected side . Embryos of Xenopus laevis were obtained by in vitro fertilisation , staged according to ( Nieuwkoop and Faber , 1967 ) and injected unilaterally into two-cell blastomeres according to standard procedures ( Sive et al . , 2000 ) . Six1-EnR was injected at the four-cell stage into single blastomeres that give rise to the dorsal ectoderm as previously described ( Brugmann et al . , 2004 ) . To obtain RNA for RNA-Seq or qPCR , both blastomeres of two-cell stage embryos were injected with either 1 ) Six1-GR ( 500 pg ) + myc-GFP ( 125 pg ) , 2 ) Eya1-GR ( 500 pg ) + myc-GFP ( 125 pg ) , or 3 ) Six1-GR ( 500 pg ) + Eya1-GR ( 500 pg ) + myc-GFP ( 125 pg ) . Each of these treatment groups was allowed to develop to early neural plate stage before being sorted under a fluorescent microscope . The lateral part of the preplacodal region ( LPR of Ahrens and Schlosser , 2005 ) was explanted from GFP positive embryos ( ~100 per biological replicate ) in 1 × MBSH ( Sive et al . , 2000 ) supplemented with 2 mM CaCl2 , 25 mg/l gentamycine ( Sigma , St Louis , Missouri ) , 400 mg/l penicillin ( Sigma ) , and 400 mg/l streptomycin sulphate ( Sigma ) . Explants were pre-treated with 0 . 1 × modified Barth’s solution ( MBS; Sive et al . , 2000 ) with cycloheximide ( CHX; final concentration 10 μg/ml ) for 30 min at 25°C . After pre-treatment , 50% of the explants were transferred to 0 . 1 × MBS with CHX + dexamethasone ( DEX; final concentration 10 μM ) and incubated for 2 hr 30 at 25°C ( Figure 1 ) when control embryos had reached stage 20 . Explants were immediately homogenised in Trizol ( Invitrogen , Carlsbad , California ) and total RNAs extracted . Isolated RNA was quality assayed in an Agilent 2100 Bioanalyzer and all samples used for sequencing had an RIN >7 . 0 . Libraries were prepared from 1 mg total RNAs and subjected to deep sequencing with Illumina Hi-Seq1000 at the EMBL Genecore facility . Paired-end ( 100 bp ) sequence reads were quality-filtered using Trimmomatic ( Bolger et al . , 2014 ) , and mapped to the Xenopus laevis genome ( XL7 . 0 ) with Bowtie2 ( version 2 . 2 . 5; Langmead and Salzberg , 2012 ) and Tophat2; ( version 2 . 0 . 13; Kim et al . , 2013 ) . An average of 65 million reads ( ~80% of quality filtered reads ) were mapped with 90% of reads properly paired in sequencing across treatment groups . Transcript models were assembled using Cufflinks2 ( version 2 . 1 . 1; Trapnell et al . , 2012 ) , and differential expression was determined using Cuffdiff2 ( version 2 . 1 . 1; Trapnell et al . , 2012 ) . Gene models were annotated against a combined Xenopus mRNA database ( X . laevis: ftp://ftp . xenbase . org/pub/Genomics/Sequences/xlaevisMRNA . fasta; X . tropicalis: ftp://ftp . xenbase . org/pub/Genomics/Sequences/xtropMRNA . fasta ) using blastn with an e-value cut-off of 1E-5 . Using this approach we were able to annotate an average of 80% of mapped reads . Initially , two samples of CHX- and CHX+DEX-treated explants were independently collected , sequenced and mapped for each treatment group ( injection of Six1-GR , Eya1-GR or Six1-GR+Eya1-GR ) , and were specified as two biological replicates in Cuffdiff . To preclude the inclusion of genes affected by DEX treatment alone , we also analysed explants taken from un-injected embryos and treated as above ( CHX vs . CHX+DEX ) . Two biological replicates of this control treatment group were included in sequencing . Presumptive direct targets of Six1 , Eya1 or Six1+Eya1 were determined by comparing Six1-GR , Eya1-GR or Six1-GR+Eya1-GR-injected embryos treated with CHX ( as controls ) against CHX+DEX-treated samples . Genes were considered to be differentially expressed if ( 1 ) the FPKM ( Fragments Per Kilobase of exon per Million fragments mapped ) for that gene was >1 in the CHX+DEX treatment group , ( 2 ) the gene was at least two-fold up-/down-regulated after CHX+DEX treatment compared to CHX treatment , ( 3 ) there was at least a two-fold difference between the control ( un-injected ) and experimental ( injected with either Six1-GR , Eya1-GR or Six1-GR+Eya1-GR ) fold change ( FC ) values in response to DEX treatment . The Pearson correlation was high for each of the treatment groups ( >0 . 9 for all pairwise comparisons ) , indicating the similarity of expression profiles between independently treated samples . As a second approach to finding genes that showed differential expression in response to DEX treatment , RNA-Seq data of several treatment groups were merged to add statistical power to the analysis . In one analysis , all replicates from our three different treatment groups were considered as equivalent to focus on genes with similar differential expression profiles across all treatment groups ( comprising the Six1+Eya1m dataset with six replicates ) . In another analysis , all treatment groups involving Six1 overexpression ( i . e . injection of Six1-GR alone or Six1-GR+Eya1-GR: Six1m with 4 replicates ) were treated as equivalent as were all treatment groups involving Eya1 overexpression ( Eya1-GR , Six1-GR+Eya1-GR: Eya1m with 4 replicates ) . This allowed us to focus on genes whose activation was limited by either Six1 or Eya1 levels . We considered a gene to be significantly differentially expressed if it passed Cuffdiff’s statistical test ( q < 0 . 05 ) in addition to meeting the criteria outlined above . Xenopus annotations were converted to their human orthologs from the Human Uniprot database , and functionally annotated using the online tools ‘PantherDB’ ( Mi et al . , 2013; http://pantherdb . org ) and ‘DAVID’ ( Huang et al . , 2009; https://david . ncifcrf . gov ) . For GSEA of placodal transcriptomes after injection of Six1 and/or Eya1 , the placodal transcriptome of un-injected , CHX treated placodal explants was specified as a background set , whereas GSEA of the transcriptome of untreated explants was conducted using the default 'human dataset' in DAVID as background . The enrichment score ( E ) for each group is reported as the geometric mean of the EASE scores ( a modified Fisher’s exact score ) that are associated with the enriched annotation terms belonging to that group ( Huang et al . , 2007 ) . RNA was extracted from explants after CHX or CHX+DEX treatment as detailed above . cDNA was synthesised using the QuantiTect Reverse Transcription Kit ( Qiagen , Hilden , Germany ) , using 500 ng total RNA according to the manufacturer’s protocol . qPCR was performed using Taqman reagents on a StepOne Plus machine ( Applied Biosystems , Foster City , California ) , using Smn2 as a reference ( Dhorne-Pollet et al . , 2013; Supplementary file 4 ) . qPCR was performed in triplicate and the entire experiment was repeated three times from independently prepared RNA . Relative Quantification ( RQ ) values and log2 fold change ( FC ) were averaged across biological replicates . The full coding region of Hes8 , Crem , FosB , Tbx15 , Atoh1 and Isl2 was synthesised from transcript models from RNA-Seq data ( KT722743; KT722744; KT722745; KT722746; KT722747; KT722748 ) by Genescript into the cloning vector pUC57 and subsequently sub-cloned into the expression vector pCS2+ using the following restriction sites: Hes8 and Crem: ClaI/EcoRI; Atoh1: XbaI; Tbx15 and FosB: BamHI/EcoRI; Isl2: EcoRI/StuI . Primers with added ClaI and EcoRI sites ( to the forward and reverse primers , respectively ) were designed ( Supplementary file 4 ) to amplify the entire coding region of Tbx6 , which was then subcloned into pCS2+ between the ClaI/EcoRI sites . Znf214 , Mab21l2-b and Pou3f2b were ordered ( pCMV-SPORT6 , Fisher Scientific , Waltham , Massachusetts; Clone IDS: 5512398 , 5515985 and 4203106 ) . Hes9 ( pCR4-TOPO ) was ordered from Source Bioscience ( Clone accession: BC169570 ) and was subcloned into the EcoRI site of pCS2+ . Embryos injected with myc-GFP were sorted under a fluorescent microscope and fixed using a standard protocol ( Sive et al . , 2000 ) . LacZ-injected embryos were fixed and then stained with X-gal solution to reveal lacZ . Wholemount in-situ-hybridisation was carried out under high stringency conditions at 60°C as previously described ( Harland , 1991 ) using digoxigenin-labelled antisense probes . Probes for Six1 ( Pandur and Moody , 2000 ) , N-tubulin ( Oschwald et al . , 1991 Sox2 ( De Robertis et al . , 1997 ) , Sox3 ( Penzel et al . , 1997 ) , Ripply3 ( Janesick et al . , 2012 ) , Hes2 ( Sölter et al . , 2006 ) , Sim1 ( Martin et al . , 2007 ) , Gbx2 ( von Bubnoff et al . , 1996 ) , Lhx5 ( Bachy et al . , 2001 ) , Sox21 ( Cunningham et al . , 2008 ) , Emx1 . 2 ( Green and Vetter , 2011 ) , Pou4f1 . 2 ( Hutcheson and Vetter , 2001 ) , and Tlx1 ( Patterson and Krieg , 1999 ) were synthesised as previously described . Primers were designed with promoter sites added ( forward , T7; reverse , SP6 ) for Hes8 , Hes9 , Gfi1a , Tbx15 , Ngn1 , Pou4f1 . 2 and Isl2 and were used to amplify a ~800 bp fragment from plasmid DNA ( Supplementary file 4 ) which was then used as a template for probe synthesis using T7 RNA polymerase to make an antisense probe . pCMV-SPORT6 with Znf214 , Mab21l2-b and Pou3f2b were linearised with SalI and antisense probes synthesised with T7 . pCS2+ vectors containing Tbx6 , FosB and Crem were linearised with BamHI and transcribed with T7 . pCS2+ with Atoh1 was linearised with NotI and transcribed with SP6 . In order to analyse the distribution of gene expression domains in finer detail , serial 40–50 μM vibratome sections were cut from selected embryos after wholemount in-situ hybridisation . Where staining with X-gal was insufficient to reveal the injected site , lacZ distribution was revealed immunohistochemically using a polyclonal rabbit anti-LacZ ( MP Biomedicals Cappel , Santa Ana , California; Cat . : 55976; 1:1000 ) and an Alexa594-conjugated anti-rabbit antibody ( 1:1000 ) . All sequencing data have been deposited in the NCBI BioProject database under BioProject PRJNA317049 . All scripts used in analysis are available at https://github . com/nriddiford/Six1-Eya1-RNA-Seq . git . | Animals that possess a backbone – also known as vertebrates – have several paired sense organs in their heads , such as the eyes and the olfactory system . These organs are thought to have arisen as vertebrates evolved from their filter-feeding ancestors and adopted an increasingly active and predatory lifestyle . The cranial placodes are tissues in the vertebrate embryo that give rise to many of these sense organs early in development . The sensory neurons that transmit information from the organs to the brain – including those that process hearing and smell – also develop from these tissues . While much is known about how the sense organs work , relatively little is known about the early processes involved in their development . Two genes have been established as crucial for the formation and later development of the sense organs . These genes encode two proteins called Six1 and Eya1 that regulate other genes , although the identity of the genes that they target in placodes was not known . Using computational and experimental approaches in the African clawed frog ( Xenopus laevis ) , Riddiford and Schlosser identified hundreds of genes that are regulated by both Six1 and Eya1 in placodes . Many of these targeted genes regulate the activity of other genes and direct important cell decisions , such as whether a stem cell should develop into a neuron . After validating several of these targets in the laboratory , Riddiford and Schlosser proposed a network of gene regulation , activated by Six1 and Eya1 , that drives sense organ development in vertebrates . While Riddiford and Schlosser provide strong evidence that Six1 and Eya1 directly regulate many of the newly identified genes , further work is required to firmly establish this . Future studies could also explore how Six1 and Eya1 drive seemingly contradictory cellular decisions , encouraging some stem cells to develop into neurons and others to maintain a stem-like state . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2016 | Dissecting the pre-placodal transcriptome to reveal presumptive direct targets of Six1 and Eya1 in cranial placodes |
As an adipokine in circulation , adiponectin has been extensively studied for its beneficial metabolic effects . While many important functions have been attributed to adiponectin under high-fat diet conditions , little is known about its essential role under regular chow . Employing a mouse model with inducible , acute β-cell ablation , we uncovered an essential role of adiponectin under insulinopenic conditions to maintain minimal lipid homeostasis . When insulin levels are marginal , adiponectin is critical for insulin signaling , endocytosis , and lipid uptake in subcutaneous white adipose tissue . In the absence of both insulin and adiponectin , severe lipoatrophy and hyperlipidemia lead to lethality . In contrast , elevated adiponectin levels improve systemic lipid metabolism in the near absence of insulin . Moreover , adiponectin is sufficient to mitigate local lipotoxicity in pancreatic islets , and it promotes reconstitution of β-cell mass , eventually reinstating glycemic control . We uncovered an essential new role for adiponectin , with major implications for type 1 diabetes .
Adiponectin is an adipocyte-derived hormone exerting pleiotropic beneficial effects on metabolism ( Ye and Scherer , 2013 ) . Increased circulating adiponectin improves the metabolic flexibility of adipose tissue and confers systemic tolerance to obesity ( Asterholm and Scherer , 2010 ) . Under normal physiological conditions , adiponectin promotes plasma lipid clearance ( Combs et al . , 2004; Qiao et al . , 2008 ) . Paradoxically however , even though many important functions have been attributed to this circulating factor in mice and humans ( Shetty et al . , 2009; Turer and Scherer , 2012 ) , adiponectin is not essential for life under normal physiological conditions . Genetic deletion of adiponectin in rodents leads to mild or moderate insulin resistance , which is exacerbated upon high-fat diet challenge ( Kubota et al . , 2002; Maeda et al . , 2002 ) . Adiponectin is required for PPARγ agonist-mediated improvements in insulin sensitivity ( Nawrocki et al . , 2006 ) . Based on many published studies , adiponectin function under normal physiological function is dispensable , and it starts to play a more prominent role under hyperglycemic and , most importantly , dyslipidemic conditions . The failure of insulin-producing β-cells is a hallmark of the pathophysiology of both type 1 and type 2 diabetes . The ongoing loss of β-cells under these conditions is associated with a failure to effectively regenerate β-cell mass . This can be attributed not only to the low capacity for replication and differentiation ( Bouwens and Rooman , 2005 ) but also the detrimental cytotoxic environment that the cells are exposed to due to the dysregulation of the balance between insulin and glucagon ( Robertson , 2009; Unger and Cherrington , 2012 ) . Insulin deficiency results in hyperglycemia and hyperlipidemia , both of which trigger β-cell glucotoxicity and oxidative stress ( Poitout and Robertson , 2008 ) , endoplasmic reticulum ( ER ) stress ( Fonseca et al . , 2009 ) , and lipotoxicity ( Kusminski et al . , 2009 ) . Tissue culture data suggested that adiponectin regulates β-cell viability ( Brown et al . , 2010; Holland et al . , 2011; Rao et al . , 2012 ) . Recently , we demonstrated that adiponectin protects β-cells against lipotoxicity and apoptosis , both in cultured cells and in vivo ( Holland et al . , 2011 ) . In this study , we took advantage of the PANIC-ATTAC transgenic mouse model ( Wang et al . , 2008 ) . After extensive β-cell ablation , adiponectin becomes essential for survival . In the context of insulin deficiency , the lack of adiponetin aggravates the lipoatrophy and hyperlipidemia to lethal levels . The critical role of adiponectin in maintaining minimal lipid homeostasis is recapitulated in the streptozotocin-treated ( STZ ) mouse model . Specifically , adiponectin is required for lipid uptake into subcutaneous white adipose tissue . Under normal conditions , the action of adiponectin can be mediated through enhanced lipoprotein lipase activity and intracellular fatty acid translocation . However , under insulinopenic conditions , the primary adiponectin-mediated effect relies on enhanced insulin sensitivity and endocytic activity . While insulin deficiency and widespread loss of β-cells lead to augmented intracellular lipotoxicity , adiponectin overexpressing mice effectively overcome the resulting intracellular lipotoxicity in β-cells by ameliorating lipid metabolism and thereby paving the way for β-cell mass recovery . Our findings reveal a novel role of adiponectin as a housekeeping protein under insulinopenic conditions , and augmentation of adiponectin is sufficient to promote β-cell regeneration .
To investigate the physiological role of adiponectin under insulinopenic conditions , we crossed adiponectin null mice ( Nawrocki et al . , 2006 ) to the homozygous PANIC-ATTAC background ( Wang et al . , 2008 ) . The PANIC-ATTAC transgene allows us to eliminate a defined number of β-cells . Starting with similar β-cell mass ( Figure 1A and Figure 1—figure supplement 1 ) , 8-week old male homozygous PANIC-ATTACs with adiponectin wild type ( P-Adn+/+ ) or knockout ( P-Adn−/− ) were treated with the same high-dose of dimerizer AP20187 to induce caspase-8-mediated apoptosis . 2 weeks after the initial dimerizer administration , the insulin positive areas of P-Adn+/+ and P-Adn−/− mice decreased down to <15% of their starting levels and further decreased down to <10% around 10 weeks post β-cell ablation ( Figure 1A and Figure1—figure supplement 1 ) . Different from our previous studies ( Wang et al . , 2008; Holland et al . , 2011 ) , we have used a fairly stringent ablation protocol to obtain a very high level of β-cell loss in both genotypes . In this setting , both the P-Adn+/+ and P-Adn−/− mice showed sustained glucose levels above ∼500 mg/dl ( Figure 1B ) . Their fasting insulin levels decreased to <8% of the euglycemic wild-type ( WT ) controls after parallel dimerizer treatment ( Figure 1C ) . With such intensive β-cell ablation and insulin deficiency , both the P-Adn+/+ and P-Adn−/− mice were severely glucose intolerant ( Figure 1D ) , and their glucose-stimulated insulin secretion ( GSIS ) was abolished ( Figure 1E ) . 10 . 7554/eLife . 03851 . 003Figure 1 . Adiponectin is required for minimal lipid homeostasis and survival in PANIC-ATTAC mice . Mice with four different genotypes ( wildtype [WT]; homozygous PANIC-ATTAC with adiponectin wildtype [P-Adn+/+] , adiponectin knockout [P-Adn−/−] , or transgenic adiponectin overexpressing mice [P-AdnTg/+] ) were exposed to dimerizer . ( A ) Quantitation of insulin-immunostained cell area , normalized to total pancreas area . n = 3–6 mice per condition . Source files are available in Figure 1—source data 1 . ( B ) Fasting blood glucose . n ≥ 8 mice per condition . ( C–E ) At 10 weeks after dimerizer treatment: ( C ) fasting blood insulin . ( D ) Blood glucose and ( E ) plasma insulin during an oral glucose tolerance test . ( F ) Survival curve . n = 25 ( WT ) , 29 ( P-AdnTg/+ ) , 23 ( P-Adn+/+ ) and 21 ( P-Adn−/− ) mice . ( G and H ) Blood triglycerides ( G ) and total ketone bodies ( H ) in mice under fed status . ( I ) Fat mass presented as percentage of mouse body weight . n ≥ 6 mice per condition unless specified . Data are presented as the mean ± SEM . *p < 0 . 05 , **p < 0 . 01 for P-Adn−/− vs P-Adn+/+ . #p < 0 . 05 , ##p < 0 . 01 for P-Adn+/+ vs WT . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 00310 . 7554/eLife . 03851 . 004Figure 1—source data 1 . Source files for insulin-positive cell area quantitation . The zip file contains all the 2400-dpi scanned images of insulin-immunostained pancreas sections used for quantitation of insulin-positive cell area . Insulin is stained as brown , and the whole section is counterstained as blue with hematoxylin . Folders are named after genotypes ( P-Adn++ or P-Adn−− ) and subfolders after time points ( week 0 , 2 , 5 , or 10 ) post initial dimerizer administration . Related to Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 00410 . 7554/eLife . 03851 . 005Figure 1—figure supplement 1 . Sustained β-cell ablation in PANIC-ATTAC mice . Representative immunofluorescence of glucagon ( red ) and insulin ( green ) on pancreatic islets from homozygous PANIC-ATTAC with adiponectin wild-type ( P-Adn+/+ ) or knockout ( P-Adn−/− ) mice at 0 , 2 , and 10 weeks after dimerizer treatment . Related to Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 00510 . 7554/eLife . 03851 . 006Figure 1—figure supplement 2 . Body composition of PANIC-ATTAC mice . Fat mass ( A ) and lean mass ( B ) at 0 , 2 , 5 , 10 weeks after initial dimerizer treatment . n ≥ 6 mice per condition . Data are presented as the mean ± SEM . **p < 0 . 01 for P-Adn−/− vs P-Adn+/+ . Related to Figure 1I . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 00610 . 7554/eLife . 03851 . 007Figure 1—figure supplement 3 . Food intake of PANIC-ATTAC mice . Food intake during the third week after initial dimerizer treatment . n = 4 mice per genotype . Data are presented as the mean ± SEM . Related to Figure 1I . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 007 To our surprise , under the insulinopenic conditions , P-Adn−/− mice showed a dramatically lower survival rate than P-Adn+/+ mice ( Figure 1F ) . Only 33% of the adiponectin null mice survived 9 weeks post dimerizer , with a median survival of 8 . 4 weeks , while 87% of the P-Adn+/+ mice remained alive . Adiponectin overexpressing mice ( Combs et al . , 2004 ) crossed into the homozygous PANIC-ATTAC background ( P-AdnTg/+ ) show a similar survival rate as P-Adn+/+ mice . This fivefold increased mortality in P-Adn−/− mice was associated with an extreme deterioration in lipid metabolism . Their triglyceride levels ( 6 . 8 ± 1 . 0 mM ) were 60% higher than the levels in P-Adn+/+ mice ( Figure 1G ) . More strikingly , the circulating ketone bodies in the adiponectin null mice reached an aberrantly high level ( 0 . 77 ± 0 . 15 mM ) , which was sixfold higher than the WT level and 3 . 5-fold higher than the P-Adn+/+ level ( Figure 1H ) . Prior to the aggravated hyperlipidemia , P-Adn−/− mice demonstrated a significant decrease in fat mass , to a critically low level of <5% body weight ( Figure 1I and Figure 1—figure supplement 2 ) . We observed no significant difference in food intake between P-Adn−/− and P-Adn+/+ mice ( Figure 1—figure supplement 3 ) . These data indicate that adiponectin is essential for lipid homeostasis and survival in the absence of insulin . To further elucidate the critical role of adiponectin in lipid metabolism under insulinopenia , we treated adiponectin knockout mice ( Adn−/− ) and the WT controls ( Adn+/+ ) with a high dose of streptozotocin ( STZ ) as an alternate approach to destroy β-cells . Consistent with the observations in the PANIC-ATTAC model ( Figure 1I ) , STZ-treated adiponectin null mice had significantly lower adipose tissue mass than WT mice ( Figure 2A and Figure 2—figure supplement 1 ) . The deterioration in lipid metabolism in the STZ-treated Adn−/− mice was also apparent as judged by the increase in circulating triglyceride levels in fed mice ( 2 . 3-fold ) ( Figure 2B ) , overnight fasted mice ( 3 . 6-fold ) ( Figure 2—figure supplement 2A ) , and following an oral triglyceride load ( Figure 2C ) . Non-esterified fatty acids ( NEFAs ) levels were also higher in Adn−/− mice ( 1 . 4-fold increase ) ( Figure 2—figure supplement 2B ) . The plasma lipoprotein fractionation of STZ-treated Adn−/− mice displayed elevated triglyceride and cholesterol content in VLDL ( Figure 2D , E ) . In an attempt to rescue this phenotype not only genetically , but also using a recombinant protein approach , exogenous administration of adiponectin induced a lowering of circulating triglycerides in STZ-treated Adn−/− mice ( Figure 2—figure supplement 3 ) . 10 . 7554/eLife . 03851 . 008Figure 2 . Adiponectin is essential for lipid metabolism in STZ-treated insulinopenic mice . Adiponectin wild-type ( Adn+/+ ) and knockout ( Adn−/− ) mice were treated with streptozotocin ( STZ ) to eliminate pancreatic β-cells and became hyperglycemic within 2 days after administration . ( A ) Fat mass 4 weeks after STZ treatment . ( B ) Fed triglyceride levels 3 weeks after STZ treatment . ( C ) Blood triglycerides during oral triglyceride tolerance test in mice 2 weeks after STZ treatment . ( D and E ) Pooled plasma samples from Adn+/+ and Adn−/− mice ( n = 3 ) were subjected to FPLC lipoprotein fractionation . Triglyceride ( D ) and cholesterol ( E ) contents of the 0 . 3 ml fractions were assayed . ( F ) Blood triglycerides after tail vein injection of tyloxapol . 2 weeks after STZ treatment , mice were fed or fasted for 4 hr before tyloxapol administration . ( G ) Blood glycerol after intraperitoneal injection of insulin ( 0 . 1 mU/g BDW ) in mice 3 days after STZ treatment . n ≥ 7 mice per condition unless specified . Data are presented as the mean ± SEM . *p < 0 . 05 , **p < 0 . 01 for Adn+/+ vs Adn−/− . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 00810 . 7554/eLife . 03851 . 009Figure 2—figure supplement 1 . Body composition of STZ-treated mice . Fat mass and lean mass of mice 4 weeks after STZ treatment . n = 15 ( Adn+/+ ) and 23 ( Adn−/− ) . Data are presented as the mean ± SEM . *p < 0 . 05 for Adn+/+ vs Adn−/− . Related to Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 00910 . 7554/eLife . 03851 . 010Figure 2—figure supplement 2 . Serum lipids in overnight fasted STZ-treated mice . Blood triglycerides ( A ) and NEFAs ( B ) in overnight fasted mice 1 week after STZ treatment . n = 7 mice per genotype . Data are presented as the mean ± SEM . *p < 0 . 05 , **p < 0 . 01 for Adn+/+ vs Adn−/− . Related to Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 01010 . 7554/eLife . 03851 . 011Figure 2—figure supplement 3 . Recombinant adiponectin rescues triglyceride clearance in STZ-treated adiponectin null mice . At 4 weeks after STZ treatment , Adn−/− mice were subjected to a tail vein injection of recombinant adiponectin ( 2 μg/g BDW ) or vehicle solutions ( PBS with 0 . 5 mM CaCl2 ) and measured for serum triglycerides at the indicated time points . n = 3 mice per condition . Data are presented as the mean ± SEM . Related to Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 01110 . 7554/eLife . 03851 . 012Figure 2—figure supplement 4 . Hepatic secretion rates of triglycerides . Slopes of plasma triglyceride over time after tyloxapol injection . n = 7–9 mice per condition . Data are presented as the mean ± SEM . Related to Figure 2F . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 01210 . 7554/eLife . 03851 . 013Figure 2—figure supplement 5 . Hepatic expression of metabolic genes . RT-qPCR analysis of gene expression in liver from mice 4 weeks after STZ treatment . cDNA abundances were normalized against 18S rRNA . n = 4 mice per genotype . Data are presented as the mean ± SEM . Related to Figure 2F . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 01310 . 7554/eLife . 03851 . 014Figure 2—figure supplement 6 . Serum triglycerides and NEFAs after low-dose insulin administration . Plasma triglycerides ( A ) and NEFAs ( B ) in mice after intraperitoneal injection of insulin ( 0 . 1 mU/g BDW ) . n = 7 mice per genotype . Data are presented as the mean ± SEM . *p < 0 . 05 , **p < 0 . 01 for Adn+/+ vs Adn−/− . Related to Figure 2G . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 014 We wanted to determine what the underlying mechanisms are for the exacerbated lipid metabolism in STZ-treated adiponectin knockout mice . We addressed whether changes in hepatic lipid secretion may be the underlying reason . We treated the mice with the lipoprotein lipase inhibitor tyloxapol ( WR-1339 ) and monitored serum lipid accumulation , a classical test to assess VLDL secretion . Compared to WT control mice , adiponectin null mice demonstrated only a minor trend towards an increase in serum triglyceride content ( Figure 2F ) and rate of accumulation ( Figure 2—figure supplement 4 ) . We also examined a panel of metabolic gene expression in the livers of STZ-treated Adn+/+ and Adn−/− mice and observed no significant changes ( Figure 2—figure supplement 5 ) . Insulin regulates fat metabolism in adipocytes by both enhancing FFA uptake and inhibiting lipolysis ( Holm et al . , 2000 ) . Does adiponectin play a role in suppressing lipolysis under conditions of low insulin ? To address this question , we administered a low dose of insulin to STZ-treated mice ( 0 . 1 mU/g body weight ) , and measured circulating glycerol levels under these conditions . This small amount of insulin substantially reduced serum glycerol ( Figure 2G ) , while it only minimally affected circulating triglycerides ( Figure 2—figure supplement 6A ) and NEFAs ( Figure 2—figure supplement 6B ) . Intriguingly , compared to WT mice , adiponectin null mice were resistant to the action of insulin , with minimal impact on glycerol levels ( Figure 2G ) . This suggests adiponectin is critical for insulin-mediated suppression of lipolysis under insulinopenic conditions . The enhanced lipolysis in STZ-treated adiponectin null mice might , at least in part , account for the reduced fat mass seen in the Adn−/− mice ( Figure 2A ) . Yet another site of action could be at the level of plasma lipid clearance . We examined whole body and tissue-specific uptake of circulating triglycerides using 3H-triolein . No apparent differences between WT and adiponectin null mice were noted prior to STZ treatment . In contrast , post STZ treatment , adiponectin null mice showed a 39% lower whole body clearance rate of labeled triolein compared to WT mice ( Figure 3—figure supplement 1A ) . Among the nine tissues examined , we observed significant differences of triolein uptake only in the subcutaneous white adipose tissue ( WAT , 50% lower in STZ-treated null mice than in STZ-treated WT mice ) ( Figure 3A , B ) . Consistent with a primary site of action on subcutaneous fat pads , we have reported that after long-term high-fat diet exposure , adiponectin promotes preferentially subcutaneous WAT expansion ( Asterholm and Scherer , 2010 ) . We also partitioned the tissue-specific 3H-triolein uptake into incorporated ( Figure 3—figure supplement 1B ) vs oxidized lipids ( Figure 3—figure supplement 1C ) . The difference in triolein uptake in subcutaneous WAT was primarily the result of a reduced level of incorporation . In agreement with the 3H-triolein uptake assay , histological analysis of adiponectin knockouts demonstrated a major reduction of adipocyte size in subcutaneous WAT and trends towards smaller cell size in brown adipose tissue ( BAT ) . However , no such reduction was found in gonadal WAT ( Figure 3C ) . We further confirmed the essential role of adiponectin in lipid uptake employing the PANIC-ATTAC model . 3 weeks post dimerizer , P-AdnTg/+ , P-Adn+/+ and P-Adn−/− mice were subjected to oral gavage of BODIPY-labeled fatty acids and examined for fluorescence signal in subcutaneous WAT by confocal microscopy . Adiponectin overexpression dramatically enhanced BODIPY signal in adipocytes , while adiponectin nulls showed a significant reduction ( Figure 3D and Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 03851 . 015Figure 3 . Adiponectin is critical for subcutaneous white adipose tissue lipid uptake in mice after β-cell ablation . ( A and B ) Total 3H radioactivity in adipose tissues at the end of the 3H-triolein injection experiment . Radioactivity is calculated as percentage of input , and normalized against tissue weight . Sc: subcutaneous . Gon: gonadal . Mes: mesenteric . WAT: white adipose tissue . BAT: brown adipose tissue . Mice were either controls or used 3 weeks after STZ treatment . n ≥ 6 mice per condition . Data are presented as the mean ± SEM . *p < 0 . 05 for Adn+/+ vs Adn−/− . ( C ) Representative H&E stains of sections from subcutaneous WAT , BAT , and gonadal WAT . ( D ) Representative confocal microscopy of BODIPY fluorescence signal in whole-mount subcutaneous WAT . PANIC-ATTAC mice were at 3 weeks post dimerizer and subjected to an oral gavage of BODIPY-labeled fatty acids ( 2 μg/g BDW ) 3 hr before tissue collection . ( E–G ) Representative immunofluorescence co-stain of apolipoproteins A1 ( E ) , B ( F ) , or E ( G ) ( red ) , and endomucin ( green ) in subcutaneous WAT . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 01510 . 7554/eLife . 03851 . 016Figure 3—figure supplement 1 . Tissue-specific catabolism of 3H-triolein . For the 3H-triolein chase experiment: ( A ) disappearance rate of 3H radioactivity in blood . ( B and C ) At the end of the chase experiment , 3H radioactivity in the organic ( B ) and aqueous ( C ) phases of tissue lipid extracts were measured . Data are presented as the mean ± SEM . *p < 0 . 05 for Adn+/+ vs Adn−/− . Related to Figure 3A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 01610 . 7554/eLife . 03851 . 017Figure 3—figure supplement 2 . Quantitation of BODIPY signal in subcutaneous WAT . Quantitation of BODIPY fluorescence intensity volume normalized against tissue area . n = 4 mice per phenotype . Data are presented as the mean ± SEM . **p < 0 . 01 vs P-Adn+/+ . Related to Figure 3D . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 01710 . 7554/eLife . 03851 . 018Figure 3—figure supplement 3 . ApoA1 immunofluorescence in gonadal WAT and BAT with quantitation . ( A and B ) Representative co-immunofluorescence of apolipoprotein A1 ( ApoA1 , red ) and endomucin ( green ) on gonadal white adipose tissue ( WAT , A ) and brown adipose tissue ( BAT , B ) . ( C–E ) Quantitation of ApoA1 immunofluorescence intensity volume normalized against tissue area in subcutaneous WAT ( C ) , gonadal WAT ( D ) , and BAT ( E ) . n ≥ 3 mice per condition . Data are presented as the mean ± SEM . *p < 0 . 05 , **p < 0 . 01 for STZ-treated Adn+/+ vs Adn−/− . #p < 0 . 05 for STZ-treated vs untreated Adn+/+ . Related to Figure 3E . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 01810 . 7554/eLife . 03851 . 019Figure 3—figure supplement 4 . ApoB immunofluorescence in gonadal WAT with quantitation . ( A ) Representative co-immunofluorescence of apolipoprotein B ( ApoB , red ) and endomucin ( green ) on gonadal WAT . ( B and C ) Quantitation of ApoB immunofluorescence intensity volume normalized against tissue area in subcutaneous WAT ( B ) and gonadal WAT ( C ) . n ≥ 3 mice per condition . Data are presented as the mean ± SEM . #p < 0 . 05 , ##p < 0 . 01 for STZ-treated vs untreated Adn+/+ . Related to Figure 3F . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 01910 . 7554/eLife . 03851 . 020Figure 3—figure supplement 5 . ApoE immunofluorescence in gonadal WAT with quantitation . ( A ) Representative co-immunofluorescence of apolipoprotein E ( ApoE , red ) and endomucin ( green ) on gonadal WAT . ( B and C ) Quantitation of ApoE immunofluorescence intensity volume normalized against tissue area in subcutaneous WAT ( B ) and gonadal WAT ( C ) . n ≥ 3 mice per condition . Data are presented as the mean ± SEM . ##p < 0 . 01 for STZ-treated vs untreated Adn+/+ . Related to Figure 3G . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 020 To determine whether the defective lipid uptake leads to lipoprotein accumulation in subcutaneous WAT , we examined apolipoproteins A1 , B , and E ( ApoA1 , ApoB , and ApoE , major components of high- , low- , and intermediate-density lipoproteins , respectively ) in situ by immunofluorescence ( Figure 3E–G ) . Compared to the euglycemic wild-type controls , STZ-treated animals showed elevated signals for apolipoproteins . Interestingly , most of the ApoA1 and ApoE signals , as well as part of the ApoB signal , located adjacent to endomucin , a marker for venous and capillary endothelial cells , reflecting apolipoproteins in circulation . The endomucin signal itself demonstrated increased vascular density in subcutaneous WAT of STZ-treated lipoatrophic animals , which was also observed in gonadal WAT ( Figure 3—figure supplement 3–5 ) . Furthermore , adiponectin knockout-induced ApoA1 accumulation was significant in subcutaneous WAT and BAT , but not in gonadal WAT ( Figure 3E and Figure 3—figure supplement 3 ) . Compared to STZ-treated wild-type mice , adiponectin null mice showed trends towards an increase in ApoB signal in both subcutaneous and gonadal WATs ( Figure 3F and Figure 3—figure supplement 4 ) . As for the ApoE signal , there was a trend towards an increase in subcutaneous WAT , but not in gonadal WAT ( Figure 3G and Figure 3—figure supplement 5 ) . Collectively , these data suggest an overall accumulation of apolipoproteins in the local circulation of adipose tissues of STZ-treated mice , which is exacerbated by adiponectin depletion predominantly in subcutaneous WAT . Insulin promotes lipid storage in adipose tissue via stimulating the intracellular insulin signaling cascades leading to enhanced extracellular lipoprotein lipase activity . A low dose of insulin treatment ( 0 . 2 mU/g body weight ) markedly suppressed the hormone-sensitive lipase ( HSL ) serine-660 phosphorylation , a marker positively associated with lipolysis , by 57% and induced Akt serine-473 phosphorylation by 8 . 6-fold in subcutaneous WAT of STZ-treated wild-type mice . However , STZ-treated adiponectin nulls showed a blunted response in HSL inhibition ( 41% ) , and an abolished Akt activation ( 13% ) , as compared to the same animal prior to insulin injection ( Figure 4A and Figure 4—figure supplement 1 ) . The unchanged post-heparin lipoprotein lipase activity ( Figure 4—figure supplement 2 ) is unlikely to be a major contributing factor to the impaired lipid uptake in the STZ-treated adiponectin null animals . The fatty acid translocase CD36 ( Figure 4B ) and fatty acid transport protein 1 ( FATP1 ) ( Figure 4C ) are also unlikely contributors , both of which show comparable distributions between genotypes . Expression of Scavenger Receptor Class B Member 1 ( SR-B1 ) , a high-density lipoprotein ( HDL ) receptor , was enhanced in STZ-treated adiponectin knockouts ( Figure 4D ) , which could reflect a compensatory response for the HDL accumulation in circulation ( Figure 3E and Figure 3—figure supplement 3C ) . 10 . 7554/eLife . 03851 . 021Figure 4 . Adiponectin is important for caveolar structures and Caveolin-1 expression in subcutaneous white adipose tissue of STZ-treated mice . ( A ) Western blots of insulin signaling molecules , hormone-sensitive lipase ( HSL , serine 660 phosphorylated and total ) and Akt ( serine 473 phosphorylated and total ) in inguinal subcutaneous WAT . 1 week after STZ treatment , overnight fasted mice were subjected to a tail vein injection of insulin ( 0 . 2 mU/g BDW ) . Every two adjacent lanes represent the paired fat pads from an individual mouse , before ( − ) or 5 min after ( + ) insulin injection . ( B–D ) Representative immunofluorescence of CD36 ( B ) , fatty acid transport protein 1 ( FATP1 ) ( C ) , and HDL receptor SR-B1 ( D ) on subcutaneous WAT of STZ-treated mice . ( E and F ) Transmission electron microscopy of subcutaneous white adipocytes . Vesicles <10 nm from the plasma membrane were defined as ‘PM associated’ . ( E ) Representative fields of adipocyte plasma membranes . LD: lipid droplet . M: mitochondrion . EC: endothelial cell . RBC: red blood cell . Arrowheads: examples of vesicles associating with plasma membrane . ( F ) Quantitation of plasma membrane-associated vesicles normalized against membrane length . n ≥ 17 fields per condition . ( G and H ) Representative confocal co-immunofluorescence of caveolin-1 ( red ) and endomucin ( green ) on subcutaneous ( G ) and gonadal ( H ) WAT . Arrowheads: examples of caveolin-1 signal on adipocytes . Data are presented as the mean ± SEM . **p < 0 . 01 for Adn−/− vs Adn+/+ after STZ treatment . ##p < 0 . 01 for STZ-treated vs untreated Adn+/+ . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 02110 . 7554/eLife . 03851 . 022Figure 4—figure supplement 1 . Quantitation of insulin effect on phosphorylation of HSL and Akt . Western blots were quantitated for infrared signal volume subtracting local background . Phosphorylated HSL ( A ) and Akt ( B ) were first normalized against the corresponding total protein . The insulin effect on phosphorylation was then calculated by normalizing the insulin-stimulated samples against their corresponding ones before insulin treatment . Data are presented as the mean ± SEM . Related to Figure 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 02210 . 7554/eLife . 03851 . 023Figure 4—figure supplement 2 . Lipoprotein lipase activity . Pre- and post-heparin plasma was collected from STZ-treated mice and assayed for total and hepatic lipase activity , which difference was calculated as activity from lipoprotein lipase . n = 3 mice per genotype , and assay was triplicated per sample . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 02310 . 7554/eLife . 03851 . 024Figure 4—figure supplement 3 . Quantitation of mitochondrial density on electron microscopic images . Quantitation of cellular mitochondrial areas , normalized against areas of cytoplasm excluding lipid droplets . n ≥ 7 cells per condition . Data are presented as the mean ± SEM . ##p < 0 . 01 for STZ-treated vs untreated Adn+/+ . Related to Figure 4E . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 02410 . 7554/eLife . 03851 . 025Figure 4—figure supplement 4 . Quantitation of caveolin-1 immunofluorescence . Quantitation of caveolin-1 immunofluorescence signal not related to endomucin signal in subcutaneous WAT ( A ) and gonadal WAT ( B ) . n ≥ 3 mice per condition . Data are presented as the mean ± SEM . *p < 0 . 05 for Adn+/+ vs Adn−/− . Related to Figure 3G , H . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 025 Employing transmission electron microscopy , we observed a twofold increase in the linear density of vesicular structures associated with the plasma membrane of the subcutaneous adipocytes in the wild-type mice after STZ treatment . In contrast , this upregulation of vesicle density was abolished in the adiponectin nulls ( Figure 4E , F ) . These changes were consistent with the 3H-triolein uptake data , which measured the lipid uptake capacity per gram tissue ( Figure 3A ) . Interestingly , STZ treatment enhanced mitochondrial density in both Adn+/+ and Adn−/− mice ( Figure 4—figure supplement 3 ) , consistent with the 3H-triolein oxidation capacity ( Figure 3—figure supplement 1C ) . Caveolin-1 , one of the major components in the plasma membrane and trans-Golgi network , was detectable in the majority of subcutaneous adipocytes in wildtype mice after STZ treatment . In contrast , in STZ-treated adiponectin null mice , the caveolin-1 signal was almost completely depleted in subcutaneous adipocytes , and the majority of the signal was associated with the endomucin-positive endothelial cells ( Figure 4G ) under those conditions . These changes were not observed in either the euglycemic controls or the gonadal WAT ( Figure 4G , H ) . These findings suggest that under conditions of limited insulin availability , adiponectin plays an important role in potentiating insulin sensitivity , supporting endocytic activity for triglyceride , and promoting lipid storage specifically in subcutaneous WAT . The lack of adiponectin exacerbates lipoatrophy and hyperlipidemia , and this may be due–at least in part–to the diminished insulin signaling and the selective loss of the caveolin-1 complex . We subsequently wanted to investigate whether adiponectin can rescue the dyslipidemia brought about by insulin deficiency in the form of a genetic gain-of-function mutant that overexpresses adiponectin . Adiponectin overexpressing mice ( Combs et al . , 2004 ) were crossed into the homozygous PANIC-ATTAC background ( P-AdnTg/+ ) and treated with the same high-dose of dimerizer as the P-Adn+/+ and P-Adn−/− mice to induce β-cell apoptosis . The adiponectin transgenic mice sustained higher levels of plasma adiponectin than the wild-type mice , also in the absence of insulin ( Figure 5A ) . Dimerizer treatment led to a moderate decrease in whole-body fat mass by week 2 . From that point onward , the adiponectin transgenic mice showed a complete recovery in fat mass by week 5 and sustained normal fat mass through week 10 , at which point the experiment was stopped ( Figure 5B and Figure 5—figure supplement 1 ) . Early post β-cell ablation , P-AdnTg/+ mice already showed significant improvements during a triglyceride tolerance test compared to P-Adn+/+ mice ( Figure 5C ) . Furthermore , the baseline serum triglyceride levels were significantly lower in P-AdnTg/+ mice compared to P-Adn+/+ mice at all stages ( Figure 5D ) , indicative of the powerful lipid-lowering effects of adiponectin on the clearance of circulating triglycerides . Subsequently , P-AdnTg/+ mice restored their serum ketone bodies to a level ( 0 . 12 ± 0 . 02 mM ) comparable to unchallenged WT controls ( 0 . 13 ± 0 . 01 mM ) and significantly lower than both the P-Adn+/+ and P-Adn−/− mice ( Figure 5E ) . We also observed trends towards a decrease in NEFAs in adiponectin transgenic mice ( Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 03851 . 026Figure 5 . Adiponectin improves systemic lipid metabolism in PANIC-ATTAC mice . ( A ) Serum samples were collected from P-Adn+/+ ( +/+ ) and P-AdnTg/+ ( Tg/+ ) mice , 0 , 1 , and 9 weeks after initial dimerizer treatment and subjected to Western blotting for adiponectin . Equal volume of serum was loaded for each lane . Serum samples from adiponectin knockout mice ( −/− ) were included as negative controls . ( B ) Fat mass presented as percentage of mouse body weight ( BDW ) . ( C ) Blood triglyceride during oral triglyceride tolerance test in mice 2 weeks after initial dimerizer treatment . ( D and E ) Plasma triglyceride ( D ) and total ketone bodies ( E ) in mice under fed status . Data of WT , P-Adn+/+ , and P-Adn−/− at week 10 were presented in Figure 1G , H . For B to E , n ≥ 5 mice per condition . Data are presented as the mean ± SEM . *p < 0 . 05 , **p < 0 . 01 for P-AdnTg/+ vs P-Adn+/+ . #p < 0 . 05 , ##p < 0 . 01 for WT vs P-AdnTg/+ . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 02610 . 7554/eLife . 03851 . 027Figure 5—figure supplement 1 . Body composition of PANIC-ATTAC mice . Fat mass ( A ) and lean mass ( B ) at 0 , 2 , 5 , 10 weeks after initial dimerizer treatment . n ≥ 6 mice per condition . Data are presented as the mean ± SEM . #p < 0 . 05 , ##p < 0 . 01 for P-AdnTg/+ vs WT . Related to Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 02710 . 7554/eLife . 03851 . 028Figure 5—figure supplement 2 . Fed NEFA levels in PANIC-ATTAC Mice . Serum non-esterified fatty acid ( NEFA ) levels in mice under fed status . n = 4–13 mice per condition . Data are presented as the mean ± SEM . ##p < 0 . 01 for P-Adn+/+ vs WT . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 028 Our findings suggest that increasing circulating adiponectin is sufficient to improve systemic lipid metabolism and promote fat mass recovery in the context of insulinopenic diabetes . The lack of available insulin action uncovers adiponectin's importance for lipid homeostasis in the absence of insulin . We have previously shown that adiponectin enhances INS-1 β-cell survival by reducing accumulation of ceramides , a family of lipotoxic sphingolipids ( Holland et al . , 2011 ) . To investigate the in vivo roles of adiponectin on lipotoxicity in β-cells , we measured different sub-species of sphingolipids in pancreatic islets isolated from WT , P-Adn+/+ , and P-AdnTg/+ mice 5 weeks post β-cell ablation . For most individual ceramide species as well as the total ceramide content , we observed a trend in diabetic P-Adn+/+ islets towards increased ceramides as compared to islets from euglycemic WT mice and a concomitant drop of ceramides in the P-AdnTg/+ islets ( Figure 6A ) . The same trends applied to dihydro-ceramides , which are ceramide biosynthetic precursors ( Figure 6B ) . The pro-survival lipids , sphingosine-1-phosphate and sphinganine-1-phosphate were below the detection limit in all islet preparations , and sphingoid bases were not significantly altered by adiponectin ( data not shown ) . Among the hexosyl-ceramides , lactosyl-ceramides displayed the same pattern of change as ceramides overall ( Figure 6C ) , while no significant changes were observed in glucosyl-ceramides ( Figure 6D ) . As the major storage pool of sphingolipids , most of the sphingomyelins also reproduced the trends seen in ceramides among the three genotypes ( Figure 6E ) . Taken together , our data suggest increased levels of lipotoxic species in P-Adn+/+ islets , which may account , at least in part , for the further loss of β-cells from week 5 to week 10 post the initial insult ( Figure 1A ) . In contrast , adiponectin transgenic mice reduced the lipotoxic sphingolipid content to levels seen in the non-diabetic WT mice . It is likely that this phenomenon accounts for the enhanced pro-survival effects on the β-cells . 10 . 7554/eLife . 03851 . 029Figure 6 . Adiponectin reduces lipotoxic sphingolipids in regenerating PANIC-ATTAC islets . Sphingolipids were assayed by mass spectrometry in pancreatic islets isolated from mice 5 weeks after initial dimerizer treatment and normalized against protein content of islet samples . n = 3–5 samples per genotype . Sphingolipid species are categorized as ( A ) ceramides , ( B ) dihydro-ceramides , ( C ) lactosyl-ceramides , ( D ) glucosyl-ceramides , and ( E ) sphingomyelins . In every panel , the first group of columns from the left represents the sums of all species combined . Data are presented as the mean ± SEM . *p < 0 . 05 , **p < 0 . 01 for P-AdnTg/+ vs P-Adn+/+ . ##p < 0 . 01 for P-Adn+/+ vs WT . N . D . : not detected . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 029 In addition to lipotoxic lipids , hyperglycemia could impose potent negative effects via glucotoxicity and subsequent pro-apoptotic oxidative and ER stress in β-cells . We wondered whether adiponectin counterbalances these negative cellular events as well . However , as judged by immunohistochemistry on pancreatic islets , no apparent difference was observed in 8-hydroxyguanosine ( an oxidative stress marker ) , or CHOP ( a marker for pro-apoptotic ER stress ) ( data not shown ) . As in the P-Adn+/+ and P-Adn−/− mice , the high-dose dimerizer treatment resulted in >85% ablation of β-cell mass in P-AdnTg/+ mice at week 2 ( Figure 7A and Figure 7—figure supplement 1 ) . In contrast to the prolonged β-cell loss in the P-Adn+/+ and P-Adn−/− mice from week 5 to week 10 ( Figure 1A ) , the P-AdnTg/+ β-cell area showed a significant recovery in β-cell mass during this period . The transgenic mice restored their islet mass to 29% of the WT controls that did not suffer β-cell ablation ( Figure 7A and Figure 7—figure supplement 1 ) . Importantly , the recovery of β-cell mass in P-AdnTg/+ mice was preceded by improvements in both the systemic lipid metabolism ( Figure 5B–D ) and the local islet lipotoxicity ( Figure 6 ) , potentially supporting a causal relationship of adiponectin-mediated lipid improvements leading to islet mass recovery . 10 . 7554/eLife . 03851 . 030Figure 7 . Adiponectin promotes β-cell recovery in PANIC-ATTAC mice . ( A ) Quantitation of insulin-immunostained cell area normalized to total pancreas area . n = 3–7 mice per condition . Source files are available in Figure 7—source data 1 . ( B ) Fasting blood glucose . n ≥ 5 mice per condition . ##p < 0 . 01 . ( C and D ) At 2 and 9 weeks after initial dimerizer treatment , P-AdnTg/+ were subjected to an oral glucose tolerance test . Plasma glucose ( C ) and insulin ( D ) were determined . n ≥ 5 mice per condition . *p < 0 . 05 , **p < 0 . 01 . ( E ) Pancreatic islets were isolated from dimerizer-treated P-Adn+/+ and P-AdnTg/+ mice at the recovery stage , with untreated P-Adn+/+ mice as controls . Insulin secretion from islets under basal ( 3 mM ) or stimulating ( 16 mM ) glucose concentrations was measured and normalized against the DNA content of islets . n = 3–8 samples per condition . **p < 0 . 01 for dimerizer-treated P-Adn+/+ vs P-AdnTg/+ . ##p < 0 . 01 for dimerizer-treated vs untreated P-Adn+/+ . ( F and G ) Representative immunofluorescence ( red ) of Ki-67 ( F ) and BrdU ( G ) stains in pancreatic islets of mice 5 weeks after initial dimerizer treatment , merged with insulin ( green ) and DAPI ( blue ) . Arrowheads: Ki-67+ Insulin+ ( F ) or BrdU+ Insulin+ ( G ) cells . Insets: representative nuclear signal ( purple ) . For BrdU incorporation , mice were subjected to an i . p . injection of BrdU at the dose of 100 μg/g BDW 6 hr before sacrifice and tissue processing . Data are presented as the mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 03010 . 7554/eLife . 03851 . 031Figure 7—source data 1 . Source files for insulin-positive cell area quantitation . The zip file contains all the 2400-dpi scanned images of insulin-immunostained pancreas sections used for quantitation of insulin-positive cell area . Insulin is stained as brown , and the whole section is counterstained as blue with hematoxylin . Folders are named after genotypes ( WT or P-AdnTg+ ) and subfolders after time points ( week 0 , 2 , 5 , or 10 ) post initial dimerizer administration . Related to Figure 7A . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 03110 . 7554/eLife . 03851 . 032Figure 7—figure supplement 1 . Adiponectin promotes β-cell recovery in PANIC-ATTAC mice . Representative immunofluorescence of glucagon ( red ) and insulin ( green ) on pancreatic islets from WT and P-AdnTg/+ mice at 0 , 2 , and 10 weeks after dimerizer treatment . Related to Figure 7A . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 03210 . 7554/eLife . 03851 . 033Figure 7—figure supplement 2 . Adiponectin restores glycemic control in female and aged PANIC-ATTAC mice . Dimerizer treatment was applied to 2-month old females ( n = 4–6 mice per condition ) ( A ) and 1 . 5-year old males ( n = 3–11 mice per condition ) ( B ) . Fasting blood glucose was measured . Data are presented as the mean ± SEM . *p < 0 . 05 , **p < 0 . 01 . Related to Figure 7B . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 03310 . 7554/eLife . 03851 . 034Figure 7—figure supplement 3 . Adiponectin enhances in vivo GSIS in PANIC-ATTAC mice . Plasma C-peptide2 levels during oral glucose tolerance tests . Data are presented as the mean ± SEM . *p < 0 . 05 . Related to Figure 7C , D . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 03410 . 7554/eLife . 03851 . 035Figure 7—figure supplement 4 . Adiponectin enhances insulin content and in vitro GSIS in islets isolated from PANIC-ATTAC mice . For the glucose-stimulated insulin secretion from isolated islets: ( A ) plasma glucose of mice . ( B ) Insulin content in isolated islets normalized against DNA content . ( C ) Insulin secretion under basal ( 3 mM ) or stimulating ( 16 mM ) glucose concentration normalized against the insulin content of islets . n = 3–8 samples per condition . Data are presented as the mean ± SEM . *p < 0 . 05 for dimerizer-treated P-Adn+/+ vs P-AdnTg/+ . #p < 0 . 05 , ##p < 0 . 01 for dimerizer-treated vs untreated P-Adn+/+ . Related to Figure 7E . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 03510 . 7554/eLife . 03851 . 036Figure 7—figure supplement 5 . Quantitation of proliferating β-cells . All islets on the pancreas sections were counted and calculated for the percentage of Ki-67+ insulin+ cells ( A ) or BrdU+ insulin+ cells ( B ) within the population of insulin+ cells . n ≥ 3 mice per genotype . Data are presented as the mean ± SEM . Related to Figure 7F , G . DOI: http://dx . doi . org/10 . 7554/eLife . 03851 . 036 With the partially restored β-cell mass , P-AdnTg/+ mice significantly down-regulated their fasting blood glucose level to 199 ± 21 mg/dl ( Figure 7B ) . The restoration to euglycemia was observed in both female ( Figure 7—figure supplement 2A ) and aged male P-AdnTg/+ mice ( Figure 7—figure supplement 2B ) , supporting a sex- and age-independent effect of adiponectin in β-cell regeneration . We further examined the adiponectin-driven improvements in glucose metabolism and β-cell function in the P-AdnTg/+ mice . The mice displayed significant improvements in glucose tolerance ( Figure 7C ) , fasting insulin , and in vivo GSIS ( Figure 7D and Figure 7—figure supplement 3 ) . To directly test the β-cell function during the recovery stage , we isolated pancreatic islets from the P-Adn+/+ and P-AdnTg/+ mice and subjected them to in vitro GSIS assays , with non-treated , euglycemic mice as controls ( Figure 7—figure supplement 4A ) . The P-AdnTg/+ islets showed an ∼fivefold increase in insulin secretion under both basal conditions and upon exposure to elevated glucose levels ( Figure 7E ) . This improvement could be attributed to increases in both overall islet insulin content ( 1 . 7-fold , Figure 7—figure supplement 4B ) and exocytic activity ( 2 . 2-fold , Figure 7—figure supplement 4C ) . Enhanced proliferative activity was observed prior to the major regeneration of β-cells in the P-AdnTg/+ mice . Five weeks post ablation , P-AdnTg/+ mice showed higher ratios of Ki-67-positive ( Figure 7F and Figure 7—figure supplement 5A ) and BrdU-positive ( Figure 7G and Figure 7—figure supplement 5B ) cells in insulin-positive cells than seen in the islets of P-Adn+/+ or WT mice . The potent adiponectin-driven improvements in systemic lipid metabolism and amelioration of local lipotoxicity may be an important facilitative component towards β-cell proliferation and may contribute to the eventual recovery of islet mass .
In this study , we uncover for the first time a pathological condition under which adiponectin is indispensable for survival . When insulin levels decrease by >90% after intensive β-cell loss , adiponectin ensures the minimal homeostasis for lipid metabolism . Consistent with insulin action promoting lipid storage in adipose tissue , adiponectin exerts a similar effect on lipid uptake , but does so by distinct mechanisms . First , adiponectin mediates lipid uptake specifically in subcutaneous WAT , but not in epididymal WAT . Hepatic VLDL secretion and fatty acid metabolism are unchanged . Second , adiponectin regulates neither lipoprotein lipase activity under insulinopenic conditions nor the intracellular translocation of fatty acids . Rather , caveolin-1 is reduced , thereby reducing endocytosis of lipids . This is in agreement with our previous reports on caveolin-1 knockout mice , which show significantly lower body weight and fat mass than wild-type controls under both regular chow and high-fat diet regimen ( Razani et al . , 2002; Wernstedt Asterholm et al . , 2012 ) . Moreover , our adiponectin knockout mice with insulin deficiency recapitulate the lipodystrophic phenotypes reported in caveolin-1 null mice , including impaired triglyceride clearance , hyper-triglyceridemia , adipocyte hypotrophy , without a change in lipoprotein lipase activity ( Razani et al . , 2002 ) . Third , adiponectin potentiates insulin signaling and the suppression on lipolysis mediated by the trace levels of insulin present . It is tempting to speculate that adiponectin exerts autocrine/paracrine actions on subcutaneous adipose as a hormone , motivating future studies on signaling pathways and transcriptome/proteome analysis under insulinopenic conditions . The generation of mouse models allowing for conditional elimination of both adiponectin receptors adipoR1 and adipoR2 should be insightful . In cases where both insulin and adiponectin are depleted , fat mass becomes critically low , accompanied by exceedingly high triglycerides and ketone bodies in circulation , resulting in major mortality . Our findings also underscore adiponectin as a messenger for the crosstalk between adipose tissue and pancreatic β-cells , especially under insulinopenic conditions . Massive pancreatic β-cell failure leads to hypoinsulinemia and dyslipidemia ( Dunn , 1992 ) . The ensuing aggravated lipotoxicity further impairs β-cell function and survival ( Lupi et al . , 2002; Kusminski et al . , 2009 ) . Our findings demonstrate that adiponectin can disrupt this vicious cycle at multiple levels and , in so doing , promote potent regenerative effects on functional β-cell mass . Adiponectin improves lipid storage in adipose tissue and improves systemic lipid metabolism . These general improvements in dyslipidemia may contribute , at least in part , to the reduction of β-cell lipotoxicity in the adiponectin overexpressing mouse , as reflected by ceramide measurements . The reduced level of local lipotoxicity may mediate β-cell survival and proliferation . Overexpression of adiponectin is sufficient to regenerate β-cells and regain glycemic control . Taken together , our data highlight the link between lipid metabolism and β-cell maintenance and identify adiponectin as a key mediator of this process . This does not exclude a direct effect of adiponectin on β-cells , since adiponectin is avidly binding to β-cells in vivo and exerts potent cytoprotective effects on β-cells under these conditions ( Holland et al . , 2011 ) . The major impediments for effective β-cell regeneration may not only relate to the inherently low proliferation rate of β-cells , but also due to the systemic highly lipotoxic environment due to severe hyperlipidemia , both in type 1 diabetes as well as late stage type 2 diabetes . In vitro experiments have provided strong evidence that lipotoxicity can impair β-cell function and survival ( Maedler et al . , 2001; Rakatzi et al . , 2004; Hoppa et al . , 2009; Holland et al . , 2011 ) . Unger and colleagues demonstrated in rats that pancreatic islets transplanted into the hepatic portal area were destroyed by the local hyperlipidemic environment ( Lee et al . , 2007 ) and islets from Zucker diabetic fatty ( ZDF ) rats are subject to lipotoxic destruction via ceramide ( Shimabukuro et al . , 1998 ) . Poitout and colleagues have established that wild-type islets similarly develop ceramide-induced impairments in islet function under hyperlipidemic/hyperglycemic conditions ( Kelpe et al . , 2003 ) . In the PANIC-ATTAC model , the prolonged β-cell loss in adiponectin wild-type and adiponectin null mice is unlikely due to prolonged exposure to dimerizer , which has a half-life in mice of ∼5 hr . Rather , this is the result of glucotoxicity , lipotoxicity , oxidative stress , and/or ER stress . In the PANIC-ATTAC islets , we did not detect any differences in oxidative stress and ER stress markers between adiponectin wild-type and transgenic adiponectin overexpressing mice ( data not shown ) . However , we did see differences at the level of lipotoxic intermediates as judged by the results from the islet ceramide assays that revealed a reduction of lipotoxic ceramides as well as their precursors and derivatives . Ultimately , this improved microenvironment allows for increased recovery of islet mass .
Mice were bred and maintained on a 12-hr dark/light cycle , with ad libitum access to water and regular chow diet ( #5058; LabDiet , St . Louis , MO ) . The strains were generated and previously described by our laboratory: PANIC-ATTAC ( Wang et al . , 2008 ) , adiponectin overexpressing mice ( Combs et al . , 2004 ) , and adiponectin null mice ( Nawrocki et al . , 2006 ) . All mice were maintained on a FVB background . Body composition including fat mass and lean mass were measured with a Bruker Minispec mq10 . Food intake was recorded on individually housed mice for 5 consecutive days . All protocols for mouse use and euthanasia were reviewed and approved by the Institutional Animal Care and Use Committee of the University of Texas Southwestern Medical Center ( UTSW ) . Primer pairs for genotyping PCR were: 5′-GAAAGTGCCCAAACTTCAGAGCATTAGG-3′ and 5′-AACTGAGATGTCAGCTCATAGATGGGGG-3′ for PANIC-ATTAC; 5′-GTTCCTCTTAATCCTGCCCAGTC-3′ and 5′-CCCGGAATGTTGCAGTAGAACTTG-3′ for adiponectin transgenic; 5′-TTGGACCCCTGAACTTGCTTCACACC-3′ and 5′-GGATGCGGTGGGCTCTATGGCTTC-3′ for adiponectin knockout allele; 5′-TTGGACCCCTGAACTTGCTTCACACC-3′ and 5′-TCCTGAGTTCAATTCCCAGCACCCAC-3′ for adiponectin wildtype allele . The PCR program was: 95°C for 5 min , followed by 35 cycles of 95°C for 15 s , 62°C for 30 s , and 72°C for 30 s , ended with 72°C for 3 min . Mice were subjected to six daily intraperitoneal ( i . p . ) injections of the dimerizer AP20187 ( Clontech , Mountain View , CA ) at the dose of 0 . 5 μg/g body weight ( BDW ) /day . The dimerizer was stored at −20°C as 12 . 5 g/l solution in 100% ethanol , and freshly diluted in 2% Tween 20 with 10% polyethylene glycol 400 before injection . Mice were fasted for 6 hr and subjected to a single i . p . injection of streptozotocin ( #S1030; STZ , Sigma , St . Louis , MO ) at the dose of 135 μg/g BDW . STZ was stored at −20°C as powder and freshly diluted in ice-cold sodium citrate buffer ( 0 . 1 M , pH 4 . 5 ) before injection . Mice were euthanized by cervical dislocation following isoflurane anesthesia . Tissues were immediately collected and fixed overnight in 10% buffered formalin . Afterward , tissues were rinsed with 50% ethanol for three times , embedded in paraffin blocks by the University of Texas Southwestern Medical Center Molecular Pathology Core , and sliced for 5-μm sections . For BrdU incorporation , mice were subjected to an i . p . injection of BrdU ( 100 μg/g BDW ) 6 hr before sacrifice . Primary antibodies and dilution for immunostaining or immunofluorescence were: insulin ( #A0564; 1:500; Dako , Carpinteria , CA ) , glucagon ( #18-0064; 1:500; Invitrogen , Grand Island , NY ) , apolipoprotein A1 ( #ab20453; 1:100; Abcam , Cambridge , MA ) , endomucin ( #sc-65495; Santa Cruz Biotechnology , Santa Cruz , CA ) , apolipoprotein B ( #ab20737; 1:50 , Abcam ) , apolipoprotein E ( #sc-6384; 1:100 , Santa Cruz Biotechnology ) , CD36 ( #NB400-144; 1:100; Novus Biologicals , Littleton , CO ) , FATP1 ( 1:25 , Dr Andrea Stahl ) , SR-B1 ( #NB400-104; 1:50; Novus ) , Caveolin-1 ( #610493; 1:100; BD Biosciences , San Jose , CA ) , BrdU ( #MCA2060; 1:50; AbD Serotec , Raleigh , NC ) , and Ki-67 ( #M7249; 1:50; Dako ) . For β-cell area quantitation , at least four sections per mouse pancreas , >50 μm apart from each other , were immunostained for insulin . Whole slides were scanned as color images with an Epson Expression 10 , 000 XL photo scanner at a resolution of 2400 dpi . The insulin-positive brown area and the total pancreas area were quantitated with Adobe Photoshop , with a tolerance of 64 and 24 , respectively . Immunofluorescence sections were examined with a Zeiss Axio Observer Z1 inverted microscope or a Leica TCS SP5 confocal microscope . Electronic/optical settings for image acquisition and parameters for linear digital processing were consistent among samples within the same experiment . Fluorescence intensity and area were quantitated with ImageJ . Mice were fed ad libitum or fasted for 6 hr unless indicated otherwise . Blood was collected from a tail nick with a heparinized Microhematocrit capillary tube ( #22-362-566; Fisher Scientific , Pittsburgh , PA ) for plasma or a plain capillary ( # 22-362-574; Fisher Scientific ) for serum . Glucose was assayed with PGO enzymes ( #P7119; Sigma ) plus o-dianisidine ( #F5803; Sigma ) . In serum samples , triglyceride was assayed with Infinity Triglycerides Liquid Stable Reagent ( #TR22421; Thermo Scientific , Waltham , MA ) , NEFAs with HR Series NEFA-HR ( 2 ) ( #999-34691; #995-34791; #991-34891; #993-35191; Wako , Richmond , VA ) , glycerol with Free Glycerol Reagent ( #F6428; Sigma ) , and total ketone bodies with Autokit Total Ketone Bodies ( #415-73301; #413-73601; Wako ) . ELISA kits were used to determine insulin ( #EZRMI-13K; Millipore , Billerica , MA ) , and C-peptide2 ( Millipore EZRMCP2-21K ) . For lipoprotein fractionation , fresh plasma was pooled and subjected to FPLC and assays for triglycerides and cholesterol at the UTSW Mouse Metabolic Phenotyping Core . Mice were fasted for 4–6 hr before the tests unless indicated . Glucose tolerance test was initiated by oral gavage of dextrose ( 2 mg/g BDW ) , and plasma was collected at 0 , 15 , 30 , 60 , 120 , and 180 min for glucose assay or at 0 , 15 , 60 , 120 min for insulin and C-peptide2 assays . Triglyceride tolerance test was initiated by oral gavage of 20% Intralipid ( 10 μl/g BDW , #2B6022; Baxter , Deerfield , IL ) , and serum was collected at 0 , 1 , 2 , 3 , and 4 hr for triglyceride assay . Hepatic triglyceride secretion assay was initiated by tail vein injection of tyloxapol ( 0 . 5 mg/g BDW , #T0307; Sigma ) , and serum was collected at 0 , 1 , 2 , 3 , and 4 hr for triglyceride assays . The lipolysis suppression assay was initiated by i . p . injection of bovine insulin ( 0 . 1 mU/g BDW ) , and serum was collected at 0 , 15 , 30 , 60 , and 120 min for glycerol , triglyceride , and NEFA assays . Mice were fasted for 4–6 hr and subjected to tail vein injection of recombinant adiponectin ( 2 μg/g BDW ) . Serum was collected at 0 , 1 , 2 , 3 , and 4 hr for triglyceride assays . cDNA was synthesized from the total RNA extract with the SuperScript II Reverse Transcriptase ( #18064-014; Invitrogen ) plus the RNaseOUT Recombinant Ribonuclease Inhibitor ( #10777-019; Invitrogen ) . Quantitative real-time PCR ( qPCR ) was performed with Power SYBR Green PCR Master Mix ( #4368708; Applied Biosystems , Carlsbad , CA ) on a 7900HT Fast Real-Time PCR System ( #4329001; Applied Biosystems ) , at least triplicate . See supplementary file 1 for primer sequences . The 3H-triolein chase experiment was performed as previously described ( Kusminski et al . , 2012 ) . The 3H-triolein ( #NET431001MC; PerkinElmer , Waltham , MA ) was dried under nitrogen flow , emulsified with 5 vol of 5% intralipid in PBS by 40 s sonication , and diluted 1:10 with PBS for tail vein injection . 10 μl of the injection solution were preserved for 3H scintillation counting to calculate input , and 200 μl per mouse were injected following 3–6 hr of fasting . At 1 , 2 , 5 , 10 , and 15 min after injection , 10 μl of tail blood was collected , immediately added into a scintillation vial containing 5 ml 3a70B complete counting cocktail ( #111154; RPI Corp , Mount Prospect , IL ) , and shaken vigorously to disperse . Tissues were then immediately collected , weighed , immersed in 0 . 75 ml chloroform–methanol 2:1 mixture at 4°C overnight , and disrupted by bead vortex in a MagNA Lyser ( #03358968001; Roche ) , at 5000 rpm , 30 s for two times . The content was then mixed with 0 . 5 ml 1 M CaCl2 and centrifuge at 3000 rpm , 4°C for 30 min . The chloroform phase at the bottom , which contained the hydrophobic incorporated 3H-triolein , was transferred to a scintillation vial , air-dried completely in a fume hood , and mixed with 5 ml counting cocktail . The water–methanol phase supernatant , which contained the hydrophilic oxidized 3H-triolein , was then transferred to a scintillation vial containing 5 ml counting cocktail , and shaken vigorously to mix . All the vials were counted 5 min for 3H scintillation in a Beckman Coulter LS6500 multi-purpose scintillation counter . Four-parameter double exponential decay regression was applied to calculate the whole body clearance rate based on the 3H activity in blood . Mice were fasted for 4 hr and subjected to an oral gavage of BODIPY 500/510 C1 , C12 fatty acids ( 2 μg/g BDW , #D3823; Molecular Probes ) . 3 hrs later , mice were euthanized and subcutaneous WAT was dissected . Tissues were immediately frozen in liquid nitrogen and stored at −80°C . Tissue pieces <2 mm were excised , mounted with fluorescence mounting medium ( #S302380; Dako ) on microscope slides with cover glasses , and examined for fluorescence with argon-ion laser excitation at 488 nm on a Leica TCS SP5 confocal microscope . After overnight fasting , mice were anesthetized with isoflurane . The left inguinal subcutaneous WAT was excised . 5 mins after a tail vein injection of insulin ( 0 . 2 mU/g BDW ) , the right inguinal subcutaneous WAT was excised . Tissues were immediately frozen in liquid nitrogen and subsequently processed for protein lysate as previously described ( Ye et al . , 2010a ) . Protein separation and transfer were performed with 4–15% Mini-PROTEAN TGX Gels ( #456-1086; Bio-Rad , Hercules , CA ) , Trans-Blot Turbo Mini Nitrocellulose Transfer Packs ( #170-4158; Bio-Rad ) , and Trans-Blot Turbo Transfer Starter System ( #170-4155; Bio-Rad ) . Immunoblots were imaged and quantitated with an IRDye 800CW IgG second antibody ( #926-32211; LI-COR ) and an Odyssey CLx infrared imaging system ( LI-COR , Lincoln , NE ) . Primary antibodies included pSer660-HSL ( #4126; 1:1000; Cell Signaling , Beverly , MA ) , HSL ( #4107; 1:1000; Cell Signaling ) , pSer473-Akt ( #9271; 1:1000; Cell Signaling ) , and Akt ( #9272; 1:1000; Cell Signaling ) . Lipase activity was measured as previously described ( Razani et al . , 2002 ) , with minor modifications . Briefly , pre- and post-heparin plasma was collected from mice before and after 15 min after tail vein injection of heparin ( 1 . 5 U/g BDW , #H3393; Sigma ) , respectively . Per 0 . 2 ml reaction , 10 μl of plasma was incubated with a triglyceride emulsion containing ∼107 cpm/ml 3H-triolein ( #NET431001MC; PerkinElmer ) at 37°C water bath for 90 min for total lipase activity , or with 1 . 5 M NaCl for hepatic lipase activity . Reaction was terminated by addition of 3 . 25 ml methanol:chloroform:heptane ( 1 . 41:1 . 25:1 ) and 1 . 05 ml K2CO3 ( pH 10 . 5 ) . After vigorous agitation and centrifugation , 1 ml of the NEFA-containing aqueous phase was transferred and assayed for 3H radioactivity on a scintillation counter . Serial dilutions of a lipoprotein lipase ( 0 , 10 , 20 , 50 , and 100 ng , #L9656; Sigma ) were parallel assayed as standard controls , and the lipase activities of plasma samples were calculated as equivalents of the standard lipase . The plasma lipoprotein lipase activity was calculated by subtracting the hepatic lipase activity from the total . Tissues were processed at the UTSW Electron Microscopy Core Facility . Sections were examined with a JEOL 1200 EX electron microscope and photographed with a Sis Morada 11 MegaPixel side-mounted CCD camera . Proteins in the same volume of serum were separated by 8% SDS-PAGE , transferred to nitrocellulose membranes ( #162-0112; Bio-Rad ) , and blotted with a mouse adiponectin antibody ( Schraw et al . , 2008 ) . The membrane was subsequently processed with the Odyssey imaging system ( LI-COR ) . Mice were euthanized by cervical dislocation following isoflurane anesthesia , and pancreatic islets were isolated as previously described ( Ye et al . , 2010b ) . Immediately after euthanasia , the major duodenal papilla of the mouse was blocked with a micro bulldog clamp . Ice-cold digestion solution , that is , Hank's Balanced Salt Solution ( HBSS ) with 0 . 1 g/l Liberase TL ( #05401020001; Roche , Indianapolis , IN ) , 0 . 1 g/l DNase I ( #10104159001; Roche ) , 2 . 5 mM HEPES , 8 mM glucose , 0 . 2% BSA , pH 7 . 2–7 . 4 , was inject via the common bile duct into the pancreatic duct . The inflated pancreas was transferred into a scintillation vial with digestion solution on ice . Subsequently , the vial was incubated at 37°C water bath for 25 min and shaken vigorously to disperse the pancreas . The digested content was resuspended with ice-cold HBSS and settled on ice for 2–5 min , and supernatant was removed without disturbing the precipitate . The wash was repeated until the supernatant became clear , and the content was transferred into a 6-cm Petri dish with ice-cold HBSS plus 2 . 5 mM HEPES , 8 mM glucose , 0 . 2% BSA , pH 7 . 2–7 . 4 . Under a dissection microscope , islets of Langerhans were picked with a 200-μl pipette to new dishes until free of exocrine pancreas content . The freshly isolated islets were either frozen immediately in liquid nitrogen followed by −80°C storage for sphingolipid assays , or transferred to culture medium for in vitro insulin secretion assays . Frozen mouse pancreatic islet samples ( 70–130 islets per sample ) were homogenized in 0 . 5 ml aqueous buffer ( 25 mM HEPES , pH 6 . 8 ) using a sonic dismembrator system equipped with a 1/8-inch probe . The samples were kept on ice during the homogenization process . 50 μl of the homogenate was taken for protein determination by BCA assay , and the remaining sample was added to 2 ml of organic extraction mixture ( isopropanol/ethyl acetate 15:85; vol:vol ) . Immediately afterward , 20 µl of internal standard solution ( diluted 1:4 in ethanol , #LM6005; Avanti Polar Lipids , Alabaster , AL ) was added . The mixture was vortexed and sonicated in ultrasonic bath for 10 min at 40°C . The samples were then allowed to reach room temperature and centrifuged at 3500 rpm in Sorvall Legend XTR ( #75004521; Thermo Scientific ) . The supernatant was transferred to a new tube and the aqueous phase was re-extracted . Supernatants were combined and evaporated under nitrogen . The dried residue was reconstituted in 200 µl of HPLC solvent B ( methanol/formic acid 99:1; vol:vol containing 5 mM ammonium formate ) for LC-MS/MS analysis . The freshly isolated pancreatic islets were cultured overnight in RPMI 1640 medium with 10% fetal bovine serum , 1% antibiotics , 8 mM glucose , and 0 . 2% BSA . 10–15 islets per sample were equilibrated in 1 ml secretion assay buffer ( SAB ) with 3 mM glucose for 1 hr and then transferred to 1 ml SAB with 3 mM glucose . After 1 hr incubation , 0 . 15 ml of SAB was frozen immediately in liquid nitrogen and stored at −80°C for insulin assay , and islets were transferred to 1 ml SAB with 16 mM glucose . After 1 hr incubation , 0 . 15 ml of SAB was sampled for insulin assay as previously , and islets were picked into 200 μl of 1 M acetic acid with protease inhibitors ( #11836170001; Roche ) . Islets were subjected to 30 s sonication on ice , and 5 μl of the islet lysate were diluted 1:500 in 1 M acetic acid with protease inhibitors for insulin assay . 20 μl of 10 N NaOH was added to neutralize the remained islet lysate and then mixed with 300 μl of 100% isopropanol . DNA was precipitated after centrifugation at 13 , 000 rpm , 4°C for 15 min , rinsed briefly with 70% ethanol , and re-suspended in 100 μl 10 mM Tris-Cl ( pH 8 . 5 ) . DNA concentration was determined by SYBR green incorporation . Two-tailed student's t test was applied for all pairwise comparisons . Kaplan–Meier survival curves were compared by the log-rank test . Statistical significance was accepted at p < 0 . 05 . | Fat tissue is essential for health . Fat cells store energy and release it when it is needed; they also release hormones that are important for the health of our heart and for regulating our metabolism . One of these hormones , adiponectin , helps cells to remove fat molecules from the bloodstream . This allows the body to maintain appropriate cholesterol levels and prevents fatty build-ups from blocking blood vessels , which is associated with heart disease . Adiponectin also helps cells respond to insulin , a hormone that regulates blood sugar levels , and thus helps to prevent diabetes . Despite this hormone's important roles in health , mice that lack adiponectin can thrive under normal conditions . Adiponectin becomes essential , however , when blood sugar or fat levels are considerably altered . For example , when mice without adiponectin are fed a high fat-content diet , they become insulin-resistant . Moreover , certain diabetes drugs that boost insulin sensitivity only work if adiponectin is present in the body . Adiponectin helps to keep the β-cells that produce insulin alive . In patients with diabetes , β-cells slowly die , and this leads to a poor insulin response and an imbalance in the amount of fats and sugars in the cells . This is toxic to the β-cells; and as more β-cells die , less insulin is produced to control sugar levels , and the condition worsens . Adiponectin appears to protect the β-cells against this vicious cycle , but the details of how it does so are unclear . Ye et al . used a mouse model in which β-cells were destroyed to see what adiponectin does when insulin is in short supply . When insulin levels were extremely low , adiponectin was found to help one type of fat tissue absorb fat molecules from the bloodstream , which reduced blood cholesterol levels . It also protects fat cells from being destroyed when insulin levels are low . Ye et al . also found that mice that lack both insulin and adiponectin lose excessive amounts of fat tissue and develop high blood cholesterol levels , which lead to death . Increasing adiponectin levels in insulin-deficient mice , however , improves their blood cholesterol levels and protects β-cells from being destroyed . This allows the β-cells to begin regenerating . As the β-cells regenerate , the insulin-deficient mice developed better control over their blood sugar . Many people with type-1 diabetes ( which is caused by their own immune system destroying their β-cells ) currently rely on insulin injections and restricted diets to manage their condition . Ye et al . 's findings might lead to new strategies to restore β-cell production and blood sugar control; as such these findings will have important implications for the management of type-1 diabetes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2014 | Adiponectin is essential for lipid homeostasis and survival under insulin deficiency and promotes β-cell regeneration |
Breast cancer genomes have revealed a novel form of mutation showers ( kataegis ) in which multiple same-strand substitutions at C:G pairs spaced one to several hundred nucleotides apart are clustered over kilobase-sized regions , often associated with sites of DNA rearrangement . We show kataegis can result from AID/APOBEC-catalysed cytidine deamination in the vicinity of DNA breaks , likely through action on single-stranded DNA exposed during resection . Cancer-like kataegis can be recapitulated by expression of AID/APOBEC family deaminases in yeast where it largely depends on uracil excision , which generates an abasic site for strand breakage . Localized kataegis can also be nucleated by an I-SceI-induced break . Genome-wide patterns of APOBEC3-catalyzed deamination in yeast reveal APOBEC3B and 3A as the deaminases whose mutational signatures are most similar to those of breast cancer kataegic mutations . Together with expression and functional assays , the results implicate APOBEC3B/A in breast cancer hypermutation and give insight into the mechanism of kataegis .
Whole genome sequencing of 21 breast cancers recently revealed the presence in more than half the cancers of a novel form of localised hypermutation ( termed kataegis ) in which clusters of multiple same-strand mutations at C:G pairs are focused on multikilobase-long genomic regions with adjacent mutations within each cluster separated by one to several hundred base pairs ( Nik-Zainal et al . , 2012 ) . Although the mechanism underlying kataegis is unknown , the fact that the mutations occurred nearly exclusively at C residues preceded by a 5′-T suggested that AID/APOBEC cytidine-DNA deaminases might possibly be involved in the process since these enzymes are sensitive to the 5′-flanking sequence context ( Conticello et al . , 2007; Nik-Zainal et al . , 2012 ) . Members of the AID/APOBEC family of enzymes ( reviewed in Conticello et al . , 2007 ) deaminate cytosine in the context of a single-stranded polynucleotide substrate , and function in adaptive and innate immunity . AID acts on C residues in the DNA of the genomic immunoglobulin loci in activated lymphocytes to trigger antibody gene diversification whereas APOBEC3 family members , of which there are seven in humans , act on C residues in the DNA of viral replication intermediates ( usually in the cytoplasm ) as part of a host restriction pathway . Off-target deamination by AID results in nucleotide substitutions and genomic rearrangements in B lymphocyte tumours , some of which are implicated in oncogenesis ( reviewed by Gazumyan et al . , 2012 ) . Although AID is the only member of the AID/APOBEC family known to act physiologically on endogenous nuclear DNA , it is possible that other members of the AID/APOBEC family might occasionally get access to the nucleus and cause cancer-associated genomic damage or mutation ( Harris et al . , 2002; Beale et al . , 2004; Vartanian et al . , 2008; Stenglein et al . , 2010; Landry et al . , 2011; Nik-Zainal et al . , 2012; Nowarski et al . , 2012 ) . Here we have asked whether we could recapitulate cancer-like kataegis by expression of different AID/APOBEC enzymes in budding yeast and if so , use the tractability of yeast to gain insight into the kataegic process . The results provide insight not only into the mechanism of kataegis but also provide a strong pointer to the identity of the deaminases responsible for the kataegis observed in breast cancers .
Several members of the AID/APOBEC family members were expressed in yeast and all were found to give a significant increase in the mutation frequency as judged by the yield of colonies resistant to canavanine ( CanR ) ( Figure 1—figure supplement 1 ) . Genome sequencing , however , revealed that most CanR colonies had typically accumulated less than 10 point mutations ( >98% at C:G pairs ) during the period of AID/APOBEC induction and clonal expansion ( Figure 1A ) . A hyperactive mutant of AID ( AID*; Wang et al . , 2009 ) gave a significantly higher mutation load ( median of 25 mutations per genome ) . We therefore initially focused on the mutations in AID*-transformants ( 1078 mutations in total , of which all except 14 were at C:G pairs ) to see if there were signs of kataegis . 10 . 7554/eLife . 00534 . 003Figure 1 . AID/APOBEC-induced kataegis in yeast . ( A ) The total number of mutations detected in canavanine-resistant ( CanR ) AID/APOBEC yeast transformants , median frequency indicated . ( B ) Rainfall plot of genome-wide intermutational distances ( IMD ) in individual CanR AID* transformants . Clone identifier is indicated along the top , mutations shown as dots with the y-axis giving the distance to the next downstream mutation on the same chromosome . For each clone , dots are ordered sequentially along the genome . Dot colours represent: cluster mutations ( at C , red; G , black ) , unclustered mutations ( at C , pink; G , grey ) . Mutations at A:T ( 14 out of 1078 total ) , single mutations on individual chromosomes ( 15% of the database ) , the most downstream mutation in each chromosome and transformants without any multiply mutated chromosomes ( 5/40 ) are not depicted . Supplementary file 1B contains the location of all identified mutations . ( C ) Observed distribution of IMDs in the AID* dataset compared to a simulation assuming mutations are randomly scattered throughout the genome . ( D ) IMD plots of APOBEC3A/B/G*-expressing yeast transformants ( 28/78 transformants harboured no multiply mutated chromosome and are not depicted ) . ( E ) Detailed view of AID* mutation clusters . Each line represents an individual cluster with the clone identifier ( grey box ) , number of mutations in the cluster and total mutations in the clone indicated . Mutations are coloured as in ( B ) , a horizontal line indicates mutations that have coalesced , * indicates clusters localising within 10 kb of CAN1 . All clusters containing ≥5 mutations are depicted . ( F ) Mutation clusters identified in yeast APOBEC3 transformants . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 00310 . 7554/eLife . 00534 . 004Figure 1—figure supplement 1 . Characterisation of AID/APOBEC yeast transformants . ( A ) CanR frequencies of yeast transformants expressing different AID/APOBEC proteins . ( B ) AID/APOBEC protein levels in yeast transformants . Whole cell extracts from 24 hr galactose induced yeast transformants were immunoblotted with anti-FLAG ( M2; Sigma ) and anti-β-Actin ( ab8224; Abcam ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 004 The distances between neighbouring mutations in the AID* yeast transformants are displayed as rainfall plots ( Figure 1B ) . While the median overall intermutational distance ( IMD ) is 13 kb , it is apparent that rather than the mutations being scattered randomly over the genome , mutation distribution is bimodal ( Figure 1C ) . Dividing the mutations into two groups using k-means cluster analysis ( Hartigan and Wong , 1979 ) reveals that one group exhibit a median IMD of 156 kb with a distribution of distances that is as expected for a set of individual mutations randomly scattered over the yeast genome as judged by Monte Carlo simulation ( Figure 1C ) . We designate these as singlet mutations: they account for 52% of the total mutations . The remaining 48% of the mutations are much more closely spaced than would be expected on a random basis . We designate these as proximal mutations , which are separated from each other by a median IMD of only 727 bp with >99% of them being within 8 . 5 kb of their closest neighbour . The overwhelming majority of the proximal mutations in the AID* transformants do not occur as isolated mutational pairs but , rather , are found in clusters . Thus , if we define proximal mutations as a pair of mutations that are located <8 . 5 kb apart ( a distance that excludes 99% of the singlet mutations ) and define a cluster as a stretch of DNA containing ≥5 proximal mutations , we find that 75% of the AID*-induced proximal mutations are actually parts of clusters . These clusters typically extend over 6–15 kb ( with the full range detected being 1 . 8–30 kb ) and contain anything up to 26 mutations ( Figure 1E ) . This clustering is far in excess of anything that would be expected on a random basis . The level of mutation clustering observed with AID* is such that more than one-third of the transformants analysed ( 16/40 ) contain at least one mutation cluster . In affected clones , a quarter to two-thirds of all the mutations in the cell are concentrated in a small number of clusters that account for <0 . 2% of the entire genome . Similar clusters were also observed in yeast cells transformed with APOBEC3A and APOBEC3B as well as with the hyperactive APOBEC3G mutant APOBEC3G* ( Figure 1D , F ) . Like the cancer kataegis , the clustered mutations in the various yeast transformants showed a strong tendency towards strand polarity; mutations within a cluster occur predominantly at either a C residue or a G residue with over 88% of mutations being strand coordinated ( Figure 1D , F ) . Exploring the mutational spectra , we find that the majority ( 76% ) of the mutations in the yeast AID* transformants are C→T transitions , although transversions do occur and these are preferentially associated with the kataegic stretches ( Figure 2A and Table 1 ) . Transversions account for 54% of the kataegic mutations in the AID* transformants but for only 13% of the unclustered substitutions ( Table 1 ) . The same bias towards transversion mutations in the kataegic stretches is also observed in the APOBEC3A , 3B and 3G* transformants ( Figure 2A and Table 1 ) . 10 . 7554/eLife . 00534 . 005Figure 2 . Yeast kataegic clusters are associated with transversions , are reduced by UNG-deficiency and can be triggered by a double strand DNA break . ( A ) IMD plots of AID*/APOBEC transformants reveal preferential association of nucleotide transversions with kataegic clusters . Mutation datasets and presentation are as in Figure 1B but with transition mutations represented by yellow dots and transversions by blue dots . Density plots depict the overall distribution of transition ( Tn ) and transversion ( Tv ) mutations at C:G pairs . ( B ) IMD plots of AID*/APOBEC-expressing ung1Δ yeast transformants , depicted as in Figure 2A . Density plots compare the distributions of IMDs in AID* transformants of ung1Δ and wild type yeast . ( C ) All mutation clusters identified in AID*/APOBEC3A-transformants of ung1Δ yeast depicted as in Figure 1E . ( D ) Kataegis localised to a double strand break . Mutations in the vicinity of the CAN1 locus of ( I-SceI+APOBEC3G* ) transformants of either control cells or of a KanMX-ISceIRS derivative carrying a CAN1-proximal I-SceI recognition sequence . The I-SceI cut site is marked with an arrow . All CAN1 mutations in control cells and 33/36 CAN1 region mutations in KanMX-ISceIRS cells occur at the canonical APOBEC3G CC context . Two-thirds of the CAN1 region mutations in the KanMX-ISceIRS cells were transversions . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 00510 . 7554/eLife . 00534 . 006Figure 2—figure supplement 1 . Canavanine resistance frequencies of yeast transformants carrying an I-SceI recognition sequence ( I-SceIRS ) . CanR mutation frequency of yeast cells carrying an I-SceIRS 420 bp downstream of the CAN1 polyadenylation site ( KanMX-ISceIRS cells ) which have been induced for APOBEC3G* and/or I-SceI expression . * indicated p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 00610 . 7554/eLife . 00534 . 007Figure 2—figure supplement 2 . IMD plots of mutations from CanR ( I-SceI+APOBEC3G* ) transformants of control or KanMX-ISceIRS cells IMD plots coloured as in Figure 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 00710 . 7554/eLife . 00534 . 008Table 1 . Pattern of nucleotide substitutions at C:G pairs in AID*/APOBEC yeast transformants . All mutations at C:G pairs were computed as substitutions at CDOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 008AID*ung1Δ AID*rev1Δ AID*A3AA3BA3G*KataegicUnclusteredTotalC→T ( % ) 46877699 . 3100798178C→G ( % ) 4711210 . 20171620C→A ( % ) 7230 . 50432Total2747901064208877130121560 Whereas C→T transitions will likely arise through direct replication over uracils generated by cytidine deamination , transversions are presumably due to replication over abasic sites created through uracil excision by uracil-DNA glycosylase ( UNG ) . The transversions exhibit a strong ( 4- to 10-fold ) bias for C→G rather than C→A substitutions ( Table 1 ) suggesting that the replication over the abasic site could be catalysed by REV1 since this translesion polymerase ( by virtue of its deoxycytidyl nucleotide transferase activity ) inserts C opposite abasic sites ( Nelson et al . , 1996 ) . Indeed , deficiency in either REV1 or UNG led to a dramatic fall in the proportion of transversion mutations ( Table 1 ) . Deficiency in UNG also resulted in a fourfold increase in the average total mutation load in AID* transformants ( Supplementary file 1B ) . This presumably reflects diminished repair of the AID/APOBEC-generated uracils . There was an overall decrease in average total mutation load in AID* transformants of REV1 deficient yeast that might reflect the possible non-catalytic roles of REV1 during DNA damage repair ( Sale et al . , 2012 ) . Since UNG is required for the transversion mutations that are enriched in kataegic stretches , we asked whether UNG itself is required for kataegis . We found that the increased mutation load in AID* ung1Δ transformants was accompanied by a dramatic shift away from mutational clustering ( Figure 2B ) . Despite the fourfold ‘increase’ in mutation load , the percentage of mutations that are <8 . 5 kb from their neighbour ( proximal mutations ) actually ‘falls’ from accounting for 48% of the AID* mutations in wild type cells to 18% in ung1Δ transformants . Similarly , using the same criterion to distinguish clustered mutations in both datasets ( ≥5 linked mutations separated from their neighbour by <8 . 5 kb ) , 274 of the 1064 mutations observed in AID* wild type transformants are found within clusters compared to 28 of the 2088 mutations in the AID* ung1Δ transformants ( Supplementary file 1B ) . Thus , the median overall IMD actually ‘increases’ from 13 kb in AID* wild type transformants to 41 kb in the ung1Δ cells despite the increase in mutation load . These shifts do not simply reflect a fall in the proportion of clustered mutations due to the increased total mutation load . There is also an absolute fall in the amount of kataegis as judged by either the average number of clustered mutations per yeast transformant ( 6 . 9 in the wild type background vs 1 . 5 in the ung1Δ transformants ) or by the frequency of kataegic events ( 26 kataegic stretches in 40 AID* transformants in the wild type background vs 4 kataegic stretches in 19 AID* transformants in ung1Δ background ) ( Supplementary file 1B ) . Thus , it is evident that kataegis is substantially reduced in the UNG-deficient background , but not completely lost: a few residual clusters ( which exhibit evident strand polarity or bipolarity ) are still detected ( Figure 2C ) . The sensitivity of kataegis to UNG-deficiency indicates that kataegis is , at least in part , triggered through the generation of abasic sites . Cleavage at abasic sites by apyrimidinic endonucleases will lead to occasional double-stranded DNA breaks: kataegis could result from AID/APOBEC deamination of single-stranded DNA exposed during the resection phase of break repair . To determine whether the DNA break repair process predispose to kataegis , we introduced a target site for the restriction endonuclease I-SceI immediately downstream of the polyadenylation site of the CAN1 locus , and asked whether co-expression of I-SceI together with the APOBEC3G* deaminase increased the likelihood of kataegis in the vicinity . We chose to use APOBEC3G* for this experiment since it gave a good mutation load but a lower proportion of kataegic mutations than AID* ( Supplementary file 1B ) : any enhancement of kataegis would therefore be more readily detectable . Consistent with previous findings ( Poltoratsky et al . , 2010 ) , induction of I-SceI expression resulted in an increased frequency of deaminase-dependent selectable mutation at the linked CAN1 locus ( Figure 2—figure supplement 1 ) . More importantly , in the presence of APOBEC3G* , induction of a double-strand break increases the probability that mutations in its vicinity are kataegic ( Figure 2D ) . The mutation clusters in the breast cancers were analysed in the same way as the yeast clusters . Most of the cancers identifiable by rainfall plots as harbouring major regions of kataegis also contain clusters comprising smaller numbers of same-strand nucleotide substitutions at 5′-T-C dinucleotides ( Figure 3—figure supplements 2–4 ) . There is some diversity amongst the breast cancers with respect to the frequency/nature of the kataegic stretches . The main outlier is tumour PD4107a which carries a dense array of highly mutated ( and transition-restricted ) kataegic clusters coincident with extensive genomic rearrangement in a 14 Mb region of chromosome 6 ( Nik-Zainal et al . , 2012 ) . Overall , the kataegic clusters in the breast cancers are distributed over a similar range of lengths to those detected in the yeast transformants ( Figure 3A ) but the yeast clusters do typically contain a twofold to fivefold lower density of mutations ( a mean inter-mutational distance of 1220 bp within the AID* yeast kataegic stretches compared to 209 bp in PD4107a , 335 bp in PD4103a and 763 bp in PD4199a ) . 10 . 7554/eLife . 00534 . 009Figure 3 . Comparison of kataegic mutations in yeast AID/APOBEC transformants with those in breast cancers . ( A ) Comparison of the length , number of mutations and polarity of yeast kataegic clusters with those in breast cancers . The degree of strand polarity is indicated by colour intensity . The breast cancer data ( Nik-Zainal et al . , 2012 ) are a compilation from three tumours ( PD4103a , PD4107a , PD4199a ) chosen for their large number of clusters . ( B ) Context of the genome wide mutated C bases in yeast AID/APOBEC transformants with total numbers of mutations in each dataset indicated . ( C ) Context of the kataegic and singlet mutated C bases in selected breast cancers . Analyses of all sequenced breast cancers are presented in Figure 3—figure supplements 2–4 . ( D ) Similarity of sequence contexts of C mutations in breast cancer kataegic stretches compared to those of deaminase-induced C mutations in yeast . ( D1 ) Identity of the base at the −2 position of TC mutations in cancer kataegic regions and in APOBEC3A/B yeast transformants . The base compositions were normalised to the genomic base composition of the −2 base at TC dinucleotides . ( D2 ) Sequence contexts similarity p-value at positions ( −1 plus −2 ) to the mutated Cs . The contexts of all Cs throughout the yeast and human genomes are included for comparison . Mutation context of wild type versions of AID and APOBEC3G are shown in Figure 3—figure supplement 1 . Analysis of additional yeast transformants and breast cancers is shown in Figure 3—figure supplements 2–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 00910 . 7554/eLife . 00534 . 010Figure 3—figure supplement 1 . Mutation context of hyperactive APOBEC3G and AID are identical to the wild type proteins . Mutation context of CanR transformants expressing APOBEC3G and AID . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 01010 . 7554/eLife . 00534 . 011Figure 3—figure supplement 2 . Analysis of kataegic stretches and mutation distributions of 21 breast cancers . Analysis of tumours PD3851a to PD4086a . For each tumour , somatic mutations are displayed as intermutational distance ( IMD ) plots ( top left ) , as described in Figure 1B . Dots are coloured according to kataegic mutations ( black ) or singlet mutations ( red ) . The sequence context of C mutations in unclustered and kataegic locations are shown ( lower plots ) , with the number of mutations indicated , coloured as in Figure 3B . The length and number of mutations within each cluster is displayed ( top right ) , with the dot colour intensity indicating polarity . Data are taken from Nik-Zainal et al . ( 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 01110 . 7554/eLife . 00534 . 012Figure 3—figure supplement 3 . Analysis of kataegic stretches and mutation distributions of 21 breast cancers . Analysis of tumours PD4088a to PD4194a . For each tumour , somatic mutations are displayed as intermutational distance ( IMD ) plots ( top left ) , as described in Figure 1B . Dots are coloured according to kataegic mutations ( black ) or singlet mutations ( red ) . The sequence context of C mutations in unclustered and kataegic locations are shown ( lower plots ) , with the number of mutations indicated , coloured as in Figure 3B . The length and number of mutations within each cluster is displayed ( top right ) , with the dot colour intensity indicating polarity . Data are taken from Nik-Zainal et al . ( 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 01210 . 7554/eLife . 00534 . 013Figure 3—figure supplement 4 . Analysis of kataegic stretches and mutation distributions of 21 breast cancers . Analysis of tumours PD4198a to PD4248a . For each tumour , somatic mutations are displayed as intermutational distance ( IMD ) plots ( top left ) , as described in Figure 1B . Dots are coloured according to kataegic mutations ( black ) or singlet mutations ( red ) . The sequence context of C mutations in unclustered and kataegic locations are shown ( lower plots ) , with the number of mutations indicated , coloured as in Figure 3B . The length and number of mutations within each cluster is displayed ( top right ) , with the dot colour intensity indicating polarity . Data are taken from Nik-Zainal et al . ( 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 013 The vast majority of the breast cancer kataegic mutations occur at C residues preceded by a T ( Figure 3—figure supplements 2–4 ) . In tumours PD4103a , PD4107a and PD4199a , over 91% of the kataegic C mutations are preceded by a T ( Figure 3C and Figure 3—figure supplement 2–4 ) . However , any sensitivity to the identity of the base at position −2 is exceedingly mild ( average across the kataegic stretches in these three tumours is A:C:G:T = 32:20:19:29 compared to the human genome average of 30:20:20:30 ) ( Figure 3C ) . Previous experiments in which AID/APOBEC deaminases have been used to mutate specific bacterial or retroviral gene targets have revealed that individual deaminases show characteristic flanking nucleotide preferences . However , none of the deaminases analysed to date ( AID , APOBEC1 , APOBEC3C , 3DE , 3F and 3G ) has been shown to exhibit a preference that accords with the breast cancer kataegic mutations . Their flanking sequence preferences ( reviewed in Conticello et al . , 2007 ) are either radically different ( e . g . , AID prefers A/G at −1; APOBEC3G prefers C at −1 ) or else they do not show the high ( >90% ) preference for T at −1 coupled to a relative indifference to the base at −2 . The mutation spectra obtained in yeast allow the consensus motifs for individual deaminases to be refined owing to the large number of potential target sequences interrogated when mutational specificity is analysed on a genome-wide basis ( Figure 3B ) . With AID , APOBEC3C and APOBEC3G the yeast data essentially confirm the previously identified sensitivity to nucleotides located at positions −1 and −2 ( AID: 5′-WRC , APOBEC3C: 5′-TYC , APOBEC3G: 5′-CCC ) whilst allowing more precise quantitation of the degree of preference . With regard to APOBEC3A and APOBEC3B , the results reveal that ( consistent with earlier studies on APOBEC3B; Bishop et al . , 2004 ) , both enzymes strongly prefer a T at position −1 ( 91% ) . However , the yeast studies reveal that unlike other deaminases , both APOBEC3A and APOBEC3B show mild discrimination with regard to the bases located at position −2 ( APOBEC3A , A:C:G:T = 25:26:7:42; APOBEC3B , A:C:G:T = 35:14:20:31 ) ( Figure 3B ) . Comparing the contexts of the mutations obtained with the individual deaminases in yeast to those of the kataegic mutations in the cancers reveals that APOBEC3B has a signature that fits extremely well with the kataegis in PD4107a and PD4103a whereas APOBEC3A fits better with PD4199a ( p values in all three cases <0 . 005 ) ( Figure 3D ) . Interestingly , a marked bias towards a 5′-T is also seen amongst the individual singlet C mutations in several of the breast tumours ( e . g . , PD4199a , PD4005a and PD4120a; Figure 3C and Figure 3—figure supplement 2–4 ) . Although APOBEC3A has been shown to be capable of causing genomic damage in mammalian cells ( Vartanian et al . , 2008; Stenglein et al . , 2010; Landry et al . , 2011 ) , the same has not been shown for APOBEC3B . We find that induction of APOBEC3B expression in stably transfected human KBM7 cells ( like that of APOBEC3A ) results in loss of viability as well as in genomic DNA damage as judged by the induction of γH2AX ( a marker of the DNA damage response ) and of 53BP1 foci ( which identify DNA breaks ) ( Figure 4A , B ) . 10 . 7554/eLife . 00534 . 014Figure 4 . DNA damage by APOBEC3 family members and expression in breast cancer cell-lines . ( A ) Effect of enforced APOBEC expression on cell viability . ( A1 ) Stable transfectants of KBM7 cells that inducibly express APOBEC proteins were incubated with inducer ( doxycyclin ) and viability was monitored after 72 hr . ( A2 ) Expression of FLAG-tagged APOBECs after 24 hr doxycyclin treatment . ( B ) Enforced expression of APOBEC3A and APOBEC3B leads to induction of histone γH2AX and of 53BP1 foci . The percentage of cells ( B1 ) positive for histone γH2AX expression quantified by flow cytometry and ( B2 ) exhibiting punctate rather than diffuse 53BP1 staining quantified by immunofluorescence microscopy , with the foci number per cell indicated . ( C ) Expression of APOBEC family members in six human breast cancer cell-lines as well as in HEK293 cells was analysed by qRT-PCR of total cellular RNA . Expression is shown relative to the average of housekeeping genes HPRT and HMBS . The effect of phorbol ester ( PMA ) and interferon alpha ( INFα ) treatment on APOBEC3A and APOBEC3B levels is also shown . In all cases , * indicates p<0 . 1 , ** indicates p<0 . 0001 compared to control ( unpaired t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00534 . 014 Although kataegis could easily have resulted from a transient spike in deaminase expression during tumour development , it was interesting to ascertain whether APOBEC3A or APOBEC3B expression could be detected or induced in breast cancer-derived cells . RNA analysis revealed that although several APOBEC3s can be expressed in individual breast cancer cell-lines , the highest and broadest pattern of expression was evident with APOBEC3B ( Figure 4C ) . Consistent with studies in other cell-types ( Madsen et al . , 1999; Koning et al . , 2009; Stenglein et al . , 2010 ) , the expression of APOBEC3A and APOBEC3B in some of the breast cancer cell-lines could be enhanced by treatment with phorbol ester or interferon alpha .
Expression of AID/APOBECs cytidine deaminases in yeast generates mutations across the genome , a proportion of which are found in clusters . Since completing this work , two other groups have also demonstrated that cytidine deaminases can generate such clustered mutations ( Chan et al . , 2012; Lada et al . , 2012 ) . Here we extend on these findings , demonstrating the similarity of yeast and breast cancer kataegis , use yeast genetics to gain insight into the mechanism of kataegis and provide evidence identifying the individual APOBECs likely responsible for the kataegis in the breast cancers . The stimulation of local kataegis in yeast by the induction of an I-SceI break indicates that the process occurs during DNA break repair , most likely by AID/APOBEC-catalysed deamination of cytidines exposed on single-stranded DNA during the resection phase of homology-mediated repair . The lengths of the kataegic stretches ( mostly in the range 6–15 kb ) are in the same order as the extent of resection observed during yeast DNA break repair ( reviewed in Paull , 2010 ) although the occurrence and detection of kataegis is likely to bias towards longer stretches . The distances separating adjacent mutations within the yeast kataegic stretches ( averaging about 1 . 2 kb in the AID* dataset ) might in part reflect that the deaminase both jumps and slides on ssDNA , acting with possibly low efficiency at each encountered cytidine as proposed by Goodman ( Chelico et al . , 2006 ) . In the absence of an induced double-strand break , kataegis in AID/APOBEC-expressing yeast was greatly dependent on UNG , likely reflecting that kataegis under these circumstances was dependent on breaks generated through the processing of abasic sites . That some residual kataegis is still observed in the absence of UNG might well reflect that breaks will occasionally occur spontaneously through other means . The finding that a double-strand break can be the nucleating lesion for kataegis in this yeast experimental system is consistent with the close association of kataegis and rearrangements in breast cancer ( Nik-Zainal et al . , 2012 ) . Whereas the yeast data demonstrate that double-strand breaks can nucleate kataegis , it is probable that APOBEC-catalysed kataegic deamination in exposed stretches of single-stranded DNA in the cancer cells might itself lead to DNA breaks . It has long been known that recombinational repair of double-strand breaks in yeast is associated with an increased frequency of local mutations with implication of error-prone polymerases ( Strathern et al . , 1995 ) . In our experiments , the signatures of the mutations associated with the I-SceI break ( see Figure 2D legend ) implicate APOBEC3 activity rather than error-prone polymerases as the source of mutations during the double-strand break repair . More recently Gordenin and colleagues have shown that extensive clusters of mutations can be induced in yeast by alkylating agents acting on single-stranded DNA ( Roberts et al . , 2012 ) . Thus , the AID/APOBEC-mediated kataegic hypermutation , driven by these endogenous mutagens , can be viewed as a specialised , albeit dramatic , example of localised hypermutation caused by exposure of single-stranded DNA during homologous recombination , along the lines proposed by Roberts ( Roberts et al . , 2012 ) . It is striking that transversions in yeast are specifically associated with kataegic stretches whereas the unclustered mutations in the same cells are restricted to transitions . The reason for this is a matter for speculation but we suspect the singlet uracils largely encounter UNG as part of the base-excision repair process ( which would be non-mutagenic ) ; the C→T transition mutations would be the result of direct replication over the non-excised uracil . In contrast , the action of UNG on uracil in a stretch of exposed single-stranded DNA may yield an abasic site that is replicated over by a translesion polymerase rather than repaired . The yeast experiments indicate that kataegis can be triggered by DNA breaks , whether generated through the joint action of the deaminase and UNG or by other processes . The same likely holds true for the breast cancer kataegis . However , there is no reason why kataegis should be restricted to such initiating triggers . One can well imagine that other processes that cause significant exposure of single-stranded DNA ( e . g . , DNA spooling caused by replication fork stalling [Lopes et al . , 2006]; R-loop structures generated during transcription of suitable target sequences [Aguilera and Gómez-González , 2008] ) could predispose to kataegis . Such mechanisms , or spontaneously-arising DNA breaks , could underlie the presence of kataegis in UNG-deficient cells ( this work and Lada et al . , 2012 ) . A more extensive study of the genetic dependence of kataegis and of the localisation of the kataegic stretches in yeast may give insight into such possibilities . Comparison of the yeast and breast cancer data reveals that the kataegic stretches in both sets extend over a similar range of lengths but with the cancer kataegis displaying a twofold to fivefold higher average mutation density . This could reflect differences in deaminase activity in the two organisms . It also appears that those cancers which harbour kataegic stretches comprising larger numbers of mutations additionally contain multiple clusters with smaller numbers of T-C mutations ( Figure 3—figure supplements 2–4 ) . The marked bias towards a 5′-T seen amongst some cancer singlet C mutations suggests that kataegis might be signalling a much wider implication of APOBEC-mediated deamination in genome-wide mutagenesis in some tumours . The mutation data obtained in yeast reveal APOBEC3B and APOBEC3A as the only deaminases characterised whose target specificity matches the breast cancer kataegic mutations , arguing very strongly for an involvement of these deaminases in cancer kataegis . The implication of APOBEC3A fits with data from others revealing that enforced expression of APOBEC3A ( as well as APOBEC3C and 3H ) can lead to mutation of human papilloma viral DNA ( Vartanian et al . , 2008 ) as well as of transfected plasmid DNA ( Stenglein et al . , 2010 ) . Enforced expression of APOBEC3A has also been shown to lead to genomic damage in the nucleus ( Landry et al . , 2011 ) . The target-specificity data implicating APOBEC3B in the breast cancer mutation is not only supported by our demonstration that its enforced expression can yield DNA damage but also by the fact that it is well expressed in breast cancer cell lines . Furthermore , after submission of this manuscript , Burns et al . , have demonstrated that APOBEC3B expression also correlates with a T-C mutator phenotype in many primary breast cancer tumours ( Burns et al . , 2013 ) . Thus , APOBEC3B and/or APOBEC3A are the deaminases likely responsible for the breast cancer hypermutation although it remains possible that other APOBEC3s might contribute to genome mutation in other tumours . With regard specifically to kataegis , given that double strand breaks are a common feature of tumour development , it will obviously be interesting to discover whether whole genome sequencing of other tumour types also reveals evidence of kataegic hypermutation and whether , in light of the fact that the AID/APOBEC family has undergone considerable expansion in primates , such kataegic hypermutation might also have contributed more generally to recent genome evolution .
Yeast strain BY4741 ( MATa; his3Δ1; leu2Δ0; met15Δ0; ura3Δ0 ) and the ung1Δ::kanMX4 derivative were from Euroscarf ( Frankfurt , Germany ) . The rev1Δ::LEU2 derivative was generated by homologous recombination to remove the open reading frame of REV1 using a LEU2 cassette generated by PCR using 157-bp 5′ homology arm and 200-bp 3′ homology arm . The CAN1::KanMX-ISceIRS strain was generated by inserting a 1 . 4-kb module containing the I-SceI-recognition site and the KanMX selection cassette ( Wach et al . , 1994 ) immediately after its poly-A site by homologous recombination . Correct integration of the targeting constructs was confirmed by PCR . Yeast transformants expressing galactose-inducible human AID/APOBEC proteins were generated by transformation with the appropriate pRS426-derived expression vectors ( Christianson et al . , 1992 ) in which C-terminally-FLAG-tagged AID/APOBEC cDNAs flanked by a GAL1 promoter and tADH polyA site had been inserted between the HindIII and XhoI sites . The cDNAs encoded the full-length human wild type polypeptides except that AID* and A3G* correspond to upmutants AID-7 . 3 and A3G-T283I in Wang et al . , 2009 , with a FLAG-tagged A3G* comprising just the second deaminase domain used in the I-SceI experiments . For these experiments , the I-SceI-ORF with an N-terminal HA tag and 3xNLS ( Johnson et al . , 1999 ) was cloned between the EcoRI and XhoI sites in pSH62 ( Gueldener et al . , 2002 ) . For canavanine resistance assays , single yeast colonies ( at least 12 independent colonies for each experiment ) that had been grown overnight in glucose medium to repress expression from the GAL1 promoter were diluted 1:100 into galactose-containing medium and grown for 2 days at 30°C before serial dilutions were plated onto canavanine-selection or viability plates . Colonies were counted after 3 days growth . For I-SceI-break induction , individual colonies were grown overnight in glucose medium before dilution 1:10 into raffinose-containing medium . After 4 hr growth , galactose was added to 2% and cells were cultured for a further 2 days before serial dilution and plating as above . APOBEC3G* was used in the I-SceI experiments as it gave a good mutation load but a lower proportion of kataegic mutations than AID* ( Supplementary file 1B ) . Induction of protein expression both with and without the raffinose step gave similar mutation rates . For genome sequence determination , individual CanR colonies selected as above were subcloned by streaking out on selective plates , grown for 3 days in canavanine selection media ( 10 ml ) and DNA prepared using Gentra Puregene Yeast/Bact . Kit ( Qiagen Ltd , Manchester , UK ) following manufacturers instructions . Short insert 500-bp library construction , flowcell preparation and cluster generation was in accordance with the Illumina no-PCR library protocol ( Kozarewa et al . , 2009 ) . 100-bp paired-end sequencing was performed on Illumina Hiseq 2000 analysers as described in the Illumina Genome Analyzer operating manual . Short insert 2 × 100 bp paired-end reads were aligned to the reference yeast genome ( SacCer_Apr2011/sacCer3 ) using BWA ( Li and Durbin , 2009 ) . An average of approximately 25-fold sequence coverage was achieved for each yeast genome . A bespoke substitution-calling algorithm , CaVEMan ( manuscript in preparation ) was used for calling somatic substitutions where these were identified as alleles present in an AID/APOBEC-transformant genome but absent in the parental BY4741 genome . All high-confidence mutations included in this analysis were present in more than 0 . 5 variant allele fractions but were more frequently present in all reads reporting that genomic position . Post-processing filters were developed to improve the specificity of substitution calling . These filters removed false positive variants that were generated by genomic features resulting in mapping errors or systematic sequencing artefacts . All substitutions were visually assessed using a genome browser in order to ensure a high specificity of mutation-calling . K-cluster analysis ( Hartigan and Wong , 1979 ) was used to divide intermutational distances ( IMDs ) into two groups , which we designated distal and proximal . An IMD which excluded 99% of the distal group was then used as a threshold for cluster calling . For all the yeast analysis , the IMD threshold was set using the combined dataset of mutations from the wild type transformants . For the breast cancer analysis , IMDs combined from all tumours were used for threshold setting ( except PD4120 because of its much higher mutation load ) . A cluster was called when a minimum of 5 adjacent mutations were identified each with IMDs below the threshold . This 5 mutation threshold was chosen since such clusters are likely to arise with a probability of <0 . 001 by randomly scattered mutations . Sequence contexts were compared in pairwise fashion with the Tomtom motif comparison tool using Sandelin-Wasserman similarity ( MEME Suite; http://tools . genouest . org/tools/meme/cgi-bin/tomtom . cgi ) and are displayed as p-values . APOBEC expressing vectors were generated by cloning the appropriate C-terminally FLAG-tagged cDNAs into a self-inactivating retroviral plasmid . The self-inactivating retroviral plasmid was generated by cloning a pTRE- ( pTRE-TIGHT; Clontech , Saint-Germain-en-Laye , France ) -IRES-GFP ( pMX-IG ) cassette into the BglII- and 3′LTR XbaI site of pMSCVpuro . The tetracycline transactivator ( TET-ON; Clontech ) was cloned into a modified pMSCVpuro ( Clontech ) which contained an IRES-mCherry cassette at the BglII–ClaI site , to generate pTET-ON-ImC . A derivative of the KBM7 human myelocytic leukemia line that stably expressed TET-ON protein was established by retroviral infection with virus particles produced from 293 cells that had been co-transformed with pTET-ON-ImC and packaging vectors using GeneJuice ( Merck KGaA , Darmstadt , Germany ) according to manufacturers instructions . This KBM7[pTET-ON-ImC] cell-line was then superinfected with pMSCV/APOBEC retrovirus to yield derivatives expressing the AID/APOBEC proteins under doxycyclin-inducible control . Expression of the FLAG-tagged AID/APOBEC proteins in the KBM7 transfectants was monitored by Western blot analysis of whole cell lysates after 24 hr of doxycyclin induction using HRP-conjugated anti-FLAG antibody M2 ( A8592; Sigma , Gillingham , UK ) , probing with anti-lamin antibody ( ab16048; Abcam , Cambridge , UK ) as a loading control . Stable derivatives of KBM7 cells harbouring regulatable APOBEC proteins were induced for 72 hr with doxycyclin ( inducer ) and viability measured by flow cytometry by DAPI exclusion . γH2AX and 53BP1 induction and localisation was analysed by flow cytometry and confocal immunofluorescence after 24hr induction with doxycyclin; caspase inhibitor ( 20 μM Z-VAD-FMK; Promega , Southampton , UK ) was included in the cultures for γH2AX expression analysis to maintain cell viability . For γH2AX staining , ethanol-fixed cells were stained sequentially for 1 hr with anti-γH2AX ( 05-636; Millipore , Watford , UK ) and Alexa568-conjugated anti-mouse IgG ( A-11004; Invitrogen Life Technologies Ltd , Paisley , UK ) prior to resuspension in PBS containing 5 μg/ml DAPI and flow cytometry . For 53BP1 staining , cells were allowed to adhere to poly-L-lysine-coated cover slips and stained using anti-53BP1 ( NB100-304; Novus Biologicals , Cambridge , UK ) and Alexa 568-conjugated anti-rabbit IgG ( A-11011; Invitrogen ) prior to mounting with DAPI . 20-30 fields per sample were imaged with a Bio-Rad Radiance 2100 confocal microscope using a 63x oil immersion objective . Images were processed using ImageJ ( default settings ) , and cells were scored as exhibiting either diffuse or punctate staining with punctate cells further scored for the number of foci . Breast cancer cell lines were kindly provided by Dr Kerstin Meyer ( Cancer Research Institute , Cambridge , United Kingdom ) and RNA extracted using RNeasy Plus Mini Kit ( Qiagen ) . cDNA was prepared using GoScript Reverse Transcription System ( Promega ) prior to APOBEC expression quantification by qPCR using QuantiFast SYBR Green PCR Kit using an ABI ViiA-7 system ( Applied Biosystems , Paisley , UK ) . The primers ( which were selected for specificity and equivalent amplification on APOBEC ORF templates ) are given in Supplementary file 1C . | The genomes of cancer cells contain mutations that are not present in normal cells . Some of these prevent cells from repairing their DNA , while others give rise to tumours by causing cells to multiply uncontrollably . Moreover , some of the mutations in breast cancer cells occur in clusters—a phenomenon known as kataegis ( from the Greek for ‘thunderstorm’ ) . Kataegic mutations occur almost exclusively at a cytosine preceded by a thymine . This suggests that a family of proteins called AID/APOBEC enzymes—which remove amine groups from cytosines—may be involved in generating these mutations . In this study , Taylor et al . confirm this possibility by showing that expressing individual members of the AID/APOBEC family of enzymes in yeast cells increases the mutation frequency and induces kataegis . The kataegis triggered by the AID/APOBEC enzymes could be localised through the introduction of double-stranded breaks into the DNA: Taylor et al . suggest that this might happen because repairing the breaks exposes single-stranded DNA , which the AID/APOBEC enzymes then act upon . By comparing the mutations induced in the yeast cells with those observed in breast cancer cells , Taylor et al . identified APOBEC3B as the enzyme most likely to be responsible for kataegis in breast cancer ( with APOBEC3A also a strong candidate in some cancers ) . Moreover , they showed that APOBEC3B was highly expressed in breast cancer cell lines , and that APOBEC3B and APOBEC3A can also cause DNA damage in human cells . Taken together , the findings provide key insights into the mechanism by which kataegis arises , and identify two proteins likely to contribute to the mutations seen in breast cancer . Further work is now required to determine whether these enzymes also give rise to mutations in other forms of cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2013 | DNA deaminases induce break-associated mutation showers with implication of APOBEC3B and 3A in breast cancer kataegis |
Type II CRISPR immune systems in bacteria use a dual RNA-guided DNA endonuclease , Cas9 , to cleave foreign DNA at specific sites . We show here that Cas9 assembles with hybrid guide RNAs in human cells and can induce the formation of double-strand DNA breaks ( DSBs ) at a site complementary to the guide RNA sequence in genomic DNA . This cleavage activity requires both Cas9 and the complementary binding of the guide RNA . Experiments using extracts from transfected cells show that RNA expression and/or assembly into Cas9 is the limiting factor for Cas9-mediated DNA cleavage . In addition , we find that extension of the RNA sequence at the 3′ end enhances DNA targeting activity in vivo . These results show that RNA-programmed genome editing is a facile strategy for introducing site-specific genetic changes in human cells .
Methods for introducing site-specific double-strand DNA ( dsDNA ) breaks ( DSBs ) in genomic DNA have transformed our ability to engineer eukaryotic organisms by initiating DNA repair pathways that lead to targeted genetic re-programming . Zinc-finger nucleases ( ZFNs ) and transcription activator-like effector nucleases ( TALENs ) have proved effective for such genomic manipulation but their use has been limited by the need to engineer a specific protein for each dsDNA target site and by off-target activity ( Urnov et al . , 2010; Bogdanove and Voytas , 2011 ) . Thus , alternative strategies for triggering site-specific DNA cleavage in eukaryotic cells are of great interest . Research into genome defense mechanisms in bacteria showed that CRISPR ( Clustered Regularly Interspaced Short Palindromic Repeats ) /Cas ( CRISPR-associated ) loci encode RNA-guided adaptive immune systems that can destroy foreign DNA ( Bhaya et al . , 2011; Terns and Terns , 2011; Wiedenheft et al . , 2012 ) . The Type II CRISPR/Cas systems require a single protein , Cas9 , to catalyze DNA cleavage ( Sapranauskas et al . , 2011 ) . Cas9 generates blunt DSBs at sites defined by a 20-nucleotide guide sequence contained within an associated CRISPR RNA ( crRNA ) transcript ( Gasiunas et al . , 2012; Jinek et al . , 2012 ) . Cas9 requires both the guide crRNA and a trans-activating crRNA ( tracrRNA ) that is partially complementary to the crRNA for site-specific DNA recognition and cleavage ( Deltcheva et al . , 2011; Jinek et al . , 2012 ) . Recent experiments showed that the crRNA:tracrRNA complex can be redesigned as a single transcript ( single-guide RNA or sgRNA ) encompassing the features required for both Cas9 binding and DNA target siterecognition ( Jinek et al . , 2012 ) . Using sgRNA , Cas9 can be programmed to cleave double-stranded DNA at any site defined by the guide RNA sequence and including a GG protospacer-adjacent ( PAM ) motif ( Sapranauskas et al . , 2011; Jinek et al . , 2012 ) . These findings suggested the exciting possibility that Cas9:sgRNA complexes might constitute a simple and versatile RNA-directed system for generating DSBs that could facilitate site-specific genome editing . However , it was not known whether such a bacterial system would function in eukaryotic cells . We show here that Cas9 can be expressed and localized to the nucleus of human cells , and that it assembles with sgRNA in vivo . These complexes can generate double stranded breaks and stimulate non-homologous end joining ( NHEJ ) repair in genomic DNA at a site complementary to the sgRNA sequence , an activity that requires both Cas9 and the sgRNA . Extension of the RNA sequence at its 3′ end enhances DNA targeting activity in vivo . Further , experiments using extracts from transfected cells show that sgRNA assembly into Cas9 is the limiting factor for Cas9-mediated DNA cleavage . These results demonstrate the feasibility of RNA-programmed genome editing in human cells .
To test whether Cas9 could be programmed to cleave genomic DNA in vivo , we co-expressed Cas9 together with an sgRNA designed to target the human clathrin light chain ( CLTA ) gene . The CLTA genomic locus has previously been targeted and edited using ZFNs ( Doyon et al . , 2011 ) . We first tested the expression of a human-codon-optimized version of the Streptococcus pyogenes Cas9 protein and sgRNA in human HEK293T cells . The 160 kDa Cas9 protein was expressed as a fusion protein bearing an HA epitope , a nuclear localization signal ( NLS ) , and green fluorescent protein ( GFP ) attached to the C-terminus of Cas9 ( Figure 1A ) . Analysis of cells transfected with a vector encoding the GFP-fused Cas9 revealed abundant Cas9 expression and nuclear localization ( Figure 1B ) . Western blotting confirmed that the Cas9 protein is expressed largely intact in extracts from these cells ( Figure 1A ) . To program Cas9 , we expressed sgRNA bearing a 5′-terminal 20-nucleotide sequence complementary to the target DNA sequence , and a 42-nucleotide 3′-terminal stem loop structure required for Cas9 binding ( Figure 1C ) . This 3′-terminal sequence corresponds to the minimal stem-loop structure that has previously been used to program Cas9 in vitro ( Jinek et al . , 2012 ) . The expression of this sgRNA was driven by the human U6 ( RNA polymerase III ) promoter ( Medina and Joshi , 1999 ) . Northern blotting analysis of RNA extracted from cells transfected with the U6 promoter-driven sgRNA plasmid expression vector showed that the sgRNA is indeed expressed , and that their stability is enhanced by the presence of Cas9 ( Figure 1D ) . 10 . 7554/eLife . 00471 . 003Figure 1 . Co-expression of Cas9 and guide RNA in human cells generates double-strand DNA breaks at the target locus . ( A ) Top: schematic diagram of the Cas9-HA-NLS-GFP expression construct . Bottom: lysate from HEK293T cells transfected with the Cas9 expression plasmid was analyzed by Western blotting using an anti-HA antibody . ( B ) Fluorescence microscopy of HEK293T cells expressing Cas9-HA-NLS-GFP . ( C ) Blue line indicates the sequence used for the guide segment of CLTA1 sgRNA . Top: schematic diagram of the sgRNA target site in exon 7 of the human CLTA gene . The target sequence that hybridizes to the guide segment of CLTA1 sgRNA is indicated by the blue line . The GG nucleotide protospacer adjacent motif ( PAM ) is highlighted in yellow . Black lines denote the DNA binding regions of the control ZFN protein . The translation stop codon of the CLTA open reading frame is highlighted in red for reference . Middle: schematic diagram of the sgRNA expression construct . The RNA is expressed under the control of the U6 Pol III promoter and a poly ( T ) tract that serves as a Pol III transcriptional terminator signal . Bottom: sgRNA-guided cleavage of target DNA by Cas9 . The sgRNA consists of a 20-nt 5′-terminal guide segment ( blue ) followed by a 42-nt stem-loop structure required for Cas9 binding ( red ) . Cas9-mediated cleavage of the two target DNA strands occurs upon unwinding of the target DNA and formation of a duplex between the guide segment of the sgRNA and the target DNA . This is dependent on the presence of a GG dinucleotide PAM downstream of the target sequence in the target DNA . Note that the target sequence is inverted relative to the upper diagram . ( D ) Northern blot analysis of sgRNA expression in HEK239T cells . ( E ) Surveyor nuclease assay of genomic DNA isolated from HEK293T cells expressing Cas9 and/or CLTA sgRNA . A ZFN construct previously used to target the CLTA locus ( Doyon et al . , 2011 ) was used as a positive control for detecting DSB-induced DNA repair by non-homologous end joining . DOI: http://dx . doi . org/10 . 7554/eLife . 00471 . 00310 . 7554/eLife . 00471 . 004Figure 1—figure supplement 1 . Mutated alleles of the CLTA gene in HEK293T cells as a result of Cas9-induced NHEJ . Four types of indels were identified by Sanger sequencing in 68 clonal amplicons . The protospacer sequence is highlighted in grey with the putative cleavage site indicated by the arrow . Dashes represent deleted bases . DOI: http://dx . doi . org/10 . 7554/eLife . 00471 . 004 Next , we investigated whether site-specific DSBs are generated in HEK293T cells transfected with Cas9-HA-NLS-mCherry and the CLTA1 sgRNA . To do this , we probed for minor insertions and deletions in the locus resulting from imperfect repair by DSB-induced NHEJ using the Surveyor nuclease assay ( Qiu et al . , 2004 ) . The region of genomic DNA targeted by Cas9:sgRNA is amplified by PCR and the resulting products are denatured and reannealed . The rehybridized PCR products are incubated with the mismatch recognition endonuclease Cel-1 and resolved on an acrylamide gel to identify Cel-1 cleavage bands . As DNA repair by NHEJ is typically induced by a DSB , a positive signal in the Surveyor assay indicates that genomic DNA cleavage has occurred . Using this assay , we detected cleavage of the CLTA locus at a position targeted by the CLTA1 sgRNA ( Figure 1E ) . A pair of ZFNs that target a neighboring site in the CLTA locus provided a positive control in these experiments ( Doyon et al . , 2011 ) . To determine if either Cas9 or sgRNA expression is a limiting factor in the observed genome editing reactions , lysates prepared from the transfected cells were incubated with plasmid DNA harboring a fragment of the CLTA gene targeted by the CLTA1 sgRNA . Plasmid DNA cleavage was not observed upon incubation with lysate prepared from cells transfected with the Cas9-HA-NLS-GFP expression vector alone , consistent with the Surveyor assay results . However , robust plasmid cleavage was detected when the lysate was supplemented with in vitro transcribed CLTA1 sgRNA ( Figure 2A ) . Furthermore , lysate prepared from cells transfected with both Cas9 and sgRNA expression vectors supported plasmid cleavage , while lysates from cells transfected with the sgRNA-encoding vector alone did not ( Figure 2A ) . These results suggest that a limiting factor for Cas9 function in human cells could be assembly with the sgRNA . We tested this possibility directly by analyzing plasmid cleavage in lysates from cells transfected as before in the presence and absence of added exogenous sgRNA . Notably , when exogenous sgRNA was added to lysate from cells transfected with both the Cas9 and sgRNA expression vectors , a substantial increase in DNA cleavage activity was observed ( Figure 2B ) . This result indicates that the limiting factor for Cas9 function in HEK293T cells is the expression of the sgRNA or its loading into Cas9 . 10 . 7554/eLife . 00471 . 005Figure 2 . Cell lysates contain active Cas9:sgRNA and support site-specific DNA cleavage . ( A ) Lysates from cells transfected with the plasmid ( s ) indicated at left were incubated with plasmid DNA containing a PAM and the target sequence complementary to the CLTA1 sgRNA; where indicated , the reaction was supplemented with 10 pmol of in vitro transcribed CLTA1 sgRNA; secondary cleavage with XhoI generated fragments of ∼2230 and ∼3100 bp fragments indicative of Cas9-mediated cleavage . A control reaction using lysate from cells transfected with a ZFN expression construct shows fragments of slightly different size reflecting the offset of the ZFN target site relative to the CLTA1 target site . ( B ) Lysates from cells transfected with Cas9-HA-NLS-GFP expression plasmid and , where indicated , the CLTA1 sgRNA expression plasmid , were incubated with target plasmid DNA as in ( A ) in the absence or presence of in vitro-transcribed CLTA1 sgRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 00471 . 005 As a means of enhancing the Cas9:sgRNA assembly in vivo , we next tested the effect of extending the presumed Cas9-binding region of the guide RNA . Two new versions of the CLTA1 sgRNA were designed to include an additional 4 or 10 base pairs in the helix that mimics the base-pairing interactions between the crRNA and tracrRNA ( Figure 3A ) . Additionally , the 3′-end of the guide RNA was extended by five nucleotides based on the native sequence of the S . pyogenes tracrRNA ( Deltcheva et al . , 2011 ) . Vectors encoding these 3′ extended sgRNAs under the control of either the U6 or H1 Pol III promoters were transfected into cells along with the Cas9-HA-NLS-GFP expression vector and site-specific genome cleavage was tested using the Surveyor assay ( Figure 3B ) . These results suggest that the 3′-extended sgRNAs support more efficient Cas9 function in vivo , although more quantitative comparisons will be necessary to confirm this conclusion . 10 . 7554/eLife . 00471 . 006Figure 3 . 3′ extension of sgRNA constructs enhances site-specific NHEJ-mediated mutagenesis . ( A ) The construct for CLTA1 sgRNA expression ( top ) was designed to generate transcripts containing the original Cas9-binding sequence v1 . 0 ( Jinek et al . , 2012 ) , or sequences extended by 4 base pairs ( v2 . 1 ) or 10 base pairs ( v2 . 2 ) . ( B ) Surveyor nuclease assay of genomic DNA isolated from HEK293T cells expressing Cas9 and/or CLTA sgRNA v1 . 0 , v2 . 1 or v2 . 2 . A ZFN construct previously used to target the CLTA locus ( Doyon et al . , 2011 ) was used as a positive control for detecting DSB-induced DNA repair by non-homologous end joining . DOI: http://dx . doi . org/10 . 7554/eLife . 00471 . 006
Besides serving as an invaluable research tool , targeted genome engineering in cells and organisms could potentially provide the path to revolutionary applications in human therapies , agricultural biotechnology and microbial engineering . Methods of modifying the genome exploit endogenous DNA repair pathways that are initiated by the introduction of site-specific dsDNA cleavages . The results presented here provide a straightforward system of RNA-guided site-specific dsDNA cleavage using the Cas9 protein from a Type II bacterial CRISPR system to promote genome editing in human cells . Our data show that a codon-optimized version of Cas9 , when programmed by an appropriate sgRNA , successfully assembles into Cas9 targeting complexes to trigger site-specific DNA cleavage and repair by NHEJ . The efficiency of NHEJ-induced mutagenesis at the CLTA locus investigated here is consistently in the range of 6–8% . This frequency is lower than that found for a ZFN pair that recognizes a nearby target sequence , but is within the range of frequencies observed more generally with ZFNs and TALENs ( Bogdanove and Voytas , 2011 ) . Our data suggest that sgRNA expression and/or its assembly into Cas9 , rather than Cas9 expression , localization or folding , presently limits Cas9 function in human cells . Higher efficiencies of Cas9-mediated genome targeting could be achieved by optimization of the sgRNA construct design , its expression levels or its subcellular localization . We note that the sgRNAs used in this study are not thought to be 5′-capped or 3′-polyadenylated , which may have reduced their stability in vivo . This and other 5′ and 3′ end modifications might provide alternative approaches to enhancing Cas9:sgRNA assembly and activity in cells . Nonetheless , the levels of targeting observed in this study have been obtained with a minimal system that relies on simple base pairing to a guide RNA , in contrast to the ZFN and TALEN proteins , which require a new protein to be engineered for each new cleavage site . RNA-guided genome editing would thus offer distinct advantages due to the simplicity of the sgRNA design . Our results thus provide the framework for implementing Cas9 as a facile molecular tool for diverse genome editing applications . Although not tested explicitly in this study , a powerful feature of this system is the potential to program Cas9 with multiple sgRNAs in the same cell , either to increase the efficiency of targeting at a single locus , or as a means of targeting several loci simultaneously . Such strategies would find broad application in genome-wide experiments and large-scale research efforts such as the development of multigenic disease models . As an inexpensive and rapid mechanism for triggering site-specific genome modification , the programmable Cas9:sgRNA system could potentially transform next-generation genome-scale studies .
The sequence encoding Streptococcus pyogenes Cas9 ( residues 1–1368 ) fused to an HA epitope ( amino acid sequence DAYPYDVPDYASL ) , a nuclear localization signal ( amino acid sequence PKKKRKVEDPKKKRKVD ) was codon optimized for human expression and synthesized by GeneArt ( Regensburg , Germany; DNA and protein sequences shown in supplementary file 1 ) . Ligation-independent cloning ( LIC ) was used to insert this sequence into a pcDNA3 . 1-derived GFP and mCherry LIC vectors ( vectors 6D and 6B , respectively , obtained from the UC Berkeley MacroLab ) , resulting in a Cas9-HA-NLS-GFP and Cas9-HA-NLS-mCherry fusions expressed under the control of the CMV promoter . Guide sgRNAs were expressed using expression vector pSilencer 2 . 1-U6 puro ( Life Technologies , Carlsbad , CA ) and pSuper ( Oligoengine , Seattle , WA ) . RNA expression constructs were generated by annealing complementary oligonucleotides to form the RNA-coding DNA sequence and ligating the annealed DNA fragment between the BamHI and HindIII sites in pSilencer 2 . 1-U6 puro and BglII and HindIII sites in pSuper . HEK293T cells were maintained in Dulbecco's modified eagle medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) in a 37°C humidified incubator with 5% CO2 . Cells were transiently transfected with plasmid DNA using either X-tremeGENE DNA Transfection Reagent ( Roche Applied Science , Indianapolis , IN ) or Turbofect Transfection Reagent ( Thermo Scientific , Waltham , MA ) with recommended protocols . Briefly , HEK293T cells were transfected at 60–80% confluency in 6-well plates using 0 . 5 μg of the Cas9 expression plasmid and 2 . 0 μg of the RNA expression plasmid . The transfection efficiencies were estimated to be 30–50% for Turbofect ( Figures 1E and 2A , B ) and 80–90% for X-tremegene ( Figure 3B ) , based on the fraction of GFP-positive cells observed by fluorescence microscopy . 48 hr post transfection , cells were washed with phosphate buffered saline ( PBS ) and lysed by applying 250 μl lysis buffer ( 20 mM Hepes pH 7 . 5 , 100 mM potassium chloride [KCl] , 5 mM magnesium chloride [MgCl2] , 1 mM dithiothreitol [DTT] , 5% glycerol , 0 . 1% Triton X-100 , supplemented with Roche Protease Inhibitor cocktail ) and then rocked for 10 min at 4°C . The resulting cell lysate was divided into aliquots for further analysis . Genomic DNA was isolated from 200 μl cell lysate using the DNeasy Blood and Tissue Kit ( Qiagen , Hilden , Germany ) according to the manufacturer's protocol . HEK293T , transfected with the Cas9-HA-NLS-GFP expression plasmid , were harvested and lysed 48 hr post transfection as above . 5 μl of lysate were electrophoresed on a 10% SDS polyacrylamide gel , blotter onto a PVDF membrane and probed with HRP-conjugated anti-HA antibody ( 1:1000 dilution in 1× PBS; Sigma , St . Louis , MO ) . The Surveyor assay was performed as previously described ( Qiu et al . , 2004; Miller et al . , 2007; Doyon et al . , 2011 ) . Briefly , the human clathrin light chain A ( CLTA ) locus was PCR amplified from 200 ng of genomic DNA using a high fidelity polymerase , Herculase II Fusion DNA Polymerase ( Agilent Technologies , Santa Clara , CA ) and forward primer 5′-GCAGCAGAAGAAGCCTTTGT-3′ and reverse primer 5′-TTCCTCCTCTCCCTCCTCTC-3′ . 300 ng of the 360 bp amplicon was then denatured by heating to 95°C and slowly reannealed using a heat block to randomly rehybridize wild type and mutant DNA strands . Samples were then incubated with Cel-1 nuclease ( Surveyor Kit; Transgenomic , Omaha , NE ) for 1 hr at 42°C . Cel-1 recognizes and cleaves DNA helices containing mismatches ( wild type:mutant hybridization ) . Cel-1 nuclease digestion products were separated on a 10% acrylamide gel and visualized by staining with SYBR Safe ( Life Technologies ) . Quantification of cleavage bands was performed using ImageLab software ( Bio-Rad , Hercules , CA ) . The percent cleavage was determined by dividing the average intensity of cleavage products ( 160–200 bps ) by the sum of the intensities of the uncleaved PCR product ( 360 bp ) and the cleavage product . Mutations ( indels ) resulting from Cas9-induced DNA repair reactions were identified by Sanger sequencing of cloned amplicons . Briefly , the 360-bp PCR amplicon was treated with Taq DNA polymerase ( NEB ) and dATP to attach an ‘A’ base to the 3′ end of the DNA fragment . The DNA was then ligated into pGEM-T Easy vector ( Promega , Madison , WI ) according to the manufacturer's protocol . Transformants were grown on Luria Broth agar plates containing 100 μg/ml of ampicillin , 70 μg/ml 5-bromo-4-chloro-indolyl-β-D-galactopyranoside and 100 μM isopropyl β-D-1-thiogalactopyranoside . Random white colonies were selected for Sanger sequencing ( Quintara Biosciences , Albany , CA ) . Representative DNA sequences are shown in Figure 1—figure supplement 1 . Guide RNA was in vitro transcribed using recombinant T7 RNA polymerase and a DNA template generated by annealing complementary synthetic oligonucleotides as previously described ( Sternberg et al . , 2012 ) . RNAs were purified by electrophoresis on 7 M urea denaturing acrylamide gel , ethanol precipitated , and dissolved in DEPC-treated water . RNA was purified from HEK293T cells using the mirVana small-RNA isolation kit ( Ambion ) . For each sample , 800 ng of RNA were separated on a 10% urea-PAGE gel after denaturation for 10 min at 70°C in RNA loading buffer ( 0 . 5× TBE [pH 7 . 5] , 0 . 5 mg/ml bromophenol blue , 0 . 5 mg xylene cyanol and 47% formamide ) . After electrophoresis at 10 W in 0 . 5× TBE buffer until the bromophenol blue dye reached the bottom of the gel , samples were electroblotted onto a Nytran membrane at 20 V for 1 . 5 hr in 0 . 5× TBE . The transferred RNAs were cross-linked onto the Nytran membrane in UV-Crosslinker ( Strategene ) and were pre-hybridized at 45°C for 3 hr in a buffer containing 40% formamide , 5× SSC , 3× Dernhardt's solution ( 0 . 1% each of ficoll , polyvinylpyrollidone , and BSA ) and 200 µg/ml Salmon sperm DNA . The pre-hybridized membranes were incubated overnight in the prehybridization buffer supplemented with 5′-32P-labeled antisense DNA oligo probe at 1 million cpm/ml . After several washes in SSC buffer ( final wash in 0 . 2× SCC ) , the membranes were imaged phosphorimaging . Cell lysates were prepared as described above and incubated with CLTA-RFP donor plasmid ( Doyon et al . , 2011 ) . Cleavage reactions were carried out in a total volume of 20 μl and contained 10 μl lysate , 2 μl of 5× cleavage buffer ( 100 mM HEPES pH 7 . 5 , 500 mM KCl , 25 mM MgCl2 , 5 mM DTT , 25% glycerol ) and 300 ng plasmid . Where indicated , reactions were supplemented with 10 pmol of in vitro transcribed CLTA1 sgRNA . Reactions were incubated at 37°C for 1 hr and subsequently digested with 10 U of XhoI ( NEB ) for an additional 30 min at 37°C . The reactions were stopped by the addition of Proteinase K ( Thermo Scientific ) and incubated at 37°C for 15 min . Cleavage products were analyzed by electrophoresis on a 1% agarose gel and stained with SYBR Safe . The presence of ∼2230 and ∼3100 bp fragments is indicative of Cas9-mediated cleavage . | The ability to make specific changes to DNA—such as changing , inserting or deleting sequences that encode proteins—allows researchers to engineer cells , tissues and organisms for therapeutic and practical applications . Until now , such genome engineering has required the design and production of proteins with the ability to recognize a specific DNA sequence . The bacterial protein , Cas9 , has the potential to enable a simpler approach to genome engineering because it is a DNA-cleaving enzyme that can be programmed with short RNA molecules to recognize specific DNA sequences , thus dispensing with the need to engineer a new protein for each new DNA target sequence . Now Jinek et al . demonstrate the capability of RNA-programmed Cas9 to introduce targeted double-strand breaks into human chromosomal DNA , thereby inducing site-specific genome editing reactions . Cas9 assembles with engineered single-guide RNAs in human cells and the resulting Cas9-RNA complex can induce the formation of double-strand breaks in genomic DNA at a site complementary to the guide RNA sequence . Experiments using extracts from transfected cells show that RNA expression and/or assembly into Cas9 is the limiting factor for the DNA cleavage , and that extension of the RNA sequence at the 3′ end enhances DNA targeting activity in vivo . These results show that RNA-programmed genome editing is a straightforward strategy for introducing site-specific genetic changes in human cells , and the ease with which it can programmed means that it is likely to become competitive with existing approaches based on zinc finger nucleases and transcription activator-like effector nucleases , and could lead to a new generation of experiments in the field of genome engineering for humans and other species with complex genomes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2013 | RNA-programmed genome editing in human cells |
In epithelial collective migration , leader and follower cells migrate while maintaining cell–cell adhesion and tissue polarity . We have identified a conserved protein and interactors required for maintaining cell adhesion during a simple collective migration in the developing C . elegans male gonad . LINKIN is a previously uncharacterized , transmembrane protein conserved throughout Metazoa . We identified seven atypical FG–GAP domains in the extracellular domain , which potentially folds into a β-propeller structure resembling the α-integrin ligand-binding domain . C . elegans LNKN-1 localizes to the plasma membrane of all gonadal cells , with apical and lateral bias . We identified the LINKIN interactors RUVBL1 , RUVBL2 , and α-tubulin by using SILAC mass spectrometry on human HEK 293T cells and testing candidates for lnkn-1-like function in C . elegans male gonad . We propose that LINKIN promotes adhesion between neighboring cells through its extracellular domain and regulates microtubule dynamics through RUVBL proteins at its intracellular domain .
In epithelial collective migration , interconnected cells migrate together in various configurations , such as sheets , branches , chains , and clusters , to produce organs of diverse shapes and possess both epithelial and mesenchymal characteristics ( Montell , 2001; Haas and Gilmour , 2006; Ewald et al . , 2008; Zelenka and Arpitha , 2008 ) . The cells develop apico-basal polarity and cell–cell adhesion as an epithelial tissue , but cells at leading edge of the group are also capable of migration . Many of the components involved in individual cell migration also affect collective cell migration ( Rorth , 2011 ) , such as response to external guidance cues ( Klämbt et al . , 1992; Haas and Gilmour , 2006; Bianco et al . , 2007; Pozzi and Zent , 2011 ) and establishment of front–back polarity ( Prasad and Montell , 2007; Janssens et al . , 2010; Ng et al . , 2012; Law et al . , 2013; Lebreton and Casanova , 2014 ) . Collective cell migration , however , additionally depends on the ability of cells to coordinate and follow the leader cells . Cell–cell adhesion molecules such as cadherins ( Cai et al . , 2014; Menko et al . , 2014 ) and tissue organization through the planar cell polarity pathway ( Muñoz-Soriano et al . , 2012 ) impact the collective migratory ability by coordinating cytoskeleton movement . Effective collective migration therefore requires not only components promoting motility but also those that contribute to tissue integrity and coordination . The Caenorhabditis elegans male gonad is shaped by a collective cell migration during larval development . It has a simple organization of one migratory leader cell , the linker cell ( LC ) , that is followed by a stalk of adherent , passive follower cells that can be visualized in live animals ( Kimble and Hirsh , 1979; Kato and Sternberg , 2009 ) . After the migration leads the elongating gonad from its origin at the mid-body to the cloaca opening near the posterior end of the body , the gonad completes its differentiation into the mature structure . The migratory linker cell ( LC ) is a hybrid of mesenchymal and epithelial-like characteristics , while the follower somatic cells are epithelial-like . The cellular organization of the migrating male gonad is similar to the migrating branches in lung , trachea , and vascular development , in which interconnected cells organize into stalks behind leader tip cells ( Affolter et al . , 2009; Eilken and Adams , 2010 ) . As with other branching structures ( Ikeya and Hayashi , 1999; Llimargas , 1999 ) , Notch signaling is required to specify roles between leader and follower cells in the C . elegans gonad ( Kimble and Hirsh , 1979; Greenwald et al . , 1983 ) . However , unlike other systems , the role of the leader and follower is simplified , as they are not interchangeable once established ( Kimble , 1981 ) . Investigation into genes required for the migration of C . elegans gonadal leader cells has revealed similarities to other cell migrations , including their responding to netrin and Wnt guidance cues ( Hedgecock et al . , 1990; Merz et al . , 2001; Cabello et al . , 2010 ) , binding to the extracellular matrix ( ECM ) through integrin receptors , and remodeling of surrounding ECM using metalloproteases ( Blelloch and Kimble , 1999; Nishiwaki et al . , 2004 ) . However , little is known about the interaction between cells to promote effective collective migration . We have identified a new protein , LINKIN , required for maintaining tissue integrity through cell adhesion and apical polarization . LINKIN is a previously uncharacterized transmembrane protein conserved among metazoans . We identified seven atypical FG–GAP domains in LINKIN that may fold into a β-propeller domain resembling the α-integrin ligand-binding domain . We show that the C . elegans LINKIN protein , LNKN-1 , is localized to membranes of interconnected cells , most pronouncedly at apical surfaces and cell–cell contacts . In particular , LNKN-1 is required for adhesion among collectively migrating gonadal cells in C . elegans , although it is also expressed in many other interconnected tissues . Taking advantage of the conservation between C . elegans and human LINKIN , we performed SILAC based mass spectrometry on a human cell line and functional testing in C . elegans to identify potential interactors of LINKIN . Members of the highly conserved AAA+ ATPase family , RUVBL1 and RUVBL2 , and the cytoskeletal protein α-tubulin physically interacted with LINKIN and were required for collective gonadal migration . Our data support a function for LINKIN as an adhesion molecule that uses its extracellular domain to bind molecules on the surface of neighboring cells and its intracellular domain to regulate microtubule dynamics .
The developing male gonad is a collective cell migration consisting of a chain of passively migrating somatic and germ cells led by a migratory somatic cell , the linker cell ( LC ) ( Figure 1A–C ) . After migration , the interconnected somatic cells behind the LC differentiate during the transition from the fourth larval ( L4 ) stage to the adult into a mature gonad structure , a tube comprising the vas deferens and seminal vesicle . Behind the somatic gonad are the proliferating germ cells , arranged from the newest in the distal region to the most developed closest to the somatic gonad . Capping the distal end of the gonad are the two male distal tip cells , which maintain the mitotic germ cells . To form this gonad shape during the L2 through L4 stages of the larval development , the LC leads the elongating gonad from the mid-body region to the cloaca opening in the posterior body , where it dies after completing the migration . During the L3 stage of the migration , the somatic cells of the vas deferens and seminal vesicle precursors divide from seven to 53 cells to form the elongating gonad . The developing somatic gonad has epithelial-like characteristics consisting of strong intercellular connections and a developing apical domain running down the core of the gonad ( Figure 1A ) . The somatic cells have a radial symmetry around this core , and as they proliferate , the daughter cells are incorporated into the chain while maintaining this configuration . 10 . 7554/eLife . 04449 . 003Figure 1 . The collective migration of the male gonad in wild-type animals and its disruption in lnkn-1 mutants . ( A ) Intact wild-type male gonad shape is generated by the collective migration of somatic and germ cells . The migration of the leader cell , the linker cell ( LC; green ) , pulls the interconnected follower cells . The somatic gonad consists of the LC , the vas deferens precursors ( yellow ) , the seminal vesicle precursors ( blue ) , and the distal tip cells ( orange ) . The germ cells ( purple ) follow behind most of the somatic gonad . At the beginning of its migration in the early L2 stage , the gonad is a small cluster of cells in the ventral mid-body of the animal ( top left panel ) . As the LC migration defines the shape of the mature gonad , the gonad expands through the proliferation of the interconnected follower cells ( bottom left panel ) . Longitudinal and transverse sections of the vas deferens precursor cells reveal the apical domain ( red ) running through the somatic gonad core ( right panels ) . ( B–F ) Nomarski micrographs of gonads superimposed with fluorescence images of YFP-tagged LC ( green fluorescence , black arrow ) . Gonad is outlined in the same color scheme as ( A ) . ( B , C ) Wld-type L3 and L4 stage gonads . ( D ) The connection between the LC and the gonad was severed by ablating cells immediately behind the LC and examined 6 hr later . The LC alone has continued to migrate along its normal path , while the gonad no longer elongates after being severed from the LC . ( E ) In the L3 stage lnkn-1 ( gk367 ) mutant , the gonad starts to show thinning of follower cells ( yellow arrows ) behind the LC ( black arrow ) . ( F ) By the mid-L4 stage , gonad ( yellow arrows ) has stopped migrating at the point where the LC detached , while the LC ( black arrow ) has continued to migrate . In this and subsequent figures , anterior ( A ) is to the left , posterior ( P ) is to the right , dorsal ( D ) is to the top , and ventral ( V ) is to the bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 003 We examined the requirement of different gonadal cell types for migration . By ablating the LC ( 15/15 ) , we confirmed the findings of Kimble ( 1981 ) that the LC is necessary for migration and is not replaced by a follower cell taking on its migratory role . Without the LC , the gonad stopped elongating but continued to balloon through cell proliferation . The LC , however , was capable of migrating alone if the somatic cells around it were ablated in the L3 stage ( 8/10 ) . In this case , the LC alone migrated along its normal course while the gonad stopped elongating at the point of LC detachment ( Figure 1D ) . Since LC migration was slower than normal , the LC often did not complete its migration by the L4-to-adult transition . The cause of the slower migration may be due to the missing contribution of gonadal cells or other factors such as drag from scarred tissue . The germ cells add to gonadal mass but are not necessary for migration , as the LC reaches the cloaca even when germ cell precursors are ablated . We discovered the lnkn-1 mutant during a process of identifying new genes involved in LC migration by utilizing a database of expression patterns reported by the Genome BC C . elegans Gene Expression Consortium ( Hunt Newberry et al . , 2007 ) . Since the consortium only reports hermaphrodite expression patterns , we searched their database for genes expressed in the migratory leader cells for the hermaphrodite gonad , the distal tip cells ( DTCs ) , which functionally are the closest cells to the male LC ( Kimble and Hirsh , 1979 ) . We reasoned that the gonadal leader cells of both sexes may use partially overlapping genes for their migrations . One of the genes reported to be expressed in the hermaphrodite DTCs was ZK637 . 3 ( WBGene00014023 ) , a conserved gene of unknown function that had an available deletion mutant , gk367 . After obtaining this mutant and observing unusual male gonad defects , we decided to investigate this gene further . We have renamed this gene from tag-256 ( temporarily assigned gene-256 , ZK637 . 3 ) to lnkn-1 ( LiNKiNg-1 ) . In males homozygous for lnkn-1 ( gk367 ) , gonadal cells near the LC became detached during gonad migration ( n = 29/30 ) such that the LC continued to migrate , either alone or with a few remaining follower cells , but the rest of the gonad did not follow . This detachment resulted in a partially elongated gonad and , further ahead along the normal path , a detached LC alone or with a few adherent follower cells ( Figure 1E , F ) . This phenotype is similar to that of the gonad with ablated follower cells behind the LC ( Figure 1D ) . The position of detachment was variable but usually occurred within a few cell lengths behind the LC , suggesting that the pulling force generated by the LC may have caused detachment . Although the LC continued to migrate along its normal course after detachment and occasionally completed the migration , the male was sterile since the gonad did not connect to the cloaca opening . In lnkn-1 ( gk367 ) mutant hermaphrodites , the gonadal leader DTCs remained connected but migrated a shorter distance than the wild type and their shape appeared elongated and strained . The mutant also recessively caused maternal effect lethality ( n = 30/30 ) . The male gonad phenotype in lnkn-1 ( gk367 ) mutants suggested that lnkn-1 is required for cell–cell interaction rather than LC migration . LNKN-1 was a conserved , poorly characterized protein predicted to be a type I single-pass transmembrane protein of 599 amino acids ( AA ) , consisting of a 19 AA signal sequence , 533 AA extracellular domain , 23 AA transmembrane domain , and 24 AA intracellular domain ( Figure 2 ) . We were able to identify homologs of LNKN-1 back to an early branching animal phylum , Placozoa , as well as in fungi , and have called this protein family LINKIN . The presence of LINKIN in Plasmodium falciparum has previously been noted ( Kaczanowski and Zielenkiewicz , 2003 ) . LINKIN is conserved in Metazoa from Placozoa Trichoplax adhaerens , a basal metazoan , to vertebrates including human . A protein alignment of LINKIN from Homo sapiens ( ITFG1/TIP , 612 AA ) , Mus musculus ( ITFG1/TIP , 610 AA ) , Drosophila melanogaster ( CG7739 , 596 AA ) , and C . elegans ( LNKN-1 , 599 AA ) revealed that all orthologs have similar protein lengths and domain organizations ( Figure 2B ) . 10 . 7554/eLife . 04449 . 004Figure 2 . LINKIN protein domains and sequence are conserved across diverse metazoan species . ( A ) C . elegans LNKN-1 is a single-pass transmembrane protein of 599 amino acids ( AA ) . Conserved protein motifs include seven atypical FG–GAP domains ( orange boxes ) and an extracellular region proximal to the transmembrane domain ( light blue box ) . The gk367 genomic lesion results in the deletion of 92 AAs based on cDNA sequencing . ( B ) LINKIN sequence from divergent animals ( Homo sapiens , Mus musculus , Drosophila melanogaster , and Caenorhabditis elegans ) were aligned using Clustalw . The intracellular domain shows high conservation in all examined species . ‘*’ indicates identical AA , ‘:’ indicates strong similarity , and ‘ . ’ indicates weak similarity . Signal peptide is boxed in purple , extracellular domain in blue , transmembrane domain in gray , and intracellular domain in green . Sequence deleted by gk367 mutation is underlined in red . FG–GAP domains are highlighted in orange , and the Dx ( D/N ) xDxxxD calcium-binding motif contained within each FG–GAP domain is indicated with red letters . FG–GAP domains were defined as a region from 8 AA N-terminal of the calcium-binding domain to 18 AA C-terminal of the calcium-binding domain , based on annotation by uniprot . org of the second , third , and fifth FG–GAP domains in M . musculus LINKIN . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 004 Overall , the protein sequence between H . sapiens and C . elegans excluding the signal sequence is 26% identical ( 154 AA ) and 61% similar ( 365 AA ) . However , LINKIN has a highly conserved intracellular domain , which is 62 . 5% ( 15/24 AA ) identical and 87 . 5% similar ( 21/24 AA , clustalo analysis ) . In particular , the last eight amino acids are identical in all four species ( Figure 2B ) . Despite its high conservation , a BLAST search of the intracellular domain alone did not identify strong similarities with domains in other proteins . Protein motifs in LINKIN were largely unknown , with the only ascribed motif being an FG–GAP domain found in one copy in H . sapiens and three in M . musculus ( uniprot . org ) . The FG–GAP is a domain that occurs in seven copies in α-integrins and folds into a seven-bladed β-propeller structure that serves as its ligand-binding domain ( Springer , 1997; Xiong et al . , 2002 ) . We have identified seven atypical FG–GAP domains in the N-terminal of the extracellular domain , based on sequence similarities to the annotated LINKIN FG–GAP domains from H . sapiens and M . musculus and to α-integrin FG–GAP domains ( Figure 2 , ‘Materials and methods’ ) . FG–GAP domains have a loosely conserved Phe-Gly and Gly-Ala-Pro sequence , which are separated by sequence that can include a calcium-binding motif . A comparison of all human α-integrin FG–GAP domains showed that their calcium-binding motif has a strong DxxxDxxxD signature ( D = Asp , x = AA; Chouhan et al . , 2011 ) . We found a strong DxxxDxxxD signature in all seven calcium-binding domains of LINKIN , suggesting that these are similar to FG–GAP domains of α-integrins ( Figure 2 ) . The significance of finding seven FG–GAP domains in the N-terminal of LINKIN is the possibility that LINKIN , like α-integrins , uses a seven-bladed β-propeller structure to bind ligand . We also identified a highly conserved extracellular region adjacent to the transmembrane domain , which is a yet unrecognized protein domain ( Figure 2 ) . LINKIN has a dozen predicted N-linked glycosylation sites ( uniprot . org ) . Taken together , our investigation shows that LINKIN is a conserved transmembrane glycoprotein pre-dating Metazoa and potentially containing a seven-bladed , β-propeller , ligand-binding domain . We examined the expression pattern and subcellular localization of LNKN-1 in C . elegans , particularly in the male gonad . Previously , lnkn-1 localization was categorized to be in cell membrane , when examined by GFP-tagging in a screen of muscle-related genes ( Meissner et al . , 2011 ) . The only other investigation into LINKIN observed that in mammals the extracellular domain functions as a secreted protein that modulates T-cell dependent immune response ( Fiscella et al . , 2003 ) . We made both extracellularly and intracellularly YFP-tagged versions of LNKN-1 expressed under its natural regulatory specific promoter ( Figure 3E , F , Figure 3—figure supplement 2M , N ) . Both YFP::LNKN-1 and LNKN-1::YFP are similarly localized to the plasma membrane of many cells . LNKN-1 begins to be expressed in all somatic gonadal cells of the male , including the LC , the vas deferens precursor cells , and seminal vesicle precursor cells , starting in the early L3 stage and continuing through adulthood ( Figure 3E , F ) . It is also expressed in all somatic gonadal cells of the hermaphrodite , including the distal tip cells , anchor cell , uterine precursor cells , and spermatheca precursor cells ( Figure 3—figure supplement 2 ) . Other expression occurs in pharynx , pharyngeal-intestinal valve , intestine , excretory cell and canal , seam cells , a specialized subset of hypodermal cells , the vulval precursor cells of the hermaphrodite , and hook precursor cells in the male ( Figure 3—figure supplement 2 ) . YFP-tagged LNKN-1 is localized to the plasma membrane , exhibiting stronger localization to the sides of cell–cell contact in tissues such as the intestine , seam , and gonad . This broad expression is consistent with our observations that lnkn-1 is expressed in the LC but not enriched ( Schwarz et al . , 2012 ) . 10 . 7554/eLife . 04449 . 005Figure 3 . LNKN-1 localizes to the apical and lateral plasma membrane . ( A–D ) Immunofluorescence staining of dissected male gonads using antibodies against LNKN-1 extracellular domain ( A ) and intracellular domain ( C ) shows localization to the plasma membrane with enrichment at lateral and apical regions ( arrows ) . The antibody against the extracellular domain also labeled cytoplasmic puncta ( A ) , and the antibody against the intracellular domain also labeled the nucleus ( C ) , but these may be due to non-specific staining since they were present in gonads from lnkn-1 RNAi-treated and mutant animals ( Figure 3—figure supplement 1 ) . ( B ) An overlay of the image from ( A ) and an image of the same gonad stained with DAPI . ( D ) An overlay of the image from ( C ) and an image of the same gonad stained with DAPI . In both ( B ) and ( D ) , the cells are outlined in white and the apical domain is highlighted in yellow . ( E and F ) Nomarski and epifluorescence images of a live animal expressing YFP-tagged LNKN-1 show that LNKN-1::YFP is expressed in the plasma membrane of gonadal cells but has spread to the basolateral domain . Bracket marks the male somatic gonad . ( G and H ) LNKN-1 ( mutant ) ::YFP , in which wild-type lnkn-1 cDNA construct from ( F ) is replaced by lnkn-1 ( gk367 ) mutant cDNA , does not localize to the plasma membrane . Nomarski ( G ) and epifluorescence ( H ) images are of male somatic gonad from a live animal expressing LNKN-1 ( mutant ) ::YFP . Scale bar represents 10 μm . Anterior is to the left , posterior is to the right , dorsal is to the top , and ventral is to the bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 00510 . 7554/eLife . 04449 . 006Figure 3—figure supplement 1 . LNKN-1 antibodies specifically label LNKN-1 protein in the plasma membrane . Representative epifluorescence images show dissected male gonads from lnkn-1 RNAi-treated ( A ) , lnkn-1 mutant ( B ) , or wild-type ( C ) males stained with antibody against the extracellular domain of C . elegans LNKN-1 . Gonads in ( A ) , ( B ) , and ( C ) were treated to the same staining conditions and the images were taken with a 200 ms exposure time . Gonads from lnkn-1 RNAi-treated ( D ) , lnkn-1 mutant ( F ) , and wild-type ( E and G ) males were stained with an antibody against the intracellular domain of C . elegans LNKN-1 under identical conditions . Images of lnkn-1 RNAi-treated ( D ) and its control wild-type ( E ) gonads were taken with 100 ms exposure time . Images of lnkn-1 mutant ( F ) and its control wild-type ( G ) gonads were taken at 200 ms exposure time . Plasma membrane staining , particularly in the apical domain ( arrows ) , is reduced in RNAi-treated animals and eliminated in lnkn-1 mutants . Scale bar represents 10 μm in all images . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 00610 . 7554/eLife . 04449 . 007Figure 3—figure supplement 2 . Expression pattern for YFP-tagged LNKN-1 . LNKN-1::YFP localizes to the plasma membrane . In addition to the male somatic gonad ( Figure 3E , F ) , expression was seen in: ( A and B ) seam cells , ( C and D ) pharynx ( bracket ) , pharyngeal-intestinal valve ( arrow ) , and excretory cell ( arrowhead ) , ( E and F ) hermaphrodite somatic gonad ( bracket ) and vulval precursor cells ( arrows ) , ( G and H ) the hermaphrodite distal tip cell , ( I and J ) intestine ( bracket ) , male hook precursor cells ( arrowhead ) , and tail cells ( arrow ) , and ( K and L ) excretory canal ( arrows , longitudinal fluorescent line ) and intestine in the background . ( M and N ) YFP fused to the beginning of the extracellular domain of LNKN-1 , in YFP::LNKN-1 constructs , also shows the same plasma membrane localization . Bracket marks the male somatic gonad . ( O and P ) Dissected , fixed gonad from LNKN-1::YFP expressing animals ( O ) and the same gonad stained with DLG-1 antibody , an apical domain marker . LNKN-1::YFP is expressed uniformly in the plasma membrane and does not show enrichment in the apical domain as marked by DLG-1 antibody . In contrast , LNKN-1 antibody staining shows enriched LNKN-1 localization to the apical domain ( Figure 3A–D ) . In all figures , the top panel is a Nomarski micrograph and the bottom panel is an epifluorescence image of the same region . Scale bars represent 10 μm except in I and J , where they represent 20 μm . Anterior is left , posterior is right , dorsal is up , and ventral is down . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 00710 . 7554/eLife . 04449 . 008Figure 3—figure supplement 3 . lnkn-1 RNAi silencing reduces LNKN-1 protein and causes gonad cell detachment defects . ( A ) A male treated with lnkn-1 RNAi exhibits gonadal defects in the L4 stage . The gonad ( outlined in yellow ) stretches thin and detaches from the LC ( yellow arrow ) . ( B and C ) Nomarski and fluorescence images of a male expressing LNKN-1 tagged with YFP ( LNKN-1::YFP ) show expression in gonadal cells ( outlined in yellow ) . ( D and E ) Treatment of LINKIN::YFP expressing animals with lnkn-1 RNAi reduces LINKIN::YFP expression in the gonad ( outlined in yellow ) to undetectable levels . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 008 To examine the localization of the native protein , two polyclonal antibodies were raised in rabbit against the entire extracellular domain ( 533 AA ) of LNKN-1 and against a peptide derived from the last 17 AA of the short intracellular domain and affinity-purified with the respective antigens . Immunofluorescence staining was performed on dissected gonads and intestines , which greatly improves antibody penetration over whole animals ( Figure 3A–D ) . Staining of other dissected parts of the worm confirmed expression in the pharynx , excretory canal , and seam cells , indicating that the tissue expression pattern of LNKN-1::YFP is accurate . However , there was an important difference in subcellular localization between the native and YFP-tagged proteins: the antibodies showed stronger localization of LNKN-1 to the apical and lateral domain of the gonad ( Figure 3A , B ) and intestine , while YFP-tagged LNKN-1 is uniformly distributed in the plasma membrane . While both antibodies show heavier localization to the apical and lateral plasma membrane , the extracellular domain antibody shows additional staining in large cytoplasmic puncta ( Figure 3A ) , and the intracellular domain antibody shows nuclear staining ( Figure 3B , Figure 3—figure supplement 1G ) . In lnkn-1 mutants and lnkn-1 RNAi-treated animals , we do not see plasma membrane staining with antibodies against either the extracellular or intracellular domain , but we do see non-specific staining in the cytoplasm and nucleus ( Figure 3—figure supplement 1 ) . This indicates that the membrane staining is due to LNKN-1 localization , but the cytoplasmic and nuclear stainings are due to non-specific binding of other proteins . The antibodies also stain the plasma membrane in the germ cells , where YFP expression could not be determined because germ cells do not readily express transgenes . The two antibodies reveal the true localization of native LNKN-1 to be in the plasma membrane with a preference for apical and lateral domains . Since this was the first reported use of the lnkn-1 ( gk367 ) deletion allele , we characterized it molecularly . The genomic locus of lnkn-1 spans 3640 nucleotides and is the second gene in an operon . The gk367 deletion of lnkn-1 excises 393 base pairs ( bp ) of genomic DNA starting soon after the signal sequence and ending in an intron . Since a possibility existed that an in-frame mRNA could be transcribed from the deletion , we characterized the truncated mRNA through RT-PCR and sequencing . The lnkn-1 cDNA sequence resulting from the gk367 deletion is missing the last 18 bp of the signal sequence and the first 258 bp of the extracellular domain ( Figure 2 ) . The larger size of the cDNA deletion than would be predicted based on the genomic lesion indicates that an alternate splice site was used when the lesion removed the usual splice donor; however , the product cDNA is still in-frame . Since mRNA was being transcribed in lnkn-1 mutants , we investigated whether protein was being expressed . We generated a lnkn-1 promoter::lnkn-1 ( mutant cDNA ) ::yfp construct , in which yfp was fused to mutant lnkn-1 cDNA and expressed using its 5′ genomic region . We found that LNKN-1 ( mutant ) ::YFP is in fact expressed but shows cytoplasmic rather than plasma membrane expression ( Figure 3G , H ) . This confirms that a truncated protein is being produced from the lnkn-1 deletion locus but is mislocalized and unlikely to have its normal function . To test whether the lnkn-1 deletion mutant functions as a null despite producing a truncated protein , we examined the phenotype produced by lnkn-1 RNAi silencing . Males treated with lnkn-1 RNAi produced gonadal defects that were milder than the mutant ( Figure 3—figure supplement 3A ) . While only 11% ( 4/35 animals ) had detached gonadal cells , an additional 17% ( 6/35 animals ) had ‘stringy’ gonads , in which fewer cells remained attached and were stretched from pulling by the LC . We also performed lnkn-1 RNAi on animals expressing LNKN-1::YFP to ensure that the RNAi was effective ( Figure 3—figure supplement 3B–E ) . LNKN-1::YFP was absent from the gonad in all animals ( n = 28 ) , but was retained in a few tissues including the pharynx and excretory canal , likely because these tissues produce higher levels of LNKN-1 or were more resistant to the effects of RNAi . Since RNAi effectively reduces but does not eliminate the function of LNKN-1 , we interpret the similar but stronger phenotype of the mutant to indicate that the mutant is in fact a loss-of-function allele . We investigated the requirement of various domains of LNKN-1 for rescuing the mutant phenotype . We were able to completely rescue the lnkn-1 mutant , including maternal effect lethality and gonad adhesion defects , using a genomic construct ( Figure 4A ) . Since lnkn-1 is the second gene in an operon , this construct contains 4 . 5 kb of genomic region upstream of lnkn-1 start site , the lnkn-1 gene , and the lnkn-1 3′ UTR ( Figure 4A ) . We also made a cDNA construct using the same 5′ region of lnkn-1 fused to lnkn-1 cDNA and unc-54 3′ UTR , a common 3′ UTR for C . elegans constructs . The lnkn-1 cDNA was able to rescue the gonad defect but not maternal effect lethality ( Figure 4B ) , possibly because it does not contain all regulatory elements for complete tissue expression or is silenced in the germline . 10 . 7554/eLife . 04449 . 009Figure 4 . Rescue of lnkn-1 mutant phenotypes requires the full-length lnkn-1 gene . ( A ) A full-length genomic construct containing 4 . 5 kb of 5′ regulatory region , lnkn-1 coding region , and 3′ UTR rescues both gonad detachment defects in the male and maternal lethality . ‘+’ indicates rescue and ‘−’ indicates no rescue . Micrograph shows the posterior body of a male lnkn-1 ( gk367 ) mutant that has been rescued for gonad detachment by a full-length genomic lnkn-1 construct . ( B ) 4 . 1 kb of genomic promoter region fused to lnkn-1 cDNA and unc-54 3′ UTR rescues male gonad defect but not maternal lethality . ( C and D ) Constructs with YFP inserted within the extracellular domain ( C ) or the intracellular domain ( D ) of lnkn-1 cDNA did not rescue lnkn-1 mutants . ( E–H ) 4 . 1 kb of genomic 5′ control region fused to partial domains of lnkn-1 ( cDNA ) do not rescue . lnkn-1 constructs of extracellular domain only ( E ) , extracellular and transmembrane domain ( F ) , transmembrane and intracellular domain ( G ) , and intracellular only ( H ) also did not rescue lnkn-1 mutants . ( I ) LC-specific expression of lnkn-1 using lag-2 control sequences also does not rescue . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 009 We also investigated whether other lnkn-1 constructs can rescue the mutant ( Figure 4 ) . Full-length lnkn-1 with yfp fused to either the intracellular or extracellular domain did not rescue the mutant ( Figure 4C , D ) . This is not surprising considering that YFP tagging prevents correct localization of LNKN-1 to the apical domain ( see above ) . While both YFP-tagged proteins were expressed well in wild-type animals , intracellularly fused LNKN-1::YFP is not well-tolerated in lnkn-1 mutant heterozygotes and is lost within a few generations . Partial domains of lnkn-1 were also not able to rescue the mutant; we attempted rescue with constructs containing only the secreted extracellular domain , the extracellular and transmembrane domain , the transmembrane and intracellular domain , and the cytoplasmic intracellular domain ( Figure 4E–H ) . Lastly , we expressed full-length lnkn-1 under a LC promoter to test a gonadal non-cell autonomous effect , but this also did not rescue the mutant ( Figure 4I ) . We conclude that the intact protein , with intact extracellular and intracellular domains , is required for the function of LNKN-1 . Since no interactors were previously known , we used a proteomics approach to identify binding partners in order to better understand the function of LNKN-1 through its interactions . Based on the hypothesis that the highly conserved intracellular sequence suggests both a required function for this domain and a potential for its binding partners to be conserved , we decided to identify LINKIN interactors by mass spectrometry using a human cell line . The advantage of using a human cell line compared to whole C . elegans is that it is a homogeneous cell type and the conditions for SILAC mass spectrometry are established . By using SILAC ( stable isotope labeling by amino acids in cell culture ) , experiment and control immunoprecipitates can simultaneously be analyzed by mass spectrometry , enabling both quantitation and background subtraction . SILAC mass spectrometry was performed on immunoprecipitates from human LINKIN-Myc-expressing HEK 293T cells without isotopic labeling and from non-transfected cells with heavy isotopic labeling . Based on the ratio of ‘light’ to ‘heavy’ isotopes for each protein , enrichment through specific binding to LINKIN over background non-specific binding was determined . 484 proteins with at least two unique peptides were identified by LC/MS/MS from the immunoprecipitate with LINKIN , excluding common contaminants ( Supplementary file 1 ) . As validation for successful immunoprecipitation , LINKIN was itself one of the proteins with highest enrichment in LINKIN-Myc-expressing cells over control cells ( 35-fold enrichment; Figure 5A ) . 10 . 7554/eLife . 04449 . 010Figure 5 . Interactors of LINKIN are RUVBL1 , RUVBL2 , and α-tubulin . ( A ) Graph represents human LINKIN interactors identified by mass spectrometry . ITFG1 ( human LINKIN ) co-immunoprecipitates from SILAC treated HEK 293T cells with ITFG1-Myc expression were compared to unlabeled cells without ITFG1-Myc expression . ( B–F ) RNAi knockdowns in C . elegans of ruvb-1 , ruvb-2 , α-tubulin , and β-tubulin show the same gonad cell detachment as lnkn-1 mutant . Gonad is outlined in yellow and LC is marked by cytoplasmic YFP . Figures are an overlay of Nomarski and fluorescence images . ( G–K ) Western blots of ITFG1 ( human LINKIN ) co-immunoprecipitates probed with antibodies against RUVBL1 , RUVBL2 , α-tubulin , β-tubulin , and control β-actin show that LINKIN interacts with RUVBL1 , RUVBL2 , and α-tubulin . Myc immunoprecipitation was performed on non-transfected cells ( left column ) and ITFG1-Myc transfected cells ( right column ) . Equal protein loading was determined by Ponceau S staining . ( L–S ) Lysates from cells transfected with Myc-ITFG1 , Flag-RUVBL1 , and HA-RUVBL2 ( labeled ‘T’ ) or with Myc control ( labeled ‘C’ ) were separated into a cytoplasmic and a membrane fraction . The cytoplasmic fraction ( L–O ) and membrane fraction ( P–S ) were immunoprecipitated using Flag-RUVBL1 and assayed by Western blot for interactors . ITFG1 is only detected in the membrane fraction ( P ) . RUVBL1 ( M ) , RUVBL2 ( N ) , and α-tubulin ( O ) interact in the cytoplasmic fraction even without ITFG1 . ITFG1 ( P ) , RUVBL1 ( Q ) , RUVBL2 ( R ) , and α -tubulin ( S ) interact in the membrane fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 010 Our aim was to identify , among the many LINKIN interactors , those functioning in maintaining gonadal cell attachment . We performed an assay in C . elegans based on the hypothesis that some of the binding partners of human LINKIN would also be conserved in C . elegans . Our approach was to assign C . elegans homologs to the ITFG1-interacting genes and perform an RNAi screen in C . elegans , seeking genes with a similar gonadal cell detachment phenotype as the lnkn-1 mutant . There were 68 proteins ( excluding LINKIN ) identified by mass spectrometry that were >fivefold enriched in ITFG1-Myc immunoprecipitates over control , and 45 of them had at least one C . elegans homolog ( Supplementary file 2 ) . 40 genes were available in existing RNAi libraries and screened . The silencing of three genes , ruvb-1/RUVBL1 ( also known as Pontin ) , ruvb-2/RUVBL2 ( also known as Reptin ) , and tba-2/α-tubulin , caused a similar gonadal defect to lnkn-1 , ( Figure 5B–E ) . ruvb-1/RUVBL1 and ruvb-2/RUVBL2 are highly conserved members of the AAA+ ATPase superfamily of proteins that often function together in a hexameric ring complex ( Matias et al . , 2006; Gorynia et al . , 2011 ) . α-tubulin together with β-tubulin forms microtubules , which as part of the cell cytoskeleton have roles in cell mechanics and transport of cellular components ( Etienne-Manneville , 2013 ) . β-tubulin was also among the highly enriched human gene interactors , but the homologous C . elegans β-tubulin did not produce a gonadal phenotype by RNAi . We tested the other five C . elegans β-tubulin genes and found that tbb-2 has a gonadal detachment defect ( Figure 5F ) . Having identified RUVBL1 , RUVBL2 , α- and β-tubulin as potential interactors of LINKIN that were also involved in the same biological process as LNKN-1 in cell adhesion , we wanted to confirm their physical interaction . Binding between RUVBL1 and RUVBL2 into heteromeric multimers ( Gorynia et al . , 2011 ) and binding between RUVBLs and microtubules have been reported ( Gartner et al . , 2003; Dobreva et al . , 2008; Ducat et al . , 2008 ) . To test physical interaction between LINKIN and each of RUVBL1 , RUVBL2 , α- and β-tubulin , we performed Western blots on co-immunoprecipitates from ITFG1-Myc-expressing HEK 293T cells . Probing with antibodies against RUVBL1 , RUVBL2 , α-tubulin , and β-tubulin , we found that RUVBL1 , RUVBL2 , and α-tubulin bound LINKIN ( Figure 5G–I ) . However , we did not observe β-tubulin binding LINKIN ( Figure 5J ) . As a control for binding specificity of abundant cytoskeletal proteins , we also probed with an anti-β-actin antibody and found that β-actin did not bind LINKIN ( Figure 5K ) . We next investigated whether these interactions occur at the plasma membrane . We isolated the membrane and cytoplasmic fractions from HEK293T cells transfected with plasmids expressing ITFG1-Myc , Flag-RUVBL1 , and HA-RUVBL2 . By Western blot analysis , we showed that RUVBL1 , RUVBL2 , and α-tubulin were present in both cytoplasmic and membrane fractions , but ITFG1 was only present in the membrane fraction . We then immunoprecipitated with Flag-RUVBL1 from both cytoplasmic and membrane fractions and determined by Western blot analysis which proteins interacted with RUVBL1 ( Figure 5L–S ) . ITFG1 , RUVBL2 , and α-tubulin interacted with RUBVL1 in the membrane fraction ( Figure 5P–S ) . In the cytoplasmic fraction , RUVBL1 , RUVBL2 and α-tubulin interact even in the absence of ITFG1 ( Figure 5L–O ) . These results showed that the interaction between ITFG1 , RUVBL1 , RUVBL2 , and α-tubulin occurs at the membrane , likely the plasma membrane based on C . elegans LINKIN localization . We next investigated whether LNKN-1 interactors also localize to the plasma membrane . To examine RUVB-1 and RUVB-2 localization in the gonad , we generated polyclonal rabbit antibodies against full-length C . elegans RUVB-1 and RUVB-2 proteins , which were affinity-purified using the respective full-length proteins ( Figure 6A–B ) . Strong staining for both RUVB-1 and RUVB-2 was observed in the cytoplasm and nucleus of gonadal cells ( Figure 6A , B ) . We demonstrated the specificity of our anti-RUVB-1 and anti-RUVB-2 antibodies by comparing their staining in dissected gonads from wild-type animals ( n = 6 ) and ruvb-1 ( n = 6 ) or ruvb-2 ( n = 8 ) RNAi-treated animals ( Figure 6—figure supplement 1 ) . We observed a consistent decrease in staining in the cytoplasm and nucleus of the gonads from RNAi-treated animals , indicating that RUVB-1 and RUVB-2 actually localize to both locations . Anti-α-tubulin antibody stained a dense network of microtubule fibers throughout the cytoplasm but particularly in the cell cortex of all gonadal cells ( Figure 6C ) . The microtubules were more densely aligned along the developing apical domain of the gonad . Based on the Western blot analysis of membrane fractionated cells described above and the antibody staining results in C . elegans , the localization of LINKIN , RUVBL1 , RUVBL2 , and α-tubulin intersect at the cytoplasmic face of the plasma membrane , which agrees with the adhesion function of LINKIN . 10 . 7554/eLife . 04449 . 011Figure 6 . Antibodies against RUVB-1 , RUVB-2 , and α-tubulin show localization in C . elegans male gonads . Dissected male gonads stained with antibody against C . elegans RUVB-1 ( A ) and RUVB-2 ( B ) show localization in cytoplasm and nucleus . Figure 6—figure supplement 1 shows that both cytoplasmic and nuclear stainings are specific to RUVB-1 or RUVB-2 proteins and can be reduced by ruvb-1 or ruvb-2 RNAi . ( C ) Dissected male gonad stained with antibody against α-tubulin shows network of microtubule fibers throughout the gonad , with stronger localization to the cell cortex and apical domain ( arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 01110 . 7554/eLife . 04449 . 012Figure 6—figure supplement 1 . The anti-RUVB-1 antibody specifically labels RUVB-1 protein . Representative epifluorescence images show dissected gonads from wild-type ( A ) or ruvb-1 RNAi-treated ( B ) males stained with antibody against C . elegans RUVB-1 and gonads from wild-type ( C ) or ruvb-2 RNAi-treated ( D ) males stained with antibody against C . elegans RUVB-2 . The pairs of gonads in A and B and the gonads in C and D were treated to identical staining conditions and imaged at the same exposure times . Both the cytoplasmic and nuclear staining are reduced in RNAi-treated animals . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 012 Since the known functions for RUVB-1 and RUVB-2 include transcriptional regulation ( Jha and Dutta , 2009 ) and for α-tubulin include protein transport ( Miller et al . , 2009 ) , we investigated whether RUVB-1 , RUVB-2 , or α-tubulin is required for either LNKN-1 expression or localization . We used RNAi to reduce ruvb-1 , ruvb-2 , and tba-2 function in LNKN-1::YFP animals and found that RNAi silencing of these genes did not affect LNKN-1::YFP function or localization ( Figure 7 ) . For ruvb-1 and ruvb-2 RNAi-treated animals , we also examined LNKN-1 expression by immunofluorescence staining and found no difference from untreated animals . 10 . 7554/eLife . 04449 . 013Figure 7 . LNKN-1 expression and localization are not dependent on ruvb-1 , ruvb-2 , or tba-2 function . Representative epifluorescence images of LINKIN::YFP animals that were either not treated ( A ) , or treated with RNAi against ruvb-1 ( B ) , ruvb-2 ( C ) , or tba-2 ( D ) . The expression of LINKIN::YFP and its localization to the plasma membrane is similar in RNAi-treated and untreated animals . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 013 RUVBLs are members of many complexes ( Rosenbaum et al . , 2013 ) , but their only known interaction with microtubules has been at the mitotic spindle ( Gartner et al . , 2003; Ducat et al . , 2008 ) . We therefore investigated whether LINKIN or cleaved domains of LINKIN may also localize to the spindle during cell division . There are precedents for membrane-associated proteins involved in cell adhesion , such as ILK and β-catenin , to localize to the spindle with RUVBLs and microtubules ( Kaplan et al . , 2004; Fielding , et al . , 2008; Dobreva et al . , 2008 ) . We first examined microtubule localization , which is known to redistribute during mitosis to a formation that radiates out from the mitotic spindle and attaches to the kinetochore and cell cortex . In fixed dissected gonads from L3 stage animals , by using an antibody against α-tubulin to identify microtubules and DAPI staining to identify condensed chromosomes of dividing cells , we observed the expected microtubule redistribution to the mitotic spindle ( Figure 8A , B ) . The usually dense network of microtubules found in the cytosol and cell periphery of non-mitotic cells was replaced in mitotic cells by radial microtubules . We next examined whether LNKN-1 or domains of LNKN-1 were redistributed during cell division . Both antibodies against LNKN-1 extracellular ( Figure 8C , D ) and intracellular domains ( Figure 8E , F ) showed that localization of LNKN-1 remains unchanged at the plasma membrane during mitosis . These observations imply that there is a temporary loss of interaction between α-tubulin , RUVBL proteins , and LNKN-1 at the membrane and the interactions must be reestablished after cell division . 10 . 7554/eLife . 04449 . 014Figure 8 . LNKN-1 remains at the plasma membrane during cell division . ( A , C , E ) In DAPI-stained dissected male gonads , dividing cells ( white bracket ) were identified by their condensed chromosomes . ( B ) Anti-α-tubulin staining shows that microtubules redistribute during cell division to radiate out from the mitotic spindle . ( D and F ) Staining with anti-LNKN-1 antibody against either the extracellular domain ( D ) or intracellular domain ( F ) shows that LNKN-1 remains at the membrane during cell division . LNKN-1 localization is the same in dividing cells ( white bracket ) and non-dividing neighbors . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 014 To determine whether gonad detachment in lnkn-1 mutants involves the dissolution of mature cell–cell junctions , we examined cell junction formation using the apical junction marker , AJM-1::GFP , and gap junction marker innexin , INX-5::GFP ( Figure 9 ) . AJM-1 localizes to apical adhesion junctions in a complex with cadherins ( Köppen et al . , 2001 ) . In the tube-shaped adult gonad , we observed AJM-1::GFP lining the entire lumen composed of the apical surfaces of the surrounding somatic gonad cells ( Figure 9C , D ) . This apical accumulation begins in the mid-L4 stage as puncta ( Figure 9A , B ) , but it is not present in the L3 stage when gonad detachment usually occurs in the lnkn-1 mutant . INX-5 shows expression and localization to cell–cell junctions in a few cells in the distal somatic gonad also starting in the L4 stage and becoming stronger in the adult ( Figure 9E–H ) . Results based on both adhesion markers are consistent with the timeline of gonadal development , which dictates that cell division and rearrangement occurs during the L3 stage , while differentiation into the mature structure occurs during the latter half of the L4 stage up until the transition into adulthood . Cell dissociation in lnkn-1 mutants occurs at a stage when cell junctions are being rearranged; during this growth phase , adhesion-promoting genes like lnkn-1 may have a greater effect than later after the establishment of other more secure junctions . 10 . 7554/eLife . 04449 . 015Figure 9 . Mature cell–cell junctions form during the L4 stage . ( A and B ) Adherens junction marker , AJM-1::GFP , begins to localize as puncta to the apical region of the gonad in the L4 stage ( arrows ) . ( C and D ) In the adult gonad , AJM-1::GFP lines the apical junctions . ( E and F ) INX-5::YFP , an innexin expressed in the male gonad , begins to be expressed and accumulate at gap junctions in the L4 stage in a cluster of somatic gonadal cells ( arrow ) . ( G and H ) INX-5::YFP expression becomes stronger in the adult . Neither AJM-1::GFP nor INX-5::YFP expresses in the gonad in the L3 stage . For each image pair , the top panel is a Nomarski image and bottom is a fluorescence image . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 015
We have demonstrated that LINKIN is a unique transmembrane glycoprotein that functions as an adhesion molecule . This approximately 600 AA protein has a large extracellular domain of approximately 530 AA and a short 22 AA intracellular domain . We found a number of notable features about LINKIN . First , it is a protein pre-dating metazoans . Second , the size of the protein and organization of the domains have not changed during its evolution . The seven atypical FG–GAP domains , the extracellular region proximal to the transmembrane domain , and the intracellular sequence are all conserved . The identical sequence at the C-terminal of LINKIN suggested that some of its intracellular interactors may also be conserved proteins . In fact , LINKIN interactors , RUVBL1 , RUVBL2 , and α-tubulin are all highly conserved proteins that likely pre-date LINKIN . Third , among the genomes we searched , including those of H . sapiens , M . musculus , D . melanogaster , C . elegans , and T . adhaerens , LINKIN has no paralog . Even the highly conserved short 22 AA intracellular domain is not found in other proteins . The lack of paralogs is surprising considering that other highly conserved adhesion molecules like integrins and cadherins have numerous paralogs and have expanded into superfamilies ( Angst et al . , 2001; Takada et al . , 2007 ) . The presence of seven atypical FG–GAP domains in the extracellular domain suggests that LINKIN might fold into a seven-bladed β-propeller . The significance of this prediction is that this structure resembles the ligand-binding domain of the adhesion molecule α-integrin . As with α-integrins , the β-propeller structure for LINKIN is located towards the amino-terminal of the extracellular domain . LINKIN is expressed on the surface of each adherent cell and might use the β-propeller domain to bind ligands on a neighboring cell to promote adhesion . The C . elegans gonad shape develops through a collective migration of epithelial-like cells lead by the LC . One of the striking phenotypes of the lnkn-1 mutant was cell detachment in the migrating gonad . In lnkn-1 mutants , gonadal cell dissociation occurs in the L3 stage . At this stage , although the adhesions are strong enough that these cells cannot be mechanically dissociated without lysing , they have not yet formed mature adherens junctions . The expression of adhesion junction markers , AJM ( apical junction molecule ) -1::GFP , and gap junction marker , INX ( innexin ) -5::GFP , was absent in the L3 stage , when gonad cells begin detaching in lnkn-1 mutants; their expression only begins in the L4 stage and grows stronger in the fully differentiated adult gonad . The role of adhesion molecules , such as LINKIN , may therefore be more important before other , possibly stronger , adhesions are formed . The gonad detachment defect of lnkn-1 mutants likely results from the combined effects of the absence of a functional LINKIN adhesion molecule , the lack of other permanent adhesion structures in the L3 stage , and the force generated by the migrating LC . LNKN-1 was present in many C . elegans tissues that possess apical/basal polarity , including the intestine , excretory canal , vulva , hook cells , and gonad; its localization was stronger on the apical and lateral sides for the tissues examined . YFP-tagged to either the extracellular or intracellular domain of LNKN-1 changes its localization from apical- and lateral-biased to uniform plasma membrane localization , indicating that apical localization is an active process requiring the function of both extracellular and intracellular domains of the protein . YFP-tagging also disrupts the function of LINKIN , as neither YFP-tagged LINKN-1 construct rescued the mutant phenotype , suggesting that apical localization may be necessary for LNKN-1 function . Known adhesion molecules have preferential localization at the plasma membrane . Integrins are often enriched in the basal domain since they bind extracellular matrix ( Schoenenberger et al . , 1994 ) , while cadherins are enriched in lateral domains ( Halbleib and Nelson , 2006 ) . Specialized adhesion molecules like claudins and occludins localize to tight junctions ( González-Mariscal et al . , 2003 ) and connexins to gap junctions ( Evans and Martin , 2002 ) . LINKIN may be an adhesion molecule for cell–cell contacts and apical junctions . Our rescue experiments also showed that secreted forms and partial domains of LNKN-1 do not provide function . The only previous study of LINKIN showed that the extracellular domain of human LINKIN functions as a secreted protein to modulate T-cell activation in cell culture and graft-versus-host disease model ( Fiscella et al . , 2003 ) . Our experiments indicate that in the context of C . elegans collective migration , a secreted extracellular domain is insufficient to rescue the detachment defect . Our results suggest that , in the many C . elegans and human tissues expressing LINKIN , its function could originally have been as a transmembrane adhesion molecule , which in vertebrates conceivably has expanded to include a secreted form . Although we do not yet know the binding partner on the extracellular side , we have made progress in identifying conserved interactors on the intracellular side—RUVBL1 , RUVBL2 and α-tubulin . RUVBL1 and RUVBL2 are highly conserved members of the AAA+ ATPase superfamily , which use the energy harvested from ATP hydrolysis to perform mechanical functions on macromolecules . In bacteria where they were first identified , RUVBs have a function as a DNA helicase at holiday junctions; but in more complex organisms , the RUVBLs have additional diverse functions ( Jha and Dutta , 2009 ) . Among its non-nuclear roles , RUVBLs function in R2TP co-chaperone complex assembly ( Kakihara and Houry , 2012 ) and in spindle assembly by nucleating microtubules and localizing components to the mitotic spindle ( Gartner et al . , 2003; Ducat et al . , 2008 ) . We have shown through limited cell fractionation that LINKIN is present in the membrane , where it interacts with RUVBL proteins and α-tubulin . Although we showed that LINKIN does not localize to the mitotic spindle , the latter role of RUVBLs as regulators of microtubule assembly and complex formation may be most relevant to its function with LINKIN and α-tubulin . Based on previously known RUVBL functions , we propose that RUVBL1 and RUVBL2 form a heterometric ring structure that promotes assembly of a LINKIN complex and nucleation of microtubules ( Figure 10 ) . 10 . 7554/eLife . 04449 . 016Figure 10 . Model for the function of LINKIN , RUVBL1 , RUVBL2 , and microtubule proteins in cell–cell adhesion . ( A ) The male gonad shape is generated by a collective migration of the leader LC ( green ) and follower somatic cells ( blue ) . LINKIN ( purple ) is a transmembrane glycoprotein expressed in the plasma membrane of all gonadal cells , with enrichment at apical and lateral domains ( dark purple ) . LNKN-1 is required for cell–cell adhesion during gonadal migration . ( B ) The interface between two adherent cells boxed in ( A ) is shown in more detail . LINKIN integrates interactions with neighboring cells on the cell exterior with connections to the microtubule cytoskeleton on the cell interior . On the extracellular side , the β-propeller domain ( purple heptagon ) of LINKIN binds an unidentified partner ( white oval ) on the adjacent cell membrane . The highly conserved intracellular domain of LINKIN binds RUVBL1 , RUVBL2 , and α-tubulin at the intracellular face of the plasma membrane . Based on RUVBL1 and RUVBL2's ability to form stacked hetero-hexameric rings ( Gorynia et al . , 2011 ) and regulate microtubule nucleation and dynamics ( Gartner et al . , 2003 ) , we propose that they assist in interaction between LINKIN and microtubules . DOI: http://dx . doi . org/10 . 7554/eLife . 04449 . 016 Most adhesion receptors interact with the cell cytoskeleton through their intracellular domain ( Juliano , 2002 ) ; LINKIN interacts with microtubules . Microtubules are known to play important functions in cell migration and tissue organization ( Gauthier-Rouvière et al . , 2004; Etienne-Manneville , 2013 ) . They provide structure and rigidity to tissues so that they can withstand high compression forces ( Brangwynne et al . , 2007 ) , and the bundled cortical microtubules of gonadal cells likely provide such a structure . Microtubules are also involved in creating polarity and trafficking components along polarized tracks . The subunits have a plus and minus end polarity growing out from the centrosome , which in turn can create a front–back polarization in migratory cells through selective stabilization of fibers ( Wadsworth , 1999 ) and polarized trafficking to the membrane ( Miller et al . , 2009 ) . Microtubules serve as tracks to transport cadherin-containing vesicles to specific areas of the plasma membrane to establish cell adhesion ( Mary et al . , 2002 ) . LINKIN may serve to anchor the microtubule cytoskeleton to particular domains of the plasma membrane . Conversely , microtubules and microtubule-associated motor proteins may transport LINKIN to select domains of the plasma membrane , helping to establish LINKIN localization to apical domains and cell–cell contacts . As investigations into other cell adhesion molecules have revealed numerous important functions in animal development ( Thiery , 2003; Halbleib and Nelson , 2006 ) , LINKIN's expression in many human and C . elegans tissues suggests that future studies will demonstrate its involvement in many processes . We have demonstrated an adhesion function for the transmembrane glycoprotein LINKIN and have identified interactors RUVBL1 , RUVBL2 , and α-tubulin , which support a model for LINKIN regulating the microtubule cytoskeleton . Considering that the interactions between LINKIN , RUVBL1 , RUVBL2 , and α-tubulin were identified using a human cell line and were also required for gonad cell adhesion in a nematode worm , these interactions may have a conserved molecular function in Metazoa including human . We are proposing that LINKIN functions in maintaining tissue integrity through cell–cell adhesion and apically polarization , but further studies are necessary to show the generality of this function and to elucidate differences between the roles of LINKIN and other adhesion receptors .
C . elegans strains were cultured at room temperature using standard protocols unless indicated otherwise ( Brenner , 1974 ) . Strain VC877 tag-256 ( gk367 ) /hT2 was obtained from the Caenorhabditis Genetics Center ( CGC ) . Other alleles and transgenes used in this study are him-5 ( e1490 ) ( Hodgkin et al . , 1979 ) | PS4730 syIs128 [lag-2::YFP]; him-5 ( e1490 ) ( Kato and Sternberg , 2009 ) | syIs78 [ajm-1::GFP] ( Gupta et al . , 2003 ) | inx-5::GFP | PS6018 unc-119 ( ed4 ) ; him-5 ( e1490 ) syEx1130[tag-256::TAG-256::YFP ( 20 ng/µl ) + unc-119 ( + ) ( 70 ng/µl ) + Bluescript] | PS6372 tag-256 ( gk367 ) ; him-5 ( e1490 ) ; syEx1184[tag-256::TAG:256 ( 5 ng/µl ) + myo-2::mCherry] . Total RNA was extracted from 25 lnkn-1 ( gk367 ) mutant animals using a TRIzol extraction method ( Chomczynski and Sacchi , 1987 ) but modified for small quantities of worms ( Morimoto group , http://groups . molbiosci . northwestern . edu/morimoto/research/Protocols/IX . %20C . %20elegans/B . %20Extraction/2 . %20TotalRNA . pdf ) . This was followed by a reverse transcription reaction using SuperScriptIII first-strand synthesis supermix ( Invitrogen , Waltham , MA ) following manufacturer's instructions and PCR using lnkn-1 primers to the beginning and end of the gene . The amplified product was submitted for DNA sequencing ( Laragen ) . See Supplementary file 3 . For lnkn-1 mutant rescue experiments , a DNA mixture of 5 ng/μl or 1 ng/μl of lnkn-1 rescuing construct , 7 ng/μl of myo-2::dsRed , and 150 ng/μl of 1 kb ladder ( NEB ) as carrier DNA , was injected into adult gonads of PS6372 lnkn-1 ( gk367 ) /hT2; him-5 ( e1490 ) hermaphrodites . Genecards . org was used to identify the C . elegans homologs of the human genes identified by mass spectrometry . C . elegans genes were screened using an RNAi protocol previously described ( Kamath et al . , 2001 ) , with a few modifications . Single RNAi bacterial colonies were grown for 6–8 hr in LB with carbenicillin selection ( 100 μg/ml ) . Carbenicillin ( 25 μg/ml ) and IPTG ( 1 mM ) were spread on NGM plates just prior to adding 200 μl of RNAi bacterial culture , and plates were dried at RT overnight . The following day , eggs were harvested from gravid adults by bleaching and placed on plates containing RNAi bacteria . Animals were grown at RT and L4 stage males were scored by Nomarski and fluorescence microscopy for detached gonad . The RNAi bacteria were obtained from the Vidal library ( Rual et al . , 2004 ) when the gene was available , and the Ahringer library ( Kamath et al . , 2003 ) otherwise . Equal amounts of cell lysate were mixed with SDS sample buffer ( 4× ) and heated at 90°C for 5 min . After centrifugation at 3000×g for 30 s , sample ( 25 μg ) was loaded on a 4–12% or 4–20% SDS-PAGE . Proteins were transferred to nitrocellulose membranes and stained with Ponceau S . The nitrocellulose membranes were blocked with 5% milk/TBST for 5 min to remove Ponceau S and for additional 30 min . Primary antibodies were prepared in 3% milk/TBST and incubated for 1 hr at room temperature on a shaker . After removing the primary antibody , membranes were washed three times with 3% milk/TBST for 5 min each . Primary antibodies were used including rabbit polyclonal antibodies against human RUVBL1 and RUVBL2 ( Proteintech ) , anti-β-tubulin and anti-β-actin . Secondary antibodies were incubated for 1 hr at room temperature on a shaker . Membranes were washed three times with TBST with 5 min each time . ECL Plus ( GE healthcare ) was used to detect signal . HEK 293T cells were transfected with plasmids encoding ITFG1-Myc ( RC204773 , Origene ) , Flag-RUVBL1 ( 51635 , Addgene , Cambridge , MA ) , and HA-RUVBL2 ( 51636 , Addgene ) or with control Myc-vector . Mem-PER Plus Membrane Protein Extraction Kit ( Thermo Fisher Scientific ) was used to isolate membrane and cytoplasmic fractions from harvested cells . Flag-RUVBL1 was immunoprecipitated from the two fractions using Flag-beads ( Sigma-Aldrich , St . Louis , MO ) . Immunoprecipitation/Western blot analysis was performed as described above . A soluble LNKN-1 protein containing the entire extracellular domain was expressed and purified from S2 cells by the Caltech protein expression facility . Rabbit polyclonal antibodies were generated against this LNKN-1 extracellular domain protein and a 17 AA intracellular domain peptide , Ac-C ( Ahx ) DRYERQQQSHRFHFDAM-OH ( QCB , Hopkinton , MA ) . The antibodies were affinity-purified using their antigens , either the extracellular domain protein or the intracellular domain peptide . Rabbit polyclonal antibodies were also generated against the entire RUVB-1 and RUVB-2 proteins ( Proteintech ) and affinity-purified using the RUVB-1 and RUVB-2 proteins . Specificity of all antibodies was tested by staining tissues from C . elegans that were either treated with RNAi against the antigen gene or mutant for the gene . Reduction in staining was observed for each of these antibodies ( Figure 3—figure supplement 1 , Figure 6—figure supplement 1 ) . Gonads were dissected from larval stage males following Chan and Meyer's ‘Protocol 21: Antibody staining of C . elegans gonads’ ( Shaham , 2006 ) . A final concentration of 2% paraformaldehyde was used and phosphate buffered saline was substituted for sperm salts . Primary antibodies were used at 1:250 dilution for antibody against LNKN-1 extracellular domain , RUVB-1 , and RUVB-2 and at 1:200 dilution for LNKN-1 intracellular domain . Mouse monoclonal antibodies against α-tubulin ( 12G10 , supernatant , Developmental Studies Hybridoma Bank , Iowa City , IA ) and DLG-1 ( DLG-1 , supernatant , DSHB ) were used at a 1:100 dilution . Secondary antibodies against rabbit ( Alexa Fluor 594 Goat Anti-Rabbit IgG , Life Technologies , Carlsbad , CA ) and mouse ( Alexa Fluor 594 Goat Anti-Mouse IgG , Life Technologies ) were used at 1:500 dilution . Tissues were mounted on slides using Vectashield mounting media containing DAPI ( Vector Laboratories , Burlingame , CA ) . LINKIN sequences from H . sapiens , M . musculus , D . melanogaster ( uniprot . org ) , C . elegans ( wormbase . org ) were aligned using clustalw ( http://www . genome . jp/tools/clustalw/ ) and clustalo ( http://www . ebi . ac . uk/Tools/msa/clustalo/ ) . T . adhaerens LINKIN is TRIADDRAFT_52570 but the gene prediction is imperfect ( metazoa . ensembl . org ) . The FG–GAP domain , first identified in α-integrin , contains a loosely conserved sequence of Phe-Gly and Gly-Ala-Pro . A motif frequently found between the FG and GAP sequences is a DxDxDG calcium-binding motif ( D = Asp , G = Gly , x = AA; Chouhan et al . , 2011 ) . While DxDxDG is a common calcium-binding motif , variations on this motif depend on the particular protein family ( Rigden and Galperin , 2004 ) . A comparison of all human α-integrin FG–GAP domains showed that their calcium-binding motif has a strong DxxxDxxxD signature , which contains a more weakly conserved DxDxDG sequence as its first 6 AAs ( Chouhan et al . , 2011 ) . This DxxxDxxxD signature is different from DxDxDG calcium-binding regions of other proteins like the EF-hand protein family ( Rigden and Galperin , 2004 ) . The FG–GAP sequence was only loosely conserved and not always identifiable , but the DxxxDxxxD calcium-binding domain was highly conserved in LINKIN . | In animals , cells can move from one place to another to shape tissues , heal wounds , or defend against invading microbes . A cell may move alone or it may be attached to others and move as part of a group . One member of the group leads this ‘collective migration’ , but it is not known how the cells are able to stick to each other and move together . Collective migration takes place in the male gonad—the organ that makes sperm cells—in larvae of the nematode worm C . elegans . As the gonad matures , a group of cells form a simple chain that can move together . Kato et al . found that a protein called LINKIN must be present for this to happen . LINKIN is found in the membrane that surrounds animal cells . One section of the protein—called the β-propeller—sits on the outside surface of the membrane . The structure of the β-propeller is similar to a section of another protein—called α-integrin—that also allows cells to attach , suggesting LINKIN may work in a similar way . LINKIN is found in many animals , so Kato et al . searched for proteins that can interact with it in human cells . This search revealed three proteins that can interact with LINKIN and are required for the cells to move together . Two of the proteins control elements of the internal scaffolding of the cell: this scaffolding , which is known as the cytoskeleton , is involved in moving the cells . The experiments suggest that LINKIN coordinates the process of binding together with the changes in the cytoskeleton that are needed to allow the cells to move as one . The next challenge is to understand how LINKIN changes the internal program of the cells to achieve this . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2014 | LINKIN, a new transmembrane protein necessary for cell adhesion |
Flying squirrels are the only group of gliding mammals with a remarkable diversity and wide geographical range . However , their evolutionary story is not well known . Thus far , identification of extinct flying squirrels has been exclusively based on dental features , which , contrary to certain postcranial characters , are not unique to them . Therefore , fossils attributed to this clade may indeed belong to other squirrel groups . Here we report the oldest fossil skeleton of a flying squirrel ( 11 . 6 Ma ) that displays the gliding-related diagnostic features shared by extant forms and allows for a recalibration of the divergence time between tree and flying squirrels . Our phylogenetic analyses combining morphological and molecular data generally support older dates than previous molecular estimates ( ~23 Ma ) , being congruent with the inclusion of some of the earliest fossils ( ~36 Ma ) into this clade . They also show that flying squirrels experienced little morphological change for almost 12 million years .
Flying squirrels ( Sciurinae , Pteromyini ) are the only group of gliding mammals to have achieved a significant diversity ( 52 species in 15 genera ) and wide geographical distribution across Eurasia and North America ( Koprowski et al . , 2016 ) . They have been classically regarded as a distinct subfamily among the Sciuridae ( McKenna and Bell , 1997; McLaughlin , 1984; Simpson , 1945 ) , and even sometimes considered a separate family derived from a different group than the remaining sciurids ( De Bruijn and Ünay , 1989; Forsyth Major , 1893; Mein , 1970 ) . The fact that presumed fossil flying squirrels are at least as old as ( or maybe even older than ) the oldest tree squirrels ( 36 . 6 – 35 . 8 Ma ) may support the latter hypothesis . However , flying squirrels are currently recognized as a monophyletic clade , as supported by a set of synapomorphies in the wrist ( Thorington , 1984 ) . The carpal anatomy of flying squirrels is unique , being related to the structures that support the patagium and their particular gliding position , which is different from that of all other gliding mammals ( Thorington , 1984; Thorington and Darrow , 2000 ) . Molecular phylogenies indicate that flying squirrels ( tribe Pteromyini ) are nested within tree squirrels ( subfamily Sciurinae ) and likely diverged as recently as the latest Oligocene–early Miocene ( 23 ± 2 . 1 Ma ) ( Fabre et al . , 2012; Mercer and Roth , 2003; Steppan et al . , 2004 ) . Notwithstanding , the pteromyin fossil record suggests a much older split . Indeed , one of the earliest sciurids , Hesperopetes thoringtoni from the late Eocene ( 36 . 6 – 35 . 8 Ma ) of North America , has been related to the lineage leading to flying squirrels according to dental morphology ( Emry and Korth , 2007 ) . In the light of molecular results , it was conceded that Hesperopetes unlikely represented a pteromyin and was not assigned to any squirrel subfamily ( Emry and Korth , 2007 ) . On the other hand , this genus appears to have been closely related to Oligopetes ( Emry and Korth , 2007 ) , an earliest Oligocene ( ca . 34 – 31 Ma ) purported flying squirrel from Europe and Pakistan ( Cuenca Bescós and Canudo , 1992; De Bruijn and Ünay , 1989; Heissig , 1979; Marivaux and Welcomme , 2003 ) . Hesperopetes is last recorded during the earliest Oligocene ( Orellan; Korth , 2017 ) , coinciding with the oldest record of Sciurion ( Bell , 2004 ) , yet another alleged flying squirrel . Isolated cheek teeth are the only material available for all these taxa , which have been related to flying squirrels exclusively based on dental morphology . In fact , the whole fossil record of flying squirrels almost exclusively consists of isolated cheek teeth and a few mandibular and maxillary fragments . Unfortunately , dental features commonly used to recognize flying squirrels are not unique but also present in other sciurids ( Thorington et al . , 2005 ) , so it is uncertain if any of the extinct ‘flying’ squirrels belonged to this group . Furthermore , if any of the oldest ( late Eocene–early Oligocene ) forms truly represented a pteromyin this would imply a discrepancy of more than 10 Myr between molecular and paleontological data . Contrary to dental material , postcranial remains do show diagnostic characters of the pteromyins ( Thorington , 1984; Thorington et al . , 2005; Thorington and Darrow , 2000 ) . Therefore , they are of utmost importance to clarify the assignment of extinct ‘flying’ squirrels and calibrate their divergence date from other sciurids . Yet , these have not been described and are rarely preserved in the fossil record . Here we report a remarkably complete skeleton of a Miocene squirrel that displays the gliding-related diagnostic features shared by extant pteromyins and allows for a recalibration of the time of origin and diversification of the group . The fossil record of ‘flying’ squirrels is further discussed in the light of this new finding and the results of our phylogenetic analyses .
The described partial skeleton ( IPS56468; Figure 1 , Videos 1 , 3D model in Supplementary file 1 ) was recovered at Abocador de Can Mata site ACM/C5-D1 ( els Hostalets de Pierola , Catalonia , Spain; see Materials and Methods ) , with an estimated age of 11 . 63 Ma ( Alba et al . , 2017 ) . The recovered remains were found partly articulated ( Figure 2—figure supplement 1 ) and comprise more than 80 complete and fragmentary bones including the skull ( Figure 6—figure supplement 1 ) and elements of the fore- and hindlimbs ( Figures 2–5 , Figure 2 , Table 1 ) . Additional material , including a second cranium ( Figure 6—figure supplement 2 ) , has been recovered from the same horizon and other roughly coeval ACM localities ( Table 2 ) . The specimens are assigned to Miopetaurista neogrivensis based on diagnostic cheek tooth morphology ( Figure 3; for detailed description and comparisons of cheek teeth morphology see Appendix 3 . 1 ) . In the ACM localities a second genus of ‘flying’ squirrel , Albanensia , is recorded , but Miopetaurista is clearly distinguished by its larger size , and several morphological features . The diagnostic characters of M . neogrivensis comprise: its large size; the presence of a complete entolophid and the frequent occurrence of a short mesolophid in the lower molars; and the large mesostyle in the P4 ( Casanovas-Vilar et al . , 2015; Mein , 1970 ) . Miopetaurista neogrivensis has only been reported from La Grive L5 ( type locality ) and L3 in France , from Bellestar ( Seu d’Urgell Basin , also in Catalonia ) , and from several sites from the Vallès-Penedès Basin ( Casanovas-Vilar et al . , 2015 ) . This species is extremely rare , being represented by just a few isolated cheek teeth in most of the Vallès-Penedès sites . Extant and fossil flying squirrels have been classified into different groups according to the complexity of dental morphology ( Mein , 1970 ) . The cheek teeth of Miopetaurista show a simple occlusal pattern , with enamel wrinkling only in the lower molars and no additional lophules ( Figure 3 , Appendix 3 . 1 ) . This pattern clearly differs from the more complex one of other large-sized flying squirrels , such as Aeretes and Petaurista ( Mein , 1970; Thorington et al . , 2002 ) . Therefore , Miopetaurista has been included within the group that comprises Aeromys and the small-sized flying squirrels , which do show simple dental patterns ( Mein , 1970 ) . However , our phylogenetic analyses ( see below ) show that M . neogrivensis is the sister taxon of extant Petaurista , a genus that would belong to a completely different group according to dental classification ( Mein , 1970 ) . Considering dental morphology Petaurista is assigned to a group characterized by its complex dental pattern with additional transverse lophules which would also comprise the genera Aeretes , Belomys , Eupetaurus and Trogopterus , among others ( Mein , 1970 ) . This clearly illustrates that dental characters , although useful to diagnose the different species and genera , should not warrant high consideration for disentangling the phylogenetic relationships between flying squirrels . Among the recovered postcranial material , a complete scapholunate and the dorsal end of the pisiform ( Figure 4 and Video 2 ) are the most diagnostic elements of pteromyins , because they form the functional complex associated with the extension of the gliding membrane ( Thorington , 1984; Thorington et al . , 2005; Thorington et al . , 2002; Thorington and Darrow , 2000 ) . The styliform cartilage , which supports the patagium in all members of the group , attaches to the pisiform and is extended when the wrist is radially abducted and dorsiflexed ( Thorington , 1984; Thorington and Darrow , 2000 ) . The pisiform of M . neogrivensis displays an elevated process for the articulation with the scapholunate ( Figure 4 and Video 2 ) . This is characteristic of pteromyins , serving as a stabilizer of the styliform cartilage , whereas in other squirrels this bone articulates only with the triquetrum and the distal end of the ulna . Moreover , in the scapholunate of M . neogrivensis , the articular surface for the radius is much more convex than in tree squirrels , thus resembling the flying squirrel condition , which enables a greater radial abduction . Therefore , the proximal wrist joint morphology of M . neogrivensis indicates that this species belongs to the pteromyin clade and provides the oldest evidence of gliding locomotion in sciurids ( see also Appendix 3 . 3 ) . The latter is further confirmed by other postcranial adaptations shared with extant pteromyins ( Thorington et al . , 2005 ) , including the elongated and slender limb bones with reduced muscular attachments ( Figure 5 and Figure 5—figure supplement 1 ) , which enhance joint extension during gliding ( Thorington et al . , 2005 ) , as well as the elongated lumbar vertebrae ( Figure 2; see Appendix 3 . 3 for a detailed description and comparisons of the postcranial elements ) . The elongation of lumbar vertebrae and limbs determines the size and shape of the patagium and dictates important aerodynamic features , such as the decreased wing loading of flying squirrels ( Thorington and Heaney , 1981; Thorington and Santana , 2007 ) . Based on morphological ( Thorington et al . , 2002 ) and molecular data ( Mercer and Roth , 2003 ) , flying squirrels are divided into two distinct subtribes: Pteromyina , comprising large-sized forms , and Glaucomyina , for the small-sized ones ( Thorington and Hoffmann , 2005 ) . The skeleton of M . neogrivensis morphologically resembles that of the Pteromyina , further being comparable in size to their largest representatives . Body mass was estimated by means of an allometric regression of body mass vs . skull length in extant sciurids ( see Materials and Methods ) , resulting in 1339 g ( 50% confidence intervals 1116 – 1606 g thus being in the range of most species of the extant giant flying squirrel Petaurista ( about 1200 – 2000 g; Thorington et al . , 2012 ) . The long bones are almost indistinguishable of Petaurista . The skull , which was virtually reconstructed from two well-preserved specimens ( Figure 6 , Figure 6—figure supplement 1-2 , Video 3 , Table 3 ) , is strikingly similar in size and morphology to that of the other large-sized flying squirrels , particularly Aeromys and Petaurista ( for a detailed morphological description of the skull and comparisons see Appendix 3 . 2 ) . These genera are characterized by their short and wide rostrum , moderately inflated bullae and relatively wide posterior region of the skull . Other morphological similarities include the robust and long postorbital process that partially encloses the orbit , the well-developed jugal process in the zygomatic arch and the presence of two septa in the tympanic cavity ( Video 4; Appendix 3 . 2 ) . Most of the smaller flying squirrels show more elongate muzzles , slender or shorter postorbital processes and , in some cases , a higher number of transbullar septa . The proximal carpal bones of M . neogrivensis not only unambiguously indicate that it is a flying squirrel , but also allow assigning it to the Pteromyina ( Figure 4 and Video 2 ) . The pisiform displays a distinct spur ( triquetral process ) that fits between the palmar surfaces of the scapholunate and the triquetrum . This process is completely lacking in the Glaucomyina ( Thorington et al . , 2002; Thorington and Darrow , 2000 ) ( Figure 4 and Video 2; see also Appendix 3 . 3 ) . Both subtribes are also distinguished by the origin of the tibiocarpalis muscle , which runs from the ankle to the tip of the styliform cartilage , defining the edge of the patagium . In the Glaucomyina the tibiocarpalis originates from a tuberosity on the distal tibia which is lacking in M . neogrivensis and the Pteromyina ( Thorington et al . , 2002 ) . In the latter , the tibiocarpalis originates from the metatarsals instead . The assignment of M . neogrivensis to the subtribe Pteromyina is further confirmed by a total evidence phylogenetic analysis combining morphological and molecular data ( see Materials and Methods ) . This analysis relied on 35 extant species of sciurids plus Aplodontia used as outgroup ( both with molecular and morphological data ) and two fossils ( M . neogrivensis and the oldest-known tree squirrel , Douglassciurus jeffersoni , both represented only by morphological data; see also Appendices 1 . 1 and 2 ) . Our analysis strongly supports M . neogrivensis as the sister taxon of the Petaurista spp . clade and indicates a divergence date between flying and tree squirrels ranging from the late Eocene to the late Oligocene ( 95% highest posterior density [HPD] interval 36 . 5 – 24 . 9 Ma; Figure 7 ) . Such range is congruent with previous molecular estimates ( Fabre et al . , 2012; Mercer and Roth , 2003 ) but also supports older dates , as old as the oldest records of purported pteromyins ( Emry and Korth , 2007 ) . The Pteromyina and Glaucomyina would have diverged between the late Oligocene–early Miocene ( HPD interval 27 . 1 – 18 . 1 Ma; Figure 7 ) . We independently tested the estimates of Pteromyini/Sciurini divergence and the onset of Pteromyini crown diversification by means of a node dating analysis ( see Materials and Methods ) of the extant Sciurinae using two different calibration points , one for each tribe ( Figure 7—figure supplement 1 , Table 4; see also Appendix 1 . 2 ) . Estimates for many nodes are somewhat younger using this alternative approach , but mostly overlap with the younger half of HPD intervals of the total evidence analysis ( Figure 7 ) . The age range for the Sciurini/Pteromyini split spans from the mid Oligocene to early Miocene ( HPD interval 30 . 6 – 17 . 4 Ma; Figure 7—figure supplement 1 ) . Therefore , divergence estimates derived using independent methods overlap for the late Oligocene .
The two independent phylogenetic analyses estimate divergence dates ranging from 36 . 5 to 17 . 4 Ma ( late Eocene–early Miocene; Figure 7 and Figure 7—figure supplement 1 ) . These broad ranges are congruent with a previous molecular phylogenetic analysis of the sciurids ( Mercer and Roth , 2003 ) , which placed the split between tree and flying squirrels near the Oligocene/Miocene boundary ( 23 ± 2 . 1 Ma ) , although allowing for substantially older dates . Total evidence analysis in particular provides older estimates ( 36 . 5 – 24 . 9 Ma ) than node dating ( 30 . 6 – 17 . 4 Ma ) , but these still marginally overlap with those of previous molecular results . The different results may arise from the different selection of calibration points ( see Table 4 ) . In that study ( Mercer and Roth , 2003 ) the age of Douglassciurus ( ca . 36 Ma ) was assigned to the base of the sciurid crown radiation ( i . e . the origin of extant major clades ) to calibrate the phylogenetic tree whereas the root of our phylogenetic trees is calibrated using multiple points ( see Materials and Methods ) . A few molecular phylogenetic studies ( Montgelard et al . , 2008; Tapaltsyan et al . , 2015 ) , generally dealing with the whole rodent order and including only a few sciurid genera , have provided older ages for the divergence of flying squirrels dating back to the late Eocene and earliest Oligocene ( 34 . 5 – 30 . 9 Ma ) . Again , this may be attributed to the different selection of calibration points , which in most cases consider paleontological data from other rodent groups . Finally , rodent diversification has been analyzed using a molecular supermatrix that included 98% of extant squirrel genera ( Fabre et al . , 2012 ) . This study found a late Oligocene divergence date between pteromyins and sciurins , which is perfectly congruent with our results . Total evidence 95% highest posterior density ( HPD ) interval for the pteromyin divergence is very broad ( 36 . 5–24 . 9 Ma; Figure 7 ) , being consistent with the range of Oligopetes ( Cuenca Bescós and Canudo , 1992; De Bruijn and Ünay , 1989; Heissig , 1979; Marivaux and Welcomme , 2003 ) , the oldest records of Sciurion ( Bell , 2004 ) and even the earliest ( late Eocene ) occurrences of Hesperopetes ( Korth , 2017 ) ( Figure 8 ) . In contrast , node dating analysis estimates a younger HPD interval that would exclude the older records of these genera ( 30 . 6 – 17 . 4 Ma; Figure 7—figure supplement 1 ) . According to our results , late Eocene to early Oligocene alleged flying squirrels might indeed belong to this group , but this is only supported by total evidence analysis ( Figure 7 ) . Therefore , our analyses are not conclusive to this regard and further fossil data are required to elucidate the phylogenetic position of these older ‘flying’ squirrels . Although allowing for older and younger ages , the independently derived estimates are generally older than previous molecular results ( Mercer and Roth , 2003 ) and overlap for the late Oligocene , which should be considered the most likely time for pteromyin divergence . Except for the partial skeleton of Miopetaurista neogrivensis described here , no diagnostic postcranial material has been recovered for any other extinct ‘flying’ squirrel , so their assignment to the pteromyins is doubtful . Here we discuss the fossil record of purported pteromyins in the light of our phylogenetic results , but in these instances our conclusions depend of the correctness of the tribe assignment . Our analyses indicate that the initial diversification of flying squirrels into the two extant subtribes ( Glaucomyina and Pteromyina ) occurred between the latest Oligocene and the middle Miocene , with estimates derived from both methods overlapping for the early Miocene ( Figure 7 and Figure 7—figure supplement 1 ) . Such time interval coincides with the almost simultaneous earliest records of new genera of ‘flying’ squirrels in both North America and Europe . In Europe , these include the small-sized genus Blackia , which is first recorded near the Oligocene/Miocene transition ( biozone MP30 ) ( Engesser and Storch , 2008 ) at around 23 . 3 – 23 . 0 Ma ( Figure 8 ) . According to dental morphology this genus appears to be closely related to the older North American Sciurion ( Skwara , 1986 ) , so in case their pteromyin affinities were confirmed this would argue for a North American origin of the group . Besides Blackia three additional genera are recorded in the European early Miocene ( De Bruijn , 1999 ) : Aliveria , Miopetaurista and Neopetes ( Figure 8 ) . These occurrences date back to biozone MN3 , corresponding to the early Miocene , yielding a relative age of 20 . 0 – 18 . 0 Ma . A species of the extant genus Hylopetes has been erected based on material from site of Oberdorf ( Austria ) , correlated to the MN4 ( 18 . 0 – 17 . 0 Ma ) ( De Bruijn , 1998 ) . Such occurrence would pull back the range of this genus into the early Miocene and has been taken for granted in some recent molecular studies ( Lu et al . , 2013 ) even though it blatantly disagrees with previous molecular results ( Mercer and Roth , 2003 ) . Indeed , some paleontologists have argued that the characters justifying the ascription of the Austrian and other material to Hylopetes are symplesiomorphies shared with many other flying squirrel genera ( e . g . Petinomys , Glaucomys ) and assign this material to the extinct ‘flying’ squirrel genera Neopetes and Pliopetes ( Casanovas-Vilar et al . , 2015; Daxner-Höck , 2004 ) . In North America , aside from Sciurion the larger-sized genus Petauristodon is also present from the late Oligocene to the late Miocene ( Goodwin , 2008 ) ( Figure 8 ) . While Sciurion is already known since the early Oligocene ( 33 . 5-33 . 0 Ma ) ( Bell , 2004 ) , the oldest record of Petauristodon is a single molar dated between 25 . 9 and 23 . 8 Ma from the John Day Formation of Oregon ( Korth and Samuels , 2015 ) . Subsequent records date back to the early Miocene ( ca . 19 Ma; Goodwin , 2008 ) . Interestingly , the specimens ascribed to Sciurion and Petauristodon were previously referred to the European genera Blackia and Miopetaurista , respectively ( Goodwin , 2008 ) . Finally , in Asia the genera Meinia ( Qiu , 1981 ) , Parapetaurista ( Qiu , 1981 ) and Shuanggouia ( Qiu and Liu , 1986 ) have been reported from the early Miocene ( Shanwangian ) Shanwang and Xiacaowan formations of North and Eastern China , with an estimated age of ca . 18 – 16 Ma or slightly younger ( Qiu et al . , 2013b ) ( Figure 8 ) . Even though fossil evidence can only justify the ascription of Miopetaurista to the pteromyins , it is worth noting that Europe records highest ‘flying’ squirrel diversity at the time ( Figure 8 ) , which has led to the suggestion that the group may well have originated there and immediately dispersed into Asia and North America near the Oligocene/Miocene transition ( Lu et al . , 2013 ) . Notwithstanding , this hypothesis is challenged by the fossil record of ‘flying’ squirrels since the oldest occurrences are in North America . If certain Oligocene forms can be ultimately assigned to the pteromyins this would confirm an opposed model with an early origin and initial diversification in North America and a later dispersal into Eurasia by the latest Oligocene . Initial flying squirrel diversification and dispersal across the Northern Hemisphere coincided with a period of high mean global temperatures during the early Miocene that peaked between 17 to 15 Ma , during the so-called Mid-Miocene Climatic Optimum ( Zachos et al . , 2001 ) ( Figure 8 ) . Humid warm-temperate broadleaf and mixed forests , resembling those existing in the southeastern coast of Asia , characterized the mid latitudes in Eurasia and North America ( Pound et al . , 2012 ) . These forests provided a suitable habitat for flying squirrels and would have contributed to their initial radiation and dispersal into different continents . Our phylogenetic analyses show that some extant genera , such as Petaurista , Pteromys and Glaucomys , diverged approximately at that time ( Figure 7 and Figure 7—figure supplement 1 ) , thus agreeing with previous molecular results ( Arbogast et al . , 2017; Mercer and Roth , 2003 ) . The warm early Miocene phase was followed by a gradual cooling and the reestablishment of permanent major ice sheets on Antarctica by about 10 Ma ( Zachos et al . , 2001 ) . Warm-temperate forests were still dominant in the mid latitudes during the middle and the beginning of the late Miocene . However , their distribution became more restricted , particularly in Western North America and Central Asia , where they began to be replaced by cooler and drier biomes ( Pound et al . , 2012 ) . ‘Flying’ squirrels are particularly diverse in Europe during this interval , with as many as five different species co-occurring in a single site ( Casanovas-Vilar et al . , 2015 ) . In addition , certain genera such as Miopetaurista , Pliopetaurista and Albanensia , are widely distributed ( Casanovas-Vilar et al . , 2015 ) ( Figure 8 ) . Miopetaurista and Pliopetaurista are also know from the beginning of the late Miocene of Amuwusu ( Inner Mongolia , China ) ( Qiu et al . , 2013b ) , whereas Albanensia or a closely related form would occur in the latest Miocene of Shihuiba ( Yunnan , China ) ( Qiu , 2002 ) . Our phylogenetic analyses show that most extant Asian flying squirrel genera diverged during the interval from 15 to 10 Ma ( Figure 7 and Figure 7—figure supplement 1 ) , thus coinciding with this time of high diversity and geographic range extension . On the contrary , ‘flying’ squirrel diversity in North America stayed at very low levels , with only the genera Petauristodon and Sciurion known from the middle and the beginning of the late Miocene ( Goodwin , 2008 ) ( Figure 8 ) . Open habitats ( mixed scrubland-grassland ) housing a remarkable diversity of grazing mammals became widespread in central North America at that time , even though C4-dominated grasslands did not spread until the late Miocene ( ca . 7 Ma ) ( Strömberg , 2011 ) . These unsuitable habitats would have hampered the radiation of flying squirrels there . Throughout the later Miocene , warm-temperate forests continued reducing their extension and intermingled with drier and cooler biomes in Eurasia ( Pound et al . , 2012 ) . In the Mediterranean regions , open habitats corresponding to woodlands and scrublands , already occurred in the southern Iberian Peninsula and Turkey since the middle Miocene ( Fauquette et al . , 2007; Strömberg , 2011 ) , but would increase their extension at that time . Furthermore , the characteristic Mediterranean rainfall seasonality ( summer drought ) appeared during the Pliocene ( Suc , 1984 ) . In Central Europe , deciduous forests increasingly replaced warm-temperate mixed ones , which became restricted to coastal areas of the Mediterranean and the Paratethys ( Mosbrugger et al . , 2005 ) . In Asia , major physiographical changes such as the uplift of the Himalaya-Tibetan Plateau affected atmospheric circulation and resulted in increased aridification of large areas of the continent ( Zhisheng et al . , 2001 ) . Carbon stable isotope records of soil carbonates and ungulate tooth enamel indicate that C4 grasses rise to dominance in ecosystems between 8 – 7 Ma in Pakistan ( Cerling et al . , 1993; Strömberg , 2011 ) and later in China ( Strömberg , 2011 ) . All these environmental changes had profound effects in flying squirrel diversity and biogeography . The group became increasingly rarer over much of its former range and several genera ( Forsythia , Albanensia , Miopetaurista , Blackia ) disappeared during the late Miocene and the Pliocene ( De Bruijn , 1999 ) ( Figure 8 ) . In Asia , fossil occurrences during the late Miocene are mostly confined to the southeast ( Yunnan province in China , Thailand ) ( Lu et al . , 2013 ) . In North America , only the species Miopetaurista webbi is known by very scarce remains from the latest Miocene and the Pliocene of Florida ( Goodwin , 2008 ) ; Robertson , 1976 ) , being the only record of this genus outside Eurasia ( Figure 8 ) . It is not surprising that this last record of a large-sized flying squirrel in North America comes from Florida , an area that is still characterized by a humid subtropical climate with abundant densely forested areas . The extant American flying squirrel genus Glaucomys , which today inhabits temperate deciduous forests and boreal coniferous forests , is recorded for the first time already in the Pleistocene ( Ruez , 2001 ) although it would have diverged significantly earlier according to molecular results ( Figure 7 and Figure 7—figure supplement 1; Arbogast et al . , 2017; Mercer and Roth , 2003 ) . The Pleistocene records of flying squirrels mostly correspond to extant genera and species ( Lu et al . , 2013 ) ( Figure 8 ) . This is again congruent with our results , which show that most extant species had already diverged during the Pliocene and some even at the latest Miocene ( Figure 7 and Figure 7—figure supplement 1 ) . During the Pleistocene , Glaucomys is the only flying squirrel known from North America . In Asia , the genera Aretes , Belomys , Petaurista , Pteromys and Trogopterus have been recorded from several sites of south and eastern China ( Lu et al . , 2013 ) . In addition , Belomys has also been reported from Thailand and Pteromys from Japan ( Lu et al . , 2013 ) . The Pleistocene records of all these genera are located within their current geographical range and generally correspond to extant species . In the European Pleistocene , flying squirrels are represented by the genera Neopetes , Petauria and Pliopetaurista ( Figure 8 ) . Quite surprisingly , there are no fossil records of Pteromys , the only flying squirrel genus still extant in Europe . Neopetes comprises species formerly included in the extant genus Hylopetes ( Jackson and Thorington , 2012 ) . Petauria is a large-sized flying squirrel known from middle Pleistocene fissure fillings and cave deposits of Germany and Poland . The available cheek teeth show several striking morphological similarities with the extant Petaurista , including: smooth enamel with numerous lophules , particularly in the basin of the lower cheek teeth; prominent mesostylid in the lower molars; and presence of a well-developed postero-lingual re-entrant fold in the upper molars . We agree with the opinion of some authors that this genus is a junior subjective synonym of Petaurista ( Jackson and Thorington , 2012; Thorington et al . , 2005 ) . Therefore , the geographical range of Petaurista during the middle Pleistocene included Europe . Whether the genus originated in Europe or Asia cannot be resolved , since its closest relative , Miopetaurista , shows a similarly broad geographical range . It is worth noting that Miopetaurista is remarkably similar to Petaurista , to the point that their postcranial skeleton is virtually identical , even in specificities such as the more reduced lateral epicondylar ridge of the humerus or the wider patellar groove in the femur as compared to other ( generally smaller ) flying squirrels ( for more detailed comparisons see Appendix 3 . 3 ) . Cranial morphology evidences a close affinity between both genera , with details such as the development of the postorbital processes or the short and wide rostrum , being surprisingly similar . The only differences lie in cheek tooth morphology . Miopetaurista shows a relatively simple morphology , with faint enamel in the upper cheek teeth and no additional longitudinal lophules . In contrast , Petaurista presents relatively higher-crowned teeth and a much more complex morphology with additional longitudinal lophules ( Mein , 1970; Thorington et al . , 2002 ) . In addition , the upper cheek teeth show a well-defined distolingual flexus , a feature that only occurs in the M3 in Miopetaurista . Dental morphological differences apart , it is worth noting that large-sized flying squirrels must be regarded as a very conservative group , having experienced little morphological changes since the late middle Miocene . Concerning the genus Petaurista , our analyses recognize the three main species groups ( Figure 7 and Figure 7—figure supplement 1 ) that had already been recognized by previous molecular analyses ( Li et al . , 2013 ) . The first group solely includes Petaurista leucogenys , a species endemic to Japan . It is found to have diverged during the late Miocene ( ca . 11 – 6 Ma ) , significantly earlier than the remaining species . The second clade includes Petaurista petaurista and Petaurista philippensis , the most widely distributed species in the genus , together with Petaurista xanthotis , endemic to south-central China and Tibet . Finally , the third clade comprises species mostly occurring in southern and central China ( Petaurista albiventer , Petaurista alborufus , Petaurista yunanensis ) and Indonesia ( Petaurista elegans ) , with Petaurista hainana , which is endemic to Hainan Island ( southeastern China ) . These two Petaurista subclades diverged during the late Miocene and estimates for the divergence of most species range from the late Pliocene to the early Pleistocene , that is coinciding with global cooling after 3 Ma and the start of northern hemisphere glaciation ( Zachos et al . , 2001 ) . Other flying squirrel genera , such as Hylopetes , Petinomys and Glaucomys show a similar pattern , with estimates for the divergence of many extant species ranging from the late Pliocene to the early Pleistocene . In the present , flying squirrels are widely distributed across the northern hemisphere , but only one genus lives in North America ( Glaucomys ) and a single species ( Pteromys volans ) occurs in Europe . Very few species are endemic to central and northern Asia . In contrast , flying squirrels are diverse in the tropical and subtropical forests of the Indo-Malayan region , which apparently acted both as refugia and diversification center ( also for tree squirrels ) since the late Miocene ( Lu et al . , 2013; Mercer and Roth , 2003 ) . Miopetaurista neogrivensis is the oldest unquestionable flying squirrel and dates back to the middle/late Miocene boundary ( 11 . 6 Ma ) . Its diagnostic wrist anatomy indicates that the two subtribes of flying squirrels had already diverged at that time . Moreover , this new fossil allows for a recalibration of flying squirrel time of origin and diversification , generally providing somewhat older estimates than previous molecular analyses . These differ according to the phylogenetic method used , total evidence analysis estimates an interval of 36 . 6 – 24 . 9 Ma while node dating results in a younger estimate of 30 . 6 – 17 . 4 Ma . Therefore , we cannot rule out that at least some of the oldest ( ca . 36 Ma ) fossils tentatively identified as flying squirrels may indeed belong to this group . However , the estimates of both independent phylogenetic approaches overlap for the late Oligocene ( 31 – 25 Ma ) , which should be considered the most likely interval for flying squirrel divergence . The two flying squirrel subtribes are found to have diverged during the early Miocene ( 22 – 18 Ma ) while most extant genera would do so during the Miocene , although they are not recorded until the Pleistocene . Miopetaurista neogrivensis is estimated to have diverged from Petaurista spp . , its sister taxon , between 18 . 8 – 12 . 4 Ma , the oldest boundary overlapping with the earliest record of the genus Miopetaurista ( 18 – 17 Ma ) . Perhaps not surprisingly , the skeletons of both genera show little differences . Sciurids are often regarded as a morphologically conservative group and flying squirrels are no exception having experienced few morphological changes for almost 12 million years . 3D surface models of the described material of M . neogrivensis ( reconstructed skeleton , reconstructed skull , carpal bones ) are available at MorphoBank https://morphobank . org/index . php/Projects/ProjectOverview/project_id/3108 . All other data files are provided as supplementary data .
The partial skeleton of Miopetaurista neogrivensis ( IPS56468 ) and all other described material ( Figure 2—figure supplement 1-2 ) is housed in the Institut Català de Paleontologia Miquel Crusafont ( ICP ) in Sabadell ( Barcelona , Catalonia , Spain ) . IPS56468 was found in 2008 during paleontological surveillance of excavation works at the Can Mata landfill ( Abocador de Can Mata [ACM] , els Hostalets de Pierola , Catalonia , Spain ) . It was unearthed from a rich fossiliferous horizon ACM/C5-D1 ( ACM Cell 5 , sector D , locality 1 ) in two different blocks with some elements in anatomical connection ( Figure 2; Figure 7—figure supplement 1 ) . The excavation of ACM/C5-D1 provided additional material of M . neogrivensis ( Table 2 ) . A partial , dorsoventrally crushed skull ( IPS88677 ) was also recovered from stratigraphically close locality ACM/C8-Af ( ACM Cell 8 , sector A , locality f ) ( Alba et al . , 2017 ) . The ACM series is located in the Vallès-Penedès Basin , an elongated half-graben filled mostly by continental deposits during the Miocene , which are rich in vertebrate fossils ( Figure 9 ) . The ACM composite series ranges from ca . 12 . 6 to 11 . 5 Ma , the age of the paleontological localities being well constrained thanks to high-resolution litho- , bio- and magnetostratigraphical data ( Alba et al . , 2017 ) . ACM/C5-D1 is correlated to chron C5r . 2n ( 11 . 657 – 11 . 592 Ma ) ( Alba et al . , 2017 ) , and its interpolated age is 11 . 64 Ma ( Alba et al . , 2017 ) . Dental terminology , abreviations and measurement methods for sciurid cheek teeth follow Casanovas-Vilar et al . , 2015 and references therein . A virtual three-dimensional ( 3D ) cranial reconstruction of M . neogrivensis was performed based on both IPS56468h and IPS88677 ( Figure 6—figure supplement 1-2 , Video 3 ) . The specimens were analyzed separately by microfocus X-ray computed tomography ( μCT ) at the Multidisciplinary Laboratory of the ‘Abdus Salam’ International Centre of Theoretical Physics ( Trieste , Italy ) , using a system specifically designed for the study of archaeological and paleontological materials ( Table 5 ) . Raw data from each scanning were imported ( as stack of TIFF 8-bit files ) to Avizo 7 . 0 and Rhinoceros 5 . 0 for segmentation , repositioning , mirroring and visualization . Each cranial bone or bone fragment was segmented virtually removing the surrounding matrix using semiautomatic thresholding tools and obtaining individual 3D digital models . Up to 64 3D models were generated for both specimens prior to repositioning and mirroring them to assemble an almost complete skull . Reconstruction primarily relied upon IPS56468h , but used 3D bone models from IPS88677 for elements that were particularly damaged or missing from the former ( Table 3 ) . Due to poor preservation of some fragments , it was necessary to import some of the 3D models to Rhinoceros 5 . 0 to repair the meshes , split the model to keep only the well-preserved regions , and use , if available , the mirrored region from the other side of the same skull , or alternatively take it from IPS88677 ( mirrored if necessary ) . The 3D models were repositioned using Avizo 7 . 0 based on bilateral symmetry and fracture congruence . An almost complete 3D virtual skull model was finally assembled using a total of 41 3D models of cranial bones ( Figure 6 , Table 3 ) . Besides cranial restoration , the two mandibles belonging to IPS56468 were scanned using a 3D desktop laser scanner ( NextEngine ) at high definition and with a dimensional accuracy of 0 . 13 mm ( Macro Mode ) . Mandibles were exported to Rhinoceros 5 . 0 to repair meshes and were repositioned and joined to the resulting skull model using Avizo 7 . 0 . In addition , the posterior region of the skull of the extant flying squirrel Petaurista petaurista was scanned to compare its inner ear morphology to that of M . neogrivensis ( Video 4 ) . The specimen is kept at the Naturalis Biodiversity Center ( NBC , Leiden , the Netherlands ) and was scanned at this institution with a µCT scanner , using a Skyscan system model 1172 ( Bruker company , Belgium ) ( Table 5 ) . The 3D surface of the postcranial skeleton of M . neogrivensis was digitized by means of photogrammetry using a variable number of sets of two-dimensional images . Within each set , individual images were captured with a minimum overlap of 66% using a digital single-lens reflex camera with a 50 mm f/2 lens . Aperture and focus were constant to ensure consistency between sets of photographs . Each set was then aligned and scaled to create a 3D model using the photogrammetric software Agisoft PhotoScan Professional . Individual surface models were exported to the 3D software Pixologic ZBrush version 4R7 , which was used to merge the different meshes and improve the quality of the final merged 3D surface . Missing areas were modeled according to mirrored bones when available . Missing bone elements and broken areas in the available material were reconstructed using P . petaurista as reference . The morphology of the axial skeleton follows that of P . petaurista but has been modified according to the proportions of the recovered vertebrae . Finally , carpal bones of extant Sciurus vulgaris , P . petaurista and Hylopetes sagitta , as well as those from M . neogrivensis were digitized for visualization and comparison . All extant specimens are stored in the collections of the NBC and were scanned there using a µCT scanner Skyscan system model 1172 ( Table 5 ) . Raw CT data from each scanning were imported directly ( as stack of TIFF 16-bit files ) to Avizo 7 . 0 . Body mass ( BM , in g ) was estimated using an allometric regression of BM vs . skull length ( SL , in mm ) , which is the estimator most tightly correlated with BM in extant rodents ( Bertrand et al . , 2016 ) . SL was measured in M . neogrivensis from the 3D reconstruction based on crania IPS56468h and IPS88677 simultaneously ( 69 . 80 mm ) . The allometric regression was performed using log-transformed ( ln ) mean species data , as it is customary when predicting BM ( Ruff , 2003 ) . Sexes were treated separately ( when possible ) to avoid potentially confounding effects of body size dimorphism . The regression was computed for sciurids only to avoid biases due to allometric grade shifts among rodent families . SL sex-species means were calculated from published individual values ( Bertrand et al . , 2016 ) or measured by one of the authors ( I . C . V . ) . BM average data were computed from individual published values ( Bertrand et al . , 2016 ) or taken from the literature ( Hayssen , 2008; Thorington and Heaney , 1981; Zahler , 2001 ) . The ordinary least-squares regression method , selected as the most suitable for prediction ( Ruff , 2003; Smith , 1994 ) , yielded the following equation with SPSS v . 17: ln BM = 4 . 369 ln SL – 11 . 384 ( N = 33 , p<0 . 001 , r = 0 . 972 , SEE = 0 . 256 ) . The logarithmic detransformation bias was corrected using the quasimaximum likelihood estimator ( Ruff , 2003; Smith , 1993 ) : QMLE = 1 . 033 . 50% confidence intervals ( CI ) for the prediction were computed . To infer the phylogenetic placement of M . neogrivensis and assess its impact on the dating estimates of the phylogeny of flying squirrels , we performed a new phylogenetic analysis of tribe Pteromyini by means of two alternative approaches: total evidence ( Ronquist et al . , 2012 ) and node dating ( Ronquist et al . , 2016 ) . These approaches relied on the combination of morphological and molecular datasets , or on molecular datasets alone , respectively . For the molecular dataset , we retrieved from GenBank the combination of the four genes more frequently used in previous phylogenetic studies of the Sciuridae ( 12S , 16S , cytochrome b , and irbp ) ( Mercer and Roth , 2003 ) , which were downloaded for all Sciurinae species available in GenBank ( Supplementary file 2 ) . This dataset comprises 58 of the 89 extant Sciurinae species ( Koprowski et al . , 2016 ) and all genera but Biswamoyopterus . The Pteromyini are represented by 29 species , comprising 65% of extant diversity . We also obtained the sequence for Aplodontia rufa , sole extant member of the Aplodontiidae ( the sister group of the Sciuridae; Fabre et al . , 2012; Huchon et al . , 2002 ) , which was used as outgroup in the total evidence analysis . Each gene was aligned using two procedures: ribosomal coding genes were aligned by means of MAFFT v6 ( Katoh et al . , 2002 ) and protein coding genes were aligned using the translation alignment algorithm implemented in TranslatorX ( using MAFFT to align proteins ) ( Abascal et al . , 2010 ) . Poorly-aligned regions in the ribosomal coding genes were eliminated with Gblocks ( Castresana , 2000 ) under low stringency options ( Talavera and Castresana , 2007 ) . The morphological dataset consisted of 105 characters comprising the dentition , skull and postcranial skeleton of 36 extant taxa plus two fossil squirrels: M . neogrivensis and Douglassciurus jeffersoni ( Appendix 2; Supplementary file 3 ) . Douglassciurus is chosen because it is the earliest squirrel represented by abundant postcranial material and is currently recognized as an outgroup to all other sciurids ( Appendix 1 . 1 ) . The character list is mostly based in Thorington et al . , 2002 with numerous additions of diagnostic pteromyin characters from the limb bones after Thorington et al . , 2005 . Finally , a set of new characters was also added to resolve the relationships of A . rufa and D . jeffersoni with other taxa ( Appendix 2 , Supplementary file 3 ) . For the total evidence dating we used MrBayes v . 3 . 2 . 6 ( Ronquist et al . , 2012 ) . The analysis included 36 species for which we had both molecular and morphological data plus two fossil species with only morphological data . The molecular data consisted in a total of 3345 base pairs ( bps ) distributed in each gene as follows: 12S ( 505 bps ) , 16S ( 521 bps ) , cytb ( 1140 pbs ) , irbp ( 1179 pbs ) . The protein coding genes were split in codon positions and the most appropriate partitioning scheme and the model of molecular evolution for each partition were estimated by means of the software Partitionfinder ( Lanfear et al . , 2012 ) ( Table 6 ) . The five partitions derived from Partitionfinder were then concatenated with the previously described morphological dataset . We modeled node ages and tree topology using a fossilized birth and death process ( Gavryushkina et al . , 2017; Heath et al . , 2014; Stadler , 2010; Zhang et al . , 2016 ) with broad priors on speciation ( exp[10] ) , extinction ( beta[1 , 1] ) , and fossilization ( beta[1 , 1] ) ( Pyron , 2017; Zhang et al . , 2016 ) . The analysis also relied on a relaxed-clock with the independent gamma-rates ( IGR ) model ( Ronquist et al . , 2012 ) , with a broad prior for the variance increase parameter ( exp[10]; Pyron , 2017 ) . Following Pyron ( Pyron , 2017 ) we used 1/mean of the root age and exp ( 1/mean of the root age ) to generate a wide clock rate prior compatible with both morphological and molecular clocks . For the root age we used the interval 52 . 7–45 . 7 Ma as a uniform prior . The lower bound of this prior was informed by the most recent age estimate ( MRE ) of the oldest Aplodontiidae ( Table 4; see also Appendix 1 . 1 ) and we calculated the upper bound using the algorithm proposed by Hedman , 2010 , which calculates the probability distribution of the age of a clade , given the ages of the oldest fossil representatives of the outgroups of that clade . We used the upper limit of the 95% confidence interval for the age of the root as a plausible upper bound for the root . This was informed by the MRE for the minimum age of the oldest Gliridae ( 47 . 4 Ma , outgroup to Aplodontiidae+ Sciuridae ( Blanga-Kanfi et al . , 2009; Fabre et al . , 2012; Huchon et al . , 2002 ) and the MRE for the minimum age of Erlianomys combinatus , the oldest Myodonta ( 53 . 9 Ma , outgroup to Gliridae +Aplodontiidae + Sciuridae; Blanga-Kanfi et al . , 2009; Fabre et al . , 2012; Huchon et al . , 2002 ) ; Table 4 , Appendix 1 . 1 ) . The maximum age oldest rodent was used as the maximum possible age of the root ( 56 Ma; Table 4 , Appendix 1 . 1 ) . Partitions and nucleotide substitution models for the molecular data were estimated by means of Partitionfinder v . 2 . 1 . 1 ( Lanfear et al . , 2012 ) with linked branch lengths , a Bayesian Information Criterion ( BIC ) model of selection and a greedy search algorithm ( Table 6 ) . For the morphological data ( only variable characters included ) , we used a k-state Markov ( Mkv ) model ( Lewis , 2001 ) with a rate variation modeled by means of a discrete gamma model . We distinguished between ordered and unordered characters . We also used an alternative Bayesian program , BEAST v . 1 . 8 . 4 ( Drummond et al . , 2012 ) , to provide an independent estimate of the time of the Pteromyini/Sciurini divergence and the onset of Pteromyini crown diversification by means of node dating . This analysis included all species of Sciurinae available in GenBank ( 58 taxa ) and consisted in 3225 pbs distributed in each gene as follows: 12S ( 383 bps ) , 16S ( 523 bps ) , cytb ( 1140 pbs ) , irbp ( 1179 pbs ) . Protein coding genes were subsequently split in codon positions and we used Partitionfinder to determine the best set of molecular partitions and models of molecular evolution ( Table 6 ) . We estimated the tree in time units using an uncorrelated lognormal clock applied to each of the five partitions derived from Partitionfinder and two different calibration points ( one located in each tribe; Table 4 ) . Miopetaurista neogrivensis from ACM/C5-D1 ( 11 . 6 Ma ) provides a minimum age for the divergence between Petaurista and the remaining Pteromyina , while the oldest record of the genus Sciurus ( dating back to 13 . 6–10 . 3 Ma ) gives minimum age for the Sciurus/Tamiasciurus split ( Table 4 , Appendix 1 . 2 ) . Soft maxima for both calibrations were calculated as described in Hedman , 2010 . The soft maximum for the calibration point in the Pteromyina was informed by the MRE of the oldest Heteroxerus , ( 25 . 0 Ma ) , the earliest Xerinae ( sister group to Sciurinae ) , and by the oldest estimate of D . jeffersoni ( 35 . 8 Ma ) , basal to crown Sciuridae ( Table 4 , Appendix 1 . 2 ) . For the calibration point in the Sciurini , the soft maximum was informed by the age of M . neogrivensis and the ages of oldest Heteroxerus and D . jeffersoni stated above . Following the ‘consistent approach’ ( Hedman , 2010 ) the minimum age of the oldest Aplodontiidae ( 45 . 7 Ma ) is taken as the maximum possible age of both calibration points . Both MrBayes and BEAST analyses relied on four independent runs of 25 , 000 , 000 generations , each sampled every 4000 generations . Each of the runs in MrBayes relied on four Markov chain Monte Carlo ( MCMC ) chains . | Mammals can walk , hop , swim and fly; a few , like marsupial sugar gliders or colugos , can even glide . With 52 species scattered across the Northern hemisphere , flying squirrels are by far the most successful group that adopted this way of going airborne . To drift from tree to tree , these small animals pack their own ‘parachute’: a membrane draping between their lower limbs and the long cartilage rods that extend from their wrists . Tiny specialized wrist bones , which are unique to flying squirrels , help to support the cartilaginous extensions . The origin of flying squirrels is a point of contention: while most genetic studies point towards the group splitting from tree squirrels about 23 million years ago , the oldest remains – mostly cheek teeth – suggest the animals were already soaring through forests 36 million years ago . However , recent studies show that the dental features used to distinguish between gliding and non-gliding squirrels may actually be shared by the two groups . In 2002 , the digging of a dump site in Barcelona unearthed a peculiar skeleton: first a tail and two thigh bones , big enough that the researchers thought it could be the fossil of a small primate . In fact , and much to the disappointment of paleoprimatologists , further excavating revealed that it was a rodent . As the specimen – nearly an entire skeleton – was being prepared , paleontologists insisted that all the ‘dirt’ attached to the bones had to be carefully screen-washed . From the mud emerged the minuscule specialized wrist bones: the primate-turned-rodent was in fact Miopetaurista neogrivensis , an extinct flying squirrel . Here , Casanovas-Vilar et al . describe the 11 . 6 million years old fossil , the oldest ever found . The wrist bones reveal that the animal belongs to the group of flying squirrels that have large sizes . Evolutionary analyses that combined molecular and paleontological data demonstrated that flying squirrels evolved from tree squirrels as far back as 31 to 25 million years ago , and possibly even earlier . In addition , the results show that Miopetaurista is closely related to Petaurista , a modern group of giant flying squirrels . In fact , their skeletons are so similar that the large species that currently inhabit the tropical and subtropical forests of Asia could be considered living fossils . Molecular and paleontological data are often at odds , but this fossil shows that they can be reconciled and combined to retrace history . Discovering older fossils , or even transitional forms , could help to retrace how flying squirrels took a leap from the rest of their evolutionary tree . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2018 | Oldest skeleton of a fossil flying squirrel casts new light on the phylogeny of the group |
Positional information is fundamental to animal regeneration and tissue turnover . In planarians , muscle cells express signaling molecules to promote positional identity . At the ends of the anterior-posterior ( AP ) axis , positional identity is determined by anterior and posterior poles , which are putative organizers . We identified a gene , nr4A , that is required for anterior- and posterior-pole localization to axis extremes . nr4A encodes a nuclear receptor expressed predominantly in planarian muscle , including strongly at AP-axis ends and the poles . nr4A RNAi causes patterning gene expression domains to retract from head and tail tips , and ectopic anterior and posterior anatomy ( e . g . , eyes ) to iteratively appear more internally . Our study reveals a novel patterning phenotype , in which pattern-organizing cells ( poles ) shift from their normal locations ( axis extremes ) , triggering abnormal tissue pattern that fails to reach equilibrium . We propose that nr4A promotes pattern at planarian AP axis ends through restriction of patterning gene expression domains .
Metazoans display a large diversity of developmental modes and adult forms . Processes that govern the generation of form , collectively known as patterning , act to regulate cell identity , location , and number ( Wolpert , 1969 ) . The mechanisms by which pattern is established and maintained in adult tissues , however , are poorly understood . Planarians are freshwater flatworms capable of remarkable feats of whole-body regeneration and offer the opportunity , as a model system , to generate important insights into the molecular and cellular mechanisms that can generate and maintain adult tissue pattern . The planarian anterior-posterior ( AP ) axis has been a target for study of positional information in adult biology . RNA interference ( RNAi ) approaches have identified genes with regionalized expression domains that are important in AP patterning ( Forsthoefel and Newmark , 2009; Adell et al . , 2010; Reddien , 2011 ) . Genes with such constitutive regionalized expression and association with pathways with planarian patterning roles are called position control genes or PCGs ( Witchley et al . , 2013; Scimone et al . , 2016; Fincher et al . , 2018 ) . PCGs are predominantly expressed in planarian muscle ( Witchley et al . , 2013; Scimone et al . , 2016; Fincher et al . , 2018; Scimone et al . , 2018 ) . A number of PCGs encode evolutionarily conserved signaling proteins , such as members of the FGFRL family and Wnt/ß-catenin pathway components ( Reddien , 2011 ) . For example , inhibition of the FGFRL gene ndl-3 and the Wnt gene wntP-2 led to the formation of ectopic mouths and pharynges in the trunk ( Lander and Petersen , 2016; Scimone et al . , 2016 ) . These findings support a model in which planarian muscle serves as a source of positional information that patterns surrounding tissues , for instance by influencing the specification of resident stem cells called neoblasts and/or the localization of the specified progeny cells of neoblasts ( Reddien , 2011; Witchley et al . , 2013; Scimone et al . , 2017; Wurtzel et al . , 2017; Atabay et al . , 2018; Hill and Petersen , 2018 ) . Some PCGs are expressed in small groups of cells at the extreme ends of the primary axis called the anterior and posterior poles . Organizers , commonly studied in embryogenesis , are groups of cells that influence the fate of their neighboring cells for the production of tissue pattern ( Spemann and Mangold , 1924; Lemaire and Kodjabachian , 1996; De Robertis et al . , 2000 ) . Planarian poles function as putative organizers in planarian head and tail patterning ( Scimone et al . , 2014; Vogg et al . , 2014; Owlarn and Bartscherer , 2016; Oderberg et al . , 2017; Reddien , 2018 ) . The anterior pole expresses notum , which encodes a Wnt inhibitory protein ( Petersen and Reddien , 2011 ) , the transcription factor ( TF ) -encoding genes foxD , zic-1 , islet , pitx ( Hayashi et al . , 2011; Currie and Pearson , 2013; März et al . , 2013; Scimone et al . , 2014; Vásquez-Doorman and Petersen , 2014; Vogg et al . , 2014 ) , and follistatin , which encodes an Activin-inhibitory protein ( Gaviño et al . , 2013; Roberts-Galbraith and Newmark , 2013 ) . The posterior pole expresses wnt1 ( Petersen and Reddien , 2008 ) , the TF-encoding genes islet1 ( Hayashi et al . , 2011; März et al . , 2013 ) and pitx ( Currie and Pearson , 2013; März et al . , 2013 ) , and wnt11-2 ( Adell et al . , 2009; Gurley et al . , 2010 ) . Ablation of the anterior pole through inhibition of pole-specifying genes resulted in head patterning defects , including cyclopia and fused brain lobes ( Felix and Aboobaker , 2010; Blassberg et al . , 2013; Chen et al . , 2013; Currie and Pearson , 2013; März et al . , 2013; Scimone et al . , 2014; Vásquez-Doorman and Petersen , 2014; Vogg et al . , 2014 ) . Ablation of the posterior pole or inhibition of posterior-pole PCGs resulted in tail-patterning defects , such as stunted tails and fused ventral nerve cords ( Petersen and Reddien , 2009; Adell et al . , 2009; Gurley et al . , 2010; Hayashi et al . , 2011; Currie and Pearson , 2013; März et al . , 2013 ) . We used tissue-fragment and single-cell RNA sequencing to determine the anterior- and posterior-pole transcriptomes , respectively , and expanded the known repertoire of genes with enriched expression in these putative organizers . Inhibition of identified genes uncovered a novel patterning role for the gene nr4A , which encodes a planarian ortholog of the broadly conserved NR4A nuclear receptor . nr4A is expressed broadly in muscle , including strongly in both poles and other muscle cells at the ends of the AP axis . nr4A inhibition results in a novel patterning abnormality in which PCG expression domains shift away from the head and tail tips , followed by similar shifts in differentiated tissues during homeostatic turnover in uninjured animals . The nr4A phenotype represents a patterning abnormality in which a stable anatomical pattern is not reached and instead iterative duplication of tissues continues to occur . This phenotype reveals how homeostatic placement of pattern-generating cells at a correct location is essential for the maintenance of stable tissue pattern . We conclude NR4A is a transcription factor that helps regulate tissue pattern at both ends of the AP body axis .
To comprehensively identify genes expressed in the planarian poles , we performed RNA sequencing of anterior and posterior poles from both uninjured animals and regenerating trunks 72 hr post amputation ( hpa ) . To obtain the anterior-pole transcriptome , we surgically isolated anterior poles ( Figure 1A ) . Differential gene expression analysis identified 203 genes with significantly higher ( padj <0 . 05 ) expression in pole-containing pieces compared to flanking pieces from uninjured animals ( Figure 1A , Supplementary file 1A ) . In the anterior blastema ( an unpigmented outgrowth of differentiating tissue ) , 86 genes had pole-enriched expression ( padj <0 . 05 ) ( Figure 1A , Supplementary file 1B ) . 51 genes had enriched expression in the poles of both uninjured and regenerating animals and these included the previously published anterior-pole genes foxD , notum , zic-1 , and follistatin ( Figure 1A , Supplementary file 1A , B ) . Because the posterior pole is diffuse , its excision contains substantial non-pole tissue . Therefore , we utilized single-cell RNA sequencing of cells from uninjured tail tips and from 72 hpa posterior blastemas to obtain the posterior-pole transcriptome ( Figure 1—figure supplement 1A ) . We used qRT-PCR to identify FACS-isolated single cells expressing wnt1 and the muscle marker collagen to identify pole cells ( wnt1+; collagen+ ) and non-pole muscle cells ( wnt1-; collagen+ ) ( Figure 1—figure supplement 1A ) . From 768 cells , 11 posterior pole cells were identified ( six from uninjured animals and five from blastemas ) ; 90 non-pole muscle cells ( 43 from uninjured animals and 47 from blastemas ) were used as controls ( Figure 1—figure supplement 1B ) . Using single-cell differential expression ( SCDE ) analysis ( Kharchenko et al . , 2014 ) , we identified 198 genes with significantly higher expression ( p<0 . 05 ) in posterior-pole cells compared to non-pole posterior muscle cells ( Figure 1—figure supplement 1B , Supplementary file 1C ) . wnt1 , pitx , and islet1 were in the top 16% of those 198 genes ranked by enrichment in posterior-pole cells ( Supplementary file 1C ) . We assessed the in vivo expression of 133 candidate anterior-pole genes and 96 candidate posterior pole genes by whole-mount in situ hybridization ( WISH ) and identified 12 new genes expressed in the anterior pole ( Figure 1B ) and 10 new genes expressed at the posterior pole ( Figure 1—figure supplement 1C ) . The new pole genes included those encoding a predicted secreted factor ( kallmann1 ) , cell-surface receptors ( ror1 , ephr4 , ephr5 , pcdh9 , dcc , ddr2 ) , and transcription factors ( islet2 , musculin , nr4A ) ( Figure 1B , Figure 1—figure supplement 1C ) . Interestingly , some of these genes , such as kallmann1 , ephr5 , and dd_20026 were expressed in both anterior and posterior poles , identifying molecular similarity between these structures ( Figure 1B , Figure 1—figure supplement 1C , Supplementary file 1A-C ) . We confirmed the expression of the genes in anterior pole cells by their co-expression with the anterior pole markers notum and foxD ( Figure 1C ) . We utilized RNAi to determine the functions of new pole genes and identified a novel phenotype following inhibition of the gene nr4A . nr4A encodes the S . mediterranea homolog to proteins of the NR4A nuclear receptor family , which are found broadly in metazoans ( Bridgham et al . , 2010; Escriva et al . , 2004 ) ( Figure 2—figure supplement 1 ) . RNAi of nr4A resulted in the progressive formation of posterior ectopic eyes in uninjured animals ( Figure 2A ) . After initiation of nr4A RNAi , the distance between the original eyes and the head tip progressively decreased . By 6 weeks of nr4A RNAi , an additional eye pair emerged just posterior to the original eyes . As time progressed , the original and second eye pairs became closer to the head tip and decreased in size . A third eye pair , posterior to the second eye pair then appeared , resulting in animals with six eyes . This iterative process continued with ongoing nr4A inhibition , generating animals with as many as 10 eyes after 16 weeks of RNAi . By 12 weeks of nr4A RNAi , uninjured animals developed dorsal outgrowths at their tail tips , suggesting that nr4A influences both head and tail tip pattern ( Figure 2B ) . After transverse amputation of heads and tails , nr4A ( RNAi ) trunk fragments were capable of regenerating heads with normal eye number and tails without outgrowths ( Figure 2A , B ) , indicating that the nr4A phenotype is most readily apparent during tissue turnover associated with maintenance of anatomical pattern . Using tissue-specific RNA probes and fluorescence in situ hybridization ( FISH ) , we detected patterning abnormalities of multiple differentiated tissues in both the head and the tail of 9 week nr4A ( RNAi ) animals . In addition to ectopic eyes , labeled by an RNA probe to opsin , the heads of nr4A ( RNAi ) animals developed internal ( along the AP axis ) ectopic foci of epidermal cells normally restricted to the animal periphery at the dorsal-ventral ( DV ) -median plane ( laminB+ and NB . 22 . 1e+ cells ) ( Figure 2C , Figure 2—figure supplement 2 ) . Marginal adhesive gland cells ( mag-1+ ) are normally present in a prepharyngeal domain posterior to the eyes and along the body margin posteriorly from this zone ( Zayas et al . , 2010 ) ( Figure 2C ) . The distance from the original eyes to the posterior end of the mag-1+ domain was significantly greater in nr4A ( RNAi ) animals than in control animals ( Figure 2C , Supplementary file 1D ) . FISH with neural markers ChAT and dd_3524 , which are expressed in the inverted ‘U’-shaped cephalic ganglia , showed a loss of the head region between the apex of the head and the anterior aspect of the brain following nr4A RNAi ( Figure 2C ) . This phenotype is consistent with the shortening of the distance between the original eyes and the apex of the head , and the iterative appearance of ectopic posterior eyes with time . These results suggest that nr4A inhibition causes a loss of head tip pattern involving shortening of the distance between the anterior end of differentiated tissues and the head tip and posterior expansion of anteriorly restricted tissues . Similar to the changes we observed in the head , ectopic foci of DV boundary epidermal cells ( laminB+ and NB . 22 . 1e+ ) appeared internally ( along the AP axis ) in the tails of nr4A ( RNAi ) animals ( Figure 2D ) . The secretory mag-1+ cells normally restricted to the body margin around the tail periphery also appeared internally along the midline of tails following nr4A RNAi ( Figure 2D ) , resulting in aberrant tail-tip patterning . These internally mis-localized tissues indicate a similarity between the defects at the anterior and posterior extremities . The mislocalization of differentiated tissues in nr4A ( RNAi ) animals demonstrates that nr4A activity is required for normal restriction of tissues to their proper domains at both extreme ends of the AP axis . The patterning defects in nr4A ( RNAi ) animals led us to examine the expression of patterning genes ( PCGs ) in these animals . The expression of several genes like ndl-4 , sFRP-1 , notum , and foxD at the anterior end of the AP axis defines the head extremity ( Figure 3A ) . In uninjured nr4A ( RNAi ) animals analyzed at 9 weeks of RNAi , these expression domains were retracted from the head tip and expanded posteriorly ( Figure 3A , Figure 3—figure supplement 1 , Supplementary file 1D ) . foxD+; notum+ anterior pole cells were no longer present in a cluster at the head tip , but instead were found scattered between the eyes ( Figure 3A ) . Similar to the anterior pole domain , the expression domains of PCGs that are broadly expressed in the head and pre-pharyngeal regions , ndl-2 , ndl-5 , ndl-3 , and wnt2 , were also shifted . Specifically , the anterior boundary of their expression domains was shifted posteriorly from the head tip following nr4A RNAi ( Figure 3A , Figure 3—figure supplement 1 , Supplementary file 1D ) . When considered relative to an anatomical landmark - the original eyes - the anterior boundaries of ndl-2 and ndl-5 expression domains , for example , were significantly shifted more posteriorly compared to those in control animals ( Figure 3A , Figure 3—figure supplement 1 , Supplementary file 1D ) . ndk expression was not excluded from the head tip in nr4A ( RNAi ) animals , but was also expanded more posteriorly ( Figure 3A , Figure 3—figure supplement 1 , Supplementary file 1D ) . Whereas nr4A RNAi impacted the expression of many head PCGs , inhibition of head PCGs fz5/8-4 , ndk , wntA , and foxD , which are required for anterior patterning ( Cebrià et al . , 2002; Scimone et al . , 2014; Vogg et al . , 2014; Scimone et al . , 2016 ) , did not lead to detectable nr4A expression changes ( Figure 3—figure supplement 2 ) . Amputated nr4A ( RNAi ) animals were able to regenerate anterior poles by 72hpa ( Figure 3B ) . However , at 7 days post amputation ( 7dpa ) the anterior-pole-cell cluster was absent from the head tip and instead was disorganized and shifted posteriorly ( Figure 3B ) . This anterior-pole positioning phenotype is similar to the anterior-pole phenotype we observed in uninjured nr4A RNAi ( Figure 3A ) . In the posterior , wnt1 is normally expressed linearly along the midline of the tail tip ( the posterior pole ) . At 9 weeks post-initiation of nr4A RNAi , the posterior-most expression domain of wnt1 was absent , leaving a gap between the tip of the tail and wnt1 expression ( Figure 3C , Figure 3—figure supplement 1 , Supplementary file 1D ) . Furthermore , the localized expression of wnt11-2 at the tail tip was also abrogated by nr4A RNAi ( Figure 3C , Supplementary file 1D ) . We also observed an increase in the area of wnt11-1 expression in the tails of nr4A ( RNAi ) animals ( Figure 3C , Figure 3—figure supplement 1 , Supplementary file 1D ) . In the midbody region , nr4A RNAi did not significantly affect PCG expression domains . Specifically , the location of the posterior boundaries of the expression domains of ndl-2 , ndl-3 , ndl-5 , and wnt2 were normal compared to the control ( Figure 3—figure supplement 3A , Supplementary file 1D ) , despite the anterior boundaries of their expression domains being shifted ( Figure 3A , Figure 3—figure supplement 1 , Supplementary file 1D ) . Additionally , the length of ptk7 expression domain , defined by the distance between its anterior and posterior boundaries in the midbody , was not significantly altered by nr4A RNAi ( Figure 3—figure supplement 3A , Supplementary file 1D ) . We also examined PCG expression pattern on the DV and medial-lateral axes of nr4A ( RNAi ) animals . The dorsal-specific expression of bmp4 was largely intact in nr4A ( RNAi ) animals , with an increase in its expression in the esophagus compared with its expression in control animals ( Figure 3—figure supplement 3B ) . There was no ectopic ventral bmp4 expression in nr4A ( RNAi ) animals ( Figure 3—figure supplement 3B ) . The midline expression domains of slit and ephR1 were broadened medial-laterally and were disorganized in the head of nr4A ( RNAi ) animals ( Figure 3A ) . slit expression was also reduced in the tail of nr4A ( RNAi ) animals , but its midline expression in the rest of the body was comparable to that of control animals ( Figure 3—figure supplement 3C ) . This suggests nr4A inhibition does not cause a gross midline-patterning defect , but rather disrupts slit expression specifically in the head and tail , likely associated with disruption in the pattern of anterior and posterior poles , which play important roles in midline specification and patterning ( Hayashi et al . , 2011; Currie and Pearson , 2013; März et al . , 2013; Scimone et al . , 2014; Vásquez-Doorman and Petersen , 2014; Vogg et al . , 2014; Oderberg et al . , 2017 ) . Taken together , our PCG expression analysis showed that nr4A inhibition causes anterior PCGs to shift posteriorly and posterior PCGs to shift anteriorly ( Figure 3D ) while largely sparing midbody , midline , and DV PCG expression domains outside of the head and tail ends . Planarian muscle cells have important roles in expressing positional information to control adult body pattern maintenance and regeneration ( Witchley et al . , 2013; Scimone et al . , 2016; Scimone et al . , 2017 ) . Consistent with its role in patterning , nr4A was predominantly expressed in planarian muscle in both intact animals and regenerating blastemas ( Figure 4A ) . A higher proportion of muscle cells expressed nr4A in the head and tail than in the pre-pharyngeal region ( Figure 4B ) . Although its expression was enriched at the ends of the AP axis , including in the posterior pole ( Figure 4C ) , nr4A was also expressed broadly in body-wall muscle ( Figure 4A ) . Mapping nr4A expression onto single-cell expression data ( Wurtzel et al . , 2015 ) showed that nr4A was most highly expressed in muscle cells , with some expression in neoblasts ( Figure 4D ) . By using a single-cell sequencing dataset of different muscle cell subtypes ( Scimone et al . , 2018 ) , we found that nr4A was expressed in longitudinal , circular , and DV muscle fiber types , but not in intestinal muscle cells ( Figure 4E , Figure 4—figure supplement 1 ) . Similarly , SCDE analysis of nr4A-expressing muscle cells using large-scale single-cell RNA sequencing data ( Fincher et al . , 2018 ) showed no enrichment for fiber-specific transcripts ( Supplementary file 1E ) . During regeneration , nr4A expression was upregulated by 48 hpa and increased in level at 72 hpa at both anterior- and posterior-facing wound sites ( Figure 4F , G ) . This expression time-course behavior in regeneration was similar to that of many anterior and posterior PCGs ( Wenemoser et al . , 2012; Wurtzel et al . , 2015 ) ( Figure 4G ) . However , unlike canonical muscle-expressed pattering factors , nr4A was upregulated at both anterior- and posterior-facing wounds – having both anterior- and posterior-PCG-like expression dynamics during regeneration ( Figure 4G ) . Because nr4A encodes a transcription factor , we reasoned transcriptional changes in the head and tail tips of nr4A ( RNAi ) animals , such as changes in muscle-expressed patterning genes , might precede and explain the patterning phenotype observed . To identify such gene expression changes , we collected head and tail regions for RNA sequencing at several time points before gross anatomical changes could be detected: 2 , 3 , 4 , and 5 weeks following the initiation of nr4A RNAi ( Figure 5A ) . We identified 56 and 41 genes with significant expression changes in the head and tail ( padj <0 . 05 ) , respectively , in at least one time point in nr4A ( RNAi ) animals compared to controls ( Figure 5A ) . The majority of these genes ( 49/56 in the head and 36/41 in the tail ) were downregulated and most ( 32/56 in the head and 19/41 in the tail ) were genes known to have muscle-enriched expression from available single-cell RNA sequencing data on wild-type animals ( Wurtzel et al . , 2015 ) ( Figure 5A ) . 16 genes were significantly affected by nr4A in both the head and the tail . A high proportion ( 11/16 ) of those ( e . g . , dd_9565 ( col21a1 ) and dd_11601 ( qki ) ) were genes known to be expressed in muscle ( Wurtzel et al . , 2015 ) . We verified the decrease in the expression of many of these genes following nr4A RNAi with FISH ( Figure 5B , Figure 5—figure supplement 1A ) . Whereas nr4A RNAi reduced the body-wide expression of some genes ( e . g . , dd_9565 ( col21a1 ) , dd_1706 ( psapl1 ) ) , it inhibited the expression of other genes more specifically in the head and tail ( e . g . , dd_2972 , dd_11683 ( ca12 ) , dd_950 ( vim ) ) ( Figure 5B , Figure 5—figure supplement 1A ) . One of them , dd_950 ( vim ) , was expressed in the epidermis ( Videos 1 and 2 ) . We also analyzed the expression patterns of genes with expression affected by nr4A RNAi using available RNA sequencing data from different AP axis segments ( Stückemann et al . , 2017 ) ( Figure 5—figure supplement 1B ) . One-third of the nr4A-regulated genes , including 9 of 16 genes affected by nr4A in both the head and the tail , were enriched in their expression in the head and the tail compared to midbody regions ( Figure 5—figure supplement 1B ) . Among the genes with muscle-enriched expression affected by nr4A RNAi were several PCGs ( Figure 5A ) . Following initiation of RNAi , ndl-2 and ndl-5 were downregulated in the head ( by week 3 and 4 , respectively ) , wnt11-1 was upregulated in the tail ( by week 2 ) , and nlg-8 was downregulated in the tail ( by week 3 ) . These findings are consistent with our FISH studies of the nr4A ( RNAi ) phenotype ( Figure 3A , C and Figure 5—figure supplement 1C ) . These genes encode FGFRL ( ndl-2 , –5 ) , Wnt ( wnt11-1 ) , and Bmp-regulatory ( nlg-8 ) proteins and have been implicated in patterning ( Molina et al . , 2007; Gurley et al . , 2010; Lander and Petersen , 2016; Scimone et al . , 2016 ) . These data show that nr4A is required for the expression of numerous muscle-expressed genes at the planarian head and tail ends , including those encoding extracellular matrix proteins and PCGs . Because a number of genes with muscle-enriched expression were downregulated following nr4A RNAi , we examined muscle fibers in these animals using the 6G10 muscle antibody ( Ross et al . , 2015 ) . By 9 weeks of RNAi , when multiple ectopic eyes were present , uninjured nr4A ( RNAi ) animals showed loss of longitudinal , circular , and diagonal muscle fibers and a decreased number of collagen+ muscle cells in the head compared to control animals ( Figure 6A–C , Figure 6—figure supplement 1A ) . By contrast , muscle fibers were not significantly affected in the trunk or tail of nr4A ( RNAi ) animals and collagen+ muscle cell numbers were normal in nr4A ( RNAi ) animal tails ( Figure 6A–C Figure 6—figure supplement 1A ) . Similar analyses of nr4A ( RNAi ) animals at an earlier time point ( 4 weeks of RNAi ) showed no changes in muscle fibers or collagen+ cell numbers in the head or tail compared to control animals ( Figure 6A , C , Figure 6—figure supplement 1A–B ) . EdU labeling was performed at 9 weeks of nr4A RNAi to examine new muscle cell production . The numbers of new muscle cells ( EdU+; collagen+ ) in the head tip of nr4A ( RNAi ) animals , but not in the posterior head , trunk , or tail , were significantly decreased compared to similar regions ( by location and area ) in control animals ( Figure 6D ) . By contrast , there were no significant differences in the total amount of EdU incorporation ( all EdU+ cells ) in those regions between nr4A ( RNAi ) and control animals ( Figure 6—figure supplement 1C ) . Comparing the region anterior to the original eyes in nr4A ( RNAi ) animals to the region anterior to the eyes in control animals showed greater reduction in new muscle cell number and reduction in total EdU incorporation ( Figure 6E , Figure 6—figure supplement 1D ) . This finding is consistent with the head tip ( distance between head apex and original eyes ) size reduction observed in nr4A ( RNAi ) animals ( Figure 2A , C ) . By contrast , the numbers of new muscle cells anterior to the eyes of controls and anterior to the newest eyes in nr4A ( RNAi ) animals were similar ( Figure 6E ) . nr4A ( RNAi ) and control animals had comparable numbers of apoptotic cells in similar head tip regions ( by location and area ) as assessed by TUNEL ( Figure 6—figure supplement 1E ) , suggesting defective progenitor migratory targeting as opposed to increased cell death underlies muscle loss at the head tip . These findings indicate that nr4A was not required in general for muscle progenitor production and differentiation throughout the body . The head-tip-specific muscle progenitor incorporation defect occurred after PCG and other gene expression changes were detected in the nr4A ( RNAi ) RNA sequencing data . Together with the observation that muscle fiber loss was also head-restricted and a late-stage phenotype ( absent at 4 weeks of RNAi ) , these findings suggest loss of muscle cells per se is not the cause of the PCG expression changes that occurred at earlier timepoints in the nr4A RNAi . Given that the RNA sequencing results suggest PCG expression shifts precede differentiated tissue pattern changes , we sought to characterize the sequence of PCG expression changes that might underlie the phenotype . Importantly , we found that PCG shifts occurred before changes were visible in differentiated tissues , including changes in head muscle ( Figure 7A–D ) . Specifically , in the head , ndl-2 and ndl-5 expression domains shifted posteriorly relative to the eyes by 3 weeks of nr4A RNAi compared to control RNAi ( Figure 7A , B , Figure 7—figure supplement 1A , B , Supplementary file 1D ) . At this time point , the numbers of muscle cells and the appearance of muscle fibers were normal in the heads of nr4A ( RNAi ) animals ( Figure 6A , C , Figure 6—figure supplement 1A , B ) . Mis-positioned anterior pole cells were the earliest abnormalities detected , becoming scattered between the eyes by two weeks after initiation of RNAi ( Figure 7A ) . This posterior shift of notum expression preceded detected changes in the expression of other head PCGs ( ndl-2 , ndl-4 , ndl-5 , sFRP-1 ) , as well as changes in the distribution of marginal adhesive gland cells ( Figure 7A , B , Figure 7—figure supplement 1A , B , Supplementary file 1D ) . Consistent with our findings that muscle fiber loss at the head tip occurs at late stages of the phenotype ( after 4 weeks - Figure 6A , B , Figure 6—figure supplement 1A , B ) , we observed no apparent differences in head tip muscle fibers between control and nr4A ( RNAi ) animals after two weeks of RNAi when pole cells were already gone from this location ( Figure 7C , Figure 7—figure supplement 1C ) . In the tail , the posterior pole shifted anteriorly by three weeks after initiation of RNAi ( Figure 7D , Supplementary file 1D ) . The clustered expression of wnt11-2 at the tail tip also decreased by this time point ( Figure 7D , Supplementary file 1D ) . However , no changes in mag-1+ cells were apparent at this time ( Figure 7D , Supplementary file 1D ) , indicating that similar to the case of the anterior head tip , PCG pattern changes precede differentiated tissue shifts at the posterior extremity of nr4A ( RNAi ) animals . Even after just one week of nr4A RNAi , we observed a quantifiable increase in the distance between notum+ cells and the head apex ( Figure 8A , Figure 8—figure supplement 1A , non-irradiated ) . Lethal irradiation kills the cycling cells ( neoblasts ) that are responsible for the production of all planarian cell types ( Bardeen and Baetjer , 1904; Dubois , 1949; Reddien et al . , 2005 ) . Irradiation suppressed this earliest detectable posterior shift of notum+ cells in nr4A ( RNAi ) animals , demonstrating that anterior-pole changes were neoblast-dependent ( Figure 8A , Figure 8—figure supplement 1A irradiated ) . SMEDWI-1 is a protein that persists in newly differentiated neoblast progeny ( Guo et al . , 2006 ) . SMEDWI-1+ ectopic anterior pole cells were present at both early ( 2 weeks ) and late ( 9 weeks ) nr4A RNAi time points ( Figure 8B ) , further indicating that new anterior pole cells are continually being targeted to ectopically posterior regions in nr4A ( RNAi ) animals . Relative to the original eyes , the anterior pole cells progressively appeared more posteriorly ( Figures 3A , 7A and 8A ) . Evidence suggests that the anterior pole has organizer-like activity that can induce surrounding cells to adopt head tip identity ( Scimone et al . , 2014; Vásquez-Doorman and Petersen , 2014; Vogg et al . , 2014; Oderberg et al . , 2017 ) . We hypothesized that rapid changes in anterior-pole-cell distribution following nr4A inhibition could be an important component in the development of the nr4A ( RNAi ) patterning phenotype . New anterior pole cell formation requires the gene foxD ( Scimone et al . , 2014; Vogg et al . , 2014 ) , providing a method to genetically ablate the anterior pole . Uninjured animals were subjected to either foxD or control RNAi for 6 weeks to eliminate the anterior pole ( a subgroup , "pole-cut animals" , was subjected to head tip excision to facilitate anterior pole removal ) . Subsequently , nr4A RNAi was added to foxD and control ( RNAi ) groups ( 10 weeks of nr4A RNAi for intact animals and 12 weeks for pole-cut animals ) to assess whether ectopic pole cell placement was required for the nr4A ( RNAi ) phenotype . A greater proportion of foxD; nr4A ( RNAi ) animals lacked ectopic eyes than did control; nr4A ( RNAi ) animals ( Figure 8C ) . This result was consistently observed in intact and pole-cut animal groups , and in biological replicates ( Figure 8—figure supplement 1B , C ) . qPCR of animals at the end of the double RNAi period showed similar levels of nr4A expression inhibition between foxD; nr4A ( RNAi ) and control; nr4A ( RNAi ) groups ( Figure 8D , Figure 8—figure supplement 1D ) . Taken together with the fact that a posterior shift in anterior pole cells was the earliest observed change in nr4A ( RNAi ) animals , these results indicate that the head patterning defects of nr4A RNAi are mediated , at least in part , by the mis-positioning of anterior pole cells . The iterative appearance of ectopic eyes in nr4A ( RNAi ) animals prompted us to ask if a stable , but different , tissue pattern ever emerged in these animals . Tracking the fates of the eyes in individual animals after extended periods of continuous nr4A inhibition ( as long as 21 weeks ) showed that the original eyes ( and even older ectopic eyes ) eventually faded and disappeared at the head tip as the phenotype progressed ( Figure 9A , Figure 9—figure supplement 1 ) . As this process unfolded , eye progenitors iteratively shifted their targeting to the most posterior set of eyes , leaving the anterior eyes to decay . Specifically , there were significantly more new eye cells ( opsin+; SMEDWI+ ) in the newest ( posterior ) ectopic eyes than in older ( more anterior ) ectopic eyes ( Figure 9B ) . The numbers of new eye cells produced in a given unit of time were similar between the eyes of control animals and the newest , posterior ectopic eyes in nr4A ( RNAi ) animals ( Figure 9B ) . These findings suggest that eye progenitor targeting continued to shift posteriorly after nr4A inhibition , iteratively producing ectopic eyes without reaching a stable tissue-pattern state .
The planarian anterior and posterior poles have been the subjects of recent intense study because of their roles in patterning the head and tail ( Reddien , 2011; Owlarn and Bartscherer , 2016; Reddien , 2018 ) . The poles are specialized muscle cells localized at the midline and at the extreme ends of the AP axis and are specified by transcription factors as discrete structures at the animal ends . Given the role for poles in influencing the pattern of neighboring tissues , the poles have been regarded as organizers in planarian regeneration ( Hayashi et al . , 2011; Currie and Pearson , 2013; März et al . , 2013; Scimone et al . , 2014; Vásquez-Doorman and Petersen , 2014; Vogg et al . , 2014; Vogg et al . , 2016 ) . Using bulk and single-cell RNA sequencing , we identified many new genes expressed in the anterior and posterior poles , including genes expressed in both poles that suggest functional similarities in these cells . For example , dd_13188 ( kallmann1 ) is expressed highly specifically in both poles and encodes an extracellular matrix protein implicated in olfactory neuron migration in humans ( Rugarli et al . , 1993; Rugarli , 1999 ) . Genes with unique functions in organizers in development and regeneration are of interest for understanding how these regions form and pattern neighboring tissue , and RNA-sequencing approaches like the one defined here could identify the transcriptomes of these regulatory regions of embryos and animals broadly . Planarian nr4A is specifically expressed in muscle and is required for AP axial patterning . nr4A genes are widely conserved in the animal kingdom , and to our knowledge this is the first report of a role in tissue patterning for a member of the NR4A-family of nuclear receptors . The most striking aspect of the activity of nr4A in planarians is its role in homeostatic patterning at both ends of the AP axis . Patterning phenotypes affecting both ends of the AP axis of animals are rare . In planarians , islet1 is known to be required for both anterior and posterior pole formation ( Hayashi et al . , 2011; März et al . , 2013 ) . Here , we find a unique patterning phenotype affecting pole positioning and differentiated tissue pattern at both extreme ends of the AP axis . NR4A-subfamily members belong to the broader superfamily of nuclear receptors specific to metazoans ( Bridgham et al . , 2010; Escriva et al . , 2004 ) . Although members of many other subfamilies of nuclear receptors bind to steroid hormones , retinoic acids , fatty acids , and prostaglandins , NR4A-family members are orphan receptors that have been proposed to regulate transcription in a ligand-independent manner ( Paulsen et al . , 1992; Bridgham et al . , 2010 ) . Whereas vertebrates possess three NR4A members ( NR4A1 , NR4A2 , and NR4A3 ) , many protostomes have only one ( Bertrand et al . , 2004 ) . DHR38 , the Drosophila NR4A ortholog , mediates an atypical ligand-independent ecdysteroid-signaling pathway responsible for molting and metamorphosis ( Baker et al . , 2003 ) . In vertebrates , three members of the NR4A family are immediate-early stress response genes induced by a host of physiological signals such as growth factors , cytokines , prostaglandins , neurotransmitters , and phorbol esters ( Maxwell and Muscat , 2006; Safe et al . , 2016 ) . NR4A1 , NR4A2 , and NR4A3 are expressed in a variety of tissues , such as the brain , liver , gonads , skeletal muscle , kidney , fat , lung , and endocrine glands ( Maxwell and Muscat , 2006 ) . In the fields of developmental biology and regeneration , little is known about the role of NR4A transcription factors . nr4A was expressed strongly in the planarian head and tail , but was also expressed in muscle cells throughout the body . After amputation , nr4A was upregulated at anterior-facing wounds , with timing similar to anterior PCGs , and at posterior-facing wounds , with timing similar to posterior PCGs . RNA-sequencing data from nr4A ( RNAi ) animals revealed that many of the genes dependent on nr4A for their expression were specifically expressed in muscle cells , the majority of which encode extracellular matrix ( ECM ) components ( collagens and metalloproteinase ) and signaling proteins . Several of these genes predicted to encode ECM components displayed enriched expression in both the head and tail tips . Although it is unknown which genes are direct transcriptional targets of the NR4A protein , genes that experienced changes in expression early in the course of RNAi are more likely to be directly dependent on nr4A . The planarian AP axis is associated with a continuum of distinct and overlapping PCG expression domains ( Forsthoefel and Newmark , 2009; Adell et al . , 2010; Reddien , 2011 ) . PCGs promote the production of varying cell types and tissue patterns along the AP axis , for example by promoting eye formation in the head and pharynx formation in the midbody ( Lander and Petersen , 2016; Scimone et al . , 2016 ) . Progenitors for differentiated tissues are specified by the expression of TF genes in neoblasts , generating specialized neoblasts ( Reddien , 2013; Zhu and Pearson , 2016 ) . These specialized neoblasts are specified coarsely on the AP axis and produce progenitors that migrate to specific locations or "target zones" ( TZs ) ( Atabay et al . , 2018 ) . For example , eye-specialized neoblasts are specified broadly in the head and pre-pharyngeal regions and migrate to precise locations in the head where eyes form and are maintained ( Lapan and Reddien , 2011; Lapan and Reddien , 2012 ) . Altering PCG pattern with RNAi can result in mis-targeting of progenitors . For example , inhibition of the laterally expressed wnt5 gene causes eye progenitors to be targeted more laterally than in the wild type , resulting in ectopic lateral eyes ( Atabay et al . , 2018 ) . In addition to migratory targeting of progenitors by extrinsic cues , self-organization of progenitors into their target organ/tissue has a major influence on progenitor targeting ( Atabay et al . , 2018 ) . For example , in instances when an organ and the PCG pattern are discordant ( such as during morphallaxis ) , progenitors can be incorporated into such an organ despite the fact that it is in the incorrect position ( Atabay et al . , 2018; Hill and Petersen , 2018 ) . We propose a model for the nr4A ( RNAi ) phenotype based upon this prior knowledge and the work presented here . The model involves ectopically posterior tissues in the head and ectopically anterior tissues in the tail manifesting from shifts of the AP positional information away from AP axis ends . This model is described in Figure 9C and below: Focusing on the head , the AP PCG axis terminus demarcated by the anterior pole shifts internally from the AP anatomical terminus ( the head tip ) in nr4A ( RNAi ) animals . These pole cells influence gene expression in neighboring muscle so that the entire PCG pattern for the head shifts posteriorly . This results in an animal with a frameshift between the positional information map and the anatomy ( Figure 9C ) . PCG expression shifting changes the target-zone locations for progenitors . For example , once the target zone for the eye has shifted sufficiently posterior to avoid the self-organizing influence of the original eyes , ectopic posterior eyes appear ( Figure 9C ) . At this point the original eyes stop receiving progenitors and shrink , as was observed . The head tip shrinks for similar reasons - without the PCG extremity zone coincident with the head tip , less progenitor targeting to the tip leads to its decay during natural tissue turnover ( Figure 9C ) . This patterning abnormality continues in an iterative process: new sets of posterior eyes continue to appear and anterior eye sets and the head tip shrink . What underlies this iterative process ? We propose that the continuous out-of-register placement of pattern-organizing cells ( new pole cells that are part of naturally occurring tissue turnover ) in relation to the rest of positional information underlies this process ( Figure 9C ) . Anterior pole progenitors would continue to be targeted too posteriorly in the changing PCG map . This continues to influence PCG pattern , shifting it even farther posteriorly . Consequently , the extremity continues to decay and the target zone for eye progenitors continues to move posteriorly , with sets of eyes iteratively appearing posteriorly , and anterior sets of eyes decaying because of a lack of progenitor targeting for their renewal . Tracking new eye progenitor incorporation supported this model ( Figure 9B ) . This process would continuously occur , with equilibrium between PCG pattern and anatomy pattern not attainable . This novel type of patterning phenotype suggests that breaking the concordance between new pattern-organizing cells and pattern itself can lead to a phenotype of continuous pattern-anatomy shifting . This model depends on PCG-pattern shifts preceding differentiated tissue changes , as was observed . As early as one week after nr4A RNAi , the anterior pole began to shift away from the head tip , followed by the posterior pole shifting away from the tail tip at 3 weeks of RNAi . Between 3 and 4 weeks of RNAi , a host of other head and tail PCGs became progressively excluded from head and tail tips . Ectopic eyes , gland , or DV-boundary epidermal cells all appeared later , between 6 and 12 weeks of RNAi . This model is also supported by the finding that ectopic anterior pole cells in nr4A ( RNAi ) animals were newly specified from neoblasts and were required for the nr4A ( RNAi ) phenotype . Ultimately , muscle progenitor targeting to the head tip was reduced in nr4A ( RNAi ) animals , but this occurred after pole and other PCG pattern shifts initiated . This reduction could reflect a general defect in head tip targeting of muscle and other progenitors , perhaps first manifested by the shift in pole progenitor placement , which triggered pattern changes in head muscle . The nr4A ( RNAi ) phenotype is distinct from prior RNAi phenotypes affecting the number of planarians eyes or pattern at the ends of the AP axis . Unlike notum RNAi in uninjured animals , which results in anterior ectopic eyes ( Atabay et al . , 2018; Hill and Petersen , 2018 ) , nr4A RNAi generates posterior ectopic eyes . In ndk RNAi , ectopic eyes appear in very posterior regions ( including the trunk and tail ) ( Cebrià et al . , 2002 ) , whereas in nr4A RNAi , ectopic posterior eyes are restricted to the head . Furthermore , ndk RNAi does not lead to a posterior shift in the anterior pole ( Scimone et al . , 2016 ) . Finally , although islet1 inhibition also affects both poles , its patterning defect during regeneration ( versus during homeostasis for nr4A RNAi ) leads to midline collapse ( e . g . , cyclopia ) ( Hayashi et al . , 2011; März et al . , 2013 ) that is not observed in nr4A ( RNAi ) animals . Why would the iterative tissue treadmill-like process of the nr4A ( RNAi ) phenotype stay regionally restricted , without progressing through the body ? Whereas PCG expression in the head tip retracted posteriorly , expression boundaries of PCGs outside of the head ( e . g . , the posterior expression boundary of ndl-2 , ndl-5 , and wnt2 ) did not move posteriorly . Therefore , if head progenitors in general are still made at normal rates ( as seen for eye progenitors ) , one possibility is that more progenitor targeting towards the posterior head could result in growth counterbalancing head tip shrinking . Our findings regarding the nr4A ( RNAi ) phenotype are consistent with and provide additional evidence for the presence of PCG-defined progenitor target zones that both maintain proper tissue structures during homeostasis and instruct tissue formation at new locations as PCG zones shift , for example , in regeneration ( Atabay et al . , 2018; Hill and Petersen , 2018 ) . Together , our data show that nr4A is necessary to maintain the correct location of both the anterior- and posterior-PCG expression zones , consequently maintaining the proper pattern of positional information and differentiated tissues in the head and the tail . We conclude that nr4A is a novel adult patterning gene that helps muscle promote the patterning of both the anterior and posterior extreme ends of the primary planarian body axis .
Asexual strain CIW4 of Schmidtea mediterranea ( RRID: NCBITaxon:79327 ) was used for all experiments . Animals were starved 7–14 days before experiments . Cut animal fragments were diced with a scalpel , homogenized using Qiagen TissueLyser III , and RNA was extracted using TRIzol ( Life Technologies ) according to the manufacturer’s protocol . Between 0 . 5 and 1 μg of RNA was used for cDNA sequencing library synthesis using TruSeq RNA Library Prep Kit V2 ( Illumina ) following the manufacturer’s protocol . Libraries were sequenced on Illumina HiSeq for an average sequencing depth of 20–30 million reads per replicate sample . Libraries were mapped to the dd_Smed_v4 transcriptome ( http://planmine . mpi-cbg . de ) using bowtie 1 ( Langmead et al . , 2009 ) with -best alignment parameter . Pairwise differential expression analysis was performed using DESeq ( Anders and Huber , 2010 ) ( RRID: SCR_000154 ) . Sequencing data and DESeq analysis available in NCBI/GEO ( accession GSE121048 ) . Cut animal fragments were diced with a scalpel , macerated with collagenase treatment , stained with Hoechst ( 1:25 ) and propidium iodide ( 1:5000 ) and 2C ( differentiated , non-dividing ) viable ( Hoechst+; propidium iodide- ) cells were sorted ( Hayashi et al . , 2006 ) one cell per well by fluorescence activated cell sorting on 96-well plates containing 5 μL of Buffer TCL with 1% 2-mercaptoethanol . Single-cell RNA sequencing libraries were prepared via the SmartSeq2 method , as previously described ( Picelli et al . , 2013; Picelli et al . , 2014 ) . Briefly , a poly-dT oligo and a template-switching oligo were used for reverse transcription . After cDNA amplification , single-cell cDNA libraries were fragmented and tagged using the Nextera XT kit ( Illumina ) . Single-cell cDNA libraries were then qRT-PCR screened for wnt1 and collagen expression with the primers provided in Supplementary file 1F . Posterior pole cells with wnt1 expression ( 11 ) and muscle cells with collagen but no wnt1 expression ( 90 ) were selected for were sequencing on Illumina HiSeq for an average of 1–2 million reads per cell . Differential expression analysis of gene expression between pole cells and non-pole cells was performed using the Single-Cell Differential Expression ( SCDE , RRID: SCR_015952 ) method , as described ( Kharchenko et al . , 2014 ) . Sequencing data and SCDE analysis available in NCBI/GEO ( accession GSE121048 ) . Previously published and homology-verified planarian genes are in lowercase italics by themselves . Genes not analyzed for homology by phylogenetics but that have strong ( E-value <0 . 05 ) human best-BLAST matches have gene identifiers with "dd_" followed by their Smedv4 . 1 Dresden transcriptome assembly contig number and italicized human best-BLAST match in parentheses . Genes that only have Dresden transcriptome assembly contig identifiers have no human , mouse , or C . elegans best BLAST matches . Nuclear receptor family phylogenetic tree was constructed from 114 total protein sequences ( including the translated Schmidtea mediterranea dd_12229 nr4A mRNA sequence ) from representative nuclear receptors of all six nuclear receptor subfamilies from eight different species plus Schmidtea mediterranea . Their accession numbers are in Supplementary file 1G . Protein sequences were aligned by their conserved DNA-binding and ligand-binding domains via maximum likelihood method using PhyML with 1000 bootstrap replicates . Trees were visualized and formatted in FigTree . Primers used to clone nr4A and its targets are listed in Supplementary file 1F . Genes were cloned from cDNA into the pGEM vector ( Promega ) and transformed into E . coli DH10B by heat shock . Bacteria were plated on agarose plates containing 1:500 carbenicillin , 1:200 Isopropylthio-β-D-galactoside ( IPTG ) , and 1:625 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside ( X-gal ) for overnight growth . Colonies were screened by colony PCR and gel electrophoresis . Plasmids were extracted from positive ( white ) colonies and subsequently validated by Sanger sequencing . Single RNAi dsRNA was transcribed in vitro ( Promega reagents ) from PCR-generated templates with flanking T7 promoters . It was then precipitated in ethanol , annealed after resuspension in water , and mixed with planarian food ( liver ) . Each animal was fed 2 µL of the liver containing 4–7 µL/mL dsRNA twice every week , with 2–3 days between each feeding . Animals were then fixed seven days after the last feeding . The total amount of dsRNA per feeding per animal was kept constant as described before . Control RNAi used dsRNA against the C . elegans unc-22 transcript . WISH with nitro blue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate ( NBT/BCIP ) was performed as described ( Pearson et al . , 2009 ) . FISH and post-antibody binding washes and tyramide development were performed as described ( King and Newmark , 2013 ) . Briefly , animals were killed in 5% NAC and treated with proteinase K ( 2 μg/ml ) . Animals were hybridized with RNA probes at 1:800 dilution overnight at 56°C . Samples were then washed twice in pre-hybridization buffer , 1:1 pre-hybridization: 2X SSC , 2X SSC , 0 . 2X SSC , and PBS with Triton-X ( PBST ) in that sequence . Blocking was performed in 5% Western Blocking Reagent ( Roche ) plus 5% heat-inactivated horse serum diluted in PBST solution when anti-DIG , anti-FITC , or anti-DNP antibodies were used . Post-antibody binding washes and tyramide development were performed as described ( King and Newmark , 2013 ) . Light images were taken with a Zeiss Discovery Microscope . Fluorescent images were taken with a Zeiss LSM700 Confocal Microscope . Fiji/ImageJ ( RRID: SCR_002285 ) was used for FISH co-localization analyses and PCG domain quantifications . Animals were fixed in 4% formaldehyde solution as for in situ hybridization and treated as described ( King and Newmark , 2013 ) . The muscle antibody 6G10 ( Ross et al . , 2015 ) ( RRID: AB_2619613 ) was used at 1:1000 dilution in PBSTB ( 0 . 1% TritonX , 0 . 1% BSA ) , and an anti-mouse-Alexa conjugated antibody ( Life Technologies ) was used at a 1:500 dilution . The rabbit anti-SMEDWI-antibody ( RRID: AB_2797418 ) was used at 1:1000 dilution in PBSTx ( 0 . 1% TritonX ) with 10% horse serum , and an anti-rabbit-HRP antibody was used at 1:300 dilution in PBSTx ( 0 . 1% TritonX ) with 10% horse serum . All quantification was performed in a condition-blind manner . Statistical analyses were done using Welch’s t-tests for comparisons between two groups or using Brown-Forsythe and Welch ANOVA tests with Dunnett’s T3 multiple comparisons test for comparisons involving more than two groups , unless otherwise specified . For the proportion of muscle cells expressing nr4A , the number collagen+; nr4A + cells was divided by the number of collagen+ cells in head tip , pre-pharyngeal region , and tail tip in a 200 μm by 200 μm area through a 15 μm stack ( Figure 4—source data 1 ) . The number circular and longitudinal muscle fibers in each animal was counted within a 100 μm x 100 μm image area ( Figure 6—source data 1 , Figure 6—figure supplement 1—source data 1 ) . For muscle cell quantification , collagen+ cells were counted in a in a 190 μm wide by 125 μm high rectangular area centered around the midline of the tip of the heads and tails in size-matched control and nr4A ( RNAi ) animals ( at least four animals per condition ) , through the entire thickness ( DV axis ) of the animal ( Figure 6—source data 2 ) . The numbers of EdU+; collagen+ and EdU+ cells were counted in size- and location-matched areas in control and nr4A ( RNAi ) animals through a 72 μm stack for head , posterior head , and tail regions , and a 33 μm stack in the trunk regions ( Figure 6—source data 3 , Figure 6—figure supplement 1—source data 2 ) . Quantification of TUNEL+ cells was done in size- and location-matched head tip areas in control and nr4A ( RNAi ) animals through the entire thickness ( DV axis ) of the animal ( Figure 6—figure supplement 1—source data 3 ) . Quantification of PCG expression domain shifts between control and nr4A ( RNAi ) animals was done in ImageJ ( at least three animals per condition ) ( Supplementary file 1D ) . Before PCG domain quantification , animal eyes were occluded by black boxes to de-identify their RNAi condition when possible . Images were then randomized and the positions of the anterior or posterior boundaries of PCG expression domains were designated by another individual drawing lines at the boundaries . Distances between the midpoint of the boundary lines to head tip , anterior edge of original eyes , or tail tip were calculated and normalized to animal length . For sFRP-1 , wnt11-1 , and wnt11-2 , curves following their expression boundaries were drawn and areas of their expression domains were calculated and normalized to the entire area of their respective animals . A one-tailed Welch’s t-test was used to test for PCG boundary recession in nr4A ( RNAi ) animals . The presence of wnt11-2 cluster of expression at the tail tip was blindly scored as present or absent , and a Fisher’s Exact Test was used for statistical analysis . For anterior-pole-to-head-tip quantifications after one week of RNAi , the distance of each notum+; collagen+ cell to the apex of the head was measured on Fiji/ImageJ . This distance was divided by the distance between the eyes within each animal for normalization ( Figure 8—source data 1 ) . Kruskal-Wallis test with Dunn’s multiple comparisons test was used for statistical analysis of anterior-pole-cell-to-head-apex distances between control and nr4A ( RNAi ) animals ( at least five animals per condition ) . For newly incorporated eye cell quantifications , all opsin+; SMEDWI-1+ cells in each eye were counted in control , 2-eye-pair nr4A ( RNAi ) animals , and 3-eye-pair nr4A ( RNAi ) animals ( Figure 9—source data 1 ) . Cells were included in the count if located within the eye itself or within approximately one cell diameter of the eye . Heads and tails of animals subjected to 2 , 3 , 4 , and 5 weeks of control or nr4A RNAi were surgically isolated and separately processed for RNA sequencing ( Figure 5A ) , as described in Bulk RNA sequencing in Materials and methods . Three biological replicates of heads and tails were collected per time point , with six animals pooled within each biological replicate . Gene expression in nr4A ( RNAi ) heads and tails were compared with gene expression in control ( RNAi ) heads and tails , respectively , within each RNAi time point , as described in Bulk RNA sequencing in Materilas and methods . Sequencing data and DESeq analysis available in NCBI/GEO ( accession GSE121048 ) . Three RNAi groups , control RNAi , control; nr4A RNAi , and foxD; nr4A RNAi were selected to test the effect of genetic pole ablation on the nr4A ( RNAi ) phenotype . Control RNAi used dsRNA against the C . elegans unc-22 transcript . Animals in the foxD; nr4A ( RNAi ) group were subjected to only foxD RNAi for 6 weeks ( two feedings/week ) before the addition of nr4A RNAi . During this initial single RNAi period , animals in control and control; nr4A ( RNAi ) groups received only control RNAi , with nr4A RNAi added to the control; nr4A ( RNAi ) group at the same time as the nr4A RNAi in the foxD; nr4A ( RNAi ) group . At the end of the single RNAi period , the anterior poles from some animals were excised with a scalpel as described for anterior pole sequencing . These pole-cut animals were allowed to recover for 3 days before initiating the double RNAi . Double RNAi was performed for a total of 10 weeks ( with two feedings/week ) for intact animals and 12 weeks for pole-cut animals , with one-half of the total amount of dsRNA being from each kind dsRNA ( see single RNAi in Materials and methods ) . Two biological replicates of double RNAi with uninjured animals and two biological replicates of double RNAi with pole-cut animals were performed . Six trunk fragments per biological triplicate were isolated for each of the RNAi groups ( control , control; nr4A , and foxD; nr4A ) . The uninjured and pole-cut animal groups were analyzed separately . Tissues were homogenized using Qiagen TissueLyser III , and RNA was extracted using TRIzol according to the manufacturer’s protocol , and synthesized into cDNA with reverse transcriptase per manufacturer’s protocol . qPCR primers for nr4A ( Supplementary file 1F ) were designed to amplify regions outside of the dsRNA sequences for the respective genes . Primers for the housekeeping gene gapdh ( Supplementary file 1F ) were used for expression normalization . Expression levels were calculated by the double delta-Ct method , with normalized nr4A expression levels in control; nr4A , and foxD; nr4A ( RNAi ) groups compared to normalized nr4A expression levels in the control ( RNAi ) group . F-ara-EdU ( Sigma ) dissolved in DMSO at 50 mg/mL was mixed with liver paste ( 3:1 liver: planarian water ) for a final concentration of 0 . 5 mg/mL . Animals were fed once with EdU and liver mix and fixed 8 days after as in in situ hybridization . Development was performed with click chemistry , as described ( Salic and Mitchison , 2008; Ji et al . , 2017 ) . ApopTag Red In Situ Apoptosis Detection Kit ( Millipore ) was used for the TUNEL assay . Animals were fixed as for in situ hybridization , incubated overnight at 37°C in terminal deoxynucleotidyl transferase diluted in reaction buffer , washed in stop/wash buffer , rinsed in PBST ( 0 . 3% TritonX ) , and blocked for 30 min in PBSTx containing 5% heat-inactivated horse serum and 5% Western Block Reagent ( Roche ) . Animals were then developed in anti-digoxigenin rhodamine conjugate diluted in block solution overnight at 4°C in the dark . Animals were given lethal irradiation dosage of 6000 rads using a Gammacell 40 dual 137cesium source . Bulk and single-cell RNAi sequencing of planarian poles , and bulk RNA sequencing of nr4A RNAi animal heads and tails . Li , DJ , McMann , CL , Reddien PW ( 2018 ) available at the NCBI Gene Expression Omnibus ( Accession no . : GSE121048 ) . | Many animals are able to regenerate tissue that has been lost through illness or injury . Flatworms called planarians have long been used to study tissue regeneration because of their remarkable ability to completely regenerate their whole body from small pieces of tissue . Furthermore , the stem cells of adult planarians continually produce new cells to replace dying cells in a process called tissue turnover . For regeneration and tissue turnover to be successful , it is important for the new cells to form in the right location in the body; for example , new eye cells need to form in the head . Genes known as position control genes are active in muscle at specific locations along the body of a flatworm to regulate both regeneration and tissue turnover . However , it was not clear how these genes coordinate with stem cells to produce new cells in the correct positions in the body . Li et al . examined the effects of a gene known as nr4A that is particularly active in muscle at the head and tail ends of planarians . Using a technique called RNA interference to decrease the activity of nr4A in planarians disrupted the patterns of tissues at each end of the flatworms . Over time , the activity of the position control genes also became restricted to locations progressively farther away from the head and tail . As a result , cells that were intended to replace tissues in the head or tail were deposited increasingly far away from these locations . For example , new eyes formed repeatedly in the planarians , with each set farther away from the head tip than the last . Li et al . propose that these disruptions of normal tissue patterning ensue because the cells that organize such patterns at the ends of the planarian ( the poles ) are themselves misplaced within the existing body pattern . The nr4A gene can be found in a wide range of animal species . Understanding how this gene affects tissue patterns in planarians could therefore also help researchers to discover how adult tissue patterns form and are maintained in animals more generally . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine"
] | 2019 | Nuclear receptor NR4A is required for patterning at the ends of the planarian anterior-posterior axis |
Single-cell analysis has revealed that transcription is dynamic and stochastic , but tools are lacking that can determine the mechanism operating at a single gene . Here we utilize single-molecule observations of RNA in fixed and living cells to develop a single-cell model of steroid-receptor mediated gene activation . We determine that steroids drive mRNA synthesis by frequency modulation of transcription . This digital behavior in single cells gives rise to the well-known analog dose response across the population . To test this model , we developed a light-activation technology to turn on a single steroid-responsive gene and follow dynamic synthesis of RNA from the activated locus .
Steroid receptors coordinate a diverse range of responses in higher eukaryotes and are involved in a wide range of human diseases ( Gronemeyer et al . , 2004 ) . Steroid receptor response elements are present throughout the human genome and modulate chromatin remodeling and transcription in both a local and long-range fashion ( John et al . , 2011 ) . As such , steroid receptor-mediated transcription is a paradigm of genetic control in the metazoan nucleus . Moreover , the ligand-dependent nature of these transcription factors makes them appealing targets for therapeutic intervention , necessitating a quantitative understanding of how receptors control output from target genes . The classic sigmoidal dose response of steroid-regulated gene products belies the complexity of a system which relies on an intricate , multi-step sequence of events to initiate transcription from a target gene ( McKenna et al . , 2009 ) . Depending on the particular steroid receptor , a partial list of events required to activate the target gene includes ligand-binding , receptor dimerization , nuclear translocation , eviction of co-repressors ( i . e . , histone deacetylases ) , recruitment of chromatin modifying enzymes ( histone acetyltransferases , ATP-dependent remodeling enzymes , methyltransferases ) , and eventually recruitment of the basal transcription machinery . Despite this complexity , the dose-response is deceptively simple: it often takes the form of a simple Hill function with a coefficient of unity , which has traditionally been interpreted to imply that ligand-binding is the only rate-limiting step in the activation pathway ( Ong et al . , 2010 ) . This description models the dose response as a continuum where each cell transcribes RNA at rates proportional to the dosage level . However , single-cell studies of gene expression demonstrate that this description is incorrect . First , gene expression is not uniform over a population for a given dose , but shows variation from cell to cell , an observation which was first made for a steroid-responsive MMTV reporter gene ( Ko et al . , 1990 ) . The authors demonstrated that the observed analog dose response was in fact a digital dose response ( on or off ) when viewed in single cells . Likewise , previous studies initially demonstrated that enhancers increase the probability of activation of a cell but not the strength of activation in the cell ( Moreau et al . , 1981; Weintraub , 1988; Moon and Ley , 1991; Walters et al . , 1995 ) . Second , expression is only a snapshot of temporally evolving gene activity ( Ross et al . , 1994; White et al . , 1995; Harper et al . , 2011; Suter et al . , 2011 ) . Thus , a cell is counted as activated or not activated dependent on the moment it is observed . Thus far , all single-cell measurements on metazoans suggest that genes are transcribed in ‘bursts’ of transcription , meaning that short periods of RNA synthesis are interspersed by long periods of inactivity . The causes of transcriptional bursting are unknown . Third , since molecules involved in regulating transcription are usually present at low copy number , this leads to stochastic fluctuations ( ‘noise’ ) and hence gene expression variation across the population ( Larson et al . , 2009 ) . Finally , dynamic interactions between upstream regulators and chromatin add another level of complexity to the molecular events occurring during transcriptional activation ( McNally et al . , 2000; Darzacq et al . , 2009; Ong et al . , 2010 ) . Under such conditions , the observed dose response does not result solely from ligand-binding but rather the composite result derived from many coupled reactions . In summary , population models of gene activation are too coarse to explain activation in single cells . Moreover , tools do not exist whereby the activity of single genes in single cells can be directly manipulated and measured . In this work , we describe an approach for activating a steroid-receptor in order to achieve high temporal and spatial precision and to measure the activity of a responsive gene in the same cell over time . In contrast to the ensemble approach derived from observations of cell populations , we have developed a single-molecule kinetic approach for interrogation of a single gene . This approach is based on photoactivation of a steroid receptor ligand followed by observation of pre-mRNA synthesis at an active locus . The system consists of an exogenous reporter gene under control of the ecdysone receptor which is activated by the agonist ponasterone A ( No et al . , 1996 ) . The real-time behavior of the gene is visualized using a bacteriophage capsid protein which binds MS2 RNA stem loops with high affinity to label nascent pre-mRNA in living cells ( Bertrand et al . , 1998 ) . We demonstrate experimentally how the ensemble steroid dose response arises from the stochastic behavior of individual genes . These results suggest that the response element controls the frequency of gene activity but affects neither the duration of the active period nor the actual rate of transcripts produced during an active period . By using a caged ligand that could be uncaged by a light pulse ( Lee et al . , 2009; Lin et al . , 2002 ) we measured the impulse-response of the gene and determined that a single pulse of active ligand resulted in a corresponding burst of polymerase activity several hours later . Further , this photoactivatable ligand has the property of being an antagonist in the caged state and an agonist in the uncaged state , enabling a precise window for kinetic perturbation in single cells . Thus , we were able to propose and validate a stochastic model of steroid-receptor activity for a reporter gene which provides a new framework for studying this ubiquitous mechanism of eukaryotic gene regulation .
First , we directly observed RNA synthesis at an active locus in a clonal cell line using single-molecule live-cell microscopy . Consistent with previous studies which directly observe transcriptional activity of single genes in eukaryotes we observed transcriptional bursting ( Chubb et al . , 2006; Stevense et al . , 2010; Yunger et al . , 2010 ) . We then measured the bursting dynamics in the microscope for three different concentrations of the steroid Ponasterone A ( PA ) ( 3 . 125 μM , 12 . 5 μM , 50 μM , control: 0 . 0 μM ) , which is the high affinity ligand for the chimeric ecdysone receptor . An example time series is shown for 50 μM PA ( Figure 2A–F , Videos 1–5 ) . The first transcription sites were visible several hours after induction ( Figure 2—figure supplement 1 ) . At steady state ( 24 hr after induction ) for each dose , single gene transcription sites were captured at 15 min intervals for ∼ 15 hr , resulting in representative transcription intensity traces shown in Figure 2E , F ( red , green lines , Video 4 ) . As is evident from the images and the intensity analysis , individual cells show punctuated periods of transcriptional activity . The intensity traces were fit using a two-state hidden Markov model ( Figure 2E , F , black lines , 50 μM; Figure 2—figure supplement 2 , 12 . 5 μM , 3 . 125 μM ) ( Lee , 2009 ) , allowing the distribution of on-times and off times to be determined ( Figure 2G , H ) . We observed that the duration of inactivity ( off-time ) decreased in a dose-dependent manner ( Figure 2I , gray bars ) . However , the duration of transcription activity ( the on-time ) did not significantly change over the dose response curve ( Figure 2I , black bars ) . Moreover , the average peak brightness did also not change significantly with [PA] ( Figure 2J ) . In the case of treatment with a vehicle ( [PonA] = 0 . 0 μM ) , occasional spurious spots are detected by the tracking algorithm ( Figure 2I , 0 . 0 μM , black bar ) , and the off-time is statistically indistinguishable from the duration of the entire time-lapse video ( Fig . 2I , 0 . 0 μM , gray bar , compared to red-dashed line ) . Since the brightness of the nascent RNA signal reflects the number of individual RNA at the site of transcription ( Zenklusen et al . , 2008; Larson et al . , 2011 ) ( Figure 1F ) , these data indicate the average number of polymerases which fire during an active period is insensitive to [PA] . This live-cell analysis was also repeated for an additional clone with equivalent results ( Figure 2—figure supplement 3 and below ) . In summary , direct measurements of transcription activity suggest that changing steroid levels affects the frequency of active periods but neither the duration of the active period nor the number of polymerases fired during the active period . Even for this highly-induced artificial reporter gene , the gene is active approximately 50% of the time at saturation ( average on time = 38 ± 6 min; average off time = 35 ± 4 min ) and shows a total gene cycling time of approximately 80 min . 10 . 7554/eLife . 00750 . 004Figure 2 . Dynamic observation of integrated reporter genes shows transcriptional bursting . ( A–D ) Time-lapse images of active transcription in two separate nuclei . TS are visualized as punctate fluorescent spots where MS2-YFP has coalesced onto multiple MS2-binding sites in nascent pre-mRNA . Scale bar = 4 μm . ( E and F ) Quantification of TS intensity for the upper cell/TS ( red ) and lower cell/TS ( green ) , respectively . The left axis is the integrated intensity of the TS; the right axis is the normalized intensity which is used to fit the intensity trace to a hidden Markov model ( black lines ) ( Lee , 2009 ) . Examples of on and off states are designated by the arrows on panel E . ( G and H ) Histogram of on and off times , respectively . The red lines are exponential fits to the experimental distribution with decay time of 36 ± 6 min and 35 ± 4 min in panels G and H , respectively ( N = 40 cells ) . [PA] = 50 μM , panels A–H . ( I ) Average on ( black ) and off ( gray ) duration for four different [PA] obtained from fitting experimental data to an exponential distribution ( Figure 2—figure supplement 2 ) . The red dashed line indicates the duration of the experiment , which sets the upper limit for off-time values . ( J ) Average TS intensity for four different [PA] ( N = 40 cells for each [PA] ) . Error bars are the SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 00410 . 7554/eLife . 00750 . 005Figure 2—figure supplement 1 . Time-dependent induction of reporter gene . Cells are induced at t = 0 with 50 μM Ponasterone A at 37°C . z-stacks are acquired every hour , and the number of cells with a transcription site is recorded . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 00510 . 7554/eLife . 00750 . 006Figure 2—figure supplement 2 . Dynamic observation of integrated reporter genes shows transcriptional bursting . ( A and B ) Quantification of TS intensity for a cell treated with 12 . 5 μM and 3 . 125 μM PA , respectively . The left axis is the integrated intensity of the TS; the right axis is the normalized intensity which is used to fit the intensity trace to a hidden Markov model ( black lines ) . ( C and D ) Histogram of on- and off-times for 12 . 5 μM PA , respectively . ( E and F ) Histogram of on- and off-times for 3 . 125 μM PA , respectively . The red lines are exponential fits to the experimental distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 00610 . 7554/eLife . 00750 . 007Figure 2—figure supplement 3 . Dose dependence of on- and off- times for an alternative single cell clone . Average on ( black ) and off ( gray ) duration for three different [PA] obtained from fitting experimental data to an exponential distribution . Error bars are the SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 00710 . 7554/eLife . 00750 . 008Video 1 . Time-lapse sequence of reporter gene response to 50 μM PA shows a single pulse of transcription . The nascent transcription site is visualized as the coalescence of MS2-YFP coat protein on newly synthesized pre-mRNA . Each frame is the maximum projection of 18 z-steps acquired at 0 . 5 μm intervals . Exposure time = 400 ms . Frame interval: 15 min . Total duration: 15 hr . T = 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 00810 . 7554/eLife . 00750 . 009Video 2 . Time-lapse sequence of reporter gene response to 50 μM PA shows a gene which is on for almost the entire duration of the video . The nascent transcription site is visualized as the coalescence of MS2-YFP coat protein on newly synthesized pre-mRNA . Each frame is the maximum projection of 18 z-steps acquired at 0 . 5 μm intervals . Exposure time = 400 ms . Frame interval: 15 min . Total duration: 15 hr . T = 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 00910 . 7554/eLife . 00750 . 010Video 3 . Time-lapse sequence of reporter gene response to 50 μM PA for multiple cells showing uncorrelated transcription dynamics . The nascent transcription site is visualized as the coalescence of MS2-YFP coat protein on newly synthesized pre-mRNA . Each frame is the maximum projection of 18 z-steps acquired at 0 . 5 μm intervals . Exposure time = 400 ms . Frame interval: 15 min . Total duration: 15 hr . T=37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 01010 . 7554/eLife . 00750 . 011Video 4 . Time-lapse sequence of reporter gene response to 50 μM PA for cells in Figure 2 . The nascent transcription site is visualized as the coalescence of MS2-YFP coat protein on newly synthesized pre-mRNA . Each frame is the maximum projection of 18 z-steps acquired at 0 . 5 μm intervals . Exposure time = 400 ms . Frame interval: 15 min . Total duration: 15 hr . T = 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 01110 . 7554/eLife . 00750 . 012Video 5 . Post-processing time-lapse sequence of reporter gene response to 50 μM PA for cells in Figure 2 . The nascent transcription site is visualized as the coalescence of MS2-YFP coat protein on newly synthesized pre-mRNA . Each cell is individually segmented and re-cropped to a separate image file , resulting in a time-lapse sequence where the cell is always centered in the frame . Each frame is the maximum projection of 18 z-steps acquired at 0 . 5 μm intervals . Exposure time = 400 ms . Frame interval: 15 min . Total duration: 15 hr . T = 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 012 We then repeated this experiment on the same plasmid that was transiently transfected instead of genomically integrated by lentivirus ( Figure 3 , Videos 6 and 7 ) . The transfected template , which is present at many copies in the nucleus and which is not integrated into chromatin , shows an extremely rapid induction ( <20 min ) and robust transcription throughout the time series . We were unable to detect transcriptional bursting . These data indicate that chromatin integration is a primary determinant of transcription kinetics for the hormone response , as has been suggested in previous studies ( Archer et al . , 1992 ) . 10 . 7554/eLife . 00750 . 013Figure 3 . Dynamic observation of transiently transfected reporter genes shows fast induction and no bursting . Time-lapse images of transiently transfected cells induced with 50 μM PA at t = 0 min . Each cell contains multiple copies of the reporter plasmid , visible as diffraction limited spots ( Video 6 ) . Scale bar = 4 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 01310 . 7554/eLife . 00750 . 014Video 6 . Time-lapse sequence of a transfected reporter . Multiple nascent transcription sites are visualized as the coalescence of MS2-YFP coat protein on newly synthesized pre-mRNA emerging from the plasmids . Each frame is a single z-plane . Exposure time = 200 ms . Frame interval: 10 min . Total duration: 2 hr . T = 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 01410 . 7554/eLife . 00750 . 015Video 7 . Time-lapse sequence of a transfected reporter . Multiple nascent transcription sites are visualized as the coalescence of MS2-YFP coat protein on newly synthesized pre-mRNA emerging from the plasmids . Each frame is a single z-plane . Exposure time = 200 ms . Frame interval: 30 min . T = 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 015 Using the lentivirus approach results in quasi-random insertion of the transgene throughout the genome . Since the kinetics of transcription depend strongly on transgene insertion ( Figure 3 ) , we sub-cloned three different lines , measured the dose response to [PA] , and fit these data to a three-parameter Hill equation for the steroid response ( Figure 4A ) : ( 1 ) activity=Amax[PA]hEC50+[PA]hwhere Amax is the amount of reporter mRNA present at saturation , [PA] is the concentration of ligand , EC50 is the half-maximal response , and h is the Hill coefficient ( Ong et al . , 2010 ) . Although the saturation value for the number of mRNA/cell differed substantially for each insertion site ( Figure 4A , 960 ± 60 mRNA/cell , 430 ± 50 mRNA/cell , 180 ± 10 mRNA/cell , blue , green , red curves respectively ) , the Hill coefficient did not change significantly ( 1 . 2 ± 0 . 2 , 1 . 3 ± 0 . 3 , 1 . 2 ± 0 . 2 ) , nor did the EC50 ( 8 . 0 ± 1 . 4 μM , 10 ± 3 . 2 μM , 9 . 1 ± 1 . 5 μM ) . We also compared the protein dose response to the RNA dose response for one of the cell lines and observed similar behavior ( Figure 4B ) . Since the population dose response for this reporter gene is statistically indistinguishable from a first-order Hill response , we also show a fit to a two-parameter Hill equation with the Hill coefficient fixed at unity ( Figure 4B , black line , Amax = 190 ± 10 mRNA , EC50 = 9 . 6 ± 2 . 0 μM ) . In summary , the population dose response is well-described by a first-order Hill response function where the Hill coefficient and the EC50 are invariant with insertion site , but the saturation level ( Amax ) shows fivefold variation among different clones . 10 . 7554/eLife . 00750 . 016Figure 4 . Steady state distribution of mRNA from the steroid-activated reporter gene indicates frequency modulation of transcription . ( A ) Dose response of the reporter gene . Total cellular reporter mRNA is plotted as a function of [PA] for three different clones isolated from a population of lentiviral-infected cells ( clone 1 , red; clone 2 , green; clone 3 , blue ) . Each data point is the average value of mRNA/cell determined by automated image segmentation and spot counting ( N = 60 cells , error bars are SEM ) . The fits are three-parameter Hill functions: Amax = 960 ± 60 mRNA/cell , 430 ± 50 mRNA/cell , 180 ± 10 mRNA/cell; h = 1 . 2 ± 0 . 2 , 1 . 3 ± 0 . 3 , 1 . 2 ± 0 . 2; EC50 = 8 . 0 ± 1 . 4 μM , 10 ± 3 . 2 μM , 9 . 1 ±1 . 5 μM , blue , green , red curves respectively . ( B ) Dose response of mRNA and protein as function of PA concentration for clone 1 . The left axis ( red circles ) is the number of mRNA counted per cell in the microscope as in panel A . The error bars are the SEM for N = 60 cells for each concentration . The right axis is the normalized protein level , determined from the quantification of the Western blot shown above . α-Tubulin is used as the loading control . The data are fit with a two-parameter Hill function with h = 1 . 0 ( black line ) : Amax = 190 ± 10 mRNA; EC50 = 9 . 6 ± 2 . 0 μM . ( C ) Stochastic model of gene induction and the resulting polymerase density . In the random telegraph model ( upper panel ) , the gene exists in an inactive state which is non-permissive to transcription or an active state from which transcripts are produced ( on-state indicated by red line ) . The rate of transition to the active state is a; the rate of transition to the inactive state is b; the rate of initiation from the active state is c . Individual initiation events are indicated by vertical green lines . Lower panel: the RNA polymerase II loading that would result from the telegraph process shown in the upper panel . Each initiation event results in the loading of a polymerase , and that polymerase will have a dwell time determined by the time necessary to synthesize the nascent transcript . Note that even when the gene is in the active state , it is possible that no polymerases are present ( comparing black occupancy trace with red gene activity trace ) . ( D–F ) Fluorescence in situ hybridization at 3 . 125 , 12 . 5 , and 50 μM induction with PA . Gray = Cy3 oligos; blue = DAPI . Scale bar = 4 μm . ( G–L ) The steady-state distribution of mRNA/cell for 1 . 56 μM , 3 . 125 μM , 6 . 25 μM , 12 . 5 μM , 25 . 0 μM , and 50 μM , respectively . The bin size of the histogram is 100 mRNA . The theory is the full-solution to the Master Equation for the scenario shown in panel C ( ‘Materials and methods’ ) . The on rate a determined from the fit varies with the dose , but the parameters b/d and cd are kept constant at 1 . 5 and 340 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 016 We next sought to build and test a quantitative kinetic model capable of elucidating how the classical analog dose response in a population ( Figure 4A , B ) relates to the time-dependent digital behavior observed directly in single cells ( Figure 2 ) . We used the ‘Random Telegraph model’ , which is a paradigm for stochastic transcription and quantitatively describes the distribution of RNA/cell in a variety of organisms ( Peccoud and Ycart , 1995; Larson et al . , 2009 ) . The key feature of this model is the hypothesis that the gene toggles between an active state where transcripts can be synthesized and an inactive state where no transcripts are produced . It is important to note that the gene can be ‘active’ even when no nascent RNA is detected , owing partly to the stochastic initiation events within an active state ( Figure 4C ) and partly to the placement of the MS2 cassette in the middle of the gene ( Figure 1A ) . The biochemical determinant ( s ) of this active state is unknown and is likely to vary between genes . Consequently , it is not possible at present to do a direct measurement of the active state . However , the distribution of total cellular reporter mRNA reflects the dynamics of this active-inactive switching , so we measured the steady state distribution of total cellular reporter mRNA over a population using single-molecule FISH in fixed cells . For each cell line , the variation of total cellular mRNA within a population is indicative of infrequent bursts of transcriptional activity ( Golding et al . , 2005; Chubb et al . , 2006; Raj et al . , 2006; Voss et al . , 2006; Zenklusen et al . , 2008 ) ( Figure 4D–F ) . At low PA induction levels ( 3 . 12 μM , Figure 4D ) , there is a large heterogeneity of cellular mRNA in the population: some cells contain dozens of mRNAs and others have none . Transcription sites are infrequent . At medium induction levels ( 12 . 5 μM , Figure 4E ) , most cells contain mRNA , and many are actively transcribing . At high levels of induction ( 50 μM , Figure 4F ) , cells contain a high abundance of mRNA in the cytosol , and most cells show an active transcription site in the nucleus . At each concentration of the steroid , we fit the full distribution of mRNA/cell to the steady-state solution for the random telegraph model ( Figure 4G–L , N = 60 cells at each dose ) . The steady state distribution function used for fitting is a function of three unitless kinetic parameters ( a/d , b/d , c/d ) corresponding to the rate of activation ( a ) , the rate of inactivation ( b ) , and the rate of firing once the gene is active ( c ) , each normalized by the RNA decay rate , d ( Figure 4C ) . These parameters correspond to the frequency of bursts , the duration of bursts , and the initiation rate from a burst , respectively . At steady state , the mean level of RNA 〈N〉 is: ( 2 ) 〈N〉=cd ( aa+b ) where a/ ( a+b ) corresponds to the fraction of time the gene spends in the active state . There is a functional similarity between the stochastic random telegraph equation ( Equation 2 ) and the deterministic Hill equation ( Equation 1 ) : Amax corresponds to c/d , EC50 corresponds to b , and [PA] corresponds to a . This correspondence indicates that non-cooperative behavior for the hormone response ( h = 1 , Equation 1 ) is mathematically equivalent to frequency modulation . In other words , the rate of switching to the active state ( a ) is the only variable which changes with ligand concentration . In Figure 4G–L we show the distribution of mRNA/cell compared to the theoretical fit arising from the telegraph model as a function of dose . Indeed , these data are consistent with a model where the only variable which varies with dose is the rate of activation ( a ) , while the other parameters ( b , c ) are globally fit for the entire dose response for a particular cell line ( b/d = 1 . 5 , c/d =340 ) . When the same analysis is carried out on cell lines with different insertion sites ( i . e . , Figure 4A ) , the dose-dependent activation rate ( a ) and the dose-independent inactivation rate ( b ) remain invariant , but the initiation rate ( c ) changes with the cell line . Thus , the absolute number of polymerases loaded during an active period does not depend on steroid level but rather some extrinsic factor such as local chromatin environment or abundance of some component of the core transcriptional machinery . This is evident when comparing the episomal genes , which do not show frequency modulation . In summary , both fixed-cell and live-cell single-molecule measurements are consistent with a model where the steroid is a robust frequency modulator . We emphasize that the frequency change observed here is in response to changes in the time-invariant level of steroid . Conversely , there is a growing body of evidence that the dynamics of the upstream activator itself may in fact carry biological information and modulate downstream expression levels ( Cai et al . , 2008; Ashall et al . , 2009; Stavreva et al . , 2009; Purvis et al . , 2012 ) . To investigate the temporal relationship between steroid availability and the kinetics of transcription , we required a ligand that could be activated at a precise time to induce transcription . We utilized a caged PA that consists of PA modified at a critical OH residue with a photolabile organic ligand ( Figure 5A ) ( Lin et al . , 2002 ) . 10 . 7554/eLife . 00750 . 017Figure 5 . Caged PA is an ecdysone receptor antagonist . ( A ) Molecular structures of the two forms of caged PA: DMNB-PonA and CNB-PonA . ( B ) Dose dependent competition between caged PA and unmodified PA . The competition experiment consists of an overnight incubation in various doses of caged PA ( CNB PonA = red triangle; DMNB-PonA = gray triangle ) , followed by incubation with un-modified PA . The expression is quantified by Western blot . The induced CFP-SKL marker is the lower green band ( 24 kD ) , the constitutively expressed MS2-YFP is the upper green band ( 45 kD ) , and the α-tubulin is the red band ( 50 kD ) . ( C ) Quantification of caged PA inhibition ( gray = DMNB-PonA; red = CNB-PonA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 01710 . 7554/eLife . 00750 . 018Figure 5—figure supplement 1 . Caged-PA can be switched from inactive to active with UV photolysis . ( A ) Dose-response of CFK-SKL production as measured by western blot for PA ( black circles and line ) and the DMNB-PA ( red-circles and line ) . The solid lines are fit to a first-order Hill response function . ( B ) The caged-PA can be activated with UV light . Caged PA was irradiated in a cuvette with 300 , 000 pulses at 75 μJ/pulse using the same laser as used for in vivo uncaging . The photolyzed sample was then added the culture dish for 1 hr and then removed . The cells were washed 2 × in fresh media and allowed to incubate overnight . CFP-SKL production was assayed by Western blot . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 01810 . 7554/eLife . 00750 . 019Figure 5—figure supplement 2 . Caged-PA competes with unmodified PA . Cells are incubated with 5 μM PA overnight , resulting in the appearance of transcription sites . At t = 0 hr , time-resolved measurements of transcription site intensity begin . At t = 2 hr , caged PA is added at 50 μM , and transcription site intensity is monitored subsequently for 9 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 019 We observed empirically that we were unable to activate single genes in the presence of the caged reagent and hypothesized that the caged PA possessed antagonist properties . We therefore examined the dose-dependent induction properties of the reporter gene in the presence and absence of the caged reagent by performing an in vivo competition experiment . First , we confirmed that both forms of the caged PA were inactive by measuring a dose response of the caged ligands ( Figure 5—figure supplement 1 ) . Second , the cells were incubated with two different forms of the caged PA ( DMNB-PonA and CNB-PonA ) ( Figure 5A ) . After overnight incubation , the cells were exposed to 25 μM of the active , unmodified PA for 30 min . The cells were then washed and returned to the incubator for 8 hr to allow the buildup of protein ( Figure 5B ) . At 100 μM of DMNB-PA , we observed a fivefold reduction in protein levels , indicating that the caged ligand blocks the activity of the active , uncaged form in vivo ( Figure 5C , gray , Figure 5—figure supplement 2 ) . When the DMNB-PonA and uncaged PA are present in equimolar amounts , the competition is alleviated , and the protein output from the reporter gene is close to the levels observed in the absence of the caged ligand . Thus , we show that this ligand has the unique property of being an agonist in the uncaged state and an antagonist in the caged state . The benefit of a reagent that can be switched from repressive to activating is that the effect of small amounts of the agonist are suppressed by the presence of the antagonist . Such a feature is especially important for a reagent that is membrane-permeable and might diffuse to neighboring cells in the tissue . Thus , the caged compound has built-in protection against ‘leakiness’ and is capable of acting in a highly localized spatial manner when uncaged . The method of activation , therefore , has to be all or none: the balance between agonist/antagonist must be shifted past a threshold in order for gene activation to occur . A typical experimental design is schematized in Figure 6A . The nucleus is activated with a UV laser at 349 nm . The above-threshold dose of uncaging light ( 40 pulses , 100 Hz , 2 J/cm2 total dose at the sample ) is delivered in approximately 400 ms over the area of the nucleus ( ∼80 μm2 ) . Now , all the receptors in the nucleus are active , leading to the initiation of transcription at the single locus . Concurrently , any activating ligand that diffuses away encounters neighboring cells that are in the repressed state due to the presence of the antagonist . Once the activation occurs in the target cell , the diffusion into the cell of unactivated antagonist coupled with the rapid turnover of receptor-bound agonist ( Stavreva et al . , 2004 ) results in antagonist-bound receptors that bind to the response element and repress transcription . 10 . 7554/eLife . 00750 . 020Figure 6 . Photoactivation of single genes in vivo relates agonist kinetics to transcription dynamics . ( A ) Schematic uncaging experiment in tissue culture . The target cell is photolyzed with a laser that has the physical dimensions of a single nucleus ( dotted circle ) . The transcription site in that cell becomes active , but transcription sites in neighboring cells are still repressed due to the antagonistic effects of caged PA . Thus , even though PA will diffuse through the membrane ( indicated by hazy blue circle ) , the neighboring cells will not activate . ( B–E ) Time-lapse images of expression for an uncaged cell ( ligand = DMNB-PonA ) . The uncaging spot is designated by a dotted blue circle . The transcription site in the activated cell is evident at 165 min and persists for 60 min . By 465 min , the peroxisomes are visible . Each frame is a z-stack of 30 images taken at 0 . 5 μM increments . Frame interval = 15 min . Images are maximum projected z-stacks . To maintain viability , only the YFP channel is imaged during acquisition , with CFP imaging utilized as an endpoint assay . Scale bar = 4 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 020 The time evolution of a typical activated cell is shown in Figure 6 ( Video 8 ) . The cell is uncaged at t = 0 ( Figure 6B ) ; the transcription site is visible at 2 . 5 hr ( Figure 6C ) and persists in the on state for 1 . 5 hr afterwards ( Figure 6D ) , consistent with the duration of the on-state measured previously ( Figure 2G ) . At 15 hr , the CFP peroxisomes are observed to assay whether the cell has activated ( Figure 6E ) . In this example , only the central cell was photolyzed , and is the only cell that displays expression of the CFP protein construct . By contrast , the two non-photolyzed neighboring cells show no evidence of CFP expression . The photolysis protocol above results in: 1 ) an average 106 +/− 36 min between uncaging and the appearance of a transcription site which then persisted for 35 +/− 12 min , 2 ) 40% success rate for protein expression , and 3 ) a single on-state in 70% of those cells which activated , with the remainder showing either no detectable on-period ( 13% ) or multiple periods ( 17% ) ( N = 30 ) . Moreover , we also observe cell division after uncaging , which is a measure of viability after UV photolysis ( Video 9 ) . In sum , the photoactivation data suggest that a single pulse of ligand results in a single active period of transcription . 10 . 7554/eLife . 00750 . 021Video 8 . Time-lapse sequence of reporter gene uncaging . The nascent transcription site is visualized as the coalescence of MS2-YFP coat protein on newly synthesized pre-mRNA . Cells were incubated in 100 μM DMNB-PA overnight . The media were then removed , the cells rinsed 3 × with fresh media without DMNB-PA , and then irradiated with 40 pulses at 100 Hz . Each frame is the maximum projection of 28 z-steps acquired at 0 . 5 μm intervals . Exposure time = 200 ms . Frame interval: 15 min . Total duration: 8 hr . T = 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 02110 . 7554/eLife . 00750 . 022Video 9 . Time-lapse sequence of reporter gene uncaging . The nascent transcription site is visualized as the coalescence of MS2-YFP coat protein on newly synthesized pre-mRNA . Cells were incubated in 100 μM DMNB-PA overnight . The media were then removed , the cells rinsed 3 × with fresh media without DMNB-PA , and then irradiated with 40 pulses at 100 Hz . In this instance , the uncaged cell proceeded through mitosis , and only one of the daughter cells shows activation . Each frame is the maximum projection of 28 z-steps acquired at 0 . 5 μm intervals . Exposure time = 200 ms . Frame interval: 30 min . Total duration: 15 hr . T = 37° . DOI: http://dx . doi . org/10 . 7554/eLife . 00750 . 022
The work presented here addresses the specific question as to how transcription factor availability translates into transcriptional activity . By a variety of approaches , we show that the frequency of initiation is the variable that determines RNA levels , but only when the gene is integrated into the chromatin . By measuring the synthesis of pre-mRNA directly , we connected the dynamics of the gene with the availability of an upstream transcription factor , the ligand-bound steroid receptor . We demonstrated how steroid receptors , which are present in all metazoans , can control transcription through frequency modulation . This digital frequency modulation of transcription in single cells manifests as an analog dose response in cellular mRNA and protein in a population . Importantly , the ability to activate the transcription factor ( steroid receptor ) using light provided a precise temporal start to measure the time needed for the gene to respond , an unexpectedly long time ( 2 hr ) . Hence our observations clarify the mechanics behind these observations and extend the understanding to the intrinsic cycling on and off for a gene . One dose of a transcription factor yields one cycle of transcription . The precise timing of the transcriptional response places constraints on the regulation of the gene: one can infer regulatory principles based on the observed synthesis of pre-mRNA ( Pedraza and Paulsson , 2008 ) . A stochastic model of kinetic rates allowed us to break down the steroid transcriptional response into distinct processes , which could be measured independently in the microscope . Although a number of studies have inferred transcriptional bursting by measuring events which are downstream of transcription , such as cellular RNA or protein production , ( Blake et al . , 2003; Raser and O’Shea , 2004; Becskei et al . , 2005; Bar-Even et al . , 2006; Newman et al . , 2006; Raj et al . , 2006; Zenklusen et al . , 2008; Singh et al . , 2010; Batenchuk et al . , 2011; Skupsky et al . , 2010; So et al . , 2011; Suter et al . , 2011 ) , we are able to unambiguously resolve the properties of individual bursts and relate these measured kinetic steps to steroid abundance . For this ecdysteroid-responsive reporter gene in mammalian cells , the time between on-states decreased at higher doses of PA . The duration of the active state and the transcription rate from the active state were invariant . However , the ‘on’ times and ‘off’ times were both exponentially distributed ( Figure 2G , H ) , which suggested that both processes were determined by a single rate-limiting step . The simplest biochemical model to explain this result is that the state of the gene is determined entirely by the on rate and off rate of the active ligand-receptor complex from the response element . The corollary to this model is that the actual transcription rate ( c ) from the active state is determined by factors other than the response element . For example the ability of trans-acting factors to access cis-acting sequences in the core promoter may determine the firing rate from the active state . In agreement with this supposition , the dynamic frequency modulation by the steroid was invariant with the insertion site , but the transcription rate varied over fivefold with insertion site . In this way , genetic regulation can be conceptualized as the superposition of individual parts , with distal enhancers and response elements controlling frequency of activation and proximal promoters determining the firing rate . Future experiments will be able to directly test this hypothesis using further development of the single-molecule approaches described here . This simple model must address the fact that the exchange rate of SRs on chromatin has been observed to be extremely dynamic . Fluorescence photobleaching recovery measurements on tandem gene arrays in single cells record dwell times on the order of seconds to minutes ( Darzacq et al . , 2009 ) , while the duration of the active transcriptional state in this study is on the order of an hour . One possibility is that SR binding is actually much longer , and FRAP data on arrays is dominated by non-specific interactions and is not capable of detecting the small percentage of receptors which might be stably bound . Another possibility is that SR binding is in fact short-lived but initiates a cascade of events involved in activation of the gene ( Herschlag and Johnson , 1993; Hager et al . , 2006; Metivier et al . , 2006 ) . In support of this latter view , we observe that there was a delay between photoactivation of the ligand and the appearance of the first transcription sites ( Figure 6 ) . This delay was comparable in size to the average duration of the off period . In contrast , a non-chromatinized or weakly chromatinized template showed extremely rapid induction ( <20 min ) . It is plausible that chromatin remodeling is inefficient at this reporter gene , possibly because the promoter lacks cis-acting elements which might aid in a more rapid chromatin remodeling response . Taken together , these results support the notion that opening of chromatin interposes a delay between the docking of active ligand-receptor complex on the response element and the eventual synthesis of pre-mRNA ( Janicki et al . , 2004 ) . The synergy between DNA binding factors and chromatin modifying proteins is essential for control of gene expression ( Segal and Widom , 2009 ) , and the role of chromatin in a bimodal model of gene regulation has been proposed previously ( Archer et al . , 1992; Di Croce et al . , 1999 ) . Moreover , recent evidence points strongly toward chromatin as directly controlling the duration of the on and off states . Histone methylation has been implicated in the transmission of transcriptional frequency between mother and daughter cells ( Muramoto et al . , 2010 ) . Manipulation of histone acetylation changes transcriptional cycle timing ( Harper et al . , 2011; Suter et al . , 2011 ) . Histone deacetylases and nucleosome-remodeling complexes are cyclically recruited during transcriptional activation of an estrogen-responsive gene ( Métivier et al . , 2003 ) . Importantly , FRAP studies on histones and genome-wide pulse-chase measurements on histones suggest a timescale of turnover ( ∼hour ) that is more consistent with the time scales of transcriptional activity observed in this study ( Kimura and Cook , 2001; Dion et al . , 2007; Deal et al . , 2010 ) . The bimodal ‘telegraph’ model is likely a simplification of the number of biological steps involved in gene activation , but the quantitative agreement between this model and our data suggests that at steady state , the requirements for re-activating a gene that has cycled off may depend on only a few rate-limiting steps . In summary , we have described a comprehensive approach for unraveling the dynamic behavior of genetic control in single cells . By using a combination of nascent RNA visualization , single molecule microscopy , computational modeling , and light activation , we were able to construct a quantitative model of steroid-receptor mediated activation in which SRs control the frequency of the transition to an active transcriptional state in a dose-dependent manner . This model is based on a stochastic , kinetic description that reflects the molecular events occurring in single-cells .
The chimeric ecdysone receptor ( pERV3 , Stratagene , Agilent , Santa Clara , CA ) was introduced into U2-OS cells by nucleofection ( Amaxa , Lonza , Allendale , NJ ) , and stable insertions were selected by screening for G418 resistance followed by isolation of single colonies . Response to ecdysteroids was assayed using a luciferase reporter ( pEGSH , Stratagene ) transiently transfected into resulting cell lines . The most responsive clone was then used for all subsequent cell line derivation . The reporter gene ( Figure 1A ) was cloned into the pHAGE lentiviral vector backbone described elsewhere ( Mostoslavsky et al . , 2006 ) . Viral particles were produced from 293-T packaging cells in two 15 cm plates . The supernatant was harvested for three consecutive days . The collected supernatant was spun at max speed in a clinical centrifuge to pellet large debris and then filtered through a 0 . 45 μm filter before transfer to a sterile ultracentrifuge tube SW28 . Viral particles were pelleted by spinning at 100 , 000 × g for 1 . 5 hr at 4°C . The entire collected supernatant was used to infect one dish of 5000 cells , resulting in infection efficiencies of ∼90% . The MS2 coat protein was also cloned into the same pHAGE backbone . The coat protein is driven off the human ubiquitin C promoter , contains a nuclear localization sequence and was described previously ( Fusco et al . , 2003 ) . Lentivirus collection and infection were carried out according to the above protocol . Unless otherwise stated , U2-OS cells were cultured in low-glucose DMEM ( Gibco , Life Technologies , Grand Island , NY ) supplemented with 10% FBS and 1% penicillin/streptomycin . Reporter RNA was visualized by hybridization of a 20-mer ssDNA oligo labeled with Cy3 at both ends to the MS2 repeats . Therefore , each 20-mer contains two dye molecules . The protocol for RNA FISH has been described elsewhere ( Femino et al . , 1998 ) and is used here with slight modification . Briefly , the coverslips with adherent U2-OS cells are washed three times with Hanks balanced salt solution and once with PBS followed by fixation with 4% paraformaldehyde for 10 min at room temperature . The cells were washed twice with PBS for 5 min each and permeabilized with 0 . 5% Triton X-100 in PBS for 10 min . The cells were washed with PBS and incubated twice in 10% formamide in 2 × SSC for 5 min before the hybridization is started . Hybridization was carried out for 4 hr at 37°C in 10% formamide and 2 × SSC in the presence of competitor DNA and BSA . The probe concentration was 0 . 1–1 μM . After hybridization , the cells were washed twice in 10% formamide in 2 × SSC for 20 min at 37° and two times 1 hr in PBS at room temperature . The nucleus was stained for 10 s with 1 ml of 0 . 5 μg/ml DAPI followed by 2 × wash in PBS for 5 min . Coverslips were mounted using ProLong Gold antifade reagent ( Invitrogen , Life Technologies , Grand Island , NY ) and imaged 24 hr later . Wide-field fluorescence images were obtained on a BX61 epi-fluorescence microscope ( Olympus , Center Valley , PA ) using a PlanApo 60 × , 1 . 4 NA oil-immersion objective . The microscope is equipped with an X-Cite 120 PC light source ( EXFO , Mississauga , Canada ) for illumination , Uniblitz shutters ( Vincent Associates , Rochester , NY ) , filter sets 31000 ( DAPI ) , 41007a ( Cy3 ) and 41008 ( Cy5 ) ( Chroma Technologies , Brattleboro , VT ) and a MS-2000 Microscope Stage Controller ( Applied Scientific Instrumentation , Eugene , OR ) . Digital images were acquired with a CoolSNAP HQ camera ( Photometrix , Tucson , AZ ) controlled by Metamorph 7 . 6 . 3 software as a 3D stack of 18 images at 0 . 3 μm steps with a 1 × 1 binning . Exposure times were 500 ms for Cy3 and Cy5 and 30 ms for DAPI . The fields for imaging are selected in the DAPI channel to minimize bias . All image analysis was performed using custom-written software in IDL on maximum intensity projections of three-dimensional image stacks . Spot detection using a Gaussian mask ( Thompson et al . , 2002 ) was carried out using the Localize software package described previously ( Trcek et al . , 2012 ) . Nucleus segmentation was achieved by applying an intensity threshold to the maximum intensity projection of the grayscale image . The threshold was adjusted manually to ensure accurate masking . The algorithm detected and removed nuclei that are either too small or not fully contained within the image in order to generate a binary mask . Cells were segmented using intensity thresholding in combination with the watershed operator . We used the following primary antibodies: β-tubulin ( Rockland , Gilbertsville , PA ) and GFP ( Roche , Branchburg , NJ ) , and secondary antibodies: donkey anti-rabbit conjugated to IRDye 800 ( Rockland ) and donkey anti-mouse conjugated to Alexa Fluor 680 ( Invitrogen A21102 ) . We washed cells in ice-cold PBS and lysed them at room temperature for 2 min in 1 ml lysis buffer per 10-cm dish ( 50 mM Tris-HCl ( pH 8 . 0 ) , 50 mM NaCl , 1% NP40 , 5 mM DTT , 1 mM PMSF and half a mini tablet protease inhibitor ) . We spun the lysate 15 min at 14 , 000 × g at 4°C and loaded the supernatant on a Nupage 4–12% bis-tris gel using MOPS running buffer ( Invitrogen ) . After transfer on a nitrocellulose membrane in Nupage transfer buffer ( 25 V for 1 . 5 hr ) , we blocked nonspecific interactions by incubating the blot overnight at 4°C in PBS supplemented with 5% nonfat dry milk and 1% BSA . After that , we rinsed the membrane and incubated it for 1 hr with the primary antibodies in PBS supplemented with 1% BSA ( dilutions: 1:2500 mouse anti–Actb; 1:5 , 000 mouse anti–β-tubulin ) . We then washed the blot five times 10 min in PBS with 0 . 3% Tween-20 before incubation for 30 min with the secondary antibody ( 1:10 , 000 in PBS with 1% BSA ) . We then washed the membrane five times in PBS with 0 . 3% Tween-20 , before exposure on an Odyssey infrared imaging system ( two-color detection ) . Band intensities were quantified using custom software in IDL . Ecdysone and Ponasterone A were purchased from AG Scientific ( San Diego , CA ) . The purity of ponasterone A was determined to be ∼75% by analytical HPLC . All other reagents and solvents were purchased from Sigma-Aldrich ( St . Louis , MO ) . Silica gel 60 ( 40 mm , Fischer ) was employed for column chromatography . DMNB-Ecdysone and 2-bromo-2- ( 2-nitrophenyl ) acetic acid were synthesized by reported method ( Chang et al . , 1995; Lin et al . , 2002 ) . 1D and 2D NMR spectra were recorded on a Bruker DRX-500 . High-resolution mass spectra were obtained at the Mass Spectrometry Facility , the University of North Carolina at Chapel Hill , Chapel Hill , NC . A suspension of Ponasterone A ( 10 mg , 21 μmol ) and dibutyltin oxide ( 6 . 7 mg , 27 μmol ) in anhydrous methanol ( 3 ml ) was heated to reflux for 3 hr under argon . After the solvent was removed under reduced pressure , the residue was subsequently azeotroped with anhydrous benzene ( 3 × 2 ml ) . The resulting stannylene acetal was further dried in vacuo for 2 hr before addition of 3 Å molecular sieves ( 50 mg ) , CsF ( 13 mg , 83 μmol ) , 1-bromomethyl-4 , 5-dimethoxy-2-nitrobenzene ( 11 mg , 38 μmol ) , and anhydrous DMF ( 1 ml ) . After the reaction mixture was stirred at room temperature overnight , the solvent was evaporated under reduced pressure . The resulting residue was purified by silica gel column chromatography ( chloroform/methanol: 20:1 ) to afford caged-product as an off-white solid . DMNB-ponasterone A needs further purification by HPLC due to the impurity in the purchased ponasterone A ( 8 mg , 57% ) ( retention time 46 . 9 min on a Prevail C18 5 μ column 250 mm × 22 mm monitored at 242 nm; a 30 min linear gradient from 95% A [water] to 50% B [acetonitrile] , followed by 50% B for 5 min with the flow rate of 8 ml/min ) . The purity of the product was determined to be >99% by analytical HPLC ( retention time 43 . 4 min on an Apollo C18 5 μ column 250 mm × 4 . 6 mm , monitored at 242 nm; a 15 min linear gradient from 95% A [water] to 50% B [acetonitrile] , followed by 50% B for 5 min and a 15 min linear gradient from 50% A [water] to 95% B [acetonitrile] with the flow rate of 1 ml/min ) . 1HNMR ( 500 MHz , CD3OD ) : δ 0 . 91 ( s , 3H ) , 0 . 94 ( s , 3H ) , 0 . 95 ( s , 3H ) , 1 . 01 ( s , 3H ) , 1 . 20 ( s , 3H ) , 1 . 2 ( m , 2H ) , 1 . 5 −2 . 1 ( m , 15H ) , 2 . 44 ( m , 2H ) , 3 . 08 ( m , 1H ) , 3 . 11 ( m , 1H ) , 3 . 70 ( d , J = 11 . 5 Hz , 1H ) , 3 . 93 ( s , 3H ) , 3 . 99 ( s , 3H ) , 4 . 29 ( s , 1H ) , 4 . 92 ( d , J = 15 . 0 Hz , 1 H ) , 5 . 04 ( d , J = 15 . 0 Hz , 1H ) , 5 . 64 ( s , 1H ) , 7 . 45 ( s , 1H ) , 7 . 72 ( s , 1H ) . 13CNMR ( 125 MHz , CD3OD ) : δ 16 . 6 , 19 . 6 , 20 . 1 , 21 . 4 , 22 . 0 , 22 . 9 , 27 . 8 , 29 . 0 , 30 . 3 , 31 . 1 , 31 . 3 , 33 . 2 , 33 . 7 , 36 . 3 , 37 . 9 , 47 . 2 , 48 . 2 , 49 . 0 , 50 . 6 , 55 . 4 , 55 . 5 , 64 . 2 , 66 . 9 , 76 . 2 , 76 . 4 , 76 . 6 , 83 . 8 , 107 . 7 , 110 . 5 , 120 . 8 , 130 . 1 , 139 . 7 , 147 . 9 , 153 . 6 , 166 . 5 , 204 . 8 . HRMS ( ESI+ ) calculated for C36H53NO10Cs+: 792 . 2724 Found: 792 . 2716 ( −1 . 0 ppm ) . CsF ( 13 mg , 83 μmol ) was added to a solution of the stannylene acetal of ponasterone A/Ecdysone ( 21 μmol ) , 3 Å molecular sieves ( 50 mg ) , 2-bromo-2- ( 2-nitrophenyl ) acetic acid ( 11 mg , 38 μmol ) in anhydrous DMF ( 1 ml ) . After the reaction mixture was stirred at room temperature overnight , the solvent was filtered to remove 3 Å molecular sieves and evaporated under reduced pressure . The resulting residue was dissolved in DMSO and purified by HPLC . White solid ( 7 mg , 52% ) ( retention time 15 . 1 min on a Prevail C18 5 μ column 250 mm × 4 . 6 mm , monitored at 242 nm a 40 min linear gradient from 30% A [water] to 95% B [acetonitrile] , with the flow rate of 8 ml/min ) . 1HNMR ( 500 MHz , CD3OD ) : δ 0 . 91 ( s , 3H ) , 0 . 94 ( s , 3H ) , 0 . 95 ( s , 3H ) , 1 . 01 ( s , 3H ) , 1 . 20 ( s , 3H ) , 1 . 2 ( m , 2H ) , 1 . 5 −2 . 1 ( m , 15H ) , 2 . 44 ( m , 2H ) , 3 . 08 ( m , 1H ) , 3 . 11 ( m , 1H ) , 3 . 60 ( d , J = 11 . 5 Hz , 1H ) , 4 . 15 ( s , 1H ) , 4 . 92 ( m , 1H ) , 5 . 72 ( s , 1 H ) , 7 . 46 ( t , J = 9 . 5 Hz , 1H ) , 7 . 62 ( t , J = 9 . 5 Hz , 1H ) , 7 . 75 ( d , J = 9 . 5 Hz , 1H ) , 7 . 93 ( d , J = 10 . 0 Hz , 1H ) . HRMS ( ESI+ ) calculated for C34H49NO8Na+ ( M–CO2 ) : 622 . 3356 Found: 622 . 3355 ( −1 . 0 ppm ) . White solid ( 9 mg , 65% ) ( retention time 34 . 5 min on a Prevail C18 5 μ column 250 mm × 22 mm monitored at 242 nm; a 30 min linear gradient from 95% A [water] to 50% B [acetonitrile] , followed by 50% B for 5 min and a 15 min linear gradient from 50% A [water] to 95% B [acetonitrile] with the flow rate of 8 ml/min ) . 1HNMR ( 500 MHz , CD3OD ) : δ 0 . 80 ( s , 3H ) , 0 . 89 ( s , 3H ) , 1 . 14 ( s , 3H ) , 1 . 21 ( s , 6H ) , 1 . 22–1 . 98 ( m , 17H ) , 2 . 39 ( m , 2H ) , 3 . 11 ( m , 1H ) , 3 . 60 ( d , J = 11 . 5 Hz , 1H ) , 4 . 15 ( s , 1H ) , 4 . 92 ( m , 1H ) , 5 . 72 ( s , 1 H ) , 7 . 46 ( t , J = 9 . 5 Hz , 1H ) , 7 . 61 ( t , J = 9 . 5 Hz , 1H ) , 7 . 74 ( d , J = 9 . 5 Hz , 1H ) , 7 . 92 ( d , J = 10 . 5 Hz , 1H ) . HRMS ( ESI+ ) calculated for C34H49NO9Cs+ ( M–CO2 ) : 748 . 2461 Found: 748 . 2496 ( +4 . 6 ppm ) . The number of mRNA/cell was binned into histograms and fit with the analytical solution to the random telegraph model ( Raj et al . , 2006 ) using Mathematica ( Wolfram Research , Champaign , IL ) . The probability was summed over the bin size to generate the proper normalization . Unless otherwise indicated , there are three fitting parameters: a/d , b/d , c/d , corresponding to the normalized gene activation rate , gene inactivation rate , and initiation rate , respectively . Live-cell fluorescence images were obtained on a BX61 epi-fluorescence microscope ( Olympus ) modified for simultaneous UV photolysis . The system consists of an X-Cite 120 PC light source ( EXFO , Mississauga , Canada ) for imaging , an Explorer 339 nm pulsed laser for photolysis ( Newport , Santa Clara , CA ) , and ASI shutters and automated stage ( Applied Scientific Instrumentation ) . The laser is capable of operating from single-shot to 1 kHz at 120 μJ/pulse . All data was acquired at 37°C using a Delta T stage incubator and objective heater ( Bioptechs , Butler , PA ) . The UV and visible excitation were focused through a U-Apo 40 × 1 . 35 NA oil lens , and the emission was collected through the same objective . Excitation , emission , and dichroic optics were custom designed for simultaneous UV and visible excitation ( Chroma ) . The microscope was controlled by IPLab ( BioVision , Exton , PA ) , and images were acquired on an Orca R2 CCD ( Hamamatsu , Hamamatsu , Japan ) . Time-lapse images of active transcription were analyzed using previously described methods ( Larson et al . , 2011 ) . First , nuclei are segmented and individual cell videos are separated out from the raw data . The intermediate which results is a time-lapse image stack where the fluorescent nucleus is always centered in the frame ( Video 5 ) . Next , the diffraction-limited spots are fit using a Gaussian mask algorithm ( Thompson et al . , 2002 ) to generate a list of positions . The complete time-dependent trajectories were then constructed using a tracking algorithm ( Crocker and Grier , 1996 ) . We used a Hidden Markov Model fitting algorithm taken directly from Lee et al . to analyze the resulting intensity traces and extract the duration of active and inactive states ( Lee , 2009 ) . | The process by which a gene is expressed as a protein consists of two stages: transcription , which involves the DNA of the gene being copied into messenger RNA ( mRNA ) ; and translation , in which the mRNA is used as a template to assemble amino acids into a protein . Transcription and translation are controlled by many interlinked pathways , which ensures that genes are expressed when and where required . One of these regulatory pathways involves steroid receptors . The binding of a steroid molecule to its receptor causes the receptor to move into the nucleus and interact with a specific gene , triggering transcription of that gene . When measured at the level of the whole organism , this transcriptional response is dose-dependent—the more steroid molecules that are present , the greater the amount of transcription . However , this is not the case in single cells , in which transcription is either activated or not . This ‘on/off’ behaviour is also seen over time: steroid-activated transcription occurs in bursts , separated by periods of inactivity . To unravel the molecular mechanism behind this phenomenon , Larson et al . created a light-activated form of the ligand that activates a specific steroid receptor . Using this molecule , they were able to switch transcription of the gene controlled by that receptor on and off . They then used fluorescent proteins to label the mRNA and protein molecules that were produced as a result . They found that activating the steroid receptor increases the likelihood of transcription occurring inside a cell , but not the duration of individual bursts of transcriptional activity , nor the amount of mRNA produced during each burst . Activation of a steroid receptor seems to control transcription by reducing the length of time each cell spends in the ‘off’ state between bursts . Larson et al . incorporated their findings into a model that also takes into account the natural variability in levels of transcription between cells , and found that this could explain how the digital ( on/off ) control of transcription at the cellular level leads to analogue , dose-dependent control at the level of a whole organism . These findings should lead to further insights into how transcription is controlled at the molecular level . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2013 | Direct observation of frequency modulated transcription in single cells using light activation |
Adult-born neurons are continually produced in the dentate gyrus but it is unclear whether synaptic integration of new neurons affects the pre-existing circuit . Here we investigated how manipulating neurogenesis in adult mice alters excitatory synaptic transmission to mature dentate neurons . Enhancing neurogenesis by conditional deletion of the pro-apoptotic gene Bax in stem cells reduced excitatory postsynaptic currents ( EPSCs ) and spine density in mature neurons , whereas genetic ablation of neurogenesis increased EPSCs in mature neurons . Unexpectedly , we found that Bax deletion in developing and mature dentate neurons increased EPSCs and prevented neurogenesis-induced synaptic suppression . Together these results show that neurogenesis modifies synaptic transmission to mature neurons in a manner consistent with a redistribution of pre-existing synapses to newly integrating neurons and that a non-apoptotic function of the Bax signaling pathway contributes to ongoing synaptic refinement within the dentate circuit .
Continual neurogenesis in the adult dentate gyrus ( DG ) produces new granule cells ( GCs ) that integrate into the hippocampal circuit by establishing synapses with existing neurons ( Espósito et al . , 2005; Ge et al . , 2006; Toni et al . , 2008; Dieni et al . , 2013 ) . During a transient period of maturation , new GCs exhibit intrinsic and synaptic properties distinct from mature GCs , potentially underlying the contribution of neurogenesis to memory encoding ( Schmidt-Hieber et al . , 2004; Ge et al . , 2007; Aimone et al . , 2011; Marín-Burgin et al . , 2012; Dieni et al . , 2013; Brunner et al . , 2014; Dieni et al . , 2016 ) . Yet computational models also suggest that remodeling of pre-existing circuits by continual neurogenesis can degrade established memories ( Weisz and Argibay , 2012; Chambers et al . , 2004 ) , a possibility that has recently gained experimental support from the observation that neurogenesis facilitates ‘forgetting’ ( Akers et al . , 2014; Epp et al . , 2016 ) . Circuit remodeling could occur by synaptic redistribution , wherein existing terminals that synapse onto mature GCs are appropriated by newly integrating GCs . This possibility is supported by anatomical evidence that immature dendritic spines transiently receive a high proportion of synapses from multiple-synapse boutons ( Toni et al . , 2007; Toni and Sultan , 2011 ) . Furthermore , dramatically increasing the number of new neurons does not alter the density of spines and synapses in the molecular layer , suggesting a readjustment of synaptic connections ( Kim et al . , 2009 ) . Yet whether synaptic integration of new GCs is accompanied by changes in synaptic function and structure of mature GCs is not known . The number of integrating new GCs can be selectively altered by genetic manipulations targeted to adult stem cells that regulate the survival of progeny ( Enikolopov et al . , 2015 ) . Adult-born neurons undergo a period of massive cell death during the first weeks after cell birth that is rescued by deletion of the pro-apoptotic protein Bax ( Sun et al . , 2004; Kim et al . , 2009 ) , and conditional Bax deletion in Nestin- expressing progenitors enhances the number of adult-born neurons without affecting other cell populations ( Sahay et al . , 2011; Ikrar et al . , 2013 ) . Similarly , inducible expression of the diphtheria toxin receptor in Nestin-expressing stem cells allows selective ablation of adult-born neurons ( Arruda-Carvalho et al . , 2011 ) . These approaches have been used to identify contributions of adult born neurons in hippocampal-based behaviors , with the understanding that behavioral outcomes could either reflect unique functions of adult-born neurons themselves or homeostatic adaptions within the network ( Singer et al . , 2011 ) . Physiological stimuli like exercise and environmental enrichment also enhance dentate neurogenesis , yet it is unclear whether genetically targeted manipulations of neurogenesis mimic the circuit function in the same manner as physiological stimuli . To identify network adaptions resulting from synaptic integration of new GCs , here we tested how manipulating the number of adult-born GCs affects perforant path-evoked excitatory synaptic currents ( EPSCs ) in mature GCs . We measured synaptic transmission to pre-existing mature GCs in response to selective genetic manipulations of Nestin-expressing stem cells , using inducible Bax deletion to enhance , or diphtheria toxin-induced ablation to reduce , the number of new neurons . We also tested synaptic transmission to immature GCs and mature GCs with Bax deletion to investigate potential non-apoptotic functions of the Bax signaling pathway in synaptic function ( Jiao and Li , 2011; Ertürk et al . , 2014 ) . Finally , we tested whether enhancing neurogenesis by a physiological stimulus likewise alters excitatory transmission to mature neurons . Our results show that selectively manipulating the number of immature GCs modifies synaptic function of mature GCs in a manner consistent with synaptic redistribution , with an inverse relationship between the number of new neurons and perforant-path evoked EPSCs . In contrast , enhancing neurogenesis via the non-selective paradigm of environmental enrichment generates a net increase in functional connectivity of mature neurons . Together these results demonstrate the capacity of mature GCs to alter synaptic function in response to genetic and experiential circuit manipulations .
We sought to test synaptic transmission to mature GCs after selectively enhancing the number of integrating new GCs by manipulating cell survival , given that most proliferating DG progenitors and newborn neurons undergo apoptosis ( Sierra et al . , 2010 ) . Cell death of progenitors and new GCs requires the pro-apoptotic protein Bax , a member of the BCL-2 family of proteins in the intrinsic apoptotic pathway ( Sun et al . , 2004 ) . Both germ line and conditional Bax deletion block cell death of adult-generated GCs without altering proliferation or the gross structural integrity of the DG ( Sun et al . , 2004; Kim et al . , 2009; Sahay et al . , 2011 ) . As previously described ( Sahay et al . , 2011; Ikrar et al . , 2013 ) , we increased the population of adult-born GCs by crossing inducible Nestin-CreERT2 mice with a Bax conditional knockout mouse line to selectively block apoptotic cell death in proliferating cells and their progeny ( Materials and methods; Figure 1—figure supplement 1A ) . Four-to-six weeks after tamoxifen-induced recombination at two months of age , we compared the number of new GCs and synaptic responses from pre-existing mature GCs in hippocampal slices from BaxKOimmature mice ( referred to as BaxKOim ) and controls ( Figure 1A ) . We crossed some BaxKOim mice with a transgenic reporter line that labels early postmitotic GCs ( Overstreet et al . , 2004 ) to reveal a ~40% increase in the number of newborn GCs and overtly normal dentate structure ( Figure 1B , C ) . 10 . 7554/eLife . 19886 . 003Figure 1 . Increasing neurogenesis reduces EPSCs in mature GCs . ( A ) The experimental timeline showing recording 4–6 weeks after tamoxifen ( TMX ) -induced Bax deletion in Nestin-expressing progenitors . ( B ) Confocal images of newborn neurons expressing eGFP in fixed sections ( 50 μm ) from control and BaxKOim mice . ( C ) Stereological cell counts of eGFP+ newborn cells revealed neurogenesis was enhanced by 41% ( control 16 , 881 ± 1422 cells , n = 4; BaxKOim 23 , 756 ± 2166 cells , n = 4; unpaired t-test p=0 . 038 ) . ( D ) Schematic showing experimental paradigm , with simultaneous fEPSPs and whole-cell recordings of EPSCs from mature GCs . All experiments were performed in the presence of picrotoxin to block GABAA receptor-mediated currents . ( E ) Examples of fEPSPs ( top ) with the fiber volley ( FV , top inserts ) and EPSCs ( bottom ) in slices from control and BaxKOim mice . Synaptic responses were evoked by increasing intensity stimulation by a patch pipette placed in the middle molecular layer . fEPSPs and EPSCs were binned by FV amplitude . Stimulus artifacts are blanked for clarity . ( F ) The fEPSP versus FV plot illustrates the effectiveness of FV normalization , with fEPSP increasing linearly with axonal recruitment . There was no difference in fEPSPs in slices from BaxKOim and control mice ( two-way ANOVA , 0 . 076 ) . FVs are binned by 100 μV and each symbol denotes the mean and SEM of 10–38 responses from 15 control and 14 BaxKOim slices ( with four responses in the largest 300–400 μV FV control bin ) . ( G ) Left , a decrease in synaptic strength to mature GCs was revealed by the EPSC plotted against FV amplitude ( two-way ANOVA , Fgenotype ( 1 , 167 ) =54 . 41 p<0 . 0001; p<0 . 05 for all bins with Bonferroni post-tests ) . Right , the overall EPSC/FV ratio was reduced in BaxKOim slices ( unpaired t-test , n = 86 , 95 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 00310 . 7554/eLife . 19886 . 004Figure 1—figure supplement 1 . Generation of BaxKOimmature mice . ( A ) Schematic illustrating tamoxifen ( TMX ) -induced excision of the lox-p flanked Bax locus to generate BaxKOimmature mice ( BaxKOim ) . Control mice included Nestin-Cre-/Baxfl/fl , Nestin-Cre-/Baxfl/+ and Nestin-Cre+/Bax+/+ mice that all received TMX . ( B ) There was no difference in the averaged EPSC/FV ratio between different control genotypes ( n = 4 , 2 , nine experiments; one-way ANOVA , Fgenotype ( 2 , 12 ) =0 . 61 , p=0 . 561 ) . ( C ) The EPSC amplitude in mature GCs from Nestin-Cre-/Baxfl/fl control mice was significantly greater than Nestin-Cre+/Baxfl/fl mice , confirming that differences in EPSCs persist comparing only Baxfl/fl genotypes ( two-way ANOVA , Fgenotype ( 1 , 106 ) =27 . 42 , p<0 . 0001 , n = 11–26 responses from 9 controls and 14 BaxKOim slices ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 00410 . 7554/eLife . 19886 . 005Figure 1—figure supplement 2 . No change in fEPSPs or mature GC intrinsic excitability in BaxKOim mice . ( A ) There was no difference in FVs or fEPSPs between BaxKOim and control mice ( 2-way ANOVA , p=0 . 478 or 0 . 998 ) . Note that individual responses that exhibit saturation drop out of the analysis , such that there are fewer values at high stimulus intensities ( see Materials and methods ) . Each symbol represents 3–14 responses . ( B ) An example of a mature GC targeted for whole cell recordings that was reconstructed after recording . The intrinsic properties of GCs were tested using step current injections in current clamp prior to voltage clamp experiments . ( C ) Mature unlabeled GCs in BaxKOim mice had similar intrinsic properties as in control mice , including input resistance , action potential amplitude and action potential frequency , measured at 100 pA current injections ( unpaired t-tests , p=0 . 872 , 0 . 893 and 0 . 572 respectively , n = 15 controls and 14 BaxKOim ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 005 To assess excitatory transmission from entorhinal cortex across the population of GCs and onto individual mature GCs , we stimulated the medial perforant path while simultaneously recording field excitatory postsynaptic potentials ( fEPSPs ) and excitatory postsynaptic currents ( EPSCs ) from mature GCs ( Figure 1D , E ) . All experiments were performed in the GABAA receptor antagonist picrotoxin to isolate glutamatergic synaptic responses . There was no difference in fiber volleys ( FVs; a measure of axonal activation ) or fEPSPs between slices from BaxKOim and control mice ( Figure 1—figure supplement 2A ) ( Sahay et al . , 2011 ) , as well as no difference in fEPSPs when responses were binned by the FV to account for differences in the number of stimulated axons across slices ( Figure 1F ) . We targeted mature GCs located near the mid or outer edge of the granule cell layer and confirmed their maturity by morphology and intrinsic membrane properties ( Figure 1—figure supplement 2B , C ) . Interestingly , we found that mature GCs in BaxKOim mice exhibited smaller EPSCs than mature GCs in controls across all FV amplitudes ( Figure 1G , left ) , and an overall lower EPSC/FV ratio ( Figure 1G , right ) . There was no difference in the EPSC/FV ratio between mature GCs in Cre+ and Cre- controls , and the difference in EPSCs persisted when only Baxfl/fl genotypes were analyzed ( Figure 1—figure supplement 1B , C ) . Thus mature GCs in BaxKOim slices had reduced excitatory transmission . To assess the pre- or postsynaptic locus of reduced EPSCs in mature GCs from BaxKOim mice , we first compared the paired-pulse ratio ( PPR ) , a measure of presynaptic release probability . There was no difference in the PPR of evoked EPSCs at an interstimulus interval of 100 ms ( Figure 2A ) , implying that adult-born neurons do not regulate transmission to mature GCs by secreting a factor that alters the release probability . However , mature GCs in BaxKOim mice displayed a lower frequency of spontaneous EPSCs ( sEPSCs ) with no change in amplitude ( Figure 2B ) , suggesting a reduction in the number of active synapses with no change in postsynaptic responsiveness . Furthermore , using Sr2+ to desynchronize evoked release in order to detect single site EPSCs ( Bekkers and Clements , 1999; Rudolph et al . , 2011; Williams et al . , 2015 ) , we found a reduction in the frequency but not the amplitude of desynchronized events ( Figure 2C ) . Thus , enhanced numbers of newly generated neurons were associated with reduced excitatory synaptic transmission to mature GCs that appeared to be mediated by fewer functional synapses . 10 . 7554/eLife . 19886 . 006Figure 2 . Fewer functional synapses on mature GCs in BaxKOim mice . ( A ) The paired-pulse ratio of evoked EPSCs ( 100 ms ISI ) was similar in BaxKOim and control mature GCs ( unpaired t-test p=0 . 90; n = 15 controls , 14 BaxKOim ) . ( B ) Spontaneous EPSCs in mature GCs from BaxKOim mice had lower frequency and similar amplitudes as sEPSCs in mature GCs from control mice ( unpaired t-test p=0 . 027 and 0 . 79 , respectively; n = 11 controls , 9 BaxKOim ) . ( C ) Asynchronous EPSCs were generated by desynchronizing synaptic release with 1 mM Ca2+ and 4 mM Sr2+ . Uniquantal aEPSCs were detected following the synchronous EPSC . Left , 40 traces overlaid with examples of averaged aEPSCs . Middle , there was no difference in the average amplitude across genotypes but a reduction in the frequency of aEPSCs ( 1743 events in 8 GCs from controls , 1015 events in 9 GCs from BaxKOim; unpaired t-test p=0 . 51 and 0 . 007 respectively ) . There was no difference in the average rise time or decay of aEPSCs ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 006 To further examine the locus of change , we assessed the PPR of EPSCs in mature GCs across a range of interstimulus intervals ( 20–1000 ms ) . In this protocol , mature GCs in BaxKOim and control mice again exhibited similar passive and active properties ( Figure 3—figure supplement 1 ) . The PPR was mildly depressing ( Petersen et al . , 2013 ) , with no difference in ratios between genotypes ( Figure 3A ) , as previously reported using fEPSPs ( Sahay et al . , 2011 ) . During the recordings , we filled GCs with biocytin for posthoc spine analysis , focusing on dendrite segments in the middle molecular layer where medial perforant path synapses are located ( Figure 3B ) . Consistent with reduced evoked and sEPSCs , there was a robust reduction in the density of spines in mature GCs from BaxKOim mice compared to controls ( Figure 3C ) . We classified spines by shape ( mushroom , thin , stubby ) to determine the percentage of each spine type in control and BaxKOim mice . There was a slight increase in the percentage of stubby spines in BaxKOim mice ( Figure 3D ) , with no significant difference in the percentage of thin and mushroom spines . Together , these results support the functional data showing that increasing the number of newborn GCs decreases synaptic transmission to mature GCs by reducing the number of synapses . 10 . 7554/eLife . 19886 . 007Figure 3 . Mature GCs in BaxKOim mice exhibit low spine density . ( A ) There was no difference in the paired-pulse ratio of EPSCs in mature GCs from BaxKOim and control mice across a range of interstimulus intervals ( 2-way ANOVA , p=0 . 31 , n = 8 , 12 mature GCs ) . ( B ) Examples of reconstructed mature GCs from the recordings in ( A ) . Red boxes indicate regions used for spine analysis . ( C ) Left , example images of dendritic spines from mature GCs . Scale bar , 10 μm . Middle , the density of dendritic spines was lower in BaxKOim mice ( 14 ± 0 . 8 spines/10 μm , 936 total spines counted on 15 dendritic segments in two control mice; 10 ± 0 . 6 spines/10 μm , 676 total spines on 12 dendritic segments from 3 BaxKOim mice; p=0 . 0007 unpaired t-test ) . ( D ) Classifying spines as stubby , thin and mushroom revealed a significant increase in the percentage of stubby spines in mature GCs from BaxKOim mice ( p=0 . 04 unpaired t-test ) with no change in the percentage of thin spines ( p=0 . 07 unpaired t-test ) or mushroom spines ( p=0 . 45 unpaired t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 00710 . 7554/eLife . 19886 . 008Figure 3—figure supplement 1 . Intrinsic properties of mature GCs for PPR and spine analysis . The intrinsic properties of unlabeled GCs in the PPR experiments and spine analysis confirmed their maturity . There were no differences in input resistance , AP amplitude and AP frequency , measured at 200 pA current injection ( unpaired t-tests , n = 8 control and 12 BaxKOim ) . The patch pipette intracellular solution for these experiments was slightly modified from that reported in the Materials and methods , containing 8 rather than 28 mM Cl- and 0 . 2% biocytin . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 008 We next tested whether genetically ablating adult-generated neurons alters excitatory transmission to mature GCs . We crossed Nestin-CreERtm4 mice ( Kuo et al . , 2006a ) to Cre-inducible diphtheria toxin receptor ( iDTR ) mice ( Buch et al . , 2005; Arruda-Carvalho et al . , 2011 ) . Six weeks after tamoxifen-induced recombination , DT injections were given to ablate immature adult-born GCs in Nestin-Cre+/iDTR+ offspring ( termed Ablatedim mice; Figure 4A ) with Cre- littermates used as controls . Ten days after injections , there was a 27% reduction in the number of Dcx-expressing immature cells in the dentate of Ablatedim mice ( Figure 4B; 5601 ± 262 , n = 2 , compared to 7648 ± 332 , n = 4 , p=0 . 016 ) , noting that re-population of Dcx-expressing cells in the period after DT injection can lead to an underestimation of ablation efficiency ( Vukovic et al . , 2013; Yun et al . , 2016 ) . Performing simultaneous field and whole-cell recordings from mature GCs in Ablatedim mice and controls at 1–2 weeks after DT injections suggested no change in total synapses , assayed by the FV and fEPSP slopes ( Figure 4C , D , Figure 4—figure supplement 1A ) . We also assayed synaptic terminals by immunodetection of the vesicular glutamate transporter ( vGlut1 ) in the molecular layer , and found no differences between controls and either Ablatedim or BaxKOim slices ( Figure 4—figure supplement 1B ) . Furthermore , there was no change in the fEPSP normalized to the FV ( Figure 4D ) . However , there was enhanced synaptic transmission to individual mature GCs , shown by larger EPSC amplitudes across FVs ( Figure 4E ) and an overall larger EPSC/FV ratio ( 2 . 2 ± 0 . 1 in control compared to 3 . 7 ± 0 . 4 in Ablatedim mice; n = 42 , 47 respectively , p=0 . 001 unpaired t-test ) . The change in synaptic strength was not associated with any changes in the intrinsic properties of mature GCs ( Figure 4—figure supplement 2A ) . These results suggest that reducing the number of immature GCs increases the strength of synaptic transmission to mature GCs , an effect that cannot be explained by altered inhibition as GABAA receptors were blocked in these experiments ( Singer et al . , 2011; Temprana et al . , 2015; Drew et al . , 2016 ) . There was no difference in PPR , suggesting that release probability was unchanged ( Figure 4—figure supplement 2B ) . We were unable to detect differences in the average frequency or amplitude of sEPSCs in mature GCs from Ablatedim mice ( Figure 4—figure supplement 2C ) , making it unclear whether reduced EPSCs resulted from pre- or postsynaptic mechanisms . Since the frequency of spontaneous activity in GCs is low , the threshold for detecting differences in synaptic function using spontaneous activity may be higher than for evoked transmission with FV normalization , and it appears that neurogenesis was altered by a greater degree in BaxKOim mice compared to Ablatedim mice ( ~40% versus 25% change in new neuron number ) . However , we also cannot rule out the possibility that separate pools of synaptic vesicles contribute to differences between results obtained with evoked and spontaneous assays ( reviewed in Kavalali , 2015 ) . 10 . 7554/eLife . 19886 . 009Figure 4 . Ablating neurogenesis increases synaptic transmission to mature GCs . ( A ) Experimental timeline showing ablation of immature GCs that are <6 weeks of age . Recordings from mature GCs were done 1–2 weeks after ablation . ( B ) Confocal images of Dcx-expressing immature neurons in control and Ablatedim mice . ( C ) Example of fEPSPs ( top ) with fiber volleys ( FV , top insets ) and simultaneously recorded EPSCs from mature GCs ( bottom ) in control and Ablatedim mice . ( D ) There was no difference in the fEPSP slope versus FV between Ablatedim and control mice ( two-way ANOVA p=0 . 879 , each symbol represents 8–22 responses from 7 control and 7 Ablatedim mice; FVs were binned by 75 μV ) . ( E ) The EPSC amplitude plotted against FV was larger in mature GCs from Ablatedim mice compared to controls ( two-way ANOVA , Fgenotype ( 1 , 91 ) =30 . 31 p<0 . 0001; ***p<0 . 001 Bonferonni post-test ) . There was an increase in the overall EPSC/FV ratio in mature GCs from Ablatedim mice ( unpaired t-test , p=0 . 0008 , n = 42 , 47 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 00910 . 7554/eLife . 19886 . 010Figure 4—figure supplement 1 . No change in FV , fEPSP slope or vGlut1 expression . ( A ) There was no difference in the FV ( left ) or fEPSPs ( right ) in slices from control and . Ablatedim mice ( two-way ANOVA p=0 . 118 and 0 . 893 , n = 9 control slices and 7 Ablatedim slices ) . ( B ) There were no differences in vGlut1 expression in the molecular layer between respective controls and BaxKOim mice ( left ) or Ablatedim mice ( right ) . Top row , 20X images , yellow box shows quantification region , GCL = granule cell layer , MML = middle molecular layer , scale bar = 50 μm . Bottom row , higher magnification image used to measure fluorescence intensity , scale bar = 10 μm . ( C ) Quantification of corrected total fluorescence intensity ( CTFI ) showing no difference in the amount of Vglut1 signal between BaxKOim ( 48 , 807 ± 2 , 554 , n = 11 images ) and control ( 44 , 964 ± 1 , 283 , n = 20; unpaired t-test p=0 . 239 ) or between Ablatedim ( 48 , 247 ± 2 , 171 , n = 13 images ) and controls ( 53 , 787 ± 1 , 856 , n = 11; unpaired t-test p=0 . 093 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 01010 . 7554/eLife . 19886 . 011Figure 4—figure supplement 2 . Unlabeled GCs in Ablatedim mice have mature intrinsic properties and no change in PPR or sEPSCs . ( A ) Intrinsic properties of mature GCs were assayed by current injections ( as in Figure 1—figure supplement 2 ) . Intrinsic properties of mature GCs were similar in Ablatedim and control mice , including input resistance ( unpaired t-test p=0 . 12 ) , action potential ( AP ) amplitude measured from threshold ( p=0 . 10 ) and AP frequency , measured in response to 50 pA current injections ( p=0 . 74 , n = 9 control , 7 Ablatedim ) . ( B ) Control and Ablatedim mice had similar PPR when stimulating at a 100 ms interval ( unpaired t-test p=0 . 317 , n = 9 control and 7 Ablatedim ) . ( C ) Example traces from spontaneous EPSC recordings ( left ) . Both the frequency ( middle ) and amplitude ( right ) of spontaneous events was the same in control and Ablatedim mice ( unpaired t-tests p=0 . 807 and 0 . 312; n = 9 control and 7 Ablatedim ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 011 In summary , manipulating the number of immature GCs was inversely associated with excitatory synaptic strength of mature GCs . These manipulations did not affect global measures of axonal activation , synaptic strength or presynaptic terminals , suggesting that changing the number of newly generated neurons did not alter the total number of afferent axons or synapses . The idea that global measures of basal synaptic transmission and release probability are independent of the number of dentate GCs is in agreement with prior results in the conditional BaxKO ( Sahay et al . , 2011 ) as well as the observation that perforant path synapse density is unaltered in germline BaxKO mice which exhibit dramatically enhanced numbers of dentate GCs ( Kim et al . , 2009 ) . Together these results support the idea that synaptic integration of newborn GCs involves a redistribution of existing synapses from old to new cells ( Tashiro et al . , 2006; Toni et al . , 2007; McAvoy et al . , 2016 ) . One assumption inherent to this idea , however , is that synaptic integration of newborn neurons is unaffected by manipulating their number , such that the increase in new cell number is paralleled by an increase in the total number of new synapses . We therefore sought to confirm synaptic integration of BaxKO immature GCs by crossing BaxKOim and control mice with a tdTomato reporter line ( Ai14 ) to target BaxKO and BaxWT immature GCs for recordings ( Figure 5A ) . The input resistance is a measure of cell maturity ( Overstreet-Wadiche and Westbrook , 2006; Dieni et al . , 2013 ) and as expected , labeled immature GCs ( six weeks post-tamoxifen ) had higher input resistance than mature GCs , with no difference between genotypes ( Figure 5B ) . This confirms that the immature GCs were at a similar stage of maturation and is consistent with the similar dendrite development reported in this model ( Sahay et al . , 2011 ) . FVs and fEPSP slopes were the same between genotypes , replicating the results of Figure 1 and further suggesting a similar level of axonal activation and number of total synapses after conditional Bax deletion ( Figure 5—figure supplement 1 ) . Consistent with the low excitatory connectivity of immature GCs ( Dieni et al . , 2016 ) , in control mice the EPSC/FV ratio of immature GCs ( 1 . 24 ± 0 . 07 n = 80 ) was lower than the EPSC/FV ratio in mature GCs ( 2 . 44 ± 0 . 16 n = 86 , p<0 . 0001 unpaired t-test ) . But unexpectedly , simultaneously recorded fEPSPs and EPSCs revealed that EPSCs in BaxKO immature GCs were significantly larger than EPSCs in BaxWT immature GCs across FV bins , and the overall EPSC/FV ratio was greater ( Figure 5C ) . Thus BaxKO immature GCs showed enhanced synaptic transmission compared to WT immature GCs . The PPR of EPSCs in immature GCs was similar between genotypes and there was not a significant difference in the frequency or amplitude of sEPSCs ( Figure 5—figure supplement 2 ) , again noting that the low frequency of spontaneous activity in immature GCs ( Mongiat et al . , 2009; Dieni et al . , 2016 ) makes it difficult to interpret the lack of change in sEPSCs . These results confirm that BaxKO immature GCs acquired synapses during integration and , in fact , suggest Bax deletion promotes the synaptic integration of new GCs . 10 . 7554/eLife . 19886 . 012Figure 5 . Bax deletion enhances EPSCs in adult born neurons . ( A ) Whole cell recordings were made from immature GCs in control and BaxKOim slices at six weeks post-tamoxifen injection , using picrotoxin to isolate glutamatergic EPSCs . Simultaneous fEPSPs were recorded in the molecular layer as in Figure 1 . ( B ) Immature GCs in control and BaxKOim tdT mice had a similar input resistance that was higher than mature GCs ( n = 12 , 12 , 16 , respectively; one-way ANOVA p=0 . 0004 , *p<0 . 05 , ***p<0 . 0001 Bonferroni post hoc test ) . ( C ) Left , examples of fEPSPs ( top ) and EPSCs ( bottom ) recorded in immature GCs . Middle , an increase in synaptic transmission to immature BaxKO GCs was revealed by the EPSC plotted against fiber volley ( two-way ANOVA , Fgenotype ( 1 , 143 ) =18 . 55 p<0 . 0001 , n = 12 control tdT , 12 BaxKOim tdT; *p<0 . 05 with Bonferroni post-tests ) . Right , the EPSC/FV ratio for all stimulus intensities ( control 1 . 24 ± 0 . 07 , n = 80; BaxKOim1 . 59 ± 0 . 09 , n = 75; unpaired t-test p=0 . 0029 ) . ( D ) Schematic showing simultaneous recordings from adjacent tdT- ( BaxWT ) and tdT+ . ( BaxKO ) GCs in slices from BaxKOim tdT mice at 16 weeks after tamoxifen . ( E ) Adult-generated BaxKO GCs had larger EPSCs than simultaneously recorded unlabeled mature GCs . EPSCs were normalized to the maximum amplitude of the unlabeled ( BaxWT ) GC in each slice ( two-way ANOVA , Fgenotype ( 1 , 94 ) =11 . 59 p=0 . 001 , n = 6 pairs ) , scale bars: 10 ms , 100 pA . Comparing raw EPSCs between pairs of unlabeled and tdT+ GCs across all stimulus intensities confirmed EPSCs were larger in tdT+ GCs ( not shown , paired t-test , p<0 . 0013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 01210 . 7554/eLife . 19886 . 013Figure 5—figure supplement 1 . No change in FV or fEPSP in slices from BaxKOim tdT mice . There was no difference in the FVs ( left ) or fEPSPs ( middle ) in BaxKOim mice with tdT-labeled immature neurons ( two-way ANOVA p=0 . 535 and 0 . 345 , stim intensity binned by 10 V ) . The fEPSP slope plotted against the FV also suggested no change in total synapse number ( right , two-way ANOVA p=0 . 210 , FVs binned by 100 μV ) . Data from 12 control slices and 12 BaxKOim slices . These results replicate those from Figure 1 , showing that addition of tdT expression in immature GCs has no effect on the measures obtained in BaxKOim mice . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 01310 . 7554/eLife . 19886 . 014Figure 5—figure supplement 2 . No change in PPR or sEPSCs in immature GCs from BaxKOim tdT mice . ( A ) A representative immature GC filled with biocytin during recording . Note the small dendritic tree compared to mature GCs ( shown in Figure 1—figure supplement 2; Dieni et al . , 2013 , 2016 ) . ( B ) Immature GCs in control and BaxKOim tdT cells had similar paired-pulse ratio ( 100 ms ISI ) . Unpaired t-test p=0 . 345; n = 12 control tdT and 11 BaxKOim tdT . ( C ) Example traces from spontaneous EPSC recordings ( left ) . The frequency ( middle ) and amplitude ( right ) of spontaneous events were similar in control and BaxKOim tdT cells ( unpaired t-tests , p=0 . 203 and 0 . 525 respectively , n = 9 control tdT and 10 BaxKOim tdT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 01410 . 7554/eLife . 19886 . 015Figure 5—figure supplement 3 . No differences in intrinsic properties between adult-born mature tdT+ ( Bax-/- ) and unlabeled mature GCs . There were no differences in the input resistance ( unpaired t-test , p=0 . 51 , control n = 21 , BaxKOimn = 14 ) , AP amplitude or AP frequency ( unpaired t-tests , p=0 . 21 and 0 . 69 respectively , control n = 13 , BaxKOimn = 11 ) between unlabeled mature GCs and tdT+ adult-born GCs in BaxKOim tdT mice . Note that this data is from16 weeks after TMX induced recombination , such that adult-born tdT+ GCs have mature intrinsic properties . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 01510 . 7554/eLife . 19886 . 016Figure 5—figure supplement 4 . Global Bax levels are unaltered in BaxKOim hippocampus . Representative hippocampal western blot from control ( Con ) , BaxKOim ( KOim ) , and germ line Bax-/- hippocampal lysates for Bax protein and control β-tubulin . Band intensity was quantified and normalized to control protein . Unpaired t-test , p=0 . 28 ( n = 3 mice , ± SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 016 To further test the role of Bax in excitatory transmission to postmitotic GCs , we compared synaptic activity of adult-born BaxKO and unlabeled GCs at 16 weeks after tamoxifen-induced recombination , well after excitatory synaptic integration is complete ( Mongiat et al . , 2009 ) . We directly compared EPSCs using simultaneous recordings from neighboring BaxWT ( tdT- ) and BaxKO ( tdT+ GCs; Figure 5D ) . In this paradigm , FV normalization is unnecessary because the number of stimulated axons is the same for both recorded cells . To compare across cell pairs with different numbers of stimulated fibers in each slice , we normalized EPSCs to each BaxWT GC . Consistent with a role of Bax suppressing synaptic depression , EPSCs in BaxKO GCs were larger than EPSCs in BaxWT GCs ( Figure 5E ) . There was no difference in the mature intrinsic properties of BaxWT and BaxKO GCs , again showing that Bax deletion does not alter intrinsic cell properties ( Figure 5—figure supplement 3 ) . Thus , enhanced synaptic transmission in Bax deficient GCs persists when adult-born neurons are fully mature . Our results show that Bax deletion increases excitatory synaptic integration of adult born GCs , consistent with growing evidence that the Bax/caspase signaling cascade has non-apoptotic functions in synaptic plasticity ( Unsain and Barker , 2015 ) . Prior work suggests that Bax activation is an intermediary step between NMDAR-Ca2+ influx and local activation of caspase-3 , which in turn is necessary and sufficient for LTD and subsequent spine pruning ( Li et al . , 2010; Jiao and Li , 2011; Ertürk et al . , 2014; Sheng and Ertürk , 2014 ) . The high level of Bax mRNA throughout the adult dentate gyrus ( Lein et al . , 2007 ) raises the possibility that this pathway contributes to activity-dependent synaptic remodeling of mature GCs in addition to controlling the number of integrating new GCs via apoptosis . Given that synaptic strength may depend on Bax expression , we tested whether overall Bax levels are altered in BaxKOim mice . Western blot analysis revealed no difference in Bax protein levels in hippocampal lysates from BaxKOim and control mice , showing that deletion of Bax from a small percentage of GCs does not lead to widespread changes in Bax protein ( Figure 5—figure supplement 4 ) . To further probe the synaptic function of Bax , we next tested whether enhanced synaptic strength persists in mature neurons when Bax is deleted from postmitotic GCs throughout development . We generated conditional BaxKO in postmitotic GCs ( termed BaxKOmature ) using POMC-Cre to direct recombination in dentate GCs throughout development ( Gao et al . , 2007; Figure 6—figure supplement 1 ) . Expression of tdTomato ( tdT ) reporter revealed that most , but not all , NeuN-expressing GCs in the granule cell layer expressed Cre and that NeuN-lacking proliferating progenitors in the subgranular zone were Cre negative ( Figure 6A ) , consistent with transient activity of the POMC promoter in early postmitotic GCs ( Overstreet-Wadiche et al . , 2006; Overstreet et al . , 2004 ) . We compared EPSCs in simultaneous recordings from neighboring tdT+ ( BaxKO ) and tdT- ( BaxWT ) mature GCs ( Figure 6B ) , again normalizing EPSCs to each WT cell to compare EPSCs across cell pairs . EPSCs in BaxKO GCs were larger than EPSCs in BaxWT GCs across a range of stimulus intensities ( Figure 6C ) . To confirm that the increase in EPSC amplitude resulted from Bax deletion , we repeated the experiment in POMC-Cre/BaxWT/tdT mice ( Figure 6D , E ) . EPSCs were the same in neighboring tdT+ and tdT- mature GCs ( Figure 6F ) , indicating that the difference shown in Figure 6C requires the Baxfl/fl genotype . Thus , Bax deletion from immature GCs decreases EPSCs in mature GCs via a non-cell autonomous mechanism ( Figures 1–3 ) , whereas here we show a cell-autonomous effect of Bax deletion that increases EPSCs in mature GCs ( Figure 6 ) . These counterintuitive results could occur if Bax deletion generates presynaptic actions that are most evident when Bax is deleted from a large population of GCs . We addressed potential presynaptic alterations in BaxKOmat mice by testing the Ca2+-dependence of synaptic transmission . However , we found no difference in presynaptic function as assessed by comparing EPSC amplitudes and PPRs across a range of extracellular Ca2+ concentrations ( Figure 6—figure supplement 2 ) . 10 . 7554/eLife . 19886 . 017Figure 6 . Bax deletion increases EPSCs and spine density of mature GCs . ( A ) Confocal image of fixed tissue from a BaxKOmat /tdTomato mouse showing tdT ( red ) and NeuN ( blue ) . Note the larger fraction of tdT+ GCs compared to ( D ) , consistent with enhanced survival of GCs that lack Bax ( confirmed in Figure 7A ) . ( B ) Adjacent tdT- ( BaxWT ) and tdT+ ( BaxKO ) mature GCs were recorded simultaneously . ( C ) Examples of EPSCs in tdT- and tdT+ mature GCs to the same stimuli . EPSCs were normalized to the maximum EPSC of the unlabeled cell in each slice . EPSCs were larger in tdT+ GCs ( two-way ANOVA , Fgenotype ( 1 , 198 ) =21 . 14 p<0 . 0001 , n = 12 cell pairs ) . ( D ) Confocal image of fixed tissue from a BaxWT/POMC-Cre+/tdTomato mouse , in which both tdT+ and unlabeled GCs are BaxWT ( red tdT , blue NeuN ) . ( E ) Adjacent tdT+ and unlabeled mature GCs were recorded simultaneously . ( F ) There was no difference in EPSCs between BaxWT tdT+ and unlabeled cells ( two-way ANOVA p=1 . 0 , n = 8 cell pairs ) , confirming the difference in panel C requires the Bax-/- genotype . ( G ) Posthoc dendrite reconstructions ( top ) revealed higher spine density in BaxKO GCs from ( A ) ( 10 . 5 ± 0 . 53 spines/10 μm in BaxWT compared with 17 . 60 ± 1 . 3 BaxKO , unpaired t-test p<0 . 0001 ) with no change in spine head diameter ( unpaired t-test , p=0 . 7 , n = 21 segments from 5 BaxWT , 18 segments from 9 BaxKO ) . Lower images illustrate spine analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 01710 . 7554/eLife . 19886 . 018Figure 6—figure supplement 1 . Generation of BaxKOmature mice . Schematic illustrating the conditional excision of the lox-p flanked Bax locus using POMC-Cre/Baxfl/+ mice to generate BaxKOmature mice ( BaxKOmat ) . Controls were POMC-Cre-/Baxfl/fl . Additional breeding to reporter mice was used in the experiments shown in the indicated figures: POMC-eGFP to label newborn GCs or Ai14 ( tdTomato ) to identify recombined BaxKO cells . In the latter case , controls expressing tdTomato were POMC-Cre+/Bax+/+ . . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 01810 . 7554/eLife . 19886 . 019Figure 6—figure supplement 2 . No difference in Ca2+-dependence of evoked EPSCs in BaxKOmat mice . ( A ) Evoked EPSCs in mature GCs in control and BaxKOmat mice were tested across a range of extracellular Ca2+ concentrations , with the average EPSC from 50 trials in each [Ca2+] normalized to the amplitude of EPSCs in 2 mM Ca2+ . As expected , the EPSC amplitude was highly sensitive to extracellular Ca2+ , and there was no difference between EPSCs across genotype ( p=0 . 44 , n = 4–11 cells per [Ca2+] ) . ( B ) The PPR ( 100 ms ISI ) was inversely related to [Ca2+] , but there was no difference between genotypes ( p=0 . 08 , n = 4–17 cells per [Ca2+] ) . Normalizing the PPR in each cell to the PPR in 2 mM Ca2+ likewise revealed no difference between genotypes ( p=0 . 11 ) , suggesting that the Ca2+ dependence of release is similar in control and BaxKOim mice . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 019 Since Bax activation is necessary and sufficient to activate caspase-3 , which acts as a mediator of activity-dependent hippocampal LTD and synaptic pruning ( Li et al . , 2010; Jiao and Li , 2011; Ertürk et al . , 2014; Lo et al . , 2015 ) , we wondered whether enhanced synaptic transmission to BaxKO GCs resulted from a deficit in synaptic pruning . We analyzed dendritic spines in BaxKO and BaxWT GCs by filling cells with biocytin during recordings . Posthoc analysis revealed a significant increase in the density of spines in BaxKO mature GCs , with no change in head diameter ( Figure 6G ) . Together these results show that loss of Bax in GCs generates a persistent enhancement of synaptic transmission consistent with a deficit in synaptic pruning . Based on the above results , we predicted that neurogenesis-induced loss of synapses from mature GCs might require intact Bax signaling to allow synaptic pruning . We thus assayed neurogenesis-induced synapse loss from mature GCs in BaxKOmat mice , where most mature GCs lack Bax . First , we confirmed that BaxKO in newly postmitotic GCs increases the number of integrating new neurons by assessing neurogenesis using POMC-eGFP expression . Consistent with the later period of cell death that occurs in newly postmitotic GCs ( Sierra et al . , 2010 ) , we found that the number of newborn integrating neurons was enhanced to a similar degree as observed in BaxKOim mice ( Figure 7A ) . However , neurogenesis-induced suppression of synaptic transmission to mature GCs was absent , since the evoked EPSC was similar to controls across all stimulus intensities and the average EPSC/FV ratio was unchanged ( Figure 7B ) . Similar to Ablatedim and BaxKOim mice , there was no difference in axonal activation or total synapse number , measured by the FV amplitude and fEPSP slope versus FV , respectively ( Figure 7—figure supplement 1A ) . Intrinsic properties of mature GCs were the same in BaxKOmat and control mice , showing that Bax deletion does not affect these measures of cellular excitability ( Figure 7—figure supplement 1B ) . There was also no difference in the PPR , sEPSC frequency or sEPSC amplitude between mature GCs in control and BaxKOmat mice ( Figure 7—figure supplement 2 ) . However , there was considerable variability in EPSC/FV ratios and sEPSC frequencies in mature GCs from BaxKOmat mice , potentially indicative of the heterogeneous population of BaxWT and BaxKO GCs ( as in Figure 6A ) with mixed susceptibility to neurogenesis-induced synapse impairment . Together , these results suggest that neurogenesis-induced loss of synaptic strength to mature GCs requires intact Bax signaling . 10 . 7554/eLife . 19886 . 020Figure 7 . Neurogenesis-induced loss of synaptic transmission requires Bax in mature GCs . ( A ) Confocal images of newborn neurons expressing eGFP in BaxKOmat mice . Stereological cell counts revealed neurogenesis was enhanced by ~48% ( control 17 , 910 ± 900 cells , n = 7; BaxKOmat26 , 508 ± 2728 cells , n = 6; unpaired t-test ) , similar to enhanced neurogenesis in BaxKOim mice . ( B ) Left , examples of fEPSPs ( top ) and EPSCs in mature GCs ( bottom ) from control and BaxKOmat mice . Middle , there was no difference in EPSCs across FVs ( two-way ANOVA p=0 . 990 , n = 12 control , 19 BaxKOmat GCs ) or in the EPSC/FV ratio ( unpaired t-test p=0 . 387 , n = 88 control , 129 BaxKOmat ) , although there was greater variability in the BaxKOmat group ( CV = 52% vs . 43% ) consistent with a mixed population of mature Bax-/- and Bax+/+ GCs ( as shown in Figure 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 02010 . 7554/eLife . 19886 . 021Figure 7—figure supplement 1 . No differences in FVs , fEPSP slopes and intrinsic properties of mature GCs in BaxKOmat mice . ( A ) There was no difference between slices in BaxKOmat and control mice in FV amplitudes , fEPSP slopes , or fEPSP slope plotted against FVs ( two-way ANOVA , p=0 . 13 , 0 . 39 and 0 . 21 respectively ) . FVs were binned by 100μV , control n = 13 slices , BaxKOmatn = 18 slices ) . ( B ) There were no differences in the input resistance ( unpaired t-test p=0 . 32 , control n = 21 , BaxKOmatn = 14 ) , AP amplitude or AP frequency ( unpaired t-tests . p=0 . 84 and 0 . 051 respectively , control n = 13 , BaxKOmatn = 11 ) in mature GCs from BaxKOmat mice . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 02110 . 7554/eLife . 19886 . 022Figure 7—figure supplement 2 . Similar PPR and sEPSCs in mature GCs from BaxKOmat mice . ( A ) The PPR of EPSCs in mature GCs was similar in BaxKOmat and controls ( unpaired t-test , p=0 . 115 , n = 12 controls and 19 BaxKOmat ) . ( B ) The frequency of spontaneous EPSCs in mature GCs from BaxKOmat mice was similar to controls ( unpaired t-test p=0 . 715 ) but with higher variance ( p=0 . 04 , not shown ) . The sEPSC amplitude was similar ( p=0 . 142 , n = 9 control , 13 BaxKOmat ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 022 Our experiments revealed that selective manipulations of adult-born neurons are sufficient to alter functional synaptic transmission to mature neurons , raising the question of whether enhancing neurogenesis by physiological stimuli likewise affects synaptic function of mature neurons . One long-established strategy to enhance neurogenesis is housing rodents with environmental enrichment ( EE ) that includes exploration of novel objects , social interactions , and running wheels . EE enhances both the number of newborn GCs and their synaptic integration ( van Praag et al . , 1999; Tashiro et al . , 2007; Ambrogini et al . , 2010; Chancey et al . , 2013; Bergami et al . , 2015 ) , as well as altering structural plasticity in the dentate and other brain regions ( Green and Greenough , 1986; Foster et al . , 1996; Eadie et al . , 2005; Foster and Dumas , 2001; Stranahan et al . , 2007 ) . We enhanced neurogenesis by housing WT mice with EE ( Figure 8A ) , a treatment reported to generate a 1 . 5–2-fold increase in the number of integrating new GCs ( van Praag et al . , 1999; Brown et al . , 2003; Olson et al . , 2006 ) . We previously found that housing mice with running wheels alone for four weeks increases the number of POMC-eGFP labeled GCs to 146% of age-matched controls ( Overstreet et al . , 2004 ) , suggesting that EE enhances neurogenesis to a similar or greater extent as observed in BaxKOim mice ( Figure 1B , C ) . To assess the strength of excitatory transmission from entorhinal cortex across the population of GCs and onto individual mature GCs , we again stimulated the medial perforant path while simultaneously recording fEPSPs and EPSCs from mature GCs in GABAA receptor antagonists ( Figure 8B ) . As previously reported ( Green and Greenough , 1986; Foster et al . , 1996 ) , the fEPSP slope was enhanced in slices from EE mice with no difference in the FV , suggesting an increase in total synaptic strength with no change axonal excitability ( Figure 8—figure supplement 1A , B ) . Indeed , normalizing the fEPSP slope to the FV to account for differences in the number of stimulated axons across slices revealed a significant increase in the fEPSP ( Figure 8C ) . We targeted mature GCs located near the outer edge of the granule cell layer and confirmed their maturity by intrinsic membrane properties ( Figure 8—figure supplement 1C ) . Consistent with the enhanced fEPSPs , EPSCs in mature GCs were larger in slices from mice housed in EE ( Figure 8D ) , such that the overall EPSC/FV ratio was 2 . 6 ± 0 . 16 in EE compared to 1 . 6 ± 0 . 07 in control ( n = 88 , 58 respectively , p<0 . 0001 unpaired t-test ) . Enhanced synaptic strength after EE could result either from increased release probability , increased number of synapses or increase in the number of receptors per synapse . We found no difference in the PPR of evoked EPSCs , suggesting that release probability is unchanged ( Figure 8E ) , as previously reported ( Foster et al . , 1996 ) . However , the frequency of sEPSCs was increased with no change in sEPSC amplitude ( Figure 8F ) , similar to the recently reported increase in miniature EPSCs in mature GCs after EE ( Kajimoto et al . , 2016 ) . Together these results suggest that enhanced evoked EPSCs in mature GCs result from greater number of functional synapses , consistent with increased spine density in Golgi-stained ( presumably mature ) dentate GCs ( Eadie et al . , 2005; Stranahan et al . , 2007 ) . These results show that mature GCs exhibit experience-dependent synaptic enhancement that argues against the recently described restricted period for experience-dependent plasticity of dentate GCs ( Bergami et al . , 2015 ) . However , these results cannot resolve whether integration of EE-induced newborn GCs affects synaptic function of mature neurons . Increased connectivity of mature neurons is likely a parallel phenomenon independent of neurogenesis , since similar increases in synaptic transmission and spine density occur in non-neurogenic regions ( Rampon et al . , 2000; Malik and Chattarji , 2012; Jung and Herms , 2014 ) . Thus , the magnitude of increased connectivity of mature GCs could be reduced by neurogenesis-induced synaptic redistribution . Altogether , these results highlight the capacity of mature GCs to undergo changes in synaptic connectivity in response to both genetic and experiential circuit manipulations . 10 . 7554/eLife . 19886 . 023Figure 8 . Environmental enrichment increases synaptic transmission to mature GCs . ( A ) The experimental timeline showing recordings performed 4–6 weeks after EE . ( B ) Left , simultaneous fEPSPs and whole-cell recordings from mature GCs , as shown in Figure 1 . Examples of fEPSPs ( top ) with FV ( insert ) and EPSCs in mature GCs ( bottom ) in slices from control and EE mice . ( C ) Slices from EE mice exhibited an increase in the fEPSP slope plotted against FV amplitude ( two-way ANOVA , Fmanipulation ( 1 , 116 ) =9 . 59 , p=0 . 0025 , n = 11 control , 9 EE ) . FVs were binned by 75 μV . ( D ) Left , an increase in synaptic transmission to mature GCs was revealed by the EPSC plotted against FV ( two-way ANOVA , Fmanipulation ( 1 , 150 ) =52 . 88 , p<0 . 0001 , n = 11 control , 9 EE ) . *p<0 . 01with Bonferroni post-tests . Right , the overall EPSC/FV ratio was enhanced by EE ( unpaired t-test , p<0 . 0001 ) . ( E ) The paired-pulse ratio of EPSCs ( 100 ms ISI ) was similar in EE and control mice ( p=0 . 181 unpaired t-test , n = 9 controls , 9 EE ) . ( F ) Spontaneous EPSCs in mature GCs from EE mice had higher frequency ( p=0 . 0081 unpaired t-test ) but similar amplitude as sEPSCs in mature GCs from control mice ( p=0 . 46 , n = 9 controls , 10 EE ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 02310 . 7554/eLife . 19886 . 024Figure 8—figure supplement 1 . EE enhances the fEPSP with no change in fiber volley and input resistance of mature GCs . ( A ) The FV amplitude plotted across stimulation intensity revealed no difference between . slices from EE and control mice ( p=0 . 602 two-way ANOVA , n = 11 control and 13 EE ) . ( B ) In the same recordings , the fEPSP slope was significantly greater in EE mice ( p=0 . 0006 ) . ( C ) The input resistance of mature GCs in whole-cell recordings confirmed the mature status of GCs after EE . Input resistance; control 299 ± 61 MΩ , n = 11; EE 267 ± 41 MΩ , n = 10; p=0 . 18 unpaired t-test . Action potential ( AP ) amplitude measured from threshold; control 83 ± 9 mV , EE 89 ± 7 mV , p=0 . 11 unpaired t-test . There was a significant reduction in the AP frequency measured at 100 pA current injection; control 24 ± 5 Hz , EE 19 . 4 ± 4 . 5 Hz; p=0 . 048 unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 024 Immature GCs make up a small percent of total GCs , and yet when neurogenesis was selectively manipulated the change in synaptic strength to mature GCs was unexpectedly robust . To determine whether the magnitude of altered transmission to mature GCs could be explained by a redistribution of existing synapses to integrating new GCs , we made a quantitative estimate of the proportion of mature synapses that would be transferred to new GCs over the time course of our experiments . We simulated the BaxKOim condition , since in this condition we quantified excitatory input to mature GCs and immature GCs , as well as the increase in new cells induced by Bax deletion . Other parameters were based on reported rates of neurogenesis ( Chancey et al . , 2013; Gil-Mohapel et al . , 2013 ) , cell death ( Sierra et al . , 2010 ) and excitatory synaptic integration ( Dieni et al . , 2013 , 2016 ) . The simulation is based on a static number of synapses that re-distribute to immature GCs according to their number and time-dependent synaptic integration ( Figure 9—figure supplement 1 ) . The simulation showed a steep increase in the proportion of synapses occupied by immature GCs in BaxKOim mice starting at the time point when immature GCs start to integrate into the network ( Figure 9A , red line ) . The robust transfer of synapses resulted not only from the increased number of immature GCs , but also from the increased acquisition of immature synapses resulting from Bax deletion . The predicted reduction in mature synapse number ( expressed as a % ) at days 36–43 in the simulation was similar to the % change in mature EPSCs measured experimentally ( Figure 9B ) . Despite the small proportion of immature GCs within the network ( initially set at 5% ) , the continuous increase in cell number along with enhanced synaptic integration was compounded over time to attenuate synapses on pre-existing neurons to a degree that could account for the magnitude of reduced synaptic strength observed in the BaxKOim experiments . 10 . 7554/eLife . 19886 . 025Figure 9 . Simulation of neurogenesis-induced synaptic redistribution . ( A ) Distribution of synapses occupied by mature and immature GCs using quantitative synaptic transfer simulation ( see Materials and methods ) . Lines indicate the percentage of synapses on mature and immature GCs across the duration of the BaxKOim experiment , with the total number of synapses held constant . ( B ) Experimentally measured %change in EPSCs in mature GCs ( left axis ) compared to the %change in mature synapse number predicted by the simulation at time points t = 36 through t = 43 ( right axis ) . Experimental data is the mean mature GC EPSC amplitude in BaxKOim mice normalized to control from each FV bin shown in Figure 1G . ( C ) Graphic depiction of synaptic integration of adult born neurons showing that new GCs ( green ) gain EC synapses ( orange ) through two possible sources: ( C1 ) New EC terminals may form to innervate new GCs . In this case , increasing neurogenesis would increase the total number of synapses over time but the synapses per individual mature GC would remain constant . ( C2 ) . Alternatively , new GCs may take over existing EC synapses from surrounding mature GCs . In this case , the total number of synapses would remain constant over time and the number of synapses per mature GC would decrease . The reduced synaptic input to mature GCs in BaxKOim mice coupled with the apparent lack of change in total synapses ( Figures 1–3 ) supports the synaptic redistribution model . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 02510 . 7554/eLife . 19886 . 026Figure 9—figure supplement 1 . Quantitative simulation of synaptic transfer . Graphic representation of a quantitative simulation of synapse redistribution between mature and integrating new GCs . Control ( left ) and BaxKOim ( right ) conditions are illustrated at progressive time points ( 7 , 21 , 35 and 42 days ) , with t = 0 being the day of tamoxifen-induced Cre recombination . There is a static number of EC synapses defined at the beginning of the simulation , and synapses occupied by mature or immature GCs are portrayed as a percent of the total . Proliferation rate multiplied by survival determines the number of new GCs incorporating into the network on each day . As immature GCs age , they each increase in synaptic connectivity represented as the number of synapses relative to mature GCs . The sum of the number of immature GCs at each age multiplied by their number of synapses determines the number of synapses appropriated by the immature population . The proliferation rate decreases with age in both control and BaxKOim conditions , and BaxKO GCs have both increased survival and increased synaptic integration . See Materials and methods for additional parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 19886 . 026
Our results showing that increasing neurogenesis decreased synaptic transmission and spine density of mature GCs is consistent with the idea that immature neuron synaptic integration is a competitive process ( Tashiro et al . , 2006; Toni and Sultan , 2011; McAvoy et al . , 2016 ) . Anatomical analysis has suggested that multisynaptic boutons ( MSBs ) represent an intermediary structure in the transfer of functional synapses from mature to immature GCs ( Toni et al . , 2007; Toni and Sultan , 2011 ) . Although we did not find evidence for alterations in the total number of functional synapses reflecting the presence of MSBs when neurogenesis was manipulated , shared transmission from MSBs may be functionally silent due to lack of AMPA receptors on new neurons ( Wu et al . , 1996; Chancey et al . , 2013 ) , or may be below the detection limits of field potential recordings . Furthermore , recent work suggests MSBs are a common feature of mature GCs and the complexity of MSB innervation increases with GC maturation ( Bosch et al . , 2015 ) , so it is unclear how our functional results relate to prior anatomical studies . Nevertheless , our results unambiguously demonstrate that neurogenesis modifies synaptic transmission to existing mature GCs through a mechanism that involves reduced number of functional synapses . Unlike prior reports of alterations in DG excitability following selective manipulations of neurogenesis , we isolated excitatory synaptic transmission using GABAA receptor antagonists , thus our results cannot be attributed to differential recruitment of local inhibitory circuits by immature GCs ( Singer et al . , 2011; Massa et al . , 2011; Ikrar et al . , 2013; Temprana et al . , 2015 ) . In addition to such feedback inhibition , regulation of the density of mature GC excitatory synapses could potentially contribute to the counter-intuitive finding that the number of immature GCs is inversely related to the excitability of the mature network ( Ikrar et al . , 2013; Drew et al . , 2016 ) . Our interpretation that integrating new GCs acquire synapses from mature GCs relies on the assumption that modulation of EPSCs reflects changes in synapse number . Several pieces of evidence support this assumption . First , to account for differences in the number of stimulated axons across slices , we normalized EPSCs in mature GCs to the simultaneously recorded fiber volley , a common approach used in synaptic plasticity studies . Thus , differences in EPSCs cannot result from systematic differences in the number of stimulated axons . Second , reduced evoked EPSCs were accompanied by reduced frequency of sEPSCs with no change in amplitude and no change in the PPR . These characteristics are widely accepted indicators of changes in synapse number . Third , strontium-evoked asynchronous EPSCs likewise supported the idea that small EPSCs in mature GCs from BaxKOim mice resulted from fewer active synapses rather than a postsynaptic change in sensitivity . We also found no difference in the Ca2+ sensitivity of EPSCs between BaxKOmat and control mice . This suggests that Bax deletion from the majority of GCs did not affect Ca2+ dependence of release processes , making it unlikely that a secreted factor acts presynaptically to alter release following Bax manipulation . Finally , we found that the density of mature GC spines was reduced after selective enhancement of neurogenesis . Our results are consistent with a model wherein newly generated GCs usurp pre-existing synapses from mature GCs , perhaps through an activity-dependent competitive process ( Tashiro et al . , 2006 ) , yet we cannot rule out other non-competitive mechanisms by which newly generated cells affect the number of synapses on mature GCs . The recent observation that conditional suppression of spines on mature GCs enhances the integration of newborn GCs further supports the interactions between new and existing neurons ( McAvoy et al . , 2016 ) . A synaptic re-distribution model predicts that the addition of new neurons does not alter the total number of synapses within the circuit ( Figure 9C ) . We used fEPSPs as a primary measure of total synapses , and presumably the fEPSP does not change despite the loss of EPSCs in mature GCs due to the additional contribution of synapses on immature neurons . Although we did not detect differences in fEPSPs ( or vGluT expression ) , it is important to note that fEPSPs may not be particularly sensitive to synaptic density and will also be affected by intrinsic excitability . We did not detect any differences in the intrinsic excitability of mature GCs in our genetic models , but it is expected that the higher intrinsic excitability of immature neurons would enable a greater contribution to fEPSPs compared to mature GCs ( for a given number of active AMPAR-containing synapses ) . However , newborn GCs have a high fraction of silent synapses that may limit their contribution to fEPSPs ( Chancey et al . , 2013 ) . Most importantly , our interpretation of synaptic redistribution is not affected if the immature GC contribution to the fEPSP does not fully compensate for the loss of transmission to mature neurons ( that is , if the fEPSP was reduced in BaxKOim mice ) . Only an increase in the fEPSP in BaxKOim mice would lend support a synaptic addition model . Even so , changes in fEPSPs are somewhat tangential to our novel finding that EPSCs in mature GCs are altered by selective manipulations of newborn GCs . Our results indicate that Bax is required in mature GCs for neurogenesis-induced loss of transmission , suggesting that a change in the Bax signaling pathway is involved in spine loss from mature GCs . The contribution of Bax in our experiments is thus complex . We show that mature GCs exhibit a non-cell autonomous effect of Bax deletion from adult-born GCs ( Figures 1–3 , decreased EPSCs ) that is opposite to the cell-autonomous effect of Bax deletion in both cell types ( Figures 5–6 , increased EPSCs ) . Remarkably , the cell autonomous function is required for the non-cell autonomous effect ( Figure 7 ) . This complexity , however , makes sense when we consider the role of Bax signaling in both cell death and synapse pruning . We propose that the non-cell autonomous effect results from enhanced neurogenesis ( supported by the observation that ablation of neurogenesis produced the opposite outcome , Figure 4 ) , whereas the cell autonomous effect results from a contribution of the Bax pathway in synaptic depression and spine pruning . Although the Bax signaling pathway is best known in the context of programmed cell death , it also has a non-apoptotic role in synaptic plasticity that is mediated by downstream caspases , the same family of cysteine proteases that initiate cell apoptosis ( Sheng and Ertürk , 2014 ) . Caspases mediate dendritic remodeling during neural development ( Kuo et al . , 2006b; Williams et al . , 2006; Riccomagno and Kolodkin , 2015 ) , and more recent work shows that caspase-3 activation is necessary and sufficient for NMDAR-mediated AMPA receptor internalization and LTD at hippocampal synapses ( Li et al . , 2010; Jiao and Li , 2011 ) . LTD is associated with spine shrinkage and is typically considered a herald of synapse pruning ( Oh et al . , 2013; Wiegert and Oertner , 2013 ) , thus it appears that pathways mediating cellular destruction also contribute to synaptic destruction ( Sheng and Ertürk , 2014 ) . Indeed , local induction of caspase-3 activity in dendrites triggers spine elimination whereas caspase-3 KO mice exhibit increased GC spine density ( Ertürk et al . , 2014; Lo et al . , 2015 ) , similar to our results of increased spine density in BaxKO GCs . Our findings that Bax deletion enhanced synaptic strength and spine density while blocking neurogenesis-induced loss of mature GC synaptic strength are consistent with the idea that on-going synaptic refinement controls the strength of excitatory transmission and that continual neurogenesis promotes a competitive environment for redistribution of synapses ( McAvoy et al . , 2016 ) . These results have potential implications for understanding the role of neurogenesis and plasticity in DG function . First , both enhancing neurogenesis and blocking output from mature GCs improves performance on the same context discrimination task ( Sahay et al . , 2011; Nakashiba et al . , 2012 ) , suggesting that neurogenesis could contribute to DG function by modifying mature GC activity . Synaptic depression and subsequent pruning are activity-dependent processes that typically require NMDA receptor activation ( Shipton and Paulsen , 2014 ) . Hence , re-distribution of active terminals away from mature GCs could transiently sparsify population activity , if new GCs initially have insufficient excitatory connectivity to allow recruitment ( Dieni et al . , 2016 ) . Second , since eliminating Bax in progenitors leads to greater innervation as well as greater survival of neural progeny , enhancing neurogenesis by blocking the apoptotic pathway likely promotes competition to a greater extent than other methods of increasing neurogenesis . This could have implications for understanding the potential role of enhanced neurogenesis using Bax deletion in behavioral outcomes assessing pattern separation , stress resilience and forgetting ( Sahay et al . , 2011; Akers et al . , 2014; Hill et al . , 2015 ) . Finally , our results showing that deletion of Bax signaling in postmitotic GCs enhances synaptic transmission is consistent with increased activation of DG neurons observed in caspase-3-/- mice , which also show behavioral deficits in attending to relevant stimuli ( Lo et al . , 2015 ) . Together , we speculate that synaptic redistribution between immature and mature GCs may contribute to activity-dependent synaptic remodeling that allows salient stimuli to receive precedence in DG encoding and may also contribute to circuit remodeling that degrades established memories ( Weisz and Argibay , 2012; Chambers et al . , 2004; Akers et al . , 2014; Epp et al . , 2016 ) .
All animal procedures followed the Guide for the Care and Use of Laboratory Animals , U . S . Public Health Service , and were approved by the University of Alabama at Birmingham Institutional Animal Care and Use Committee ( protocol# 8674 and 10134 ) . Mice of either gender were maintained on a 12 hr light/dark cycle with ad libitum access to food and water . BaxKOimmature mice were generated by crossing heterozygous loxP-flanked Bax mice ( Jackson #006329 , the Bak1 null allele was bred out ) with Nestin-CreERt2 mice ( Jackson #016261 ) . The offspring were crossed with each other to produce Nestin-Cre+ or -/Baxfl/fl , Baxfl/+ , or Bax+/+ animals ( see Figure 1—figure supplement 1 ) . Eight week-old mice were injected with tamoxifen ( TMX , from a 20 mg/ml stock dissolved in sunflower seed oil , 75 mg/kg for three consecutive days ) to induce recombination and experiments were done 4–6 weeks post-injection . Control Nestin-Cre- or Bax+/+ genotypes received TMX injections with the same protocol . For knockdown of neurogenesis , homozygous iDTR mice ( Jackson #007900 ) were crossed with male Nestin-CreERtm4 mice provided by Chay Kuo ( Kuo et al . , 2006a ) to obtain offspring that were iDTR+ and either Nestin-Cre+ ( Ablatedimmature ) or Nestin-Cre- ( control group ) . All mice were given TMX injections between 6–8 weeks of age , followed by diphtheria toxin injections six weeks later ( DT , 16 µg/kg in sterile saline for three consecutive days ) . To conditionally delete Bax from postmitotic GCs , we crossed POMC-Cre mice ( Jackson #005965 ) with Baxfl/fl mice ( see Figure 6—figure supplement 1 ) . Conditional knockouts were maintained on a mixed 129 and C57BL/6J background using sibling controls . For counting newborn GCs , mice were crossed with POMC-eGFP transgenic mice ( Jackson #009593 ) . In some experiments , we visualized Cre-expressing cells by crossing conditional lines with Ai14 reporter mice ( Jackson #007914 ) . Tissue from homozygous germ line BaxKO mice ( Jackson #002994 ) was used to validate Bax antibodies in western blots , with Bax+/- mice crossed with each other to generate both Bax-/- and control Bax+/+ genotypes . All experiments were performed in adult P60-P120 mice . Mice were anesthetized and perfused intracardially with cold cutting solution containing ( in mM ) : 110 choline chloride , 25 D-glucose , 2 . 5 MgCl2 , 2 . 5 KCl , 1 . 25 Na2PO4 , 0 . 5 CaCl2 , 1 . 3 Na-ascorbate , 3 Na-pyruvate , and 25 NaHCO3 . The brain was removed and 300 μm horizontal slices were taken on a Vibratome 3000EP or Leica VT1200S ( Leica Biosystems , Wetzlar , Germany ) . After recovery in artificial CSF ( ACSF ) containing ( in mM ) : 125 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 2 CaCl2 , 1 MgCl2 , 25 NaHCO3 , and 25 glucose , recordings were performed at 30°C in ACSF +100 μm picrotoxin ( PTX ) to block GABAA receptors . Patch pipettes were filled with the following ( in mM ) : 115 K-gluconate , 20 KCl , 4 MgCl2 , 10 HEPES , 4 Mg-ATP , 0 . 3 Na-GTP , 7 phosphocreatine , 0 . 1 EGTA , pH 7 . 2 and 290 mOsm ( 2–4 MΩ ) . In some cases , a 0 . 2% biocytin was included in the patch pipette . Field pipettes were placed in the middle molecular layer and filled with ACSF ( 1–2 MΩ ) . A patch pipette filled with 1M NaCl ( 1 MΩ ) was used to stimulate the middle molecular layer using an isolated stimulator ( Digitimer , Letchworth Garden City , UK ) . The minimum stimulation intensity that evoked an EPSC was first established and the stimulus intensity was increased at multiples of the threshold intensity until response saturation was evident . In some experiments we tested a pre-set range of stimulus intensities , again ceasing stimulation after responses saturated . Both methods used the same range of intensities ( 0 to 100 V ) with each approach generating fewer independent observations at progressively higher stimulus intensities due to saturation of axonal recruitment . The response of 10 stimuli at each intensity was averaged . Averaged field EPSPs ( fEPSPs ) and EPSCs were binned by their corresponding fiber volley ( FV ) amplitude . This normalizes for differences in stimulus intensities across experiments and removes the parameter ‘stimulus intensity’ from data sets . Anesthetized mice were perfused intracardially with 0 . 9% NaCl or 0 . 1 M PBS and chilled 4% PFA before brains were removed and post-fixed overnight in PFA . Free-floating horizontal slices were taken on a Vibratome 1000 ( 50 μm ) . To enhance endogenous GFP expression , slices were blocked in TBS block buffer ( 0 . 1M TBS , glycine , 3% bovine serum albumin , 0 . 4% Triton X-100 and 10% normal goat serum ) and incubated overnight with anti-GFP conjugated Alexa 488 ( 1:1000 , Invitrogen , Carlsbad , CA ) . For NeuN and Dcx , slices were washed in TBST ( 50 mM Tris , 0 . 9% NaCl and 0 . 5% Triton X-100 ) and treated with antigen retrieval solution ( 10 mM sodium citrate , 0 . 5% tween 20 ) and 0 . 3% hydrogen peroxide before block with TBST +10% normal goat serum , followed by 48 hr incubation in rabbit anti-NeuN antibody ( 1:1000 , Millipore , Billerica , MA ) or rabbit anti-Dcx antibody ( 1:500 , Abcam , Cambridge , UK ) , respectively . For NeuN , this was followed by incubation of 4 hr with goat anti-rabbit Alexa 647 ( Invitrogen ) . For Dcx , a 3 hr incubation with biotinylated goat anti-rabbit ( 1:800 , Southern Biotech , Homewood , AL ) was followed by a 30 min incubation with streptavidin conjugated to Alexa Fluor 647 ( 1:200 , Invitrogen ) . Slices were mounted with Prolong Gold or VectaShield mounting medium ( Invitrogen ) . To visualize spines , acute brain slices containing biocytin-filled cells were post-fixed in 4% PFA for at least 24 hr then stained with streptavidin conjugated to Alexa Fluor 647 ( 1:1000 , Invitrogen ) . EGFP+ cells and doublecortin ( Dcx+ ) cells were counted using the optical fractionator method from every sixth slice through the entire left dentate gyrus using StereoInvestigator software ( MBF Bioscience , Williston , VT ) . Counting frame and SRS grid sizes were set to give a Gunderson coefficient of error of <0 . 1 by an investigator blinded to genotype . For Figure 3 , mature GCs from control and BaxKOim mice were patched using an internal solution that included 0 . 2% biocytin . After fixation , GC dendrites and spines were imaged on an Olympus Fluoview 300 confocal microscope with a 60X objective and a 3X digital zoom using a z-step of 0 . 1 μm . Dendritic segments that were relatively horizontal to the plane of the slice were selected for spine analysis by an investigator blinded to genotype ( avg segment length = 46 ± 7 µm in control and 56 ± 4 µm in BaxKOim mice , p=0 . 3 ) . Analysis of spine density and type was performed by an investigator blinded to genotype using NeuronStudio software ( Rodriguez et al . , 2008 ) . For Figure 6 , TdTomato+ ( Bax-/- ) or tdTomato- ( Bax+/+ ) cells were patched in alternating slices from BaxKOmature mice and processed as described above . Spine density , length and head width were analyzed using Imaris software ( Bitplane , Belfast , Northern Ireland ) ( Swanger et al . , 2011 ) . Hippocampal lysates were prepared by homogenizing flash frozen subdissected hippocampi using RIPA buffer ( 150 mM NaCl , 50 mM Tris , pH 7 . 5 , 1% Triton-X 100 , 0 . 5% sodium deoxycholate , 1% sodium dodecyl sulfate ) containing protease inhibitors ( Fisher Scientific , Hampton , NH ) . Following BCA assay ( Pierce ) , 20 μg of lysate was separated through 12% polyacrylamide gels and transferred to low-fluorescent PVDF ( Biorad , Hercules , CA ) . Membranes were blocked with casein blocking buffer ( Sigma-Aldrich , St . Louis , MO ) in Tris buffered saline with 0 . 1% Tween 20 ( TBST ) and incubated with primary antibody ( in 0 . 3% BSA in TBST ) at 4°C overnight using antibodies to detect Bax ( Fisher Scientific ) or beta-tubulin ( Developmental Studies Hybridoma Bank ) . Secondary antibodies conjugated to Alexa-680 ( Fisher Scientific ) allowed detection and quantification by scanning with an Odyssey Imaging System ( Licor Biosciences , Lincoln , NE ) . Data are expressed as mean ± SEM . We set the alpha level at 0 . 05 and accepted significant results with p<0 . 05 for all statistical tests . When determining the effect of genotype between two samples , data sets that satisfied normality criteria were analyzed with two-tailed paired or unpaired t tests , while non-normal data sets were analyzed with Mann-Whitney or Wilcoxon tests . For comparing two genotypes across multiple stimulus intensities , a two-way ANOVA was used . When EPSCs or fEPSPs were binned by FV amplitude , the number of data points varied between samples requiring an unweighted means analysis . Statistics were performed using Graphpad Prism . The purpose of the calculation is to predict the proportion of mature GC synapses that will be appropriated by immature cells over a 6-week time period in a control or BaxKOim DG . Time ( t ) is expressed in days , where t = 0 represents the starting point when 8-week-old animals are injected with TMX . New GCs are continually added to an existing network comprised of mature and immature GCs . Each new GC gains synaptic strength beginning two weeks after cell birth ( Ge et al . , 2006; Mongiat et al . , 2009; Dieni et al . , 2013 ) , acquiring innervation from a finite pool of synapses with synaptic strength defined as the number of synapses per cell . The total number of GCs was initially set at 200 , 000 ( unilateral cell count in the adult mouse DG ( Pugh et al . , 2011 ) . The number of mature GCs ( >8 weeks cell age ) was set at 95% of the total ( 190 , 000 ) , while the initial number of immature cells ( 2–8 weeks cell age ) was set at 5% of the total ( 10 , 000 ) ( Imayoshi et al . , 2008 ) . The baseline number of mature GC synapses at t = 0 was set at 100% , defined as 100 per cell , giving initial mature synapse number , SM:SM=100 ( 190 , 000×0 . 95 ) We approximated the increase in synaptic strength , Y ( t ) , of developing GCs by fitting the amplitude of evoked EPSCs in immature GCs at progressive ages ( Dieni et al . , 2013 ) by the equation:Y ( t ) =71 . 1ln ( 14+t ) −187 . 7 For example , a 2-week-old control GC receives ~5% as many excitatory synapses as a mature GC , a 5-week-old GC contains ~65% as many excitatory synapses , and an 8-week-old GC achieves ‘mature’ levels of 100% synaptic strength . To determine the initial number of immature synapses , SI ( 0 ) , we divided the number of initial immature GCs by 43 ( the number of days of maturation and thus the number of different synaptic strengths ) and multiplied this quantity by the sum of all synaptic strengths:SI ( 0 ) =10 , 000/43×∑t=143Y ( t ) This result plus the initial number of mature synapses gives the total synapses in the system:SM + SI ( 0 ) which remains static throughout the simulation ( ~19 . 6 million ) . To calculate the number of synapses appropriated by immature GCs each day , we considered cell proliferation P ( t ) , the rate of cell survival , and synaptic strength Y ( t ) . The rate of decrease in progenitor proliferation was defined by a best-fit equation ( Gil-Mohapel et al . , 2013 ) , adjusted to give ~8000 progenitor cells at t = 14 , ( stereological ki67 counts from 8-week-old mouse ) ( Chancey et al . , 2013 ) , giving the available progenitor cell number , P ( t ) :P ( t ) =4×106 ( 42+t ) −1 . 5 The survival rate for new WT cells is 20% ( Sierra et al . , 2010 ) . In the BaxKOim group , new GCs incorporating into the network at t = 14 ( 2 weeks after TMX-induced recombination ) have a survival rate of 70% ( assuming partial efficiency of Cre expression ) ( Lagace et al . , 2007 ) . The number of immature GCs added to the system per day , I ( t ) , is:I ( t ) =P ( t ) ×survival rate All immature GCs will gain synaptic strength daily . The immature synapses appropriated each day , SI ( t ) , is the cumulative sum of the surviving GCs times their respective synaptic strengths:SI ( t ) = ( I ( t ) ×Y ( 1 ) ) + ( I ( t−1 ) ×Y ( 2 ) ) + ( I ( t−2 ) ×Y ( 3 ) ) … Importantly , BaxKO GCs possess ~35% more synapses than control due to lack of Bax-dependent synapse pruning ( Figure 3E , EPSC increase at highest FV bin ) . In both groups , the cumulative number of immature synapses divided by the total synapses ( multiplied by 100 ) equals the percent synapses appropriated by the immature population:%im=SI ( t ) SM + SI ( 0 ) ×100 Since there is a static number of total synapses defined at the start of the simulation , the percent mature synapses remaining is:%mat=100−%im The %synapses occupied by all cell groups across time is plotted in Figure 7 . Since the experiment is less than eight weeks in duration , immature GCs never convert into mature GCs , and we did not account for the conversion of pre-existing WT immature GCs because that population would not differ between control and BaxKOim conditions . To calculate the predicted difference in mature synapse number in BaxKOim vs . control conditions , we took the ratio of %mat in BaxKOim to %mat in control at each time point from t = 36 through t = 43 ( multiplied by 100 ) . | Neurogenesis , the creation of new brain cells called neurons , occurs primarily before birth . However , a region of the brain called the dentate gyrus , which is involved in memory , continues to produce new neurons throughout life . Recent studies suggest that adding neurons to the dentate gyrus helps the brain to distinguish between similar sights , sounds and smells . This in turn makes it easier to encode similar experiences as distinct memories . The brain’s outer layer , called the cortex , processes information from our senses and sends it , along with information about our location in space , to the dentate gyrus . By combining this sensory and spatial information , the dentate gyrus is able to generate a unique memory of an experience . But how does neurogenesis affect this process ? As the dentate gyrus accumulates more neurons , the number of neurons in the cortex remains unchanged . Do some cortical neurons transfer their connections – called synapses – to the new neurons ? Or does the brain generate additional synapses to accommodate the newborn cells ? Adlaf et al . set out to answer this question by genetically modifying mice to alter the number of new neurons that could form in the dentate gyrus . Increasing the number of newborn neurons reduced the number of synapses between the cortex and the mature neurons in the dentate gyrus . Conversely , killing off newborn neurons had the opposite effect , increasing the strength of the synaptic connections to older cells . This suggests that new synapses are not formed to accommodate new neurons , but rather that there is a redistribution of synapses between old and new neurons in the dentate gyrus . Further work is required to determine how this redistribution of synapses contributes to how the dentate gyrus works . Does redistributing synapses disrupt existing memories ? And how do these findings relate to the effects of exercise – does this natural way of increasing neurogenesis increase the overall number of synapses in the system , potentially creating enough connections for both new and old neurons ? | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2017 | Adult-born neurons modify excitatory synaptic transmission to existing neurons |
Many organisms use free running circadian clocks to anticipate the day night cycle . However , others organisms use simple stimulus-response strategies ( ‘hourglass clocks’ ) and it is not clear when such strategies are sufficient or even preferable to free running clocks . Here , we find that free running clocks , such as those found in the cyanobacterium Synechococcus elongatus and humans , can efficiently project out light intensity fluctuations due to weather patterns ( ‘external noise’ ) by exploiting their limit cycle attractor . However , such limit cycles are necessarily vulnerable to ‘internal noise’ . Hence , at sufficiently high internal noise , point attractor-based ‘hourglass’ clocks , such as those found in a smaller cyanobacterium with low protein copy number , Prochlorococcus marinus , can outperform free running clocks . By interpolating between these two regimes in a diverse range of oscillators drawn from across biology , we demonstrate biochemical clock architectures that are best suited to different relative strengths of external and internal noise .
Extracting information from a noisy external signal is fundamental to the survival of organisms in dynamic environments ( Bowsher and Swain , 2014 ) . From yeast anticipating the length of starvation ( Mitchell et al . , 2015 ) and bacteria estimating sugar availability ( Tu et al . , 2008 ) , to dictyostelium counting cAMP pulses ( Cai et al . , 2014 ) , organisms must often infer properties of the environment that are masked by noisy irregular fluctuations in order to be well-adapted ( Siggia and Vergassola , 2013; Mora and Wingreen , 2010 ) . A striking example of regularity in environmental stimuli is the daily day-night cycle of light on earth; organisms from all kingdoms of life use circadian clocks to synchronize - or ‘entrain’ - in phase to these 24-hour periodic signals in order to anticipate and prepare for future changes in light ( Winfree , 2001 ) . The most remarkable and well-studied examples of clocks are free running circadian clocks , found in organisms ranging from the cyanobacterium S . elongatus to insects , plants and humans . Such clocks use non-linear dynamics to generate self-sustained 24-hr rhythms of a preferred amplitude even in the absence of external driving . Many salient properties have been ascribed to such free running internal rhythms ( Troein et al . , 2009; Winfree , 2001 ) . However , several organisms have only damped clocks or ‘hourglass clocks’; their response to daily changes in light is not a self-sustaining oscillation , but rather a physiological program that decays to a steady state over a day . For example , some strains of P . marinus , a smaller 0 . 5μm cyanobacterium with an estimated 50× smaller protein copy number than S . elongatus ( Bryant , 2003; Gutu et al . , 2013; Holtzendorff et al . , 2008; Dufresne et al . , 2003; Kitayama et al . , 2003 ) , appear to have such a damped ‘hourglass’ clock , despite the clock being constituted from Kai proteins similar to those in S . elgonatus . The potential benefits and drawbacks of these timing systems are not immediately obvious . In particular , it is unclear when an ‘hourglass’ clock might be sufficient or even preferred over free running clocks . Here , we compare such classes of clocks when driven by the day-night cycle of light in fluctuating conditions . One source of fluctuations are amplitude fluctuations in the external day-night signal due to weather patterns ( Gu et al . , 2001 ) or other environmental disturbances . Phase entrainment to such fluctuating environmental signals is a challenge because while amplitude fluctuations are uninformative of phase , an entrainment mechanism looking for dawn-dusk transitions might conflate such amplitude fluctuations with true variations in phase . Biomolecular clocks also face an internal source of fluctuations ( Lestas et al . , 2010 ) , due to various causes like finite copy number effects ( Tsimring , 2014 ) , bursty transcription , interactions with the cell cycle and cell division ( Teng et al . , 2013 ) . It is clear that the inability to deal with either of these fluctuations will lead to poor phase entrainment , with a host of associated fitness costs in cyanobacteria ( Woelfle et al . , 2004 ) , plants , rodents and humans ( Evans and Davidson , 2013 ) . However , it is not clear what kinds of clock architecture are best at dealing with internal and external fluctuations and whether these demands are compatible . We find that free running clocks , based on limit cycle attractors , are a double-edged sword when subject to such internally and externally fluctuating conditions . The flat direction along such continuous limit cycle attractors can selectively project out external amplitude fluctuations while retaining phase information . However , the flat direction along the attractor makes these continuous attractor-based clocks susceptible to internal fluctuations ( e . g . low protein copy number [Potoyan and Wolynes , 2014] ) . In contrast , point attractor-based damped clocks are relatively resistant to internal fluctuations because they have no flat directions . Hence such ‘hourglass’ clocks out-perform free running clocks at sufficiently high internal noise . We first demonstrate our results in diverse biochemical oscillators , drawn from the literature ( Leloup et al . , 1999; Schmal et al . , 2014; Locke et al . , 2005; Leloup and Goldbeter , 2003; Goldbeter , 1991; Goodwin , 1965; Gonze and Abou-Jaoudé , 2013; Kondepudi and Prigogine , 2014; Elowitz and Leibler , 2000; Buşe et al . , 2009; Potvin-Trottier et al . , 2016 ) on clocks in cyanobacteria , plants and mammals to cell cycle and synthetic oscillators . We complement this detailed network-based study with dynamical systems theory that explains the same trade-off in terms of the broad features common to the diverse models studied here . In all cases , our approach involves systematically deforming the dynamics to interpolate between free running and ‘hourglass’ clocks and using information theoretic measures to quantify clock performance in the presence of fluctuations . By continuously interpolating between these clock architectures , our work predicts that a survey of clock systems in different environmental niches will reveal that clock architecture vary systematically with the relative strength of external and internal fluctuations ( Laughlin , 1981 ) . Further , our work suggests intriguing forward evolution experiments in the lab where the same structured external environment can nevertheless result in distinct regulatory systems , depending on the size of internal fluctuations . Finally , the existence of ‘hourglass’ clocks are easier to overlook experimentally than free running oscillations . Hence our theoretical demonstration that ‘hourglass’ clocks have functional benefits in specific conditions highlights the importance of experiments that specifically look for such damped clocks . More broadly , our work highlights the need to experimentally probe regulatory strategies by varying different kinds of noise independently when possible , since the strategies to deal with different kinds of noise are not equivalent and can be in conflict .
Many organisms like humans and rodents have free running clocks that show self-sustained 24 hr rhythms even in constant dark or light conditions . A particularly simple and well-characterized free running clock is that found in S . elongatus where the clock dynamics can be reproduced by the post-translational dynamics of Kai ABC in vivo as well . Measuring the phosphorylation state at any one of several sites on KaiC reveals an orderly phosphorylation reaction with a period of 24 hr . As shown in Figure 1a , oscillations of a characteristic amplitude are sustained even in constant darkness or constant light , that is , in the absence of a periodic external drive . Not all organisms have a free-running clock; for example , many insects ( Saunders , 2002 ) have damped ‘hourglass’ clocks that decay to a fixed point under constant light or constant dark conditions but show oscillatory dynamics under day-night cycling ( see Figure 1b ) . In fact , a sister cyanobacterial species P . marinus has a KaiBC-protein based clock . While the details of this clock are not fully characterized , the clock lacks the KaiA-mediated negative feedback ( Dufresne et al . , 2003; Holtzendorff et al . , 2008 ) loop that enables free running oscillations in S . elongatus . Consequently , in constant light or dark conditions , the clock’s state decays to a distinct day or a night state respectively ( Holtzendorff et al . , 2008 ) . Thus , both classes of clock show regular oscillations when externally driven . With cloudless day-night cycling , both systems can synchronize in phase with the external signal ( i . e . , ‘entrain’ ) and show distinct clock states at distinct times of the day . In this way , the clock state provides the rest of the cell with an estimate of the time of the day . However , while the free running clock has a natural amplitude relatively independent of the external drive , the damped clock’s amplitude is directly set by the external drive . The day-night pattern of light on earth does not resemble the clean square wave shown in Figure 1a but is rather subject to large amplitude fluctuations during the day due to weather patterns . Such amplitude fluctuations and their spectrum have been quantified ( Gu et al . , 2001 ) and also identified as playing a critical role in several studies on the evolution and performance of circadian clocks ( Domijan and Rand , 2011; Troein et al . , 2009 ) . The biological impact of such changes in light intensity in cyanobacteria have been quantified recently ( Teng et al . , 2013 ) . The clock must entrain in phase to the external signal while ignoring such amplitude fluctuations . In addition to external fluctuations , circadian clocks also deal with the intrinsically noisy nature of biochemical reactions ( Swain et al . , 2002 ) . Sources of internal noise for clocks include finite copy number effects , bursty transcription , cell division and other sources ( Tsimring , 2014 ) . In particular , based on their relative sizes ( Dufresne et al . , 2003; Holtzendorff et al . , 2008; Bryant , 2003 ) , P . marinus is thought to have far fewer copies of the Kai clock proteins ( e . g . , ∼500 of KaiC ) than S . elongatus ( ∼O ( 10000 ) copies of KaiC [Gutu et al . , 2013; Kitayama et al . , 2003] ) . Such finite numbers of molecules is known to create significant stochasticity in oscillators in the absence of an external signal ( Potoyan and Wolynes , 2014 ) . We tested the impact of such external and internal fluctuations on the contrasting clock architectures in S . elongatus and P . marinus through simulations . We set up explicit Gillespie simulations ( Gillespie , 2007 ) of explicit biomolecular models of the post-translational Kai clock that captures the known biochemistry ( Rust et al . , 2007 ) of S . elongatus’s clock and the putative KaiBC clock ( Bryant , 2003; Holtzendorff et al . , 2008 ) in P . marinus ( Figure 1 ) . We do not include transcriptional coupling ( Zwicker et al . , 2010 ) of the clock here and focus on the core post-translational oscillator . See Appendix 1 for details . The ATP levels in these models ( Pattanayak et al . , 2014 ) were coupled to an external square wave input of period 24 hr , representing the day-night cycle of light . To model external fluctuations , we modulated the amplitude of the input square wave over a broad range of frequencies , reflecting the broad frequency spectrum quantified by the Harvard Forest database ( Moore et al . , 1996 ) . To model internal fluctuations , we varied the copy number in these Gillespie simulations . With only external fluctuations but suppressing internal fluctuations using high copy numbers , we find that the damped oscillator develops a much larger population variance than the free running clock . In contrast , at low copy number ( i . e . , high internal noise ) but with a noiseless external signal , we find the situation is reversed; the free running clock has significantly higher population variance . See Figure 1c . To study this effect quantitatively , we fixed the strength of amplitude fluctuations and increased the internal noise by decreasing the copy number of all Kai proteins in the Gillespie simulation . We measured the resulting mutual information between clock state and objective time of day . ( Mutual information is intuitively a measure of population variance along the most informative directions; see Appendix 4 for more . ) We see that the free running clock has higher precision than the damped clock at low internal noise ( high copy number ) . However , as the internal noise is increased , the precision of the free running clock drops earlier and consequently , the damped oscillator has higher precision at sufficiently high internal noise ( low copy number ) . This is shown in Figure 1d , where the precision measures the mutual information between the clock state and the time . For a fair comparison , in undriven conditions , different clocks are assumed to lose information at the same rate . While our study here was motivated by the contrasting Kai protein-based clocks in the two cyanobacterial species S . elongatus and P . marinus , we sought to test the broader validity of our results . Hence we analyzed the internal and external noise resistance in a range of eight well-studied biochemical oscillators in the literature . These models range from circadian clocks in numerous organisms - Neurospora ( Leloup et al . , 1999 ) , Arabidopsis ( Schmal et al . , 2014; Locke et al . , 2005 ) , mammalian cells ( Leloup and Goldbeter , 2003 ) - to other oscillators such as cell cycle models ( Goldbeter , 1991 ) , the Goodwin ( Goodwin , 1965; Gonze and Abou-Jaoudé , 2013 ) oscillator , the Brusselator ( Kondepudi and Prigogine , 2014 ) and the synthetic repressilator ( Elowitz and Leibler , 2000; Buşe et al . , 2009 ) - see Figure 2 . While the internal noise properties of these oscillators in undriven conditions have been studied before ( Gonze et al . , 2002 ) , here we analyzed the contrasting internal and external noise properties of these oscillators under externally driven conditions . The results are shown in Figure 2 . In each case , we set all kinetic parameters to values specified in the original publications and coupled the external driving signal in the way specified in those original publications . As in the Kai clock simulations , the external signal was a square wave with amplitude fluctuations of fixed strength . Finally , we add Langevin noise to the equations to simulate internal noise; when available , we followed the finite volume prescription for rates in the original publications or related papers to set the size of Langevin noise for each reaction . Simulation and model details are in Appendix 2 . These models here all exhibit a Hopf bifurcation as kinetic parameters are tuned . The publications ( Leloup et al . , 1999; Schmal et al . , 2014; Locke et al . , 2005; Leloup and Goldbeter , 2003; Goldbeter , 1991; Goodwin , 1965; Gonze and Abou-Jaoudé , 2013; Kondepudi and Prigogine , 2014; Buşe et al . , 2009 ) identified a parameter which when tuned leads to a Hopf bifurcation; that is , on one side of the bifurcation , we find damped oscillations while on the other side , we find free running oscillations of increasing amplitude . We picked three points along this parameter; the green and purple data correspond to free running oscillations of large and smaller natural amplitude relative to the size of the external drive . The red data corresponds to a choice of parameters on the other side of the Hopf bifurcation , that is , to damped oscillations . For the red data , we chose μ such that the relaxation timescale was comparable to the period of the external driving force , much as in the Kai model of P . marinus . The damped oscillator is a useful clock only when the relaxation timescale is comparable to the period . In each case , we observed the same trade-off as seen in the Kai system; free running oscillations of large amplitude relative to the external drive ( green ) were most precise when only subject to external noise but are most fragile to internal noise . Damped oscillations in the same oscillator models are more robust and thus are preferable at sufficiently high internal noise . We find that intermediate amplitude free running oscillations show intermediate noise properties . Consequently , we can continuously trade-off resistance to internal noise for resistance to external noise by changing the amplitude of free running oscillations relative to the strength of the external drive . We have demonstrated a trade-off between external and internal noise resistance in diverse clocks . While it is possible to trace the origin of this trade-off to specific features of each clock , here , we take a different approach based on dynamical systems theory . Dynamical systems theory has been use to make fruitful general predictions about biological clocks since Winfree’s analysis of phase singularities ( Winfree , 2001 ) . In a similar vein , we use dynamical systems theory to show this trade-off emerges due to basic features of free running and damped clock dynamics and can thus be expected to hold broadly . Free running clocks are phenomenologically well-described by a limit cycle attractor , a non-linear oscillator of fixed amplitude ( Winfree , 2001 ) . While such descriptions have been used for numerous biochemical oscillators , limit cycle dynamics can be experimentally seen in molecular detail for the KaiABC clock in S . elongatus as shown in Figure 3a ( reproduced from [Leypunskiy et al . , 2017] ) . The clock follows distinct limit cycle dynamics during the day ( orange data ) and night ( black data ) ( Leypunskiy et al . , 2017; Pattanayak et al . , 2014 ) , with the day cycle positioned at higher phosphorylation levels due to a higher ATP/ADP ratio . The Kai model and indeed the diverse range of biochemical oscillators in Figure 2 show such a change in the limit cycle between day and night conditions . Here , we build a minimal model of such free-running clocks using circular day and night limit cycles of radius R in a plane . The limit cycle is defined by the dynamics τrelaxr˙=r-r3/R2 , θ˙=ω about its own center; but the center of the limit cycle itself moves along the x axis in Figure 3b as ( ρ ( t ) L , 0 ) where ρ ( t ) ∈[0 , 1] is the normalized light intensity level at time t and L is a measure of the physiological changes between day and night ( e . g . , ATP/ADP ratio change in S . elongatus ) . Thus , for example in Figure 3b , the system follows the blue dynamics at night and then after dawn it relaxes to the orange day attractor on a time scale τrelax . Note that R represents the amplitude of free-running oscillations while L represents the strength or amplitude of the external driving signal . In contrast , damped clocks are phenomenologically well-described by a day-time and a night-time point attractor with slow relaxation dynamics between them ( Figure 3c ) , modeled as r˙=-r/τrelax , θ˙=ω about an attractor point whose location varies with current light levels as ( -ρ ( t ) L , 0 ) . Here we assume 2τrelax∼24 hrs as in P . marinus ( Holtzendorff et al . , 2008 ) ; if relaxation were faster and completed before the day is over , the clock cannot resolve all times of the day . Here , we will also consider a family of limit cycle clocks of varying R/L to interpolate between large-R/L limit cycles and point attractors ( approximated by R/L=0 ) . We begin with the performance of different clocks in the presence of external intensity fluctuations . Weather patterns cause large fluctuations in the intensity of light over a wide range of time-scales as shown in Figure 4a . Much like with biochemical circuits , we subject an in silico population of dynamical system clock models to different realizations of such noisy weather patterns . When subject to weather fluctuations , we see in Figure 4b that the population variance of clock states for limit cycles at given times ( purple ) is fundamentally limited by the spacing between the day and night limit cycles . Point attractors develop larger overlapping population distributions at different times . We can geometrically understand the daytime phase variance increase σclouds2 in terms of the phase lag ΔΦ due to a single , say 2 . 4 hr dark pulse , administered during the day . Figure 4c shows that the deviation in trajectory for limit cycle clocks ( purple ) is fundamentally limited by the presence of a continuous attractor . In contrast , for the point attractor , the trajectory is in free fall towards the night point attractor , with no limit cycle to arrest such a fall . Consequently , the geometrically computed phase shift ΔΦ due to the particular dark pulse shown in Figure 4c is much smaller for limit cycles ( ΔΦ∼0 . 5 hr for the R , L geometry shown ) than for point attractors ( ΔΦ∼4 hr ) ( see Appendix 5 ) . In fact , this contrast in ΔΦ between limit cycles and point attractors holds for dark pulses of any duration and time of occurrence . The contrast is even greater at small L/R since ( ΔΦ ) 2∼ ( L/R ) 2 for small L/R , as shown geometrically in Appendix 5 and confirmed in simulations that average over random weather conditions ( Figure 4d ) . Hence , limit cycles are more resistant to external fluctuations than point attractors . To complete the analysis , we note that phase variance increases additively during the day and falls multiplicatively at dusk ( and dawn ) , that is , σ2→dayσ2+σclouds2→dusk ( σ2+σclouds2 ) /s2→night ( σ2+σclouds2 ) /s2→dawn ( σ2+σclouds2 ) /s4 . Solving for steady state phase variance ( σ2= ( σ2+σclouds2 ) /s4 ) , we obtain ( 1 ) σlimitcycle2 , ext∼ΔΦ2/ ( s4-1 ) . where we have equated σclouds2 to ΔΦ2 for a typical dark pulse Here , s2 represents the variance drop during a dawn/dusk entrainment . As shown in Appendix 5 for external noise ( and in Figure 5 for internal noise ) , this factor s , can be geometrically explained by the slope of the circle map relating the two cycles Leypunskiy et al . , 2017; we find that s2-1∼L/R for large-R/L limit cycles . Plugging this and ΔΦ2∼ ( L/R ) 2 into Equation1 , we see that σ2→L/R→0 for large -R/L cycles . Figure 4e shows that the precision ( i . e . , mutual information between clock state and time ) computed from random weather simulations agrees with this theory; clock precision drops as we interpolate from limit cycles to point attractors by changing L ( with equivalent results for changing R ) . Internal noise due to finite copy number effects in biochemical networks can be modeled exactly using the Gillespie method used in Figure 1 . In the context of our dynamical systems model , we follow Gillespie , 2007 and add Langevin noise to all dynamical variables of the system of strength ϵint∼1/N , where N is the overall copy number , with the ratios of different species assumed fixed ( see Appendix 3 ) . Such a Langevin approach is an approximation Gillespie , 2007 to the exact Gillespie method used in Figure 1 . We simulated a population of clocks in externally noiseless day-night light cycles but with internal Langevin noise . We see in Figure 5b that limit cycle populations have significantly higher variance of clock state due to internal noise than point attractors , in contrast to Figure 4b with external noise alone . We can understand the weakness of limit cycle attractor relative to the point attractor in terms of diffusion during day/night balanced by dawn/dusk transitions . The flat direction along the limit cycle attractor cannot contain diffusion caused by the Langevin noise during the day/night; hence the phase variance increases by σ2→σ2+ϵint2Tday during a day of length Tday ( and similarly at night ) . Dawn and dusk times reduce the phase variance σ2→σ2/s2 as the trajectories originating on , say , the day cycle converge on the night cycle ( see Figure 5c and Leypunskiy et al . , 2017; Monti and Lubensky , 2017 ) . In fact , we can compute this variance drop s2 entirely through geometric considerations . We define the circle map ϕ=P ( θ ) as relating originating points θ near dusk on the day cycle to final points on the night cycle ϕ after relaxation ( experimentally characterized in Leypunskiy et al . , 2017 ) . Then s-1=dP ( θ ) /dθ . Figure 5c shows that this slope s-1=dP ( θ ) /dθ , geometrically computed in the SI , agrees with the dawn/dusk variance drop in Langevin simulations and scales as s2-1∼L/R for large R/L . Thus , the population phase variance changes asσ2→Dayσ2+ϵint2Tday→Dusk ( σ2+ϵint2Tday ) /s2→Night ( σ2+ϵint2Tday ) /s2+ϵint2Tday→Dawn ( ( σ2+ϵint2Tday ) /s2+ϵint2Tday ) /s2 . Assuming T=Tday=Tnight and solving for steady-state phase variance ( σ2= ( ( σ2+ϵint2Tday ) /s2+ϵint2Tday ) /s2 ) , we obtain ( 2 ) σcycle2 , int∼ϵint2Ts2-1 Consequently , as the cycles become large ( large R/L ) , the dawn/dusk variance drop vanishes as s2-1∼L/R→0 while diffusion along the flat direction still adds ϵint2T to the variance during each day and each night; hence large-R/L limit cycles have large σcycle2 , int and thus low precision . ( Unlike with external noise , internal noise introduces a diffusion length scale and hence changing L and R are not equivalent . To make a fair comparison , we fix R and internal noise while changing L in Figure 5e; see Appendix 3 for more detail about other equivalent choices ) . Note that Equation 2 is invalid for strictly undriven clocks ( i . e . , s=1 ) ; such clocks show a variance that increases indefinitely with time . Here , we focus on driven clocks , which always settle to the finite variance given by Equation 2 . In contrast , for point attractors , the population variance stays constant during the day-night cycle and is shown to beσpoint2 , int∼ϵint2τrelaxin the SI , which matches Langevin simulations ( Figure 5d ) . Since τrelax∼Tday to have distinct clock states through the day ( Figure 3 ) , we find σcycle2 , int≥σpoint2 , int . In summary , in both cases , population variance is reduced by the geometric ‘curvature’ of the dynamics , that is , convergence of nearby trajectories . Point attractor trajectories experience a constant curvature of 1/τrelax . But limit cycle clocks experience such ‘curved’ off-attractor dynamics only at dawn and dusk , which is offset by dephasing ( Mihalcescu et al . , 2004; Gonze et al . , 2002 ) during long periods of zero curvature on the limit cycle ( day/night ) . Hence limit cycles underperform point attractors under high internal noise . We now subject the clock systems to both internal and external noise at the same time . We find results ( see Figure 6a ) that parallel those for explicit molecular models of biochemical oscillators studied in Figure 2 . Large-R/L limit cycles outperform other clocks in filtering out external noise when internal noise is low , but their precision degrades more rapidly than other clocks as internal noise ϵint2∼1/N is increased . Point attractors have poor precision with only external noise but do not significantly degrade with internal noise and outperform all other clocks at high internal noise . At comparable strengths of internal and external noise , limit cycles with an intermediate value of R/L are most precise . In the SI , we show that the optimal geometry is set by the ratio of internal and external noise strength , ( 3 ) ( L/R ) optimal=ϵintϵext . In the SI , also we show that , under certain simplifying assumptions , Equations 1 and 2 can be combined to give an explicit trade-off relationship , ( 4 ) σint2σext2∼Qwhere Q=ϵint2ϵext2 and where σint2 is the population angular variance of the clock state due to internal noise when driven by a noiseless external signal and σext2 is the population angular variance in the absence of internal noise due to amplitude fluctuations of the external signal . Note that angular variance is a better indicator than variance because we want to know how well the system can tell time . Equation 4 makes our trade-off explicit and also clarifies which parameters are varied and which parameters are held fixed in this trade-off . As long as Q is held fixed , we allow all other parameters to vary – for example , the overall strength of the external drive L , the size of the cycle R , and as discussed in the SI , all other parameters characterizing the normal form of limit cycles near a Hopf bifurcation . However , in holding Q fixed , our trade-off does assume that the strength of the external fluctuations ϵext – that is , the fractional size of amplitude fluctuations in the external signal – is held fixed . Similarly , we hold ϵint2 , the phase diffusion constant , fixed – that is , we are comparing clocks that would show the same population phase variance ( in units of radians ) over the same time in undriven conditions . See Appendix 3 for alternative comparisons and other details . Another measure of clock quality is the entrainment speed , that is , the time taken to reach steady state population variance , starting from a population uniformly distributed in clock phase . In Figure 6b , we see that with external noise only , the most precise clocks ( i . e . , small-L/R limit cycles ) are the slowest to entrain because they retain a longer history of the external signal , allowing them to average out external noise better . But strikingly , such a speed-precision trade-off is absent if internal noise is high . In Figure 6c , only internal noise is present and the external signal has no fluctuations . We see that clocks most robust to internal noise are also the fastest to entrain . Intuitively , the phase of slow entraining clocks is affected by the cumulative effect of internal fluctuations over a longer period of time . With both external and internal noise present , clocks with intermediate entraining speed - that is , intermediate ( L/R ) optimal=ϵint/ϵext - will have the highest precision .
Free running circadian clocks are a remarkable result of evolution in a changing but predictable environment and are thought to provide numerous benefits ( Troein et al . , 2009 ) . Here , we showed that the limit cycle attractor underlying such a clock is able to effectively project out weather-related amplitude changes that are perpendicular to the flat direction of the attractor . Similar roles for the flat direction of continuous attractors in projecting out external ( or input ) fluctuations have been explored in neuroscience ( Burak and Fiete , 2012 ) ; Seung ( 1996 ) , for example , for head and eye motor control and spatial navigation . However , we also found that the same flat direction becomes a vulnerability with internal fluctuations since such fluctuations cannot be restricted to be perpendicular to the attractor . We confirmed the trade-off between resistance to external and internal noise in diverse models of biochemical clocks and oscillators , using parameters inferred from experimental data by the original publications ( Leloup et al . , 1999; Schmal et al . , 2014; Locke et al . , 2005; Leloup and Goldbeter , 2003; Goldbeter , 1991; Goodwin , 1965; Gonze and Abou-Jaoudé , 2013; Kondepudi and Prigogine , 2014; Elowitz and Leibler , 2000; Buşe et al . , 2009 ) . The trade-off in each of these models can be given explanations that are specific to those models; for example , one can identify specific bottlenecks for external and internal noise in these models ( Cheong et al . , 2011 ) . However , we have provided an alternative kind of analysis based on the geometry of the dynamical systems involved . Such an explanation misses aspects specific to each clock - for example , how specific biologically tuneable parameters in each model affect internal and external noise resistance . However , the dynamical systems picture has the advantage in that it identifies the common origin of the trade-off across these systems . Such a dynamical systems picture has been fruitful in making other general but falsifiable predictions in biology ( Gan and O'Shea , 2017; Leypunskiy et al . , 2017; Corson and Siggia , 2017 ) , going back to Winfree’s phase singularity ( Winfree , 2001 ) . Our dynamical systems theory shows that the critical parameter for noise resistance is the strength of the external driving relative to the amplitude of free running oscillations , captured by the geometric ratio L/R in our analysis . Weak driving provides resistance to external noise while strong driving provides resistance to internal noise . While our dynamical systems theory involve planar circular limit cycles , the models in Figure 2 have complex non-planar non-circular limit cycles and yet reproduce our trade-off . Finally , while the internal noise discussed here is set by finite copy number , this dependence is not essential to the results here . Any source of disturbance ( e . g . , bursty transcription ) that perturbs the phase of the oscillator in constant light conditions is equivalent to internal noise . Similarly , external noise can involve any kind of fluctuation ( e . g . , multiplicative fluctuations , phase fluctuations ) of the external signal that does not result in a persistent phase shift of the external signal itself . Our work suggests that the damped oscillators are not merely poor cousins of the remarkable free running oscillators found for example , in S . elongatus . At the low protein copy numbers such as those found in P . marinus , damped clocks keep time more reliably than free running clocks . ( Low copy number has been linked to a trend towards reduced genome size and cell size in P . marinus [Bryant , 2003] . ) In addition to the noisy internal environment of P . marinus , the external environment might also play a role in selecting a damped clock; P . marinus is typically found in the open ocean , where the external environment may be more regular than the fresh water habitat of S . elongatus . In addition to P . marinus , damped oscillators are found elsewhere in biology , often in specific physiological conditions ( Saunders , 2002; Kidd et al . , 2015 ) . Understanding the benefits and drawbacks of such damped oscillators in different conditions is critical since such oscillations are easily overlooked experimentally , in comparison to free running oscillations . While numerous upstream and downstream considerations can modify ( Rand et al . , 2004; Domijan and Rand , 2011 ) the ultimate biological impact of fluctuations , we find that the core oscillator’s geometry in itself can continuously trade off protection against external fluctuations for protection against internal fluctuations in the diverse range of models studied here . Note added in proofs: The study of Monti et al . ( 2018 , in press ) independently arrived at the conclusion that free running clocks based on limit-cycles are more robust to external noise . Experiments in Chew et al . ( 2018 , in press ) suggest that the free running clock in S . elongatus is less robust to internal noise than the hourglass clock in P . marinus .
We incorporated most of our methods in Results and Discussion . For details beyond those presented in Results , please see Appendices . Code to simulate the systems can be found at https://github . com/WeerapatP/elife_tradeoff_clocks ( Pittayakanchit , 2018; copy archived at https://github . com/elifesciences-publications/elife_tradeoff_clocks ) . | The daily rising and setting of the sun is perhaps the most predictable pattern on Earth . Many organisms , from ancient bacteria to animals and plants , have evolved internal biological clocks to anticipate specific events such as dusk and dawn . However , biological clocks also need to continue working when faced with irregularities – both arising from within the organism and from external factors , such as a passing cloud that darkens the sky . Some organisms , including humans , have a so-called ‘free-running’ clock that generates a 24-hour rhythm , and keeps ticking even in the absence of any time triggers . Others , such as certain cyanobacteria , have an ‘hourglass’ clock that is not self-sustained – rather , these clocks show a simple response to the sunrise ( or sunset ) that would gradually perish without another sunset ( or sunrise ) . So far , it has been unclear why organisms have different kinds of clocks and if one type of clock is better suited for some conditions than others . Here , Pittayakanchit , Lu et al . analyzed and compared mathematical models of clocks in a variety of organisms , from cyanobacteria and fungi to plants and animals . The results revealed that internal and external irregularities put opposing pressures on biological clocks . Free-running clocks are more precise and more robust to external fluctuations , but more susceptible to internal ones . In contrast , hourglass clocks can remain accurate when internal irregularities are high but can be disturbed by external ones . Biological clocks affect the health of the entire organism and faulty clocks are implicated in numerous diseases . The study of Pittayakanchit , Lu et al . showed that the optimal architecture of a biological clock depends on the balance of irregularities in the external and internal environment of an organism . A next step will be to understand whether an organism can change its clock architecture while the environment changes . A better understanding of how biological clocks are regulated may help us find ways to tune faulty clocks to account for both the external environment and the internal state of an organism . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"physics",
"of",
"living",
"systems",
"computational",
"and",
"systems",
"biology"
] | 2018 | Biophysical clocks face a trade-off between internal and external noise resistance |
Heterogeneity and functional specialization among skin-resident macrophages are incompletely understood . In this study , we describe a novel subset of murine dermal perivascular macrophages that extend protrusions across the endothelial junctions in steady-state and capture blood-borne macromolecules . Unlike other skin-resident macrophages that are reconstituted by bone marrow-derived progenitors after a genotoxic insult , these cells are replenished by an extramedullary radio-resistant and UV-sensitive Bmi1+ progenitor . Furthermore , they possess a distinctive anti-inflammatory transcriptional profile , which cannot be polarized under inflammatory conditions , and are involved in repair and remodeling functions for which other skin-resident macrophages appear dispensable . Based on all their properties , we define these macrophages as Skin Transendothelial Radio-resistant Anti-inflammatory Macrophages ( STREAM ) and postulate that their preservation is important for skin homeostasis .
Macrophages are myeloid cells highly specialized in pathogen clearance and antigen capture in lymphoid tissues , where they participate in the first line of immune defense against exogenous threats , extending a bridge between innate and adaptive immunity ( Varol et al . , 2015 ) . Homeostasis is also maintained in a number of peripheral non-lymphoid tissues by the phagocytic , anti-microbial and tissue remodeling activities of specialized resident macrophages , including alveolar macrophages in lungs , interstitial histiocytes in connective tissue , osteoclasts in bone , microglia in brain , Kupffer cells in liver , peritoneal , intestinal and adipose tissue macrophages ( Murray and Wynn , 2011 ) . Likewise , macrophages contribute to the barrier function of the skin . Unlike migratory epidermal Langerhans cells and dermal dendritic cells , which are mobilized after topical antigen uptake to perform their tolerogenic or immunogenic role within the draining lymph nodes ( Henri et al . , 2010 ) , skin-resident macrophages remain in the dermis , contributing to pathogen clearance , tissue repair and the resolution of inflammation ( Pasparakis et al . , 2014 ) . A certain degree of heterogeneity among skin-resident macrophages has been previously reported relative to their ontogeny as well as to the functional specialization of specific subsets ( Abtin et al . , 2014; Hoeffel et al . , 2015; Schulz et al . , 2012; Tamoutounour et al . , 2013 ) . Herein , we define for the first time a clear dichotomy among the skin-resident macrophages based on their differential sensitivity to γ-irradiation . Radio-resistant and radio-sensitive skin macrophages are distinctively polarized already in steady state as revealed by transcriptional analysis . Consequently , these two macrophage subsets are functionally specialized in a cell-autonomous manner even though both subsets are exposed to mostly shared environmental cues . In particular , the radio-resistant macrophage subset comprises a novel type of perivascular macrophages that gain access to the vascular lumen , are highly phagocytic and possess anti-inflammatory properties even in the presence of pro-inflammatory stimuli . Moreover , they are renewed from a local Bmi1+ progenitor and become outcompeted over time by bone marrow-derived resident macrophages . Finally , the preferential depletion of these skin transendothelial radio-resistant anti-inflammatory macrophages ( STREAMs ) using diphtheria toxin-OVA nanoparticles evidences their involvement in tissue repair and remodeling , as well as the inability of their radio-sensitive counterparts to act as functional surrogates .
To visualize the microvascular network of the skin , we performed intravital imaging of the upper dermis of the mouse ear under steady-state conditions using a minimally invasive model based on confocal microscopy ( Auffray et al . , 2007 ) . The vasculature was highlighted by injecting i . v . a non-permeable vascular tracer , high molecular weight ( HMw; 2 MDa ) TRITC-dextran . Remarkably , a population of perivascular cells became readily visible during the first hours after HMw dextran injection , suggesting that these mostly sessile cells were constantly taking up intravascular dextran ( Figure 1A and Video 1 ) . This observation was further confirmed by monitoring the uptake of two differently labeled HMw dextrans injected with a lapse of 24 hr ( Figure 1—figure supplement 1A ) . Dextran administration followed by the subsequent injection of a non-phagocytosable tracer ( vivotag ) revealed the presence of dextran+ protrusions as non-stained regions in the vivotag channel projecting into the vessel lumen ( Figure 1B ) . The protrusions were flapping inside the vessels , presumably deflected by the vascular flow , and even contacting circulating cells ( Figure 1—figure supplement 1B and Videos 2–4 ) . The phenotypic characterization of the TRITC-dextran+ perivascular cells identified them mostly as macrophages ( i . e . , they were either CD45+ F4/80high CD11c- CD11b+ or CD45+ F4/80high CD11c- CD64+ cells by FACS [Figure 1C and Figure 1—figure supplement 1C] and CD68+ cells by whole-mount immunofluorescence staining [Figure 1—figure supplement 1D] ) . Flow cytometry analysis of skin single-cell suspensions from 12-week-old animals showed that the 96 . 67% ± 1 . 02% of all TRITC-dextran+ cells were CD45+ and the 90 . 41% ± 2 . 06% of CD45+ TRITC-dextran+ cells corresponded to macrophages ( CD64+ ) . In addition , TRITC-dextran+ macrophages represented 41 . 28% ± 5 . 66% of total skin macrophages . Next , we performed a mesoscopic analysis of murine ears to characterize the spatial distribution of skin macrophages . This analysis revealed their preferential localization around dermal blood microvessels rather than lymphatics ( Figure 2—figure supplement 1A and Video 5 ) . Interestingly , TRITC-dextran+ macrophages were mostly apposed to capillaries and venules , possibly because the dense smooth muscle layer wrapping arterioles prevented the access of macrophages to the arteriolar lumen ( data not shown ) . Moreover , endothelial cells did not detectably capture dextran under our experimental conditions , ruling out the possibility of the endothelium making accessible the intravascular dextran to the perivascular macrophages by transcytosis ( Figure 2—figure supplement 1B ) . 10 . 7554/eLife . 15251 . 003Figure 1 . Dermal perivascular macrophages capture intraluminal dextran . ( A ) ( Left ) Representative frames from Video 1 are shown . Briefly , image acquisition of ear mouse dermis started 30 min after i . v . injection of HMw TRITC-dextran and lasted for 5 hr . ( Right ) The surface plots show the fluorescence intensity of a representative cell at the starting time and 3 hr 15 min later ( white insets in corresponding video frames; a . u . , arbitrary units ) . ( B ) ( Upper left ) Three-dimensional ( 3D ) rendering of the fluorescence signals obtained in a C57BL/6 ear dermis after i . v . administration of HMw TRITC-dextran ( red , injected 16 hr before imaging ) and vivotag ( blue , injected at the time of imaging ) . ( Upper right ) The panels show a representative xy plane split into the different channels with an orthogonal section beneath , obtained along the yellow cross-section line . ( Lower ) The boxed region in the 3D rendering is shown tilted and at higher magnification . Yellow-dotted line marks endothelial perimeter in the red channel . Black asterisks mark void spaces corresponding to intravascular cells in the blue channel . ( C ) Phenotypic analysis of the dextran+ cells in the ear skin using the pan-leukocyte marker CD45 and the macrophage marker CD64 . Representative FACS dot plots of control and HMw TRITC-dextran-injected mice are shown ( n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 00310 . 7554/eLife . 15251 . 004Figure 1—figure supplement 1 . Spatio-temporal analysis of endothelial-protruding macrophages under steady-state conditions . ( A ) 'Dextran pulse-chase' assay . A C57BL/6 mouse was sequentially injected with red and green HMw dextran ( 24 hr interval ) and then sacrificed 48 hr after the beginning of the experiment . A maximal projection of the mouse ear dermis split in green and red channels as well as a zoomed merged image of the inset are shown . Some of the vessels in the inset have been depicted with white dotted lines for reference . ( B ) Orthogonal sectioning of the isosurface rendering from data shown in Figure 1B . The pink arrow highlights the interaction of a dextran+ cell ( red ) with an intravascular cell ( black hole not filled with the blue vascular tracer ) . ( C ) Representative FACS analysis describing the specific identification of skin macrophages by means of two equivalent staining strategies that only differ in the use of either CD11b or CD64 ( MΦ: macrophages , DC: dendritic cells ) . ( D ) Whole-mount staining of the ear of a C57BL/6 mouse injected i . v . with HMw TRITC-dextran ( red ) , using anti-CD68 ( macrophage marker , blue ) and anti-CD31 ( vascular marker , green ) . Split channels and the merged composite of the maximal projection image obtained from a z-stack of the dermis are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 00410 . 7554/eLife . 15251 . 005Video 1 . Observation of intravascular dextran uptake by dermal non-migratory perivascular cells . Video sequence illustrating the progressive phagocytosis of intravascular dextran by perivascular cells ( mostly sessile ) under homeostatic conditions . A C57BL/6 animal was injected i . v . with HMw TRITC-dextran and immobilized by means of long-term anesthesia . Intravital imaging started 30 min after injection . The frames in the video sequence are maximal projections of a confocal z-stack ( 207 planes , 129 . 66 μm in depth ) acquired every 5 min over 5 hr ( only shown first 3 hr 15 min ) . White asterisks highlight cells of particular interest . The video also shows the clearance of dextran from the bloodstream and the absence of paravascular permeability under homeostatic conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 00510 . 7554/eLife . 15251 . 006Video 2 . Dermal perivascular cells protrude into dermal microvessels and contact intravascular leukocytes . Video animation of a 3D rendering from a z-stack acquired in vivo in the dermis of a mouse ear . The mouse was injected i . v . with HMw TRITC-dextran 16 hr before microscopic observation and injected i . v . with vivotag ( as vascular tracer ) just before the experiment . Intravital imaging with confocal sectioning ( 1 z-stack , 65 planes , 40 . 28 μm in depth ) was performed . The image shows several perivascular dextran+ cells around a vessel highlighted in blue ( vivotag staining ) . Remarkably , two adjacent perivascular cells introduce their protrusions into the vascular lumen to contact intravascular leukocytes , seen as holes not filled with the vascular tracer . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 00610 . 7554/eLife . 15251 . 007Video 3 . Scanning movement of the protrusions of dermal perivascular cells . Frames represent maximal projections of a z-stack ( 9 sections acquired every 10 μm ) obtained repeatedly ( delay time 7 s 89 ms ) over a short time period ( 9 min 30 s ) . The video shows the scarce motility of dextran-capturing cells . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 00710 . 7554/eLife . 15251 . 008Video 4 . Magnified detail from Video 3 . Crop and zoom from Video 3 for detailed observation . The white asterisk marks a cell with an intravascular protrusion that flaps inside the vessel due to blood flow and the arrowhead points to the rearward movement of a dextran-loaded phagocytic vesicle from the vessel and along the protrusion towards the perivascular cell body . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 00810 . 7554/eLife . 15251 . 009Video 5 . Mesoscopic analysis of macrophage organization in relation to lymphatic and blood vessels . A whole mount-staining of a C57BL/6 ear dermis followed by tiled z-stack acquisition of the whole organ were performed . The video animation progresses from the mesoscopic analysis of the whole organ ( ear ) towards the detailed microscopic organization of lymphatics ( LYVE-1+ , red ) , vascular endothelium ( CD31+ , green ) and tissue resident macrophages ( CD68+ , white ) . Note the evident organization of macrophages around the complex endothelial network formed around hair follicles ( minimal units of the skin ) instead of around lymphatics . Blue corresponds to DAPI staining of nuclei , to facilitate observation of ear anatomy . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 009 We next studied how the macrophage protrusions gain access to the vascular lumen . We detected dextran+ macrophages embracing vessels from outside the pericyte sheath ( visualized as GFP+ cells around vessels using a nestin-eGFP reporter mouse ) , whereas their protrusions extended across the basement membrane ( stained with anti-collagen IV ) and between pericytes to reach the endothelial wall ( whole-mount immunofluorescence stainings in Figure 2A and Figure 2—figure supplement 1C ) . We also visualized dextran+ macrophage protrusions aligned with endothelial junctions in vivo by intravital microscopy by injecting i . v . an anti-CD31 antibody to stain the luminal surface of dermal microvasculature ( Figure 2—figure supplement 1D and Video 6 ) . Further analysis using orthogonal sectioning of whole-mount immunofluorescence stainings indicated that the macrophages insert their protrusions in the intravascular space through endothelial intercellular junctions ( Figure 2B ) . Interestingly , TRITC-dextran uptake was minimal in VE-cadherin-α-catenin knock-in mice ( Figure 2C ) . In these animals , endothelial adherens junctions are stabilized ( Schulte et al . , 2011 ) , which prevents paravascular permeability and leukocyte paracellular but not transcellular diapedesis . The ability of dermal macrophages to access the intraluminal space is restricted in these animals , supporting the observation that macrophages can reach the vascular lumen crossing the endothelial junctions and excluding the possibility that their protrusions reach the vessel lumen via a transcellular route through endothelial cells . Finally , since the capture of dextran by skin myeloid cells is mediated by the mannose receptor ( CD206 ) ( Wollenberg et al . , 2002 ) , we could selectively stain the intraluminal protrusions of these dermal perivascular macrophages in vivo by injecting i . v . anti-CD206 and quantified the amount of endothelial-protruding macrophages by FACS ( Figure 2D ) . Altogether these data indicate the existence of hitherto unidentified dermal perivascular macrophages able to protrude across endothelial junctions to reach the vascular lumen in homeostatic conditions . 10 . 7554/eLife . 15251 . 010Figure 2 . Dermal perivascular macrophages protrude across endothelial junctions into microvessels . ( A ) ( Left ) 3D reconstruction combining fluorescence signal and isosurface rendering of the ear dermis of a nestin-eGFP animal injected i . v . with HMw dextran . The inset is shown magnified in the central panel with a dextran+ macrophage protruding into the vascular lumen highlighted with a white arrow . Another protruding macrophage , marked with an asterisk in the 3D reconstruction image , is shown in a zoomed single confocal plane on the right . ( B ) ( Left ) Whole-mount staining of a C57BL/6 ear using anti-CD31 ( red ) and anti-CD68 ( green ) antibodies . ( Center ) The inset is shown as a 3D rendering with higher magnification . ( Right ) The same inset was tilted and further analyzed by orthogonal sectioning . The white cross-section lines are localized along the cell body and the intravascular protrusion of the macrophage on the right side . ( in , intraluminal; out , extravascular ) . ( C ) Comparison of dextran capture in wt vs . VE-cadherin-α-cat mice under steady-state conditions . Animals were injected i . v . with HMw TRITC-dextran and with HMw FITC-dextran 24 hr later , before imaging . ( Left ) Representative intravital images show the dermis of each phenotype , containing dextran+ cells ( red ) and vasculature ( green ) . ( Center ) The bar histogram represents the mean fluorescence intensity ( MFI ) of the TRITC-dextran signal ± SD obtained from five 20x fields of view of 2 animals of each phenotype ( a . u . , arbitrary units ) . ( Right ) Flow cytometry analysis of dextran capture by macrophages in wt and VE-cad-α-cat mice . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( **p-value < 0 . 005 , ***p-value < 0 . 001 ) . ( D ) In vivo staining of the intraluminal protrusions of endothelium-protruding macrophages . Mice were injected i . v . with either an antibody against CD206 ( mannose receptor , involved in dextran uptake ) or an isotype control antibody and sacrificed 3 min later . Then , animals were perfused with PBS and ears were processed for FACS analysis . Single-cell suspensions were stained with anti-CD45 , anti-F4/80 , anti-CD64 and a different anti-CD206 clone . Representative FACS dot plots are depicted and the bar histogram shows the specific detection of macrophage staining in vivo ( MΦ: macrophages ) . Data are mean values ± SD ( n = 4 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( ***p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 01010 . 7554/eLife . 15251 . 011Figure 2—figure supplement 1 . Spatial organization of skin macrophages relative to lymphatic and blood vessels . ( A ) Mesoscopic analysis of the spatial distribution of skin-resident macrophages relative to microvasculature and lymphatics . ( Left ) Whole-mount staining of a C57BL/6 mouse ear using anti-CD68 ( white ) , anti-CD31 ( green ) , anti-LYVE-1 ( red ) and DAPI . The maximal projection of the tiled z-stacks is shown . ( Right ) Split channels from the region in the inset are shown with higher magnification . The bar histogram represents the percentage of macrophages ( CD68+ cells ) proximal to either lymphatics or blood vessels obtained by 3D computational analysis of distances . ( B ) A single-cell suspension of ears from a C57BL/6 animal injected i . v . with HMw TRITC-dextran was stained with an specific marker for endothelium ( CD31 ) and analyzed by FACS . ( C ) Maximal projection of a z-stack of skin dermis from a nestin-eGFP mouse ear showing macrophages ( CD68+ , white ) on the outer surface of vessel walls , which are defined by basal lamina staining ( collagen IV , red ) , pericytes ( eGFP+ cells covered with collagen ) and endothelium ( CD31 , blue ) . The boxed area is shown as a magnified view with split channels . ( D ) Intravital imaging of the dermis of a chimeric animal ( B6 ACTB-eGFP ( donor ) /C57BL/6 ( host ) ) that was injected i . v . with HMw TRITC-dextran and anti-CD31 Alexa647 ( extracted from Video 6 ) . ( Left ) A single confocal plane is shown . The white circle and arrowheads highlight dextran+ macrophages whose protrusions are aligned with endothelial junctions . ( Right ) The red signal of these cells is displayed with zoom . Endothelial junctions are depicted with white dotted lines . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 01110 . 7554/eLife . 15251 . 012Video 6 . In vivo imaging of dextran+ macrophages aligned with endothelial junctions . The video sequence shows a confocal z-stack of the dermis of a mouse ear ( 65 planes , 40 . 28 μm in depth ) . The animal was chimeric ( reconstitution with B6 ACTB-eGFP BM cells ( green ) in a C57BL/6 host ) and was injected i . v . with HMw TRITC-dextran ( red ) 16 hr before imaging . For in vivo staining of endothelial junctions , an anti-CD31 antibody coupled to Alexa 647 ( blue ) was injected i . v . at the beginning of the experiment . The white circle and arrowheads highlight the area of interest , where the macrophage protrusions are posed in close contact with endothelial junctions . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 012 In an attempt to visualize endothelial-protruding macrophages without dextran labeling , we first analyzed whether these macrophages selectively expressed CX3CR1 , a marker associated to certain tissue-resident macrophage subsets including intestinal macrophages and microglia ( Varol et al . , 2015 ) . However , dextran+ macrophages did not express fluorescent reporter protein in Cx3cr1gfp/+ mice , neither in other myeloid reporter strains such as Lyz2Cre:Rosa26YFP and Langerin-eGFP ( Figure 3—figure supplement 1A ) . Then , we generated chimeric animals by reconstituting lethally irradiated C57BL/6 mice with β-actin-GFP+ hematopoietic progenitors , since it has been previously reported the absence of radio-resistant macrophages in the skin ( Bogunovic et al . , 2006 ) . Unexpectedly , dextran+ dermal macrophages did not express GFP in these chimeras ( Figure 3A and Video 7 ) , indicating that they were not replaced by BM-derived macrophages after irradiation and suggesting that they could be a specialized radio-resistant subset of skin-resident macrophages never characterized before . In fact , we monitored the presence of perivascular macrophages able to phagocytose intravascular HMw dextran 7d after lethal γ-irradiation without BM reconstitution ( Figure 3—figure supplement 1B ) . To further assess their radio-resistant nature , we reconstituted lethally irradiated CD45 . 2 host mice with CD45 . 1 BM progenitors . Reconstitution of the hematopoietic system in blood and lymphoid organs was complete after 12 weeks , with a negligible residual contribution of host cells to the myeloid compartment . In contrast , the chimerism rate in the skin myeloid subsets exhibited a higher contribution of the host populations , which could not be ascribed to blood reconstitution ( Figure 3—figure supplement 1C–D ) . The host radio-resistant CD45 . 2+ macrophages efficiently captured HMw dextran under steady-state conditions and were found mostly positioned around microvessels ( Figure 3B and Figure 3—figure supplement 1E ) . To validate this observation in a homeostatic context devoid of irradiation , the uptake of intravascular dextran was also analyzed in parabiotic mice maintained for 6 months with a shared circulation . The results obtained were similar to those of chimeric animals , inasmuch as only host macrophages captured intraluminal dextran ( Figure 3C ) . 10 . 7554/eLife . 15251 . 013Figure 3 . Endothelium-protruding macrophages are radio-resistant macrophages independent of BM supply . ( A ) ( Left ) Representative intravital image ( maximal projection of a z-stack ) , with zoomed view aside , of the ear dermis of a chimeric B6 ACTB-eGFP ( donor , green ) /C57BL/6 ( host ) animal pre-treated with HMw TRITC-dextran ( red ) and injected with vivotag ( blue ) at the time of imaging . The 3D colocalization analysis between green and red signals was estimated ( Pearson's coefficient in colocalized volume = 0 . 1074 ) . ( Right ) Representative FACS dot plots showing the analysis of the TRITC-dextran+ ( red ) , the GFP+ ( green ) and the double positive subsets in these chimeric animals as well as the macrophage content in the TRITC-dextran+ subset . ( B ) ( Left ) Averaged frequency of CD45 . 1+ ( donor ) and CD45 . 2+ ( host ) skin-resident macrophages in chimeric animals reconstituted for 3 months . ( Right ) Representative FACS analysis of the dextran+ subset in chimeric CD45 . 1 ( donor ) /CD45 . 2 ( host ) animals treated with HMw TRITC-dextran . The bar histogram shows the contribution of host and donor macrophages to the pool of dextran+ macrophages in the skin . Values are mean ± SD ( n=4 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( ****p<0 . 0001 ) . ( C ) Parabionts were injected i . v . with HMw TRITC-dextran and sacrificed 3d later . Ears were harvested and processed for FACS analysis . ( Left ) Averaged frequency of CD45 . 1+ and CD45 . 2+ skin-resident macrophages in each partner of the parabiotic pairs ( n = 4 ) after 6 months of parabiosis . ( Right ) Representative dot plots and bar histogram showing the contribution of host macrophages and macrophages derived from partner to the pool of dextran+ macrophages in the skin . Values are mean ± SD ( n = 4 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( ****p<0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 01310 . 7554/eLife . 15251 . 014Figure 3—figure supplement 1 . Analysis of tissue-resident macrophage subsets in reporter and chimeric animals . ( A ) Dextran+ macrophages do not express GFP in several myeloid reporter mouse lines . Cx3cr1gfp/+ , Lyz2-Cre:Rosa26YFP ( also known as LysM-Cre:Rosa26YFP ) and Langerin-eGFP animals were injected HMw TRITC-dextran i . v . and intravital imaging of ear dermis in steady-state was performed . Panels either show representative confocal planes , maximal projections or 3D reconstructions . White dashed lines or white dots depict vessel walls and white arrows highlight contacts of protruding macrophages and intravascular GFP+ cells . ( B ) Intravital imaging of ear dermis from a mouse injected with HMw TRITC-dextran i . v . 7d after lethal γ-irradiation . ( C ) Representative FACS analysis of the frequency of CD45 . 1+ ( donor ) and CD45 . 2+ ( host ) myeloid and T-lymphoid subsets in different organs ( blood , lymph nodes and spleen ) of a chimeric animal after 3 months of reconstitution . ( D ) Representative FACS analysis of skin donor and host myeloid subsets in the same chimeric animal . ( E ) Images represent the maximal projection of a z-stack acquired in a fixed ear dermis of a chimeric mouse , showing CD45 . 2 ( green ) , CD68 ( blue ) and CD31 ( red ) stainings . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 01410 . 7554/eLife . 15251 . 015Video 7 . Dextran+ macrophages are not from BM origin . Video animation showing the maximal projection of a confocal z-stack obtained from the tail skin of a chimeric animal ( B6 ACTB-eGFP ( donor ) /C57BL6 ( host ) ) injected i . v . with HMw TRITC-dextran . The animation alternatively shows the position of dextran+ macrophages ( red ) and BM-derived hematopoietic cells ( GFP+ , green ) relative to vessels ( CD31+ in white , whole-mount staining ) and finally highlights the lack of colocalization of red and green channels . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 015 Interestingly , a recent study described the existence of a subset of GFP+ perivascular macrophages in the DPE-GFP mouse strain , in which GFP is driven under the control of the distal and proximal CD4 enhancers and CD4 promoter ( details on the generation of this mouse line described in [Mempel et al . , 2006] ) . These GFP+ macrophages provide guidance cues for the recruitment of inflammatory cells such as neutrophils during intradermic bacterial infection ( Abtin et al . , 2014 ) . To analyze whether the dermal endothelial-protruding macrophages correspond to the GFP+ macrophages described above , we injected HMw TRITC-dextran in DPE-GFP animals . Remarkably , ~80% of cells that acquired circulating HMw dextran were GFP– in the skin of DPE-GFP mice ( Figure 4A ) . Furthermore , chimerism experiments in which we reconstituted lethally irradiated CD45 . 2 DPE-GFP hosts with CD45 . 1 hematopoietic progenitors revealed a dramatic decrease in the number of GFP+ dermal macrophages one month after the genotoxic insult , indicating their radio-sensitive nature ( Figure 4B ) . Altogether these results allow us to establish a clear dichotomy between DPE-GFP+ and dextran-capturing perivascular macrophages . 10 . 7554/eLife . 15251 . 016Figure 4 . The perivascular endothelium-protruding macrophages are distinct from the perivascular DPE-GFP+ macrophages . ( A ) ( Left ) Maximal projection of the ear dermis of a DPE-GFP animal injected i . v . with HMw TRITC-dextran , fixed and stained with anti-CD31 . The 3D colocalization analysis of the green ( GFP+ perivascular MΦ ) and red ( dextran+ perivascular MΦ ) signals was estimated ( Pearson's coefficient in colocalized volume = 0 . 1901 ) . ( Upper right ) FACS analysis of macrophages in ear skin from DPE-GFP animals injected i . v . with HMw TRITC-dextran . ( Lower right ) Bar histogram showing contribution of GFP- and + subsets to the pool of dextran+ macrophages . Values are mean ± SD ( n = 3 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( * p<0 . 05 ) . ( B ) DPE-GFP CD45 . 2 animals were lethally irradiated and reconstituted with wt congenic CD45 . 1 BM . Animals were sacrificed 1 month later and the content of skin GFP+ MΦ in their skin was analyzed by FACS and compared with untreated DPE-GFP CD45 . 2 animals . The bar histogram on the right shows the % of GFP+ macrophages out of total skin macrophages . Data are mean ± SD ( n = 4 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( * p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 016 In order to find the source for the skin-resident radio-resistant macrophages described in this work , we explored whether they are long-lived , endowed with self-renewal capacity , or derived from an extramedullary progenitor as alternative possibilities . We first tackled their sensitivity to UV irradiation , since other radio-resistant immune subset in the skin , the Langerhans cells ( LC ) , is UV-sensitive ( Merad et al . , 2002 ) . Chimeric CD45 . 1/CD45 . 2 ( donor/host ) animals were exposed to UVA-B irradiation ( 5 J/cm2 ) , and ear skin was analysed after one month . UV irradiation had a clearly negative effect on the survival and renewal of host compared with donor macrophages ( Figure 5A ) , indicating that radio-resistant macrophages are UV-sensitive . However , the radio-resistant macrophages did not express langerin and are clearly distinguishable from epidermal and dermal langerin+ dendritic cells ( Figure 3—figure supplement 1A ) . 10 . 7554/eLife . 15251 . 017Figure 5 . The skin radioresistant macrophages are an UV-sensitive population endowed with in situ renewal potential from an extramedullary progenitor . ( A ) Ears from chimeric animals ( CD45 . 1 ( donor ) /CD45 . 2 ( host ) ) were analyzed by flow cytometry 4 weeks after irradiation with UV-A/B or no irradiation . The left bar histogram shows the percentage of remaining macrophages of each haplotype after UV irradiation relative to untreated controls ( n = 3/group ) . The second bar histogram depicts the relative content of each subset out of the total pool of tissue-resident macrophages in treated and non-treated animals . Data are means ± SEM . Statistical significance was assessed by two-way ANOVA analysis with Sidak’s post-test ( left ) and one-sample t-test referred to value = 100 ( right ) ( n . s . not significant , **p-value < 0 . 01 ) . ( B ) Ki-67 staining reveals the lack of self-renewal capacity of radioresistant host macrophages in steady-state . Chimeric mice ( CD45 . 1-donor/CD45 . 2-host ) reconstituted for 3 months were sacrificed and skin processed and stained with anti-Ki-67 for FACS analysis . Tissue-resident macrophages from host ( CD45 . 2+ and CD64+ ) and donor ( CD45 . 1+ and CD64+ ) as well as Langerhans cells ( LC , CD45 . 2+ CD326+ and MHC-II+ , used as control subset endowed with self-renewal capacity ) were analyzed . Values are mean ± SEM ( n = 4 ) . Statistical significance was assessed by one-way ANOVA with Dunnett’s post-test ( *p-value < 0 . 05 , **p-value < 0 . 005 ) . ( C ) BrdU tracing revealed the slower turnover rate of host radioresistant macrophages . Chimeric mice ( CD45 . 1-donor/CD45 . 2-host ) were treated with BrdU in drinking water ad libitum . Animals were sacrificed 8 d later and skin processed for BrdU staining . Tissue-resident macrophages from host and donor as well as Langerhans cells were analyzed . Values are mean ± SD ( n = 4 ) . Statistical significance was assessed by one-way ANOVA with Tukey’s post-test ( **p-value < 0 . 005 , ****p-value < 0 . 0001 ) . ( D ) Fluctuations over time in the relative content of host and donor macrophages in the homeostatic skin of chimeric animals ( n = 5/group ) . Data are mean ± SD . Statistical significance was assessed by two-way ANOVA with Sidak’s post-test ( *p-value < 0 . 05 , **p-value < 0 . 01 , ***p-value < 0 . 005 , ****p-value < 0 . 0001 ) . ( E ) Ears from tamoxifen-treated Bmi1-IRES-Cre-ERT2 Rosa26 YFP reporter mice were analyzed by flow cytometry 5d after treatment . Representative FACS analysis of the identified YFP+CD45+ fraction is shown . ( F ) Whole-mount staining of an ear from a tamoxifen-treated Bmi1-IRES-Cre-ERT2 Rosa26 YFP mouse . The boxed area highlights a YFP+ perivascular cell protruding into a vessel , shown at high magnification in the accompanying panels . The white-dotted line marks the vasculature in the red channel . ( G ) Bmi1-IRES-Cre-ERT2 Rosa26 YFP CD45 . 2 host mice were chimerized with bone marrow from CD45 . 1 wt donors . Ears of these chimeric mice were analyzed by FACS and a subset of YFP+CD11b+F4/80high myeloid cells was detected . ( H ) FACS analysis of Bmi1-IRES-Cre-ERT2 Rosa26 YFP chimeric mice injected with HMw dextran . The bar histogram represents the percentage of YFP+ dextran+ macrophages respect to untreated control ( n = 4 ) . Data are mean ± SD . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( **p-value < 0 . 01 ) . ( I ) Representative intravital images of the ear of a chimeric mouse generated as in G , and injected with HMw TRITC-dextran . White arrows in the green channel mark YFP+dextran+ perivascular cells . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 01710 . 7554/eLife . 15251 . 018Figure 5—figure supplement 1 . Selective recovery of host macrophages in chimeric animals treated with clodronate liposomes intradermally . ( A ) Mouse hind footpads were injected intradermally with clodronate or PBS liposomes , and animals were sacrificed 24 hr later . ( Upper ) Representative dot plots show the partial depletion of the skin macrophage pool after clodronate treatment . ( Lower ) Bar histogram comparing the percentage of tissue-resident macrophages in PBS- and clodronate-treated skins . ( B ) Mouse hind footpads were intradermally injected either with clodronate or PBS liposomes and animals were sacrificed 24 hr or 7d after treatment ( Upper ) The bar histograms show the percentage of macrophages of each haplotype remaining after the indicated post-clodronate times relative to untreated controls ( n = 3/group ) . ( Lower ) The bar histogram shows the relative content of each type out of the total pool of tissue-resident macrophages in treated and non-treated ears . ( n = 4 ) . Statistical significance was assessed ( upper ) by unpaired two-tailed Student’s t-test ( n . s . not significant , p-value = 0 . 05 ) and ( lower ) by two-way ANOVA with Sidak’s post-test ( *p<0 . 05 , ****p<0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 01810 . 7554/eLife . 15251 . 019Figure 5—figure supplement 2 . Absence of stemness gene expression in skin resident macrophages in steady-state . ( A ) qPCR analysis of the expression of stemness-related genes in skin macrophages , dendritic cells and a skin CD34+ cell pool . Data are means of normalized values obtained from two independent samples . Bars represent mean ± S . E . M . in all cases . ( B ) qPCR analysis of Bmi1 gene expression in the different subsets of tissue-resident macrophages expressed as fold change using as housekeeping gene Gapdh . We used LSK cells from an inducible Bmi1-/- animal treated with tamoxifen as negative control and long-term hematopoietic stem cells from a wt animal as positive control . Data are means of normalized values obtained from two independent samples . Bars represent mean ± SD . ( C ) Assessment of driver induction in the skin of Bmi1-IRES-Cre-ERT2 Rosa26 YFP animals . Maximal projections of z-stacks of ear dermis and epidermis from a chimeric wt CD45 . 1/Bmi1 cre-IRES ERT2 Rosa26YFP CD45 . 2 animal ( donor/host ) treated with tamoxifen and injected with HMw TRITC-dextran . The white arrow points to YFP+ cells with a morphology compatible with keratinocytes in the epidermis . The white arrowhead indicates an YFP+ cell with dendritic morphology that corresponds to a γδ epidermal T cell ( FACS analysis not shown ) . The white asterisks ( green channel ) mark selected YFP+ dextran+ perivascular cells found in the dermis . ( D ) Bmi1-IRES-Cre-ERT2 Rosa26 YFP reporter mice were intradermally injected in the ears either with PBS or clodronate liposomes and simultaneously i . p . injected with tamoxifen . After 15d , ears were harvested and their content in YFP+ macrophages ( defined as CD64+ F4/80+ cells ) was analyzed by flow cytometry . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( *p-value < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 019 Next , we explored whether the radio-resistant macrophages proliferate in situ . We first treated chimeric animals intradermally with clodronate liposomes to deplete skin macrophages . Monitoring at 24 hr after injection showed depletion of both macrophage subsets , whereas after 7d we observed selective partial recovery of host macrophages , consistent with an in situ renewal of these cells ( Figure 5—figure supplement 1 ) . Then , we performed Ki-67 staining in the skin of host and donor macrophages from chimeric animals using LC as control cells endowed with self-renewal potential ( Chorro et al . , 2009 ) ( Figure 5B ) . The results point out to host macrophages as the only cell type assayed unable to proliferate by itself in steady-state . We complemented the study analysing the incorporation of BrdU administered ad libitum during 8d in these subsets ( Figure 5C ) . The results indicate that host radio-resistant macrophages have a slower turnover compared to BM-derived donor macrophages and LC . Altogether these findings suggest that skin radio-resistant macrophages are a long-lived differentiated population that originates from an in situ proliferating progenitor . Interestingly , the relative content of resident macrophages from donor and host changed dynamically over time in favor of the dominant BM-derived pool , which exhibits higher proliferative capacity ( Figure 5D ) . To identify the extramedullary source for dermal radio-resistant macrophages , we performed a fate mapping analysis in adult skin . First , we performed an analysis of the expression of a panel of stemness-related genes demonstrating that skin-resident macrophages possess no pluripotent potential per se in steady-state ( Figure 5—figure supplement 2A ) . Then , we employed a reporter mouse ( Bmi1-IRES-Cre-ERT2 Rosa26 YFP ) driven by the promoter of Bmi1 , gene involved in the maintenance of adult stemness ( Sangiorgi and Capecchi , 2008 ) , which enabled us to trace adult stem cells and their progeny in skin ( Figure 5—figure supplement 2B ) . Macrophages contain very low levels of the Bmi1 transcript ( Figure 5—figure supplement 2C ) and are devoid of Bmi1 protein in steady-state ( Sienerth et al . , 2011 ) . However , YFP+ CD45+ F4/80high CD11b+ CD11c- macrophages were detected in the homeostatic skin of Bmi1 reporter animals not earlier than 5d after i . p . injection of tamoxifen , suggesting that they are indeed derived from a Bmi1+ progenitor ( Figure 5E ) . Generation of YFP+ macrophages was increased during macrophage replenishment after depletion with s . c . clodronate treatment ( Figure 5—figure supplement 2D ) . Moreover , microscopy analysis of the dermis of these reporter animals detected perivascular YFP+ cells protruding into vessels ( Figure 5F ) . To exclude the contribution of BM-derived hematopoietic stem cells in this system , we generated chimeric CD45 . 1 C57BL/6 wt//Bmi1-IRES-Cre-ERT2 Rosa26 YFP CD45 . 2 ( donor/host ) mice . YFP+ dextran+ macrophages were detected in the skin of tamoxifen-treated HMw dextran-injected chimeric mice by FACS analysis ( Figure 5G–H ) . Intravital experiments confirmed the existence of YFP+ dextran+ macrophages under steady-state conditions in chimeras lacking tamoxifen-inducible hematopoietic cells of BM origin ( Figure 5I ) . Together these results confirm the in situ generation of radio-resistant macrophages from a skin pluripotent progenitor independent of BM . Our results strongly imply the coexistence of two distinct skin-resident macrophage pools in homeostasis based on their differential capacity to uptake circulating macromolecules , sensitivity to γ-irradiation and progenitor source . To further characterize these two populations , we performed a detailed phenotypic characterization of host radio-resistant and donor BM-derived macrophages in chimeric mice after 3 months of reconstitution . Both subsets were F4/80high , CD11b+ and expressed similar levels of other myeloid markers such as Ly6C and CD115 , as well as of the prototypic macrophage markers CD68 and CD64-MertK ( whose combined expression is associated with mature tissue macrophages [Gautier et al . , 2012] ) ( Figure 6A and Figure 6—figure supplement 1A ) . Moreover , both subsets were CD11c- and Ly6G- ( Figure 6—figure supplement 1A ) . There was also no consistent or sizeable difference in the expression of the angiogenesis-related receptor Tie-2 , the costimulatory molecules CD40 and CD70 or MHC-I . However , we found a higher mean expression of the costimulatory molecule CD86 and the mannose receptor ( CD206 ) in the host radio-resistant subset , whereas MHC-II was elevated in the donor BM-derived subset ( Figure 6A ) . Nevertheless , the broad range of expression of these differential markers within each macrophage subset precluded their use to dissect accurately the heterogeneity found among the skin-resident macrophages ( example provided for MHC-II in Figure 6—figure supplement 1B ) . 10 . 7554/eLife . 15251 . 020Figure 6 . Phenotypic and RNASeq analyses of donor and host skin-resident macrophages from chimeric animals . ( A ) Flow cytometry analysis of the surface expression of macrophage-specific markers ( CD64 , MerTK , CD68 ) , as well as CD86 , CD206 and MHC-II in donor and host macrophages from chimeric animals in steady-state . Values correspond to the normalized average GeoMean ± SD ( n = 3 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( n . s . not significant , *p-value < 0 . 05 , **p-value < 0 . 01 ) . ( B ) ( Left ) Compact view of the hierarchical clustering of the normalized expression profiles of the protein coding sequences with p-value <= 0 . 05 and consistent expression levels across replicates . ( Center and right ) Detailed view of the hierarchical clustering showing gene annotation . Arrows highlight relevant genes upregulated in host ( blue ) and donor ( red ) . ( C ) Vulcano plot representing the log2FC vs –log10 p-value transformation . Highlighted points correspond to genes with a p-value smaller than 0 . 05 ( blue for genes more expressed in host than donor and red for genes with the opposite expression profile ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 02010 . 7554/eLife . 15251 . 021Figure 6—figure supplement 1 . Extended phenotypic analysis of host and donor skin-resident macrophages . ( A ) Flow cytometry analysis of the surface expression of myeloid-specific markers ( CD11c , CD115 , Ly6C ) , as well as Tie-2 , CD40 , CD70 and MHC-I in donor and host macrophages from chimeric animals in steady-state . Values correspond to the normalized average GeoMean ± SD ( n = 3 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( n . s . not significant , **p-value < 0 . 01 ) . ( B ) MHC-II expression does not accurately discriminate radio-resistant host from BM-derived donor macrophages . Single-cell suspensions obtained by enzymatic digestion and gentle dissociation of the ear skin were gated to exclude doublets and dead cells ( DAPI staining ) . Then , macrophages were gated as CD45+ , F4/80high , CD11b+ , CD11c- and the expression of MHC-II was separately analyzed in the host ( CD45 . 2+ ) and donor ( CD45 . 1+ ) subsets . MHC-IIlow macrophages correspond to the P4 subset and MHC-IIhigh macrophages to the P5 subset in the classification reported by Tamoutounour and colleagues . Representative dot plots are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 02110 . 7554/eLife . 15251 . 022Figure 6—figure supplement 2 . GSEA of the RNASeq data . Mitochondrial function in skin macrophages . ( A ) Selected enrichment plots of Kegg pathways selectively increased in host or donor macrophages from the GSEA in Supplementary file 2 are shown . ( B ) Representative FACS histograms show the mitochondrial content and function in skin-resident macrophage subsets of chimeric mice . Briefly , ears were collected and both halves separated and incubated with MitoTracker Orange CMTMRos . Then , tissue was processed and single-cell suspensions analyzed by flow cytometry for mitochondrial content–Tomm20 staining- ( left ) and mitochondrial membrane potential ( function ) –mitotracker- ( right ) . Values are normalized average GeoMean ± SD ( n = 3 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( n . s . not significant , *p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 022 To find a specific marker for radio-resistant macrophages and gain further insight into the differences between the two tissue-resident macrophage subsets , we sorted donor and host macrophages from the skin of chimeric animals for high-resolution transcriptome profiling , using deep-sequencing technology ( RNA sequencing ( RNA-Seq ) ) ( Wang et al . , 2009 ) . Of 17 , 741 genes compiled , we found significant differences in 744 , 135 corresponding to protein coding sequences with similar normalized expression levels across replicates ( Figure 6B–C , and Supplementary file 1 ) . A comprehensive Gene Set Enrichment Analysis ( GSEA ) of all genes expressed in at least one cell type showed 36 Kegg pathways upregulated in host radio-resistant macrophages ( 14 of them enriched at a nominal p-value < 1% ) vs . 149 in donor macrophages ( 26 pathways enriched at a nominal p-value < 1% ) . The GSEA analysis confirmed striking differences in metabolism between the two resident macrophage populations in homeostatic skin ( Figure 6—figure supplement 2A and Supplementary file 2 ) . Host radio-resistant macrophages displayed a metabolic gene profile based on fatty acid oxidation and oxidative phosphorylation , whereas donor BM-derived macrophages showed a glycolytic gene profile . Such metabolic differences found at the gene expression level were confirmed by assessing mitochondrial activity ex vivo . Direct staining of the skin with an indicator of mitochondrial membrane potential ( MitoTracker Orange CMTMRos ) revealed elevated mitochondrial function in host radio-resistant macrophages , although both macrophage subsets contained comparable amounts of mitochondria per cell ( as indicated by similar levels of Tomm20 , a translocase of the outer mitochondrial membrane ) ( Figure 6—figure supplement 2B ) . Thus , the metabolic profiles of donor and host macrophages in homeostatic skin mirrored the metabolic reprogramming of pro-inflammatory classically activated and anti-inflammatory alternatively activated macrophages ( Galvan-Pena and O'Neill , 2014 ) , respectively . In addition to metabolic differences , we further validated the differential expression observed for several immune-related genes using a microfluidic-based multiplex qRT-PCR . Donor-derived macrophages showed increased expression of genes such as H2-Eb1 ( coding for MHC-II ) , Ccr2 , Ilβ , Il6 and Il10 . Importantly , the latest three genes code for cytokines related to the pro-inflammatory response in macrophages ( Rodriguez-Prados et al . , 2010 ) . Host macrophages showed upregulated expression of Cd86 , Il15 and Hmox among others ( Figure 7A ) . Accordingly , we detected elevated protein expression of heme oxigenase-1 ( encoded by Hmox ) and of other anti-inflammatory markers such as arginase I ( Figure 7B ) as well as higher production of TGF-β ( Figure 7C ) in host macrophages in resting conditions . 10 . 7554/eLife . 15251 . 023Figure 7 . Immune transcriptional profile and polarization potential of STREAMs . ( A ) Comparative qPCR analysis of immune response-related genes in donor and host skin macrophages from chimeric mice . Data represent average logFC values ± SEM ( n = 4 ) . ( B ) Immunostaining of anti-inflammatory markers arginase I and heme oxygenase-1 in cytospin samples of isolated macrophage subsets ( donor and host from chimeric animals ) . ( C ) ELISA quantification of TGF-β secretion by anti-inflammatory and pro-inflammatory macrophages at steady-state . Data are mean value ± SD ( n = 4 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( *p<0 . 05 ) . ( D ) Sorted pro-inflammatory and anti-inflammatory MΦ were left untreated or treated with 1 ng/ml LPS for 24 hr . Then , a multiplex flow cytometry-based analysis of culture supernatants was performed . Data are mean ± SD ( n = 3–5 ) . Statistical significance was assessed by two-way ANOVA with Bonferroni’s post-test ( n . s . not significant , *p-value < 0 . 05 , ** p-value < 0 . 01 , ***p-value < 0 . 005 ) . ( E ) ELISA quantification of IL-10 secretion by anti-inflammatory macrophages ( 50 . 000 cells/100 µl RPMI medium ) at steady-state and after IL-4 stimulation ( 20 ng/ml ) for 24 hr . Data are mean value ± SD ( n = 4 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( ***p-value < 0 . 005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 02310 . 7554/eLife . 15251 . 024Figure 7—figure supplement 1 . Gene profile of skin macrophages from non-treated and γ-irradiated animals . LPS receptor expression in skin macrophages . ( A ) Representative qPCR analysis of pro-inflammatory genes in samples from skin-resident MHClow and MHCIIhigh macrophage subsets in steady state or remaining skin-resident macrophages 7d after γ-irradiation . ( B ) Representative FACS dot plots showing the expression of the LPS receptors TLR4-MD2 complex and RP105 in skin macrophages from control C57BL/6 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 024 Importantly , the immune gene signatures of the dermal macrophages expressing low MHC-II and never exposed to γ-irradiation and of the remaining dermal macrophages in the skin 7d after lethal γ-irradiation showed lack of upregulation of proinflammatory genes ( Figure 7—figure supplement 1A ) , correlating with the observations made in radio-resistant host macrophages in chimeric animals . In contrast , dermal macrophages expressing high MHC-II and never exposed to γ-irradiation possess a pro-inflammatory profile as observed for donor BM-derived macrophages in chimeric animals ( Figure 7—figure supplement 1A ) . These results suggest that the prior lethal irradiation to generate BM chimeras did not alter the immune gene expression profiles studied herein . Hence , our results indicate that the radio-resistant subset of skin-resident macrophages is skewed towards an anti-inflammatory phenotype already in steady-state and predict a certain degree of functional specialization for them . We have coined this novel subset of macrophages as Skin Transendothelial Radio-resistant Anti-inflammatory Macrophages ( STREAM ) based on all the specific hallmarks found in our study . The divergent immune-related transcriptional profiles observed in the two distinct macrophage subsets coexisting in homeostatic skin prompted us to investigate whether they are committed to specific pro- or anti-inflammatory functions in a cell-autonomous manner and independently of polarizing stimuli . To examine this possibility , we isolated both macrophage subsets treated them ex vivo with LPS , a stimulus for which both subsets expressed the prerequisite receptors ( Figure 7—figure supplement 1B ) . Interestingly , the pro-inflammatory macrophages became activated and produced substantial levels of CCL4 , CCL5 , CXCL1 , TNF-α and IL-6 ( among other pro-inflammatory mediators examined ) in comparison to control conditions , whereas the anti-inflammatory subset did not produce significant amounts of these cytokines and chemokines in response to LPS ( Figure 7D ) . However , the anti-inflammatory macrophages produced IL-10 in response to IL-4 stimulation ( Figure 7E ) . These results suggest that the anti-inflammatory macrophages have a dearth of functional plasticity , being refractory to pro-inflammatory challenges but responsive to anti-inflammatory stimuli . They also emphasize the fact that the different subsets of skin-resident macrophages could be already committed even since steady-state to perform non-redundant functions . To explore the function of STREAMs , we searched for a method to deplete them selectively from the skin . For this purpose , we studied the capacity of STREAMs to capture nanoparticles coated with ovalbumin ( OVA ) , another ligand for the mannose receptor ( Burgdorf et al . , 2006 ) that is highly expressed on their surface . In this manner , we ensured the active capture of the particulate OVA ( OVA-coated fluorescent nanoparticles ( FITC fluospheres ) , Φ=0 . 2 μm ) from the bloodstream , preventing its passive diffusion into the perivascular tissue that could allow for extravascular phagocytosis . We first observed that OVA fluospheres injected i . v . were specifically captured by skin macrophages ( Figure 8A ) , and this process was abrogated after the depletion of skin macrophages by s . c . clodronate treatment ( Figure 8B ) . Intravital imaging followed by flow cytometry evaluation confirmed that the macrophages able to capture OVA fluospheres were dextran+ in C57BL/6 wt animals ( Figure 8C , Figure 8—figure supplement 1A and Videos 8–9 ) as well as radio-resistant in CD45 . 1/CD45 . 2 ( donor/host ) C57BL/6 chimeras ( Figure 8D–E ) . The availability of blood-borne nanoparticles in peripheral circulation is limited by the filtering action of mainly the liver and spleen , and therefore the amount of OVA-coated beads captured by skin STREAMs was low compared with that of splenic macrophages ( Figure 8—figure supplement 1B ) ; however , this amount was increased in γ-irradiated and splenectomized animals ( Figure 8—figure supplement 1A ) . 10 . 7554/eLife . 15251 . 025Figure 8 . STREAMs capture blood-borne OVA nanoparticles at steady-state . ( A ) Representative flow cytometry analysis of OVA-FITC fluosphere capture by skin macrophages in C57BL/6 mice . The peaks observed in the histogram correspond to the number of fluospheres uptake per cell . ( B ) Mice were intradermally injected with PBS ( control ) or clodronate liposomes in the hind footpads and were injected i . v . with OVA-adsorbed FITC fluospheres 72 hr later , followed by analysis after 4d . Representative dot plots show the amount of FITC+ cells for each treatment and histograms show the proportion of MΦ in the FITC+ fraction . ( C ) ( Left ) A C57BL/6 mouse was first injected i . v . with HMw TRITC-dextran and with OVA-FITC fluospheres 16 hr later . Imaging began 24 hr after the last injection . A representative frame of the experiment extracted from Video 8 with zoomed detail aside is shown . White dotted lines depict vessel walls . White asterisks mark FITC+ intravascular monocytes that have phagocytosed OVA fluospheres . White arrowheads point to intraluminal particles retained by dextran+ STREAMs and white arrows point to extravascular particles already phagocytosed by STREAMs . ( Center and right ) Split channels are shown with higher magnification as well as a 3D colocalization mask of both channels . ( D ) ( Left ) Absolute number of macrophages of each haplotype ( CD45 . 1 ( donor ) or CD45 . 2 ( host ) ) that have captured OVA-coated fluospheres in treated chimeric animals ( n = 4 ) . Data are means ± SEM . ( Right ) Representative per-cell uptake of fluospheres by macrophages of each haplotype . ( E ) Single-cell images of sorted skin macrophage subsets of a chimeric mouse ( CD45 . 1/CD45 . 2 , donor/host ) injected i . v . with OVA-FITC fluospheres . Images were obtained using an imaging flow cytometer . Notably , host macrophages were mostly filled with melanin as confirmed with Fontana-Masson staining ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 02510 . 7554/eLife . 15251 . 026Figure 8—figure supplement 1 . Comparison of OVA-FITC fluosphere uptake either by skin dextran+ macrophages in control , γ-irradiated and splenectomized animals or by spleen macrophages and skin macrophage subsets in chimeric mice . ( A ) Control , γ-irradiated and splenectomized mice were injected i . v . with HMw TRICT-dextran and , 16 hr later , with OVA-coated FITC-fluospheres . Then , animals were sacrificed 3d later and dextran+ cells were analyzed for their content in fluospheres . Representative FACS dot plots are shown . ( B ) Chimeric CD45 . 1 ( donor ) /CD45 . 2 ( host ) animals were injected i . v . with OVA-FITC fluospheres and sacrificed 3d later . FITC+ macrophage subsets from spleen and ear were plotted on the same histogram . The splenic macrophages displayed a wide range of phagocytic capacity ( from 1 fluosphere up to more than 10 per cell ) , whereas skin host macrophages could mostly phagocytose 1–2 fluospheres per cell . Skin donor macrophages were unable to phagocytose intravascular fluospheres . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 02610 . 7554/eLife . 15251 . 027Video 8 . In vivo uptake of blood-borne OVA-FITC fluospheres by STREAMs , example 1 . A C57BL/6 mouse was first injected i . v . with HMw TRITC-dextran and with OVA-FITC fluospheres 16 hr later . Imaging began 24 hr after the last injection . The frames in the video sequence correspond to a z-stack ( 4 sections acquired every 10 μm ) obtained every 3 . 300 s over a total period of 16 min . Interestingly , an intravascular STREAM’s protrusion is detected at the beginning of the video sequence ( white arrow ) . The video sequence shows free-flowing FITC-fluospheres as well as the intravascular cells that have phagocytosed them ( most probably monocytes ) . Some of the FITC+ cells transiently interact with endothelium at sites coinciding with STREAM’s intravascular protrusions . Intraluminal FITC-fluospheres retained in such intravascular protrusions can be also observed . Furthermore , FITC-fluospheres that have been previously phagocytosed by dextran+ STREAMs are detected in the extravascular tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 02710 . 7554/eLife . 15251 . 028Video 9 . In vivo uptake of blood-borne OVA-FITC fluospheres by STREAMs , example 2 . The i . v . injection of HMw TRITC-dextran was carried out 16 hr prior to the experiment and the i . v . injection of OVA-FITC fluospheres preceded the beginning of imaging acquisition . The frames from the video sequence are maximal projections of a z-stack ( 4 sections acquired every 10 μm , 30 μm in depth ) obtained every 2 . 27 sec over a total period of 11 min 30 s . The video shows a dextrandim STREAM that simultaneously phagocytoses several OVA-coated FITC fluospheres . The subsequent retrograde transport of phagosomes from the vessel towards the cell body is also shown . The white arrow in the merged image points to the place where fluospheres are captured from vasculature and the cell area is depicted in green and red channels for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 028 Based on the specific ability of STREAMs to capture OVA nanoparticles , we next set up a method to deplete them selectively by administering i . v . OVA-coated nanoparticles ( NP ) adsorbed with diphtheria toxin ( DT ) . In contrast to the i . v . administration of clodronate liposomes , which failed to eradicate skin STREAMs ( data not shown ) , this tailored experimental strategy depleted preferentially STREAMs preserving most of the other skin macrophages that had no access to the blood-borne toxin-coated NPs ( Figure 9—figure supplement 1 ) . Then , the potential anti-inflammatory function of STREAMs in tissue repair was interrogated using a model of wound healing in combination with DT-OVA-NP treatment . Animals were treated either with OVA-NP or DT-OVA-NP ( 2 . 5 μl/g per dose ) 1d before and 2d after wounding and were sacrificed at d5 , observing a clear decrease in the macrophage population ( Figure 9A , D ) . At this mid-stage of repair , macrophage depletion was associated with a significant delay in the healing process ( measured as % of area of initial wound ) ( Figure 9B ) and the histological analyses revealed a marked reduction in the formation of granulation tissue , concomitant reduced angiogenesis as well as delayed re-epithelialization ( Figure 9C–D ) . Next , we administered a sole dose ( 2 . 5 μl/g ) of DT-OVA-NP 2d after wounding and monitored wound closure till d9 . In this manner , the macrophage depletion was attained after the initial inflammatory phase of wound healing ( Gurtner et al . , 2008 ) . The Masson trichrome staining demonstrated a clear defect in collagen deposition in the animals treated with DT-OVA-NP ( Figure 10A ) . This defect correlated with the disordered distribution of myofibroblasts in these animals observed with the α-smooth muscle actin ( SMA ) staining ( Figure 10B ) . These results contrasted with the abundant collagen deposition and the parallel organization of myofibroblasts in OVA-NP-treated animals ( Figure 10A–B ) . This noticeable phenotype observed in macrophage function-related parameters indicates that STREAMs could exert a prominent role as the macrophages involved in orchestrating tissue remodeling and repair in skin , as anticipated by their anti-inflammatory gene and protein expression profile . 10 . 7554/eLife . 15251 . 029Figure 9 . Macrophage depletion using DT-OVA-NP hampered wound healing . ( A ) ( Left ) Animals were first injected i . v . with either OVA-NP or DT-OVA-NP ( 2 . 5 μl/g ) and , 1d later , the wound healing assay began . Two days after wounding , animals were treated with another similar dose of OVA-NP ( control ) or DT-OVA-NP . Then , mice were sacrificed 3d later ( mid-stage repair ) and their ears were collected to analyze skin macrophages . ( Right ) Bar histogram showing the percentage of MΦs out of total CD45+ cells in the ears of OVA-NP- or DT-OVA-NP-treated animals . Data are mean ± SEM ( n = 4–7 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( **p-value < 0 . 01 ) . ( B ) Analysis of mid-stage repair ( 5d after wounding ) . The plot illustrates the decrease of the wound area over time expressed as percentage of the initial wound . Data are mean SEM , n = 32 wounds/group . Statistical significance was assessed by two-way ANOVA analysis with Sidak’s post-test ( **p-value < 0 . 01 , ****p-value < 0 . 0001 ) ( C ) Histological analysis of wounds at mid-stage of repair from OVA-NP- and DT-OVA-NP-treated animals . Hematoxylin-eosin staining of representative samples is shown . ( Hp ep: hyperproliferative epithelium , ep tg: epithelial tongue , G: granulation tissue , d: dermis , p: panniculus carnosus , a: adipose tissue , ml: muscle layer ) . ( D ) ( Left ) Representative immunofluorescence staining of vessels ( CD31 staining ) and macrophages ( Mac-3 staining ) in tissue sections of wounds at mid-stage ( 5d ) from OVA-NP- or DT-OVA-NP-treated animals . ( Right ) Quantification of CD31 and Mac-3 stained area within the granulation tissue is shown . Data are mean ± SD ( n = 4–6 ) . Statistical significance was assessed by unpaired two-tailed Student’s t-test ( **p<0 . 01 , ***p<0 . 005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 02910 . 7554/eLife . 15251 . 030Figure 9—figure supplement 1 . Selective deletion of STREAMs after DT-OVA-NP treatment . Bar histogram showing the fluctuations of anti- and pro-inflammatory MΦ subsets 48 hr after DT-OVA-NP treatment compared to control . Data are means ± SD ( n = 3 ) . Statistical significance was assessed one-sample t-test referred to value = 100 ( *p-value < 0 . 05 , ****p-value < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 03010 . 7554/eLife . 15251 . 031Figure 10 . Macrophage depletion using DT-OVA-NP impaired collagen deposition and myofibroblast organization within the wound . ( A ) Wound healing assay until closure with administration of DT-OVA-NP or OVA-NP 2d after injury . After sacrifice , tissue was stained with hematoxylin-eosin ( upper right ) or analyzed for its content in collagen using Masson trichrome staining ( lower right ) followed by relative intensity quantification ( n = 6/group ) ( lower left ) . Statistical significance was assessed by an unpaired two-tailed Student’s t-test ( **p<0 . 01 ) . ( Ep: epidermis , d: dermis ) . ( B ) Analysis of the myofibroblast distribution in the wound area at the time of closure ( contraction phase ) in OVA-NP and DT-OVA-NP mice . Immunohistochemical staining of α-SMA was performed . ( Left ) View of the complete wound , ( right ) zoom in the granulation tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 03110 . 7554/eLife . 15251 . 032Figure 10—figure supplement 1 . Skin-resident macrophages different from STREAMs are not critically involved in tissue repair . ( A ) Analysis of CCR2 expression in skin host and donor macrophages from CD45 . 1 ( host ) /CD45 . 2 ( donor ) chimeric animals . Representative FACS dot plot is shown . ( B ) A Ccr2-/- mouse was injected i . v . with HMw dextran and , 16 hr later , with a vascular tracer ( vivotag ) . Intravital imaging revealed the presence of dextran+ STREAMs ( red ) around the vasculature ( white ) . A maximal projection of a z-stack obtained from the ear dermis is depicted . ( C ) Wound healing assay until wound closure performed in Ccr2+/+ and Ccr2-/- animals . ( Left ) The plots illustrate the decrease of the wound area over time expressed as percentage of the initial wound . Data are mean SEM , n = 28 wounds/group . Statistical significance was assessed by two-way ANOVA analysis with Sidak’s post-test ( differences were not found significant ) . ( Right ) Relative quantification of collagen deposition at the contraction phase ( n = 6 ) . Statistical significance was assessed by an unpaired two-tailed Student’s t-test ( n . s . not significant ) . ( D ) Immunohistochemical staining of α-SMA showing the distribution of myofibroblasts in the wound area at the time of closure in representative tissue sections of the two groups of animals . ( E ) Representative FACS dot plots showing the content of skin donor and host macrophages in the chimeric Ccr2+/+ ( donor ) / Ccr2+/+ ( host ) and Ccr2-/- ( donor ) / Ccr2+/+ ( host ) mice . ( F ) Wound healing assay until wound closure performed in chimeric Ccr2+/+/ Ccr2+/+ vs . Ccr2-/-/ Ccr2+/+ ( donor/host ) animals . ( Left ) The plots illustrate the decrease of the wound area over time expressed as percentage of the initial wound . Data are mean ± SEM , n = 28 wounds/group . Statistical significance was assessed by two-way ANOVA analysis with Sidak’s post-test ( differences were not found significant ) . ( Right ) Relative quantification of collagen deposition at the contraction phase ( n = 12/group ) . Statistical significance was assessed by an unpaired two-tailed Student’s t-test ( n . s . not significant ) . ( G ) Immunohistochemical staining of α-SMA showing the distribution of myofibroblasts in the wound area at the time of closure in representative tissue sections of the different chimeric animals . DOI: http://dx . doi . org/10 . 7554/eLife . 15251 . 032 Next , we performed wound healing experiments abrogating monocyte infiltration in the skin and depleting the radiosensitive subset of skin macrophages to further study the role of STREAMs in such scenario . We first analyzed the healing process in Ccr2-/- animals , wherein the repopulation of skin from blood monocytes is compromised ( Boring et al . , 1997; Serbina and Pamer , 2006 ) . Interestingly , STREAMs are CCR2- ( Figure 10—figure supplement 1A ) and are normally localized around dermal vessels in Ccr2-/- mice ( Figure 10—figure supplement 1B ) . Importantly , we did not observe substantial differences in wound closure , collagen deposition , and myofibroblast distribution in Ccr2-/- mice compared to Ccr2+/+ animals ( Figure 10—figure supplement 1C–D ) . Next , we benefited from the radio-resistant nature of STREAMs to deplete skin radio-sensitive macrophages by lethal γ-irradiation and prevented their replenishment by limiting the affluence of peripheral blood monocytes after reconstitution with Ccr2-/- hematopoietic progenitors ( Willenborg et al . , 2012 ) ( Figure 10—figure supplement 1E ) . We could not find differences in the healing capacity of the Ccr2-/-/Ccr2+/+ ( donor/host ) chimeras ( whose skin was defective in radio-sensitive macrophages , the repopulation from BM was prevented but the pool of STREAMs was intact ) compared to that of control Ccr2+/+/Ccr2+/+ chimeras , using all parameters assayed previously ( Figure 10—figure supplement 1F–G ) . Altogether these results strengthen the idea of the selective role for STREAMs in tissue repair .
In this study we have identified a unique subset of dermal perivascular macrophages , skin transendothelial radio-resistant anti-inflammatory macrophages ( STREAM ) , which capture blood-borne macromolecules by extending protrusions into the vascular lumen at steady-state . The observation of this phenomenon in chimeric animals allowed us to further characterize STREAMs as a radio-resistant subset of skin macrophages , whose turnover is ensured by an extramedullary source . We could also determine that STREAMs are committed to perform anti-inflammatory functions , lacking the plasticity to be reprogrammed in the presence of a potent pro-inflammatory stimulus . Finally , the selective depletion of STREAMs highlighted their function in tissue repair and remodeling processes to regain skin homeostasis . It is well established that specialized macrophage subtypes , such as subcapsular sinus macrophages in lymph nodes and Kupffer cells in liver , can traverse fenestrated endothelial sinusoids to capture lymph- or blood-borne particulate antigens and participate in the initiation of innate and adaptive immunity ( Huang et al . , 2013; Junt et al . , 2007; Lee et al . , 2010; Wong et al . , 2013 ) . Dendritic cells are also able to extend transepithelial projections to capture foreign particulate antigens from airways and gut mucosa , or traverse the vasculature in pancreatic islets of Langerhans ( Calderon et al . , 2008; Chang et al . , 2013; Farache et al . , 2013; Hammad et al . , 2009; Rescigno et al . , 2001 ) . However , some of these specialized dendritic cells have been redefined as macrophages ( Calderon et al . , 2014; Mazzini et al . , 2014 ) . CX3CR1+ myeloid cells in the central nervous system could also gain access to vasculature ( Barkauskas et al . , 2013 ) and mast cells were recently reported to capture blood-borne IgE , albeit the underlying mechanism remains unknown ( Cheng et al . , 2013 ) . Herein , we report that STREAMs are unique macrophages in that they traverse the endothelial junctions of non-fenestrated dermal vasculature to take up circulating macromolecules under homeostatic conditions . STREAM’s protrusions only emerge through the endothelial cell-cell junctions and do not use a transcellular route , as illustrated using a genetic approach ( the VE-cad-α-cat transgenic mouse line , which display impregnable , highly stable adherens junctions ) . Moreover , unlike subcapsular sinus macrophages in lymph nodes , which act like fly paper retaining lymph-borne viral particles on their surface for subsequent transfer to B cells ( Junt et al . , 2007 ) , STREAMs internalize particulate antigens , such as OVA-FITC Fluospheres , and might play a role in their cross-presentation to tissue-resident memory CD8+ T cells . The existence of radio-resistant dendritic cells in the skin has been previously reported ( Bogunovic et al . , 2006; Merad et al . , 2002 ) , whereas the presence of radio-resistant dermal macrophages has remained elusive hitherto . This study reveals that STREAMs are a radio-resistant subset that coexists in steady state with other skin macrophages of radio-sensitive nature . In this regard , the radio-resistant macrophages described in ( Haniffa et al . , 2009 ) might be the human homolog of this mouse subset , extending the generality and impact of our findings . The chimerism , parabiosis and fate mapping experiments also indicate that STREAMs are maintained locally in the skin throughout adult life independently of circulatory precursors , as opposed to radio-sensitive macrophages that are renewed at a low rate from blood in steady state and replaced by donor BM progenitors after a genotoxic insult . Thus , our data further extent the previous knowledge on the minimal blood exchange experienced by tissue-resident macrophages in other tissues ( Hashimoto et al . , 2013; Yona et al . , 2013 ) . The renewal potential and the lifespan of tissue-resident macrophages are still controversial . Somes studies have described the potential of macrophages to be self-perpetuated under certain conditions ( for example , in the presence of IL-4 during a parasitic infection [Jenkins et al . , 2011] or in a M-CSF- or GM-CSF-dependent manner after macrophage depletion [Hashimoto et al . , 2013] ) . STREAMs are unable to self-renew in steady-state and their turnover from a Bmi1+ adult progenitor occurs at a low basal rate , becoming outcompeted over time by other skin-resident macrophages with faster turnover . Hence , our results concur with previous studies reporting a dual origin for skin macrophages with a prenatally-derived CCR2-independent subset and a postnatal BM-derived CCR2-dependent subset , being the prenatally-derived subset displaced by the BM-derived subset with age ( Jakubzick et al . , 2013; Tamoutounour et al . , 2013 ) . Interestingly , this dual origin has been also observed in other tissues such as heart ( Molawi et al . , 2014 ) . Unraveling the ontogenetic source for tissue-resident macrophages is another complex matter . There are several landmark studies describing the existence of different macrophage precursors during development . The yolk sac-derived Myb-independent F4/80bright macrophages seeded in the embryonic skin prior to birth and renewed independently of definitive hematopoiesis ( Schulz et al . , 2012 ) might correspond to the developmental precursor for STREAMs and the c-Myb dependent fetal liver monocytes that give rise to macrophages with self-renewal capacity ( Hoeffel et al . , 2015 ) might be the precursors of the radiosensitive macrophages ( those susceptible of replenishment by BM precursors after a genotoxic insult ) . However , the elucidation of skin macrophage ontogeny deserves further thorough investigation in future studies . Regarding phenotypic characterization , a classification of the skin mononuclear phagocyte system based on phenotypic markers has recently identified heterogeneity among the skin-resident macrophages , although not related to radio-sensitivity . These authors distinguish two distinct skin-resident macrophage subsets defined both as CD11b+ CD11c- CD64+ MerTK+ Ly6Clow CCR2- but differing on the expression of MHC-II ( Tamoutounour et al . , 2013 ) . They suggest that both subsets possess analogous transcriptional profiles , share functions , are radio-sensitive and equally dependent on BM supply , and could represent sequential stages of differentiation ( the MHC-IIlow subset ( P4 ) maturing into MHC-IIhigh ( P5 ) macrophages ) . As demonstrated in Figure 6A and Figure 6—figure supplement 1B , the expression of MHC-II in skin macrophages cannot precisely discriminate the two independent subsets of macrophages with distinct radio-sensitivity , different polarization profiles and non-overlapping functions described in our work . Therefore , the classification criteria used in this work allows better resolution of the skin macrophage subsets , compared to classifications relying on subset-defining markers that might not adequately reflect the existing heterogeneity within the skin niche . Metabolic adaptation is a key feature in macrophage activation , instrumental for their function in homeostasis , immunity , and inflammation ( Martinez et al . , 2013 ) . In this regard , the GSEA of our RNASeq data highlight clear differences in the metabolic gene profiles of host radio-resistant and BM-derived macrophages . We found an enrichment in genes involved in glycerolipid , glycerophospholipid and linoleic acid metabolism , as well as in glycosphingolipid biosynthesis in the gene profile of radio-resistant macrophages . These results are compatible with the energetic requirements and the regulation of membrane fluidity needed for phagocytosis and are also crucial for tissue remodeling ( Biswas and Mantovani , 2012 ) . Other genes enriched in this subset are related to fibrogenesis and tissue repair such as Tgfb1 and genes of the Hedgehog signaling pathway , as well as genes related to taurine and hypotaurine metabolism , being consistent this with an antioxidant protective role . Conversely , the transcriptional profile of the BM-derived macrophages includes an over-representation of genes associated with the carbohydrate and retinol metabolisms such as the pentose phosphate pathway , with the ABC transporters and the JAK-STAT and WNT signaling pathways , in accordance with their pro-inflammatory potential . Moreover , a gene encoding a regulator of the metabolic switch from oxidative phosphorylation to aerobic glycolysis ( Trap1 ) ( Yoshida et al . , 2013 ) is specifically expressed in BM-derived macrophages . Our study in the homeostatic skin along with others describing the existence of polarized macrophages already in steady state in heart , testis and pancreas ( Calderon et al . , 2015; DeFalco et al . , 2014; Pinto et al . , 2012 ) help to reformulate the long-held paradigm of macrophage polarization , which defined the functional specialization of macrophages from a steady-state non polarized status mainly as a consequence of integrating external stimuli ( Gordon and Pluddemann , 2013; Sica and Mantovani , 2012 ) . Interestingly , the study by Calderon and colleagues ( Calderon et al . , 2015 ) highlights the imprinting of distinct features in macrophages by their anatomical localization within the steady-state pancreas . The macrophages in the islets possess a pro-inflammatory M1 phenotype and protrude into the fenestrated vessels , whereas the stromal milieu sets the response of the macrophages in favor of the M2 phenotype . This pattern of regional specialization might not be applicable to the skin where , e . g . , perivascular macrophages with pro-inflammatory ( Abtin et al . , 2014 ) and anti-inflammatory ( STREAMs ) phenotypes coexist in the same areas . However , although both subsets mostly share environmental stimuli , STREAMs can get access to the intravascular milieu and might receive selective signals that could contribute to maintain their phenotype . Therefore , the essential role of the environment in controlling tissue-specific macrophage identities cannot be dismissed as highlighted in recent epigenetics studies ( Gosselin et al . , 2014; Lavin et al . , 2014 ) . The preferential depletion of anti-inflammatory STREAMs has a deleterious impact in the wounded skin , indicating that they could participate in orchestrating the pro-reparative functions traditionally associated to the so-called alternatively activated ( M2 ) macrophages ( Das et al . , 2015 ) . Conversely , we observed that the rate of wound closure was neither affected in the Ccr2-/- animals ( as previously reported by Willenborg et al . , 2012 ) nor in chimeric Ccr2-/-/Ccr2+/+ mice , indicating that infiltrating monocytes and radiosensitive tissue-resident macrophages could be mostly dispensable for the reparative phase of the healing process in the skin . Thus , the coexistence of macrophage subsets performing non-redundant roles in homeostatic skin highlights the need of maintaining equilibrium between both pools to preserve the skin fitness . The lack of functional redundancy could be of clinical significance , and may contribute to better understand concomitant skin pathologies arising from radio-induced bone marrow ablation therapies and certain inflammatory skin disorders , particularly those aggravated with age .
Male either littermates or age-matched ( 8-to-12-week-old ) mice on the C57BL/6J background were used for intravital imaging , phenotypic characterization , transcriptional analysis and functional experiments . Chimeric animals were generated as follows: 8-to-12-week-old C57BL/6 wt male mice ( CD45 . 2 haplotype ) , were lethally γ-irradiated with 2 doses of 6 . 5 Gy and transplanted with a mixture of 5 x 106 BM cells from either B6 SJL mice ( CD45 . 1 haplotype ) , or B6 ACTB-eGFP mice . Alternatively , irradiated B6 SJL mice ( CD45 . 1+ ) were reconstituted with BM from Ccr2+/+ or Ccr2-/- ( CD45 . 2+ ) mice . Animals were used for experimental procedures after 12 to 18 weeks of reconstitution . Details of the generation and characterization of the gene-targeted mouse expressing VE-cadherin-α-catenin protein can be found in ( Schulte et al . , 2011 ) . Inducible Bmi1-IRES-Cre-ERT2 Rosa26 YFP mice were generated by crossing the Bmi1CreER/+ strain ( Sangiorgi and Capecchi , 2008 ) with Rosa26YFP/+ reporter mice . Bmi1-IRES-Cre-ERT2 Rosa26 YFP double heterozygous mice were injected with tamoxifen 5d prior to analysis . Tamoxifen ( Sigma-Aldrich [St Louis , MO] ) was dissolved in corn oil ( Sigma-Aldrich ) to a final concentration of 20 mg/ml , and mice were i . p . injected every 24 hr on three consecutive days with tamoxifen at a concentration of 9 mg per 40 g body weight . The nestin-eGFP reporter strain was a kind gift from Dr . G . Enikolopov ( Mignone et al . , 2004 ) . DPE-GFP mice were generated as described in ( Mempel et al . , 2006 ) . Langerin-eGFP , Cx3cr1-GFP , Lyz2-Cre:Rosa26YFP and Ccr2-/- mice were kindly provided by Dr . B . Malissen , Dr A . Hidalgo , Dr . M . Ricote and Dr . C . Ardavin , respectively . Mice were kept in pathogen-free conditions in the CNIC Animal Unit , Madrid . Animal studies were approved by the local ethics committee and by the Division of Animal Protection of Comunidad de Madrid ( approved protocols PROEX 159/15 and 160/15 ) . All animal procedures conformed to EU Directive 2010/63EU and Recommendation 2007/526/EC regarding the protection of animals used for experimental and other scientific purposes , enforced in Spanish law under Real Decreto 1201/2005 . Ears were taped to the center of a coverslip and attached with high vacuum grease . Hairless areas were examined using a HCX PL APO lambda blue 20 . 0X 0 . 70 IMM UV objective ( multi-immersion , glycerol ) coupled to an inverted microscope ( DMI6000; Leica Microsystems GmbH [Wetzlar , Germany] ) equipped with a confocal laser-scanning unit ( TCS-SP5; Leica ) . Non-invasive intravital imaging procedures were carried out in a thermostatic chamber at 37ºC . For short-term studies ( 1–2 hr ) , animals were first anesthetized by i . p . injection of ketamine ( 50 mg/kg ) , xylazine ( 10 mg/kg ) , and acepromazine ( 1 . 7 mg/kg ) , and repeated half-doses were administered when needed over the course of the experiments . For long-term studies ( time-lapse imaging > 2 hr ) , animals were first anesthetized by i . p . injection of urethane ( 1 . 2 gr/kg ) dissolved in PBS , followed by s . c . injection of 1/10 of the initial dose dissolved in 200 μl PBS for rehydration after several hours . As vascular tracers , 100 μl of 2 MDa FITC- or TRITC-dextran ( 1% w/v , which corresponds to a concentration of 5 μM; anionic lysine fixable [Molecular Probes , Thermo Fisher Scientific , Waltham , MA] ) , or 25 μg Vivotag ( VisEn Medical , PerkinElmer [Waltham , MA] ) dissolved in PBS were i . v injected . For in vivo staining of the vasculature , an anti-CD31 antibody ( Fitzgerald Industries [North Acton , MA] ) was labeled in house with an Alexa Fluor 647 monoclonal antibody labeling kit ( Molecular Probes ) , and 25 μg were injected i . v . 16 hr after dextran administration . Images were obtained using bidirectional scanning mode and the acquisition of the different channels was sequential to avoid fluorescence bleed-through artifacts . Z-stack images were acquired with optimal confocal sectioning ( spaced 0 . 63 μm along the z-axis ) . For time-lapse acquisition , z-sections were obtained at 10 μm intervals up to a depth of 70–100 μm , using a pinhole aperture > airy 1 ( confocal acquisition ) . The acquisition rate was ~2–3 sec/z-stack . Live mice anesthetized with a mixture of zoletil and dontor were surgically opened and perfused with 1% paraformaldehyde ( PFA ) in PBS at a continuous infusion rate of 7 ml/min , using a programmable syringe ( Harvard Aparatus ) . Different parts of the skin were collected and post-fixed in 1% PFA for 1 hr at room temperature ( RT ) . Tissues were stained as described in ( Baluk et al . , 2007 ) . Briefly , tissues were incubated overnight ( O/N ) at RT with PBS containing 0 . 3% Triton X-100 , 5% goat serum ( Jackson ImmunoResearch Europe Ltd . [Suffolk , UK] ) and primary antibodies . The anti-murine primary antibodies used were hamster anti-CD31 clone 2H8 ( Chemicon , EMD Millipore [Billerica , MA] ) , rat anti-CD68 clone FA-11 and polyclonal rabbit anti-collagen IV ( Abd Serotec , Bio-rad [Hercules , CA] ) , polyclonal rabbit anti-CD31 and polyclonal chicken anti-GFP ( Abcam [Cambridge , UK] ) , polyclonal rabbit anti-LYVE-1 ( ReliaTech [Wolfenbüttel , Germany] ) , and biotin-conjugated CD45 . 2 clone 104 ( BD Pharmingen , BD Biosciences [San Diego , CA] ) . Samples were then thoroughly washed with 0 . 3% Triton X-100 in PBS and stained with appropriate secondary antibodies O/N at RT . Secondary antibodies used were DyLight 405 donkey anti-rat , Cy3 donkey anti-chicken and Alexa Fluor 647 or 488 goat anti-armenian hamster ( Jackson Immunoresearch ) , Rhodamine-X goat anti-rabbit , Alexa 647 goat anti-rabbit , Alexa 647 chicken-anti-rat , Alexa 488 donkey-anti-rat and Alexa 647 or 488 goat-anti-mouse ( Molecular Probes ) . After the last round of washing , labeled samples were fixed in 1% PFA for 30 min and mounted for microscopy analysis in Prolong Gold antifade reagent ( Molecular Probes ) . Confocal z-stacks up to a depth of 100 μm ( including epidermis and upper dermis ) were obtained using a LSM 700 laser scanning microscope equipped with a LD LCI Plan/Apochromat 25x/0 . 8 Imm Korr DIC M27 objective ( Carl Zeiss AG [Oberkochen , Germany] ) . Confocal z-stack images from fixed samples or intravital experiments were processed to obtain maximal projections , orthogonal sectionings , 3D reconstructions or isosurface renderings , fluorescence intensity profiles , colocalization analyses and 3D computational analyses of distances using Imaris 7 . 3 . 1 ( Bitplane [Belfast , UK] ) , Volocity 5 . 5 . 1 ( PerkinElmer ) , and ImageJ 1 . 42q ( NIH , [Bethesda , MD] ) . Images from in vivo time-lapse acquisitions were analyzed and videosequences set up and edited either with Imaris or Volocity softwares . Animals were injected i . v . with 100 μl 1% HMw TRITC-dextran and , after a lapse of 24 hr , with 100 μl 1% HMw FITC-dextran . Animals were sacrificed 24 hr after the last injection and ears were processed for imaging . Mice were injected retro-orbitally with 10 μg of Alexa 488 anti-CD206 ( clone C068C2 , BioLegend [San Diego , CA] ) or Alexa 488 control isotype antibody ( rat IgG2b , κ ) and sacrificed 3 min later . Then , animals were perfused with PBS and ear skin was processed for flow cytometry . Counter-staining of CD206 was performed with Alexa 647 anti-CD206 clone MR5D3 ( BioLegend ) . Skin samples ( ears and hind footpads ) were digested either with 2 mg/ml crude collagenase type IA ( Gibco , Thermo Fisher Scientific [Waltham , MA] ) in PBS for 2 hr or with 0 . 25 μg/ml Liberase TM Research Grade ( Roche Holding AG [Basel , Switzerland] ) in RPMI medium for 1 hr at 37°C . Samples were then mechanically disrupted , washed and filtered . Single-cell suspensions were incubated with anti-mouse FcRII/III ( clone 2 . 4G2 ) for 10 min at 4°C in PBS containing 0 . 05% BSA and 0 . 05 mM EDTA , and then stained with the following antibodies: Pecy7 anti-F4/80 clone BM8 , APC or APC-eFluor 780 anti-CD45 . 2 clone 104 , PE anti-CD115 clone AFS98 , PE anti-Tie-2 clone Tek4 , and APC-eFluor 780 Streptavidin ( eBioscience [San Diego , CA] ) ; v450 Horizon anti-CD45 Clone 30-F11 , v450 Horizon or PerCP-Cy5 . 5 anti-CD45 . 1 clone A . 20 , PE anti-CD11c clone HL3 , FITC or APC or APCCy7 anti-MHC-II clone M5/114 . 15 . 2 , APC anti-MHC-I clone AF6-88 . 5 , PE anti-Ly6C clone HK1 . 4 , PECy7 or Alexa Fluor 647 anti-CD64 clone X54-5/7 . 1 , PE anti-CD70 clone FR70 , PE anti-CD86 clone GL-1 , FITC anti-CD40 clone 3/23 , and FITC anti-CD11b clone M1/70 ( BD Pharmingen ) ; APC anti-CCR2 and biotinylated MertK ( R&D Systems Inc . [Minneapolis , MN] ) ; Streptavidin Alexa Fluor 350 ( Molecular Probes ) ; APC or PerCP-Cy5 . 5 anti-CD68 clone FA-11 , and Alexa Fluor 647 CD326 clone G8 . 8 ( BioLegend ) ; Alexa Fluor 647 anti-CD206 ( AbD serotec ) ; and rabbit polyclonal anti-Tomm20 clone FL-145 ( Santa Cruz Biotechnology [Dallas , Tx] ) . For intracellular staining , samples were first fixed and permeabilized with Fix & Perm before antibody incubations in Wash & Perm buffer ( BD Biosciences [San Diego , CA] ) . For Ki-67 staining , samples were treated following manufacturer’s protocol ( Alexa Fluor 647 anti-Ki-67 , BD Pharmingen ) . Single-cell suspensions were analyzed with a FACSCanto II HTS cytometer or a FACSAria II SORP sorter ( BD Biosciences ) , and data were processed with FlowJo 7 . 6 . 3 and BD FACSDiva v . 6 . 1 . 3 . softwares . In some cases , sorted samples were further analyzed by combining flow cytometry and imaging with Flowsight ( Amnis , EMD Millipore [Billerica , MA] ) . For parabiosis , gender- and weight-matched mice were anesthetized i . p . with a mixture of ketamine ( 100 mg/kg ) , xylazine ( 10 mg/kg ) , and acepromazine ( 3 mg/kg ) . Hair was removed from the lateral aspects of the mice by shaving and hair removal cream . A longitudinal skin incision was made from the olecranon to the knee joint on opposing sides of each mouse of the parabiotic pair . Animals were placed side-by-side and the right olecranon of one animal was attached to the left olecranon of the other by a double suture . The equivalent procedure was performed for the knee joints to further secure the parabiotic pair . The dorsal and ventral skins were then approximated continuous 6–0 suture and by staples along the lateral abdominal aspect of the mice . Parabiosis was maintained for 6 months and , then , one of the partners in the parabiotic pair was injected retro-orbitally with 200 μl of HMw TRITC-dextran ( 5 μM ) and animals were sacrificed 3d later . Chimerism in skin-resident macrophages was determined as well as the capture of intraluminal dextran by host and partner macrophages . Animals were exposed to UV irradiation using a 1000 watt xenon arc solar simulator ( Oriel by Newport [Irvine , CA] ) equipped with an Oriel 81 , 017 filter ( Colipa [Oudergem , Belgium] ) . UVB and UVA irradiance measurements were performed using an IL-1700 radiometer ( International Light Technologies Inc . [Peabody , MA] ) equipped with SED240/UVB-1/TD and SED033/UVA/TD photodetectors . The radiometer was calibrated with a Solar-Scope spectroradiometer ( Solatell [Croydon , UK] ) . Animals were irradiated with 5 J/cm2 UVA-UVB except for their left ear that was covered to be used as non-treated control . Then , they were allowed to recover for 1 month before sacrifice . Clodronate ( Roche ) or PBS liposomes were injected s . c . ( 20 μl/hind footpad ) . Skin samples were processed for flow cytometry analysis after 24 hr or 7d . Mice were allowed to drink BrdU-treated water ( 0 . 8 mg/ml ) for 8d and then were sacrificed . BrdU incorporation in skin cell subsets was analyzed by flow cytometry ( FITC BrdU Flow Kit , BD Pharmingen ) . Ex vivo staining of mitochondria activity was conducted by flow cytometry following a previous protocol ( Johnson and Rabinovitch , 2012 ) . Briefly , ears were excised and split in half . The tissue was immediately incubated in the staining solution ( 20 nM of MitoTracker Orange CMTMRos [Molecular Probes] diluted in HBSS1X with 5% FBS ) for 20 min at 37°C , 5% CO2 . Then tissues were digested and dye was maintained at 10 nM in all the following steps before FACS analysis . cDNA was synthesized and amplified directly from cells using the Smarter Ultra Low RNA kit ( Clontech Laboratories Inc . [Mountain View , CA] ) . Amplified cDNAs ( 10 ng ) were fragmented using Covaris E220 ( Covaris [Woburn , MA] ) to an average fragment size of approximately 150 pb . Index-tagged sequencing libraries were generated from the fragmented cDNAs using the TruSeq RNA Sample Preparation v2 Kit ( Illumina Inc . [San Diego , CA] ) , starting from the End Repair step . Libraries were quantified using a Nanodrop spectrophotometer ( Thermo Fisher Scientific [Waltham , MA] ) and their size distributions were determined using the Bioanalyzer DNA-1000 Kit ( Agilent ) . Libraries were sequenced on the Genome Analyzer IIx ( Illumina ) following the standard sequencing protocol with the TruSeq SBS Kit v5 ( Illumina ) . Fastq files containing the sequencing reads for each library were extracted and demultiplexed using Casava v1 . 8 . 2 ( Illumina ) . Reads were pre-processed with Cutadapt ( Martin , 2011 ) , to remove both TruSeq adaptor and SMARTer primer sequences . The resulting reads were mapped on the mouse transcriptome ( Ensembl gene-build GRCm38 . v70 ) and genome , using RSEM v1 . 2 . 3 ( Li and Dewey , 2011 ) and Bowtie2 v2 . 0 . 6 ( Langmead and Salzberg , 2012 ) , respectively . The observation that a significant fraction of reads could be mapped on the genome , but not on the transcriptome , in close proximity to dA/dT-rich sequences , suggested that traces of contaminant genomic DNA could have been artifactually amplified by poly ( A ) priming . To avoid overestimation of transcript read counts , read clusters mapped on the genome at less than 50 bp downstream of dT-rich or upstream dA-rich regions were discarded . Filtered reads were mapped again on the transcriptome , and quantified , with RSEM . Only genes with at least 2 counts per million in at least 2 samples were considered for statistical analysis . Data were then normalized and differential expression tested using the Bioconductor package EdgeR ( Robinson et al . , 2010 ) . Hierarchical Clustering was run using Genesis Software ( Sturn et al . , 2002 ) on the normalized expression profiles of the subset of genes with a p-value smaller than 0 . 05 and similar behavior in the two replicates of each cell type . Gene Set Enrichment Analysis ( GSEA ) ( Subramanian et al . , 2005 ) was run on the list of genes expressed in at least one cell type . The RNASeq dataset is available at http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? token=gzijkssyjraldqd&acc=GSE50566 ( authors: Francisco Sanchez-Madrid and Olga Barreiro; year of publication: Sep 4th 2013; title: Homeostatic skin contains two different subsets of resident macrophages with distinct origin and gene profile; GEO Accession Number GSE50566 ) . RNA was extracted from sorted macrophage populations obtained from chimeric CD45 . 1 ( donor ) -CD45 . 2 ( host ) animals using the Absolutely RNA Nanoprep Kit ( Agilent Technologies [Santa Clara , CA] ) . RNA quantity and quality were determined using a 2100 Bioanalyzer ( Agilent Technologies ) and a Nanodrop-1000 Spectrophotometer ( Thermo Fisher Scientific ) . Each RNA sample was amplified using the MessageAmp II aRNA Amplification Kit ( Ambion ) . Total amplified mRNA ( 30 ng ) was converted to cDNA and loaded on a TaqMan Array Micro Fluidic Card ( Applied Biosystems , Life Technologies , Thermo Fisher Scientific [Foster City , CA] ) . Relative gene expression was calculated with the ABI Prism 7900 HT Sequence Detection System ( Applied Biosystems ) and Qbase software ( Biogazelle [Gent , Belgium] ) , using Actb , B2m , Hprt1 , Gusb and Gapdh genes as reference targets for the analysis of the immune system and inflammation gene signature array and 18S , Raf1 , Ctnnb1 , Eef1a1 for the analysis of the stem cell gene signature array . A comparative gene expression analysis was carried out by calculating ΔlogCt from the average logCt for every gene from each type of tissue-resident macrophages . Sorted skin anti- and pro-inflammatory macrophage populations were obtained from the skin of C57BL6 mice . Cells were seeded ( 40 , 000 cells/100 µl RPMI medium ) and stimulated in vitro with LPS ( 1 µg/ml ) for 24 hr . Supernatants were used for analysis of cytokine secretion using Mouse Inflammation ( 17-Plex ) multiplex kit ( ANTIGENIX AMERICA Inc [Huntington Station , NY] ) . TGF-β and IL-10 secretion were analyzed with an ELISA kit ( eBioscience ) . Sorted donor and host skin macrophages of quimeric mice were diluted in PBS containing 0 . 5% BSA and centrifuged in a cytospin centrifuge for 10 min at 800 rpm . Samples were stained with rabbit polyclonal anti-arginase I clone H-52 ( Santa Cruz Biotechnology ) or rabbit polyclonal anti-heme oxygenase-1 ( Chemicon ) following a standard immunofluorescence protocol . Then , samples were analyzed in the LSM 700 laser scanning microscope described above . FITC fluospheres ( Ø 0 . 2 μm ) ( Fluospheres , sulfate microspheres , Molecular Probes ) were diluted in PBS to a final concentration of 0 . 5% solids and adsorbed with 1 mg/ml endotoxin-free OVA ( Calbiochem , EMD Millipore [Billerica , MA] ) following the manufacturer’s instructions . Then , 200 μl of the colloidal dissolution were injected i . v . in either untreated , lethally γ-irradiated or splenectomized animals . Seventy-two hours later , animals were sacrificed and manually perfused with PBS to eliminate remaining fluospheres from the vasculature . In some cases , hind footpads were treated with clodronate or PBS liposomes 72 hr before the injection of fluospheres . Clodronate or PBS liposomes ( ClodronateLiposomes . org ) were injected s . c . ( 20 μl/hind footpad ) . For splenectomy , animals were anesthetized with a mixture of ketamine/xylazine ( 100/10 mg/Kg ) , the left lateral side was depilated and a small incision was made to remove the organ after ligation of the splenic vessels . Then , animals were allowed to recover from surgery for 1 month before proceeding with experiments . Home-made ultrasmall superparamagnetic iron oxide nanoparticles ( Groult et al . , 2015 ) ( abbreviated as NPs , 1 ml , 3 mg Fe/ml ) with 10 nm of core , 40 nm of hydrodynamic size and −35 mV of ζ-potential were incubated with endotoxin-free OVA ( 500 μl , 10 mg/ml ) for 1 hr at R . T . Non-adsorbed OVA remnants were washed out and further incubation of OVA-NP with diphtheria toxin ( DT ) ( 50 μl , 1 mg/ml ) ( 3 hr at R . T . ) were carried out , followed by extensive washing . Then , mice were injected i . v . with 1 dose ( 2 . 5 μl/g ) or 2 doses ( spaced 3d apart ) of this colloidal suspension and skin samples were analyzed by FACS and immunohistochemistry at different time-points . The back of mice was depilated and left to recover homeostasis ( 24 hr ) . Then , animals were anesthetized with isoflurane by inhalation . Four skin excisions of 5 mm in diameter were carried out using a bio-punch ( Kai Industries Co . , ltd [Seki , Japan] ) . Wounds were routinely measured every 24 hr until the sacrifice of the animals . Animals were injected i . v . either DT-OVA-NP or OVA-NP before or during the wound healing assay . After sacrifice , wounded tissue was embedded in paraffin and processed for histological analysis . Standard Hematoxylin-Eosin and Masson trichrome stainings were performed in wound sections . Alternatively , samples were processed for standard immunohistochemistry staining using anti-αSMA ( clone 1A4 , Sigma-Aldrich ) or for immunofluorescence using anti-Mac-3 ( clone M3/84 , Santa Cruz Biotechnology ) , anti-CD31 ( rabbit polyclonal anti-mouse/human , Abcam ) , and DAPI . Stained samples were analyzed on a Leica DM2500 microscope coupled to a Leica DFC420 camera using 5x , 10x and 20x objectives . Intensity quantification of collagen deposition and quantification of macrophages and vessels in the granulation tissue were performed using ImageJ 1 . 42q . Normality of data distribution was assessed with the Kolmogorov-Smirnov test . Statistical significance was calculated either with an unpaired two-tailed Student’s t-test , one-sample t-test , one-way ANOVA followed by either Tukey’s or Dunnett’s post-tests or two-way ANOVA followed by Sidak’s or Bonferroni post-test as indicated . All statistical analyses were carried out with GraphPad Prism v5 ( GraphPad Software Inc . [La Jolla , CA] ) . | The skin forms an essential barrier that defends our body from external damage . For this reason , it is important to understand the complex mechanisms involved in healing wounds and maintaining healthy skin . This could allow us to find effective treatments for skin diseases such as atopic dermatitis and psoriasis . Immune cells called macrophages can protect the body in different ways depending on the signals they receive . Their protective roles include killing microbes that may cause disease , and helping to repair damaged tissues . Barreiro et al . have now investigated the roles of the macrophages in the skin by means of a number of complementary techniques , including a method called intravital microscopy that can image cells in a living organism . The experiments revealed that a division of labor exists among the macrophages that reside in the skin of mice . Some macrophages help to trigger inflammatory responses in the skin . These immune cells disappear after being exposed to ionizing radiation ( such as that used to treat cancer ) but can be replaced via a bone marrow transplant . Other macrophages that help to repair tissues can survive being exposed to ionizing radiation but cannot resist high levels of ultraviolet light . Both types of macrophages perform unique and essential roles , and both types are necessary for maintaining healthy skin . Barreiro et al . also discovered that the skin macrophages that help to repair tissues have the ability to move into blood vessels and take up substances from the blood . A question for future investigation is whether the macrophages perform this scavenging process to trigger a protective immune response in the nearby skin . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"immunology",
"and",
"inflammation"
] | 2016 | Pivotal role for skin transendothelial radio-resistant anti-inflammatory macrophages in tissue repair |
CTCF and cohesin are key drivers of 3D-nuclear organization , anchoring the megabase-scale Topologically Associating Domains ( TADs ) that segment the genome . Here , we present and validate a computational method to predict cohesin-and-CTCF binding sites that form intra-TAD DNA loops . The intra-TAD loop anchors identified are structurally indistinguishable from TAD anchors regarding binding partners , sequence conservation , and resistance to cohesin knockdown; further , the intra-TAD loops retain key functional features of TADs , including chromatin contact insulation , blockage of repressive histone mark spread , and ubiquity across tissues . We propose that intra-TAD loops form by the same loop extrusion mechanism as the larger TAD loops , and that their shorter length enables finer regulatory control in restricting enhancer-promoter interactions , which enables selective , high-level expression of gene targets of super-enhancers and genes located within repressive nuclear compartments . These findings elucidate the role of intra-TAD cohesin-and-CTCF binding in nuclear organization associated with widespread insulation of distal enhancer activity .
The mammalian genome is organized into stereotypical domains , averaging ~700 kb in length , called Topologically Associating Domains ( TADs ) ( Dixon et al . , 2012; Nora et al . , 2012 ) . TADs are insulated chromatin domains whose genomic boundaries are often retained across tissues ( Dixon et al . , 2012 ) and have been conserved during mammalian evolution ( Vietri Rudan et al . , 2015; Dixon et al . , 2015 ) . TADs provide a stable genomic architecture that constrains enhancer-promoter contacts , while allowing for dynamic tissue-specific interactions that stimulate gene expression within TADs , thereby linking chromatin structure and positioning to gene expression ( Dowen et al . , 2014; Sexton et al . , 2007 ) . Hi-C , an unbiased genome-wide chromosome conformation capture method ( Lieberman-Aiden et al . , 2009 ) , identifies TADs based on their insulation from inter-domain interactions and by the increased frequency of intra-domain interactions that occurs within individual TADs ( Dixon et al . , 2012; Nora et al . , 2012 ) . TADs show substantial overlap with features of nuclear organization identified using other approaches , including replication domains , lamina-associated domains , and A/B chromatin compartments ( Dixon et al . , 2015; Pope et al . , 2014; Nora et al . , 2013 ) . TADs impact gene expression by insulation , which limits a given gene’s access to regulatory regions ( Le Dily et al . , 2014 ) . While TAD structures are often shared across tissues within a species , some individual TADs show tissue-specific differences in their spatial positioning within the nucleus , and in their overall activity , transcription factor ( TF ) binding patterns , and patterns of expression of individual genes ( Dixon et al . , 2015 ) . It is unclear to what extent these large megabase-scale chromatin structures exert regulatory control over the multiple , often variably-expressed , genes found within their boundaries . Two key protein factors , CCCTC-binding factor ( CTCF ) and the multi-subunit cohesin complex , are the primary architects of nuclear organization in mammals ( Ong and Corces , 2014; Sanborn et al . , 2015; Guo et al . , 2015 ) . CTCF and cohesin cooperatively engage genomic DNA via a loop extrusion complex , which is dynamically mobile within TAD boundaries and may help organize TAD structure ( Sanborn et al . , 2015; Fudenberg et al . , 2016; Rao et al . , 2014 ) . CTCF is an 11 zinc finger protein that stably binds DNA and can serve as an insulating enhancer-blocker and a modulator of 3D chromatin structure ( Phillips and Corces , 2009 ) . Sites bound by both cohesin and CTCF ( cohesin-and-CTCF ( CAC ) sites ) are associated with insulator function ( Dowen et al . , 2014; Zuin et al . , 2014 ) and are found at TAD boundaries ( Dixon et al . , 2012; Nora et al . , 2012 ) . In contrast , sites bound by cohesin but not CTCF ( cohesin-non-CTCF ( CNC ) sites ) are found at tissue-specific promoters and enhancers ( Kagey et al . , 2010 ) and may help to stabilize large TF complexes ( Faure et al . , 2012 ) . CAC complexes are also associated with topoisomerase-IIβ ( Top2b ) , which presumably relieves the torsional strain of the extrusion complex ( Uusküla-Reimand et al . , 2016 ) . Complete knockout of either CTCF or cohesin is embryonic lethal ( Heath et al . , 2008; White et al . , 2013; Xu et al . , 2010 ) , whereas partial depletion of CTCF or cohesin results in altered gene expression but has more limited phenotypic impact , increasing radiation sensitivity , DNA repair defects , and cell cycle arrest ( Ong and Corces , 2014; Xu et al . , 2010; Moore et al . , 2012 ) . Complete removal of CTCF or cohesin-related factors , achieved using inducible degradation systems , leads to a complete loss of virtually all loop structures in a highly dosage-dependent manner ( Nora et al . , 2017; Rao et al . , 2017; Schwarzer et al . , 2017 ) . Mutations affecting CAC loop anchors are frequently seen in cancer and lead to dysregulation of adjacent genes , evidencing the functionality of these loops ( Ji et al . , 2016; Katainen et al . , 2015; Fujimoto et al . , 2016 ) . However , there are many more CAC sites within TADs than at TAD boundaries , and it is not clear what factors differentiate loop-forming CAC sites at TAD boundaries from other CAC sites in the genome . Chromatin interactions can be studied by Hi-C analysis , which under standard conditions provides a resolution of 25–100 kb and has been used to study nuclear organization at the level of megabase-scale TAD structures . However , high resolution Hi-C datasets obtained using extreme deep sequencing ( >25 billion reads ) have led to two key discoveries ( Rao et al . , 2014 ) . First , ~90% of DNA loops ( ‘loop domains’ , defined as local peaks in the Hi-C contact matrix ) are associated with both CTCF binding and cohesin binding , and 92% of such loops involve inwardly oriented CTCF anchors ( Rao et al . , 2014 ) . Thus , loop anchors are bound at asymmetric CTCF motifs that face the loop interior . This previously unappreciated feature of CTCF loops facilitates the identification of such loops in silico ( Sanborn et al . , 2015; Oti et al . , 2016 ) . Furthermore , expression of neighboring genes changes in a predictable manner when CTCF anchors are inverted or deleted by CRISPR/Cas9 genomic editing ( Dowen et al . , 2014; Sanborn et al . , 2015; Guo et al . , 2015 ) . Second , extreme deep sequencing Hi-C studies identify a much larger number of shorter loops than previously recognized ( ~10 , 000 loops with a median size of 185 kb ) ( Rao et al . , 2014 ) , many of which represent complex nested structures ( e . g . , isolated cliques ) ( Sanborn et al . , 2015 ) . The ability to distinguish between such substructures has led to predictions ranging from 103 to 106 loops per genome , depending on the 3C-based analysis method and the cutoff values employed ( Sanborn et al . , 2015; Handoko et al . , 2011; Fullwood et al . , 2009; Jin et al . , 2013 ) . The presence of nested loop structures may be a general feature of topological nuclear organization , and the ability to detect such structures is dependent on the method , resolution , and computational approach ( Handoko et al . , 2011; Fullwood et al . , 2009; Jin et al . , 2013; Hnisz et al . , 2016; Weinreb and Raphael , 2016 ) . While sub topologies within TADs have been observed , it is unknown whether those interactions represent enhancer-promoter loops or other looped structures , and whether they are mediated by cohesin , mediator , or other architectural proteins ( Zuin et al . , 2014; Sofueva et al . , 2013 ) . Short , <200 kb CTCF-anchored loops , termed chromatin contact domains or super-enhancer domains , have been identified in mouse embryonic stem cells ( mESCs ) by ChIA-PET experiments that select for CTCF and cohesin binding sites ( via immunoprecipitation of Smc1 ) ( Handoko et al . , 2011; Tang et al . , 2015 ) , and are enriched for tissue-specific genes and enhancers ( Dowen et al . , 2014; Handoko et al . , 2011 ) . However , these genomic regions represent a minority of CTCF-anchored DNA loops , and likely do not fully represent all of the nuclear topological domains evident in high resolution Hi-C maps ( Rao et al . , 2014; Rao et al . , 2017; Rowley et al . , 2017 ) . Given the inability to identify CAC-anchored intra-TAD loops from standard , low resolution Hi-C data , we sought to build on the above advances and develop a computational method to predict such subTAD-scale loops by using only 2D ( CTCF and cohesin ChIP-seq binding activity ) and 1D ( CTCF motif orientation ) information . Here we define intra-TAD loops anchored by cohesin and CTCF , and that contain at least one gene , which represent a superset encompassing super-enhancer and polycomb domains ( Dowen et al . , 2014 ) . These CAC-mediated intra-TAD loops are mechanistically distinct from short range enhancer-promoter loops , and from longer range genomic compartmentalization ( Rao et al . , 2017; Schwarzer et al . , 2017; Stevens et al . , 2017 ) , whose impact on gene expression in mouse liver is also discussed . Here we present , and then validate in three mouse tissues and two human cell lines , a computational method to identify intra-TAD loops genome-wide . We elucidate the structural and functional features of the intra-TAD loops identified , and those of the better-established TADs , including their impact on gene expression in a mouse liver model . We show that , mechanistically , intra-TAD loops are anchored by loop extrusion CAC complexes that are shared across tissues and show strong conservation . Further , we demonstrate that , at a functional level , intra-TAD loops insulate repressive chromatin mark spread and thereby enable selective expression of genes at a high level compared to their immediate genomic neighbors , notably genes targeted by super-enhancers , and genes that are otherwise found in repressive nuclear compartments . These findings reveal how intra-TAD loops harness many of the same mechanisms as TAD-scale loops but in ways that allow for greater local control of gene expression .
We present , and then validate in multiple mouse and human cell models , a computational method that uses 2D ( ChIP-seq ) and 1D ( DNA sequence ) information to predict 3D-looped intra-TAD structures anchored by cohesin and CTCF ( CAC sites ) , and we provide evidence that the intra-TAD loops predicted underpin a general mechanism to constrain the interactions between distal enhancers and specific gene targets . While select instances of CAC-mediated loop insulation within TADs have been described ( Dowen et al . , 2014; Willi et al . , 2017; Hanssen et al . , 2017 ) , our work establishes that this phenomenon is a more general feature of genomic organization and regulation than previously appreciated . The intra-TADs described here are nested , CAC-anchored loops whose formation may be a result of extrusion complex pausing within larger domains ( i . e . , TADs ) ; these loops act to constrain the promoter contacts available to a given distal enhancer , and correspondingly , the distal enhancer contacts available to a given promoter ( Hnisz et al . , 2016 ) . We also provide evidence that the loop-forming CTCF sites , but not other CTCF sites , are highly insular . This insulation is apparent from the blockage of repressive histone mark spread and by the inhibition of chromatin contacts across intra-TAD loop and TAD boundaries . The impact of this insulation is highlighted for super-enhancer regions , such as the super-enhancer upstream of Alb , where local insulation by CAC-anchored intra-TAD loops both enables and constrains strong near-cis interactions , which facilitate the high expression of Alb and presumably also other liver-expressed genes regulated by super-enhancers . Weaker trans interactions with distal active regions were also observed , and are likely driven by a distinct mechanism , such as aggregation of transcription factories or super-enhancers ( Rao et al . , 2017; Osborne et al . , 2004 ) . Genomic interactions occur at three levels: ( 1 ) compartmentalization , where inactive regions localize to the nuclear periphery and active chromatin compartments aggregate toward the center of the nucleus in cis or trans in a largely cohesin-independent manner , as proposed in the transcription factory model ( Rao et al . , 2014; Seitan et al . , 2013; Osborne et al . , 2004; Lieberman-Aiden et al . , 2009 ) ; ( 2 ) CAC-dependent looping , which generates many tissue-invariant scaffolds along the linear chromosome ( Dixon et al . , 2012; Sanborn et al . , 2015; Hnisz et al . , 2016 ) ; and ( 3 ) enhancer-promoter looping within CAC-loops , which may be directed by cohesin non-CTCF ( CNC ) sites , mediator , or tissue-specific TFs ( Dowen et al . , 2014; Kagey et al . , 2010; Faure et al . , 2012 ) . If TADs define the broad domain within which a cohesin-driven extrusion complex generally operates , then we have presented a simple method to identify loops within this region that form as a result of dynamic loop extrusion movement and pausing at additional loop anchors . We have used the term intra-TAD loops , also referred to as sub-TADs , to highlight their subdivision of TAD-internal genomic space , although they are functionally similar to loop domains , isolated cliques , and insulated neighborhoods , which tend to overlap or be contained within TADs ( Sanborn et al . , 2015; Rao et al . , 2014; Hnisz et al . , 2016 ) . Our computational method cannot predict CTCF-independent loops , such as those mediated cohesin alone ( enhancer-promoter loops ) , although such loops are likely constrained by CAC driven intra-TADs , as was highlighted by our Albumin 4C-seq results . The method for CAC-mediated intra-TAD loop identification described here builds on the strong preference for inward-facing CTCF motifs evident from high resolution Hi-C data ( Sanborn et al . , 2015; Rao et al . , 2014 ) , and will be most useful for the identification of intra-TAD CAC loops for the large number of cell lines and tissues that lack high resolution Hi-C data . In these cases , intra-TAD loop domains cannot be identified because there is not sufficient local enrichment to calculate a corner score with the arrowhead algorithm ( Rao et al . , 2014 ) . Further , while we used TAD boundaries from standard resolution liver Hi-C data to filter out longer CAC loops , the frequent conservation of TADs across both tissues and species ( Dixon et al . , 2012; Vietri Rudan et al . , 2015 ) broadens the applicability of our method to cell types , and perhaps to new species , for which Hi-C data is not available and TAD boundaries have not been determined . Thus , even in the absence of TAD coordinates , our method identifies TAD and intra-TAD looping events , which may provide an invaluable first approximation for understudied organisms . As we have tuned our parameters to identify loop structures comparable in size and number to those found previously in mouse and human models , the parameters used to filter an initial set of loop anchors may need to be adjusted for other model organisms . We have used both CTCF and cohesin peak strength as the primary predictor of intra-TAD loop strength , which is a reasonably good predictor of interactions ( Sanborn et al . , 2015; Oti et al . , 2016 ) . An alternative machine learning approach to predicting CTCF/cohesin-mediated interactions , posted as an on-line preprint during review of our manuscript ( Kai et al . , 2017 ) , uses data from up to 77 genomic-derived features to predict CTCF-mediated loops in three human cell lines . A key finding from this work was that cohesin strength was consistently the most predictive feature of CTCF loops , followed by CTCF binding strength ( Kai et al . , 2017 ) . This method also captures loops that lack convergent CTCF motif orientation , which represent as few as 8% of the total for loop domains ( Rao et al . , 2014 ) , or as many as 20% in the case of Insulated Neighborhoods ( Ji et al . , 2016 ) . However , the identification of this subset of loops comes at the expense of requiring a minimum of 16 features for a given cell type , whereas our approach only requires three features ( CTCF motif , CTCF ChIP-seq , and cohesin ChIP-seq data ) . Importantly , the three features used by our method represent 3 of the top four predictive features identified in ( Kai et al . , 2017 ) . The computational method presented here , which was validated in both mouse and human cell models , provides a practical alternative to using high resolution Hi-C libraries for the identification of intra-TAD loops . High resolution Hi-C requires extremely deep sequencing , which is costly , both in terms of computational and experimental laboratory resources , and has only been achieved for a small number of cell lines ( Rao et al . , 2014; Jin et al . , 2013; Bonev et al . , 2017 ) . Strategies to reduce the need for extreme deep sequencing to identify interactions at high resolution have been proposed ( Weinreb and Raphael , 2016; Martin et al . , 2015; Zhang et al . , 2017 ) , and are beginning to make higher resolutions possible in more systems , however , the sequencing depth and cost will likely remain out of reach for many labs . Antibody enrichment for select genomic regions followed by chromosome conformation capture , as implemented in ChIA-PET , is an experimental alternative to intra-TAD prediction . ChIA-PET and other 3C-based antibody enrichment methods select for genomic regions that are highly bound by the protein ( s ) of interest ( e . g . , CTCF and cohesin ) , and can therefore identify ‘many to many’ interactions , instead of the ‘all to all’ interactions identified by Hi-C; these methods are therefore more practical than Hi-C , in terms of their sequencing depth requirements ( Fullwood et al . , 2009 ) . However , ChIA-PET still requires ~10 fold more extensive deep sequencing per sample ( ~400 million reads ) than is needed to obtain the CTCF and cohesin ChIP-seq data utilized in our computational analysis to identify intra-TAD loops . Further , as ChIA-PET uses antibody to select for genomic regions bound by CTCF and/or cohesin , it is difficult to differentiate strength of antibody binding to the anchor proteins from strength of chromatin interaction between the anchors . Of the various CTCF loops described in the literature , insulated neighborhoods are most similar to the intra-TAD loops described here . Insulated neighborhoods are proposed to rectify the observation of smaller and more abundant loops , evident in ChIA-PET datasets , with the established TAD model of large loops from Hi-C experiments ( Dowen et al . , 2014; Rao et al . , 2014; Tang et al . , 2015 ) . The TAD and intra-TAD loop anchors identified here together comprise 27% of all liver CTCF binding sites , consistent with the 30% of murine ESC CTCF peaks that overlap insulated neighborhood anchor regions ( Hnisz et al . , 2016 ) . The precise mechanism that differentiates these CTCF sites , which anchor intra-TAD and TAD loops , from the larger number of non-anchor CTCF binding sites present in any given tissue is unknown . Further , it is unclear what role the typically weaker remaining ~70% of CTCF sites play in organizing the nucleus . Some of these non- ( intra- ) TAD anchor CTCF sites may serve other , unrelated functions , given the ability of CTCF to interact with other TFs , bind RNA , and regulate splicing mechanics ( Lutz et al . , 2000; Ross-Innes et al . , 2011; Sun et al . , 2013; Saldaña-Meyer et al . , 2014; Shukla et al . , 2011 ) . Alternatively , some of these CTCF sites may anchor loops present in only a minority of cells in the population analyzed , which would account for their overall weaker signals . Early single cell Hi-C experiments suggested that TADs are present in virtually all individual cells ( Nagano et al . , 2013 ) , however , more recent studies indicate cell-to-cell variability in TADs within a given cell population , although the presence of distinct active and inactive genomic compartments is common across most individual cells ( Stevens et al . , 2017; Wang et al . , 2016 ) . Truly high-resolution elucidation of single cell intra-TAD structures may not be possible due to the intrinsic limitation of two potential ligation events per fragment in any given cell . We found that CAC sites are found at insulators and also at promoters , which we defined as DNase hypersensitive sites ( DHS ) with high a histone-H3 K4me3/K4me1 ratio , whereas CNC sites are primarily at enhancers and weak enhancers . Others find that promoters , when defined as the set of all TSS upstream regions ( including those not at a DHS ) , are bound by cohesin alone ( Kagey et al . , 2010; Faure et al . , 2012 ) . Further , we found that CTCF-bound open chromatin regions distal to promoters ( insulator-DHS ) show features that distinguish them from other classes of open chromatin ( promoter-DHS and enhancer-DHS ) , including the absence of enhancer marks and their general conservation across tissues . Thus , these insulator-DHS regions are not simply enhancers with CTCF bound . Supporting this , insulator regions consistently show less intrinsic enhancer activity than weak enhancers in in vivo enhancer screens ( Vanhille et al . , 2015 ) . It is less clear what role CTCF binding in the absence of cohesin plays in the nucleus , as we found such sites lack insulating activity and also lack strong directional interactions . As CTCF binding is always intrinsically directional , due to its non-palindromic motif , the absence of directional interactions from CTCF-non-cohesin sites suggests that the directionality of interactions with CTCF sites at TAD and intra-TAD loop anchors is conferred by other factors associated with the extrusion complex , such as cohesin ( Sanborn et al . , 2015; Fudenberg et al . , 2016 ) or Top2b ( Uusküla-Reimand et al . , 2016 ) . However , our findings suggest that the interactions of Top2b involve binding to cohesin , and not CTCF , as indicated by the high frequency of CNC sites bound by Top2b vs . very low frequency of Top2b binding at CTCF-non-cohesin sites ( Figure 2—figure supplement 3B ) . Furthermore , binding by Top2b does not distinguish TAD anchors from intra-TAD loop anchors . Indeed , by all metrics tested , we found no TF or motif that differentiates TAD anchors from intra-TAD loop anchors , although the existence of some unknown differentiating factor cannot be ruled out . Cohesin can stabilize large protein complexes comprised of up to 10 distinct TFs at enhancers ( Faure et al . , 2012 ) , and could thus facilitate the binding of other unknown proteins to the loop extrusion complex . Cohesin is continuously recycled throughout the genome by loading and release factors ( Busslinger et al . , 2017 ) , and so it is unclear how insulator activity is effectively maintained at TAD and intra-TAD loop anchors in such a dynamic environment . We found that CNC sites , which are primarily found at enhancers , consistently show the least insulation of repressive histone marks , just as they show the least insulation of chromatin contacts . This provides further evidence that TAD and intra-TAD loop anchors are functionally unique sites , and are not a moonlighting feature of CTCF bound to enhancer regions . Furthermore , while enhancers are strongly enriched for genetic non-coding variants , genetic variations at loop anchors are rare ( Hnisz et al . , 2016 ) . Mutations that occur at loop anchors can result in dramatic phenotypes like polydactyly or tumorigenesis ( Lupiáñez et al . , 2015 ) and often occur in cancer ( Ji et al . , 2016 ) . Disruption of specific , individual CAC-mediated loop anchors using genomic editing tools results in aberrant chromatin contacts and misregulation of neighboring genes in a largely predictable manner , although some redundancy may occur when multiple nearby anchors are present ( Dowen et al . , 2014; Hnisz et al . , 2016; Willi et al . , 2017 ) . The computational method for intra-TAD loop discovery , described here , is a substantial improvement over prior implementations of computational loop prediction ( Sanborn et al . , 2015; Oti et al . , 2016 ) . The loops we identified were longer and fewer in number ( ~9500 vs ~60 , 000 ) , showed much stronger insulation of chromatin interactions and greater insulation of repressive histone marks , and displayed considerably greater overlap with cohesin-mediated loops identified by ChIA-PET using antibody to the cohesin subunit Smc1 . Key features of our computational method include the consideration of both CTCF and cohesin binding strength , as noted above , as well as TAD structure and consistency across biological replicates . Our use of both CTCF and cohesin binding strength in predicting intra-TAD loops is supported by a recent study of CTCF sites nearby the mouse α-globin gene cluster , where the presence of CTCF alone was not sufficient to predict DNA loop interactions , and where insulation by individual CAC sites ranged widely – from none to moderate to very strong insulation – in direct proportion to the strength of CTCF binding , as revealed by deletion of individual CTCF sites ( Hanssen et al . , 2017 ) . Furthermore , we developed a simple extension of our method that predicts TAD anchors when given a set of TAD boundaries ( Supplementary file 1C ) , and thereby overcome the limitation in identifying TAD anchors from low resolution , standard sequencing depth Hi-C datasets . We were thus able to identify well-defined inter-TAD regions , which we found are enriched for unique gene ontologies , notably , housekeeping genes with ribosomal , nucleosome , and mitosis-related functions . A further extension of our findings would be the explicit use of the intra-TAD and refined-TAD loop coordinates defined here to improve gene target assignments for distal regulatory elements , based on the insulating capacity of these CAC-anchored looped domains . Such an approach may be beneficial for the many model systems where distal enhancer activity is the clear driver of tissue specificity or a given disease state ( Hnisz et al . , 2016 ) . The ability to identify intra-TAD loops based solely on CTCF motif orientation and CTCF and cohesin ChIP-seq binding data , and then use these loops to improve gene target assignments for distal regulatory elements is likely to constitute a substantial improvement over ‘nearest gene’ and other , more nuanced target assignment algorithms , such as GREAT ( Raviram et al . , 2016 ) . In conclusion , our studies reveal that while TAD structures are readily apparent in routine Hi-C experiments , their structural organization and functional impact on the genome is not unique . Structurally , the 9 , 543 TAD-internal sub-loops that we identified for mouse liver have strong cohesin-and-CTCF-bound anchors and appear to be formed by the same loop extrusion mechanism responsible for TAD formation . Functionally , we hypothesize that these intra-TAD loops contribute to nuclear architecture as intra-TAD scaffolds that further constrain enhancer-promoter interactions . We further show that these intra-TAD loops maintain key properties of TADs , most notably insulation of chromatin interactions and insulation of repressive histone mark spreading . The insulation provided by intra-TAD loops may enable high expression of super-enhancer target genes , as illustrated for Alb in mouse liver , as well as high expression of individual genes within otherwise inactive TADs , as exemplified by Scd1 and the many other single gene intra-TAD loops that we identified . Given the increasing interest in interactions of genes with distal enhancers and other intergenic sequences , the rapid and cost-effective method described here for identification of intra-TAD structures that constrain long-range chromatin interactions may prove invaluable in many areas of genomic research .
Adult male and female CD-1 mice ( ICR strain; RRID:MGI:5659424 ) were purchased from Charles River Laboratories ( Wilmington , MA ) and were housed in the Boston University Laboratory Animal Care Facility . Animals were treated using protocols specifically reviewed for ethics and approved by Boston University's Institutional Animal Care and Use Committee ( IACUC; protocol 16–003 ) . Livers were collected from 8-week-old mice euthanized by cervical dislocation and rinsed in cold PBS . Livers were homogenized in a Potter-Elvehjem homogenizer using high sucrose homogenization buffer ( 10 mM HEPES ( pH 7 . 5 ) , 25 mM KCl , 1 mM EDTA , 2 M sucrose , 10% glycerol , 0 . 05 mM DTT , 1 mM PMSF , 0 . 15 mM spermine , 0 . 2% ( v/v ) spermidine , 1 mM Na orthovanadate , 10 mM NaF , and Roche Complete Protease Inhibitor Cocktail ) to prevent aggregation of nuclei and preserve chromatin structure . The resulting slurry was transferred on top of a 3 ml cushion of homogenization buffer followed by ultracentrifugation at 25 , 000 RPM for 30 min at 4°C in an SW41 Ti rotor to pellet the nuclei and remove cellular debris . The supernatant was carefully decanted to remove liquid , and residual solid debris was removed from the tube walls using a sterile spatula and a dampened Kimwipe . Nuclei were resuspended in 1 ml of crosslinking buffer ( 10 mM HEPES buffer ( pH 7 . 6 ) , 25 mM KCl , 0 . 15 mM 2-mercaptoethanol , 0 . 34 M sucrose , 2 mM MgCl2 ) and transferred to a 1 . 5 ml Eppendorf tube . To ensure consistent crosslinking , tubes were incubated for 3 min at 30°C prior to the addition of formaldehyde to a final concentration of 0 . 8% ( v/v ) . Samples were incubated in a 30°C water bath for 9 min with periodic mixing . Crosslinking was halted by the addition of 110 μl of 1 M glycine , followed by a 5 min incubation at room temperature . The crosslinked material was layered on top of 3 ml of high sucrose homogenization buffer and then centrifuged as above . The crosslinked nuclear pellet was resuspended at 4°C in 1 ml of 1X Radioimmunoprecipitation assay ( RIPA ) buffer ( 50 mM Tris-HCl , pH 8 . 0 , 150 mM NaCl , 1% IPEGAL , 0 . 5% deoxycholic acid ) containing 0 . 5% SDS and protease inhibitors until homogenous by gentle pipetting . Crosslinked nuclei in RIPA buffer containing 0 . 5% SDS were transferred to 15 ml polystyrene tubes ( BD Falcon # 352095 ) for sonication using a Bioruptor Twin instrument ( UCD-400 ) according to the manufacturer’s instructions . Briefly , samples were sonicated at 4°C for 30 s ON and 30 s OFF at high intensity for a total of 75 cycles . Sonicated material was transferred to 1 . 5 ml Eppendorf tubes , and large debris was cleared by centrifugation at 18 , 000 x g for 10 min at 4°C . The bulk of this material was snap frozen in liquid nitrogen and stored at −80°C for immunoprecipitation , except that a small aliquot ( 15 μl ) was removed to quantify material and ensure quality by gel electrophoresis , as follows . Aliquots from each sample were adjusted to 0 . 2 M NaCl , final concentration , then incubated for 6 hr at 65°C . After a three-fold dilution in nuclease-free water , 5 μg of RNase A ( Novagen: #70856 ) was added and samples were incubated for 30 min at 37°C . Samples were then incubated for 2 hr at 56°C with 20 μg of Proteinase K ( Bioline; BIO-37084 ) . This material was then quantified in a dilution series using PicoGreen assay ( Quanti-iT dsDNA Assay Kit , broad range , Invitrogen ) and analyzed on a 1% agarose gel to ensure the bulk of material was within 100–400 bp . Immunoprecipitation of crosslinked , sonicated mouse liver chromatin and downstream steps were as described previously ( Sugathan and Waxman , 2013 ) . Protein A Dynabeads ( 30 μl; Invitrogen: 1002D ) were incubated in blocking solution ( 0 . 5% bovine serum albumin in PBS ) with 5 μl of antibody to CTCF ( Millipore #07–729; RRID:AB_441965 ) or to the cohesin subunit Rad21 ( Abcam #992; RRID:AB_2176601 ) for 3 hr at 4°C . As a control , 1 μl of non-specific rabbit IgG was used ( Santa Cruz: sc-2027 ) . Bead immune-complexes were washed with blocking solution , followed by overnight incubation with 70 μg of liver chromatin . After washing with 1X RIPA ( containing 0 . 1% SDS ) and reverse crosslinking as described above , DNA was purified using the QIAquick Gel Extraction Kit ( Qiagen #28706 ) and quantified on a Qubit instrument with high sensitivity detection ( Invitrogen DNA HS# Q32854 ) , with ChIP yields ranging from 1 to 25 ng . Samples were validated by qPCR using primers shown in Supplementary file 4B . ChIP libraries were prepared for sequencing using NEBNext Ultra II DNA Library Prep Kit for Illumina according to the manufacturer’s instructions ( NEB , cat . #E7645 ) . All samples were subjected to double-sided SPRI size selection prior to PCR amplification ( Agencourt AMPure XP; Beckman Coulter: A63882 ) . Samples were assigned unique barcodes for multiplexing , and subjected to 8 rounds of PCR amplification with barcoded primers ( NEB , cat . #E7335 ) . Samples were sequenced either on an Illumina Hi-Seq 4000 instrument at the Duke Sequencing Core or an Illumina Hi-Seq 2000 instrument at the MIT BioMicroCenter , giving 50 bp single end reads at a depth of ~11–19 million reads per sample . A total of four CTCF and three Rad21 ( cohesin ) ChIP-seq samples were analyzed , representing four male mouse livers . The fourth liver CTCF sequencing sample , sample G133_M9 , did not have a matching cohesin ChIP-seq dataset from the same liver sample , and was therefore matched to a merged sample comprised of all three cohesin ChIP-seq replicates ( merged at the fastq file level , with processing described below ) . Raw and processed sequencing data are available at GEO accession number GSE102997 . Sequence reads were demultiplexed and mapped to the mouse genome ( build mm9 ) using Bowtie2 ( version 2 . 2 . 9 ) , allowing only uniquely mapped reads . Peaks of sequencing reads were identified using MACS2 ( version 2 . 1 . 1 ) as regions of high signal over background . Peaks were filtered to remove blacklisted genomic regions ( www . sites . google . com/site/anshulkundaje/projects/blacklists ) . Genomic regions called as peaks that contain only PCR duplicated reads , defined as >5 identical sequence reads that do not overlap any other reads , were also removed . All BigWig tracks for visualization in a genome browser were normalized for sequencing depth , expressed as reads per million mapped reads ( RPM ) using Deeptools ( version 2 . 3 . 3 ) . Unless otherwise indicated , all pairwise comparisons presented in the Figures were performed using a Kolmogorov–Smirnov test , where **** indicates p≤0 . 0001 , ***p≤0 . 001 , **p≤0 . 01 , and *p≤0 . 05 . Motifs within CTCF peak regions were identified using MEME Suite ( version 4 . 10 . 0; FIMO and MEME-ChIP options ) . FIMO was used to assign CTCF motif orientation and motif scores for CAC sites and to discover individual motif occurrences . De novo motif discovery was carried out using MEME-ChIP using default parameters ( Figure 2—figure supplement 3C , D ) . Similar results were obtained using Homer ( version 4 . 8 ) . Alternative CTCF motifs were downloaded from CTCFBSDB 2 . 0 ( http://insulatordb . uthsc . edu/download/CTCFBSDB_PWM . mat ) , however , these did not substantially change any results performed using the core JASPAR motif ( MA0139 . 1 ) . These motifs were explicitly used in Figure 2—figure supplement 3D , where no difference between intra-TAD loop anchor and TAD anchor motif usage was observed . Four male and four female mouse livers were processed for Albumin-anchored 4C-seq analysis using published protocols , with some changes for primary tissue ( van de Werken et al . , 2012 ) . To adapt the protocol for liver , care was taken to rapidly isolate single liver cells or nuclei suspensions prior to crosslinking . Specifically , two approaches to crosslinking were taken and both gave similar results . One male mouse liver and one female mouse liver sample were processed through the crosslinking step as described for the ChIP protocol , above , prior to quantification of nuclei . The other liver samples ( 3 males and 3 females ) were crosslinked as follows . Half of a liver ( ~0 . 5 g ) was dissected from each mouse , the gall bladder was removed , and the liver was rinsed with PBS . The liver was then minced and rapidly processed with 10 strokes in a Dounce homogenizer in PBS containing protease inhibitors ( PBS-PI; 1X Roche Complete Protease Inhibitor Cocktail; Roche #11697498001 ) . The resulting slurry was passed through a 40-micron cell strainer ( Corning #431750 ) , then pelleted and rinsed with PBS-PI ( centrifugation at 1 , 300 RPM for 5 min at 4°C ) . Following an additional spin , the cell pellet was resuspended well in 9 ml of PBS-PI at room temperature . 270 μl of 37% formaldehyde was added to give a final concentration of 1% , and crosslinking was carried out for 10 min with nutation at room temperature . The remaining formaldehyde was quenched with 1 . 25 ml of 1 M glycine . Crosslinked cells were pelleted and rinsed with PBS twice ( as above ) prior to lysis . The supernatant was removed following the second wash , and cell pellets were resuspended in 8 ml of lysis buffer ( 50 mM Tris ( pH 7 . 5 ) , 150 mM NaCl , 5 mM EDTA , 0 . 5% NP-40 , 1% TX-100 , and 1X Complete Protease Inhibitor Cocktail ) and incubated on ice for 40 min with occasional mixing . Lysed cells were spun down at 2 , 000 RPM for 5 min at 4°C then washed twice with PBS-PI , as above . Nuclei were pelleted , quantified using an Invitrogen Countess instrument , and snap frozen in 10 million nuclei aliquots . Primary digestion of 10 million nuclei with 50 , 000 U of DpnII ( NEB: #R0543 ) was performed overnight at 37°C in 450 μl of NEBuffer 3 ( NEB: #B7003S; 100 mM NaCl , 50 mM Tris-HCl , 10 mM MgCl2 , 1 mM DTT , pH 7 . 9 ) with agitation at 900 RPM . After confirming digestion by agarose gel electrophoresis , DpnII was inactivated with SDS ( 2% , final concentration ) and the samples then diluted 5-fold in 1X ligation buffer ( Enzymatics #B6030 ) . 200 U of T4 DNA ligase was added and primary ligation was carried out overnight at 16°C ( Enzymatics #L6030 ) . Ligation was confirmed by analysis of a small aliquot on an agarose gel , and reverse crosslinking was conducted by overnight incubation with 600 μg proteinase K at 65°C . After RNase A digestion and phenol/chloroform cleanup , samples underwent secondary digestion with 50 U of Csp6I ( Fermentas #ER0211 ) overnight at 37°C in 500 μl of 1X Buffer B ( Fermentas: #BB5; 10 mM Tris-HCl ( pH 7 . 5 ) , 10 mM MgCl2 , 0 . 1 mg/ml BSA ) . Csp6I was then heat inactivated for 30 min at 65°C . Samples were diluted 10-fold and secondary ligation was carried out as above , overnight at 16°C . The final PCR template was purified by phenol/chloroform clean up , followed by QiaPrep 2 . 0 column cleanup ( Qiagen #27115 ) . PCR reactions were performed using inversely-oriented 4C primers specific to the Alb promoter ( sequences shown in bold , below ) with dangling 5’ half adapter sequences ( Reading primer: ACACTCTTTCCCTACACGACGCTCTTCCGATCTGGTAAGTATGGTTAATGATC; Non-reading primer: GACTGGAGTTCAGACGTGTGCTCTTCCGATCTCTCTTTGTCTCCCATTTGAG ) . This design has two advantages: 1 ) the addition of barcodes in a secondary reaction allows a primer to be reused across samples; and 2 ) it avoids barcoding at the start of 4C read , which would reduce the mappable read length available for downstream analyses . 4C templates were amplified using Platinum Taq DNA polymerase ( Invitrogen #10966026 ) under the following conditions: 94°C for 2 min , 25 cycles at ( 94°C 30 s , 55°C 30 s , 72°C 3 min ) , then 4°C hold . A total of eight liver samples were analyzed ( four males , M1-M4; and four females , F1-F4 ) . For liver samples M3 , M4 , F3 and F4 , eight identical PCR reactions for each liver , processed in parallel , were prepared and then pooled to limit the impact of PCR domination in any single reaction . For liver samples M1 , M2 , F1 and F2 , two PCR reactions for each liver were sequenced separately then pooled at the fastq file level for downstream analyses . We observed that the 8 PCR pool reactions gave a more reproducible profile than the single PCR reactions . After pooling , 4C samples were purified using AMPure XP beads ( Beckman Coulter: #A63882 ) at a 1 . 5:1 ratio of beads to sample , washed with 75% ethanol , dried , and resuspended in 0 . 1X TE buffer to elute the DNA . 4C-seq samples were multiplexed and PCR amplified using standard NEB barcoded primers ( NEB #E7335 ) , as was done for ChIP-seq library preparations , but for five additional PCR cycles rather than the eight cycles used for ChIP libraries ( total of 30 cycles of PCR per sample: 25 cycles with viewpoint-specific primers followed by five cycles with viewpoint-generic barcoded NEB primers ) . 4C libraries were sequenced on an Illumina Hi-Seq 2500 instrument at the New York Genome Center giving 125 bp long paired end reads . Samples were each sequenced to a depth of ~1–5 million reads . Raw and processed sequencing data are available at GEO under accession number: GSE102998 . All Alb viewpoint 4C-seq reads were filtered to ensure a match for the bait primers used , then trimmed using a custom script to remove the first 20 nt of each read ( Source code 1 ) . Reads were then mapped to the mouse mm9 reference genome using the Burrows-Wheeler aligner ( bwa-mem ) allowing for up to two mismatches . The package r3Cseq ( Thongjuea et al . , 2013 ) was used to analyze the distribution of signal in cis , both near the bait and along chromosome 5 . Reads were counted per restriction fragment to obtain the highest possible resolution . Data shown in the main text figures are for the intersection of four male and four female mouse livers , merged according to sex using the intersection option , meaning that the 4C interactions shown are those present in all four samples for a given sex . This produces both a normalized read depth signal ( reads per million per restriction fragment ) and an associated p-value for the interaction , taking into account distance from viewpoint and reproducibility across replicates ( Figure 6—figure supplement 1A , C ) . A comprehensive view of all replicates is presented in Figure 6—figure supplement 1A . For all pairwise comparisons , a correlation between samples was calculated genome-wide using the UCSC utility bigWigCorrelate with default settings . To account for high signal immediately surrounding the viewpoint , this analysis was only conducted for regions > 10 kb from the viewpoint fragment . To analyze more distal cis interactions , we first calculated the normalized 4C signal observed per TAD along chromosome 5 , in units of RPKM per TAD . We observed a robust logarithmic decay of signal with increasing distance from the viewpoint TAD ( Figure 6C; R2 = 0 . 719 ) . Interacting TADs were defined according to observed over expected 4C signal relative to this background model . Interacting TADs were designated as follows: high , defined as regions with >2 fold enrichment over this background model ( observed/expected ) ; medium , defined as 1 . 5 to 2-fold enrichment; and low , between 1 . 5-fold enrichment and 1 . 5-fold depletion . Non-interacting TADs showed >1 . 5 fold depletion of signal . For the Alb viewpoint , we identified 17 high , 20 medium , 128 low , and 30 non interacting cis TADs along chromosome 5 . We required a more comprehensive background model to analyze interactions in trans . The tool 4Cker was used for its adaptive windowing and Hidden Markov model approach ( Raviram et al . , 2016 ) . The count tables from r3C-seq were merged by sex and imported , then trans analysis was conducted with the recommended parameters ( k = 20 ) . The default output identifies three classes of regions: interacting , low-interacting , and non-interacting . For our analysis , the interacting group was divided into two equal-number groups: high-interacting and medium-interacting , based on 4C interaction strength in the male liver samples . Trans interacting regions tend to be large ( median size of 1 . 8 Mb ) , therefore trans interacting TADs were defined as TADs wholly contained within these interacting regions . This corresponds to a total of 659 ( high ) , 618 ( medium ) , 969 ( low ) , and 77 ( non ) interacting trans TADs genome-wide . Cohesin-and-CTCF ( CAC ) sites were defined as CTCF peaks that were present in at least 2 of 4 individual mouse liver samples and that overlapped with a cohesin peak in any liver sample . CAC sites were scanned for a CTCF motif ( JASPAR motif MA0139 . 1 ) within the CTCF peak coordinates using the FIMO tool in the MEME Suite ( version 4 . 10 . 0 ) . For a given CAC site , the highest scoring motif occurrence for the canonical core CTCF motif ( MA0139 . 1 ) was considered . A ( + ) strand orientation indicates that the motif is found on the ( + ) genomic strand ( Watson strand ) . Each CAC site was represented by two different scores: a CTCF score = p * ( m/10 ) , where p is the CTCF peak strength ( MACS2 score ) and m is the CTCF motif score , as determined by FIMO; and a cohesin score = p * ( m/10 ) , where p is the cohesin ( Rad21 ) peak strength ( MACS2 score ) and m is the CTCF motif score , as determined by FIMO . We modified a published algorithm for CTCF-mediated loop prediction ( Oti et al . , 2016 ) to predict intra-TAD loop structures . Key modifications to the algorithm include the following: incorporation of cohesin ChIP-seq data in scoring , based on the finding that CTCF signal in the absence of cohesin is not sufficient to predict chromatin interactions ( Hanssen et al . , 2017 ) ; consideration of TAD structure , TSS overlap , and consistency across biological replicates when filtering to obtain the final set of predicted loops; and a final target set of approximately 10 , 000 intra-TAD loops , based on experimental results from high resolution Hi-C analyses ( Rao et al . , 2014 ) . First , CAC sites were identified from mouse liver ChIP-seq data for Rad21 and CTCF , obtained as described above . Next , each chromosome was scanned for putative intra-TAD loops , formed between a ( + ) anchor [upstream anchor , that is , CAC site with a CTCF motif ( JASPAR motif MA0139 . 1 ) on the ( + ) strand] at the start of a loop and a ( - ) anchor [downstream anchor , that is , CAC site with a CTCF motif on the ( - ) strand] at the end of a loop , as described for prediction of intra-chromosomal CTCF loops in ( Oti et al . , 2016 ) . Scanning was initiated from the beginning of each chromosome , and a list of putative ( + ) anchors was generated . Next: ( 1 ) if the next CAC site encountered was a ( - ) anchor , the pair of ( + ) and ( - ) anchors was recorded as a putative intra-TAD loop . The ( + ) anchor was paired with all subsequent , downstream ( - ) anchors until another ( + ) anchor was encountered , at which point the list of putative intra-TAD loops was closed , ending with the last ( - ) anchor . Alternatively , ( 2 ) if the next CAC sites encountered were ( + ) anchors , then all such ( + ) anchors were retained as putative upstream anchors , until the next ( - ) anchor was reached , and then all such ( + ) anchors were paired ( i . e . , assigned to loops ) with all of the subsequent , downstream ( - ) anchors until a new ( + ) anchor was encountered , as described under ( 1 ) , at which point the list of putative loops was closed , ending with the last ( - ) anchor . A new scan for putative intra-TAD loops was then initiated in a linear fashion , starting from the next ( + ) anchor until all chromosomes were scanned and a set of putative intra-TAD loops was obtained . Chromosome scanning for putative intra-TAD loops was then repeated as described above after removing 10% of the CAC sites – those with the lowest CTCF scores ( defined above ) . Chromosome rescanning was repeated iteratively until the number of putative intra-TAD loops decreased to as close to 20 , 000 as possible ( removing the lowest scoring loops if needed so that all replicates had exactly 20 , 000 loops prior to merging ) . A parallel series of iterative scans was carried out , except that 10% of the CAC sites with the lowest cohesin scores ( defined above ) were removed at each iteration , to generate a second set of ~20 , 000 putative intra-TAD loops . The intersection of the two sets of 20 , 000 putative intra-TAD loops was then determined . The same iterative process of intra-TAD loop prediction was carried out independently for each of the n = 4 individual mouse livers , based on an analysis of matched CTCF and cohesin ( i . e . , Rad21 ) ChIP-seq datasets for each liver . Thus , for each liver sample , a single putative intra-TAD loop set was generated from the intersection of two sets of predicted CAC-based loops , one using the CTCF score and the other using the cohesin ( Rad21 ) score; these scores were calculated using MACS2 scores for CTCF and cohesin ( Rad21 ) , respectively , together with the CTCF motif score m value , as described above . The overlap of these two putative intra-TAD loop sets was approximately 80% , and ranged from 15 , 999 to 16 , 892 loops for a given liver sample . Additional filters were then applied to remove intra-TADs that did not contain either a protein-coding TSS or a liver-expressed multi-exonic lncRNA TSS ( as defined in [Melia et al . , 2016] ) , as we were primarily interested in the impact of intra-TADs on gene expression and regulation . Putative intra-TAD loops that overlapped >80% of the length of a TAD , or whose ( + ) and ( - ) anchors are both TAD anchors ( defined below ) were also excluded , as these loops could not be distinguished from TAD loops . These two filters further reduced the putative intra-TAD loop sets to approximately 63% of the original 20 , 000 ( ranging from 12 , 395 to 12 , 962 loops across the four liver samples ) . A single merged ChIP-seq dataset ( merged at the fastq file level , separately for CTCF and for Rad21 datasets ) was treated as a fifth dataset . It was run through the full pipeline , above , and then sequentially intersected with the set of putative intra-TAD loops predicted for each individual liver to obtain a final set of 9543 intra-TAD loops identified in all four livers and also present in the 5th dataset ( merged sample ) . Each intra-TAD loop was assigned an intra-TAD loop score equal to the geometric mean of the ( + ) anchor’s CAC site CTCF score and that of its ( - ) anchor . A second intra-TAD loop score , equal to the geometric mean of the ( + ) anchor’s CAC site cohesin score and that of its ( - ) anchor , was also assigned . The CTCF and cohesin scores reported for each loop in the final intra-TAD loop lists ( Supplementary file 1B ) are those obtained from the merged sample . Custom scripts for intra-TAD loop prediction are available in Source code 1 . Loop predictions for two other mouse cell types ( mESC and NPC ) and in two human cell lines ( GM12878 and K562 ) was carried out as described above , with the following modifications during filtering . For the mouse cells , ChIP-seq data from biological replicates ( n = 4 for mESC and n = 3 for NPC ) was obtained from public sources ( see below ) for CTCF and cohesin ( ChIP for the subunit Smc1 ) . Further , TADs from the same cell type were used to filter based on TAD overlap ( using TAD boundaries from [Bonev et al . , 2017] ) . TSS overlap used the same definitions as above ( RefSeq and multi-exonic lncRNA TSS defined in mouse liver [Melia et al . , 2016] ) . For human loop predictions , cohesin ( ChIP for the subunit Rad21 ) and CTCF ChIP-seq data were obtained from biological replicates for K562 cells and for GM12878 cells ( n = 5 for both cell lines ) , and overlap with Refgene TSS ( hg19 ) was used to filter the merged loops , as TADs were not defined in ( Rao et al . , 2014 ) . Supplementary file 4A provides further details on data sources and accession numbers . Gene expression values for liver-expressed protein coding genes are log2 ( FPKM + 1 ) values for adult male mouse liver from ( Melia et al . , 2016 ) . Liver-expressed non-coding genes are expressed in FPKM based on the gene models and expression values from ( Melia et al . , 2016 ) . To express the tissue specificity of a gene’s expression across a panel of 21 mouse tissues ( including liver ) , we used Tau , which was shown to be the most robust in a recent study ( Kryuchkova-Mostacci and Robinson-Rechavi , 2017 ) . Testis was excluded from this analysis because a large proportion of testis-expressed genes are highly tissue specific . For each tissue , the maximum FPKM per gene between the two replicates was used . These FPKM values were log transformed and a Tau value , ranging from 0 to 1 , was calculated , where one represents high tissue specificity: τ = [ ∑ni = 1 ( 1−yi ) ] / ( n−1 ) , where yi = xi / [ max1≤i≤n ( xi ) ] , n is the number of tissues , and xi is the expression of the gene in tissue i . Hi-C data was processed using the HiC-Pro package ( version 2 . 7 . 0 ) ( Servant et al . , 2015 ) for mapping and read filtering , followed by Homer ( version 4 . 8 ) for downstream analyses such as PCA analysis and aggregate contact profiles . Biological replicates were merged to increase read depth . The default Homer background model was used for all datasets , where the expected frequency of interactions takes into account read depth between interacting bins and genomic distance . PCA was conducted using Homer with the command ‘runHiCpca . pl -res 10000 -cpu 4 -genome mm9’ to generate genome wide eigenvalues at 10 kb resolution . The values changed marginally at 20 , 40 , or 50 kb , but the sign of the eigenvalue was unaffected , that is , there was no impact on whether a TAD was assigned as A compartment or B compartment . Published TAD coordinates in mouse liver ( Vietri Rudan et al . , 2015 ) were converted from mouse genome mm10 to mm9 using liftover with default parameters . Each TAD was then divided into 100 equal-sized bins using the Bedtools command makewindows . Next , these bins were compared to the peak positions of various publicly available ChIP-seq datasets using the Bedtools coverage command , and the number of peaks per bin was counted . This resulted in a string of 100 values for each TAD , representing the number of ChIP-seq peaks per bin , where the first value is the start of the TAD and the last value is the end of the TAD . Conducting this analysis across all TADs yielded a matrix of 3617 rows ( one per TAD ) x 100 columns ( one per bin ) . To generate the aggregate profiles shown in Figure 1A–1E , and in Figure 1—figure supplement 1B–E , the sum of each column was taken and then normalized to account for differences in total peak count for the different samples , factors , and chromatin marks analyzed . Normalization was conducted by taking the average of the center five bins ( bins 48–52 ) and dividing the bin sums by this normalizing factor . This allows the y axis to represent bin enrichment relative to the center of the TAD , as shown . TAD boundaries were defined at single nucleotide resolution as the end of one TAD and the start of another ( as defined in [Vietri Rudan et al . , 2015] ) , thus excluding the start of the first TAD in each chromosome and the end of the last TAD . In contrast , all references to ‘TAD anchors’ refers to the CTCF sites most likely to be anchoring TAD loops based on distance from the boundary and proper orientation ( as described in TAD Anchor Identification , below ) . Data sources for all ChIP-seq , GRO-seq , Hi-C , and other datasets are described in Supplementary file 4 . H3K9me3 , H3K27me3 , H2AK5ac , and H3K36me3 marks were processed from the raw sequencing data ( fastq files ) through the standard ChIP-seq pipeline , above . H3K9me3 and H3K27me3 mark data were expressed as log2 ( ChIP/IgG signal ) . Lamina-associated domain coordinates and GRO-seq data were downloaded as pre-processed data . Heat maps were generated using Deeptools reference point , with a bin size of 10 kb . TAD boundaries were grouped according to k-means clustering ( k = 4 ) using signal within a 1 Mb window from three datasets: H3K9me3 , H2AK5ac , and the eigenvalue of the Hi-C PCA analysis ( above ) . Based on these clusters , TADs were classified as active , weak active , weak inactive , or inactive , as follows . If both the start and end boundary of a given TAD were classified as active , then the TAD was designated active . Specifically , a TAD was considered ‘active’ if the boundary at the start of a TAD fell into clusters 1 or 2 ( as marked in Figure 1F ) and the boundary at the end of the same TAD fell into clusters 1 or 3 . The corresponding metric was applied to identify inactive TADs . If the activity status of the start and the end of a given TAD were not in agreement , then the TAD was designated weakly active if the median Hi-C PC1 eigenvalue within the TAD was positive , or weakly inactive if the median Hi-C PC1 eigenvalue was negative . Gene expression and tissue specificity metrics represent expression or Tau values of genes whose TSS overlap active or inactive TADs . Contact profiles around TAD , intra-TAD loop , and non-loop-anchor CTCF sites were generated using Homer ( v4 . 9 ) using the command analyzeHiC and the options ‘-size 500000 -hist 5000’ to generate interaction profiles for 1 Mb windows around CTCF sites with 5 kb resolution . TAD and intra-TAD loop anchors were split into left and right anchors when found at the start and at the end of the predicted loop , respectively . Non-anchor CTCF sites were defined as other CTCF sites , based on the merged CTCF sample , that also contained a CTCF motif . Left and right groupings were determined based on the orientation of the strongest CTCF motif within the non-anchor peak regions . The inward bias index ( IBI ) was modified from the more genome-wide directionality index ( DI ) described in ( Dixon et al . , 2012 ) . Both DI and IBI use a chi-squared statistic to determine the extent to which Hi-C reads from a given region have a strong upstream or strong downstream bias . While DI is genome wide , IBI focuses on the directionality of cis interactions ( within 2 Mb ) from a 25 kb window immediately downstream of a CTCF peak relative to the motif orientation . A large positive value indicates a strong interaction bias towards the loop center , as the motif orientation would predict . Values close to zero indicate a roughly equal distribution of interactions upstream and downstream . By orienting the sign of the IBI value relative to the CTCF motif directionality , we were able to group left and right loop anchors together . Virtual 4C plots ( Figure 3B and Figure 3—figure supplement 1 ) and Hi-C screenshots ( Figure 2—figure supplement 5A–CF , and Figure 2—figure supplement 7A–C ) were generated using the 3D Genome Browser ( http://promoter . bx . psu . edu/hi-c/index . html ) . Virtual 4C plots used mESC Hi-C with 10 kb resolution and a 25 kb viewpoint for ±250 kb of the selected region . Screenshots were generated for mouse ( mESC , CH12 , and NPC ) and human ( GM12878 and K562 ) cells using raw signal and 10 kb resolution . All Hi-C datasets used were publicly available for mouse ( Rao et al . , 2014; Bonev et al . , 2017 ) and human cells ( Rao et al . , 2014 ) ( Supplementary file 4 ) . TAD anchors were predicted for mouse liver using a modified version of the intra-TAD loop prediction algorithm . The merged list of CTCF peaks was filtered to only consider peaks that were found across all four biological replicates , that contained CTCF motifs , and that were within 50 kb of a TAD boundary , as defined previously for mouse liver ( Vietri Rudan et al . , 2015 ) . This 50 kb distance was chosen based on the ambiguity of binned Hi-C data to more accurately determine the precise TAD boundary . Then , for each TAD boundary , all pairs of ( + ) and ( - ) CTCF peaks were considered and scored based on their combined distance to the called TAD boundary . Pairs of ‘+/-” CTCF peaks that were comprised of a ( + ) anchor upstream of a ( - ) anchor ( i . e . , CTCF peak pairs that were not divergently oriented ) were considered an invalid combination to define the end of one TAD and the beginning of the next TAD , and were not considered . The valid pairs with the shortest combined distance to the previously defined liver TAD boundary ( Vietri Rudan et al . , 2015 ) were retained and all others were removed . If no valid pair for a TAD boundary was identified , the single CTCF peak closest to the TSD boundary was retained as the TAD anchor . A complete listing of TAD anchors is found in Supplementary file 1C , and a listing of inter-TAD regions and associated gene ontology analysis is presented in Supplementary file 3 . We sought to compare the relative insulation of loops identified by our computational approach to alternative loops identified using the original core algorithm of ( Oti et al . , 2016 ) . This provides an objective measure to compare the performance of each computational method in identifying TAD-like loops and loop anchors within TADs . To this end , we used the complete mouse liver CTCF peak list from the merged CTCF sample as input and implemented the loop prediction algorithm exactly as described previously ( 60% proportional peak cutoff , CTCF signal +motif scores as above , retaining only loops < 200 kb ) ( Oti et al . , 2016 ) . As summarized in Figure 2—figure supplement 1B , this analysis yielded many more loops ( 60 , 678; '60 k loop set' ) than we obtained using our method ( 9543 intra-TAD loops ) . Furthermore , the loops in the 60 k loop set were shorter ( median size of 61 kb ) , and they showed less overlap with cohesin-mediated loops present in the mESC ChIA-PET dataset ( 25 . 5% versus 63 . 2% overlap for our set of intra-TAD loops ) . 59% of the intra-TAD loops characterized in our study are found in the 60 k loop set . To characterize loops unique to the 60 k loop set , we had to first filter out anchors found intra-TAD or TAD loops ( to insure that each list was mutually exclusive , as above ) . Any 60 k loop anchor within 50 kb of a TAD boundary was excluded from downstream analysis . We also excluded any 60 k loop with at least one anchor that coincided with an intra-TAD loop anchor . These mutually exclusive lists of intra-TAD loop anchors and the filtered set of 60 k loop anchors ( 25 , 983 loop anchors in total; representing a subset of ‘Non Anchor CTCF’ in the main text , and referred to as ‘26 k loop anchors’ in Figure 2—figure supplement 1C , D ) were then compared based on insulation of repressive histone marks ( see ‘Repressive histone mark insulation’ , below ) and Hi-C interaction profiles ( see ‘Additional Hi-C analysis’ , above ) . Figure 2—figure supplement 1B , C compares the insular features of intra-TAD loop anchors to those of the set of 26 k alternative loop anchors , which are not intra-TAD loop or TAD anchors . CTCF ChIP-seq data for 15 non-liver tissues from the ENCODE Project were downloaded ( https://genome . ucsc . edu/cgi-bin/hgTrackUi ? db=mm9&g=wgEncodeLicrTfbs ) and intersected with replicates to form a single peak list for each tissue ( Shen et al . , 2012 ) . These single peak lists per tissue were then compared to liver CTCF peaks using the Bedtools multiinter command with the –cluster option to generate a union CTCF peak list for all tissues with a score representing the number of tissues in which a peak is present . ‘Lone’ CTCF ( CTCF sites lacking cohesin bound ) , other/non-anchor CAC sites , TAD anchors , and intra-TAD loop anchors were compared to this list to generate the histograms in Figure 2C ( see also , Supplementary file 1C ) . Knockdown-resistant cohesin binding sites in liver were defined as Rad21 ChIP-seq peaks found in both wild-type ( WT ) and Rad21+/- mouse liver , with knockdown-sensitive sites defined as Rad21 peaks found in WT liver that are absent in Rad21+/- liver ( Faure et al . , 2012 ) . Similarly , knockdown-resistant cohesin binding sites in MEFs were defined as Smc1a ChIP-seq peaks present in both WT ( Kagey et al . , 2010 ) and Stag1-knockout MEFs ( Remeseiro et al . , 2012 ) . Knockdown-sensitive sites were defined as Smc1a peaks found in WT MEFs that are absent in Stag1-knockout MEFs . Phastcons 30-way vertebrate conservation scores were downloaded from the UCSC table browser and converted to BigWig tracks using ucscutils ( version 20130327; ftp://hgdownload . cse . ucsc . edu/goldenPath/mm9/phastCons30way/vertebrate ) . Comparisons to mESC Smc1a ChIA-PET and Smc1a Hi-ChIP datasets ( Dowen et al . , 2014; Mumbach et al . , 2016 ) were based on merged replicates , and reciprocal overlaps with intra-TAD loops were required ( Bedtools intersect –wa –u –r –f 0 . 8 –a intraTADloops . bed –b mESC . bed ) . The mESC Smc1 ChIA-PET dataset was filtered to define ‘CTCF-CTCF’ interactions as those with both anchor regions overlapping CTCF peak present in a minimum of 2 replicates ( total of 3 ) . Any remaining interactions were considered as CNC-mediated enhancer-promoter interactions for the analysis shown in Figure 2—figure supplement 4C . To determine if a TAD or intra-TAD loop anchor CAC showed more insulation , or less insulation , than other classes of CTCF or cohesin binding sites , we used Jensen Shannon Divergence ( JSD; [Fuglede and Topsoe , 2004] ) to quantify the insulation of H3K27me3 and H3K9me3 ChIP-seq signals . Specifically , regions 10 kb upstream and 10 kb downstream of each peak in a given peak list ( i . e . , TAD anchors , CNC , etc . ) were each divided into 50 bins of 200 bp each . The number of H3K27me3 , H3K9me3 , or IgG ChIP-seq reads within each bin was tallied , resulting in a vector of 50 + 50 values for each peak region . These were then compared to two test vectors representing complete ( maximal ) insulation: fifty 0’s followed by fifty 1’s , and fifty 1’s followed by fifty 0’s . These are theoretical representations of low signal upstream of the peak followed by high signal downstream , and vice versa . Using a custom python script ( Source code 1 ) , the similarity between the experimentally derived vector and each of the test vectors was calculated , where a lower value represents less divergence from the test vector . The cumulative frequency distribution per group ( anchors , CAC , CNC , etc . ) is presented for the most similar test vector per peak in Figure 3D and E ( K27me3 and K9me3 ) and Figure 3—figure supplement 2D ( IgG ) . Heat maps show ChIP signal Z-transformed data across all CTCF-bound regions . The ~70 , 000 open chromatin regions ( DHS ) previously identified in mouse liver ( Ling et al . , 2010 ) were classified based on ChIp-seq signals for H3K4me1 , H3K4me3 , and CTCF within 1 kb of each DHS summit , obtained using the refinepeak option in MACS2 . The general schematic is shown in Figure 4—figure supplement 1A . Promoter DHS were defined as DHS with a > 1 . 5 fold ratio of H3K4me3 relative to H3K4me1 ChIP-seq signal; enhancer DHS were defined as DHS with a < 0 . 67 fold ratio of H3K4me3 relative to H3K4me1 ChIP-seq signal , calculated as reads per million for each factor . Both DHS sets were filtered to remove DHS with <4 reads per million for both marks after subtracting IgG signal ( Figure 4A ) . These cutoff values leave two remaining DHS groups , one with a roughly equal ratio between the two histone-H3 marks , and one with low signal ( <4 reads per million ) for both marks . The former DHS were classified as weak promoter DHS , based on their close proximity to RefSeq TSS and the low expression of neighboring genes ( Figure 4—figure supplement 1B , C ) . The remaining DHS group , characterized by low ChIP-seq signals , was largely intergenic but showed weak to undetectable levels of canonical histone marks . Low signal regions that overlapped a CTCF site with higher CTCF ChIP-seq signals than H3K4me1 signals were classified as insulators ( Figure 4—figure supplement 1A ) . The remaining regions were designated weak enhancer-DHS based on their distance from TSS and their low levels of H3K27ac ChIP-seq signal compared to the enhancer-DHS group ( Figure 4—figure supplement 1B ) . The majority of promoter-DHS and weak promoter-DHS were <1 kb from a TSS ( Figure 4—figure supplement 1B ) . To compare the level of expression for genes with promoter-DHS or weak promoter-DHS ( Figure 4—figure supplement 1C ) , the TSS was required to be within 10 kb of the DHS summit . Any gene with both a weak promoter-DHS and a promoter-DHS within 10 kb was categorized as being regulated by a promoter-DHS; thus , there was no overlap between weak promoter-DHS-regulated genes and promoter-DHS-regulated genes . All available mouse tissue DNase-seq peak regions were downloaded from the ENCODE Project website ( https://www . encodeproject . org/ ) ( Shen et al . , 2012 ) . ENCODE mm9 blacklist regions ( https://sites . google . com/site/anshulkundaje/projects/blacklists ) were removed , and the lists were merged to form a single reproducible peak list for each tissue , as follows . Due to variable replicate numbers across tissues , the following cutoffs were used to form merged DHS lists for each tissue . If a tissue had only two replicates ( as was the case for 12 of the 20 non-liver tissues ) , we required that the DHS be present in both replicates . If a tissue had 3 or 4 replicates , then the DHS were required to be present in all or all but one replicate ( this was the case for 7 of the 20 non-liver tissues ) . For whole brain tissue , the merged peak list required that a DHS was present in at least 5 of the 7 replicates . These regions were compared to each other using the Bedtools multiinter command with the –cluster option to generate a union DHS peak list for all tissues , where the score column represents the number of non-liver tissues in which a given region was found . For all liver DHS assigned to one of the above five DHS classes ( Supplementary file 2A ) , each liver DHS summit was mapped to this all tissue union peak list , allowing only one match per summit up to 150 nt away . If a given liver DHS summit was >150 nt from the nearest DHS in any other tissue , it was given a score of ‘0’ for liver-specificity . Otherwise the score represents the number of mouse tissues that the closest DHS was found in . Super-enhancers were identified using the ROSE ( Ranked Order of Super Enhancers ) software package ( http://younglab . wi . mit . edu/super_enhancer_code . html ) . ROSE takes a list of enhancer regions and mapped read positions as input to identify highly active clusters of enhancers . Default options were used , including 12 . 5 kb as the maximum distance for grouping ( stitching ) enhancers into putative super-enhancers , as well as reads per million normalization for all H3K27ac ChIP signal used for ranking enhancer clusters . The set of all enhancer-DHS and weak enhancer-DHS regions from the five class DHS model described above ( Supplementary file 2A ) was used as the region input list . A set of 19 publicly available H3K27ac ChIP-seq datasets from mouse liver was used as signal input ( see Supplementary file 4 for sample information ) . This set includes datasets for male , female ( Sugathan and Waxman , 2013 ) , and circadian time course ( male only; [Koike et al . , 2012] ) mouse liver datasets . A strict intersection of super-enhancers identified across all 19 samples was used to define a set of 503 ‘core’ super-enhancers in mouse liver using the Bedtools multiintersect command , as shown in Figure 4—figure supplement 2A . Any enhancer cluster ( i . e . , constituent enhancers within 12 . 5 kb , as above ) not identified as a super-enhancer in any sample was termed a typical enhancer and considered as individual constituents only . Gene targets for enhancers were assigned as the nearest gene ( based on TSS position up to a maximum distance cutoff of 10 or 25 kb , as specified . Gene expression values and tissue specificity were defined as described above . Aggregate plots were generated using Deeptools ( version 2 . 3 . 3 ) . In Figure 4F , the scale-regions option of Deeptools was used to scale super-enhancers and typical enhancers to their median sizes of 44 kb and 1 kb , respectively . Figure 4—figure supplement 2B used the reference-point mode of Deeptools and shows GRO-seq signal that overlaps eRNA loci as defined previously ( Fang et al . , 2014 ) . Super-enhancer and typical enhancer coordinates for mESC and ProB cells are from ( Whyte et al . , 2013 ) . Data generated and used in this study has been deposited in the Gene Expression Omnibus ( GEO ) under accession number GSE102999 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE102999 ) . ChIP-seq data are available under the subseries GSE102997 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE102997 ) . 4C-seq data are available under the subseries GSE102998 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE102998 ) . Published datasets used in this study are listed in Supplementary file 4 . | The human genome contains the complete set of DNA instructions – including all genes – needed to build and maintain an organism . To fit all of this genetic information in the cell’s nucleus , the DNA is neatly wrapped around so-called histone proteins , which help to package the genetic material into chromatin , which forms thread-like structures , the chromosomes . Chromatin is further folded into large DNA loops held together by an anchor protein , CTCF , and by a second protein , cohesin , whose ring-shaped structure ties each loop at its base . DNA segments that are within the same loop may interact frequently , whereas those outside the loop rarely do . Many of these large DNA loops are further pinched off into sub-loops . These sub-loops may help a cell fine-tune whether a gene needs to be turned on or off by limiting the contact between genes and the DNA regions that regulate the activity of genes . Knowing where these DNA sub-loop are located is very important for understanding how each gene is controlled . However , this can be very costly to determine , and therefore , is only known for a few cell types . Now , Matthews and Waxman tackle this issue by creating a computer model that can correctly predict many of these sub-loops . The method used experimental data obtained from mouse liver cells to identify the locations of CTCF and cohesin . The results showed that DNA sub-loops in the liver cells can shield genes from regulatory DNA segments outside the looped area . For example , a small sub-loop that contains a single gene related to obesity is highly active , even though the large DNA loop containing the sub-loop is an otherwise inactive gene region . Similarly , certain genes critical for liver function are positioned within sub-loops containing DNA regions that greatly enhance the gene activity in liver cells . This allows the selected genes to be highly active – unlike other genes that are close by but outside the sub-loop . This new approach will make it easier and cheaper to discover DNA loops and sub-loops across the genome . A better knowledge of where these loops form may also allow us to better understand how genes are turned on and off in different types of cells , and in response to biological stimuli or environmental stresses . This may also help understand and treat conditions that arise from mutations that disrupt the boundaries of DNA loops or sub-loops , which can allow certain DNA segments to activate the wrong genes and can lead to developmental defects and diseases such as cancer . | [
"Abstract",
"Introduction",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"computational",
"and",
"systems",
"biology"
] | 2018 | Computational prediction of CTCF/cohesin-based intra-TAD loops that insulate chromatin contacts and gene expression in mouse liver |
Microglia play key roles in regulating synapse development and refinement in the developing brain , but it is unknown whether they are similarly involved during adult neurogenesis . By transiently depleting microglia from the healthy adult mouse brain , we show that microglia are necessary for the normal functional development of adult-born granule cells ( abGCs ) in the olfactory bulb . Microglial depletion reduces the odor responses of developing , but not preexisting GCs in vivo in both awake and anesthetized mice . Microglia preferentially target their motile processes to interact with mushroom spines on abGCs , and when microglia are absent , abGCs develop smaller spines and receive weaker excitatory synaptic inputs . These results suggest that microglia promote the development of excitatory synapses onto developing abGCs , which may impact the function of these cells in the olfactory circuit .
Microglia are critically important for normal brain development in the embryonic and early postnatal stages ( Hammond et al . , 2018 ) . Originally thought to be primarily involved in injury and disease , many recent studies have implicated microglia in diverse neurodevelopmental functions ( Tremblay et al . , 2011; Salter and Beggs , 2014; Wu et al . , 2015; Hong et al . , 2016 ) . However , much less is known about what role microglia might play in the healthy adult brain , even during the process of adult neurogenesis , which can be thought of as an extension of developmental processes throughout the lifespan . During early postnatal development , microglia have been implicated in the regulation of synaptic development , including activity-dependent synaptic pruning ( Stevens et al . , 2007; Schafer et al . , 2012; Tremblay et al . , 2010; Paolicelli et al . , 2011; Gunner et al . , 2019 ) on one hand and promotion of synaptic development and maturation on the other ( Hoshiko et al . , 2012; Zhan et al . , 2014; Miyamoto et al . , 2016; Nakayama et al . , 2018 ) . Although microglia seem well-positioned to perform similar roles to facilitate the integration of adult-born neurons into circuits in the adult brain in the dentate gyrus ( DG ) and olfactory bulb ( OB ) ( Ekdahl , 2012; Rodríguez-Iglesias et al . , 2019 ) , most studies on microglial regulation of adult neurogenesis to date have focused on early stages of the process occurring in the neurogenic niches . For example , hippocampal adult neurogenesis is impaired in models of neuroinflammation ( Monje et al . , 2003; Ekdahl et al . , 2003 ) and in immune-deficient mice ( Ziv et al . , 2006 ) . Microglia regulate adult neurogenesis in the subgranular zone of the DG through ongoing phagocytosis of apoptotic neuroblasts ( Sierra et al . , 2010 ) , although they do not seem to be similarly involved in the subventricular zone ( SVZ ) or rostral migratory stream ( RMS ) ( Kyle et al . , 2019; but see Ribeiro Xavier et al . , 2015 for counterevidence ) . Most of what is known about microglial involvement in later stages of neurogenesis is based on injury and disease models . Microglial activation via sensory deafferentation in the OB decreases the number of adult-born neurons ( Lazarini et al . , 2012 ) and their spine density ( Denizet et al . , 2017 ) . In addition , lipopolysaccharide injection or CX3CR1 knockout activates microglia in the hippocampus and alters both inhibitory ( Jakubs et al . , 2008 ) and excitatory ( Bolós et al . , 2018 ) synapses onto adult-born neurons in the DG . These studies suggest that microglia can modulate the synaptic integration of adult-born neurons under inflammatory conditions , but raise the question of whether they are similarly involved in the healthy adult brain . A recent study documented increased activity in principal neurons in the OB after microglial depletion ( Reshef et al . , 2017 ) . Here we investigate the cellular and circuit mechanisms behind this effect by analyzing the functional development of adult-born granule cells ( abGCs ) in the absence of microglia . We demonstrate that microglial depletion during , but not after abGC development reduces activity in abGCs that make inhibitory connections with principal cells in the OB . We show that microglia normally interact with spines in developing abGCs , and the volume of these spines is reduced when microglia are ablated . This is accompanied by a reduction in the amplitude of excitatory but not inhibitory inputs to abGCs , suggesting that microglia are essential for proper integration of abGCs in adult circuits .
We labeled cohorts of abGCs born using lentiviral injection into the RMS ( Consiglio et al . , 2004; Livneh and Mizrahi , 2012 ) of adult 8–12 week old mice . To visualize interactions between microglia and abGCs , we performed time-lapse in vivo two-photon imaging of the dendrites of dTomato-labeled abGCs in the external plexiform layer ( EPL ) of the OB over the first four weeks after injection in CX3CR1-GFP +/- mice , in which microglia are labeled with GFP ( Video 1 , Figure 1A ) . Consistent with previous observations ( Nimmerjahn et al . , 2005; Tremblay et al . , 2010 ) , we found that microglial processes were highly motile and occasionally appeared in close proximity to labeled dendritic spines ( Video 2 , Figure 1B ) . To quantify whether microglia preferentially interact with dendritic spines ( defined as colocalization of a microglial process with at least 5% of the area of a spine head , see Materials and methods , Analysis of microglia-spine interactions ) on abGCs compared to encountering them by chance during the course of continuous motility , we compared the frequency of interactions between microglial processes and spine heads in the actual imaging data with the frequency of interactions in a series of images in which the microglia channel was arbitrarily shifted with respect to the dendritic imaging channel ( ‘Offsets’ ) . Microglia exhibited an impressive degree of motility , interacting with 38 . 5% of abGC dendritic spines classified as ‘mushroom’ spines ( Figure 1C ) and 27 . 2% of spines classified as ‘filopodial’ spines ( Figure 1H ) during the course of our 30–90 min imaging sessions , which was not significantly different from the offset data ( p=0 . 13 and p=0 . 39 , respectively ) ( Figure 1D , I ) . However , we found that microglia interacted significantly more often with individual mushroom spines than predicted by chance ( Data: mean 0 . 15 ± 0 . 00039 interactions/10 min vs . Offsets: mean 0 . 12 ± 0 . 016 interactions/10 min , p=0 . 048 ) ( Figure 1E ) though the length of individual interactions was not significantly greater ( p=0 . 96 ) ( Figure 1F ) . In contrast , microglia did not interact with filopodial spines at levels above chance ( Figure 1I–L ) . Microglia did not cover significantly more of the spine than predicted by chance during interactions with either spine type ( Mushroom: p=0 . 72 , Filopodial: p=0 . 84 ) ( Figure 1G , L ) . These results suggest that microglia specifically interact with spines that likely contain functional synapses ( Whitman and Greer , 2007 ) , positioning them to influence synaptic stabilization and maturation during the early development of abGCs . To assess whether microglial functions are essential for the development of abGCs , we ablated microglia during the entire time course of abGC development , beginning three weeks before lentiviral labeling ( Figure 2A ) . Microglial depletion using the CSF1R inhibitor PLX5622 ( hereafter PLX ) formulated in chow as previously described ( Elmore et al . , 2014 ) efficiently ablated microglia from the OB ( 85% ablation as assessed by immunostaining , 96% ablation as assessed by flow cytometry ) within one week and depletion could be maintained at similar levels for up to nine weeks with ongoing delivery ( Figure 2B , Figure 2—figure supplements 1 and 2 ) . We found no evidence of any largescale inflammatory response to microglial depletion , as assessed by immunostaining of glial fibrillary acidic protein ( GFAP ) ( Figure 2—figure supplement 3 ) , consistent with previous reports of the effects of PLX on the whole brain ( Elmore et al . , 2014 ) and the olfactory bulb ( Reshef et al . , 2017 ) . At five to six weeks post injection , when abGCs have reached a functionally mature state ( Wallace et al . , 2017 ) , we used two-photon imaging to visualize abGC dendrites in vivo . Microglial depletion did not affect the overall number of adult-born neurons in the OB ( Figure 2—figure supplement 4 ) , consistent with other reports , ( Reshef et al . , 2017; Kyle et al . , 2019 ) , and we could readily identify dTomato-labeled abGC dendrites in control and PLX-treated mice . Since GCs in the OB are axonless and their release sites are located at dendodendritic synapses on spines in the EPL ( Rall et al . , 1966 ) , we chose to record calcium responses in these dendrites . We first recorded responses in anesthetized mice to a panel of 15 monomolecular odors ( Materials and methods , Odor stimulation , Table 1 ) , while simultaneously imaging morphology in the dTomato channel to aid in region of interest identification and image alignment ( Figure 2C ) . AbGC responses to odors were sparse as previously described ( Figure 2D; Wallace et al . , 2017 ) , but across the population we could identify dendrites responding to most of the odors in our panel ( Figure 2E ) . We characterized responses by taking the mean ΔF/Fσ value over a five-second period following the onset of a two-second odor stimulus and plotted the cumulative distribution of dendritic responses across all odors . The distribution was shifted left toward lower responsiveness in PLX-treated mice ( p=2 . 56e-08 ) while the noise distributions constructed from blank trials were not different ( p=0 . 96 ) ( Figure 2F ) . Dendrites in PLX-treated mice also responded to fewer odors ( for threshold response criteria , see Materials and methods , In vivo imaging analysis , Thresholds ) ( median ( IQR ) : Control: 3 ( 1–6 ) , PLX: 1 ( 0–4 ) , p=1 . 16e-04 ) ( Figure 2G ) . We also found that lifetime sparseness ( Willmore and Tolhurst , 2001 ) , ( bounded between 0 and 1 , a low score indicates a sparser representation ) was lower in dendrites in PLX-treated mice ( median ( IQR ) : Control: 0 . 18 ( 0 . 067–0 . 32 ) , PLX: 0 . 067 ( 0–0 . 25 ) , p=4 . 18e-04 ) ( Figure 2H ) . These effects were also significant when we performed hierarchical bootstrapping ( Saravanan et al . , 2019 ) to take into account the fact that we imaged dozens of dendrites from each mouse , with dendrites in PLX-treated mice responding to fewer odors ( mean Control: 3 . 6 ± 0 . 38 , PLX: 2 . 8 ± 0 . 26 , p=0 . 050 ) ( Figure 2—figure supplement 5B ) and having lower median response amplitudes ( median Control: 0 . 14 ± 0 . 039 , PLX: 0 . 070 ± 0 . 016 , p=0 . 0011 ) ( Figure 2—figure supplement 5D ) . However , there was no difference in the median amplitude of responses above threshold ( p=0 . 77 ) ( Figure 2—figure supplement 5F ) , suggesting that the difference in overall responses was mostly mediated by an increase in the proportion of dendrites that did not respond to any of the odors in our panel , which was significantly higher in PLX-treated mice ( Control: 0 . 14 ± 0 . 039 , PLX: 0 . 32 ± 0 . 039 , p=0 . 0013 ) ( Figure 2—figure supplement 5E ) . While imaging in anesthetized mice allows better control of breathing rate , brain motion , and possible motivational influences on brain state , granule cell odor representation is significantly different in awake mice ( Kato et al . , 2012; Wienisch and Murthy , 2016; Wallace et al . , 2017 ) . Therefore , we also imaged abGC dendrites in awake mice and found similar effects as in anesthetized mice ( Figure 3A , Figure 3—figure supplement 1 ) . Dendrites in PLX-treated mice had lower responsiveness ( p=0 . 037 ) ( Figure 3B ) , responding to a lower median number of odors ( median ( IQR ) Control: 1 ( 0–6 ) , PLX: 0 ( 0–3 ) , p=0 . 052 ) ( Figure 3C ) and having lower lifetime sparseness ( median ( IQR ) Control: 0 . 067 ( 0–0 . 28 ) , PLX: 0 ( 0–0 . 16 ) , p=0 . 056 ) ( Figure 3D ) . Interestingly , while response time courses were similar between control and PLX-treated mice in the anesthetized state ( p=0 . 30 ) , allowing us to characterize responses with a simple mean across the odor analysis period , principal components analysis of the ΔF/Fσ traces for all cells’ responses to all odors revealed different response time courses in awake mice ( p=0 . 004 ) ( Figure 3—figure supplement 1 ) . These differences were likely not due to changes in active sampling of odors since sniffing rates were not different during baseline or odor presentation periods ( mean Baseline: Control = 3 . 49 Hz , PLX = 3 . 52 Hz , p=0 . 91; Odor: Control = 4 . 00 Hz , PLX = 3 . 94 Hz , p=0 . 81 ) ( Figure 3—figure supplement 2 ) . To ensure that our analysis of response amplitudes was not complicated by this possible change in response timing , we also applied an event detection analysis method ( using a sliding window across the response period to detect increases in fluorescence above a noise threshold , see Materials and methods , In vivo Imaging analysis , Event detection ) to the awake data and found similar results with dendrites in PLX mice still characterized by responses to a lower median number of odors ( p=0 . 037 ) ( Figure 3—figure supplement 1 ) . We next wondered whether the effect of microglial depletion was specific to abGCs developing in the absence of microglia or whether it might affect abGCs more generally . To address this question , we modified our experimental timeline to label abGCs and wait three months for them to mature fully ( Figure 4A ) before imaging their responses to the same set of odors ( Figure 4B ) in the same mice before and after three weeks of PLX5622 . In this case , we found no significant differences in the distribution of responses ( p=0 . 45 ) ( Figure 4C ) , median number of odors evoking a significant response ( median ( IQR ) Control: 1 . 5 ( 0–6 ) , PLX: 2 ( 0–5 ) , p=1 . 00 ) ( Figure 4D ) , or lifetime sparseness ( median ( IQR ) Control: 0 . 096 ( 0–0 . 31 ) , PLX: 0 . 099 ( 0–0 . 27 ) , p=0 . 86 ) ( Figure 4E ) . We verified that our imaging paradigm was stable since there was also no change in responses when we imaged the same mice for two sessions three weeks apart without any PLX treatment ( Figure 4—figure supplement 1 ) . Even after 9 weeks of PLX treatment , the level of responsiveness remained stable in mature abGCs ( Figure 4—figure supplement 2 ) . Since we found that microglial depletion reduces the functional responses of abGCs , we wondered if there were accompanying changes in excitatory synapses made on abGCs . We studied spines on the apical dendrites of abGCs in the EPL since our in vivo imaging showed lower calcium responses to odors in these dendrites , which could reflect fewer or weaker synaptic inputs . Four weeks after lentiviral labeling , we examined spines on apical dendrites in abGCs in tissue sections from control and PLX-treated mice ( Figure 5A , B ) . We found higher spine density ( median ( IQR ) : Control: 0 . 30 ( 0 . 21–0 . 36 ) spines/µm , PLX: 0 . 40 ( 0 . 31–0 . 48 ) spines/µm , p=0 . 039 ) ( Figure 5C ) , but smaller spine head volumes in PLX-treated mice ( mean Control: 0 . 37 ± 0 . 03 μm3 , PLX: 0 . 31 ± 0 . 02 μm3 , p=0 . 057 ) ( Figure 5D ) . Furthermore , most of the increase in spine density in PLX-treated mice was due to an increase in the density of filopodial spines that had similar spine head volumes as in control mice ( Figure 5—figure supplement 1 ) . In contrast , mushroom spines had significantly smaller head volumes in PLX-treated mice ( mean Control: 0 . 45 ± 0 . 04 μm3 , PLX: 0 . 36 ± 0 . 03 μm3 , p=0 . 036 ) ( Figure 5—figure supplement 1 ) . We next investigated the electrophysiological correlates of the observed differences in spine head size by recording spontaneous excitatory postsynaptic currents ( sEPSCs ) in abGCs . We used the same timeline as our in vivo imaging experiments , namely microglial depletion beginning three weeks before lentiviral labeling , continuing until electrophysiological recordings from labeled cells in brain slices at five to six weeks post injection ( Figure 6A , B ) . We found a similar frequency of sEPSCs in cells from control and PLX-treated mice ( mean Control: 30 . 8 ± 2 . 2 Hz , PLX: 30 . 6 ± 2 . 4 Hz , p=0 . 48 ) ( Figure 6C ) , but their amplitude was reduced ( mean Control: 10 . 8 ± 0 . 6 pA , PLX: 9 . 0 ± 0 . 4 pA , p=0 . 007 ) ( Figure 6D ) . Passive membrane properties including membrane resistance ( median Control: 597 MΩ , PLX: 532 MΩ , p=0 . 31 ) and capacitance ( median Control: 14 . 2 pF , PLX: 13 . 6 pF , p=0 . 61 ) were unchanged , signifying no differences in cell surface area or resting membrane properties ( Figure 6—figure supplement 1 ) . We also confirmed that our recording conditions were consistent by verifying that series resistance and the distance of the recorded cells from the mitral cell layer were not different between groups ( Figure 6—figure supplement 1 ) . To check whether there might be accompanying changes in inhibition that could offset or augment the observed changes in excitation , we also recorded spontaneous inhibitory postsynaptic currents ( sIPSCs ) in the same cells ( Figure 6E ) . We found no difference in the frequency ( p=0 . 77 ) ( Figure 6F ) or amplitude of sIPSCs ( p=0 . 79 ) ( Figure 6G ) , suggesting that abGCs that mature in the absence of microglia receive weaker excitatory inputs without noticeable accompanying changes in inhibition . Since there was no significant change in functional responses in abGCs that matured before microglia ablation , we checked whether synaptic inputs were also unchanged in this condition using the same experimental timeline as before and recording sEPSCs in abGCs that experienced three weeks of microglial depletion after three months of maturation ( Figure 7A ) . There was no significant change in the frequency ( p=0 . 76 ) ( Figure 7B ) or amplitude of sEPSCs ( p=0 . 09 ) ( Figure 7C ) . Inhibitory inputs were also unchanged ( Figure 7—figure supplement 1 ) . These results suggest that microglial depletion only affects synaptic inputs to abGCs when it occurs during the first five to six weeks of the cells’ development rather than after maturation .
Several technical caveats must be acknowledged to allow proper interpretation of our study . While ablation of microglia with PLX5622 is highly efficient , it necessarily involves a period of microglial cell death , and it is unknown how this debris may be cleared or what effects it may have on other cell types . We have attempted to minimize these effects by beginning experiments at least three weeks after initiation of PLX5622 treatment , when the number of microglia should have reached steady state levels ( Elmore et al . , 2014 ) , but we cannot exclude the possibility of long-term effects resulting from microglial cell death . Encouragingly , there seems to be no upregulation of cytokines or GFAP with PLX treatment generally ( Elmore et al . , 2014 ) or in the OB , specifically ( Figure 2—figure supplement 3; Reshef et al . , 2017 ) in contrast to other methods for microglial depletion , which may cause a significant increase in GFAP+ astrocytes and a cytokine storm ( Bruttger et al . , 2015 ) . However , we acknowledge that the array of possible inflammatory responses to microglial disruption is still poorly understood ( Liddelow and Barres , 2017 ) and may involve both responses to microglial cell death as well as reactions to the absence of normal homeostatic cues . Future work will be needed to characterize changes in crosstalk among different cell populations after microglial depletion and implications for the inflammatory environment . However , inflammation in the OB has been shown to generally reduce adult-born neuron numbers ( Lazarini et al . , 2012 ) and spine density ( Denizet et al . , 2017 ) as opposed to increasing spine density and reducing excitatory synapse strength as we show here , so it is unlikely that undetected inflammation relating to microglial depletion can fully account for the results we observe . We also note that microglial depletion with PLX5622 is brain-wide . AbGCs receive local inputs from within the OB as well as feedback from other cortical areas and neuromodulatory inputs ( Lepousez et al . , 2013 ) , so we cannot unambiguously attribute the effects on abGCs to changes within the OB . Future work will be necessary to investigate whether there is a change in the balance of distal ( predominantly feedforward ) and proximal ( mostly feedback ) inputs or instead a more general effect on excitatory synapse maturation and/or maintenance . AbGCs become responsive to odor stimuli soon after they arrive in the OB and then undergo a period of functional refinement during which their initially broadly tuned responses become more selective ( Wallace et al . , 2017 ) ( although there may also be a subpopulation of abGCs that shows broadening of responses with maturation Quast et al . , 2017; Wallace et al . , 2017 ) . Although the mechanism behind this increase in stimulus selectivity is not well-understood , it may involve both a decrease in some aspects of excitability ( Carleton et al . , 2003; Nissant et al . , 2009 ) , especially dendritic excitability ( Wallace et al . , 2017 ) , as well as selective reorganization of synaptic inputs such that mature abGCs become more responsive to a particular glomerular module at the expense of other inputs . We propose that abGCs in PLX-treated mice experience a normal drop in excitability ( likely regulated by cell-intrinsic mechanisms ) without a concomitant selective strengthening of specific synaptic inputs , leading to sparser odor responses ( Figure 8 ) . This is consistent with the higher density of more filopodial-like spines in PLX-treated mice ( characteristic of abGCs at an earlier stage of maturation Breton-Provencher et al . , 2014 ) along with smaller mushroom spines ( Figure 5 ) and lower amplitude sEPSCs ( Figure 6 ) . It is also possible that abGCs in PLX-treated mice could have generally lower excitability , but we regard this as unlikely since excitability has been mainly shown to affect survival rather than synaptic inputs in abGCs ( Lin et al . , 2010 ) , in contrast to what we observe here . Future work will be needed to disentangle the different aspects of abGC maturation and more fully elucidate which are affected by microglia . Though the effects on synaptic development that we observe in individual cells are modest ( ~15% reduction in median sEPSC amplitude in PLX-treated mice , Figure 6C ) , these effects combined nonlinearly across many synapses likely become important near response threshold in vivo . Indeed , we observed larger effects on calcium responses recorded in vivo ( median response amplitudes halved across all abGCs imaged in each PLX-treated animal , Figure 2—figure supplement 5D ) , which likely represent dendritic calcium spikes or global calcium events accompanying somatic action potentials that result from simultaneous activation of many feedforward and feedback synapses ( Egger et al . , 2005; Egger , 2008 ) . Though several studies have examined microglial motility ( Nimmerjahn et al . , 2005 ) and interactions between microglia and neuronal elements ( Wake et al . , 2009; Tremblay et al . , 2010; Sipe et al . , 2016 ) , it has been unclear whether microglia interact with synapses more often than would be expected by chance , given the dense synaptic milieu of the adult brain and the high degree of microglial process motility . Using automated methods to segment microglial processes from in vivo two-photon time-lapse imaging experiments and shuffle the resulting images for comparison , we demonstrated targeted motility towards mushroom spines on abGCs , which are likely to contain established excitatory synapses ( Whitman and Greer , 2007 ) . In the absence of microglia , we suggest that fewer spines attain mushroom morphology , leading to overall reduced spine volume and correspondingly weaker excitatory synapses . Concomitantly , we observed an increase in spine density in microglia-ablated animals , which is likely accounted for by filopodial spines which did not have functional or stable synapses ( since we did not see an accompanying increase in the frequency of synaptic currents in PLX-treated mice ) . One potential unifying hypothesis is that microglia may prune weaker synapses , allowing stronger synapses to strengthen further ( Stevens et al . , 2007; Schafer et al . , 2012; Tremblay et al . , 2010; Paolicelli et al . , 2011 ) . However , unlike other sensory systems that experience largescale pruning events during development , abGCs develop an increasing number of synapses over time ( Carleton et al . , 2003; Breton-Provencher et al . , 2014 ) . This does not preclude a role for microglia in pruning inputs to these cells ( in a situation in which synaptic formation occurs at a higher rate than elimination ) but suggests that if this occurs , it is likely to be more subtle and difficult to detect with conventional methods . Although we did not directly observe pruning of dendritic spines in abGCs in our time lapse imaging experiments , it remains possible that microglia could prune presynaptic elements ( Schafer et al . , 2012; Gunner et al . , 2019 ) ( which we did not image ) or that our frame rate was too slow to observe such events ( Weinhard et al . , 2018 ) . In addition to synaptic pruning , microglia may be involved in promoting synaptic strengthening . Deficits in microglial CX3CR1 have been linked to weaker synapses and impaired connectivity ( Zhan et al . , 2014; Reshef et al . , 2017 ) as well as delays in maturation of the AMPA/NMDA ratio ( Hoshiko et al . , 2012 ) . It is still unclear whether these effects , similar to those that we observe in the PLX ablation model , are related to differences in spine surveillance and direct microglial-neuron signaling or due to altered microglial behavior , including changes in release of moderating factors or interactions with other cell types , such as astrocytes . We expect that in the future , a new generation of tools for more specifically modifying microglial gene expression and function during defined time windows will help clarify the mechanisms by which microglia affect synaptic development . In addition , although most previous work has focused on microglial regulation of synapses , microglia may also regulate neuronal excitability through other mechanisms ( Li et al . , 2012; Peng et al . , 2019 ) . Accordingly , although we found no differences in passive membrane properties between abGCs in control versus PLX-treated animals , we cannot exclude additional possible effects of microglia on nonsynaptic aspects of abGC development , including active dendritic properties , that could also contribute to the reduced responsiveness we observe when microglia are ablated . While our results suggest that the effect of microglial depletion is specific to developing abGCs ( since our results with microglial depletion after abGC development did not reach statistical significance in most cases ) , we saw similar trends in reduced responses and excitatory inputs regardless of the timing of microglial depletion . This could be because lentiviral labeling is an imperfect method for isolating a single cohort of abGCs ( so some abGCs in the ‘after development group’ may still be at an earlier stage of development – see discussion of newcomer cells in Wallace et al . , 2017 ) . Another possibility is that microglial depletion also affects mature abGCs , although to a lesser extent than developing abGCs . This could be because microglia may be involved in the general maturation or strengthening of newly formed excitatory synapses in GCs , and even mature GCs have high rates of synapse formation ( and elimination ) compared to cells in other brain regions ( Mizrahi , 2007; Sailor et al . , 2016 ) . In this scheme , developing abGCs would demonstrate the most significant phenotype due to their higher rates of spine dynamics , but even mature GCs might accrue smaller effects over time . The larger effect of microglial depletion on developing rather than mature adult-born neurons highlights the role of microglia during developmental stages and may explain why some studies have not found significant effects of microglial depletion in other areas of the healthy adult brain ( Elmore et al . , 2014; Torres et al . , 2016 ) . The reduced activity that we observe in mature abGCs after microglial depletion is consistent with increased activity in the principal cells they inhibit ( Reshef et al . , 2017 ) . Since reduced inhibition from GCs to OB principal neurons has been directly linked to an increase in the time needed to discriminate odors in challenging olfactory tasks ( Abraham et al . , 2010 ) , this alteration in the OB circuitry could have functional consequences ( Egger and Urban , 2006 ) . Furthermore , given that abGCs may have an outsized role in the plasticity that underlies complex olfactory behaviors ( Breton-Provencher et al . , 2009; Jung et al . , 2000; Mandairon et al . , 2018 ) , microglial regulation of their development may contribute to ongoing plasticity in the olfactory system .
Mice were C57BL/6J males ( Jackson Laboratories ) or CX3CR1-GFP heterozygotes ( Jackson Laboratories stock #008451 ) ( Jung et al . , 2000 ) that were 8 to 12 weeks-old at the beginning of the experiment . Mice were singly housed after chronic cranial window implantation or housed with littermates for experiments that did not require an implant . In both cases , they were housed on a 12 hr reversed light/dark cycle after window implantation or viral injection . Littermates were randomly assigned to experimental groups ( control vs . PLX ) . All procedures were performed using approved protocols in accordance with institutional ( Harvard University Institutional Animal Care and Use Committee ) and national guidelines . For all experiments involving lentiviral labeling of abGCs , we used a Tet-Off lentiviral system ( Hioki et al . , 2009 ) with one construct expressing the transactivator tTAad under control of a synapsin promotor ( lenti-STB ) and a second construct expressing structural and/or activity markers . For spine quantification in fixed tissue ( Figure 5 ) and electrophysiology ( Figures 6 and 7 ) , we used lenti-STB and lenti-dTomato produced by the Boston Children’s hospital . For imaging microglia-spine interactions ( Figure 1 ) , we used lenti-STB and lenti-dTomato-t2A-GCaMP5 and for all experiments measuring odor responses ( Figures 2 , 3 and 4 ) we used lenti-STB and lenti-dTomato-t2A-GCaMP6s produced in house . VSV-G pseudotyped lentiviral vectors were produced by transfection of human embryonic kidney cells ( HEK293FT ) with third-generation lentivirus plasmids using lipofection ( Mirus TransIT−293 ) . Supernatant was collected 48 hr after transfection and concentrated using ultrafiltration ( Centricon Plus-20 PLGC centrifuge filter units ) . Reproduced from Wallace et al . ( 2017 ) : "Mice were anesthetized with an intraperitoneal injection of ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) and body temperature was maintained at 37°C by a heating pad . A small craniotomy was made bilaterally over the RMS injection sites ( coordinates from bregma: A +3 . 3 , L +0 . 82 , from the brain surface: V-2 . 9 and −2 . 7 ) and 250 nL of lentivirus ( 1:1 mixture of both constructs or 1:50 dilution of the tTA-containing construct to achieve sparser labeling for spine quantification in Figure 5 ) was injected at each of the two depths using a pulled glass micropipette ( tip diameter approximately 10–20 um ) . ' For cranial windows , the surface of the brain was kept moist with artificial cerebrospinal fluid ( 125mMNaCl , 5mMKCl , 10mMGlucose , 10mMHEPES , 2 mM CaCl2 and 2 mM MgSO4 [pH 7 . 4] ) and Gelfoam ( Patterson Veterinary ) and a glass window consisting of two 3 mm No . one coverslips ( Warner ) glued together with optical glue ( Norland Optical Adhesive 61 ) was implanted as previously described ( Adam and Mizrahi , 2011 ) . For mice used for awake imaging , Kwik Sil ( World Precision Instruments ) was placed between the coverslip and the brain surface to reduce movement . In this case , the coverslip consisted of two 3 mm and one 4 mm No . 0 coverslips forming a plug ( Dombeck and Tank , 2014 ) with the 4 mm coverslip cut with a diamond knife to fit between the mouse’s eyes . In both cases , the edges around the coverslip were sealed with Vetbond ( 3M ) and then C and B-Metabond dental cement ( Parkell , Inc ) . A custom-made titanium headplate ( eMachineShop ) was cemented to the skull . After surgery , mice were treated with carprofen ( 6 mg/kg ) every 24 hr and buprenorphine ( 0 . 1 mg/kg ) every 12 hr for 5 days . " A custom-built two-photon microscope ( Wienisch et al . , 2011 ) was used for in vivo imaging . Fluorophores were excited and imaged with a water immersion objective ( 20x , 0 . 95 NA , Olympus ) at 950 nm using a Ti:Sapphire laser ( Mai Tai HP , Spectra-Physics ) . The point spread function of the microscope was measured to be 0 . 66 × 0 . 66×2 . 26 µm . Image acquisition and scanning were controlled by custom-written software in Labview . Emitted light was routed through two dichroic mirrors ( 680dcxr , Chroma and FF555- Di02 , Semrock ) and collected by two photomultiplier tubes ( R3896 , Hamamatsu ) using filters in the 500–550 nm range ( green channel , FF01-525/50 , Semrock ) and 572–642 nm range ( red channel , FF01-607/70 , Semrock ) . Fields of view were 75 × 75 µm square spanning 800 × 800 pixels . Z-stacks of approximately 30 µm depth with a 1 µm z step for both channels ( 16 bit ) were taken every 3 min ( 0 . 5 Hz frame rate with 3x averaging during acquisition ) for periods of 30–90 min . Two or three fields of view were imaged in each mouse . Since both channels exhibited bleed-through with our imaging parameters , the ImageJ spectral unmixing plugin ( Author: J . Walter ) was used to calculate and apply unmixing matrices for each image stack prior to further analysis . In Fiji , spine heads were delineated manually for each time point in the frame where they appeared brightest using the oval or polygon tools and ROI Manager and classified as either mushroom ( spines whose spine head was wider than the spine neck at all timepoints ) or filopodial ( without a well-defined head ) . The Weka segmentation plugin ( Arganda-Carreras et al . , 2017 ) was used to perform binary segmentation of microglial processes from background after training on five frames ( that were fully segmented manually ) selected to represent a variety of microglial morphologies and brightness variation across the three mice . To optimize our resolution , we segmented very conservatively by using z stacks to mark microglial processes only in the plane where they appeared brightest . This strategy combined with only delineating spine heads in the brightest frame means that we only detected the closest interactions between microglial processes and spine heads . The features chosen for segmentation training in Weka were Gaussian blur , Sobel filter , Hessian , and Difference of Gaussians . This approach allowed us to segment complex microglia morphology from background automatically in every image frame and obviated the need for corrections for bleaching or variations in brightness across different imaging fields . Imaging frames that were too dim to segment ( usually due to loss of immersion water ) were excluded . Each segmented image stack was checked manually to ensure that any residual bleed-through from the red channel did not appear in the segmentation . After segmentation , ROIs representing spine heads were exported to Matlab using a custom-written macro with the command getSelectionCoordinates . In Matlab , the ROIs were loaded onto the segmented image and the mean value of the binary microglia channel within each ROI at each timepoint ( 0 if there was no colocalization or up to one if all pixels were colocalized with a segmented microglial process ) was measured . The frame was quantified as containing a microglia-spine interaction if the value of colocalization was greater than 0 . 05 ( at least 5% of the spine head area overlapped by a microglial process ) . To compare the overlap between microglial processes and spine heads in the real data compared to what might be expected by chance , we iteratively translated the microglia channel relative to the marked spine head ROIs to calculate distributions for all the measured interaction parameters ( Dunn et al . , 2011 ) . To do this , we took the original segmented image stack and translated it horizontally , vertically , or horizontally and vertically first by the maximum spine width across the whole data set ( ~32 pixels ) to ensure none of the offsets would overlap the real data and then iteratively by the mean spine width ( ~10 pixels ) to ensure each offset would be as uncorrelated as possible for a total of 231 offsets . Note that this method likely underestimates the significance of interactions in the real dataset because microglia cell bodies were never colocalized with spines in the real dataset ( dendrites overlapping microglia cell bodies were not chosen for imaging ) but were likely colocalized and quantified as interacting with spines in some of the offset datasets . To produce Video 2 , frames were first registered with the MultiStackReg plugin ( Author: Brad Busse ) based on the magenta channel and bleaching was corrected with histogram matching in ImageJ ( these steps were not necessary for analysis because we segmented each image frame separately as described above ) . Animals were matched in littermate pairs before cranial window surgery . All animals with a clear region of the cranial window and visible lentiviral expression were used for imaging ( 2/4 control mice and 2/4 PLX-treated mice in the first experiment , 1/2 control mice and 1/2 PLX mice in a second experiment , and 2/4 control mice and 4/4 PLX-treated mice in a third experiment ) . Imaging was performed at 930 nm with the same two-photon microscope described above . Reproduced from our previous work ( Wallace et al . , 2017 ) : "Animals were anesthetized with an intraperitoneal injection of ketamine and xylazine ( 90% of dose used for surgery ) and body temperature was maintained at 37°C by a heating pad . Frame rates were 4 Hz , the pixel size was 0 . 5 μm , and fields of view measured 150 × 150 μm . To locate regions for imaging , a low magnification z stack ( ~300–500 μm square ) at slow scanning speed ( usually 0 . 5 Hz ) with a 1–2 μm z step was taken from the surface of the dura to the granule cell layer . Planes with many dendrites perpendicular to the imaging axis were chosen for imaging during odor stimulation . " A subset of the animals that were imaged under anesthesia were chosen for awake imaging ( all animals that had sufficiently stable cranial windows were chosen from each of the experiments ) . Animals were water-restricted beginning 1–2 days before being handled and accustomed to head-fixation in a restraining tube ( Guo et al . , 2014 ) for 1–2 days with manual delivery of water rewards ( approximately 30 min sessions each ) . They were then acclimated to the sound of the scan mirrors and odor delivery ( using the full odor set ) on the day before imaging with manual delivery of water rewards before imaging and periodically between sets of repetitions . The same protocol was repeated for 1 or 2 days of imaging for each mouse . Odor lists are found in Table 1 . Odors ( Sigma ) were delivered with a custom-built 16 channel olfactometer at a nominal volumetric concentration of 16% ( v/v ) in mineral oil and further diluted by 16 times in air to a final concentration of approximately 1% ( except for isoeugenol which was not diluted in mineral oil and therefore had a final concentration of approximately 6 . 25% ) . Odors were presented for 2 s with an interstimulus interval of 40 s with 3–5 times repetitions . The order of odor delivery was not randomized . A ‘no odor’ trial with the same parameters but in which no odor valve opened was included with each set of repetitions . Odors were delivered through a mask with balanced input and output air flow that also allowed us to record respiration ( Grimaud and Murthy , 2018 ) . The positioning of the mask was adjusted daily to ensure optimal signal to noise . A photoionization detector ( miniPID , Aurora Scientific ) was used to confirm that odor concentrations were consistent between trials with these parameters . Odors were replaced before each set of experiments . Data were analyzed offline using custom-written scripts in MATLAB ( Mathworks ) . Experimenters were blind to fluorescence changes during data analysis but not to experimental group . Dendritic ROIs from abGCs were chosen based on average intensity projections in the dTomato channel as previously described ( Wallace et al . , 2017 ) . Fields of view were non-overlapping and separated by at least 100 μm to minimize the chance of the same dendrites appearing in multiple fields of view . Z stacks of each imaging region were taken with a 2 μm z step from the surface of the dura down to the convergence of GC dendrites into a single apical dendrite . The density of labeling precluded tracing of all dendrites back to their parent cell , but we used these z stacks to ensure that multiple ROIs were not chosen from the same dendrite . Therefore , these data should be considered a sample from a population of dendrites rather than cells since some dendrites may have originated from the same cell . We previously estimated with our labeling and imaging strategy that about 1 . 4 times more dendrites than cells are represented in the dataset ( and this is almost certainly an overestimate since this was based on the first week of imaging , where labeling is significantly sparser than at later timepoints ) . For ablation after development ( Figure 4 ) , the same fields of view were imaged before and after PLX treatment . For Figure 4—figure supplement 2 , we chose matching ROIs for the two sessions ( any ROIs without similar morphology between the two sessions were excluded as described previously Wallace et al . , 2017 ) to quantify how much our imaging conditions changed between sessions without PLX treatment . However , we opted to choose ROIs independently for Figure 4 since matching ROIs always results in the exclusion of many ROIs , reducing sample size . To correct for fast lateral motion and image drift , all image frames for a given field of view were aligned to the average of the first trial using cross-correlation based on rigid body translation ( ImageJ plugin Moco ) ( Dubbs et al . , 2016 ) . Frames with out-of-frame motion were removed based on the cosine similarity between each frame and the average intensity projection of the first trial ( or the user-determined trial with the least motion for awake imaging ) . Image frames with cosine similarity that differed by more than 25% ( 30% for awake imaging ) from the mean value for the user-determined best trial for that field of view or more than 20% ( 15% for awake imaging ) from the mean of the baseline period for that trial were discarded . In some trials , immersion water dried up or laser power fluctuated , so trials were removed if their average brightness was less than half of the brightness of the average of the first three trials or if difference in brightness between the baseline and odor periods was greater than three times the standard deviation of brightness in the first trial . The entire trial was removed if , after these corrections , it contained less than 75% of the original frames during either the baseline or odor analysis period . The average intensity in the GCaMP channel was calculated for each ROI , for each frame and for each odor . A response value for each cell-odor pair was calculated as the average ΔF/Fσ value over the odor analysis period ( 5 s following odor onset ) where Fσ represents the standard deviation of fluorescence during the baseline period . We used Fσ because we found that in many cases the baseline GCaMP6s fluorescence was so low in abGC dendrites that we could not reliably subtract the background as described previously ( Wallace et al . , 2017 ) . Bleaching was corrected by fitting a single exponential to the florescence during the baseline period and taking the value at the end of the baseline period as the baseline mean , only for ROIs where the fluorescence during the last 2 . 5 s of the baseline period was greater than 1 . 1 times the fluorescence during the first 2 . 5 s . If , after correction for bleaching , baseline Fσ was greater than 30% of the mean baseline fluorescence , that ROI was considered too noisy and was removed from the analysis . Events were detected separately in each trace using the calculated ROC threshold ( see next section ) and any frames that were included in an event in any repeat were averaged across all repeats to obtain a mean event trace and the mean value in a 1 s period around the peak in this mean trace was then calculated as the average response to that odor . For latency , the mean latency across repeats was calculated for all repeats that had detected events . For all figures where a threshold was applied to the data , thresholds were calculated based on the distribution of ‘no odor’ trials . An area under the receiver operating curve analysis was performed and the lowest threshold yielding a 10% false positive rate was chosen . Thresholds were calculated for each figure by combining responses from both control and PLX-treated groups and performing ROC analysis on the combined data . For event detection , we used ROC analysis to find the optimal combined threshold for the number of frames and standard deviation above baseline and chose the threshold that gave closest to a 10% false positive rate . After applying a threshold to the data , we used the following equation to calculate lifetime sparseness: ( Willmore and Tolhurst , 2001 ) LS= ∑j=1mrjm2∑j=1mrj2mwhere m = number of odors , rj = response of the neuron to odor j . If all stimuli activate a cell nearly uniformly , LS will be close to 1 , and if only a small fraction of the stimuli activate a cell significantly , LS will be close to 0 . For any cells with all responses below threshold , we set LS = 0 , interpreting this as the sparsest possible representation . Principal component analysis of the time course of responses was performed in Matlab using centered data and singular value decomposition as described previously ( Wienisch and Murthy , 2016 ) . To compare time courses for the control and PLX-treated groups , principal components were calculated on each dataset ( all traces from all cell-odor pairs ) separately and the angle between the two spaces spanned by the coefficient vectors for the first three principal components from the beginning to the end of the odor analysis period was calculated . Then a permutation test was performed in which the group to which each trace belonged was shuffled 1000 times , and the angles between new coefficient vectors were calculated based on random division into two groups of the same size as the original datasets . The actual angle was then compared to this distribution to obtain a p value . Peaks were extracted from respiration traces using the findPeaks function in Matlab with a minimum peak distance of 10 Hz . Raincloud plots of the type in Figure 2G were created using the Matlab version of the RainCloudPlots package ( Allen et al . , 2018 ) . CSF1R inhibitor PLX5622 was generously provided by Plexxikon ( Berkeley , CA ) and mixed into standard rodent diet at 1200 mg per kilogram of chow ( Research Diets: AIN-76A diet ) . Control diet was formulated identically , but without the inhibitor . Microglial depletion was confirmed via flow cytometry . A single-cell suspension enriched for microglia was generated as previously described ( Hammond et al . , Immunity 2019 ) . Briefly , mice were deeply anesthetized with a mixture of ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) and transcardially perfused with 20 mL of cold Hank’s balanced salt solution ( HBSS , GIBCO , 14175–079 ) . Bulbs and brains were minced using a razor blade ( Electron Microscopy Science , 71960 ) and homogenized using a dounce tissue grinder ( Wheaton , 357542 ) . Microglia were enriched via centrifugation in 40% Percoll ( Sigma-Aldrich , 17-0891-01 ) at 500 g for 1 hr at 4°C . Samples were incubated for 20 min with Blue Dead Cell Stain ( Thermo Fisher , L34961 ) and Fc blocking antibody ( Rat Anti-Mouse CD16/CD32 , BD Bioscience , 553141 ) in HBSS + 2 mM EDTA . Cells were additionally stained for 20 min with antibodies against CD45 ( Biolegend , 103116 ) and CD11b ( Biolegend , 101217 ) in buffer ( HBSS + 2 mM EDTA + 0 . 5% BSA ) . Counting Beads ( CountBright , Thermo Fisher Scientific , C36950 ) were added and samples analyzed using a FACS Aria II ‘SORP’ . All events were collected until a total of 8000 counting beads had been acquired for each sample . The data were analyzed in FlowJo 10 . 2 . Mice received two intraperitoneal injections of BrdU ( Sigma , 100 mg/kg in 0 . 9% saline ) 12 hr apart . Mice were deeply anesthetized with a ketamine/xylazine mixture and perfused transcardially with 20 mL of PBS ( pH 7 . 4 ) first , followed by 30–50 mL of 4% paraformaldehyde ( diluted from 16% stock , Electron Microscopy Sciences ) in 0 . 1 M phosphate buffered saline ( pH 7 . 4 ) . Brains were removed from the skull and placed in 5 ml 4% paraformaldehyde for 2 hr . They were then rinsed with PBS and one hemisphere for each mouse was sliced coronally at 100 μm with a vibratome ( Leica ) for imaging of dTomato-labeled abGC spines ( Figure 5 ) , while the other hemisphere was sliced sagitally at 35–40 μm for immunostaining . For Iba-1 staining , 3–4 slices per mouse spanning the olfactory bulb were permeabilized and blocked with a solution containing 0 . 1% Triton X-100 ( Fisher ) , and 5% goat serum in PBS for 1 hr at room temperature or blocked with Starting Block ( ThermoFisher ) with 0 . 3% TritonX-100 for 1 hr at room temperature and then incubated overnight at 4°C with the primary antibodies rabbit anti-Iba-1 ( Wako: 019–19741 , RRID:AB_839504 ) at 1:500 or rabbit anti-GFAP ( Dako: Z0334 , RRID: AB_10013382 ) at 1:1000 and then secondary antibodies ( Alexa goat-647 anti-Rabbit ) for 2 hr at room temperature . For BrdU/NeuN staining , one of every eight slices per mouse was chosen . Slices were washed in PBS with 0 . 1% Triton X three times for five minutes each before being incubated for in 2N HCl for 10 min at room temperature and then 20 min at 37°C . They were then placed in 0 . 1M Boric Acid buffer for 15 min and washed again three times with PBS . All slices were then incubated in starting block ( ThermoFisher ) with 0 . 3% Triton X for one hour before being staining in in PBS with 0 . 3% Triton X with rat anti-BrdU ( Abcam: 6326 at 1:200 ) , and mouse anti-NeuN ( Millipore: MAB377 at 1:200 ) primary antibodies for 36–48 hr at 4°C and then secondary antibodies ( Alexa Fluor 488 and 594 at 1:200 ) for 2 hr at room temperature . Slices were treated with 0 . 2% w/v Sudan Black in 70% EtOH for 5 min before mounting . Slices were mounted with DAPI mounting media ( Vectashield DAPI ) and imaged with a confocal microscope ( LSM 880 , Zeiss ) . Reported cell densities were calculated based on distances in fixed tissue , uncorrected for volume changes due to fixation and mounting . All imaging and quantification were performed blind to the experimental group of the animal ( PLX-treated or control ) . For the 1- and 4 week timepoints , one z-stack per animal was imaged at 10X with pixel size 0 . 42 × 0 . 42×1 μm spanning the thickness of the slice . For the 9 week timepoint , two z-stacks per animal were imaged at 20X and the counts from both were averaged . Stacks were imaged with pixel size 0 . 59 × 0 . 59×1 μm spanning 10 μm and converted to maximum intensity projections . The polygon tool was used to outline the granule cell layer in each image and the area was measured . Iba-1 or GFAP positive cells were counted in this area manually using the Cell Counter plugin on maximum intensity projection images in ImageJ . Cells were counted only if the cell body was fully included in the image stack . For BrdU/NeuN , two z-stacks per OB ( one centered dorsally and one centered ventrally ) were taken at 20X with pixel size 0 . 52 × 0 . 52×0 . 89 μm spanning 9 . 8 μm . BrdU counts were performed using the automatic spots function in Imaris ( Bitplane ) with the same quality settings for spot detection for all images ( quality threshold 2370 , number of voxels threshold 524 ) . Cells were counted as positive if they were located in the granule cell layer and were also positive for NeuN . The area of the granule cell layer in each image was measured in ImageJ , and the density of BrdU/NeuN positive cells was calculated for each image and averaged for all images for each mouse . All images were 20–40 μm z-stacks with 0 . 42 z-step taken with a 63x oil immersion objective , and four fields of view were imaged per mouse . Care was taken to ensure minimal saturation ( less than 5% of pixels ) . Secondary and tertiary apical dendrites that were judged to be sufficiently bright , well-separated from adjacent or overlapping dendrites , at least 40 μm long , and extending at an angle from the imaging plane less than 45 degrees were chosen for tracing in Imaris . Each dendrite and all its dendritic protrusions less than 10 μm in length were manually traced using the Filaments function . Data from each dendrite were exported in a text file and imported into Matlab for plotting . For spine density , the Imaris property ‘Filament No . Spine Terminal Pts’ was used ( meaning that branched spines were counted by the number of spine heads rather than attachment points ) . For spine head volume , the Imaris property ‘Spine Part Volume Head’ was used . For the classification of filopodial and mushroom spines ( Figure 5—figure supplement 1 , we used the Imaris properties ‘HeadMaxDiameter’ and ‘NeckMeanDiameter’ and classified spines as filopodial if they had maximum head diameter less than 1 . 5 times the neck mean diameter . This threshold was chosen because it led to a similar ratio of mushroom to filopodial spines in controls as previously observed ( Breton-Provencher et al . , 2014 ) . Two experiments with two littermate pairs each ( control and PLX-treated ) were performed for electrophysiology . Mice were deeply anesthetized with a mixture of ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) and perfused with ice-cold modified ACSF solution ( in mM: 120 choline chloride , 25 glucose , 25 NaHCO3 , 2 . 5 KCl , 0 . 5 CaCl2 , 7 MgSO4 , 11 . 6 ascorbic acid , 3 . 1 pyruvic acid , 1 . 25 NaH2PO4 ) . Brains were removed and placed in the same ice-cold modified ACSF . Coronal slices ( 300 µm thick ) of olfactory bulbs were cut using a vibratome ( VT1000S; Leica , Germany ) . Slices were incubated in oxygenated holding solution ( in mM: 119 NaCl , 26 . 2 NaHCO3 , 1 NaH2PO4*H2O , 2 . 5 KCl , 22 glucose , 1 . 3 CaCl2 , 2 . 5 MgSO4 ) at 33°C for at least 30 min before being transferred to oxygenated ACSF ( in mM: 119 NaCl , 26 . 2 NaHCO3 , 1 NaH2PO4*H2O , 2 . 5 KCl , 22 glucose , 2 . 5 CaCl2 , 1 . 3 MgSO4 ) . Extra slices were maintained in holding solution at room temperature . Whole-cell recordings ( Bessel filtered at 2 . 2 kHz and acquired at 10 kHz ) were performed using patch pipettes filled with internal solution ( in mM: 130 cesium gluconate , 5 NaCl , 10 HEPES , 12 phosphocreatine di ( tris ) salt , 3 Mg*ATP , 0 . 2 Na*GTP , 1 EGTA , 2 . 0 Na2-ATP , 0 . 5 Na3-GTP , pH 7 . 3 ) using a Multiclamp 700B amplifier ( Molecular Devices , Palo Alto , CA ) at 36°C . Cells were visualized with dTomato and DIC with custom-built optics on a BX51WI microscope ( Olympus Optical , Tokyo , Japan ) and recorded with pClamp 10 . 3 ( Molecular Devices ) . Cell identity was confirmed by the presence of fluorescence material in the patch pipet after membrane rupture and/or cell fill with Alexa Fluor 488 , and only cells that had a proximal dendrite that extended from the soma in the direction of the EPL while remaining beneath the surface of the slice ( i . e . cells that had dendrites that did not appear to have been cut during slicing ) were targeted for patching . Patch pipets had 8–11 MΩ open tip resistance . Series resistance was not compensated . Cells were recorded in continuous 10 s sweeps for five minutes with a test pulse at the beginning of every sweep , which was used to calculate series resistance and holding current in Matlab for initial quality control . sEPSCs were recorded at −70 mV and sIPSCs were recorded at 0 mV . We waited at least one minute after breaking into the cell before beginning recording for sEPSCs and at least 30 s after switching the holding potential for sIPSCs; sEPSCs were always recorded first . Cells were recorded within 8 hr of slicing , and there did not appear to be any relationship between the time of recording and the frequency of synaptic events ( data not shown ) . Experimenters were not blind to experimental group during recording . Experimenters were blind to experimental group during analysis . Cells with an initial series resistance of <50 MΩ were used for analysis . Sweeps that deviated from the average series resistance across the first three sweeps by more than 25% or had a holding current of more than 100 pA at −70 mV were excluded . The cell was excluded entirely if less than half of the recording sweeps remained after quality control . For each cell , separate test pulses ( −10 mV , 20 ms ) with 50 repetitions at 20 kHz were recorded before and after each set of recordings , and these files were used to calculate series resistance ( based on the maximum current recorded at the beginning of the pulse ) , membrane resistance ( based on the steady state current during the last 20% of the pulse ) , and cell capacitance ( based on the time constant of an exponential fit between 20% and 80% of the current decay ) . Reported values in Figure 6—figure supplement 1 and Figure 7—figure supplement 1 are the means of these parameters from before and after the 5 min of sEPSC recordings . For sEPSCs and sIPSCs , all sweeps were concatenated for each cell ( excluding 0 . 5 s around the test pulse ) , filtered with a 60 Hz band-stop filter with five harmonics in Matlab , and exported into Mini Analysis v . 6 . 0 . 7 ( Synaptosoft ) . sEPSCs were detected with the following parameters: threshold 5 , period to search a local minimum 10000 , time before a peak to baseline 15000 , period to search a decay time 20000 , fraction of peak to find a decay time 0 . 37 , period to average a baseline 1000 , area threshold 20 , number of points to average for peak 1 , direction of peak ‘negative . ’ After the initial detection step , ‘Scan and detect double peaks’ was selected . sIPSCs were detected with the following parameters: threshold 8 , period to search a local minimum 10000 , time before a peak to baseline 6000 , period to search a decay time 20000 , fraction of peak to find a decay time 0 . 37 , period to average a baseline 1000 , area threshold 80 , number of points to average for peak 1 , direction of peak ‘positive . ’ In both cases , the detection was manually inspected , and the timepoints spanning any sections of the trace that exhibited increased noise ( typically due to fluctuations in seal quality that caused many obviously false positive events ) were noted . Event data was exported to a text file and imported into Matlab and noisy sections were excluded before further analysis . Cells that had many noisy sections were excluded . This included 3 control cells and 2 PLX cells for EPSCs and 2 control cells and 4 PLX cells for IPSCs ( Figure 6 ) as well as 3 control and 3 PLX cells for EPSCs and eight control cells and 9 PLX cells for IPSCs ( Figure 7 ) . Events that were less than 3 ms apart were considered doubly detected and excluded . The number of mice to be used in each experiment was determined in advance , and all data that could be obtained from these mice under the experimental conditions and quality standards described in each section above were included in the analyses . To determine the number of mice for two-photon imaging , we took into account our previous success rate with combined cranial windows/virus injections ( about 66% ) and aimed to image a similar number of mice as in our previous work ( 4–7 mice for different experiments , Wallace et al . , 2017 ) . For electrophysiology , we aimed to obtain data from at least 20 cells per group from 3 to 4 mice , consistent with other electrophysiological investigations in the field that were able to reveal effects of various manipulations ( Pallotto et al . , 2012; Quast et al . , 2017 ) . Littermates were randomly allocated into control or PLX-treated groups for all experiments . Raw data are shown in every figure , in the form of a scatterplot overlaid on a bar graph when N < 40 or as a cumulative distribution otherwise . Non-parametric statistical tests were used for all comparisons , so no assumption of normality of data was made . Details for each statistical test can be found in the corresponding figure legend , and the rationale behind each is described here . The two-tailed Wilcoxon rank sum test was used to compare the medians of all unpaired non-normal distributions and correspondingly , the medians are plotted with bars in the figures . A one-tailed permutation test was used to test whether means in the real data were significantly higher than those in offset imaging data in Figure 1 . A two-tailed permutation test was used to test for differences in principal components for the analysis of response time courses in Figure 3—figure supplement 1 ( details in section above ‘Two-photon imaging , Temporal dynamics’ ) . The hierarchical bootstrap ( Carpenter et al . , 2003; Saravanan et al . , 2019 ) was used to compare groups in the case of hierarchical datasets ( details below in section ‘Hierarchical bootstrap’ ) . The statistical test and p value for each test is found in the legend corresponding to each figure letter . All n values for each statistical test can be found at the end of each figure legend . In the text , we report mean ± SEM for datasets where means were compared or median ( interquartile range ) for datasets where medians were compared . For comparisons with hierarchical bootstrap , we report sampling distribution mean ±sampling distribution standard deviation . Sampling was performed 1000 times for each dataset as described ( Saravanan et al . , 2019 ) . Specifically , a sample was taken with replacement across the units of the upper level . For example , in the two-photon imaging data , a sample was taken of the individual mice that were imaged ( the sample size was equal to the number of mice ) . Next , a sample was taken with replacement across the lower level , in this case the dendrites that were imaged . We chose to sample an equal number of dendrites from each mouse , regardless of the original number of dendrites imaged , to ensure that each animal would be represented equally ( since the number of dendrites originally imaged for each animal was related to experimental considerations such as the density of viral labeling in each animal ) . For this level , we chose a sample size ( 100 dendrites ) slightly larger than the largest sample per unit in the original data to ensure a representative sample would be obtained from each mouse . We then report the mean and the standard error of the bootstrapped parameter ( either mean , median , or proportion , which is stated in each case ) across the 1000 samples . We calculate a p value directly , which represents the probability that the parameter for the control group is greater than that for the PLX-treated group ( for cases where the parameter for the PLX-treated group is significantly greater , such as the proportion of dendrites that were unresponsive in the imaging data , we report 1 – p to avoid confusion with standard p values ) . The same procedure was performed to analyze datasets for spine morphology ( upper level = dendrites , lower level = spines , sample size = 100 spines ) and electrophysiology ( upper level = cells , lower level = EPSCs or IPSCs , sample size = 5000 EPSCs or IPSCs ) . | The brain has its own population of resident immune cells known as microglia , which defend against infections and are involved in conditions such as Alzheimer’s , Parkinson’s and other diseases . In the last decade , new studies have suggested that these cells also sculpt brain circuits during early development . They can ‘eat’ weak connections between neurons , and help strong ones to mature . Most of brain ‘wiring’ happens during development , when the majority of neurons is born and connects together . However , a few brain areas can incorporate new neurons during adulthood into existing circuits . In mice for example , this process takes place in the olfactory bulb , the area that first processes smells: it is believed that new neurons connecting to existing ones helps to detect new odors . It is unclear , however , whether microglia also help to shape these connections , or if their role is confined to early development . To investigate this question , Wallace et al . gave adult mice a drug that kills only microglia , and then examined how the neurons respond when the animals are exposed to smells . The results show that the new neurons that developed without microglia responded to fewer odors . These neurons also formed weaker connections and had physical features that indicated they might not have been properly incorporated into the circuit . It may be possible to encourage new neurons to be born in brain areas that normally do not produce these cells in adulthood . Ultimately , this could potentially help to repair the damages of age or disease , but this will rely on understanding exactly how new neurons are integrated into existing brain circuits . Future work , however , is still necessary to figure out how much these new neurons could compensate for cells damaged by injury or disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2020 | Microglial depletion disrupts normal functional development of adult-born neurons in the olfactory bulb |
As the major mechanism of plant growth and morphogenesis , cell elongation is controlled by many hormonal and environmental signals . How these signals are coordinated at the molecular level to ensure coherent cellular responses remains unclear . In this study , we illustrate a molecular circuit that integrates all major growth-regulating signals , including auxin , brassinosteroid , gibberellin , light , and temperature . Analyses of genome-wide targets , genetic and biochemical interactions demonstrate that the auxin-response factor ARF6 , the light/temperature-regulated transcription factor PIF4 , and the brassinosteroid-signaling transcription factor BZR1 , interact with each other and cooperatively regulate large numbers of common target genes , but their DNA-binding activities are blocked by the gibberellin-inactivated repressor RGA . In addition , a tripartite HLH/bHLH module feedback regulates PIFs and additional bHLH factors that interact with ARF6 , and thereby modulates auxin sensitivity according to developmental and environmental cues . Our results demonstrate a central growth-regulation circuit that integrates hormonal , environmental , and developmental controls of cell elongation in Arabidopsis hypocotyl .
The high levels of developmental plasticity in higher plants relies on coordinated regulation of cell elongation by many hormonal and environmental signals , including particularly light , temperature , auxin , gibberellin ( GA ) , and brassinosteroid ( BR ) , which have major effects on cell elongation and seedling morphogenesis . Complex interplays among these signals have been observed at the genetic and physiological levels , but the molecular mechanisms underlying these interactions are not fully understood . Recent studies have shown integration of the light , temperature , BR and GA pathways through direct interactions between their target transcription regulators ( Gallego-Bartolome et al . , 2012; Oh et al . , 2012; Bai et al . , 2012b; Li et al . , 2012b ) . However , the relationship of the major growth hormone auxin with the other signals remains unclear at the molecular level . Auxin is the dominant plant growth hormone that plays key roles in nearly all developmental processes including patterning and growth responses to the environment . Auxin is essential for cell elongation responses to shade , warm temperature , and the circadian clock as well as tropic growth responses to light and gravity ( Stewart and Nemhauser , 2010; Del Bianco and Kepinski , 2011 ) . While regulation of auxin level and distribution is an important aspect of its function , the ability of auxin to regulate cell elongation also depends on developmental context and the status of other hormonal and environmental signals ( Stewart and Nemhauser , 2010; Del Bianco and Kepinski , 2011 ) . For example , auxin and brassinosteroid ( BR ) are known to be interdependent and show synergistic interactions in promoting hypocotyl elongation , and they induce highly overlapping transcriptional responses ( Goda et al . , 2004; Nemhauser et al . , 2004 ) . Normal auxin response also requires another growth-promoting hormone gibberellin ( GA ) ( Chapman et al . , 2012 ) . Furthermore , both auxin level and sensitivity are modulated by light , temperature , and circadian rhythm , through the phytochrome-interacting factor ( PIFs ) family of bHLH transcription factors ( Covington and Harmer , 2007; Nozue et al . , 2007; Koini et al . , 2009; Franklin et al . , 2011; Nozue et al . , 2011; Hornitschek et al . , 2012; Sun et al . , 2012; Li et al . , 2012a ) . The mechanisms for regulation of auxin levels and distribution through metabolism and polar auxin transport have been studied extensively , however , little is known about direct interaction between the signal transduction pathways of auxin and other signals at the molecular level . Auxin signaling induces ubiquitination and degradation of the AUX/IAA family proteins to release their inhibition of the auxin response factor ( ARF ) family of transcription factors ( Chapman and Estelle , 2009; Vernoux et al . , 2011 ) , but target genes of ARFs remain largely unknown and thus the mechanisms linking ARF activation to context-specific cellular responses remain poorly understood ( Del Bianco and Kepinski , 2011 ) . By contrast , BR acts through a receptor kinase pathway to inhibit BIN2/GSK3-mediated phosphorylation of the brassinazole resistant ( BZR ) family of transcription factors , leading to their accumulation in the nucleus and regulation of thousands of target genes ( Wang et al . , 2012 ) . Cross regulation between auxin and BR has been observed at several levels , including auxin activation of BR biosynthetic genes , BR regulation of the expression levels of auxin transporters ( Bao et al . , 2004; Mouchel et al . , 2006; Chung et al . , 2011; Yoshimitsu et al . , 2011 ) , and BIN2 phosphorylation of ARF2 ( Vert et al . , 2008 ) . However , these cross regulation mechanisms appear insufficient to explain the mutual interdependence between BR and auxin . There has been evidence that both BZR2 ( also named BES1 ) and ARF5 bind to the promoter of the auxin- and BR-activated SAUR15 gene ( Walcher and Nemhauser , 2012 ) . The functions of such interactions in the auxin-BR co-regulation of genome expression and cell elongation , however , remain unclear . Much less is known about direct interactions between auxin and the phytochrome or GA pathways . While recent studies demonstrated convergence of the BR , light , and GA pathways through interactions between PIF4 , BZR1 and the GA-inactivated repressor DELLA proteins ( Feng et al . , 2008; de Lucas et al . , 2008; Gallego-Bartolome et al . , 2012; Oh et al . , 2012; Wang et al . , 2012; Bai et al . , 2012b; Li et al . , 2012b ) , the current models suggest that auxin interacts with other signals mainly through modulation of hormone levels ( Mouchel et al . , 2006; Chung et al . , 2011; Franklin et al . , 2011; Yoshimitsu et al . , 2011; Chapman et al . , 2012; Sun et al . , 2012; Li et al . , 2012a ) . In this study , we performed genome-wide analyses of target genes of an auxin response factor ( ARF6 ) that regulates hypocotyl elongation , and we demonstrate that the majority of ARF6 target genes are also targets of BZR1 and/or PIF4 . Genetic and biochemical assays further demonstrate that these factors interact directly and bind to shared target genes cooperatively . Furthermore , the DELLA protein RGA interacts with ARF6 and blocks its DNA binding . Our study elucidates a central growth regulation circuit that explains how auxin , BR , GA , light , and temperature act together in regulating hypocotyl cell elongation and how the hormone sensitivities are modulated by environmental signals and developmental programs .
ARF6 and its closed homolog ARF8 were previously shown to redundantly regulate hypocotyl elongation in Arabidopsis ( Nagpal et al . , 2005 ) . To define the genomic targets of auxin involved in hypocotyl cell elongation , we performed chromatin-immunoprecipitation followed by sequencing ( ChIP-Seq ) analysis of target genes of ARF6 . An ARF6-Myc fusion protein was expressed from the ARF6 promoter in transgenic arf6-2;arf8-3 plants , and rescued the short-hypocotyl phenotype of the arf6;arf8 double mutant ( Figure 1—figure supplement 1A ) . ChIP-Seq analysis using anti-Myc antibody identified 2037 ARF6-binding sites in the Arabidopsis genome . Most of the ARF6 binding sites were in the gene promoter regions consistent with its molecular function as a transcription regulator ( Figure 1A ) . The 2037 binding sites were linked to 2675 neighbor genes ( Figure 1—source data 1A ) , which were considered ARF6 binding target genes . The ARF6 binding targets include 40 of the 49 early auxin-induced genes in the hypocotyl tissues after 30 min of auxin treatment ( Chapman et al . , 2012 ) , but only 1 of the 16 immediate auxin-repressed genes ( Figure 1B ) . Therefore , ARF6 appears to function mainly as a transcriptional activator , consistent with previous study ( Tiwari et al . , 2003 ) . Comparison with auxin-activated genes ( Chapman et al . , 2012 ) identified 255 ARF6 binding targets that are activated by auxin in the hypocotyl tissues . These include many genes known to promote cell elongation ( PREs , BIM1 , BEE1 and HAT2 , SAURs ) and many genes with known function in auxin response , such as AUX/IAAs , PINs and PINOID ( Figure 1—source data 2 ) . 10 . 7554/eLife . 03031 . 003Figure 1 . ARF6 ChIP-Seq analyses . ( A ) Distribution of ARF6 binding peaks relative to gene structure ( −5000 base pairs from transcription start site to +1000 base pairs downstream of 3′ end ) . ( B ) Most of the early auxin-activated genes are ARF6 targets . Numbers above the columns indicate number of genes up- or down-regulated by 30 or 120 min of auxin treatments . ( C ) Venn diagram shows significant overlap among binding target genes of BZR1 , PIF4 and ARF6 . ( D ) ChIP-reChIP assay shows that BZR1 and ARF6 co-occupy shared target promoters . The enrichment of precipitated DNA was calculated as the ratio between transgenic plants and wild type control , normalized to that of the PP2A coding region as an internal control . Error bars indicate the SD of three biological repeats . ( E ) The G-box ( CACGTG ) , HUD ( CACATG ) , canonical AuxRE ( TGTCTC ) and TGTCGG are enriched in the ARF6 binding peaks associated with auxin-activated genes . GATCG ( a random motif ) is shown as a negative control . ( F ) Percentages of auxin-activated ARF6 binding peaks that have both E-box motif and core AuxRE ( TGTC ) , only TGTC , or only E-box motifs . ( G ) Distribution of distance between E-box motifs and core AuxRE ( TGTC ) found in the ARF6 peaks associated with auxin-activated genes or total Arabidopsis genome . ( H ) ARF6 binding peaks having both E-box motifs and AuxRE have higher probability ( % ) of being associated with auxin-activated ( 30 or 120 min treatment ) genes than the ARF6 binding peaks having only AuxRE . **p<0 . 01 . ( I ) Venn diagram shows that genes activated by auxin , BR , or GA and genes repressed by light are enriched in the common binding targets of BZR1 , PIF4 and ARF6 . Numbers in the Venn diagram indicate percentage of corresponding genes ( e . g . , auxin-activated genes ) in each section . Numbers in parentheses indicate percentage of genes in total Arabidopsis genome . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 00310 . 7554/eLife . 03031 . 004Figure 1—source data 1 . ( A ) ARF6 target genes . ChIP-Seq analysis R ( CSAR ) software was used to identify binding peaks , with parameters ( backg = 10 , norm = −1 , test = 'Ratio' , times = 1e6 , digits = 2 ) ( Muino et al . , 2011 ) . Binding peaks with FDR <0 . 01 were finally defined as the ARF6 binding peak and genes having at least one ARF6 binding peak within its promoter ( −3 kb ) or coding region or 1 kb downstream from stop codon were considered direct target genes . max: maximum peak value; u3000 , u2000 , u1000: upstream 3000 , 2000 , or 1000 bp from transcription start site; d0: coding region; d1000: downstream 1000 bp from stop codon . ( B ) BZR1 target genes . ChIP-Seq experiment was performed using the BZR1p::BZR1-CFP transgenic seedlings grown in the dark for 5 days , and anti-YFP antibody . Data were analyzed as described in legend of Figure 1—source data 1A . ( C ) Previous PIF4 ChIP-seq result ( Oh et al . , 2012 ) was re-analyzed with same statistical method as described in Figure 1—source data 1A , to define PIF4 target genes . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 00410 . 7554/eLife . 03031 . 005Figure 1—source data 2 . Auxin-activated genes previously identified in hypocotyls ( Chapman et al . , 2012 ) were compared with ARF6 target genes identified by ChIP-Seq to identify the auxin-activated ARF6 target genes in hypocotyls . 30 or 120 min: genes are activated after 30 or 120 min of auxin treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 00510 . 7554/eLife . 03031 . 006Figure 1—figure supplement 1 . ( A ) ARF6-Myc regulated by ARF6 native promoter restored short hypocotyl of the arf6;arf8 double mutant . Seedlings were grown in the dark for 6 days . Representative seedlings are shown . ( B ) Representative ARF6 , BZR1 , and PIF4 binding peaks in the promoters of ARF6 , BZR1 and PIF4 common target genes ( IAA19 , SAUR15 and AT2G23170 ) and UBC30 promoter as a negative control . ( C ) Distance distribution of ARF6 and PIF4 binding peaks or ARF6 and BZR1 binding peaks in the ARF6 , BZR1 and PIF4 common target genes . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 006 To understand the relationships between auxin , BR , and phytochrome pathways in regulating genome expression , we compared the ARF6 targets with the genome targets of BZR1 and PIF4 ( Figure 1—source data 1 ) . To generate comparable data sets , we performed BZR1 ChIP-Seq analysis using dark-grown seedlings ( Figure 1—source data 1B ) , as used in the ARF6 and PIF4 ChIP-Seq ( Oh et al . , 2012 ) . Interestingly , large portions of ARF6 binding targets overlapped with either targets of BZR1 ( 51% ) or PIF4 ( 71% ) or both ( 42% ) ( Figure 1C ) . The common targets include many genes with known functions in cell elongation ( EXP8 , BIM1 , BEE1/3 , PREs , HAT2 , IBH1 , HFR1 , PAR1/2 , EXO ) and auxin response ( PINs and SAURs ) . Moreover , the binding peak patterns of ARF6 , BZR1 , and PIF4 seemed very similar on promoters of many common targets ( Figure 1—figure supplement 1B ) , and the overall binding peaks were very close to each other ( Figure 1—figure supplement 1C ) , indicating that these three transcription factors bind to same or nearby genomic locations . ChIP-reChIP assays showed that two common targets , SAUR15 and IAA19 , were recovered by sequential immunoprecipitation in plants expressing both BZR1-YFP and ARF6-Myc , but not in plants expressing ARF6-Myc only ( Figure 1D ) , indicating that ARF6 and BZR1 co-occupy these promoters . Cis-element analysis identified two types of E-box motifs , the G-box ( CACGTG ) , and the Hormone Up at Dawn ( HUD , CACATG ) motifs ( Michael et al . , 2008 ) , that are highly enriched in the ARF6 binding regions associated with the auxin-activated genes ( Figure 1E ) . These motifs are known binding sites for BZR1/2 and PIFs and are also over represented in the genomic regions bound by BZR1 and PIF4 ( Sun et al . , 2010; Oh et al . , 2012 ) . The canonical ARF binding motif , AuxRE ( TGTCTC ) , and the binding motif ( TGTCGG ) recently identified for recombinant ARF1 and ARF5 were also highly enriched among the ARF6 binding regions ( Figure 1E; Ulmasov et al . , 1999; Boer et al . , 2014 ) . About half of the auxin-activated ARF6 binding regions had both core AuxRE ( TGTC ) and the E-box motifs within the ±100 base pairs around the binding site , which is much higher than random expectation ( only 9% ) ( Figure 1F ) . The E-box motifs tend to be located close to the core AuxRE , mostly within 20 base pairs ( Figure 1G ) . Furthermore , the ARF6 binding regions having both AuxRE and E-box motifs were significantly more frequently associated with auxin-activated genes than binding regions having only AuxRE ( Figure 1H ) . Consistent with cis-elements clustering , the ARF6 targets shared by BZR1 and PIF4 had higher percentage of auxin-activated genes than the targets of ARF6 alone , and a higher percentage of BR-activated genes than genes that are targets of BZR1 only ( Figure 1I ) . Furthermore , the GA-activated genes and light-repressed genes were also highly enriched among the common targets of BZR1 , PIF4 and ARF6 ( Figure 1I ) . These results suggest that the major growth signals—auxin , BR , GA , and light—converge at shared genomic target promoters containing combinatorial cis-elements for these factors . The large number of common target genes of ARF6 , BZR1 and PIF4 raises a possibility of direct interactions among these transcription factors . Indeed , ARF6 directly interacted with both BZR1 and PIF4 through the C-terminal domain of BZR1 and the bHLH domain of PIF4 in the yeast two-hybrid assays ( Figure 2A–D ) . Both the middle and C-terminal domains of ARF6 were required for the interactions with BZR1 and PIF4 ( Figure 2—figure supplement 1 ) . Co-immunoprecipitation assays showed that ARF6 interacts with BZR1 and PIF4 in vivo ( Figure 2E ) . In addition , the ARF6–PIF4 interaction was increased by co-transfection with a gain-of-function bzr1-1D protein that is constitutively active irrespective of BR signaling ( Wang et al . , 2002; Figure 2F ) , suggesting that BZR1–PIF4 interaction enhances PIF4 interaction with ARF6 . We next examined if BZR1 and PIF4 interact with other ARFs using yeast two-hybrid assays , and the results show that both BZR1 and PIF4 specifically interacted with ARF8 , but not ARF1 and ARF7 ( Figure 2G , H , Figure 2—figure supplement 2 ) , suggesting that BZR1 and PIF4 mediate only subsets of auxin responses such as hypocotyl elongation by interacting with specific ARFs . 10 . 7554/eLife . 03031 . 007Figure 2 . ARF6 interacts with BZR1 and PIF4 . ( A ) ARF6 interacts with BZR1 in yeast two-hybrid assay . Yeast clones were grown on the synthetic dropout ( +HIS ) medium or synthetic dropout medium without histidine ( −HIS ) plus 1 mM 3AT . ( B and C ) Box diagram of various fragments of BZR and PIF4 used in ( A , D , G , H ) . ( D ) ARF6 interacts with PIF4 in yeast two-hybrid assay . ( E ) ARF6 interacts with BZR1 and PIF4 in vivo . Transgenic plants expressing the indicated fusion proteins were used for immunoprecipitation using anti-GFP antibody , and the immunoblots were proved with anti-Myc antibody to detect interaction with the Myc-tagged ARF6 protein . ( F ) BZR1 enhances the ARF6–PIF4 interaction . Arabidopsis mesophyll protoplasts were transfected to express ARF6-Myc alone or together with PIF4-GFP and bzr1-1D-Myc as indicated , and the extracted proteins were immunoprecipitated by anti-GFP antibody . Gel blots were probed with anti-Myc or anti-GFP antibody . ( G and H ) BZR1 ( G ) and PIF4 ( H ) interact with ARF6 , but not with ARF1 and ARF7 in yeast two-hybrid assays . ( I ) ARF6 DNA-binding is enhanced by bzr1-1D . Seedlings ( 35S::ARF6-Myc ( WT ) and 35S::ARF6-Myc;bzr1-1D ( bzr1-1D ) ) grown on the 2 μM PPZ in the dark for 6 days were used for ChIP assays . Error bars in the ( I and J ) indicate the SD of three biological repeats . *p<0 . 05 and **p<0 . 01 . ( J ) PIF4-OX enhances ARF6 DNA-binding . Seedlings ( 35S::ARF6-Myc ( WT ) and 35S::ARF6-Myc;PIF4-OX ( PIF4-OX ) ) grown under light were used for ChIP assays of ARF6 binding to the indicated promoters . ( K ) Box plot shows that ARF6 binding peaks having both E-box motifs and AuxRE tend to have higher ARF6 DNA-binding affinity . ARF6 DNA-binding affinity was based on the peak score from the ARF6 ChIP-Seq analysis with CSAR . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 00710 . 7554/eLife . 03031 . 008Figure 2—figure supplement 1 . ( A ) Box diagram of various fragments of ARF6 used in the yeast-two hybrid assay . ( B ) ARF6 middle and C-terminal domains are required for the interaction with BZR1 . Yeast clones were grown on the synthetic dropout ( +HIS ) or synthetic dropout without histidine ( −HIS ) plus 1 or 5 mM 3AT medium . AD: activation domain fusion vector , BD: DNA binding domain fusion vector , x: empty vector . ( C ) ARF6 middle and C-terminal domains are required for the interaction with PIF4 . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 00810 . 7554/eLife . 03031 . 009Figure 2—figure supplement 2 . ARF8 interacts with both BZR1 and PIF4 . Yeast clones were grown on the synthetic dropout ( +HIS ) or synthetic dropout without histidine ( −HIS ) plus 1 mM 3AT medium . AD: activation domain fusion vector , BD: DNA binding domain fusion vector , x: empty vector . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 00910 . 7554/eLife . 03031 . 010Figure 2—figure supplement 3 . ARF6 DNA-binding on the common targets of ARF6 and BZR1 is enhanced by BR treatment . Seedlings ( 35S::ARF6-Myc ) grown on the medium containing 2 μM PPZ ( M ) or 2 μM PPZ + 100 nM brassinolide ( BL ) were used for the ChIP assay to determine ARF6 DNA-binding . Enrichment of DNA was calculated as the ratio between transgenic plants and wild type ( Col-0 ) , normalized to that of the PP2A coding region as an internal control . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 010 To test if ARF6 forms cooperative DNA binding complexes with BZR1 and PIF4 , we measured in vivo ARF6 DNA-binding activity on the ARF6 , BZR1 , and PIF4 common targets . The ARF6 binding to the promoters of common target genes were increased by BR treatment , as observed for ARF5 on the promoter of SAUR15 ( Walcher and Nemhauser , 2012; Figure 2—figure supplement 3 ) . In addition , bzr1-1D and PIF4-OX enhanced ARF6 binding to the common targets ( SAUR15 , SAUR19 and At1g29500 ) , but not to the targets of ARF6 only ( At4g12110 and At2g40880 ) , indicating that BZR1 and PIF4 enhance ARF6 DNA-binding in vivo ( Figure 2I , J ) . Based on the signal intensity of ARF6 ChIP-Seq data , ARF6 occupancy is higher at binding regions containing both E-box motifs and AuxRE than regions containing only AuxRE ( Figure 2K ) . These results support cooperative DNA binding of BZR1/PIF4 and ARF6 . To evaluate the function of BZR1-ARF6-PIF4 interaction in regulating gene expression , we carried out RNA-Seq analyses using wild type and the iaa3/shy2-2 mutant seedlings treated with mock or BL for 4 hr . The gain-of-function iaa3/shy2-2 mutation causes auxin insensitivity by stabilizing the IAA3 protein , which interacts with and inactivates multiple ARFs including ARF6 and ARF8 ( Figure 3—figure supplement 1; Vernoux et al . , 2011 ) . We identified 2664 genes that responded to BR in wild-type plants and 4725 genes differentially expressed in the iaa3 mutant compared to wild type ( Figure 3A , Figure 3—source datas 1 and 2 ) . Expression levels of many BR-regulated genes ( 1465 , 55% ) were also affected by iaa3 , and mostly in opposite ways ( correlation coefficient = −0 . 6 ) ( Figure 3B ) . Of 2482 BZR1- and PIFs-co-regulated genes ( Oh et al . , 2012 ) , 976 genes were affected by iaa3 ( 350 expected randomly ) ( Figure 3C ) . Heat-map in Figure 3D shows that most of the co-regulated genes ( 70% ) are similarly regulated by BZR1 and PIFs , but oppositely affected by iaa3 . Gene ontology ( GO ) analysis showed that many auxin-responsive genes and genes involved in cell wall biogenesis are activated by BZR1 and PIFs but repressed by iaa3 ( Figure 3—figure supplement 2A ) . 10 . 7554/eLife . 03031 . 011Figure 3 . ARF6 , BZR1 , and PIF4 synergistically induce gene expression . ( A ) Significant overlap between BR-regulated genes and IAA3-regulated genes . ( B ) Scatter plot of log2-fold change values in the 1465 overlapping set of IAA3- and BR-regulated genes . ( C ) Significant overlap among BZR1- , PIFs- , and IAA3-regulated genes . ( D ) Heat map of the 976 genes co-regulated by BZR1 , PIFs , and IAA3 . Scale bar indicates fold changes ( log2 value ) . ( E ) Box plot representation of the 1616 BR-activated or the 1048 BR-repressed genes in the WT and iaa3/shy2-2 . ( F ) Percentage of IAA3-dependent and IAA3-independent BR-regulated genes . Genes that were not significantly affected by BR treatment in iaa3/shy2-2 are defined as IAA3-dependent BR-regulated genes . ( G ) qRT-PCR analysis of BZR1-regulated genes in etiolated seedlings grown on 2 μM PPZ medium . Similar results are obtained from two independent biological repeats . Error bars indicate the SD of three technical repeats . ( H ) qRT-PCR analysis of BR-regulated genes in the seedlings treated with either mock or 100 nM BL for 4 hr . Error bars indicate the SD of three biological repeats . ( I ) qRT-PCR analysis of auxin responsive genes in the seedlings grown on medium containing no hormone ( M ) or 1 μM picloram , an artificial auxin . Error bars indicate the SD of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 01110 . 7554/eLife . 03031 . 012Figure 3—source data 1 . BR-regulated genes in wild type and their BR-responsive expression in the iaa3 mutant . Seedlings were grown on 2 µM propiconazole medium for 5 days in the dark and treated with mock or 100 nM brassinolide ( BL ) for 4 hr . BR-regulated genes were defined by 1 . 5-fold difference between wild type ( +BL ) and wild type ( −BL ) with p-value<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 01210 . 7554/eLife . 03031 . 013Figure 3—source data 2 . Genes whose expression levels are affected in the iaa3 mutant . Seedlings of wild type and iaa3 were grown on 2 µM propiconazole medium for 5 days in the dark and treated with 100 nM brassinolide for 4 hr . The IAA3-regulated genes were defined by 1 . 5-fold difference between iaa3 and wild type with p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 01310 . 7554/eLife . 03031 . 014Figure 3—figure supplement 1 . IAA3 interacts with both ARF6 and ARF8 . Yeast clones were grown on the synthetic dropout ( +HIS ) or synthetic dropout without histidine ( −HIS ) plus 1 mM 3AT medium . AD: activation domain fusion vector , BD: DNA binding domain fusion vector , x: empty vector . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 01410 . 7554/eLife . 03031 . 015Figure 3—figure supplement 2 . ( A ) Gene ontology analysis shows that the genes involved in cell wall organization or biogenesis , and the auxin responsive genes are enriched in the BZR1 , PIFs-activated but IAA3-repressed genes . ( B ) Endogenous BZR1 phosphorylation status is not affected in the iaa3/shy2-2 mutant . Seedlings ( Col-0 ) were grown on the medium containing 2 μM BRZ for 5 days in the dark and then treated with either mock ( M ) or 100 nM brassinolide ( BL ) for 30 min . Endogenous BZR1 was detected by anti-BZR1 antibody . p-BZR1: phosphorylated BZR1 , BZR1: de-phosphorylated BZR1 . ( C ) BZR1-CFP phosphorylation status is not affected in the iaa3 mutant . Transgenic plants expressing BZR1-CFP driven by native BZR1 promoter ( BZR1p::BZR1-CFP ) in the wild type or iaa3 were grown on the regular MS medium for 5 days under white light and then treated with either mock ( M ) or 100 nM brassinolide ( BL ) for 1 hr . BZR1-CFP was detected by anti-GFP antibody . p-BZR1-CFP: phosphorylated BZR1-CFP , BZR1-CFP: de-phosphorylated BZR1-CFP . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 015 The overall effect of BR treatment on gene expression is diminished in iaa3 compared to wild type ( Figure 3E ) . Of the 1616 BR-activated genes detected in wild-type plants , only 276 genes ( 17 . 1% ) were activated by BR treatment in iaa3 , whereas a bigger portion of BR-repressed genes ( 30% of 1048 ) were still repressed by BR treatment in iaa3 ( Figure 3F , Figure 3—source data 1 ) . Reverse transcription-quantitative PCR ( RT-qPCR ) analysis of selected genes , including SAURs ( SAUR15 , SAUR26 ) , ESPANSIN ( EXP8 ) , PREs , and DWF4 , confirmed the patterns observed in the genome-wide analysis ( Figure 3G ) . Activation of the BR- and auxin-induced genes by bzr1-1D was abolished or diminished by iaa3 , but the repression of BR-repressed DWF4 expression was unaffected by iaa3 ( Figure 3G ) . Consistent with iaa3 mutation not affecting the BZR1 function in repression of gene expression , the phosphorylation or accumulation status of BZR1 was not affected in the iaa3 mutant ( Figure 3—figure supplement 2B , C ) . If the effect of iaa3 on BR responses is due to inactivation of ARFs , loss of the ARF functions would have a similar effect as iaa3 . Indeed , BR-activated genes were less activated in arf6;arf8 than in wild type , but BR-repressed genes were normally repressed in arf6;arf8 ( Figure 3H ) . The effect of arf6;arf8 is similar to but weaker than iaa3 ( Figure 3G , H ) , consistent with additional ARF factors playing overlapping role with ARF6 and ARF8 and being suppressed by iaa3 . These results indicate that BR activation of genes for hypocotyl elongation is dependent on auxin activation of ARFs , whereas BR feedback repression of BR biosynthesis genes is independent of auxin signaling . We next asked whether BZR1 and PIFs are involved in the auxin regulation of gene expression . Auxin-activated genes SAUR15 , SAUR19 , ACS5 , and At4g13790 were less activated by auxin in bri1-116 than in wild type , but bzr1-1D enhanced the auxin activation of these genes in the bri1-116 background ( Figure 3I ) , indicating that BR promotes auxin responsive genes by activating BZR1 . The auxin activation of these genes was reduced in the pif-quadruple mutant ( pifq ) lacking four PIFs ( PIF1/PIL5 , PIF3 , PIF4 and PIF5/PIL6 ) ( Shin et al . , 2009 ) and completely abolished by overexpression of PAR1 ( PAR1-OX ) , which inhibits PIF activities ( Hao et al . , 2012; Figure 3I ) , consistent with previous observation ( Roig-Villanova et al . , 2007 ) . Taken together , our genome- and gene-expression analyses show that BZR1 , PIFs , and ARFs interdependently regulate the expression of large numbers of genes , integrating BR , light , and auxin signals into a common set of transcriptome . The BZR-ARF-PIF module provides a molecular model for integrating auxin signaling with BR and phytochrome pathways . To understand the functional importance of the interactions between ARF6 , BZR1 , and PIF4 , we analyzed the effects of genetic alteration of each component on the growth responses to changes in the other activities . Hypocotyl elongation of the BR receptor mutant bri1 shows diminished response to auxin ( Figure 4A ) , as observed previously ( Nemhauser et al . , 2004 ) , but the auxin-insensitive phenotype of bri1 was fully rescued by bzr1-1D ( Figure 4A ) , indicating that BZR1 mediates BR enhancement of auxin response . The hypersensitivity of bzr1-1D to auxin was abolished by the iaa3 mutation ( Figure 4B ) , suggesting that ARF activity is required for BZR1 function . Consistently , both iaa3 and arf6;arf8 were less sensitive to BR than was wild type ( Figure 4C ) . Furthermore , the bzr1-1D;arf6;arf8 triple mutant and bzr1-1D;iaa3 double mutant showed shorter hypocotyls on the medium containing BR biosynthesis inhibitor BRZ than the bzr1-1D single mutant ( Figure 4D , E ) , indicating that ARF6/8 are required for BZR1 promotion of hypocotyl elongation . Finally , to determine whether PIFs are required for auxin response , we checked the hypocotyl response to auxin in pifq and PAR1-OX . Compared with wild type , pifq was less sensitive and PAR1-OX was almost insensitive to auxin ( Figure 4A ) . These results indicate that ARF , BZR1 and PIFs are interdependent in promoting hypocotyl elongation , consistent with their cooperative regulation of a core set of genes involved in hypocotyl cell elongation . 10 . 7554/eLife . 03031 . 016Figure 4 . ARF6 , BZR1 , and PIF4 act interdependently in promoting hypocotyl elongation . ( A ) BZR1 and PIFs are required for auxin promotion of hypocotyl elongation . Seedlings were grown on 5 μM artificial auxin picloram or mock medium . ( B ) Hypersensitivity of bzr1-1D to auxin is abolished by iaa3/shy2-2 mutation . Seedlings were grown on the medium containing 2 μM brassinazole ( BRZ ) with or without 5 μM artificial auxin picloram . ( C ) ARF6 and ARF8 are required for BR promotion of hypocotyl elongation . Seedlings were grown on medium containing 2 μM BRZ plus various concentration of BL in the dark . ( D ) The iaa3/shy2-2 mutation inhibits BZR1 promotion of hypocotyl elongation . Representative seedlings are shown in left panel and quantification of hypocotyl lengths are shown in right graph . Seedlings were grown on the 2 μM BRZ medium in the dark . ( E ) ARF6 and ARF8 are required for BZR1 promotion of hypocotyl elongation . Seedlings were grown on the 2 μM BRZ medium in the dark . All error bars in ( A–E ) indicate SD ( n = 10 plants ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 016 The stronger auxin insensitivity of PAR1-OX than pifq suggests that besides the four PIFs , additional PAR1-inactivated factors are involved in the auxin signaling . However , PAR1 did not interact with IAA3 and ARF6 ( Figure 5—figure supplement 1 ) . The Arabidopsis Interactome map showed that PAR1 interacts with another bHLH transcription factor BEE2 ( Arabidopsis Interactome Mapping Consortium , 2011 ) , which was shown to be involved in BR regulation of cell elongation ( Friedrichsen et al . , 2002 ) . Our yeast two-hybrid assays confirmed that PAR1 interacts with BEE2 and its close homolog HBI1 as well ( Figure 5A ) . Like PIF4 , BEE2 , and HBI1 also interact with ARF6 ( Figure 5B ) . Transgenic plants overexpressing a dominant repressor version of HBI1 ( HBI1-SRDX ) ( Bai et al . , 2012a ) were less sensitive to auxin than was wild type ( Figure 5C ) , supporting a positive role of HBI1 in auxin promotion of hypocotyl elongation . These results suggest that ARF6 interacts with multiple bHLH transcription factors as co-transcriptional regulators , and PAR1 attenuates auxin response through direct inactivation of these bHLH transcription factors . 10 . 7554/eLife . 03031 . 017Figure 5 . The HLH/bHLH module mediates developmental regulation of auxin sensitivity . ( A ) PAR1 interacts with BEE2 and HBI1 in yeast two-hybrid assay . ( B ) ARF6 interacts with BEE2 and HBI1 in yeast two-hybrid assay . ( C ) The pre-amiR , IBH1-OX , and HBI1-SRDX plants are less sensitive to auxin . Seedlings were grown on hormone-free or 5 μM picloram medium for 7 days . Error bars indicate SD ( n = 10 plants ) . ( D ) Auxin activation of gene expression is diminished in the pre-amiR and IBH1-OX plants . 7-day-old seedlings were treated with mock ( M ) or 1 μM IAA for 2 hr . ( E ) Young stems are more sensitive to auxin than mature stems . Young stems ( 2 cm stem from the top ) and mature stems ( 2 cm stem from the bottom ) were treated with mock ( M ) or 1 μM IAA for 2 hr . ( F ) Auxin sensitivity of mature stem is enhanced by PRE1-OX , HBI1-OX , and PIF4-OX . Numbers indicate ratios between IAA-treated and mock-treated . ( G ) The DWF4 expression is high in the young stems . ( H ) BZR1 is less phosphorylated in young stems than in mature stems . Proteins extracted from the young and mature stems of the BZR1p::BZR1-CFP transgenic plants were analyzed by anti-YFP immunoblotting . Ponceau S staining is shown for loading control . ( I ) Auxin sensitivity of mature stem is restored by bzr1-1D . Sections of mature and young stems from plants of same height were treated with IAA for 2 hr , and the expression levels of SAUR15 were analyzed by qRT-PCR . Numbers in ( E , F , I ) indicate ratios of the expression levels of IAA-treated to mock-treated . Error bars in ( D–I ) indicate the SD of three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 01710 . 7554/eLife . 03031 . 018Figure 5—figure supplement 1 . PAR1 does not interact with ARF6 and IAA3 . Yeast clones were grown on the synthetic dropout ( +HIS ) or synthetic dropout without histidine ( −HIS ) plus 1 mM 3AT medium . AD: activation domain fusion vector , BD: DNA binding domain fusion vector , x: empty vector . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 018 HBI1 was recently identified as part of a tripartite HLH/bHLH cascade , in which two non-DNA binding HLH factors , PRE1 and IBH1 , antagonize each other , and IBH1 interacts with HBI1 and inhibits HBI1 DNA-binding ( Bai et al . , 2012a ) . Therefore , PRE1 and IBH1 may modulate auxin response by altering HBI1 activity . Indeed , knock down of multiple PREs ( PRE1 , PRE2 , PRE5 and PRE6 ) by artificial micro-RNA ( pre-amiR ) or overexpression of IBH1 ( IBH1-OX ) ( Bai et al . , 2012a , 2012b ) , significantly reduced the sensitivity of transgenic plants to auxin in terms of both hypocotyl elongation and gene expression ( Figure 5C , D ) . The expression levels of IBH1 and PRE1 are developmentally regulated; PRE1 expression is high in young and growing tissues , whereas IBH1 expression is high in mature and growth-arrested tissues ( Zhang et al . , 2009; Ikeda et al . , 2012 ) . We confirmed the developmental regulation of expression of PRE1 , PRE5 and IBH1 ( Figure 5E ) , and found that auxin treatment induced bigger fold changes of the PREs and SAUR15 expression levels in young stems than in mature stems ( Figure 5E ) . The reduced auxin sensitivity of the SAUR15 gene in the mature stem was restored in the PRE1-OX , HBI1-OX and PIF4-OX plants ( Figure 5F ) , supporting that developmental regulation of PREs and IBH1 impacts auxin sensitivity through PIF4 and HBI1 . The expression level of BR biosynthesis gene DWF4 was high in young stems but low in mature stems ( Figure 5G ) . Consistent with the expression pattern of DWF4 , BZR1 was less phosphorylated in the young stems than in the mature stems ( Figure 5H ) , indicating that BZR1 is more activated in the young tissues ( He et al . , 2002; Gampala et al . , 2007; Ryu et al . , 2007 ) . Since BZR1 interacts with ARF6 to potentiate auxin response , differential BZR1 activity may also contribute to the difference in auxin sensitivity between young and mature tissues . Indeed , activation of BZR1 by the bzr1-1D mutation increased auxin sensitivity of SAUR15 expression in the mature stem ( Figure 5I ) . Together , these results suggest that developmental regulation of BZR1 and HLH factors contributes to the changes of auxin sensitivity over the progression of organ development . GA regulates cell elongation through the degradation of DELLA proteins , which inactivate BZR1 and PIFs ( Bai et al . , 2012b; de Lucas et al . , 2008; Feng et al . , 2008 ) . A comparison of ARF6 targets with BZR1 and PIF4 targets revealed that GA-activated genes are enriched in the common targets of ARF6 , BZR1 , and PIF4 ( Figure 1I ) , suggesting that ARF6 is also involved in GA response . Therefore , we tested whether ARF6 directly interacts with the DELLA protein RGA . In yeast two-hybrid assays , RGA interacted with the middle domain and , to a lesser extent , the DNA binding domain of ARF6 ( Figure 6A , Figure 6—figure supplement 1 ) . RGA also interacted with other activator ARFs ( ARF6 , ARF7 and ARF8 ) , but not repressor ARF1 ( Figure 6A ) . In addition , HA-RGA , but not HA-YFP , was pulled down by ARF6-Myc in vitro ( Figure 6B ) , and RGA-GFP was co-immunoprecipitated with ARF6-Myc in Arabidopsis protoplasts ( Figure 6C ) . These results demonstrate that RGA directly interacts with ARF6 . Since the middle domain of ARF6 also mediates the ARF6-PIF4/BZR1 interactions ( Figure 2—figure supplement 1 ) , it is likely that RGA competes with PIF4/BZR1 for interaction with ARF6 . Indeed , the ARF6-PIF4 interaction was reduced by rga-Δ17 , which is a stable form of RGA due to deletion of the N-terminal DELLA domain ( Bai et al . , 2012b; Figure 6D ) , suggesting that RGA disrupts the ARF6–PIF4 interaction . 10 . 7554/eLife . 03031 . 019Figure 6 . RGA interacts with ARF6 and blocks ARF6 binding to DNA . ( A ) RGA interacts with ARF6 , ARF7 , and ARF8 , but not ARF1 in the yeast two-hybrid assay . RGA with deletion of N-terminal 208 amino acids ( RGA-C ) was used for the assay . ( B ) RGA interacts with ARF6 in vitro . In vitro-translated HA-YFP and HA-RGA proteins were incubated with in vitro-translated ARF6-Myc protein bound to magnetic beads , and the pulled-down proteins were analyzed by immunoblot with anti-HA antibody . * indicates IgG band . ( C ) RGA interacts with ARF6 in vivo . Protein extracts from protoplasts transfected with ARF6-Myc or ARF6-Myc and RGA-GFP were immunoprecipitated with anti-GFP antibody , and analyzed by immunoblots with anti-GFP or anti-Myc antibody . ( D ) RGA disrupts the PIF4–ARF6 interaction . Arabidopsis mesophyll protoplasts were transfected to express ARF6-Myc alone or with PIF4-GFP and rga-Δ17-Myc as indicated , and the extracted proteins were immunoprecipitated by anti-GFP antibody . Gel blots were probed with anti-Myc or anti-GFP antibody . ( E ) RGA inhibits ARF6 binding to the IAA19 promoter in DNA pull-down assay . ( F ) RGA inhibits ARF6 DNA-binding ability in vivo . Protoplasts transfected with GFP-Myc ( negative control ) or ARF6-Myc with or without RGA-GFP were used for ChIP assay . Error bars indicate the s . d . of two technical repeats . Similar results were obtained in two independent experiments . ( G ) RGA inhibits ARF6 transcriptional activation activity . IAA19p::Luc was co-transfected with ARF6-GFP , RGA-GFP , or both , into Arabidopsis mesophyll protoplasts . The IAA19p::Luc activities were normalized by the 35S::renilla luciferase . Error bars indicate the s . e . of 10 biological repeats ( n = 10 ) and **p<0 . 01 . ( H ) Auxin signaling mutants are less sensitive to GA . Seedlings were grown on the 10 μM paclobutrazole with or without 1 μM GA in the dark . Error bars indicate SD ( n = 10 plants ) . ( I ) DELLA inhibits the auxin promotion of hypocotyl elongation . Seedlings were grown on MS medium for 3 days and then transferred to the medium containing mock or 5 μM picloram , with or without 10 μM paclobutrazol ( PAC ) , and incubated for 4 days . Error bars indicate SD ( n = 10 plants ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 01910 . 7554/eLife . 03031 . 020Figure 6—figure supplement 1 . ( A ) Box diagram of various fragments of ARF6 used in the yeast-two hybrid assay . ( B ) RGA interacts strongly with a middle domain of ARF6 ( ARF6 M ) and interacts weakly with a DNA binding domain of ARF6 ( ARF6 D ) . RGA with deletion of N-terminal 208 amino acids ( RGA-C ) was used for the assay . Yeast clones were grown on the synthetic dropout ( +HIS ) or synthetic dropout without histidine ( −HIS ) plus 1 mM 3AT medium . AD: activation domain fusion vector , BD: DNA binding domain fusion vector , x: empty vector . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 02010 . 7554/eLife . 03031 . 021Figure 6—figure supplement 2 . Auxin signaling mutants are less sensitive to GA . Seedlings were grown on the 10 μM paclobutrazole with or without 1 μM GA in the dark . Representative seedlings are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 021 To examine if RGA directly inhibits the DNA-binding ability of ARF6 , like it does to BZR1 and PIF4 ( de Lucas et al . , 2008; Feng et al . , 2008; Li et al . , 2012b; Gallego-Bartolome et al . , 2012; Bai et al . , 2012b ) , we performed DNA–protein pull-down assays . Biotin-labeled IAA19 promoter fragment containing AuxRE effectively pulled down MBP-ARF6 in the absence of RGA , but not after pre-incubation of MBP-ARF6 with GST-RGA ( Figure 6E ) , indicating that RGA prevents ARF6 from binding to target DNA . ChIP assay confirmed that RGA inhibits ARF6 binding to target genes in vivo ( Figure 6F ) . Transient reporter gene expression assays showed that ARF6 increases IAA19 promoter activity , but the ARF6 effect was abolished by co-transfection with RGA ( Figure 6G ) . These data demonstrate that DELLA interacts with ARF6 and inhibits ARF6 DNA-binding to modulate ARF6 target gene expression . The DELLA inhibition of ARF6 suggests that GA promotes cell elongation by enhancing auxin/ARF-mediated responses . Indeed , the GA-promotion of hypocotyl elongation was much reduced in the iaa3 and arf6;arf8 mutants compared to the wild type ( Figure 6H , Figure 6—figure supplement 2 ) , supporting the notion that auxin activation of ARFs is necessary for the GA promotion of hypocotyl elongation . In addition , the auxin promotion of hypocotyl elongation was compromised in the GA-insensitive mutant gai-1 ( Peng et al . , 1997 ) and in wild-type plants grown on the medium containing GA biosynthesis inhibitor paclobutrazol ( PAC ) ( Figure 6I ) , both of which accumulate DELLA proteins , but was nearly normal in the PAC-treated della pentuple mutant ( della ) lacking all five members of the DELLA family ( Wang et al . , 2009; Figure 6I ) , indicating that accumulation of DELLA proteins inhibits auxin sensitivity . Taken together , these data indicate that the GA-induced degradation of DELLAs allows ARF6 , together with BZR1 and PIF4 , to bind to target DNA and to activate gene expression and hypocotyl cell elongation .
The ability to make decision based on integration of large numbers of signal inputs is a feature of advanced control system . The level of such ability that has evolved for the cellular control systems of plants remains unclear . While cell elongation , as the major growth process in plants , is the target of many signaling pathways that have evolved to provide the high level of developmental plasticity in higher plants , it has been unclear whether these pathways act on independent cellular machineries involved in elongation or they are processed by a central control system into a coherent cellular decision . Although physiological synergism and genetic interdependency suggested signal integration , the essential molecular connections between the signaling pathways , particularly with the auxin pathway , have been elusive . Our findings of the BZR-ARF-PIF/DELLA ( BAP/D ) module illustrate an elegant model of signal integration , which explains the synergistic or antagonistic interactions between auxin and the other signaling pathways . Our study demonstrates that hypocotyl cell elongation is controlled by major hormonal and environmental pathways through a central circuit of interacting transcription regulators . A close relationship between auxin and BR has been suggested by their synergistic physiological effects , genetic interdependence , and overlapping genomic effects ( Goda et al . , 2004; Nemhauser et al . , 2004 ) . Several possible mechanisms have been proposed for the auxin-BR interdependence , including BIN2–ARF interaction , and co-regulation of gene expression by BZR and ARF factors ( Vert et al . , 2008; Walcher and Nemhauser , 2012 ) . Our genetic analysis demonstrated that BZR1 plays a major role in potentiating auxin responses . Consistent with our genetic data and previous analysis of BR-auxin co-regulation of the SAUR15 promoter ( Walcher and Nemhauser , 2012 ) , our genomic and biochemical experiments showed direct interaction between BZR1 and ARF6 at the promoters of a large set of genomic targets , demonstrating that the auxin and BR pathways mainly converge through BZR–ARF interaction at shared target promoters . Light antagonizes auxin , BR , and GA to inhibit hypocotyl elongation and promote photomorphogenesis . Light signaling mediated by phytochromes induces degradation of PIFs ( Leivar and Quail , 2011 ) . In this study , we show that ARF6 shares most of its binding target genes with PIF4 and they interdependently activate shared target promoters , which explains the requirement of PIFs for auxin responsive gene expression and hypocotyl elongation and the requirement of auxin for skotomorphogenesis . The incomplete overlap between targets of BZR1 , ARF6 , and PIF4 suggests that their interactions are selected and/or facilitated by specific promoters to allow co-regulation of hypocotyl elongation and photomorphogenesis , while each pathway or combination of two pathways may regulate other gene sets and developmental processes . ARF6 showed a higher level of target overlap with PIF4 than with BZR1 , suggesting a tighter integration between the environmental signal and hormonal signal than between different hormones . PIFs are considered a central hub for integrating environmental and developmental signals ( Leivar and Quail , 2011 ) . In addition to light , temperature and the circadian clock transcriptionally regulate members of PIF family ( Nozue et al . , 2007; Koini et al . , 2009 ) , and consequently alter auxin synthesis as well as auxin sensitivity through the BAP module . The integral role of PIFs in hormone-responsive gene expression is also consistent with the finding of the HUD element , a potential binding site of PIF4/5 , associated with the morning-specific phytohormone gene expression program ( Nozue et al . , 2007; Michael et al . , 2008 ) . The DNA-binding activities of PIFs are also modulated by the tripartite HLH/bHLH module , in which the PRE family of non-DNA binding HLH factors sequesters another class of HLH factors ( including IBH1 and PAR1 ) , which otherwise inhibit DNA-binding of bHLH factors including HBI1 and PIFs ( Hao et al . , 2012; Ikeda et al . , 2012; Bai et al . , 2012a ) . The HLH/bHLH module has a major effect on plant sensitivity to auxin , BR and GA , presumably by controlling PIF4 and HBI1 , which both interact with ARF6 . Considering that GA , BR , auxin , and PIF4 increase the transcription levels of several PRE members ( Chapman et al . , 2012; Oh et al . , 2012; Bai et al . , 2012b ) , the HLH/bHLH modules appear to form positive feedback loops , which potentially re-enforce the activation of the BAP module and help maintain the growing condition in the dark or in young developing organs . The increase of IBH1 expression appears to mediate , at least partly , inactivation of hormone responses in mature organs ( Figure 5 ) , whereas a decrease of HBI1 expression mediates growth arrest and defense activation in response to pathogen infection ( Fan et al . , 2014; Malinovsky et al . , 2014 ) . As such , the HLH/bHLH module also provides additional nodes for input and output . GA has been shown to promote hypocotyl elongation by removing DELLA repression of BZR1 and PIF4 . Physiological studies supported additive effects of auxin and GA , while a recent study suggested that auxin regulates GA biosynthesis to release DELLA-dependent growth repression ( Chapman et al . , 2012 ) . In this study , we show that GA promotion of hypocotyl elongation also requires auxin activation of ARFs , as the iaa3 and arf6 , arf8 mutants show severely reduced GA response in hypocotyl elongation . Similar to DELLA inhibition of BZR1 and PIFs , DELLA also inhibits DNA-binding of ARF6 . In contrast to the cooperative interactions among BZR1 , ARF6 , and PIF4 , the DELLA protein RGA inhibits ARF6–PIF4 interaction . As such , DELLA inhibits both protein–DNA and protein–protein interactions of the BAP module , providing presumably coordinated and coherent control of all three components of the BAP module . Together , our study illustrates that the major growth-regulation pathways , auxin , BR , GA , and phytochrome , converge at the BAP/D module to control hypocotyl cell elongation ( Figure 7 ) . We propose that the BAP/D module coupled with the HLH/bHLH module forms the central growth regulation network that integrates hormonal , environmental , and developmental inputs into the decisions about hypocotyl cell elongation ( Figure 7 ) . 10 . 7554/eLife . 03031 . 022Figure 7 . Diagram of the central growth regulation circuit . In the diagram , solid lines indicate protein–protein interaction or post-translational modification , and dashed lines indicate transcriptional regulation . Red lines indicate new discoveries made in this study . In the BAP module , all three transcription factors , BR-regulated BZR1 , auxin-regulated ARF6 , and light/temperature-regulated PIF4 , interact with each other and cooperatively regulate shared target genes and hypocotyl cell elongation . GA-regulated DELLA interacts with all BAP transcription factors and inhibits their DNA binding . Downstream of BAP module , the HLH/bHLH module , consisting of PRE1 , IBH1/PAR1 and HBI1/PIF , modulates BAP activities through HLH–bHLH interactions . The BAP transcription factors positively regulate PRE1 in the HLH/bHLH module forming positive feedback loops . Development and pathogen signals are integrated into the central growth regulation network through PRE1/IBH1 and HBI1 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03031 . 022 In different developmental contexts , auxin induces distinct cellular and developmental responses , such as cell division , differentiation , elongation , and organogenesis . Whether the BAP/D network or similar transcription factor modules contribute to other auxin signaling outputs remains to be elucidated by future studies . The lack of interaction of BZR1 and PIF4 with ARF7 or ARF1 indicates a level of specificity of signal integration through ARF6 and ARF8 for regulation of hypocotyls and other shoot organs where ARF6 and ARF8 play important roles . Recent studies showed that auxin also acts through a receptor kinase-mediated signaling pathway to regulate cell morphogenesis ( Xu et al . , 2014 ) ; the details of this new auxin pathway and its relationships with the ARF-mediated auxin pathway and other hormonal and environmental signaling pathways are yet to be elucidated by future studies . Apparently , hormone interactions are complex and vary with developmental context . While the BAP/D module explains how auxin , BR , GA , light , and temperature coordinately regulate cell elongation of the hypocotyl , and likely other shoot organs in Arabidopsis , mechanisms of signal integration in other developmental contexts might be different and thus remain be elucidated .
All the Arabidopsis thaliana plants used in this study were in Col-0 ecotype background , except gai-1 and della , which were in the Landsberg erecta ecotype background ( Wang et al . , 2009 ) . The arf6;arf8 double mutant ( arf6-2;arf8-3 ) was provided by Jason W Reed ( Nagpal et al . , 2005 ) . To generate ARF6p::ARF6-Myc transgenic lines , ARF6 genomic fragment including 2 . 5 kb upstream of transcription start was cloned into the gateway compatible p1390-4Myc-His vector and transformed into the Col-0 and arf6 ( −/− ) ;arf8 ( +/− ) . Transgenic plants expressing ARF6-Myc and BZR1-YFP from 35S promoter ( ARF6-Myc;BZR1-YFP ) or ARF6-Myc and PIF4-YFP from 35S promoter ( ARF6-Myc;PIF4-YFP ) were treated with 100 nM BL for 4 hr . Harvested tissues were grounded in liquid nitrogen , homogenized in IP buffer ( 50 mM Tris-Cl pH7 . 5 , 1 mM EDTA , 75 mM NaCl , 0 . 1% Triton X-100 , 5% Glycerol , 1 mM PMSF , 1x Protease Inhibitor ) . After centrifugation at 20 , 000×g for 10 min , 1 ml of supernatant was incubated for 1 hr with anti-GFP ( custom made , 5 µg ) immobilized on protein A/G agarose beads ( Pierce Biotechnology , Rockford , IL ) . The beads were then washed for three times with 1 ml of IP buffer and eluted samples were analyzed by immunoblot using anti-Myc ( Cell Signaling Technology , Beverly , MA ) and anti-GFP antibodies . For co-IP assays using Arabidopsis mesophyll protoplasts , 2 × 104 isolated mesophyll protoplast were transfected with a total 20 μg of DNA and incubated overnight . Total proteins were extracted from the protoplasts using the IP buffer , and immunoprecipitation was performed as described above . Isolated Arabidopsis mesophyll protoplasts ( 2 × 104 ) were transfected with a total 20 μg of DNA and incubated overnight . Protoplasts were harvested by centrifugation and lysed in 50 μl of passive lysis buffer ( Promega , Madison , WI ) . Firefly and Renilla ( as an internal standard ) luciferase activities were measured by using a dual-luciferase reporter kit ( Promega ) . Total RNA was extracted from seedlings treated with mock or specific hormones by using the Spectrum Plant Total RNA kit ( Sigma , St . Louis , MO ) . M-MLV reverse transcriptase ( Fermentas , Thermo Scientific , Waltham , MA ) was used to synthesize cDNA from the RNA . Quantitative real-time PCR ( qRT-PCR ) was performed using LightCycler 480 ( Roche , Basel , Switzerland ) and the Bioline SYBR green master mix ( Bioline ) . Gene expression levels were normalized to that of PP2A and are shown relative to the expression levels in wild type . Gene specific primers are listed in Supplementary file 1 . For ChIP assays , seedlings ( 35S::ARF6-Myc ) were cross-linked for 20 min in 1% formaldehyde under vacuum . The chromatin complex was isolated , resuspended in lysis buffer ( 50 mM Tris–HCl pH 8 . 0 , 10 mM EDTA , 200 mM NaCl , 0 . 5% Triton X-100 , 1 mM PMSF ) and sheared by sonication to reduce the average DNA fragment size to around 500 bps . The sonicated chromatin complex was immunoprecipitated by anti-Myc antibody ( Cell Signaling Technology ) -bound protein A agarose beads ( Millipore , Bedford , MA ) . The beads were washed with low-salt buffer ( 50 mM Tris–HCl at pH 8 . 0 , 2 mM EDTA , 150 mM NaCl , 0 . 5% Triton X-100 ) , high-salt buffer ( 50 mM Tris–HCl at pH 8 . 0 , 2 mM EDTA , 500 mM NaCl , 0 . 5% Triton X-100 ) , LiCl buffer ( 10 mM Tris–HCl at pH 8 . 0 , 1 mM EDTA , 0 . 25 M LiCl , 0 . 5% NP-40 , 0 . 5% deoxycholate ) , and TE buffer ( 10 mM Tris–HCl at pH 8 . 0 , 1 mM EDTA ) and eluted with elution buffer ( 1% SDS , 0 . 1 M NaHCO3 ) . The ARF6-bound DNA was purified by using a PCR purification kit ( Thermo Scientific ) and analyzed by ChIP-qPCR . The enrichment of DNA was calculated as the ratio between ARF6-Myc and WT samples , normalized to that of the PP2A . Primers for qPCR are listed in Supplementary file 1 . The 5-day-old dark-grown seedlings of ARF6p::ARF6-Myc;arf6;arf8 and 35S::GFP-Myc ( control ) , or BZR1p::BZR1-CFP and 35S::YFP ( control ) were used for ARF6 or BZR1 ChIP-Seq analysis , respectively , following protocols described before ( Oh et al . , 2012 ) . For ChIP-Seq library construction , 10 ng of ChIP-DNA were pooled from three biological repeats to reduce sample variation . High-throughput sequencing of ChIP-Seq libraries was carried out on an Illumina HiSeq 2000 . Sequences in Solexa FASTQ format were mapped to the Arabidopsis genome , TAIR9 , using SOAP2 . ARF6 binding peaks were identified using ChIP-Seq analysis R ( CSAR ) software with parameters ( backg = 10 , norm = −1 , test = 'Ratio' , times = 1e6 , digits = 2 ) ( Muino et al . , 2011 ) . Binding peaks with FDR < 0 . 01 were defined as the ARF6 binding peak and used in further analyses . Genes having at least one ARF6 binding peak within its promoter ( −3 kb ) , coding region or 1 kb downstream from stop codon were considered ARF6 binding target genes . ChIP-reChIP assays were performed using anti-Myc antibody first ( Cell Signaling Technology ) and then using anti-GFP antibody ( custom made ) . Precipitated DNA was quantified by qPCR . Enrichment of DNA was calculated as the ratio between transgenic plants and wild type control , normalized to that of the PP2A coding region as an internal control . All error bars indicate the SD of three biological repeats . Seedlings were grown on medium containing 2 µM propiconazole ( PPZ ) in the dark for 5 days and treated with mock or 100 nM BL for 4 hr before harvesting . Total RNA was extracted by using the Spectrum Plant Total RNA kit ( Sigma ) . Libraries were constructed by using TruSeq RNA Sample Preparation Kit ( Illumina ) according to the manufacturer's instruction . RNA-Seq analysis was performed as described previously ( Oh et al . , 2012 ) . Differentially expressed genes were defined by a 1 . 5-fold difference between samples with p<0 . 01 . The ARF6-Myc , HA-RGA and HA-YFP proteins were synthesized by TNT T7 Quick Coupled in vitro transcription/translation system ( Promega ) . The ARF6-Myc proteins were pre-incubated with anti-Myc antibody ( Cell Signaling Technology ) -bound protein A-Dynabeads ( Life Technology ) for 2 hr . After removing unbound ARF6-Myc proteins , the HA-RGA or HA-YFP proteins were incubated with the ARF6-Myc-bound Dynabeads for 1 hr in PBSN buffer ( PBS buffer + 0 . 1% NP-40 ) . The beads were washed three times with the PBSN buffer and the pulled-down proteins were analyzed by immunoblots using anti-HA antibody ( Roche ) and anti-Myc antibody ( Cell Signaling Technology ) . The MBP and MBP-ARF6 proteins were affinity-purified from Escherichia coli by using amylose resin ( NEB ) . The IAA19 promoter fragment was amplified by PCR using biotin-labeled primers ( Supplementary file 1 ) . The biotin-labeled IAA19 promoter fragment and the MBP or MBP-ARF6 proteins were incubated with streptavidin-bound agarose beads ( Sigma ) for 1 hr in IP100 buffer ( 100 mM potassium glutamate , 50 mM Tris–HCl pH 7 . 6 , 2 mM MgCl2 , 0 . 05% NP-40 ) . The beads were washed four times with the IP100 buffer and DNA-bound proteins were analyzed by an immunoblot . The ChIP-seq data used in this study may be viewed under GSE51770 . The RNA-seq data used in this study may be viewed under GSE51772 . | Plants can grow by making more cells or by increasing the size of these existing cells . Plant growth is carefully controlled , but it must be able to respond to changes in the plant's environment . Many different plant hormones and various signals from the environment—such as light and temperature—influence how and when a plant grows . The different signals that affect cell growth typically act via distinct pathways that change which genes are switched on or off inside the cells . However , the ways in which these different signals are coordinated by plants are not fully understood . Now , Oh et al . have looked at the genes that are switched on and off in response to all the major signals that regulate the growth of the first stem to emerge from the seed of Arabidopsis , a small flowering plant that is widely studied by plant biologists . Oh et al . found that the proteins that change gene expression in response to hormones or the environment bind to each other . These proteins , which are collectively called transcription factors , were also revealed to cooperate to regulate the expression of hundreds of genes: transcription factors have not been seen to behave in this way in plants before . By discovering a central mechanism that coordinates the different signals that control plant growth , these findings may guide future efforts to boost the yields of food crops and plants that are grown to make biofuels . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology"
] | 2014 | Cell elongation is regulated through a central circuit of interacting transcription factors in the Arabidopsis hypocotyl |
We present an automated method to track and identify neurons in C . elegans , called ‘fast Deep Neural Correspondence’ or fDNC , based on the transformer network architecture . The model is trained once on empirically derived semi-synthetic data and then predicts neural correspondence across held-out real animals . The same pre-trained model both tracks neurons across time and identifies corresponding neurons across individuals . Performance is evaluated against hand-annotated datasets , including NeuroPAL ( Yemini et al . , 2021 ) . Using only position information , the method achieves 79 . 1% accuracy at tracking neurons within an individual and 64 . 1% accuracy at identifying neurons across individuals . Accuracy at identifying neurons across individuals is even higher ( 78 . 2% ) when the model is applied to a dataset published by another group ( Chaudhary et al . , 2021 ) . Accuracy reaches 74 . 7% on our dataset when using color information from NeuroPAL . Unlike previous methods , fDNC does not require straightening or transforming the animal into a canonical coordinate system . The method is fast and predicts correspondence in 10 ms making it suitable for future real-time applications .
The nervous system of the nematode C . elegans is well characterized , such that each of the 302 neurons is named and has stereotyped locations across animals ( White et al . , 1986; Sulston , 1976; Witvliet et al . , 2020 ) . The capability to find corresponding neurons across animals is essential to investigate neural coding and neural dynamics across animals . Despite the worm’s overall stereotypy , the variability in neurons’ spatial arrangement is sufficient to make predicting neural correspondence a challenge . For whole-brain calcium imaging ( Schrödel et al . , 2013; Venkatachalam et al . , 2016; Nguyen et al . , 2016 ) , identifying neurons across animals is additionally challenging because the nuclear localized markers that are used tend to obscure morphological features that would otherwise assist in neural identification . An ideal method for finding neural correspondence in C . elegans should accommodate two major use cases . The first is tracking neurons within an individual across time as the animal’s head moves and deforms . Here , the goal is to be able to say with confidence that a neuron imaged in a volume taken at time t1 is the same as another neuron taken from a volume imaged at time t2 . Tracking across time is needed to extract calcium dynamics from neurons during freely moving population calcium imaging ( Venkatachalam et al . , 2016; Nguyen et al . , 2016; Lagache et al . , 2020 ) . Additionally , very fast real-time tracking will be needed to bring closed-loop techniques such as brain-machine interfaces ( Clancy et al . , 2014 ) , and optical patch clamping ( Hochbaum et al . , 2014 ) to moving animals . The second and more general use case is finding neural correspondence across individuals . Often this is to identify the name of a neuron with respect to the connectome ( White et al . , 1986 ) or a gene expression atlas ( Hammarlund et al . , 2018 ) . Even when a neuron’s name cannot be ascertained , being able to identify which neurons are the same across recordings allows researchers to study neural population codes common across individuals . For both use cases , a method to find neural correspondence is desired that is accurate , fast , requires minimal experimental training data and that generalizes across animal pose , orientation , imaging hardware , and conditions . Furthermore , an ideal method should not only perform well when restricted to neural positioning information but , should also be flexible enough to leverage genetically encoded color labeling information or other features for improved accuracy when available . Multicolor strains are powerful new tools that use multiple genetically encoded fluorescent labels to aid neural identification ( Yemini et al . , 2021; Toyoshima et al . , 2019 ) ( we use one of those strains , NeuroPAL ( Yemini et al . , 2021 ) , for validating our model ) . However , some applications , like whole-brain imaging in moving worms , are not yet easily compatible with the multicolor imaging required by these new strains , so there remains a need for improved methods that use position information alone . A variety of automated methods for C . elegans have been developed that address some , but not all these needs . Most methods developed so far focus on finding the extrinsic similarity ( Bronstein , 2007 ) between one neuron configuration , called a test , and another neuron configuration called a template . Methods like these deform space to minimize distances between neurons in the template and neurons in the test and then attempt to solve an assignment problem ( Lagache et al . , 2018 ) . For example , a simple implementation would be to use a non-rigid registration model , like Coherent Point Drift ( CPD ) ( Myronenko and Song , 2010 ) to optimize a warping function between neuron positions in the test and template . More recent non-rigid registration algorithms like PR-GLS ( Ma et al . , 2016 ) also incorporate relative spatial arrangement of the neurons ( Wen et al . , 2018 ) . Models can also do better by incorporating the statistics of neural variability . NeRVE registration and clustering ( Nguyen et al . , 2017 ) , for example , also uses a non-rigid point set registration algorithm ( Jian and Vemuri , 2011 ) to find a warping function that minimizes the difference between a configuration of neurons at one time point and another . But NeRVE further registers the test neurons onto multiple templates to define a feature vector and then finds neural correspondence by clustering those feature vectors . By using multiple templates , the method implicitly incorporates more information about the range and statistics of that individual animal’s poses to improve accuracy . A related line of work uses generative models to capture the statistics of variability across many individual worms . These generative models specify a joint probability distribution over neural labels and the locations , shapes , sizes , or appearance of neurons identified in the imaging data of multiple individuals ( Bubnis et al . , 2019; Varol et al . , 2020; Nejatbakhsh et al . , 2020; Nejatbakhsh and Varol , 2021 ) . These approaches are based on assumptions about the likelihood of observing a test neural configuration , given an underlying configuration of labeled neurons . For example , these generative models often begin with a Gaussian distribution over neuron positions in a canonical coordinate system and then assume a distribution over potentially non-rigid transformations of the worm’s pose for each test configuration . Then , under these assumptions , the most likely neural correspondence is estimated via approximate Bayesian inference . The success of generative modeling hinges upon the accuracy of its underlying assumptions , and these are challenging to make for high-dimensional data . An alternative is to take a discriminative modeling approach ( Bishop , 2006 ) . For example , recent work ( Chaudhary et al . , 2021 ) has used conditional random fields ( CRF ) to directly parameterize a conditional distribution over neuron labels , rather than assuming a model for the high-dimensional and complex image data . CRF allows for a wide range of informative features to be incorporated in the model , such as the angles between neurons , or their relative anterior-posterior positions , which are known to be useful for identifying neurons ( Long et al . , 2009 ) . Ultimately , however , it is up to the modeler to select and hand curate a set of features to input into the CRF . The next logical step is to allow for much richer features to be learned from the data . Artificial neural networks are ideal for tackling this problem , but they require immensely large training sets . Until now , their use for neuron identification has been limited . For example , in one tracking algorithm , artificial neural networks provide only the initialization , or first guess , for non-rigid registration ( Wen et al . , 2018 ) . Our approach is based on a simple insight: it is straightforward to generate very large semi-synthetic datasets of test and template worms that nonetheless are derived from measurements . We use neural positions extracted from existing imaging datasets , and then apply known , nonlinear transformations to warp those positions into new shapes for other body postures , or other individuals . Furthermore , we simulate the types of noise that appear in real datasets , such as missing or spurious neurons . Using these large-scale semi-synthetic datasets , we train an artificial neural network to map the simulated neural positions back to the ground truth . Given sufficient training data ( which we can generate at will ) , the network learns the most informative features of the neural configurations , rather than requiring the user to specify them by hand . Importantly , using semi-synthetic data also allows us to train our model even when we completely lack experimentally acquired ground truth data . And indeed , in this work , semi-synthetic data is derived exclusively from measurements that lack any ground truth correspondence either within- , or across animals . All ground truth for training comes only from simulation . Realistic synthetic , semi-synthetic or augmented datasets have been key to cracking other challenging problems in neurosicence ( Parthasarathy et al . , 2017; Yoon et al . , 2017; Sun et al . , 2018; Lee et al . , 2020; Mathis and Mathis , 2020; Pereira et al . , 2020 ) and have already shown promising potential for tracking neurons ( Wen et al . , 2018 ) . In this work , we use semi-synthetic data to train a Transformer network , an artificial neural network architecture that has shown great success in natural language processing tasks ( Vaswani et al . , 2017 ) . Transformers incorporate an attention mechanism that can leverage similarities between pairs of inputs to build a rich representation of the input sequence for downstream tasks like machine translation and sentiment prediction . We reasoned this same architecture would be well-suited to extract spatial relationships between neurons in order to build a representation that facilitates finding correspondence to neurons in a template worm . Not only is the Transformer well-suited to learning features for the neural correspondence problem , it also obviates the need to straighten ( Peng et al . , 2008 ) the worm in advance . Until now , existing methods have either required the worm to be straightened in preprocessing ( Bubnis et al . , 2019; Chaudhary et al . , 2021 ) or explicitly transformed them during inference ( Varol et al . , 2020; Nejatbakhsh et al . , 2020 ) . Straightening the worm is a non-trivial task , and it is especially error-prone for complicated poses such as when the worm rolls along its centerline . Finally , one of the main advantages of the Transformer architecture is that it permits parallel processing of the neural positions using modern GPU hardware . In contrast to existing methods , which have not been optimized for speed , the Transformer can make real-time predictions once it has been trained . This speed is a necessary step toward bringing real-time applications ( Clancy et al . , 2014; Hochbaum et al . , 2014 ) to freely moving animals .
We developed a fast deep neural correspondence ( fDNC ) model that seeks to find the correspondence between configurations of C . elegans neurons in different individuals or in the same individual across time ( Figure 1 ) . We used a deep learning artificial neural network architecture , called the transformer architecture ( Vaswani et al . , 2017 ) , that specializes at finding pairs of relations in datasets , Figure 1F . The transformer architecture identified similarities across spatial relations of neurons in a test and a template to identify correspondences between the neurons . Within a single individual , neural positions vary as the worm moves , deforms , and changes its orientation and pose . Across isogenic individuals , there is an additional source of variability that arises from the animal’s development . In practice , further variability also arises from experimental measurements: neuron positions must first be extracted from fluorescent images ( Figure 1A ) , and slight differences in label expression , imaging artifacts , and optical scattering all contribute to errors in segmenting individual neurons . We created a simulator to model these different sources of variability and used it to generate realistic pairs of empirically derived semi-synthetic animals with known correspondence between their neurons for training our model ( Figure 1B , E ) . The simulator took configurations of neuron positions that lacked ground truth from real worms as inputs and then scaled and deformed them , forced them to adopt different poses sampled from real worms , and then introduced additional sources of noise to generate many new semi-synthetic individuals . We then trained our fDNC model on these experimentally derived semi-synthetic individuals of different sizes and poses . Training our model on the empirically derived semi-synthetic data offered advantages compared to experimentally acquired data . First , it allowed us to train on larger datasets than would otherwise be practical . We trained on 2 . 304 × 105 semi-synthetic individuals , but only seeded our simulator with unlabeled neural configurations from experimentally acquired recordings of 12 individuals ( 4 × 103 volumes spread across the 12 individuals , all of which lacked ground-truth correspondence ) . Second , we did not need to provide ground truth correspondence because the simulator instead generates its own ground truth correspondence between semi-synthetic individuals , thereby avoiding a tedious and error prone manual step . Consequently , no experimentally acquired ground truth correspondence was used to train the model . Later in the work , we use ground truth information from human annotated NeuroPAL ( Yemini et al . , 2021 ) strains to evaluate the performance of our model , but no NeuroPAL strains were used for training . Importantly , the amount of test data with ground truth correspondence needed for evaluating performance is much smaller than the amount of training data that would be needed for training . Third , by using large and varied semi-synthetic data , we force the model to generalize its learning to a wide range of variabilities in neural positions and we avoid the risks of overtraining on idiosyncrasies specific to our imaging conditions or segmentation . Overall , we reasoned that training with semi-synthetic data should make the model more robust and more accurate across a wider range of conditions , orientations and animal poses than would be practical with experimentally acquired ground-truth datasets . We trained our fDNC model on 2 . 304 × 105 semi-synthetic individuals ( Figure 1C ) and then , after training , evaluated its performance on 2000 additional held-out semi-synthetic pairs of individuals which had not been accessible to the model during training , Figure 1D and Figure 2 . Model performance was evaluated by calculating the accuracy of the models’ predicted correspondence with respect to the ground truth in pairs of semi-synthetic individuals . One individual is called the ‘test’ and the other is the ‘template’ . Every neuron in the test or template , whichever has fewer is assigned a match . Accuracy is reported as the number of correctly predicted matches between test and template , divided by the total number of ground truth matches in the test and template pair . Our fDNC model achieved 96 . 5% mean accuracy on the 2000 pairs of held-out semi-synthetic individuals . We compared this performance to that of Coherent Point Drift ( CPD ) ( Myronenko and Song , 2010 ) , a classic registration method used for automatic cell annotation . CPD achieved 31 . 1% mean accuracy on the same held-out semi-synthetic individuals . Our measurements show that the fDNC model significantly outperforms CPD at finding correspondence in semi-synthetic data . For the rest of the work , we use experimentally acquired human annotated data to evaluate performance . We next evaluated the fDNC model’s performance at tracking neurons within an individual over time , as is needed , for example , to measure calcium activity in moving animals ( Venkatachalam et al . , 2016; Nguyen et al . , 2016 ) . We evaluated model performance on an experimentally acquired calcium imaging recording of a freely moving C . elegans from Nguyen et al . , 2017 in which a team of human experts had manually tracked and annotated neuron positions over time ( strain AML32 , 1514 volumes , six volumes per second , additional details are describeed in the 'Datasets' section of the 'Materials and methods . ' ) . The recording has sufficiently large animal movement that the average distance a neuron travels between volumes ( 4 . 8 µm ) is of similar scale to the average distance between nearest neuron neighbors ( 5 . 3 µm ) . The recording was excluded from training and from the set of recordings used by the simulator . We collected neuron configurations from all n time points during this recording to form n-1 pairs of configurations upon which to evaluate the fDNC model . Each pair consisted of a test and template . The template was always from the same time point t , while the test was taken to be the volume at any of the other time points . We applied the pre-trained fDNC model to the pairs of neuron configurations and compared the model’s predicted correspondence to the ground truth from manual human tracking ( Figure 3 ) . Across the pairs , the fDNC model showed a mean accuracy of 79 . 1% . We emphasize that the fDNC model achieved this high accuracy on tracking a real worm using only neuron position information even though it is trained exclusively on semi-synthetic data . We compared the performance of our fDNC model to that of CPD Registration , and to Neuron Registration Vector Encoding and clustering ( NeRVE ) , a classical computer vision model that we had previously developed specifically for tracking neurons within a moving animal over time ( Nguyen et al . , 2017; Figure 3C ) . fDNC clearly outperformed CPD achieving 79 . 1% accuracy compared to CPD’s 62 . 7% . Both CPD and fDNC predict neural correspondence of a test configuration by comparing only to a single template . In contrast , the NeRVE method takes 100 templates , where each one is a different neuron configuration from the same individual , and uses them all to inform its prediction . The additional templates give the NeRVE method extra information about the range of possible neural configurations made by the specific individual whose neurons are being tracked . We therefore compared the fDNC model both to the full NeRVE method and also to a restricted version of the NeRVE method in which NeRVE had access only to the same single template as the CPD or fDNC models . ( Under this restriction , the NeRVE method no longer clusters and the method collapses to a series of gaussian mixture model registrations [Jian and Vemuri , 2011] ) . In this way , we could compare the two methods when given the same information . fDNC’s mean performance of 79 . 1% was statistically significantly more accurate than the restricted NeRVE model ( mean 73 . 1% , p=1 . 3×10-140 , Wilcoxon signed rank test ) . The full NeRVE model that had access to additional templates outperformed the fDNC model slightly ( 82 . 9% p=1 . 5×10-102 , Wilcoxon signed rank test ) . Because CPD , NeRVE , and fDNC are all time-independent algorithms , their performance on a given volume is the same , even if nearby volumes are omitted or shuffled in time . One benefit of this approach is that errors from prior volumes do not accumulate over the duration of the recording . To visualize performance over time , we show a volume-by-volume comparison of fDNC’s tracking to that of a human ( Figure 3E ) . We also characterize model performance on a per neuron basis ( Figure 3D ) . Finally , we used fDNC to extract whole brain calcium activity from a previously published recording of a moving animal in which two well-characterized neurons AVAL and AVAR were unambiguously labeled with an additional colored fluorophore ( Hallinen et al . , 2021; Figure 3—figure supplement 1A , Video 1 ) . Calcium activity extracted from neurons AVAL and AVAR exhibited calcium activity transients when the animal underwent prolonged backward locomotion , as expected ( Figure 3—figure supplement 1B ) . We conclude that the fDNC model is suitable for tracking neurons across time and performs similarly to the NeRVE method . In the following sections , we further show that the fDNC method is orders of magnitude faster than NeRVE . Moreover , unlike NeRVE which can only be used within an individual , fDNC is also able to predict the much more challenging neural correspondence across individuals . Because it relies on an artificial neural network , the fDNC model finds correspondence for a set of neurons faster than traditional methods ( Table 1 ) . From the time that a configuration of segmented neurons is loaded onto a GPU , it takes only an average of 10 ms for the fDNC model to predict correspondence for all neurons on a 2 . 4 GHz Intel machine with an NVIDIA Tesla P100 GPU . If not using a GPU , the model predicts correspondence for all neurons in 50 ms . In contrast , on the same hardware it takes CPD 930 ms and it takes NeRVE on average over 10 s . The fDNC model may be a good candidate for potential closed-loop tracking applications because its speed of 100 volumes per second is an order of magnitude faster than the 6–10 volumes per second recording rate typically used in whole-brain imaging of freely moving C . elegans ( Nguyen et al . , 2016; Venkatachalam et al . , 2016 ) . We note that for a complete closed-loop tracking system , fast segmentation algorithms will also be needed in addition to the fast registration and labeling algorithms presented here . The fDNC model is agnostic to the details of the segmentation algorithm so it is well suited to take advantage of fast segmentation algorithms when they are developed . The fDNC model uses built-in libraries to parallelize the computations for labeling a single volume , and this contributes to its speed . In particular , each layer of the neural network contains thousands of artificial neurons performing the same computation . Computations for each neuron in a layer can all be performed in parallel and modern GPUs have as many as 3500 CUDA cores . In practice , the method is even faster for post-processing applications ( not-realtime ) because it is also parallelizable at the level of each volume . Labeling one volume has no dependencies on any previous volumes and therefore each volume can be processed simultaneously . The number of volumes to be processed in parallel is limited only by the number of volumes that can be loaded onto the memory of a GPU . When tracking during post-processing in this work , we used 32 volumes simultaneously . Having shown that fDNC performs well at identifying neurons within the same individual , we wanted to assess its capability to identify neurons across different animals , using neural position information alone , as before . Identifying corresponding neurons across individuals is crucial for studying the nervous system . However , finding neural correspondence across individuals is more challenging than within an individual because there is variability in neuronal position from both the animal’s movement as well as from development . To evaluate the fDNC model’s performance at finding neural correspondence across individuals using only position information , we applied the same semi-synthetically-trained fDNC model to a set of 11 NeuroPAL worms . NeuroPAL worms contain extra color information that allows a human to assign ground truth labels to evaluate the model’s performance . Crucially , the fDNC model was blinded to this additional color information . In these experiments , NeuroPAL color information was only used to evaluate performance after the fact , not to find correspondence . NeuroPAL worms have multicolor neurons labeled with genetically encoded fluorescent proteins ( Yemini et al . , 2021 ) . Only a single volume was recorded for each worm since immobilization is required to capture multicolor information from the NeuroPAL strain . For each of the 11 Neuropal recording , neurons were automatically segmented and manually annotated based on the neuron’s position and color features as described in Yemini et al . , 2021 ( see Figure 4A , B ) . Across the 11 animals , a human assigned a ground-truth label to a mean of 43% of segmented head neurons , providing approximately 58 labeled neurons per animal ( Figure 4C , additional details in 'Datasets' section of 'Materials and Methods' ) . The remaining segmented neurons were not confidently identifiable by the human and thus were left without ground truth labels . We selected as template the recording that contained the largest number of confidently labeled ground turth human annotated neurons . We evaluated our model by comparing its predicted correspondence between neurons in the other 10 test datasets and this template , using only position information ( no color information ) . All 11 ground-truth recordings were held-out in that they were not involved in the generation of the semi-synthetic data that had been used to train the model . We applied the synthetically trained fDNC model to each pair of held-out NeuroPAL test and template recordings and calculated the accuracy as the number of correctly predicted matches divided by the total number of ground truth matches in the pair . Across the 10 pairs of NeuroPAL recordings using position information alone , the fDNC model had a mean accuracy of 64 . 1% , significantly higher than the CPD method’s accuracy of 53 . 1% ( p=0 . 005 , Wilcoxon signed-rank test ) . We wondered whether we could better use the likelihood information about potential matches generated by the algorithm . For each neuron i in the test recording , the fDNC model computes a relative confidence with which that neuron corresponds to each possible neuron j in the template , pij . A Hungarian algorithm finds the most probable match by considering all pijs for all neurons in the test . By default we use this best match in evaluating performance . The pijs also provide the user with a list of alternative matches ranked by the model’s estimate of their respective likelihood . We therefore also assessed the accuracy for the top three most likely matches . Given i and j are ground truth matches , we asked whether the value pij is among the top three values of the set pik where k can be chosen from all the neurons in the template . We defined accuracy as the number of instances in which this criteria was met , divided by the number of ground truth matches in the test template pair . When considering the top three neurons , the fDNC model achieves an accuracy of 76 . 6% using only position information . Data quality , selection criteria , human annotation , hardware , segmentation , and preprocessing can all vary from lab to lab making it challenging to directly compare methods . To validate our model against different measurement conditions and to allow for a direct comparison with another recent method , we applied our fDNC model to a previously published dataset of 9 NeuroPAL individuals ( Chaudhary et al . , 2021 ) . This public dataset used different imaging hardware and conditions and was annotated by human experts from a different group . On this public dataset , using position information alone , our method achieved 78 . 2% accuracy while CPD achieved 58 . 9% , Figure 4F . When assessing the top three candidate accuracy , the fDNC model performance was 91 . 3% . The fDNC model performance was overall higher on the published dataset than on our newly collected dataset presented here . This suggests that our method performs well when applied to real-world datasets in the literature . We further sought to compare the fDNC model to the reported accuracy of a recent model called Conditional Random Fields ( CRF ) from Chaudhary et al . , 2021 by comparing their performance on the same published dataset from that work . There are fundamental differences between the two methods that make a direct comparison of their performance challenging . CRF assigns labels to a test worm . In contrast , fDNC assigns matches between two worms or two configurations , the test and template . To evaluate whether a match is correct using fDNC , we require a ground truth label in both test and template . Consequently , our denominator for accuracy is the intersection of neurons with ground truth labels in test and template . In contrast , the denominator for evaluating accuracy of the CRF model is all neurons with ground truth labels in the test . When applied to the same dataset in Chaudhary et al . , 2021 , fDNC had an accuracy of 78% . But for the purposes of comparison with CRF this could , in principle , correspond to an accuracy of 61 . 2–82 . 5% , depending on how well those neurons in the test that lack ground truth labels in the template were matched . These bounds are calculated for the extreme cases in which neurons with ground truth labels in the test but not in the template are either all matched incorrectly ( 61 . 2% ) or all matched perfectly ( 82 . 5% ) . Seventy-eight percent is the accuracy under the assumption that those neurons with ground truth labels in the test but not in the template are correctly matched at the same rate as those neurons with ground truth labels in both . In other words , we assume the neurons we have ground truth information about are representative of the ones we don’t . For the sake of comparison , we use this assumption to compare fDNC to the published values of CRF ( Table 2 ) . fDNC accuracy is higher than the reported performance for the open atlas variant of CRF . Under the specific assumption described above , it is also higher than the data driven atlas variant , although we note that this could change with different assumptions , and we are unable to test for statistical significance . The fDNC method also offers other advantages compared to the CRF approach in that the fDNC method is optimized for speed and avoids the need to transform the worm into a canonical coordinate system . Importantly , compared to the data-driven atlas variant of the CRF , the fDNC model has an advantage in that it does not require assembling a data-driven atlas from representative recordings with known ground-truth labels . Taken together , we conclude that the fDNC model’s accuracy is comparable to that of the CRF model while also providing other advantages . Our method only takes positional information as input to predict neural correspondence . However , when additional features are available , the position-based predictions from the fDNC model can be combined with predictions based on other features to improve overall performance . As demonstrated in Yemini et al . , 2021 , adding color features from a NeuroPAL strain can reduce the ambiguity of predicting neural correspondence . We applied a very simple color model to calculate the similarity of color features between neuron i in the test recording to every possible neuron j in the template . The color model returns matching probabilities , pijc based on the Kullback-Liebler divergence of the normalized color spectra in a pair of candidate neurons ( details described in Materials and methods ) . The color model is run in parallel to the fDNC model ( Figure 5A ) . Overall matching probabilities pijall that incorporate both color and position information are calculated by combining the color matching probabilities pijc with the position probabilities pij . The Hungarian algorithm is run on the combined matching algorithm to predict the best matches . Adding color information increased the fDNC model’s average accuracy from 64 . 1% to 74 . 7% ( Figure 5B ) when evaluated on our dataset , and improved the accuracy in every recording evaluated . The top three candidate labels attained 92 . 4% accuracy . Accuracy was calculated from a comparison to human ground truth labeling , as before . We chose a trivially simple color model in part to demonstrate the flexibility with which the fDNC model framework can integrate information about other features . Since our simple color model utilized no prior knowledge about the distributions of colors in the worm , we would expect a more sophisticated color model , for example , the statistical model used in Yemini et al . , 2021 , to do better . And indeed that model evaluated on a different dataset is reported to have a higher performance with color than our model on our dataset ( 86% reported accuracy in Yemini et al . , 2021 compared to 75% for the fDNC evaluated here ) . But that model also performs much worse than fDNC when both are restricted to use only neural position information ( 50% reported accuracy for Yemini et al . , 2021 compared to 64% for the fDNC ) . Together , this suggests the fDNC model framework can take advantage of additional feature information like color and still perform relatively well when such information is missing .
Identifying correspondence between constellations of neurons is important for resolving two classes of problems: The first is tracking the identities of neurons across time in a moving animal . The second is mapping neurons from one individual animal onto another , and in particular onto a reference atlas , such as one obtained from electron microscopy ( Witvliet et al . , 2020 ) . Mapping onto an atlas allows recordings of neurons in the laboratory to be related to known connectomic , gene expression , or other measurements in the literature . The fDNC model finds neural correspondence within and across individuals with an accuracy that is comparable or compares favorable to other methods . The model focuses primarily on identifying neural correspondence using position information alone . For tracking neurons within an individual using only position , fDNC achieves a high accuracy of 79% , while for across individuals using only position it achieves 64% accuracy on our dataset and 78% on a published dataset from another group . We expect that an upper bound may exist , set by variability introduced during the animal’s development , that ultimately limits the accuracy with which any human or algorithm can find correspondence across individuals via only position information . For example , pairs of neurons in one individual that perfectly switch position with respect to another individual will never be unambiguously identified by position alone . It is unclear how close fDNC’s performance of 64% on our dataset or 78% on the dataset in Chaudhary et al . , 2021 comes to this hypothetical upper bound , but there is reason to think that at least some room for improvement remains . Specifically , we do not expect accuracy at tracking within an individual to be fundamentally limited , in part because we do not expect two neurons to perfectly switch position on the timescale of a single recording . Therefore fDNC’s 79% accuracy within-individuals suggests room for improving within-individual correspondence , and by extension , across-individual correspondence because the latter necessarily includes all of the variability of the former . One avenue for achieving higher performance could be to improve the simulator’s ability to better capture variability of a real testset , for example by using different choices of parameters in the simulator . Even at the current level of accuracy , the ability to find correspondence across animals using position information alone remains useful . For example , we are interested in studying neural population coding of locomotion in C . elegans ( Hallinen et al . , 2021 ) , and neural correspondence at 64% accuracy will allow us to reject null hypotheses about the extent to which neural coding of locomotion is stereotyped across individuals . The fDNC model framework also makes it easy to integrate other features which further improve accuracy . We demonstrated that color information could be added by integrating the fDNC model with a simple color model to increase overall accuracy . We expect that performance would improve further with a more sophisticated color model that takes into account the statistics of the colors in a NeuroPAL worm ( Yemini et al . , 2021 ) . The fDNC model framework offers a number of additional advantages beyond accuracy . First , it is versatile and general . The same pre-trained model performed well at both tracking neurons within a freely moving individual across time and at finding neural correspondence across different individuals . Without any additional training , it achieved even higher accuracy on a publicly accessible dataset acquired on different hardware with different imaging conditions from a different group . This suggests that the framework should be applicable to many real-world datasets . The model provides probability estimates of all possible matches for each neuron . This allows an experimenter to consider a collection of possible matches such as the top three most likely . In contrast to previous methods , an advantage of the fDNC method is that it does not require the worm to be straightened , axis aligned , or otherwise transformed into a canonical coordinate system . This eliminates an error-prone and often manual step . Instead , the fDNC model finds neural correspondence directly from neural position information even in worms that are in different poses or orientations . Importantly , the model is trained entirely on semi-synthetic data , which avoids the need for large experimentally acquired ground truth datasets to train the artificial neural network . Acquiring ground truth neural correspondence in C . elegans is time consuming , error prone , and often requires manual hand annotation . The ability to train the fDNC model with semi-synthetic data derived from measurements alleviates this bottleneck and makes the model attractive for use with other organisms with stereotyped nervous systems where ground truth datasets are similarly challenging to acquire . The model is also fast and finds neural correspondence of a new neural configuration in 10 ms . The development of fast algorithms for tracking neurons are an important step for bringing real-time closed loop applications such as optical brain-machine interfaces ( Clancy et al . , 2014 ) and optical patch clamping ( Hochbaum et al . , 2014 ) to whole-brain imaging in freely moving animals . By contrast , existing real-time methods for C . elegans in moving animals are restricted to small subsets of neurons , are limited to two-dimensions , and work only at low spatial resolution ( Leifer et al . , 2011; Stirman et al . , 2011; Kocabas et al . , 2012; Shipley et al . , 2014 ) . We note that to be used in a real-time closed loop application , our fDNC model would need to be combined with faster segmentation algorithms because current segmentation algorithms are too slow for real-time use . Because segmentation can be easily paralellized , we expect that faster segmentation algorithms will be developed soon . Many of the advantages listed here stem from the fDNC model’s use of the transformer architecture ( Vaswani et al . , 2017 ) in combination with supervised learning . The transformer architecture , with its origins in natural language processing , is well suited to find spatial relationships within a configuration of neurons . By using supervised learning on empirically derived semi-synthetic training data of animals in a variety of different poses and orientations , the model is forced to learn relative spatial features within the neurons that are informative for finding neural correspondence across many postures and conditions . Finally , the transformer architecture leverages recent advances in GPU parallel processing for speed and efficiency , which is an important step toward future real-time applications .
Neural configurations acquired as part of this study have been posted in an Open Science Foundation repository with DOI:10 . 17605/OSF . IO/T7DZU available at https://dx . doi . org/10 . 17605/OSF . IO/T7DZU . The publicly accessible dataset from Chaudhary et al . , 2021 is available at https://github . com/shiveshc/CRF_Cell_ID , commit 74fb2feeb50afb4b840e8ec1b8ee7b7aaa77a426 . Datasets from Nguyen et al . , 2017 and Hallinen et al . , 2021 are publicly available in repositories associated with their respective publications . Those strains used to create new datasets presented in this work are listed in Key Resources . All strains mentioned in this study , including those involved in previously published datasets , are listed in Table 4 . All strains express a nuclear localized red fluorescent protein in all neurons . All but strains OH15495 and OH15262 also express nuclear localized GCaMP6s in all neurons . NeuroPAL ( Yemini et al . , 2021 ) strains further express many additional fluorophores . To image neurons in the head of freely moving worms , we used a dual-objective spinning-disk based tracking system ( Nguyen et al . , 2016 ) ( Yokogawa CSU-X1 mounted on a Nikon Eclipse TE2000-S ) . Fluorescent images of the head of a worm were recorded through a 40x objective with both 488- and 561 nm excitation laser light as the animal crawled . The 40x objective translated up and down along the imaging axis to acquire 3D image stacks at a rate of 6 head volumes/s . To image neurons in the immobile multi-color NeuroPAL worms ( Yemini et al . , 2021 ) , we modified our setup by adding emission filters in a motorized filter wheel ( Prior ProScan-II ) , and adding a Stanford Research Systems SR474 shutter controller ( with SR475 shutters ) to programmatically illuminate the worm with different wavelength laser light . We use three lasers of different wavelengths: 405 nm ( Coherent OBIS-LX 405 nm 100 mW ) , 488 nm ( Coherent SAPPHIRE 488 nm 200 mW ) , and 561 nm ( Coherent SAPPHIRE 561 nm 200 mW ) . Only one laser at a time reached the sample , through a 40x oil-immersion objective ( 1 . 3 NA , Nikon S Fluor ) . The powers measured at the sample , after spinning disk and objective , were 0 . 14 mW ( 405 nm ) , 0 . 35 mW ( 488 nm ) , and 0 . 36 mW ( 561 nm ) . In the spinning disk unit , a dichroic mirror ( Chroma ZT405/488/561tpc ) separated the excitation from the emission light . The latter was relayed to a cooled sCMOS camera ( Hamamatsu ORCA-Flash 4 . 0 C11440-22CU ) , passing through the filters mounted on the filter wheel ( Table 5 ) . Fluorescent images were acquired in different ‘channels’ , that is , different combinations of excitation wavelength , emission filter , and camera exposure time ( Table 6 ) . The acquisition was performed using a custom software written in LabVIEW that specifies the sequence of channels to be imaged , and controls shutters , filter wheel , piezo translator , and camera . After setting the z position , the software acquires a sequence of images in the specified channels . We extracted the position of individual neurons from 3D fluorescent images to generate a 3D point cloud ( Figure 1A ) . This process is called segmentation and the fDNC model is agnostic to the specific choice of the segmentation algorithm . Segmentation was always performed on tagRFP , never on GCaMP . For recordings of strains AML32 , we used a segmentation algorithm adopted from Nguyen et al . , 2017 . We first applied a threshold to find pixels where the intensities are significantly larger than the background . Then , we computed the 3D Hessian matrix and its eigenvalues of the intensity image . Candidate neurons were regions where the maximal eigenvalue was negative . Next , we searched for the local intensity peaks in the region and spatially disambiguated peaks in the same region with a watershed separation based on pixel intensity . For recordings of NeuroPAL strains , we used the same segmentation algorithm as in Yemini et al . , 2021 . The publicly accessible dataset from Chaudhary et al . , 2021 used in Figure 4 had already been segmented prior to our use . We developed a simulator to generate a large training set of semi-synthetic animals with known neural correspondence . The simulator takes as its input the point clouds collected from approximately 4000 volumes spread across 12 recordings of freely moving animals . Each recording contains roughly 3000 volumes . For each volume , the simulator performs a series of stochastic deformations and transformations to generate 64 new semi-synthetic individuals where the ground truth correspondence between neurons in the individuals and the original point cloud is known . A total of 2 . 304 × 105 semi-synthetic point clouds were used for training . The simulator introduces a variety of different sources of variability and real-world deformations to create each semi-synthetic point cloud ( Figure 1B , E ) . The simulator starts by straightening the worm in the XY plane using its centerline so that it now lies in a canonical worm coordinate system . Before straightening , Z is along the optical axis and XY are defined to be perpendicular to the optical axis and are arbitrarily set by the orientation of the camera . After straightening , the animal’s posterior-anterior axis lies along the X axis . To introduce animal-to-animal variability in relative neural position , a non-rigid transformation is applied to the neuron point cloud against a template randomly selected from recordings of the real observed worms using coherent point drift ( CPD ) ( Myronenko and Song , 2010 ) . To add variability associated with rotation and distortion of the worm’s head in the transverse plane , we apply a random affine transformation to the transverse plane . To simulate missing neurons and segmentation errors , spurious neurons are randomly added , and some true neurons are randomly removed , for up to 20% of the observed neurons . To introduce variability associated with animal pose , we randomly deform the centerline of the head . Lastly , to account for variability in animals’ size and orientation , a random affine transformation in the XY plane is applied that rescaled the animal’s size by up to 5% . With those steps , the simulator deforms a sampled worm and generates a new semi-synthetic worm with different orientation and posture while maintaining known correspondence . Centerlines generated by the simulator were directly sampled from recordings of real individuals . The magnitude of added Gaussian noise was arbitrarily set to have a standard deviation of 0 . 42 µm . To evaluate performance , putative matches are found between template and test , and compared to ground truth . Every segmented neuron in the test or template ( whichever has fewer ) is assigned a match . Accuracy is defined as the number of proposed matches that agree with ground truth , divided by the total number of ground truth matches . The number of ground truth matches is a property of the dataset used to evaluate our model , and is listed in Table 3 . The recently developed NeuroPAL strain ( Yemini et al . , 2021 ) expresses four different genetically encoded fluorescent proteins in specific expression patterns to better identify neurons across animals . Manual human annotation based on these expression patterns serves as ground truth in evaluating our model’s performance at finding across-animal correspondence . In Figure 5B , we also explored combining color information with our fDNC model . To do so , we developed a simple color matching model that operated in parallel to our position-based fDNC model . Outputs of both models were then combined to predict the final correspondence between neurons . Our color matching model consists of two steps: First , the intensity of each of the color channels is normalized by the total intensity . Then the similarity of color for each pair of neurons is measured as the inverse of the Kullback–Leibler divergence between their normalized color features . To calculate the final combined matching matrix , we add the color similarity matrix to the position matching log probability matrix from our fDNC model . The similarity matrix of color is multiplied by a factor λ . We chose λ=60 so that the amplitude of values in the similarity matrix of color is comparable to our fDNC output . We note the matching results are not particularly sensitive to the choice of λ . The most probable matches are obtained by applying Hungarian algorithm on the combined matching matrix . Source code in Python is provided for the model , for the simulator , and for training and evaluation . A jupyter notebook with a simple example is also provided . Code is available at https://github . com/XinweiYu/fDNC_Neuron_ID ( Yu , 2021; copy archived at swh:1:rev:19c678781cd11a17866af7b6348ac0096a168c06 ) . | Understanding the intricacies of the brain often requires spotting and tracking specific neurons over time and across different individuals . For instance , scientists may need to precisely monitor the activity of one neuron even as the brain moves and deforms; or they may want to find universal patterns by comparing signals from the same neuron across different individuals . Both tasks require matching which neuron is which in different images and amongst a constellation of cells . This is theoretically possible in certain ‘model’ animals where every single neuron is known and carefully mapped out . Still , it remains challenging: neurons move relative to one another as the animal changes posture , and the position of a cell is also slightly different between individuals . Sophisticated computer algorithms are increasingly used to tackle this problem , but they are far too slow to track neural signals as real-time experiments unfold . To address this issue , Yu et al . designed a new algorithm based on the Transformer , an artificial neural network originally used to spot relationships between words in sentences . To learn relationships between neurons , the algorithm was fed hundreds of thousands of ‘semi-synthetic’ examples of constellations of neurons . Instead of painfully collated actual experimental data , these datasets were created by a simulator based on a few simple measurements . Testing the new algorithm on the tiny worm Caenorhabditis elegans revealed that it was faster and more accurate , finding corresponding neurons in about 10ms . The work by Yu et al . demonstrates the power of using simulations rather than experimental data to train artificial networks . The resulting algorithm can be used immediately to help study how the brain of C . elegans makes decisions or controls movements . Ultimately , this research could allow brain-machine interfaces to be developed . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"physics",
"of",
"living",
"systems",
"tools",
"and",
"resources",
"neuroscience"
] | 2021 | Fast deep neural correspondence for tracking and identifying neurons in C. elegans using semi-synthetic training |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.